Original Research

In response to the public health consequences of domestic violence and the number of battered women whom physicians see in their practices,^1-4 medical organizations including the American Medical Association and the American College of Obstetricians and Gynecologists have called for physicians to act as agents of change in abused women’s lives.^5,6 In the late 1980s and early 1990s these organizations and others issued guidelines and mandates based on information from domestic violence experts that outlined how physicians should intervene.^6-8 Unfortunately, these recommendations are not specific enough and do not seem to have improved their responses to violence against women^9-11; many physicians are simply not asking women about violence,^12-17 and women whose health problems result from abuse are not receiving the health care they need.^9,10,18-20

Physicians cite many barriers to intervening with victims, including patient evasiveness and failure to disclose information, lack of time and support resources, lack of education or training, fear of offending the patient, inability to “fix it,” and frustration with lack of change in the patient’s situation or the patient’s unresponsiveness to advice.^16,18,21-28 Primary care physicians in the qualitative study by Sugg and Inui²⁸ characterized talking about domestic violence with patients as opening Pandora’s box and associated the act of even asking about domestic violence with unleashing their own fears and discomforts.

Despite the barriers, some physicians are committed to addressing the underlying health problems of abused women. How do these physicians intervene, and what motivates them to continue in their commitment? In previous work,¹⁶ we described how physicians with expertise in domestic violence identify victims. With this study we explored how physicians with experience in identifying victims tried to help.

Methods

Participants

Qualitative research commonly uses purposive sampling, a method in which the participants best suited to provide a full description of the research topic are intentionally selected. We sought a sample of physicians in the San Francisco Bay Area who had experience in identifying and intervening with victims of domestic violence. To identify important common patterns that cut across different settings^29,30 physicians from 3 medical specialties were sought: primary care (family practice and general internal medicine), obstetrics and gynecology, and emergency medicine.

We conducted our recruitment in consultation with a professional survey research organization. Thirteen physicians known to have domestic violence experience, and additional physicians selected from the yellow pages, were screened and asked to participate in a study exploring the most effective ways for the health care system to meet the needs of victims of domestic violence. Eligible participants were asked to identify other colleagues who are concerned about and treated victims of domestic violence, and these individuals were screened and asked to participate and to identify others. Physicians were eligible if they worked directly with patients 20 or more hours per week, had identified and intervened with victims of domestic violence, and were somewhat confident or very confident about addressing domestic violence issues with patients.

Recruitment ceased when the goal of 12 to 22 physicians in each medical specialty who had the relevant domestic violence experience was reached.

Focus Group Method and Data Collection

In comparison with survey or one-on-one interview formats, the focus group approach allows for a more extensive exploration of the area under discussion. Participants can collectively explore different experiences and perspectives, generate ideas, and debate and compare their ideas with those of others in the group.³¹ Six focus groups ranging in size from 6 to 11 individuals were conducted during a 3-week period in January and February 1998. Each group was facilitated by 2 moderators who were members of the research team. The sessions lasted approximately 90 minutes and were held in professional focus group settings that allowed hidden viewing. Several researchers viewed the groups from behind 2-way mirrors and completed field notes that were later compared with the observations of the moderators. Before each focus group session written informed consent was obtained from all participants and a written background survey was administered to gather demographic and practice information. Participants received a small stipend for participating. Study procedures were approved by the University of California San Francisco Committee on Human Research.

We used a semistructured guide that allowed the facilitators to follow certain topics and open new lines of inquiry when appropriate.³² Open-ended questions were formulated based on our previous interviews with survivors of domestic violence³³ and a review of the literature ( Table 1 ). Audiotapes of the focus group sessions were transcribed by research staff; the principal investigator reviewed these transcripts for accuracy.

Coding and Analysis

For initial analysis we conducted multiple readings of transcripts to identify prominent themes. The investigators independently reviewed the transcripts and then met to review and discuss differences of opinion about interpretations and to further refine themes driven by the words and phrases of the participants. Through this process and the constant comparison of new data against emerging themes, a template of open codes was constructed. The transcripts were coded and specific themes within the narratives of the participants were identified in accordance with standard qualitative analytic convention.³⁴ Coded data were organized using NUD*IST 4.0 software (Qualitative Solutions and Research; Victoria, Australia). This software helps ensure the consistency of study findings and creates an audit trail.³⁵ The data were interpreted in the context of the original focus group sessions and the current literature. The final coding scheme and analysis of the findings were reviewed, and disagreements were discussed by the team until consensus was reached.

To further enhance the credibility of the findings, a qualitative technique called member check was used.³⁶ Results were directed back to 3 research participants to confirm that their experiences and those of other participants in their focus group were reflected in the findings.

Results

Of the 80 physicians who were screened, 53 were eligible, and 45 were able to attend the focus group sessions. Their characteristics are presented in Table 2 . The participants reported that they had identified an average of 28 patients per year as having been physically abused by an intimate partner, and they thought they had helped approximately 60% of those patients.

Helpful Intervention Techniques

Even these physicians reported sometimes feeling overwhelmed, frustrated, and incompetent regarding their role in domestic violence cases. They believed, however, that addressing partner abuse was part of their job and reported various ways that they have tried to help battered women improve their situation and their health. Our data analyses revealed that the following themes were common across specialties.

Give Validating Messages. The most common aspect of intervention was validation. Whatever their approach to helping, these physicians gave compassionate messages that validated the woman’s worth as a human being and indicated that the abuse was undeserved. One participant put it this way: “Just my being there, caring about them consistently, giving another message [helped]: You are worth caring about, you are deserving, you are valuable.” Physicians tended to embed this kind of attitude and message into their interventions with abused patients, making validation the foundation of their interactions with them.

Break Through Denial and Plant Seeds for Change. Physicians reported that within the context of a trusting relationship they tried to break through the denial these women presented about the seriousness of their experiences. Some physicians reported labeling the abuse for what it is, blatantly wrong and criminal. They believed that over time they could help victims to begin to see this reality and change their situation. One participant said:

I let them know that what’s going on is outlandishly not right, that they don’t deserve to have that happen. It’s frankly illegal, and you can bring charges against someone for doing that. Sometimes people can be shocked by finding out that that’s the case. You can plant a seed about their self-esteem … and their ability over time to change that situation, but piecemeal.

Another physician reported showing women the photographs taken of past injuries to remind them of the partner’s pattern of abusive behavior: “We begin every session with: Do you remember that? Sometimes the reaction is: No, it didn’t happen that way. But the photograph just sits there.”

Listen Nonjudgmentally. Physicians described listening and attending to the whole person as central to providing good health care to all patients, especially victims of abuse. In the context of listening they reported on the need to maintain a healing attitude by banishing criticism, blame, and judgment, but agreed that achieving this was difficult and required letting go of the desire to fix it by treating the women as competent adults. One participant said:

I try to get across just from my tone of voice primarily, that I’m not judging them. Because I made that mistake quite a while ago—my judgment was right away: Well, this is terrible; you’ve got to get out. And I could watch the person psychologically fly away from me. So in order to maintain that [trusting] space, that connection with them, it’s really important for me to get clear that I’m going to listen and not judge them. And it’s all going to change on their time.

Document, Refer, and Help the Patient Plan for Safety. These physicians stated that they were careful to write down the specifics of what the woman said. In addition to medical charting, some took photographs of any injuries with color Polaroid cameras. One physician stated that for the photographs to be useful in court “you have to include their face so that a lawyer can’t argue that you’re taking a photograph of someone else.”

The most practicable example of documentation was the development of a domestic violence packet which included a body chart, an instruction list for documentation, a compartment for the color Polaroid photograph, a handy tear-out sheet for information services (resources, shelters), and a telephone number for the police.

In general when physicians knew or suspected abuse they offered information about domestic violence and referrals to local community resources, hot lines, and shelters. Some stated that patients often refused referrals and that they kept referral sheets in the waiting room so that individuals could decide on their own whether to take one.

Other physicians stated that on occasion women had made statements such as: “I can’t take that (handout) home … it’s like a flag in front of a bull.” To address this problem some physicians reported putting business-size cards with domestic violence hot line numbers (eg, local hot line numbers, shelter numbers, or community resource numbers) in all the bathrooms, sometimes the only place where the batterer could not easily follow a woman. Participants reported:

[The cards] are constantly replenished [by housekeeping] … and one of the things I tell people if they’re in an explosive situation is to put it in their shoe, in the insole.

I have a very small practice. I do only office gynecology, but I put about 10 cards a week [in the bathroom]. I would bet 2 disappear in a week. Isn’t that amazing? And this is a fairly affluent area.

Physicians described various ways they had tried to help women plan for their own immediate and ongoing safety. Some physicians talked about trying to stay aware and sensitive to the safety needs of women whose partners are controlling them through abuse, even when the partners are not currently threatening violence. One participant said:

I’ll try to role-play with them … how are they going to deal with telling their partner that they have this infection or that they really want to use this type of birth control. I’ll say, “Some people in your situation could have a fight with their significant other,” and go through predicting some possibilities. Sometimes you can see them start to close down because they know that could happen, or this is a repeated thing that they keep getting that they have no control over. So, I’ll say, “Well, I have other patients in this situation who sometimes need a safe place to go,” or I’ll talk about what somebody else did and at the same time give them some information.

Physicians also reported counseling victims to keep a suitcase packed and have 24-hour hot line numbers or contact numbers for safe places, and helping them to specify what circumstances should cause them to call the police.

Although physicians stated that acute cases were rare outside of the emergency department, they described attempts to ensure safety when the woman’s life was in immediate danger. These attempts included (1) working as a team to separate the partner from the woman (eg, the nurse talked with the abusive partner in the waiting room while the physician cared for the victim and, with the victim’s permission, called the police and a domestic violence advocate to remove her from the abusive home); (2) making excuses to separate the abuser from the victim in the immediate situation (eg, taking the woman for tests); and (3) admitting women who could not be placed in a shelter into the hospital under a false name. One physician reported that the hospital at which she had trained had a safe bed designated for victims of rape, domestic violence, and other assaults.

Using a Team Approach. In general, the physicians agreed that it takes a team approach to intervene successfully with victims of domestic violence. Some expressed frustration about accessing community referrals and discussed the benefits to victims of readily accessible resources on site. The on-site resources referred women directly to the nurse, rape crisis counselor, social worker, behavioral medicine counselor, or psychologist, who counseled the women and conducted follow-up. Some physicians without access to onsite counselors or social workers described making domestic violence part of every staff member’s educational process when they come on board.

Physicians described how intervention demanded a certain amount of flexibility of roles, with nurses and physicians playing off each other in tag-team fashion, as necessary. One physician said:

Sometimes when I finish with an exam, I’ll tell the nurse that I suspect something, so when the nurse is giving the discharge instructions, she’ll also re-approach certain kinds of issues and give the woman another opportunity to talk about [the abuse], once she has gotten dressed and composed herself. The door is closed. It’s one-on-one.

Some physicians described how their prenatal team takes advantage of a “window of opportunity” and has helped women get out of their situations and into counseling:

We have a prenatal team that really works together … our nurse, our social worker, our nutritionist, the receptionist, everybody.… It’s a real intense time. But I think once they get out of pregnancy, we really lose that ability to make a change in their lives. It’s a real window of opportunity.

Prioritize Domestic Violence. Even the committed physicians in our study expressed conflict about taking the time to intervene once they had identified abuse. Some physicians advocated dropping the medical procedure (even if that means the loss of reimbursement) to spend the rest of the patient’s time dealing with the abuse. Physicians also described prioritizing domestic violence by conducting continuing education courses and meetings for everyone in the department about rape, domestic violence, and child and elder abuse. One participant reported:

We try to create a culture of caring about domestic violence so that nurses who think they’ve recognized someone as being a nondeclared victim won’t be told, “I’m too busy” by a physician. And so when physicians say, “I think that might be a domestic violence victim, could you go talk to her?” the nurse will see that as a priority. And if anybody asks her, “How come you haven’t got that IV started?” she or he could say, “Because I was in talking to this person trying to determine whether they were a domestic violence victim.”

Small Victories Offer Positive Feedback

These physicians reported receiving little direct feedback about the effectiveness of their interventions with battered women. Yet, they also reported a range of rewards for intervening, from seeing a patient really change her life to glimpsing shifts in the way a woman thinks about herself and the relationship. One physician said:

And the rewarding piece for me comes when at some point she looks up and notices, and you can see this change of realizing that she’s cared about and then what that must mean to her, that she’s worth something. And then later on [there are] those little steps that you can see people make when they feel like they’re worth something. That’s the most ongoing and rewarding thing.

Discussion

The themes described by the purposive sample of physicians in our study offer insight into the process of intervention with victims of domestic violence and help delineate practicable examples of how to apply interventions ( Table 3 ). The behaviors described are supported by quantitative and qualitative data from battered women.^33,37

These physicians described the foundation of intervention with victims as giving victims the message that they do not deserve abuse and that they are worth caring about. Battered women themselves report that validation is an important message. In a recent survey,³⁷ battered women rated validating statements and compassion from physicians as among the most desirable interventions, equal to safety planning and offering referrals. In another study,³³ survivors of domestic violence described how validation from a health care professional had not only provided relief and comfort, but also “started the wheels turning” toward realizing the seriousness of the situation. These women reported that validation helped them, regardless of whether they had disclosed the abuse or the health care professional had identified it.

Women who are being controlled through abuse by an intimate partner live with debilitating feelings of denial, shame, and humiliation that are sometimes reinforced in health care encounters and keep victims from seeking and receiving optimal care.^33,38-41 The physicians in our study recognized these barriers and made efforts to help women break through their denial and plant seeds for change. They also made efforts—and learned through trial and error—to listen and be nonjudgmental. Both of these behaviors were rated as highly desirable by battered women.³⁷ Physician statements made within the context of a trusting relationship can serve to remind women of the seriousness of their situation. Physician behaviors that convey respect through tone of voice and body language could lessen a victim’s shame and help her make small changes over time to improve her situation and her health.

The data we presented on documenting abuse, providing referrals, and planning for safety concur with the practices recommended by Physicians for a Violence-Free Society.⁴² We suggest that health care settings develop a domestic violence packet containing a body chart, documentation instructions, and referral sheets. We also suggest they provide Polaroid cameras to document specific injuries, since pictures offer an inviolable record of the abuse.⁴² Survivors report that the process of documenting abuse can serve to validate the individual if accompanied by genuine nonjudgmental statements of concern.³³

Although the physicians in our study were aware of the need to provide victims with referrals to community resources and assess their safety needs,^6,8,42,43 they had developed their own styles of intervention and admitted that victims sometimes refused referrals. One solution offered by participants in our study is to put easy-to-hide business-size cards with local domestic violence hot-line and shelter numbers in all of the bathrooms. We also suggest that physicians continue to offer referrals time after time: repetitive offering or availability of referrals may help survivors feel like they are not alone and may reassure them that support is available within and outside the health care system when they are ready to seek it. Physicians should remember that a woman may be able to talk about the abuse long before she can actualize any change. They should also be aware that ending the relationship does not necessarily end the abuse; it may escalate it.⁴⁴ The study physicians were careful to consider safety from the battered woman’s point of view and to take preventive measures. We suggest that physicians review their options for facilitating safety (ie, availability of resources and time) and, when necessary, connect the victim by telephone to an agency trained in assessing and planning for their safety. Battered women report that they want physicians to offer referrals and help them plan for safety.³⁷

Although current guidelines call for physicians to play a large role in identifying, intervening with, and following up on cases of partner abuse,^6,45,46 the physicians in our study emphasized the need to work as a team to identify and provide optimal care to victims. This requires flexibility of roles within the health care team and ready access to on-site and community domestic violence resources. In an attempt to improve health care for victims of domestic violence, experts and researchers in the field have proposed simplifying and limiting the tasks of physicians in this area. One model uses the acronym AVDR: physicians should ask patients about abuse; provide validating messages that battering is wrong and the patient is a worthy individual; document presenting signs, symptoms, and disclosures; and refer victims to specialists in domestic violence.⁴⁷ At that point specialists on site or on call from the community would assess the patient’s safety, make appropriate safety plans, and perform other in-depth interventions.

Physicians face ever-increasing demands on their limited time, yet these physicians committed to helping battered women found multiple ways to enable them to intervene. The holistic approaches described here—using a team approach, prioritizing domestic violence, developing a culture of caring—send a powerful message of prevention and intervention to victims: Battering is not a private, shameful issue, but a health care issue of great concern to physicians. These approaches also provide health care professionals with systematic support for helping battered women, perhaps allowing committed physicians to act as agents of change in battered women’s lives.⁵

Women who are being controlled by the abusive actions of their intimate partners report that even small signs of compassion from health care professionals have made a difference to them. As stated by physicians in this study and by survivors in our previous study,³³ these acts of caring plant the seeds for change. In their efforts to help battered women, physicians must remember that incremental changes and small moments of recognition can eventually lead to major shifts in the lives of these women. Every time physicians successfully intervene with a person whose health problems are caused by abuse they have engendered a positive outcome.

Acknowledgments

Our project has been supported by the National Institute of Mental Health Grant #1 R01 MH51580. We thank the physicians who participated in the focus groups and those who participated in reviewing the study findings. We also thank Stephanie Greer and Survey Methods Group for their assistance in recruiting physician participants and organizing the focus groups; Candace Love, PhD, and Richard Carlton, MPH, for assisting the authors with moderating focus group sessions; Priscilla Abercrombie, NP, PhD, for assisting with coding the data; Karen Herzig, PhD, for assisting with the literature review; and Jennifer Fechner for transcribing the focus group session audiotapes and proofreading the manuscript.

In response to the public health consequences of domestic violence and the number of battered women whom physicians see in their practices,^1-4 medical organizations including the American Medical Association and the American College of Obstetricians and Gynecologists have called for physicians to act as agents of change in abused women’s lives.^5,6 In the late 1980s and early 1990s these organizations and others issued guidelines and mandates based on information from domestic violence experts that outlined how physicians should intervene.^6-8 Unfortunately, these recommendations are not specific enough and do not seem to have improved their responses to violence against women^9-11; many physicians are simply not asking women about violence,^12-17 and women whose health problems result from abuse are not receiving the health care they need.^9,10,18-20

Physicians cite many barriers to intervening with victims, including patient evasiveness and failure to disclose information, lack of time and support resources, lack of education or training, fear of offending the patient, inability to “fix it,” and frustration with lack of change in the patient’s situation or the patient’s unresponsiveness to advice.^16,18,21-28 Primary care physicians in the qualitative study by Sugg and Inui²⁸ characterized talking about domestic violence with patients as opening Pandora’s box and associated the act of even asking about domestic violence with unleashing their own fears and discomforts.

Despite the barriers, some physicians are committed to addressing the underlying health problems of abused women. How do these physicians intervene, and what motivates them to continue in their commitment? In previous work,¹⁶ we described how physicians with expertise in domestic violence identify victims. With this study we explored how physicians with experience in identifying victims tried to help.

Methods

Participants

Qualitative research commonly uses purposive sampling, a method in which the participants best suited to provide a full description of the research topic are intentionally selected. We sought a sample of physicians in the San Francisco Bay Area who had experience in identifying and intervening with victims of domestic violence. To identify important common patterns that cut across different settings^29,30 physicians from 3 medical specialties were sought: primary care (family practice and general internal medicine), obstetrics and gynecology, and emergency medicine.

We conducted our recruitment in consultation with a professional survey research organization. Thirteen physicians known to have domestic violence experience, and additional physicians selected from the yellow pages, were screened and asked to participate in a study exploring the most effective ways for the health care system to meet the needs of victims of domestic violence. Eligible participants were asked to identify other colleagues who are concerned about and treated victims of domestic violence, and these individuals were screened and asked to participate and to identify others. Physicians were eligible if they worked directly with patients 20 or more hours per week, had identified and intervened with victims of domestic violence, and were somewhat confident or very confident about addressing domestic violence issues with patients.

Recruitment ceased when the goal of 12 to 22 physicians in each medical specialty who had the relevant domestic violence experience was reached.

Focus Group Method and Data Collection

In comparison with survey or one-on-one interview formats, the focus group approach allows for a more extensive exploration of the area under discussion. Participants can collectively explore different experiences and perspectives, generate ideas, and debate and compare their ideas with those of others in the group.³¹ Six focus groups ranging in size from 6 to 11 individuals were conducted during a 3-week period in January and February 1998. Each group was facilitated by 2 moderators who were members of the research team. The sessions lasted approximately 90 minutes and were held in professional focus group settings that allowed hidden viewing. Several researchers viewed the groups from behind 2-way mirrors and completed field notes that were later compared with the observations of the moderators. Before each focus group session written informed consent was obtained from all participants and a written background survey was administered to gather demographic and practice information. Participants received a small stipend for participating. Study procedures were approved by the University of California San Francisco Committee on Human Research.

We used a semistructured guide that allowed the facilitators to follow certain topics and open new lines of inquiry when appropriate.³² Open-ended questions were formulated based on our previous interviews with survivors of domestic violence³³ and a review of the literature ( Table 1 ). Audiotapes of the focus group sessions were transcribed by research staff; the principal investigator reviewed these transcripts for accuracy.

Coding and Analysis

For initial analysis we conducted multiple readings of transcripts to identify prominent themes. The investigators independently reviewed the transcripts and then met to review and discuss differences of opinion about interpretations and to further refine themes driven by the words and phrases of the participants. Through this process and the constant comparison of new data against emerging themes, a template of open codes was constructed. The transcripts were coded and specific themes within the narratives of the participants were identified in accordance with standard qualitative analytic convention.³⁴ Coded data were organized using NUD*IST 4.0 software (Qualitative Solutions and Research; Victoria, Australia). This software helps ensure the consistency of study findings and creates an audit trail.³⁵ The data were interpreted in the context of the original focus group sessions and the current literature. The final coding scheme and analysis of the findings were reviewed, and disagreements were discussed by the team until consensus was reached.

To further enhance the credibility of the findings, a qualitative technique called member check was used.³⁶ Results were directed back to 3 research participants to confirm that their experiences and those of other participants in their focus group were reflected in the findings.

Results

Of the 80 physicians who were screened, 53 were eligible, and 45 were able to attend the focus group sessions. Their characteristics are presented in Table 2 . The participants reported that they had identified an average of 28 patients per year as having been physically abused by an intimate partner, and they thought they had helped approximately 60% of those patients.

Helpful Intervention Techniques

Even these physicians reported sometimes feeling overwhelmed, frustrated, and incompetent regarding their role in domestic violence cases. They believed, however, that addressing partner abuse was part of their job and reported various ways that they have tried to help battered women improve their situation and their health. Our data analyses revealed that the following themes were common across specialties.

Give Validating Messages. The most common aspect of intervention was validation. Whatever their approach to helping, these physicians gave compassionate messages that validated the woman’s worth as a human being and indicated that the abuse was undeserved. One participant put it this way: “Just my being there, caring about them consistently, giving another message [helped]: You are worth caring about, you are deserving, you are valuable.” Physicians tended to embed this kind of attitude and message into their interventions with abused patients, making validation the foundation of their interactions with them.

Break Through Denial and Plant Seeds for Change. Physicians reported that within the context of a trusting relationship they tried to break through the denial these women presented about the seriousness of their experiences. Some physicians reported labeling the abuse for what it is, blatantly wrong and criminal. They believed that over time they could help victims to begin to see this reality and change their situation. One participant said:

I let them know that what’s going on is outlandishly not right, that they don’t deserve to have that happen. It’s frankly illegal, and you can bring charges against someone for doing that. Sometimes people can be shocked by finding out that that’s the case. You can plant a seed about their self-esteem … and their ability over time to change that situation, but piecemeal.

Another physician reported showing women the photographs taken of past injuries to remind them of the partner’s pattern of abusive behavior: “We begin every session with: Do you remember that? Sometimes the reaction is: No, it didn’t happen that way. But the photograph just sits there.”

Listen Nonjudgmentally. Physicians described listening and attending to the whole person as central to providing good health care to all patients, especially victims of abuse. In the context of listening they reported on the need to maintain a healing attitude by banishing criticism, blame, and judgment, but agreed that achieving this was difficult and required letting go of the desire to fix it by treating the women as competent adults. One participant said:

I try to get across just from my tone of voice primarily, that I’m not judging them. Because I made that mistake quite a while ago—my judgment was right away: Well, this is terrible; you’ve got to get out. And I could watch the person psychologically fly away from me. So in order to maintain that [trusting] space, that connection with them, it’s really important for me to get clear that I’m going to listen and not judge them. And it’s all going to change on their time.

Document, Refer, and Help the Patient Plan for Safety. These physicians stated that they were careful to write down the specifics of what the woman said. In addition to medical charting, some took photographs of any injuries with color Polaroid cameras. One physician stated that for the photographs to be useful in court “you have to include their face so that a lawyer can’t argue that you’re taking a photograph of someone else.”

The most practicable example of documentation was the development of a domestic violence packet which included a body chart, an instruction list for documentation, a compartment for the color Polaroid photograph, a handy tear-out sheet for information services (resources, shelters), and a telephone number for the police.

In general when physicians knew or suspected abuse they offered information about domestic violence and referrals to local community resources, hot lines, and shelters. Some stated that patients often refused referrals and that they kept referral sheets in the waiting room so that individuals could decide on their own whether to take one.

Other physicians stated that on occasion women had made statements such as: “I can’t take that (handout) home … it’s like a flag in front of a bull.” To address this problem some physicians reported putting business-size cards with domestic violence hot line numbers (eg, local hot line numbers, shelter numbers, or community resource numbers) in all the bathrooms, sometimes the only place where the batterer could not easily follow a woman. Participants reported:

[The cards] are constantly replenished [by housekeeping] … and one of the things I tell people if they’re in an explosive situation is to put it in their shoe, in the insole.

I have a very small practice. I do only office gynecology, but I put about 10 cards a week [in the bathroom]. I would bet 2 disappear in a week. Isn’t that amazing? And this is a fairly affluent area.

Physicians described various ways they had tried to help women plan for their own immediate and ongoing safety. Some physicians talked about trying to stay aware and sensitive to the safety needs of women whose partners are controlling them through abuse, even when the partners are not currently threatening violence. One participant said:

I’ll try to role-play with them … how are they going to deal with telling their partner that they have this infection or that they really want to use this type of birth control. I’ll say, “Some people in your situation could have a fight with their significant other,” and go through predicting some possibilities. Sometimes you can see them start to close down because they know that could happen, or this is a repeated thing that they keep getting that they have no control over. So, I’ll say, “Well, I have other patients in this situation who sometimes need a safe place to go,” or I’ll talk about what somebody else did and at the same time give them some information.

Physicians also reported counseling victims to keep a suitcase packed and have 24-hour hot line numbers or contact numbers for safe places, and helping them to specify what circumstances should cause them to call the police.

Although physicians stated that acute cases were rare outside of the emergency department, they described attempts to ensure safety when the woman’s life was in immediate danger. These attempts included (1) working as a team to separate the partner from the woman (eg, the nurse talked with the abusive partner in the waiting room while the physician cared for the victim and, with the victim’s permission, called the police and a domestic violence advocate to remove her from the abusive home); (2) making excuses to separate the abuser from the victim in the immediate situation (eg, taking the woman for tests); and (3) admitting women who could not be placed in a shelter into the hospital under a false name. One physician reported that the hospital at which she had trained had a safe bed designated for victims of rape, domestic violence, and other assaults.

Using a Team Approach. In general, the physicians agreed that it takes a team approach to intervene successfully with victims of domestic violence. Some expressed frustration about accessing community referrals and discussed the benefits to victims of readily accessible resources on site. The on-site resources referred women directly to the nurse, rape crisis counselor, social worker, behavioral medicine counselor, or psychologist, who counseled the women and conducted follow-up. Some physicians without access to onsite counselors or social workers described making domestic violence part of every staff member’s educational process when they come on board.

Physicians described how intervention demanded a certain amount of flexibility of roles, with nurses and physicians playing off each other in tag-team fashion, as necessary. One physician said:

Sometimes when I finish with an exam, I’ll tell the nurse that I suspect something, so when the nurse is giving the discharge instructions, she’ll also re-approach certain kinds of issues and give the woman another opportunity to talk about [the abuse], once she has gotten dressed and composed herself. The door is closed. It’s one-on-one.

Some physicians described how their prenatal team takes advantage of a “window of opportunity” and has helped women get out of their situations and into counseling:

We have a prenatal team that really works together … our nurse, our social worker, our nutritionist, the receptionist, everybody.… It’s a real intense time. But I think once they get out of pregnancy, we really lose that ability to make a change in their lives. It’s a real window of opportunity.

Prioritize Domestic Violence. Even the committed physicians in our study expressed conflict about taking the time to intervene once they had identified abuse. Some physicians advocated dropping the medical procedure (even if that means the loss of reimbursement) to spend the rest of the patient’s time dealing with the abuse. Physicians also described prioritizing domestic violence by conducting continuing education courses and meetings for everyone in the department about rape, domestic violence, and child and elder abuse. One participant reported:

We try to create a culture of caring about domestic violence so that nurses who think they’ve recognized someone as being a nondeclared victim won’t be told, “I’m too busy” by a physician. And so when physicians say, “I think that might be a domestic violence victim, could you go talk to her?” the nurse will see that as a priority. And if anybody asks her, “How come you haven’t got that IV started?” she or he could say, “Because I was in talking to this person trying to determine whether they were a domestic violence victim.”

Small Victories Offer Positive Feedback

These physicians reported receiving little direct feedback about the effectiveness of their interventions with battered women. Yet, they also reported a range of rewards for intervening, from seeing a patient really change her life to glimpsing shifts in the way a woman thinks about herself and the relationship. One physician said:

And the rewarding piece for me comes when at some point she looks up and notices, and you can see this change of realizing that she’s cared about and then what that must mean to her, that she’s worth something. And then later on [there are] those little steps that you can see people make when they feel like they’re worth something. That’s the most ongoing and rewarding thing.

Discussion

The themes described by the purposive sample of physicians in our study offer insight into the process of intervention with victims of domestic violence and help delineate practicable examples of how to apply interventions ( Table 3 ). The behaviors described are supported by quantitative and qualitative data from battered women.^33,37

These physicians described the foundation of intervention with victims as giving victims the message that they do not deserve abuse and that they are worth caring about. Battered women themselves report that validation is an important message. In a recent survey,³⁷ battered women rated validating statements and compassion from physicians as among the most desirable interventions, equal to safety planning and offering referrals. In another study,³³ survivors of domestic violence described how validation from a health care professional had not only provided relief and comfort, but also “started the wheels turning” toward realizing the seriousness of the situation. These women reported that validation helped them, regardless of whether they had disclosed the abuse or the health care professional had identified it.

Women who are being controlled through abuse by an intimate partner live with debilitating feelings of denial, shame, and humiliation that are sometimes reinforced in health care encounters and keep victims from seeking and receiving optimal care.^33,38-41 The physicians in our study recognized these barriers and made efforts to help women break through their denial and plant seeds for change. They also made efforts—and learned through trial and error—to listen and be nonjudgmental. Both of these behaviors were rated as highly desirable by battered women.³⁷ Physician statements made within the context of a trusting relationship can serve to remind women of the seriousness of their situation. Physician behaviors that convey respect through tone of voice and body language could lessen a victim’s shame and help her make small changes over time to improve her situation and her health.

The data we presented on documenting abuse, providing referrals, and planning for safety concur with the practices recommended by Physicians for a Violence-Free Society.⁴² We suggest that health care settings develop a domestic violence packet containing a body chart, documentation instructions, and referral sheets. We also suggest they provide Polaroid cameras to document specific injuries, since pictures offer an inviolable record of the abuse.⁴² Survivors report that the process of documenting abuse can serve to validate the individual if accompanied by genuine nonjudgmental statements of concern.³³

Although the physicians in our study were aware of the need to provide victims with referrals to community resources and assess their safety needs,^6,8,42,43 they had developed their own styles of intervention and admitted that victims sometimes refused referrals. One solution offered by participants in our study is to put easy-to-hide business-size cards with local domestic violence hot-line and shelter numbers in all of the bathrooms. We also suggest that physicians continue to offer referrals time after time: repetitive offering or availability of referrals may help survivors feel like they are not alone and may reassure them that support is available within and outside the health care system when they are ready to seek it. Physicians should remember that a woman may be able to talk about the abuse long before she can actualize any change. They should also be aware that ending the relationship does not necessarily end the abuse; it may escalate it.⁴⁴ The study physicians were careful to consider safety from the battered woman’s point of view and to take preventive measures. We suggest that physicians review their options for facilitating safety (ie, availability of resources and time) and, when necessary, connect the victim by telephone to an agency trained in assessing and planning for their safety. Battered women report that they want physicians to offer referrals and help them plan for safety.³⁷

Although current guidelines call for physicians to play a large role in identifying, intervening with, and following up on cases of partner abuse,^6,45,46 the physicians in our study emphasized the need to work as a team to identify and provide optimal care to victims. This requires flexibility of roles within the health care team and ready access to on-site and community domestic violence resources. In an attempt to improve health care for victims of domestic violence, experts and researchers in the field have proposed simplifying and limiting the tasks of physicians in this area. One model uses the acronym AVDR: physicians should ask patients about abuse; provide validating messages that battering is wrong and the patient is a worthy individual; document presenting signs, symptoms, and disclosures; and refer victims to specialists in domestic violence.⁴⁷ At that point specialists on site or on call from the community would assess the patient’s safety, make appropriate safety plans, and perform other in-depth interventions.

Physicians face ever-increasing demands on their limited time, yet these physicians committed to helping battered women found multiple ways to enable them to intervene. The holistic approaches described here—using a team approach, prioritizing domestic violence, developing a culture of caring—send a powerful message of prevention and intervention to victims: Battering is not a private, shameful issue, but a health care issue of great concern to physicians. These approaches also provide health care professionals with systematic support for helping battered women, perhaps allowing committed physicians to act as agents of change in battered women’s lives.⁵

Women who are being controlled by the abusive actions of their intimate partners report that even small signs of compassion from health care professionals have made a difference to them. As stated by physicians in this study and by survivors in our previous study,³³ these acts of caring plant the seeds for change. In their efforts to help battered women, physicians must remember that incremental changes and small moments of recognition can eventually lead to major shifts in the lives of these women. Every time physicians successfully intervene with a person whose health problems are caused by abuse they have engendered a positive outcome.

Acknowledgments

Our project has been supported by the National Institute of Mental Health Grant #1 R01 MH51580. We thank the physicians who participated in the focus groups and those who participated in reviewing the study findings. We also thank Stephanie Greer and Survey Methods Group for their assistance in recruiting physician participants and organizing the focus groups; Candace Love, PhD, and Richard Carlton, MPH, for assisting the authors with moderating focus group sessions; Priscilla Abercrombie, NP, PhD, for assisting with coding the data; Karen Herzig, PhD, for assisting with the literature review; and Jennifer Fechner for transcribing the focus group session audiotapes and proofreading the manuscript.

In response to the public health consequences of domestic violence and the number of battered women whom physicians see in their practices,^1-4 medical organizations including the American Medical Association and the American College of Obstetricians and Gynecologists have called for physicians to act as agents of change in abused women’s lives.^5,6 In the late 1980s and early 1990s these organizations and others issued guidelines and mandates based on information from domestic violence experts that outlined how physicians should intervene.^6-8 Unfortunately, these recommendations are not specific enough and do not seem to have improved their responses to violence against women^9-11; many physicians are simply not asking women about violence,^12-17 and women whose health problems result from abuse are not receiving the health care they need.^9,10,18-20

Physicians cite many barriers to intervening with victims, including patient evasiveness and failure to disclose information, lack of time and support resources, lack of education or training, fear of offending the patient, inability to “fix it,” and frustration with lack of change in the patient’s situation or the patient’s unresponsiveness to advice.^16,18,21-28 Primary care physicians in the qualitative study by Sugg and Inui²⁸ characterized talking about domestic violence with patients as opening Pandora’s box and associated the act of even asking about domestic violence with unleashing their own fears and discomforts.

Despite the barriers, some physicians are committed to addressing the underlying health problems of abused women. How do these physicians intervene, and what motivates them to continue in their commitment? In previous work,¹⁶ we described how physicians with expertise in domestic violence identify victims. With this study we explored how physicians with experience in identifying victims tried to help.

Methods

Participants

Qualitative research commonly uses purposive sampling, a method in which the participants best suited to provide a full description of the research topic are intentionally selected. We sought a sample of physicians in the San Francisco Bay Area who had experience in identifying and intervening with victims of domestic violence. To identify important common patterns that cut across different settings^29,30 physicians from 3 medical specialties were sought: primary care (family practice and general internal medicine), obstetrics and gynecology, and emergency medicine.

We conducted our recruitment in consultation with a professional survey research organization. Thirteen physicians known to have domestic violence experience, and additional physicians selected from the yellow pages, were screened and asked to participate in a study exploring the most effective ways for the health care system to meet the needs of victims of domestic violence. Eligible participants were asked to identify other colleagues who are concerned about and treated victims of domestic violence, and these individuals were screened and asked to participate and to identify others. Physicians were eligible if they worked directly with patients 20 or more hours per week, had identified and intervened with victims of domestic violence, and were somewhat confident or very confident about addressing domestic violence issues with patients.

Recruitment ceased when the goal of 12 to 22 physicians in each medical specialty who had the relevant domestic violence experience was reached.

Focus Group Method and Data Collection

In comparison with survey or one-on-one interview formats, the focus group approach allows for a more extensive exploration of the area under discussion. Participants can collectively explore different experiences and perspectives, generate ideas, and debate and compare their ideas with those of others in the group.³¹ Six focus groups ranging in size from 6 to 11 individuals were conducted during a 3-week period in January and February 1998. Each group was facilitated by 2 moderators who were members of the research team. The sessions lasted approximately 90 minutes and were held in professional focus group settings that allowed hidden viewing. Several researchers viewed the groups from behind 2-way mirrors and completed field notes that were later compared with the observations of the moderators. Before each focus group session written informed consent was obtained from all participants and a written background survey was administered to gather demographic and practice information. Participants received a small stipend for participating. Study procedures were approved by the University of California San Francisco Committee on Human Research.

We used a semistructured guide that allowed the facilitators to follow certain topics and open new lines of inquiry when appropriate.³² Open-ended questions were formulated based on our previous interviews with survivors of domestic violence³³ and a review of the literature ( Table 1 ). Audiotapes of the focus group sessions were transcribed by research staff; the principal investigator reviewed these transcripts for accuracy.

Coding and Analysis

For initial analysis we conducted multiple readings of transcripts to identify prominent themes. The investigators independently reviewed the transcripts and then met to review and discuss differences of opinion about interpretations and to further refine themes driven by the words and phrases of the participants. Through this process and the constant comparison of new data against emerging themes, a template of open codes was constructed. The transcripts were coded and specific themes within the narratives of the participants were identified in accordance with standard qualitative analytic convention.³⁴ Coded data were organized using NUD*IST 4.0 software (Qualitative Solutions and Research; Victoria, Australia). This software helps ensure the consistency of study findings and creates an audit trail.³⁵ The data were interpreted in the context of the original focus group sessions and the current literature. The final coding scheme and analysis of the findings were reviewed, and disagreements were discussed by the team until consensus was reached.

To further enhance the credibility of the findings, a qualitative technique called member check was used.³⁶ Results were directed back to 3 research participants to confirm that their experiences and those of other participants in their focus group were reflected in the findings.

Results

Of the 80 physicians who were screened, 53 were eligible, and 45 were able to attend the focus group sessions. Their characteristics are presented in Table 2 . The participants reported that they had identified an average of 28 patients per year as having been physically abused by an intimate partner, and they thought they had helped approximately 60% of those patients.

Helpful Intervention Techniques

Even these physicians reported sometimes feeling overwhelmed, frustrated, and incompetent regarding their role in domestic violence cases. They believed, however, that addressing partner abuse was part of their job and reported various ways that they have tried to help battered women improve their situation and their health. Our data analyses revealed that the following themes were common across specialties.

Give Validating Messages. The most common aspect of intervention was validation. Whatever their approach to helping, these physicians gave compassionate messages that validated the woman’s worth as a human being and indicated that the abuse was undeserved. One participant put it this way: “Just my being there, caring about them consistently, giving another message [helped]: You are worth caring about, you are deserving, you are valuable.” Physicians tended to embed this kind of attitude and message into their interventions with abused patients, making validation the foundation of their interactions with them.

Break Through Denial and Plant Seeds for Change. Physicians reported that within the context of a trusting relationship they tried to break through the denial these women presented about the seriousness of their experiences. Some physicians reported labeling the abuse for what it is, blatantly wrong and criminal. They believed that over time they could help victims to begin to see this reality and change their situation. One participant said:

I let them know that what’s going on is outlandishly not right, that they don’t deserve to have that happen. It’s frankly illegal, and you can bring charges against someone for doing that. Sometimes people can be shocked by finding out that that’s the case. You can plant a seed about their self-esteem … and their ability over time to change that situation, but piecemeal.

Another physician reported showing women the photographs taken of past injuries to remind them of the partner’s pattern of abusive behavior: “We begin every session with: Do you remember that? Sometimes the reaction is: No, it didn’t happen that way. But the photograph just sits there.”

Listen Nonjudgmentally. Physicians described listening and attending to the whole person as central to providing good health care to all patients, especially victims of abuse. In the context of listening they reported on the need to maintain a healing attitude by banishing criticism, blame, and judgment, but agreed that achieving this was difficult and required letting go of the desire to fix it by treating the women as competent adults. One participant said:

I try to get across just from my tone of voice primarily, that I’m not judging them. Because I made that mistake quite a while ago—my judgment was right away: Well, this is terrible; you’ve got to get out. And I could watch the person psychologically fly away from me. So in order to maintain that [trusting] space, that connection with them, it’s really important for me to get clear that I’m going to listen and not judge them. And it’s all going to change on their time.

Document, Refer, and Help the Patient Plan for Safety. These physicians stated that they were careful to write down the specifics of what the woman said. In addition to medical charting, some took photographs of any injuries with color Polaroid cameras. One physician stated that for the photographs to be useful in court “you have to include their face so that a lawyer can’t argue that you’re taking a photograph of someone else.”

The most practicable example of documentation was the development of a domestic violence packet which included a body chart, an instruction list for documentation, a compartment for the color Polaroid photograph, a handy tear-out sheet for information services (resources, shelters), and a telephone number for the police.

In general when physicians knew or suspected abuse they offered information about domestic violence and referrals to local community resources, hot lines, and shelters. Some stated that patients often refused referrals and that they kept referral sheets in the waiting room so that individuals could decide on their own whether to take one.

Other physicians stated that on occasion women had made statements such as: “I can’t take that (handout) home … it’s like a flag in front of a bull.” To address this problem some physicians reported putting business-size cards with domestic violence hot line numbers (eg, local hot line numbers, shelter numbers, or community resource numbers) in all the bathrooms, sometimes the only place where the batterer could not easily follow a woman. Participants reported:

[The cards] are constantly replenished [by housekeeping] … and one of the things I tell people if they’re in an explosive situation is to put it in their shoe, in the insole.

I have a very small practice. I do only office gynecology, but I put about 10 cards a week [in the bathroom]. I would bet 2 disappear in a week. Isn’t that amazing? And this is a fairly affluent area.

Physicians described various ways they had tried to help women plan for their own immediate and ongoing safety. Some physicians talked about trying to stay aware and sensitive to the safety needs of women whose partners are controlling them through abuse, even when the partners are not currently threatening violence. One participant said:

I’ll try to role-play with them … how are they going to deal with telling their partner that they have this infection or that they really want to use this type of birth control. I’ll say, “Some people in your situation could have a fight with their significant other,” and go through predicting some possibilities. Sometimes you can see them start to close down because they know that could happen, or this is a repeated thing that they keep getting that they have no control over. So, I’ll say, “Well, I have other patients in this situation who sometimes need a safe place to go,” or I’ll talk about what somebody else did and at the same time give them some information.

Physicians also reported counseling victims to keep a suitcase packed and have 24-hour hot line numbers or contact numbers for safe places, and helping them to specify what circumstances should cause them to call the police.

Although physicians stated that acute cases were rare outside of the emergency department, they described attempts to ensure safety when the woman’s life was in immediate danger. These attempts included (1) working as a team to separate the partner from the woman (eg, the nurse talked with the abusive partner in the waiting room while the physician cared for the victim and, with the victim’s permission, called the police and a domestic violence advocate to remove her from the abusive home); (2) making excuses to separate the abuser from the victim in the immediate situation (eg, taking the woman for tests); and (3) admitting women who could not be placed in a shelter into the hospital under a false name. One physician reported that the hospital at which she had trained had a safe bed designated for victims of rape, domestic violence, and other assaults.

Using a Team Approach. In general, the physicians agreed that it takes a team approach to intervene successfully with victims of domestic violence. Some expressed frustration about accessing community referrals and discussed the benefits to victims of readily accessible resources on site. The on-site resources referred women directly to the nurse, rape crisis counselor, social worker, behavioral medicine counselor, or psychologist, who counseled the women and conducted follow-up. Some physicians without access to onsite counselors or social workers described making domestic violence part of every staff member’s educational process when they come on board.

Physicians described how intervention demanded a certain amount of flexibility of roles, with nurses and physicians playing off each other in tag-team fashion, as necessary. One physician said:

Sometimes when I finish with an exam, I’ll tell the nurse that I suspect something, so when the nurse is giving the discharge instructions, she’ll also re-approach certain kinds of issues and give the woman another opportunity to talk about [the abuse], once she has gotten dressed and composed herself. The door is closed. It’s one-on-one.

Some physicians described how their prenatal team takes advantage of a “window of opportunity” and has helped women get out of their situations and into counseling:

We have a prenatal team that really works together … our nurse, our social worker, our nutritionist, the receptionist, everybody.… It’s a real intense time. But I think once they get out of pregnancy, we really lose that ability to make a change in their lives. It’s a real window of opportunity.

Prioritize Domestic Violence. Even the committed physicians in our study expressed conflict about taking the time to intervene once they had identified abuse. Some physicians advocated dropping the medical procedure (even if that means the loss of reimbursement) to spend the rest of the patient’s time dealing with the abuse. Physicians also described prioritizing domestic violence by conducting continuing education courses and meetings for everyone in the department about rape, domestic violence, and child and elder abuse. One participant reported:

We try to create a culture of caring about domestic violence so that nurses who think they’ve recognized someone as being a nondeclared victim won’t be told, “I’m too busy” by a physician. And so when physicians say, “I think that might be a domestic violence victim, could you go talk to her?” the nurse will see that as a priority. And if anybody asks her, “How come you haven’t got that IV started?” she or he could say, “Because I was in talking to this person trying to determine whether they were a domestic violence victim.”

Small Victories Offer Positive Feedback

These physicians reported receiving little direct feedback about the effectiveness of their interventions with battered women. Yet, they also reported a range of rewards for intervening, from seeing a patient really change her life to glimpsing shifts in the way a woman thinks about herself and the relationship. One physician said:

And the rewarding piece for me comes when at some point she looks up and notices, and you can see this change of realizing that she’s cared about and then what that must mean to her, that she’s worth something. And then later on [there are] those little steps that you can see people make when they feel like they’re worth something. That’s the most ongoing and rewarding thing.

Discussion

The themes described by the purposive sample of physicians in our study offer insight into the process of intervention with victims of domestic violence and help delineate practicable examples of how to apply interventions ( Table 3 ). The behaviors described are supported by quantitative and qualitative data from battered women.^33,37

These physicians described the foundation of intervention with victims as giving victims the message that they do not deserve abuse and that they are worth caring about. Battered women themselves report that validation is an important message. In a recent survey,³⁷ battered women rated validating statements and compassion from physicians as among the most desirable interventions, equal to safety planning and offering referrals. In another study,³³ survivors of domestic violence described how validation from a health care professional had not only provided relief and comfort, but also “started the wheels turning” toward realizing the seriousness of the situation. These women reported that validation helped them, regardless of whether they had disclosed the abuse or the health care professional had identified it.

Women who are being controlled through abuse by an intimate partner live with debilitating feelings of denial, shame, and humiliation that are sometimes reinforced in health care encounters and keep victims from seeking and receiving optimal care.^33,38-41 The physicians in our study recognized these barriers and made efforts to help women break through their denial and plant seeds for change. They also made efforts—and learned through trial and error—to listen and be nonjudgmental. Both of these behaviors were rated as highly desirable by battered women.³⁷ Physician statements made within the context of a trusting relationship can serve to remind women of the seriousness of their situation. Physician behaviors that convey respect through tone of voice and body language could lessen a victim’s shame and help her make small changes over time to improve her situation and her health.

The data we presented on documenting abuse, providing referrals, and planning for safety concur with the practices recommended by Physicians for a Violence-Free Society.⁴² We suggest that health care settings develop a domestic violence packet containing a body chart, documentation instructions, and referral sheets. We also suggest they provide Polaroid cameras to document specific injuries, since pictures offer an inviolable record of the abuse.⁴² Survivors report that the process of documenting abuse can serve to validate the individual if accompanied by genuine nonjudgmental statements of concern.³³

Although the physicians in our study were aware of the need to provide victims with referrals to community resources and assess their safety needs,^6,8,42,43 they had developed their own styles of intervention and admitted that victims sometimes refused referrals. One solution offered by participants in our study is to put easy-to-hide business-size cards with local domestic violence hot-line and shelter numbers in all of the bathrooms. We also suggest that physicians continue to offer referrals time after time: repetitive offering or availability of referrals may help survivors feel like they are not alone and may reassure them that support is available within and outside the health care system when they are ready to seek it. Physicians should remember that a woman may be able to talk about the abuse long before she can actualize any change. They should also be aware that ending the relationship does not necessarily end the abuse; it may escalate it.⁴⁴ The study physicians were careful to consider safety from the battered woman’s point of view and to take preventive measures. We suggest that physicians review their options for facilitating safety (ie, availability of resources and time) and, when necessary, connect the victim by telephone to an agency trained in assessing and planning for their safety. Battered women report that they want physicians to offer referrals and help them plan for safety.³⁷

Although current guidelines call for physicians to play a large role in identifying, intervening with, and following up on cases of partner abuse,^6,45,46 the physicians in our study emphasized the need to work as a team to identify and provide optimal care to victims. This requires flexibility of roles within the health care team and ready access to on-site and community domestic violence resources. In an attempt to improve health care for victims of domestic violence, experts and researchers in the field have proposed simplifying and limiting the tasks of physicians in this area. One model uses the acronym AVDR: physicians should ask patients about abuse; provide validating messages that battering is wrong and the patient is a worthy individual; document presenting signs, symptoms, and disclosures; and refer victims to specialists in domestic violence.⁴⁷ At that point specialists on site or on call from the community would assess the patient’s safety, make appropriate safety plans, and perform other in-depth interventions.

Physicians face ever-increasing demands on their limited time, yet these physicians committed to helping battered women found multiple ways to enable them to intervene. The holistic approaches described here—using a team approach, prioritizing domestic violence, developing a culture of caring—send a powerful message of prevention and intervention to victims: Battering is not a private, shameful issue, but a health care issue of great concern to physicians. These approaches also provide health care professionals with systematic support for helping battered women, perhaps allowing committed physicians to act as agents of change in battered women’s lives.⁵

Women who are being controlled by the abusive actions of their intimate partners report that even small signs of compassion from health care professionals have made a difference to them. As stated by physicians in this study and by survivors in our previous study,³³ these acts of caring plant the seeds for change. In their efforts to help battered women, physicians must remember that incremental changes and small moments of recognition can eventually lead to major shifts in the lives of these women. Every time physicians successfully intervene with a person whose health problems are caused by abuse they have engendered a positive outcome.

Acknowledgments

Our project has been supported by the National Institute of Mental Health Grant #1 R01 MH51580. We thank the physicians who participated in the focus groups and those who participated in reviewing the study findings. We also thank Stephanie Greer and Survey Methods Group for their assistance in recruiting physician participants and organizing the focus groups; Candace Love, PhD, and Richard Carlton, MPH, for assisting the authors with moderating focus group sessions; Priscilla Abercrombie, NP, PhD, for assisting with coding the data; Karen Herzig, PhD, for assisting with the literature review; and Jennifer Fechner for transcribing the focus group session audiotapes and proofreading the manuscript.

The Charity Hospital system in Louisiana provides care for a population of medically indigent patients. The legislature has authorized Louisiana State University Medical Center-New Orleans (LSUMC–NO) to form a health maintenance organization to care for Medicaid patients and those who constitute the medically indigent population. Compiling a medical data set to manage such a population is difficult. In 1997 virtually all the medical data at LSUMC was buried in paper charts that were mostly handwritten by physicians and medical students. When our group found software for a self-administered computerized medical history, we thought this might be the methodology needed to gather essential medical data from our large population of patients. We decided to test this technology by seeking an answer to the question: Can and will a patient at Charity Hospital use a computer to give a self-administered medical history?

Methods

The walk-in clinic associated with the emergency room at Charity Hospital in New Orleans was used as the site of our study. It is open 15 hours daily during the week and 10 hours daily on weekends, and approximately 5500 patients visit this clinic each month. Patients were sampled during January, February, and March 1998 by calling the patient’s name whose record was at the bottom of the stack of charts of patients to be seen. These patients were invited to participate, and demographic data were taken and a reason for refusal was noted when possible for those who refused.

The computer system used was an IBM-compatible 486 PC with a color monitor and standard keyboard. The interview software was Instant Medical History (Primetime Software, Columbia, SC). The participants completed the brief version of the history algorithm to expedite the interview process, since our goal was not to test the software but to determine the patient’s ability to use and opinion of using a computer to give the information.

In the testing area the patient was given a consent form and a questionnaire. The patient completed the consent form and a set of items on the questionnaire before using the computer. The software used gave the patient instructions on how to input answers; the investigator typed in only the patient’s age and sex. When asked by the software to provide a reason for the visit, the patient was encouraged to select “New problem or illness” or “Follow up of previous illness.” Then the patient was offered more specific choices, such as: “Bone or muscle problem,” “Cough, cold,” and “Diabetes.” The investigators gave support for the patient’s use of the computer until selection of a system for the chief complaint was made. At that point the patients generally understood what to do to indicate the desired answer. Finding the letters on the keyboard appeared to be more of a problem than understanding the software. Interaction between the patient and the investigator was by design (none to minimal, while the patient was using the computer). After completing the history the patient answered another set of items on the questionnaire, was given the printout from the computer to share with his or her physician, and was returned to the waiting area. The general atmosphere was purposely cordial for participants and nonparticipants.

Measures

We obtained demographic data for all patients approached for our study. Refusal rates were one of the main outcomes of the study. When a patient declined to participate, a reason for refusal was noted if possible. Although literacy level was not assessed directly, illiteracy was noted when the information was volunteered.

Another major outcome was the acceptability rating of using the computer to obtain a medical history. Acceptability was determined using a questionnaire consisting of 12 items. Each item was rated on a 4-point Likert scale with higher scores indicating more acceptability.

Statistical Analysis

We analyzed the data with Microsoft Access (Microsoft Corporation, Redmond, Wash), Microsoft Excel, and the Statistical Package for the Social Sciences, version 7 (SPSS Inc, Chicago, Ill). Continuous variables (eg, age, years of education) were described with means and standard errors of the means (SEM). We analyzed the internal consistency of the acceptability questionnaire using Cronbach a and item-total correlations. The effects of categorical variables (eg, sex, race) on the continuous acceptability rating were analyzed using analysis of variance, and the relations among continuous variables were analyzed with Pearson correlation coefficients (r). Significance for all analyses was set at P <.05.

Results

The sample demographics are presented in the Table. Consistent with the clinic population, the sample was approximately 55% women. African Americans constituted 79%. Approximately a third had not completed high school, another third completed high school, and a little less than a third had continued their education beyond high school. The majority (70%) had been to this clinic previously. Ninety-nine percent spoke English as their primary language. More than half of the participants reported that they had not previously used a personal computer. Thirteen patients (13%) refused to participate. The reasons for refusal varied: 4 patients were unable to read well enough to do the study, 1 refused because of dislike for computers, and 3 felt too sick. The other 5 patients gave no reason for refusal.

The acceptability rating items were highly interrelated. Item-total correlations ranged from 0.11 to 0.58, yielding a Cronbach a of 0.75. This represents adequate internal validity and suggests that a single total score (the acceptability rating) is an adequate summary of the items. We averaged the items to obtain acceptability ratings that ranged from 2.0 to 4.0, with a mean of 3.27 (SEM=0.044). Since the maximum possible score was 4.0, this mean corresponds to very positive ratings. Only 4 participants rated the experience below 2.5. Combined with the 13 who refused to participate, only 17% of the subjects rejected or were only slightly positive about the experience.

We examined the effects of demographics and selected patient experiences on the acceptability rating. Sex and race did not have a significant influence. Age had a statistically significant impact (r=-0.27), with increasing age associated with decreasing acceptability. Education (r=0.02) and previous visits to the clinic had no impact on the acceptability ratings. Patients who had used computers before rated the experience slightly more positively (mean=3.8) than patients who reported no previous use (mean=3.6). However, the 0.2 difference may not represent a clinically meaningful increase in acceptance of completing a computerized medical history.

Conclusions

We feel that the positive reactions of 83% of all patients approached about the study is a very strong statement for supporting further examination of software programs for computerized medical interviewing. We have shown that a representative sample of the Charity Hospital patient population was capable of using a computer for a self-administered medical history, and our subjects thought it a good idea. Further study of this clinical tool should not be halted because of fear of patient resistance or refusal. We encourage our colleagues to consider the use of a computerized self-administered patient history as a timesaving cost-effective adjunct to the traditional oral history.

Acknowledgments

Colonel Gordon Black, MHA, Department of Preventive Medicine and Public Health, LSUMC-NO, provided the vision that all our patients can use a computer and demanded that we prove it. His death in November 1997 prevented him from contributing to the writing of this article.

The Charity Hospital system in Louisiana provides care for a population of medically indigent patients. The legislature has authorized Louisiana State University Medical Center-New Orleans (LSUMC–NO) to form a health maintenance organization to care for Medicaid patients and those who constitute the medically indigent population. Compiling a medical data set to manage such a population is difficult. In 1997 virtually all the medical data at LSUMC was buried in paper charts that were mostly handwritten by physicians and medical students. When our group found software for a self-administered computerized medical history, we thought this might be the methodology needed to gather essential medical data from our large population of patients. We decided to test this technology by seeking an answer to the question: Can and will a patient at Charity Hospital use a computer to give a self-administered medical history?

Methods

The walk-in clinic associated with the emergency room at Charity Hospital in New Orleans was used as the site of our study. It is open 15 hours daily during the week and 10 hours daily on weekends, and approximately 5500 patients visit this clinic each month. Patients were sampled during January, February, and March 1998 by calling the patient’s name whose record was at the bottom of the stack of charts of patients to be seen. These patients were invited to participate, and demographic data were taken and a reason for refusal was noted when possible for those who refused.

The computer system used was an IBM-compatible 486 PC with a color monitor and standard keyboard. The interview software was Instant Medical History (Primetime Software, Columbia, SC). The participants completed the brief version of the history algorithm to expedite the interview process, since our goal was not to test the software but to determine the patient’s ability to use and opinion of using a computer to give the information.

In the testing area the patient was given a consent form and a questionnaire. The patient completed the consent form and a set of items on the questionnaire before using the computer. The software used gave the patient instructions on how to input answers; the investigator typed in only the patient’s age and sex. When asked by the software to provide a reason for the visit, the patient was encouraged to select “New problem or illness” or “Follow up of previous illness.” Then the patient was offered more specific choices, such as: “Bone or muscle problem,” “Cough, cold,” and “Diabetes.” The investigators gave support for the patient’s use of the computer until selection of a system for the chief complaint was made. At that point the patients generally understood what to do to indicate the desired answer. Finding the letters on the keyboard appeared to be more of a problem than understanding the software. Interaction between the patient and the investigator was by design (none to minimal, while the patient was using the computer). After completing the history the patient answered another set of items on the questionnaire, was given the printout from the computer to share with his or her physician, and was returned to the waiting area. The general atmosphere was purposely cordial for participants and nonparticipants.

Measures

We obtained demographic data for all patients approached for our study. Refusal rates were one of the main outcomes of the study. When a patient declined to participate, a reason for refusal was noted if possible. Although literacy level was not assessed directly, illiteracy was noted when the information was volunteered.

Another major outcome was the acceptability rating of using the computer to obtain a medical history. Acceptability was determined using a questionnaire consisting of 12 items. Each item was rated on a 4-point Likert scale with higher scores indicating more acceptability.

Statistical Analysis

We analyzed the data with Microsoft Access (Microsoft Corporation, Redmond, Wash), Microsoft Excel, and the Statistical Package for the Social Sciences, version 7 (SPSS Inc, Chicago, Ill). Continuous variables (eg, age, years of education) were described with means and standard errors of the means (SEM). We analyzed the internal consistency of the acceptability questionnaire using Cronbach a and item-total correlations. The effects of categorical variables (eg, sex, race) on the continuous acceptability rating were analyzed using analysis of variance, and the relations among continuous variables were analyzed with Pearson correlation coefficients (r). Significance for all analyses was set at P <.05.

Results

The sample demographics are presented in the Table. Consistent with the clinic population, the sample was approximately 55% women. African Americans constituted 79%. Approximately a third had not completed high school, another third completed high school, and a little less than a third had continued their education beyond high school. The majority (70%) had been to this clinic previously. Ninety-nine percent spoke English as their primary language. More than half of the participants reported that they had not previously used a personal computer. Thirteen patients (13%) refused to participate. The reasons for refusal varied: 4 patients were unable to read well enough to do the study, 1 refused because of dislike for computers, and 3 felt too sick. The other 5 patients gave no reason for refusal.

The acceptability rating items were highly interrelated. Item-total correlations ranged from 0.11 to 0.58, yielding a Cronbach a of 0.75. This represents adequate internal validity and suggests that a single total score (the acceptability rating) is an adequate summary of the items. We averaged the items to obtain acceptability ratings that ranged from 2.0 to 4.0, with a mean of 3.27 (SEM=0.044). Since the maximum possible score was 4.0, this mean corresponds to very positive ratings. Only 4 participants rated the experience below 2.5. Combined with the 13 who refused to participate, only 17% of the subjects rejected or were only slightly positive about the experience.

We examined the effects of demographics and selected patient experiences on the acceptability rating. Sex and race did not have a significant influence. Age had a statistically significant impact (r=-0.27), with increasing age associated with decreasing acceptability. Education (r=0.02) and previous visits to the clinic had no impact on the acceptability ratings. Patients who had used computers before rated the experience slightly more positively (mean=3.8) than patients who reported no previous use (mean=3.6). However, the 0.2 difference may not represent a clinically meaningful increase in acceptance of completing a computerized medical history.

Conclusions

We feel that the positive reactions of 83% of all patients approached about the study is a very strong statement for supporting further examination of software programs for computerized medical interviewing. We have shown that a representative sample of the Charity Hospital patient population was capable of using a computer for a self-administered medical history, and our subjects thought it a good idea. Further study of this clinical tool should not be halted because of fear of patient resistance or refusal. We encourage our colleagues to consider the use of a computerized self-administered patient history as a timesaving cost-effective adjunct to the traditional oral history.

Acknowledgments

Colonel Gordon Black, MHA, Department of Preventive Medicine and Public Health, LSUMC-NO, provided the vision that all our patients can use a computer and demanded that we prove it. His death in November 1997 prevented him from contributing to the writing of this article.

The Charity Hospital system in Louisiana provides care for a population of medically indigent patients. The legislature has authorized Louisiana State University Medical Center-New Orleans (LSUMC–NO) to form a health maintenance organization to care for Medicaid patients and those who constitute the medically indigent population. Compiling a medical data set to manage such a population is difficult. In 1997 virtually all the medical data at LSUMC was buried in paper charts that were mostly handwritten by physicians and medical students. When our group found software for a self-administered computerized medical history, we thought this might be the methodology needed to gather essential medical data from our large population of patients. We decided to test this technology by seeking an answer to the question: Can and will a patient at Charity Hospital use a computer to give a self-administered medical history?

Methods

The walk-in clinic associated with the emergency room at Charity Hospital in New Orleans was used as the site of our study. It is open 15 hours daily during the week and 10 hours daily on weekends, and approximately 5500 patients visit this clinic each month. Patients were sampled during January, February, and March 1998 by calling the patient’s name whose record was at the bottom of the stack of charts of patients to be seen. These patients were invited to participate, and demographic data were taken and a reason for refusal was noted when possible for those who refused.

The computer system used was an IBM-compatible 486 PC with a color monitor and standard keyboard. The interview software was Instant Medical History (Primetime Software, Columbia, SC). The participants completed the brief version of the history algorithm to expedite the interview process, since our goal was not to test the software but to determine the patient’s ability to use and opinion of using a computer to give the information.

In the testing area the patient was given a consent form and a questionnaire. The patient completed the consent form and a set of items on the questionnaire before using the computer. The software used gave the patient instructions on how to input answers; the investigator typed in only the patient’s age and sex. When asked by the software to provide a reason for the visit, the patient was encouraged to select “New problem or illness” or “Follow up of previous illness.” Then the patient was offered more specific choices, such as: “Bone or muscle problem,” “Cough, cold,” and “Diabetes.” The investigators gave support for the patient’s use of the computer until selection of a system for the chief complaint was made. At that point the patients generally understood what to do to indicate the desired answer. Finding the letters on the keyboard appeared to be more of a problem than understanding the software. Interaction between the patient and the investigator was by design (none to minimal, while the patient was using the computer). After completing the history the patient answered another set of items on the questionnaire, was given the printout from the computer to share with his or her physician, and was returned to the waiting area. The general atmosphere was purposely cordial for participants and nonparticipants.

Measures

We obtained demographic data for all patients approached for our study. Refusal rates were one of the main outcomes of the study. When a patient declined to participate, a reason for refusal was noted if possible. Although literacy level was not assessed directly, illiteracy was noted when the information was volunteered.

Another major outcome was the acceptability rating of using the computer to obtain a medical history. Acceptability was determined using a questionnaire consisting of 12 items. Each item was rated on a 4-point Likert scale with higher scores indicating more acceptability.

Statistical Analysis

We analyzed the data with Microsoft Access (Microsoft Corporation, Redmond, Wash), Microsoft Excel, and the Statistical Package for the Social Sciences, version 7 (SPSS Inc, Chicago, Ill). Continuous variables (eg, age, years of education) were described with means and standard errors of the means (SEM). We analyzed the internal consistency of the acceptability questionnaire using Cronbach a and item-total correlations. The effects of categorical variables (eg, sex, race) on the continuous acceptability rating were analyzed using analysis of variance, and the relations among continuous variables were analyzed with Pearson correlation coefficients (r). Significance for all analyses was set at P <.05.

Results

The sample demographics are presented in the Table. Consistent with the clinic population, the sample was approximately 55% women. African Americans constituted 79%. Approximately a third had not completed high school, another third completed high school, and a little less than a third had continued their education beyond high school. The majority (70%) had been to this clinic previously. Ninety-nine percent spoke English as their primary language. More than half of the participants reported that they had not previously used a personal computer. Thirteen patients (13%) refused to participate. The reasons for refusal varied: 4 patients were unable to read well enough to do the study, 1 refused because of dislike for computers, and 3 felt too sick. The other 5 patients gave no reason for refusal.

The acceptability rating items were highly interrelated. Item-total correlations ranged from 0.11 to 0.58, yielding a Cronbach a of 0.75. This represents adequate internal validity and suggests that a single total score (the acceptability rating) is an adequate summary of the items. We averaged the items to obtain acceptability ratings that ranged from 2.0 to 4.0, with a mean of 3.27 (SEM=0.044). Since the maximum possible score was 4.0, this mean corresponds to very positive ratings. Only 4 participants rated the experience below 2.5. Combined with the 13 who refused to participate, only 17% of the subjects rejected or were only slightly positive about the experience.

We examined the effects of demographics and selected patient experiences on the acceptability rating. Sex and race did not have a significant influence. Age had a statistically significant impact (r=-0.27), with increasing age associated with decreasing acceptability. Education (r=0.02) and previous visits to the clinic had no impact on the acceptability ratings. Patients who had used computers before rated the experience slightly more positively (mean=3.8) than patients who reported no previous use (mean=3.6). However, the 0.2 difference may not represent a clinically meaningful increase in acceptance of completing a computerized medical history.

Conclusions

We feel that the positive reactions of 83% of all patients approached about the study is a very strong statement for supporting further examination of software programs for computerized medical interviewing. We have shown that a representative sample of the Charity Hospital patient population was capable of using a computer for a self-administered medical history, and our subjects thought it a good idea. Further study of this clinical tool should not be halted because of fear of patient resistance or refusal. We encourage our colleagues to consider the use of a computerized self-administered patient history as a timesaving cost-effective adjunct to the traditional oral history.

Acknowledgments

Colonel Gordon Black, MHA, Department of Preventive Medicine and Public Health, LSUMC-NO, provided the vision that all our patients can use a computer and demanded that we prove it. His death in November 1997 prevented him from contributing to the writing of this article.

There are many highly effective contraceptive methods available. Some, including oral birth control pills (OCPs), injectable and implantable hormones, and sterilization of both sexes, have ideal effectiveness rates higher than 98% for preventing pregnancy.

However, contraception is not always used ideally. Unplanned or unintended pregnancies do occur. In 1988, US women aged 15 to 44 years reported that 35% of their full-term pregnancies in the preceding 5 years were unintended,¹ in some populations 60% of pregnancies were reported as unintended,² and one third of these ended in abortion.³ Sixty-five percent of adolescent pregnancies are also unintended.⁴ And more than 1 million pregnancies annually are reported to have occurred from misuse of OCPs.³

Pregnancies that occur while the woman is using contraception are considered unintended. These types of pregnancies have poorer outcomes when carried to term than other pregnancies, including an increased incidence of premature birth and intrauterine growth retardation.⁵ This is an area of concern for the women, their partners, and the health care providers who help these women with contraception concerns.

The way women use and experience contraception profoundly affects its effectiveness.^2,6 OCPs have been available for more than 30 years, implantable contraceptives for approximately 8 years, and injectable contraceptives for 5 to 7 years. Today, there are women who have had greater than 95% effective hormonal contraception available all their childbearing years. The purpose of our study was to discover what patterns of contraceptive use women developed during their childbearing years and how these patterns were related to unintended pregnancies.

Methods

We obtained a convenience sampling of women who visited an urban/suburban family practice residency office that is run by an open-panel health maintenance organization (HMO), but accepts more than 60 different health plans (including Medicare, HMO Medicare, Medicaid, HMO Medicaid, and self-pay) and includes maternity care. The practice has more than 25,000 patient visits annually (60% women) who are residents of eastern Baltimore, Maryland, and Baltimore County. Women patients and women relatives or friends of patients who entered the waiting room and were aged 18 to 50 years were asked to participate in a face-to-face half-hour interview by a medical student. Two medical students talked with 396 women during June and July 1999. The interviews were recorded by identification number only.

After giving informed consent, the women spent approximately 30 minutes answering questions about marital status, education, socioeconomic status, drug and cigarette use, and first and subsequent contraceptive experiences. Each woman was given a list of 20 methods of contraception and 20 reasons for discontinuation to help her; both lists had an “other” category. She was asked what contraceptive she remembered she had used first, for how long, and when she had changed methods and why; these questions were repeated for every method used since. The questionnaire had been pretested for 2 years in 2 previous studies of more than 600 women. Only 5 women refused to participate in our survey, most often because of time constraints.

One author (J.A.R.) entered the data in Excel (Microsoft Corporation; Redmond, Wash) spreadsheets. The statisticians at the MedAtlantic Research Institute converted the spreadsheets with the Statistical Package for the Social Sciences (SPSS, Inc; Chicago, Ill). Variables were analyzed for correlation and significance by Student t and chi-square tests. The patterns of contraceptive use were analyzed, and demographic values were compared.

Results

A total of 396 women participated in our survey. They were representative of urban/suburban women in general and of women of the practice [Table 1]. The average age of the respondents was 27 years; 98% (n=389) had sex with a man.

During their lifetimes these women used an aggregate total of 1421 methods of contraception (average=3.52±1.56 methods per woman; range=1-10). The methods used are listed in [Table 2]. All of the women (except the 5 who never had sex with a man) used 1 method of contraception at least once, 370 (93%) had used 2 methods, 287 (72%) had used 3, and 217 (55%) had used 4 or more. Eighty-one women (20%) reported having used more than 1 method at a time. Condoms were the most common first method of contraception (62%). The most common contraceptive methods used overall were OCPs (81%) and condoms (78%). Thirty-four percent had tried implantable or injectable hormonal contraception at least once. And 67 women (17%) had a tubal ligation (average age=28 years; range=21-45 years).

A total of 168 (42%) of the women became pregnant while using contraception, some more than once. Forty-nine women (13%) became pregnant twice and 13 women (3%) 3 times while using contraception. One hundred three women became pregnant while taking OCPs, 78 while using condoms and 11 while using an injectable contraceptive (depot medroxyprogesterone [DMP]). The women who became pregnant while using birth control were significantly more likely to be younger and African American. They were more likely to have a history of early initiation of birth control use, and they used more methods during their lifetimes. These women were pregnant more often and were more likely to use some type of public assistance [Table 3]. The total group of African American women in our survey was more likely to be younger, have some college education, and be single. Marital status did not correlate with becoming pregnant while using birth control; married, single, and separated or divorced women were equally represented. Women who became pregnant while using birth control were not more likely to use cigarettes, marijuana, or cocaine; have more partners; or start sexual relationships at an earlier age.

The rate of actual-use effectiveness of OCPs in this population was similar to the national average. First-year use effectiveness rates could not be determined, but lifetime effectiveness rates were estimated. Three hundred nineteen women (81%) had used OCPs for an aggregate total of 1422 years (average=4.5 years per woman; range=1 month-28 years). One hundred three women (33.3%) reported that they had become pregnant while using OCPs, some more than once. This was a pregnancy rate of 7.5%. Fifty-seven percent of the women who became pregnant while taking OCPs said they stopped using them because of getting pregnant.

Three hundred four women (78%) had used condoms for an aggregate total of 1178 years (average=3.9 years per woman; range=1 month-25 years). Seventy-eight women (25.6%) reported becoming pregnant while using condoms for a pregnancy rate of 6.6%. Fifty-five percent of those women stated that becoming pregnant was the reason for stopping condom use.

Eight-two women had used DMP for an aggregate total of 77.1 years with an average of 11 months per woman. Eleven women reported that they became pregnant while using DMP, for a pregnancy rate in our study of 14%.

Two major patterns of contraceptive use during a woman’s lifetime emerged. These 2 patterns described the contraceptive choices or directions of 82% of the women. One group of 210 women (53%), who will be called the “effective contraceptors,” started with condoms and then used OCPs or DMP. Following that change, they either continued to take OCPs or changed again to a method with a higher actual effectiveness rate (ie, DMP, Norplant [Wyeth-Ayerst; St. Davids, Penn] intrauterine device, tubal ligation, vasectomy, or hysterectomy).

The other group (the “less effective contraceptors,” n=110, 29%) also started with condoms. Forty-seven of these women changed immediately to a method less effective overall than OCPs (rhythm, withdrawal, gel/foam, diaphragm, or no method); 25 changed to use OCPs, and then began to use methods less effective than OCPs. Also included in this group were 38 women who began contraception by taking OCPs and then changed to less effective methods.

The effective contraceptors were significantly less likely to become pregnant while using birth control than the less effective contraceptors. Only 37% of the effective contraceptors became pregnant; 51% of the less effective contraceptors did so (odds ratio= 1.4; [Table 4]).

Discussion

Although there has been much research into the effectiveness, side effects, and reasons for discontinuation of individual birth control methods, the personal histories of how women have used contraception has seldom been examined. We attempted to document patterns of contraception use and to relate these patterns to unintended pregnancies.

Two patterns of lifetime contraceptive behaviors in women emerged in our study. Approximately half of the women in our study showed a pattern of changing their birth control methods to more effective ones (the effective contraceptors), and approximately one fourth chose a pattern of methods that became increasingly less effective (the less effective contraceptors). Thus, it may be possible for a health care professional to be able to place a patient in a low- or high-risk group for unintended pregnancy by asking a few questions about her contraception history. It may not ever be possible to completely determine prospectively who is at risk for unintended pregnancy since even highly effective methods have inherent pregnancy rates over time. Physicians, however, should start considering a woman’s history of contraceptive methods as a primary tool for helping to prevent unintended pregnancy.

Ideal Versus Actual Effectiveness

Effectiveness of the ideal use of contraceptive methods is determined by the number of pregnancies per 100 women using the method for 1 year. OCPs (98.5%), DMP (99.7%), Norplant (99.7%), tubal ligations (99.8%), vasectomy (99%), and hysterectomy (100%) all have high ideal-use effectiveness rates.^3,7 These percentages make pregnancy while using any of these methods appear extremely unlikely. Condoms have an ideal effectiveness rate of 85% to 90%, still fairly high numbers.

However, contraception is rarely used ideally. Women forget to take a pill, forget to come in for a DMP shot, use condoms improperly, or become worried about side effects and suddenly stop using their chosen method. These mistakes are taken this into account by recording actual-use effectiveness rates. Still, OCPs have an actual effectiveness rate of 94% and condoms of 82% to 85%, and the actual effectiveness rates of tubal ligation and Norplant do not decrease from their ideal rates.³

Most women have used many forms of contraception, and in our group many changed several times during their lives. The average number of methods was more than 3, and more than half of the women used 5 or more methods. Many women had used both condoms and OCPs during their lives. The reasons for these changes should be examined more carefully (and possibly prospectively). Changing to a more effective method may indicate a lessened desire to ever become pregnant. Two studies have shown that women who want to postpone pregnancy are more likely to get pregnant than those who definitely do not want to give birth.^8,9 Women who are dissatisfied with a contraceptive method are more likely to change that method and more likely to have an unintended pregnancy.² Women who use the same contraceptive method for a long period are known to use them more efficiently.¹⁰ Similarly, women in our study who used several methods were more likely to have become pregnant while using birth control; older women were less likely to become pregnant.

Many women became pregnant while using methods that have good to excellent effectiveness rates. Although some admitted missing a pill or incorrectly using a condom, these women still felt they became pregnant while using a contraceptive method. Women taking OCPs had an actual effectiveness rate of 92.5%, close to the national average of 94%.3 Women using condoms had an effectiveness rate of 93.4%, much higher than the national average of 85%.3 This latter may be explained by the more recent trend of using condoms in addition to another method (condoms to prevent sexually transmitted diseases and another method for birth control).

Forty-two percent of the women in our study became pregnant while using contraception. One reason for this apparent paradox of high effectiveness rates and a high number of unintended pregnancies is the lifetime use of contraception. Effectiveness rates are calculated as the use of a method by 100 women for 1 year. These women all used contraception for more than 1 year. It is the natural history of OCPs with a 95% effectiveness rate that 1 in 20 women will get pregnant in 1 year, and 50 pregnancies will occur in 10 years. Usual actual use of high-effectiveness contraception still carries a significant risk of pregnancy that can be seen by the results in this population of women.

Becoming pregnant even while using good methods of birth control must be expected, explained, and understood. Physicians and their patients must not consider these pregnancies to be failures of the method used, but as inherent part of a life history of contraceptive use. Unintended pregnancies may be a consequence of using even very effective contraceptive methods.

Changing Methods

In our study, becoming pregnant while using a contraceptive method was very likely to cause the woman to discontinue using that method. More than half of the women who became pregnant while using contraception stated pregnancy was their primary reason for changing methods.

In previous studies, a physician’s involvement did not affect a woman’s use or satisfaction with contraception.¹¹ In our study most women had discussed their satisfaction with contraceptive methods with a physician. Discussion or lack of discussion did not affect methods chosen, number of methods chosen, or the chance of becoming pregnant while using birth control.

Other studies have suggested that women may choose highly effective contraception, especially irreversible contraception, because of fear of pregnancy, then have a more satisfying sexual life because this fear has been reduced.^12,13 However, in our study population, no form of contraception significantly affected a woman’s satisfaction with her sexual life.

There was a very low effectiveness rate in our population of women who used DMP. DMP is usually a very effective method that boasts ideal and actual failure rates of less than 3 in 1000 women-years (99.7%). However, 11 of the 82 women in our study reported that they became pregnant while using DMP, for an effectiveness rate of only 86%. The average duration of use of DMP in these women was only 11 months (range=1 week to 60 months). It cannot be determined from the data, but because the average length of use is so short many of these pregnancies may have occurred in women who received only 1 shot and never returned. They considered themselves users of DMP, even though the medication’s effectiveness had waned. Women may have also wanted to please the interviewer or give an answer they thought was appropriate. This may be a major bias of our method of obtaining data. It is perhaps more socially acceptable to claim to be using DMP than to admit to using no method at all, or to claim to be using condoms consistently when actually only using them occasionally.

Limitations

Our study has inherent difficulties. It was a convenience sample; women who were interviewed might have been visiting the physician to obtain a prescription method of contraception, while those who used over-the-counter or rhythm methods may not have been counted proportionally. There is also an inherent recall bias. Women may be more likely to remember a significant fact (such as a pregnancy) as the reason for changing a birth control method rather than the headaches or irregular bleeding that may have contributed to the change. The women may have been more likely to tell the interviewer they were using a birth control method that failed than to say they stopped or forgot to use their method. This would make the methods look less effective.

However, reasons for changing or using contraception are based on the women’s perceptions, so although recall bias may occur, the women’s perceptions are as important as the actual happenings. Whether a woman became pregnant while using a particular method was defined by her recall of the situation. No objective measurements (counting pills, checking charts for DMP shots) were performed. The woman’s perceptions were important for our study because they affected her subsequent use of contraceptive methods. This induced a bias, however, from the interpretation by the women. Another flaw of our study is that it was difficult to determine when a woman was using more than one method concurrently; this would give a higher effectiveness rate than either method individually and overall.

A prospective concurrent study of women’s use of contraception over time would give better answers about why women change contraception and how they use it.

Conclusions

Pregnancy must be considered a possible risk even for those women using an effective method of contraception over a lifetime. OCPs had a lifetime risk of one third for pregnancy in our study population. However, by determining a woman’s pattern of contraceptive use, the health care professional may be able to pinpoint some women who are at higher risk for unintended pregnancies. These women should be followed up more closely and urged to use more effective contraception.

There are many highly effective contraceptive methods available. Some, including oral birth control pills (OCPs), injectable and implantable hormones, and sterilization of both sexes, have ideal effectiveness rates higher than 98% for preventing pregnancy.

However, contraception is not always used ideally. Unplanned or unintended pregnancies do occur. In 1988, US women aged 15 to 44 years reported that 35% of their full-term pregnancies in the preceding 5 years were unintended,¹ in some populations 60% of pregnancies were reported as unintended,² and one third of these ended in abortion.³ Sixty-five percent of adolescent pregnancies are also unintended.⁴ And more than 1 million pregnancies annually are reported to have occurred from misuse of OCPs.³

Pregnancies that occur while the woman is using contraception are considered unintended. These types of pregnancies have poorer outcomes when carried to term than other pregnancies, including an increased incidence of premature birth and intrauterine growth retardation.⁵ This is an area of concern for the women, their partners, and the health care providers who help these women with contraception concerns.

The way women use and experience contraception profoundly affects its effectiveness.^2,6 OCPs have been available for more than 30 years, implantable contraceptives for approximately 8 years, and injectable contraceptives for 5 to 7 years. Today, there are women who have had greater than 95% effective hormonal contraception available all their childbearing years. The purpose of our study was to discover what patterns of contraceptive use women developed during their childbearing years and how these patterns were related to unintended pregnancies.

Methods

We obtained a convenience sampling of women who visited an urban/suburban family practice residency office that is run by an open-panel health maintenance organization (HMO), but accepts more than 60 different health plans (including Medicare, HMO Medicare, Medicaid, HMO Medicaid, and self-pay) and includes maternity care. The practice has more than 25,000 patient visits annually (60% women) who are residents of eastern Baltimore, Maryland, and Baltimore County. Women patients and women relatives or friends of patients who entered the waiting room and were aged 18 to 50 years were asked to participate in a face-to-face half-hour interview by a medical student. Two medical students talked with 396 women during June and July 1999. The interviews were recorded by identification number only.

After giving informed consent, the women spent approximately 30 minutes answering questions about marital status, education, socioeconomic status, drug and cigarette use, and first and subsequent contraceptive experiences. Each woman was given a list of 20 methods of contraception and 20 reasons for discontinuation to help her; both lists had an “other” category. She was asked what contraceptive she remembered she had used first, for how long, and when she had changed methods and why; these questions were repeated for every method used since. The questionnaire had been pretested for 2 years in 2 previous studies of more than 600 women. Only 5 women refused to participate in our survey, most often because of time constraints.

One author (J.A.R.) entered the data in Excel (Microsoft Corporation; Redmond, Wash) spreadsheets. The statisticians at the MedAtlantic Research Institute converted the spreadsheets with the Statistical Package for the Social Sciences (SPSS, Inc; Chicago, Ill). Variables were analyzed for correlation and significance by Student t and chi-square tests. The patterns of contraceptive use were analyzed, and demographic values were compared.

Results

A total of 396 women participated in our survey. They were representative of urban/suburban women in general and of women of the practice [Table 1]. The average age of the respondents was 27 years; 98% (n=389) had sex with a man.

During their lifetimes these women used an aggregate total of 1421 methods of contraception (average=3.52±1.56 methods per woman; range=1-10). The methods used are listed in [Table 2]. All of the women (except the 5 who never had sex with a man) used 1 method of contraception at least once, 370 (93%) had used 2 methods, 287 (72%) had used 3, and 217 (55%) had used 4 or more. Eighty-one women (20%) reported having used more than 1 method at a time. Condoms were the most common first method of contraception (62%). The most common contraceptive methods used overall were OCPs (81%) and condoms (78%). Thirty-four percent had tried implantable or injectable hormonal contraception at least once. And 67 women (17%) had a tubal ligation (average age=28 years; range=21-45 years).

A total of 168 (42%) of the women became pregnant while using contraception, some more than once. Forty-nine women (13%) became pregnant twice and 13 women (3%) 3 times while using contraception. One hundred three women became pregnant while taking OCPs, 78 while using condoms and 11 while using an injectable contraceptive (depot medroxyprogesterone [DMP]). The women who became pregnant while using birth control were significantly more likely to be younger and African American. They were more likely to have a history of early initiation of birth control use, and they used more methods during their lifetimes. These women were pregnant more often and were more likely to use some type of public assistance [Table 3]. The total group of African American women in our survey was more likely to be younger, have some college education, and be single. Marital status did not correlate with becoming pregnant while using birth control; married, single, and separated or divorced women were equally represented. Women who became pregnant while using birth control were not more likely to use cigarettes, marijuana, or cocaine; have more partners; or start sexual relationships at an earlier age.

The rate of actual-use effectiveness of OCPs in this population was similar to the national average. First-year use effectiveness rates could not be determined, but lifetime effectiveness rates were estimated. Three hundred nineteen women (81%) had used OCPs for an aggregate total of 1422 years (average=4.5 years per woman; range=1 month-28 years). One hundred three women (33.3%) reported that they had become pregnant while using OCPs, some more than once. This was a pregnancy rate of 7.5%. Fifty-seven percent of the women who became pregnant while taking OCPs said they stopped using them because of getting pregnant.

Three hundred four women (78%) had used condoms for an aggregate total of 1178 years (average=3.9 years per woman; range=1 month-25 years). Seventy-eight women (25.6%) reported becoming pregnant while using condoms for a pregnancy rate of 6.6%. Fifty-five percent of those women stated that becoming pregnant was the reason for stopping condom use.

Eight-two women had used DMP for an aggregate total of 77.1 years with an average of 11 months per woman. Eleven women reported that they became pregnant while using DMP, for a pregnancy rate in our study of 14%.

Two major patterns of contraceptive use during a woman’s lifetime emerged. These 2 patterns described the contraceptive choices or directions of 82% of the women. One group of 210 women (53%), who will be called the “effective contraceptors,” started with condoms and then used OCPs or DMP. Following that change, they either continued to take OCPs or changed again to a method with a higher actual effectiveness rate (ie, DMP, Norplant [Wyeth-Ayerst; St. Davids, Penn] intrauterine device, tubal ligation, vasectomy, or hysterectomy).

The other group (the “less effective contraceptors,” n=110, 29%) also started with condoms. Forty-seven of these women changed immediately to a method less effective overall than OCPs (rhythm, withdrawal, gel/foam, diaphragm, or no method); 25 changed to use OCPs, and then began to use methods less effective than OCPs. Also included in this group were 38 women who began contraception by taking OCPs and then changed to less effective methods.

The effective contraceptors were significantly less likely to become pregnant while using birth control than the less effective contraceptors. Only 37% of the effective contraceptors became pregnant; 51% of the less effective contraceptors did so (odds ratio= 1.4; [Table 4]).

Discussion

Although there has been much research into the effectiveness, side effects, and reasons for discontinuation of individual birth control methods, the personal histories of how women have used contraception has seldom been examined. We attempted to document patterns of contraception use and to relate these patterns to unintended pregnancies.

Two patterns of lifetime contraceptive behaviors in women emerged in our study. Approximately half of the women in our study showed a pattern of changing their birth control methods to more effective ones (the effective contraceptors), and approximately one fourth chose a pattern of methods that became increasingly less effective (the less effective contraceptors). Thus, it may be possible for a health care professional to be able to place a patient in a low- or high-risk group for unintended pregnancy by asking a few questions about her contraception history. It may not ever be possible to completely determine prospectively who is at risk for unintended pregnancy since even highly effective methods have inherent pregnancy rates over time. Physicians, however, should start considering a woman’s history of contraceptive methods as a primary tool for helping to prevent unintended pregnancy.

Ideal Versus Actual Effectiveness

Effectiveness of the ideal use of contraceptive methods is determined by the number of pregnancies per 100 women using the method for 1 year. OCPs (98.5%), DMP (99.7%), Norplant (99.7%), tubal ligations (99.8%), vasectomy (99%), and hysterectomy (100%) all have high ideal-use effectiveness rates.^3,7 These percentages make pregnancy while using any of these methods appear extremely unlikely. Condoms have an ideal effectiveness rate of 85% to 90%, still fairly high numbers.

However, contraception is rarely used ideally. Women forget to take a pill, forget to come in for a DMP shot, use condoms improperly, or become worried about side effects and suddenly stop using their chosen method. These mistakes are taken this into account by recording actual-use effectiveness rates. Still, OCPs have an actual effectiveness rate of 94% and condoms of 82% to 85%, and the actual effectiveness rates of tubal ligation and Norplant do not decrease from their ideal rates.³

Most women have used many forms of contraception, and in our group many changed several times during their lives. The average number of methods was more than 3, and more than half of the women used 5 or more methods. Many women had used both condoms and OCPs during their lives. The reasons for these changes should be examined more carefully (and possibly prospectively). Changing to a more effective method may indicate a lessened desire to ever become pregnant. Two studies have shown that women who want to postpone pregnancy are more likely to get pregnant than those who definitely do not want to give birth.^8,9 Women who are dissatisfied with a contraceptive method are more likely to change that method and more likely to have an unintended pregnancy.² Women who use the same contraceptive method for a long period are known to use them more efficiently.¹⁰ Similarly, women in our study who used several methods were more likely to have become pregnant while using birth control; older women were less likely to become pregnant.

Many women became pregnant while using methods that have good to excellent effectiveness rates. Although some admitted missing a pill or incorrectly using a condom, these women still felt they became pregnant while using a contraceptive method. Women taking OCPs had an actual effectiveness rate of 92.5%, close to the national average of 94%.3 Women using condoms had an effectiveness rate of 93.4%, much higher than the national average of 85%.3 This latter may be explained by the more recent trend of using condoms in addition to another method (condoms to prevent sexually transmitted diseases and another method for birth control).

Forty-two percent of the women in our study became pregnant while using contraception. One reason for this apparent paradox of high effectiveness rates and a high number of unintended pregnancies is the lifetime use of contraception. Effectiveness rates are calculated as the use of a method by 100 women for 1 year. These women all used contraception for more than 1 year. It is the natural history of OCPs with a 95% effectiveness rate that 1 in 20 women will get pregnant in 1 year, and 50 pregnancies will occur in 10 years. Usual actual use of high-effectiveness contraception still carries a significant risk of pregnancy that can be seen by the results in this population of women.

Becoming pregnant even while using good methods of birth control must be expected, explained, and understood. Physicians and their patients must not consider these pregnancies to be failures of the method used, but as inherent part of a life history of contraceptive use. Unintended pregnancies may be a consequence of using even very effective contraceptive methods.

Changing Methods

In our study, becoming pregnant while using a contraceptive method was very likely to cause the woman to discontinue using that method. More than half of the women who became pregnant while using contraception stated pregnancy was their primary reason for changing methods.

In previous studies, a physician’s involvement did not affect a woman’s use or satisfaction with contraception.¹¹ In our study most women had discussed their satisfaction with contraceptive methods with a physician. Discussion or lack of discussion did not affect methods chosen, number of methods chosen, or the chance of becoming pregnant while using birth control.

Other studies have suggested that women may choose highly effective contraception, especially irreversible contraception, because of fear of pregnancy, then have a more satisfying sexual life because this fear has been reduced.^12,13 However, in our study population, no form of contraception significantly affected a woman’s satisfaction with her sexual life.

There was a very low effectiveness rate in our population of women who used DMP. DMP is usually a very effective method that boasts ideal and actual failure rates of less than 3 in 1000 women-years (99.7%). However, 11 of the 82 women in our study reported that they became pregnant while using DMP, for an effectiveness rate of only 86%. The average duration of use of DMP in these women was only 11 months (range=1 week to 60 months). It cannot be determined from the data, but because the average length of use is so short many of these pregnancies may have occurred in women who received only 1 shot and never returned. They considered themselves users of DMP, even though the medication’s effectiveness had waned. Women may have also wanted to please the interviewer or give an answer they thought was appropriate. This may be a major bias of our method of obtaining data. It is perhaps more socially acceptable to claim to be using DMP than to admit to using no method at all, or to claim to be using condoms consistently when actually only using them occasionally.

Limitations

Our study has inherent difficulties. It was a convenience sample; women who were interviewed might have been visiting the physician to obtain a prescription method of contraception, while those who used over-the-counter or rhythm methods may not have been counted proportionally. There is also an inherent recall bias. Women may be more likely to remember a significant fact (such as a pregnancy) as the reason for changing a birth control method rather than the headaches or irregular bleeding that may have contributed to the change. The women may have been more likely to tell the interviewer they were using a birth control method that failed than to say they stopped or forgot to use their method. This would make the methods look less effective.

However, reasons for changing or using contraception are based on the women’s perceptions, so although recall bias may occur, the women’s perceptions are as important as the actual happenings. Whether a woman became pregnant while using a particular method was defined by her recall of the situation. No objective measurements (counting pills, checking charts for DMP shots) were performed. The woman’s perceptions were important for our study because they affected her subsequent use of contraceptive methods. This induced a bias, however, from the interpretation by the women. Another flaw of our study is that it was difficult to determine when a woman was using more than one method concurrently; this would give a higher effectiveness rate than either method individually and overall.

A prospective concurrent study of women’s use of contraception over time would give better answers about why women change contraception and how they use it.

Conclusions

Pregnancy must be considered a possible risk even for those women using an effective method of contraception over a lifetime. OCPs had a lifetime risk of one third for pregnancy in our study population. However, by determining a woman’s pattern of contraceptive use, the health care professional may be able to pinpoint some women who are at higher risk for unintended pregnancies. These women should be followed up more closely and urged to use more effective contraception.

There are many highly effective contraceptive methods available. Some, including oral birth control pills (OCPs), injectable and implantable hormones, and sterilization of both sexes, have ideal effectiveness rates higher than 98% for preventing pregnancy.

However, contraception is not always used ideally. Unplanned or unintended pregnancies do occur. In 1988, US women aged 15 to 44 years reported that 35% of their full-term pregnancies in the preceding 5 years were unintended,¹ in some populations 60% of pregnancies were reported as unintended,² and one third of these ended in abortion.³ Sixty-five percent of adolescent pregnancies are also unintended.⁴ And more than 1 million pregnancies annually are reported to have occurred from misuse of OCPs.³

Pregnancies that occur while the woman is using contraception are considered unintended. These types of pregnancies have poorer outcomes when carried to term than other pregnancies, including an increased incidence of premature birth and intrauterine growth retardation.⁵ This is an area of concern for the women, their partners, and the health care providers who help these women with contraception concerns.

The way women use and experience contraception profoundly affects its effectiveness.^2,6 OCPs have been available for more than 30 years, implantable contraceptives for approximately 8 years, and injectable contraceptives for 5 to 7 years. Today, there are women who have had greater than 95% effective hormonal contraception available all their childbearing years. The purpose of our study was to discover what patterns of contraceptive use women developed during their childbearing years and how these patterns were related to unintended pregnancies.

Methods

We obtained a convenience sampling of women who visited an urban/suburban family practice residency office that is run by an open-panel health maintenance organization (HMO), but accepts more than 60 different health plans (including Medicare, HMO Medicare, Medicaid, HMO Medicaid, and self-pay) and includes maternity care. The practice has more than 25,000 patient visits annually (60% women) who are residents of eastern Baltimore, Maryland, and Baltimore County. Women patients and women relatives or friends of patients who entered the waiting room and were aged 18 to 50 years were asked to participate in a face-to-face half-hour interview by a medical student. Two medical students talked with 396 women during June and July 1999. The interviews were recorded by identification number only.

After giving informed consent, the women spent approximately 30 minutes answering questions about marital status, education, socioeconomic status, drug and cigarette use, and first and subsequent contraceptive experiences. Each woman was given a list of 20 methods of contraception and 20 reasons for discontinuation to help her; both lists had an “other” category. She was asked what contraceptive she remembered she had used first, for how long, and when she had changed methods and why; these questions were repeated for every method used since. The questionnaire had been pretested for 2 years in 2 previous studies of more than 600 women. Only 5 women refused to participate in our survey, most often because of time constraints.

One author (J.A.R.) entered the data in Excel (Microsoft Corporation; Redmond, Wash) spreadsheets. The statisticians at the MedAtlantic Research Institute converted the spreadsheets with the Statistical Package for the Social Sciences (SPSS, Inc; Chicago, Ill). Variables were analyzed for correlation and significance by Student t and chi-square tests. The patterns of contraceptive use were analyzed, and demographic values were compared.

Results

A total of 396 women participated in our survey. They were representative of urban/suburban women in general and of women of the practice [Table 1]. The average age of the respondents was 27 years; 98% (n=389) had sex with a man.

During their lifetimes these women used an aggregate total of 1421 methods of contraception (average=3.52±1.56 methods per woman; range=1-10). The methods used are listed in [Table 2]. All of the women (except the 5 who never had sex with a man) used 1 method of contraception at least once, 370 (93%) had used 2 methods, 287 (72%) had used 3, and 217 (55%) had used 4 or more. Eighty-one women (20%) reported having used more than 1 method at a time. Condoms were the most common first method of contraception (62%). The most common contraceptive methods used overall were OCPs (81%) and condoms (78%). Thirty-four percent had tried implantable or injectable hormonal contraception at least once. And 67 women (17%) had a tubal ligation (average age=28 years; range=21-45 years).

A total of 168 (42%) of the women became pregnant while using contraception, some more than once. Forty-nine women (13%) became pregnant twice and 13 women (3%) 3 times while using contraception. One hundred three women became pregnant while taking OCPs, 78 while using condoms and 11 while using an injectable contraceptive (depot medroxyprogesterone [DMP]). The women who became pregnant while using birth control were significantly more likely to be younger and African American. They were more likely to have a history of early initiation of birth control use, and they used more methods during their lifetimes. These women were pregnant more often and were more likely to use some type of public assistance [Table 3]. The total group of African American women in our survey was more likely to be younger, have some college education, and be single. Marital status did not correlate with becoming pregnant while using birth control; married, single, and separated or divorced women were equally represented. Women who became pregnant while using birth control were not more likely to use cigarettes, marijuana, or cocaine; have more partners; or start sexual relationships at an earlier age.

The rate of actual-use effectiveness of OCPs in this population was similar to the national average. First-year use effectiveness rates could not be determined, but lifetime effectiveness rates were estimated. Three hundred nineteen women (81%) had used OCPs for an aggregate total of 1422 years (average=4.5 years per woman; range=1 month-28 years). One hundred three women (33.3%) reported that they had become pregnant while using OCPs, some more than once. This was a pregnancy rate of 7.5%. Fifty-seven percent of the women who became pregnant while taking OCPs said they stopped using them because of getting pregnant.

Three hundred four women (78%) had used condoms for an aggregate total of 1178 years (average=3.9 years per woman; range=1 month-25 years). Seventy-eight women (25.6%) reported becoming pregnant while using condoms for a pregnancy rate of 6.6%. Fifty-five percent of those women stated that becoming pregnant was the reason for stopping condom use.

Eight-two women had used DMP for an aggregate total of 77.1 years with an average of 11 months per woman. Eleven women reported that they became pregnant while using DMP, for a pregnancy rate in our study of 14%.

Two major patterns of contraceptive use during a woman’s lifetime emerged. These 2 patterns described the contraceptive choices or directions of 82% of the women. One group of 210 women (53%), who will be called the “effective contraceptors,” started with condoms and then used OCPs or DMP. Following that change, they either continued to take OCPs or changed again to a method with a higher actual effectiveness rate (ie, DMP, Norplant [Wyeth-Ayerst; St. Davids, Penn] intrauterine device, tubal ligation, vasectomy, or hysterectomy).

The other group (the “less effective contraceptors,” n=110, 29%) also started with condoms. Forty-seven of these women changed immediately to a method less effective overall than OCPs (rhythm, withdrawal, gel/foam, diaphragm, or no method); 25 changed to use OCPs, and then began to use methods less effective than OCPs. Also included in this group were 38 women who began contraception by taking OCPs and then changed to less effective methods.

The effective contraceptors were significantly less likely to become pregnant while using birth control than the less effective contraceptors. Only 37% of the effective contraceptors became pregnant; 51% of the less effective contraceptors did so (odds ratio= 1.4; [Table 4]).

Discussion

Although there has been much research into the effectiveness, side effects, and reasons for discontinuation of individual birth control methods, the personal histories of how women have used contraception has seldom been examined. We attempted to document patterns of contraception use and to relate these patterns to unintended pregnancies.

Two patterns of lifetime contraceptive behaviors in women emerged in our study. Approximately half of the women in our study showed a pattern of changing their birth control methods to more effective ones (the effective contraceptors), and approximately one fourth chose a pattern of methods that became increasingly less effective (the less effective contraceptors). Thus, it may be possible for a health care professional to be able to place a patient in a low- or high-risk group for unintended pregnancy by asking a few questions about her contraception history. It may not ever be possible to completely determine prospectively who is at risk for unintended pregnancy since even highly effective methods have inherent pregnancy rates over time. Physicians, however, should start considering a woman’s history of contraceptive methods as a primary tool for helping to prevent unintended pregnancy.

Ideal Versus Actual Effectiveness

Effectiveness of the ideal use of contraceptive methods is determined by the number of pregnancies per 100 women using the method for 1 year. OCPs (98.5%), DMP (99.7%), Norplant (99.7%), tubal ligations (99.8%), vasectomy (99%), and hysterectomy (100%) all have high ideal-use effectiveness rates.^3,7 These percentages make pregnancy while using any of these methods appear extremely unlikely. Condoms have an ideal effectiveness rate of 85% to 90%, still fairly high numbers.

However, contraception is rarely used ideally. Women forget to take a pill, forget to come in for a DMP shot, use condoms improperly, or become worried about side effects and suddenly stop using their chosen method. These mistakes are taken this into account by recording actual-use effectiveness rates. Still, OCPs have an actual effectiveness rate of 94% and condoms of 82% to 85%, and the actual effectiveness rates of tubal ligation and Norplant do not decrease from their ideal rates.³

Most women have used many forms of contraception, and in our group many changed several times during their lives. The average number of methods was more than 3, and more than half of the women used 5 or more methods. Many women had used both condoms and OCPs during their lives. The reasons for these changes should be examined more carefully (and possibly prospectively). Changing to a more effective method may indicate a lessened desire to ever become pregnant. Two studies have shown that women who want to postpone pregnancy are more likely to get pregnant than those who definitely do not want to give birth.^8,9 Women who are dissatisfied with a contraceptive method are more likely to change that method and more likely to have an unintended pregnancy.² Women who use the same contraceptive method for a long period are known to use them more efficiently.¹⁰ Similarly, women in our study who used several methods were more likely to have become pregnant while using birth control; older women were less likely to become pregnant.

Many women became pregnant while using methods that have good to excellent effectiveness rates. Although some admitted missing a pill or incorrectly using a condom, these women still felt they became pregnant while using a contraceptive method. Women taking OCPs had an actual effectiveness rate of 92.5%, close to the national average of 94%.3 Women using condoms had an effectiveness rate of 93.4%, much higher than the national average of 85%.3 This latter may be explained by the more recent trend of using condoms in addition to another method (condoms to prevent sexually transmitted diseases and another method for birth control).

Forty-two percent of the women in our study became pregnant while using contraception. One reason for this apparent paradox of high effectiveness rates and a high number of unintended pregnancies is the lifetime use of contraception. Effectiveness rates are calculated as the use of a method by 100 women for 1 year. These women all used contraception for more than 1 year. It is the natural history of OCPs with a 95% effectiveness rate that 1 in 20 women will get pregnant in 1 year, and 50 pregnancies will occur in 10 years. Usual actual use of high-effectiveness contraception still carries a significant risk of pregnancy that can be seen by the results in this population of women.

Becoming pregnant even while using good methods of birth control must be expected, explained, and understood. Physicians and their patients must not consider these pregnancies to be failures of the method used, but as inherent part of a life history of contraceptive use. Unintended pregnancies may be a consequence of using even very effective contraceptive methods.

Changing Methods

In our study, becoming pregnant while using a contraceptive method was very likely to cause the woman to discontinue using that method. More than half of the women who became pregnant while using contraception stated pregnancy was their primary reason for changing methods.

In previous studies, a physician’s involvement did not affect a woman’s use or satisfaction with contraception.¹¹ In our study most women had discussed their satisfaction with contraceptive methods with a physician. Discussion or lack of discussion did not affect methods chosen, number of methods chosen, or the chance of becoming pregnant while using birth control.

Other studies have suggested that women may choose highly effective contraception, especially irreversible contraception, because of fear of pregnancy, then have a more satisfying sexual life because this fear has been reduced.^12,13 However, in our study population, no form of contraception significantly affected a woman’s satisfaction with her sexual life.

There was a very low effectiveness rate in our population of women who used DMP. DMP is usually a very effective method that boasts ideal and actual failure rates of less than 3 in 1000 women-years (99.7%). However, 11 of the 82 women in our study reported that they became pregnant while using DMP, for an effectiveness rate of only 86%. The average duration of use of DMP in these women was only 11 months (range=1 week to 60 months). It cannot be determined from the data, but because the average length of use is so short many of these pregnancies may have occurred in women who received only 1 shot and never returned. They considered themselves users of DMP, even though the medication’s effectiveness had waned. Women may have also wanted to please the interviewer or give an answer they thought was appropriate. This may be a major bias of our method of obtaining data. It is perhaps more socially acceptable to claim to be using DMP than to admit to using no method at all, or to claim to be using condoms consistently when actually only using them occasionally.

Limitations

Our study has inherent difficulties. It was a convenience sample; women who were interviewed might have been visiting the physician to obtain a prescription method of contraception, while those who used over-the-counter or rhythm methods may not have been counted proportionally. There is also an inherent recall bias. Women may be more likely to remember a significant fact (such as a pregnancy) as the reason for changing a birth control method rather than the headaches or irregular bleeding that may have contributed to the change. The women may have been more likely to tell the interviewer they were using a birth control method that failed than to say they stopped or forgot to use their method. This would make the methods look less effective.

However, reasons for changing or using contraception are based on the women’s perceptions, so although recall bias may occur, the women’s perceptions are as important as the actual happenings. Whether a woman became pregnant while using a particular method was defined by her recall of the situation. No objective measurements (counting pills, checking charts for DMP shots) were performed. The woman’s perceptions were important for our study because they affected her subsequent use of contraceptive methods. This induced a bias, however, from the interpretation by the women. Another flaw of our study is that it was difficult to determine when a woman was using more than one method concurrently; this would give a higher effectiveness rate than either method individually and overall.

A prospective concurrent study of women’s use of contraception over time would give better answers about why women change contraception and how they use it.

Conclusions

Pregnancy must be considered a possible risk even for those women using an effective method of contraception over a lifetime. OCPs had a lifetime risk of one third for pregnancy in our study population. However, by determining a woman’s pattern of contraceptive use, the health care professional may be able to pinpoint some women who are at higher risk for unintended pregnancies. These women should be followed up more closely and urged to use more effective contraception.

The world of medicine is permeated by the influence of the pharmaceutical industry, which spends an estimated $10 billion (more than $13,000 per physician) each year on drug promotion.¹ This industry’s influence extends from medical education to clinical research and patient care.² For example, medical journals and medical conferences are supported by pharmaceutical advertising or depend on financial contributions from pharmaceutical companies to offer their programs. Pharmaceutical companies grant a variety of scholarships and subsidies. The pharmaceutical industry also reinvests a portion of its profits to support new research and development, including hundreds of clinical trials reported each year in the scientific literature.³ Up-to-date drug detailing and continuing medical education (CME) is offered to clinicians, and patient education materials are frequently made available to practices. Patients, especially those with a lower income, benefit from receiving free medication samples donated by pharmaceutical companies.

Pharmaceutical companies, health care providers, and patients represent unique interests that may at times overlap. How the interactions between physicians and pharmaceutical representatives influence prescribing habits, decision making, and physician behavior in general is a contestable topic that has not been well studied or thoroughly understood. In the last several years, medical organizations in the United States and Canada have released discussion papers and policy statements about the relationship between the medical profession and the pharmaceutical industry.³ Whether the influence of the pharmaceutical industry is ultimately beneficial or detrimental remains a contentious and complex topic shrouded in much controversy.^4-15

The interactions between the pharmaceutical industry and physicians have been discussed in a plethora of publications. Several of these articles have focused on the ethical implications of these interactions and the potential for exploitation; however, they have commonly been based on surveys and self-reporting and have often been limited to providers in academic or residency training contexts.^3,16-21 Brotzman and Mark¹⁶ encourage the development of more comprehensive policies for residency training purposes. Kelcher and coworkers¹⁹ promote the concept of a structured educational program to prepare residents for future contact with the pharmaceutical industry. Lexchin⁹ discusses the need for critical comparison when evaluating drug information received from pharmaceutical representatives.

So how do family practice clinicians approach this challenging relationship? Through direct observation of these interactions within the private practice context, we attempted to add a new and vital dimension to the assessment of this rather complex symbiosis. We describe the approaches and reactions of individual family physicians and their practices to pharmaceutical representatives.

Methods

The Prevention and Competing Demands in Primary Care Practice study was designed to examine the organizational contexts that support or inhibit the delivery of preventive services in family medicine practices. Eighteen practices were studied using a multimethod ethnographic design that involved extensive observational field notes of the office system and the clinical encounters. These notes were collected by a researcher who spent at least 4 weeks in each practice. A total of 44 physicians and 9 other clinicians were shadowed, and approximately 1600 patient encounters were directly observed. Individual depth interviews were conducted with each clinician and many of the practice staff to obtain their perceptions.

Sampling

The comparative case study design began with an initial phase of purposefully selecting practices that included both urban and rural settings and those with different intensities of preventive services delivery, (n=10) followed by a sample to confirm or refute evolving hypotheses (n=8). The practices were selected from those that had participated in the Centers for Disease Control and Prevention/Nebraska Department of Health–funded Tobacco Use Prevention in Physicians Offices study (Helen McIlvain, principal investigator). These high-volume practices had a wide variation in practice location (urban, suburban, and rural) and organizational structure (solo vs group, independent vs system affiliated). Because we noted some interesting variations among practices from different health systems in the early analyses, the replication sample ensured at least 2 practices from each of 4 major health systems, and 2 from those practices that were seen as struggling to provide preventive services in the tobacco study.

We solicited study participation by sending an invitation letter followed by a telephone call to one of the physicians in the practice. Participation was very high, requiring contact with 23 practices to obtain the 18 deemed necessary in the study design.

Data Collection

We sent a field researcher trained in qualitative methods to each practice where she used a variety of data collection methods to produce a comprehensive picture of the practice as a functioning organization. Data were collected through direct observation of the practice,²² the use of structured checklists of the office environment,²³ direct observation of patient encounters supplemented with structured checklists, the use of patient pathways,²⁴ individual depth interviews with physicians and other key staff members,²⁵ and chart audits. It took from 4 to 12 weeks for the field researcher to complete the data collection in each practice, depending on its size.

The field researcher observed and jotted details of the physical environment and functioning of the practice, then dictated extensive field notes at the end of each day. These field notes contained detailed descriptions of the clinic location and environment, patient characteristics, nursing station, examination rooms, the waiting area, physician offices, bulletin boards, posters, and patient education materials. Photographs were taken of each room, including the reception area, waiting area, nurses’ station, nurse intake room, and the examination rooms. Existing practice personnel, their roles and duties, and their relationships and interactions with other staff members were characterized in a practice genogram.²⁶ Physical office systems, including charts, flow sheets, and computer systems were described, as were functional office routines and procedures. The field researcher specifically noted the storage and organization of drug samples within the clinic, comments made by staff relating to pharmaceutical representatives, and any observed interaction between staff and representatives.

The field researcher attempted to observe 30 patient encounters for each clinician; because of patient volume and scheduling this was not always possible. Thirty or more patient encounters were observed in 42 of the 53 clinicians participating in the study, and 20 or more patient encounters were observed for all but 1 provider (who had 18 encounters). After obtaining written informed consent from the patient, the field researcher shadowed the clinician and jotted notes about each encounter for later dictation. The patient encounter field notes contained rich descriptions about any verbal patient education delivered and the context of that education, including the reason for the visit, how the visit unfolded, and how the clinician and patient interacted. The patient encounter dictation captured the distribution of drug samples, discussion between the clinician and patient regarding drug samples, prescriptions written by physicians, and instructions given by physicians.

Using the depth interviewing technique, the field researcher interviewed all family physicians within the practice, the office manager, head nurse, and other significant persons within the practice.²⁵ Questions asked by the field researcher related to overall views of medicine, general practice characteristics, the systems in place for the delivery of preventive medicine, and perceived rates of delivering preventive medicine. These interviews were tape-recorded and transcribed verbatim.

Additional data such as photographs and chart audits were collected as part of the larger study but were not included in our analysis.

Data Analysis and Interpretation

Qualitative data were transcribed, checked for accuracy by having the field worker review the manuscript, and entered into FolioViews 4.11, an infobase software package.²⁷ We began our analysis by immersing ourselves in all the data from 2 practices to understand the function of the practices and the overall richness and variation of the data collected, and to identify any mention of drug samples or interactions with pharmaceutical representatives. Using this immersion/crystallization approach, we worked to formulate hypotheses and an initial organizational scheme.²⁸ Group discussions led to the development of a codebook of key words.

Two investigators then performed computer word searches on the infobases of the 2 previously selected practices using the codebook. All relevant passages were tagged and printed, and the information retrieved was compared with data identified from an earlier in-depth reading of the transcripts from the 2 practices we initially studied. The word search technique and template organizing style was used to identify relevant portions of 4 more practices for secondary data analysis.²⁹

Finally, segments of the selected field notes and patient encounter notes that described drug samples and pharmaceutical representatives were identified and organized into preliminary tables or matrices.³⁰ After constructing matrices for the 6 selected practices we held discussions and further refined the organization and categories within the matrices. Once the matrices were refined they were used to organize relevant data identified by word searches for the rest of the 18 practices in the sample. The matrices served as the framework for data analysis, allowed the visualization of emergent patterns, and facilitated comparisons across physicians and practices.

Results

Individual clinicians and their practices displayed noticeable variation in their approaches to pharmaceutical representatives and the use of samples. Our analysis revealed patterns with respect to the types of contact they have with the pharmaceutical representatives, their use of sample medication, and the relative benefits obtained from the interactions.

Contact with the Pharmaceutical Representatives

All 18 practices had some form of contact with the pharmaceutical industry. This contact ranged from scheduled and well-organized meeting times to random interactions. Formal strategies and policies regarding drug representative interactions and the use of samples were in place in 8 of the 18 practices (44%). One other practice voiced a concern regarding the need for such a policy.

Most of the 8 clinics that had a formal plan or policy preferred a specific time (such as the lunch hour) during which the physicians and staff could meet with the representative. Some practices determined which days and times would be most convenient for their staff. One clinic made the clinician’s schedule available, thus allowing the representatives to schedule visits accordingly. One rural physician even allotted 30-minute patient appointment times to each representative to optimize the quality of these interactions. These lengthy visits facilitated more extensive drug detailing and CME, and generated an ample supply of samples, which were used to serve the needs of this particular rural community. A receptionist in another clinic was found to simply use a chalkboard to notify the staff whenever a representative had set up a presentation in the office. All of these measures appeared to be effective in minimizing confusion, distraction, and schedule disruption.

The remaining 10 practices displayed more haphazard dealings with drug representatives. These included brief hallway meetings that happened when clinicians were confronted by a representative as they exited the examination room. Other pharmaceutical representatives waited in medication sample rooms, hoping to catch the physicians as they searched for samples. These casual relationships were, at times, counterproductive. In one practice the physicians seemed oblivious to when “drug lunches” were scheduled, leading to frustration on the part of the pharmaceutical representatives; in another, the physician found herself rudely interrupted by an imposing representative during her own lunch.

Usage and Storage of Sample Medication

In the 18 practices, medication samples were used in 19.8% of the 1588 observed patient encounters. Multiple drugs were dispensed in 14.6% of the encounters in which samples were used. In only 5.1% of cases was a medication dispensed as a result of a patient’s specific request; the clinician usually initiated the distribution of samples. Drug samples were also offered to additional family members in 3.5% of encounters, sometimes even when these individuals were not accompanying the patient to the given appointment. This appeared to be particularly common in certain rural practices where the clinicians seemed sensitive to the needs of their patients and their families.

A review of the types of medication dispensed revealed noticeable trends and pointed to a prevalence of certain medication categories [Table 1]. The top 4 categories included asthma and allergy remedies, anti-infective agents, analgesics and anti-inflammatory medications, and antihypertensive drugs. These major categories accounted for more than 63% of all drugs dispensed.

The duration for which these medication samples were given varied from starter dosages (lasting 1-3 days) to complete courses of treatments (eg, a course of antibiotics) to amounts sufficient to supply the patients’ needs for several months (eg, antidepressants). The analysis of the clinician’s intent when dispensing samples identified that some were using samples to test for efficacy and tolerability while others were attempting to offer temporary relief or convenience to the patient. Also, certain physicians were clearly concerned about cost savings for their patients. The use of medication samples could be seen to represent tangible benefits to the individual patient.

Instructions accompanied the dispensing of samples in 150 (47.8%) of the cases and were predominantly verbal in nature. Dosing appeared to be the main focus, and little to no attention was given to more detailed information, such as whether the medication should be taken with meals. Drug interactions were not routinely discussed.

The majority of dispensing was done by physicians, physician assistants, and nurse practitioners. In some clinics, office staff had access to samples and were asked to consult with the physician before using these samples. Personal use of samples was documented in at least 4 practices, and in 2 of these the nurses were described as “helping themselves.” One particular clinic even voiced concerns about patients having unsupervised access and mentioned the need to develop a drug policy.

Dispensing patterns varied significantly among the clinics. Certain practices, especially those in rural, underserved communities, were found to dispense more liberally than others. These clinics appeared to be particularly sensitive to the needs of the individual patients and those of the greater community. In one practice, medication was handed out in 39.2% of encounters. On one occasion the physician waited until after the pharmaceutical representative had delivered an adequate supply of the necessary medication to give it to the patient. In contrast, clinicians in a particular suburban clinic only dispensed drugs 4.4% of the time.

However, even within a given practice, vast individual variances existed for each clinician in their approach to samples. In one clinic, 2 of the partners displayed disparate habits: One physician used samples regularly (in 41.9% of observed encounters), and the other only did so on rare occasions (in 3.2% of observed encounters). Of all the 18 sites, the physician who dispensed samples most frequently did so in 56.7% of observed encounters; the physician, however, at the other end of the spectrum dispensed medications in 2.4% of the observed patient encounters.

Of the 18 practices, 9 made use of specific closets or rooms for the storage of medication samples. One particular practice boasted an extraordinary and meticulously organized shelving system, complete with labeled bins and demarcated sections. The medical assistant in whose office these medications were stored was responsible for dispensing the ordered medications. In contrast to this efficient system, most clinics displayed very little structure in the organization and dispensing of samples. One clinic was described as having an overfilled, disorganized stack of shelves where medications were placed alongside vitamin supplements and herbal products. In this clinic staff spent vast amounts of time searching for the appropriate samples.

Only one clinic kept the sample room locked; another office had placed a notice on the door to keep it shut. It is important to note, however, that these storage places did not contain narcotics or similarly scheduled medications. In spite of Health Care Financing Administration regulations, we did not find any documentation to support which samples with which serial numbers had been distributed to patients, making effective recalls almost impossible. Stocks were most often replenished during routine drug representative visits. In one of the sites the head nurse kept in contact with the representatives and notified them when additional stocks were needed.

Benefits of These Interactions

We found that physicians and patients gained some advantage from their dealings with drug companies. The benefits the practices obtained from their contact with the pharmaceutical industry varied substantially. Other than medication samples, the most commonly observed benefits included the provision of meals, treats, and patient education materials [Table 2].

Patients also profited in a spectrum of ways. While samples represented tangible cost savings, immediate relief, and convenience to the individual patient (and occasionally to their family members), patient education materials facilitated further understanding of their diagnosis, potentially leading to a higher degree of satisfaction with their health care.

Discussion

Our study suggests that the relationship between the pharmaceutical industry and clinicians is symbiotic. It is a complex relationship that deserves objective analysis regarding its virtues and pitfalls.

Although it is essential to acknowledge the extent of the pharmaceutical industry’s influence, it is equally important to recognize and enhance the opportunities that clinicians have to direct this relationship. Our data suggest that clinics with existing policies for interactions with drug companies appear to derive more satisfaction and less frustration from their encounters. Formal meeting times eliminated unwanted interruptions and distractions from patient care and made the interactions more meaningful. Physicians who negotiated extensive and structured contact with drug representatives often received regular visits and generous supplies of both samples and patient education materials. Drug representative briefings, even though these needed to be viewed with healthy skepticism, also constituted a welcome form of CME for certain physicians.³¹

We found, however, that many offices do not have formal policies regarding pharmaceutical representatives and use questionable sample handling and storage practices. Interactions with pharmaceutical representatives can also be time consuming, potentially affecting the patient care schedule. Dispensing medications in the office setting may lead to increased responsibility and liability. We found that few clinicians provided extensive patient education about the samples they dispensed. Physicians need to be cognizant of the fact that they are bypassing the pharmacist. Directions to the patient need to be clear and comprehensive, addressing issues such as polypharmacy and drug-drug interactions. In addition, physicians should remain mindful of their own vulnerability with respect to the pharmaceutical industry’s influence over their prescribing habits. Drug detailing, even when insightful, may be biased.

Conclusions

The management of office visits with pharmaceutical representatives and the distribution of medication samples need to be addressed. Clinics with specific policies for interactions with drug companies appear to derive more satisfaction from their encounters. Practices should also maintain specific strategies when it comes to dispensing samples and should develop patterns of accountability for the handling of samples in the office.

Future research should include issues such as the impact of the pharmaceutical representative’s visit on patient finances, attitudes, disease management, and the physician-patient relationship. Further work is needed in the area of prescribing habits and the factors influencing physician decision making. More information on these topics may ultimately define and establish a relationship between the physician and the pharmaceutical industry that protects and promotes the interests of patients.

Acknowledgments

Our work was supported by a grant from the Agency for Health Care Policy and Research (1RO1 HS08776) to Benjamin Crabtree, PhD.

The world of medicine is permeated by the influence of the pharmaceutical industry, which spends an estimated $10 billion (more than $13,000 per physician) each year on drug promotion.¹ This industry’s influence extends from medical education to clinical research and patient care.² For example, medical journals and medical conferences are supported by pharmaceutical advertising or depend on financial contributions from pharmaceutical companies to offer their programs. Pharmaceutical companies grant a variety of scholarships and subsidies. The pharmaceutical industry also reinvests a portion of its profits to support new research and development, including hundreds of clinical trials reported each year in the scientific literature.³ Up-to-date drug detailing and continuing medical education (CME) is offered to clinicians, and patient education materials are frequently made available to practices. Patients, especially those with a lower income, benefit from receiving free medication samples donated by pharmaceutical companies.

Pharmaceutical companies, health care providers, and patients represent unique interests that may at times overlap. How the interactions between physicians and pharmaceutical representatives influence prescribing habits, decision making, and physician behavior in general is a contestable topic that has not been well studied or thoroughly understood. In the last several years, medical organizations in the United States and Canada have released discussion papers and policy statements about the relationship between the medical profession and the pharmaceutical industry.³ Whether the influence of the pharmaceutical industry is ultimately beneficial or detrimental remains a contentious and complex topic shrouded in much controversy.^4-15

The interactions between the pharmaceutical industry and physicians have been discussed in a plethora of publications. Several of these articles have focused on the ethical implications of these interactions and the potential for exploitation; however, they have commonly been based on surveys and self-reporting and have often been limited to providers in academic or residency training contexts.^3,16-21 Brotzman and Mark¹⁶ encourage the development of more comprehensive policies for residency training purposes. Kelcher and coworkers¹⁹ promote the concept of a structured educational program to prepare residents for future contact with the pharmaceutical industry. Lexchin⁹ discusses the need for critical comparison when evaluating drug information received from pharmaceutical representatives.

So how do family practice clinicians approach this challenging relationship? Through direct observation of these interactions within the private practice context, we attempted to add a new and vital dimension to the assessment of this rather complex symbiosis. We describe the approaches and reactions of individual family physicians and their practices to pharmaceutical representatives.

Methods

The Prevention and Competing Demands in Primary Care Practice study was designed to examine the organizational contexts that support or inhibit the delivery of preventive services in family medicine practices. Eighteen practices were studied using a multimethod ethnographic design that involved extensive observational field notes of the office system and the clinical encounters. These notes were collected by a researcher who spent at least 4 weeks in each practice. A total of 44 physicians and 9 other clinicians were shadowed, and approximately 1600 patient encounters were directly observed. Individual depth interviews were conducted with each clinician and many of the practice staff to obtain their perceptions.

Sampling

The comparative case study design began with an initial phase of purposefully selecting practices that included both urban and rural settings and those with different intensities of preventive services delivery, (n=10) followed by a sample to confirm or refute evolving hypotheses (n=8). The practices were selected from those that had participated in the Centers for Disease Control and Prevention/Nebraska Department of Health–funded Tobacco Use Prevention in Physicians Offices study (Helen McIlvain, principal investigator). These high-volume practices had a wide variation in practice location (urban, suburban, and rural) and organizational structure (solo vs group, independent vs system affiliated). Because we noted some interesting variations among practices from different health systems in the early analyses, the replication sample ensured at least 2 practices from each of 4 major health systems, and 2 from those practices that were seen as struggling to provide preventive services in the tobacco study.

We solicited study participation by sending an invitation letter followed by a telephone call to one of the physicians in the practice. Participation was very high, requiring contact with 23 practices to obtain the 18 deemed necessary in the study design.

Data Collection

We sent a field researcher trained in qualitative methods to each practice where she used a variety of data collection methods to produce a comprehensive picture of the practice as a functioning organization. Data were collected through direct observation of the practice,²² the use of structured checklists of the office environment,²³ direct observation of patient encounters supplemented with structured checklists, the use of patient pathways,²⁴ individual depth interviews with physicians and other key staff members,²⁵ and chart audits. It took from 4 to 12 weeks for the field researcher to complete the data collection in each practice, depending on its size.

The field researcher observed and jotted details of the physical environment and functioning of the practice, then dictated extensive field notes at the end of each day. These field notes contained detailed descriptions of the clinic location and environment, patient characteristics, nursing station, examination rooms, the waiting area, physician offices, bulletin boards, posters, and patient education materials. Photographs were taken of each room, including the reception area, waiting area, nurses’ station, nurse intake room, and the examination rooms. Existing practice personnel, their roles and duties, and their relationships and interactions with other staff members were characterized in a practice genogram.²⁶ Physical office systems, including charts, flow sheets, and computer systems were described, as were functional office routines and procedures. The field researcher specifically noted the storage and organization of drug samples within the clinic, comments made by staff relating to pharmaceutical representatives, and any observed interaction between staff and representatives.

The field researcher attempted to observe 30 patient encounters for each clinician; because of patient volume and scheduling this was not always possible. Thirty or more patient encounters were observed in 42 of the 53 clinicians participating in the study, and 20 or more patient encounters were observed for all but 1 provider (who had 18 encounters). After obtaining written informed consent from the patient, the field researcher shadowed the clinician and jotted notes about each encounter for later dictation. The patient encounter field notes contained rich descriptions about any verbal patient education delivered and the context of that education, including the reason for the visit, how the visit unfolded, and how the clinician and patient interacted. The patient encounter dictation captured the distribution of drug samples, discussion between the clinician and patient regarding drug samples, prescriptions written by physicians, and instructions given by physicians.

Using the depth interviewing technique, the field researcher interviewed all family physicians within the practice, the office manager, head nurse, and other significant persons within the practice.²⁵ Questions asked by the field researcher related to overall views of medicine, general practice characteristics, the systems in place for the delivery of preventive medicine, and perceived rates of delivering preventive medicine. These interviews were tape-recorded and transcribed verbatim.

Additional data such as photographs and chart audits were collected as part of the larger study but were not included in our analysis.

Data Analysis and Interpretation

Qualitative data were transcribed, checked for accuracy by having the field worker review the manuscript, and entered into FolioViews 4.11, an infobase software package.²⁷ We began our analysis by immersing ourselves in all the data from 2 practices to understand the function of the practices and the overall richness and variation of the data collected, and to identify any mention of drug samples or interactions with pharmaceutical representatives. Using this immersion/crystallization approach, we worked to formulate hypotheses and an initial organizational scheme.²⁸ Group discussions led to the development of a codebook of key words.

Two investigators then performed computer word searches on the infobases of the 2 previously selected practices using the codebook. All relevant passages were tagged and printed, and the information retrieved was compared with data identified from an earlier in-depth reading of the transcripts from the 2 practices we initially studied. The word search technique and template organizing style was used to identify relevant portions of 4 more practices for secondary data analysis.²⁹

Finally, segments of the selected field notes and patient encounter notes that described drug samples and pharmaceutical representatives were identified and organized into preliminary tables or matrices.³⁰ After constructing matrices for the 6 selected practices we held discussions and further refined the organization and categories within the matrices. Once the matrices were refined they were used to organize relevant data identified by word searches for the rest of the 18 practices in the sample. The matrices served as the framework for data analysis, allowed the visualization of emergent patterns, and facilitated comparisons across physicians and practices.

Results

Individual clinicians and their practices displayed noticeable variation in their approaches to pharmaceutical representatives and the use of samples. Our analysis revealed patterns with respect to the types of contact they have with the pharmaceutical representatives, their use of sample medication, and the relative benefits obtained from the interactions.

Contact with the Pharmaceutical Representatives

All 18 practices had some form of contact with the pharmaceutical industry. This contact ranged from scheduled and well-organized meeting times to random interactions. Formal strategies and policies regarding drug representative interactions and the use of samples were in place in 8 of the 18 practices (44%). One other practice voiced a concern regarding the need for such a policy.

Most of the 8 clinics that had a formal plan or policy preferred a specific time (such as the lunch hour) during which the physicians and staff could meet with the representative. Some practices determined which days and times would be most convenient for their staff. One clinic made the clinician’s schedule available, thus allowing the representatives to schedule visits accordingly. One rural physician even allotted 30-minute patient appointment times to each representative to optimize the quality of these interactions. These lengthy visits facilitated more extensive drug detailing and CME, and generated an ample supply of samples, which were used to serve the needs of this particular rural community. A receptionist in another clinic was found to simply use a chalkboard to notify the staff whenever a representative had set up a presentation in the office. All of these measures appeared to be effective in minimizing confusion, distraction, and schedule disruption.

The remaining 10 practices displayed more haphazard dealings with drug representatives. These included brief hallway meetings that happened when clinicians were confronted by a representative as they exited the examination room. Other pharmaceutical representatives waited in medication sample rooms, hoping to catch the physicians as they searched for samples. These casual relationships were, at times, counterproductive. In one practice the physicians seemed oblivious to when “drug lunches” were scheduled, leading to frustration on the part of the pharmaceutical representatives; in another, the physician found herself rudely interrupted by an imposing representative during her own lunch.

Usage and Storage of Sample Medication

In the 18 practices, medication samples were used in 19.8% of the 1588 observed patient encounters. Multiple drugs were dispensed in 14.6% of the encounters in which samples were used. In only 5.1% of cases was a medication dispensed as a result of a patient’s specific request; the clinician usually initiated the distribution of samples. Drug samples were also offered to additional family members in 3.5% of encounters, sometimes even when these individuals were not accompanying the patient to the given appointment. This appeared to be particularly common in certain rural practices where the clinicians seemed sensitive to the needs of their patients and their families.

A review of the types of medication dispensed revealed noticeable trends and pointed to a prevalence of certain medication categories [Table 1]. The top 4 categories included asthma and allergy remedies, anti-infective agents, analgesics and anti-inflammatory medications, and antihypertensive drugs. These major categories accounted for more than 63% of all drugs dispensed.

The duration for which these medication samples were given varied from starter dosages (lasting 1-3 days) to complete courses of treatments (eg, a course of antibiotics) to amounts sufficient to supply the patients’ needs for several months (eg, antidepressants). The analysis of the clinician’s intent when dispensing samples identified that some were using samples to test for efficacy and tolerability while others were attempting to offer temporary relief or convenience to the patient. Also, certain physicians were clearly concerned about cost savings for their patients. The use of medication samples could be seen to represent tangible benefits to the individual patient.

Instructions accompanied the dispensing of samples in 150 (47.8%) of the cases and were predominantly verbal in nature. Dosing appeared to be the main focus, and little to no attention was given to more detailed information, such as whether the medication should be taken with meals. Drug interactions were not routinely discussed.

The majority of dispensing was done by physicians, physician assistants, and nurse practitioners. In some clinics, office staff had access to samples and were asked to consult with the physician before using these samples. Personal use of samples was documented in at least 4 practices, and in 2 of these the nurses were described as “helping themselves.” One particular clinic even voiced concerns about patients having unsupervised access and mentioned the need to develop a drug policy.

Dispensing patterns varied significantly among the clinics. Certain practices, especially those in rural, underserved communities, were found to dispense more liberally than others. These clinics appeared to be particularly sensitive to the needs of the individual patients and those of the greater community. In one practice, medication was handed out in 39.2% of encounters. On one occasion the physician waited until after the pharmaceutical representative had delivered an adequate supply of the necessary medication to give it to the patient. In contrast, clinicians in a particular suburban clinic only dispensed drugs 4.4% of the time.

However, even within a given practice, vast individual variances existed for each clinician in their approach to samples. In one clinic, 2 of the partners displayed disparate habits: One physician used samples regularly (in 41.9% of observed encounters), and the other only did so on rare occasions (in 3.2% of observed encounters). Of all the 18 sites, the physician who dispensed samples most frequently did so in 56.7% of observed encounters; the physician, however, at the other end of the spectrum dispensed medications in 2.4% of the observed patient encounters.

Of the 18 practices, 9 made use of specific closets or rooms for the storage of medication samples. One particular practice boasted an extraordinary and meticulously organized shelving system, complete with labeled bins and demarcated sections. The medical assistant in whose office these medications were stored was responsible for dispensing the ordered medications. In contrast to this efficient system, most clinics displayed very little structure in the organization and dispensing of samples. One clinic was described as having an overfilled, disorganized stack of shelves where medications were placed alongside vitamin supplements and herbal products. In this clinic staff spent vast amounts of time searching for the appropriate samples.

Only one clinic kept the sample room locked; another office had placed a notice on the door to keep it shut. It is important to note, however, that these storage places did not contain narcotics or similarly scheduled medications. In spite of Health Care Financing Administration regulations, we did not find any documentation to support which samples with which serial numbers had been distributed to patients, making effective recalls almost impossible. Stocks were most often replenished during routine drug representative visits. In one of the sites the head nurse kept in contact with the representatives and notified them when additional stocks were needed.

Benefits of These Interactions

We found that physicians and patients gained some advantage from their dealings with drug companies. The benefits the practices obtained from their contact with the pharmaceutical industry varied substantially. Other than medication samples, the most commonly observed benefits included the provision of meals, treats, and patient education materials [Table 2].

Patients also profited in a spectrum of ways. While samples represented tangible cost savings, immediate relief, and convenience to the individual patient (and occasionally to their family members), patient education materials facilitated further understanding of their diagnosis, potentially leading to a higher degree of satisfaction with their health care.

Discussion

Our study suggests that the relationship between the pharmaceutical industry and clinicians is symbiotic. It is a complex relationship that deserves objective analysis regarding its virtues and pitfalls.

Although it is essential to acknowledge the extent of the pharmaceutical industry’s influence, it is equally important to recognize and enhance the opportunities that clinicians have to direct this relationship. Our data suggest that clinics with existing policies for interactions with drug companies appear to derive more satisfaction and less frustration from their encounters. Formal meeting times eliminated unwanted interruptions and distractions from patient care and made the interactions more meaningful. Physicians who negotiated extensive and structured contact with drug representatives often received regular visits and generous supplies of both samples and patient education materials. Drug representative briefings, even though these needed to be viewed with healthy skepticism, also constituted a welcome form of CME for certain physicians.³¹

We found, however, that many offices do not have formal policies regarding pharmaceutical representatives and use questionable sample handling and storage practices. Interactions with pharmaceutical representatives can also be time consuming, potentially affecting the patient care schedule. Dispensing medications in the office setting may lead to increased responsibility and liability. We found that few clinicians provided extensive patient education about the samples they dispensed. Physicians need to be cognizant of the fact that they are bypassing the pharmacist. Directions to the patient need to be clear and comprehensive, addressing issues such as polypharmacy and drug-drug interactions. In addition, physicians should remain mindful of their own vulnerability with respect to the pharmaceutical industry’s influence over their prescribing habits. Drug detailing, even when insightful, may be biased.

Conclusions

The management of office visits with pharmaceutical representatives and the distribution of medication samples need to be addressed. Clinics with specific policies for interactions with drug companies appear to derive more satisfaction from their encounters. Practices should also maintain specific strategies when it comes to dispensing samples and should develop patterns of accountability for the handling of samples in the office.

Future research should include issues such as the impact of the pharmaceutical representative’s visit on patient finances, attitudes, disease management, and the physician-patient relationship. Further work is needed in the area of prescribing habits and the factors influencing physician decision making. More information on these topics may ultimately define and establish a relationship between the physician and the pharmaceutical industry that protects and promotes the interests of patients.

Acknowledgments

Our work was supported by a grant from the Agency for Health Care Policy and Research (1RO1 HS08776) to Benjamin Crabtree, PhD.

The world of medicine is permeated by the influence of the pharmaceutical industry, which spends an estimated $10 billion (more than $13,000 per physician) each year on drug promotion.¹ This industry’s influence extends from medical education to clinical research and patient care.² For example, medical journals and medical conferences are supported by pharmaceutical advertising or depend on financial contributions from pharmaceutical companies to offer their programs. Pharmaceutical companies grant a variety of scholarships and subsidies. The pharmaceutical industry also reinvests a portion of its profits to support new research and development, including hundreds of clinical trials reported each year in the scientific literature.³ Up-to-date drug detailing and continuing medical education (CME) is offered to clinicians, and patient education materials are frequently made available to practices. Patients, especially those with a lower income, benefit from receiving free medication samples donated by pharmaceutical companies.

Pharmaceutical companies, health care providers, and patients represent unique interests that may at times overlap. How the interactions between physicians and pharmaceutical representatives influence prescribing habits, decision making, and physician behavior in general is a contestable topic that has not been well studied or thoroughly understood. In the last several years, medical organizations in the United States and Canada have released discussion papers and policy statements about the relationship between the medical profession and the pharmaceutical industry.³ Whether the influence of the pharmaceutical industry is ultimately beneficial or detrimental remains a contentious and complex topic shrouded in much controversy.^4-15

The interactions between the pharmaceutical industry and physicians have been discussed in a plethora of publications. Several of these articles have focused on the ethical implications of these interactions and the potential for exploitation; however, they have commonly been based on surveys and self-reporting and have often been limited to providers in academic or residency training contexts.^3,16-21 Brotzman and Mark¹⁶ encourage the development of more comprehensive policies for residency training purposes. Kelcher and coworkers¹⁹ promote the concept of a structured educational program to prepare residents for future contact with the pharmaceutical industry. Lexchin⁹ discusses the need for critical comparison when evaluating drug information received from pharmaceutical representatives.

So how do family practice clinicians approach this challenging relationship? Through direct observation of these interactions within the private practice context, we attempted to add a new and vital dimension to the assessment of this rather complex symbiosis. We describe the approaches and reactions of individual family physicians and their practices to pharmaceutical representatives.

Methods

The Prevention and Competing Demands in Primary Care Practice study was designed to examine the organizational contexts that support or inhibit the delivery of preventive services in family medicine practices. Eighteen practices were studied using a multimethod ethnographic design that involved extensive observational field notes of the office system and the clinical encounters. These notes were collected by a researcher who spent at least 4 weeks in each practice. A total of 44 physicians and 9 other clinicians were shadowed, and approximately 1600 patient encounters were directly observed. Individual depth interviews were conducted with each clinician and many of the practice staff to obtain their perceptions.

Sampling

The comparative case study design began with an initial phase of purposefully selecting practices that included both urban and rural settings and those with different intensities of preventive services delivery, (n=10) followed by a sample to confirm or refute evolving hypotheses (n=8). The practices were selected from those that had participated in the Centers for Disease Control and Prevention/Nebraska Department of Health–funded Tobacco Use Prevention in Physicians Offices study (Helen McIlvain, principal investigator). These high-volume practices had a wide variation in practice location (urban, suburban, and rural) and organizational structure (solo vs group, independent vs system affiliated). Because we noted some interesting variations among practices from different health systems in the early analyses, the replication sample ensured at least 2 practices from each of 4 major health systems, and 2 from those practices that were seen as struggling to provide preventive services in the tobacco study.

We solicited study participation by sending an invitation letter followed by a telephone call to one of the physicians in the practice. Participation was very high, requiring contact with 23 practices to obtain the 18 deemed necessary in the study design.

Data Collection

We sent a field researcher trained in qualitative methods to each practice where she used a variety of data collection methods to produce a comprehensive picture of the practice as a functioning organization. Data were collected through direct observation of the practice,²² the use of structured checklists of the office environment,²³ direct observation of patient encounters supplemented with structured checklists, the use of patient pathways,²⁴ individual depth interviews with physicians and other key staff members,²⁵ and chart audits. It took from 4 to 12 weeks for the field researcher to complete the data collection in each practice, depending on its size.

The field researcher observed and jotted details of the physical environment and functioning of the practice, then dictated extensive field notes at the end of each day. These field notes contained detailed descriptions of the clinic location and environment, patient characteristics, nursing station, examination rooms, the waiting area, physician offices, bulletin boards, posters, and patient education materials. Photographs were taken of each room, including the reception area, waiting area, nurses’ station, nurse intake room, and the examination rooms. Existing practice personnel, their roles and duties, and their relationships and interactions with other staff members were characterized in a practice genogram.²⁶ Physical office systems, including charts, flow sheets, and computer systems were described, as were functional office routines and procedures. The field researcher specifically noted the storage and organization of drug samples within the clinic, comments made by staff relating to pharmaceutical representatives, and any observed interaction between staff and representatives.

The field researcher attempted to observe 30 patient encounters for each clinician; because of patient volume and scheduling this was not always possible. Thirty or more patient encounters were observed in 42 of the 53 clinicians participating in the study, and 20 or more patient encounters were observed for all but 1 provider (who had 18 encounters). After obtaining written informed consent from the patient, the field researcher shadowed the clinician and jotted notes about each encounter for later dictation. The patient encounter field notes contained rich descriptions about any verbal patient education delivered and the context of that education, including the reason for the visit, how the visit unfolded, and how the clinician and patient interacted. The patient encounter dictation captured the distribution of drug samples, discussion between the clinician and patient regarding drug samples, prescriptions written by physicians, and instructions given by physicians.

Using the depth interviewing technique, the field researcher interviewed all family physicians within the practice, the office manager, head nurse, and other significant persons within the practice.²⁵ Questions asked by the field researcher related to overall views of medicine, general practice characteristics, the systems in place for the delivery of preventive medicine, and perceived rates of delivering preventive medicine. These interviews were tape-recorded and transcribed verbatim.

Additional data such as photographs and chart audits were collected as part of the larger study but were not included in our analysis.

Data Analysis and Interpretation

Qualitative data were transcribed, checked for accuracy by having the field worker review the manuscript, and entered into FolioViews 4.11, an infobase software package.²⁷ We began our analysis by immersing ourselves in all the data from 2 practices to understand the function of the practices and the overall richness and variation of the data collected, and to identify any mention of drug samples or interactions with pharmaceutical representatives. Using this immersion/crystallization approach, we worked to formulate hypotheses and an initial organizational scheme.²⁸ Group discussions led to the development of a codebook of key words.

Two investigators then performed computer word searches on the infobases of the 2 previously selected practices using the codebook. All relevant passages were tagged and printed, and the information retrieved was compared with data identified from an earlier in-depth reading of the transcripts from the 2 practices we initially studied. The word search technique and template organizing style was used to identify relevant portions of 4 more practices for secondary data analysis.²⁹

Finally, segments of the selected field notes and patient encounter notes that described drug samples and pharmaceutical representatives were identified and organized into preliminary tables or matrices.³⁰ After constructing matrices for the 6 selected practices we held discussions and further refined the organization and categories within the matrices. Once the matrices were refined they were used to organize relevant data identified by word searches for the rest of the 18 practices in the sample. The matrices served as the framework for data analysis, allowed the visualization of emergent patterns, and facilitated comparisons across physicians and practices.

Results

Individual clinicians and their practices displayed noticeable variation in their approaches to pharmaceutical representatives and the use of samples. Our analysis revealed patterns with respect to the types of contact they have with the pharmaceutical representatives, their use of sample medication, and the relative benefits obtained from the interactions.

Contact with the Pharmaceutical Representatives

All 18 practices had some form of contact with the pharmaceutical industry. This contact ranged from scheduled and well-organized meeting times to random interactions. Formal strategies and policies regarding drug representative interactions and the use of samples were in place in 8 of the 18 practices (44%). One other practice voiced a concern regarding the need for such a policy.

Most of the 8 clinics that had a formal plan or policy preferred a specific time (such as the lunch hour) during which the physicians and staff could meet with the representative. Some practices determined which days and times would be most convenient for their staff. One clinic made the clinician’s schedule available, thus allowing the representatives to schedule visits accordingly. One rural physician even allotted 30-minute patient appointment times to each representative to optimize the quality of these interactions. These lengthy visits facilitated more extensive drug detailing and CME, and generated an ample supply of samples, which were used to serve the needs of this particular rural community. A receptionist in another clinic was found to simply use a chalkboard to notify the staff whenever a representative had set up a presentation in the office. All of these measures appeared to be effective in minimizing confusion, distraction, and schedule disruption.

The remaining 10 practices displayed more haphazard dealings with drug representatives. These included brief hallway meetings that happened when clinicians were confronted by a representative as they exited the examination room. Other pharmaceutical representatives waited in medication sample rooms, hoping to catch the physicians as they searched for samples. These casual relationships were, at times, counterproductive. In one practice the physicians seemed oblivious to when “drug lunches” were scheduled, leading to frustration on the part of the pharmaceutical representatives; in another, the physician found herself rudely interrupted by an imposing representative during her own lunch.

Usage and Storage of Sample Medication

In the 18 practices, medication samples were used in 19.8% of the 1588 observed patient encounters. Multiple drugs were dispensed in 14.6% of the encounters in which samples were used. In only 5.1% of cases was a medication dispensed as a result of a patient’s specific request; the clinician usually initiated the distribution of samples. Drug samples were also offered to additional family members in 3.5% of encounters, sometimes even when these individuals were not accompanying the patient to the given appointment. This appeared to be particularly common in certain rural practices where the clinicians seemed sensitive to the needs of their patients and their families.

A review of the types of medication dispensed revealed noticeable trends and pointed to a prevalence of certain medication categories [Table 1]. The top 4 categories included asthma and allergy remedies, anti-infective agents, analgesics and anti-inflammatory medications, and antihypertensive drugs. These major categories accounted for more than 63% of all drugs dispensed.

The duration for which these medication samples were given varied from starter dosages (lasting 1-3 days) to complete courses of treatments (eg, a course of antibiotics) to amounts sufficient to supply the patients’ needs for several months (eg, antidepressants). The analysis of the clinician’s intent when dispensing samples identified that some were using samples to test for efficacy and tolerability while others were attempting to offer temporary relief or convenience to the patient. Also, certain physicians were clearly concerned about cost savings for their patients. The use of medication samples could be seen to represent tangible benefits to the individual patient.

Instructions accompanied the dispensing of samples in 150 (47.8%) of the cases and were predominantly verbal in nature. Dosing appeared to be the main focus, and little to no attention was given to more detailed information, such as whether the medication should be taken with meals. Drug interactions were not routinely discussed.

The majority of dispensing was done by physicians, physician assistants, and nurse practitioners. In some clinics, office staff had access to samples and were asked to consult with the physician before using these samples. Personal use of samples was documented in at least 4 practices, and in 2 of these the nurses were described as “helping themselves.” One particular clinic even voiced concerns about patients having unsupervised access and mentioned the need to develop a drug policy.

Dispensing patterns varied significantly among the clinics. Certain practices, especially those in rural, underserved communities, were found to dispense more liberally than others. These clinics appeared to be particularly sensitive to the needs of the individual patients and those of the greater community. In one practice, medication was handed out in 39.2% of encounters. On one occasion the physician waited until after the pharmaceutical representative had delivered an adequate supply of the necessary medication to give it to the patient. In contrast, clinicians in a particular suburban clinic only dispensed drugs 4.4% of the time.

However, even within a given practice, vast individual variances existed for each clinician in their approach to samples. In one clinic, 2 of the partners displayed disparate habits: One physician used samples regularly (in 41.9% of observed encounters), and the other only did so on rare occasions (in 3.2% of observed encounters). Of all the 18 sites, the physician who dispensed samples most frequently did so in 56.7% of observed encounters; the physician, however, at the other end of the spectrum dispensed medications in 2.4% of the observed patient encounters.

Of the 18 practices, 9 made use of specific closets or rooms for the storage of medication samples. One particular practice boasted an extraordinary and meticulously organized shelving system, complete with labeled bins and demarcated sections. The medical assistant in whose office these medications were stored was responsible for dispensing the ordered medications. In contrast to this efficient system, most clinics displayed very little structure in the organization and dispensing of samples. One clinic was described as having an overfilled, disorganized stack of shelves where medications were placed alongside vitamin supplements and herbal products. In this clinic staff spent vast amounts of time searching for the appropriate samples.

Only one clinic kept the sample room locked; another office had placed a notice on the door to keep it shut. It is important to note, however, that these storage places did not contain narcotics or similarly scheduled medications. In spite of Health Care Financing Administration regulations, we did not find any documentation to support which samples with which serial numbers had been distributed to patients, making effective recalls almost impossible. Stocks were most often replenished during routine drug representative visits. In one of the sites the head nurse kept in contact with the representatives and notified them when additional stocks were needed.

Benefits of These Interactions

We found that physicians and patients gained some advantage from their dealings with drug companies. The benefits the practices obtained from their contact with the pharmaceutical industry varied substantially. Other than medication samples, the most commonly observed benefits included the provision of meals, treats, and patient education materials [Table 2].

Patients also profited in a spectrum of ways. While samples represented tangible cost savings, immediate relief, and convenience to the individual patient (and occasionally to their family members), patient education materials facilitated further understanding of their diagnosis, potentially leading to a higher degree of satisfaction with their health care.

Discussion

Our study suggests that the relationship between the pharmaceutical industry and clinicians is symbiotic. It is a complex relationship that deserves objective analysis regarding its virtues and pitfalls.

Although it is essential to acknowledge the extent of the pharmaceutical industry’s influence, it is equally important to recognize and enhance the opportunities that clinicians have to direct this relationship. Our data suggest that clinics with existing policies for interactions with drug companies appear to derive more satisfaction and less frustration from their encounters. Formal meeting times eliminated unwanted interruptions and distractions from patient care and made the interactions more meaningful. Physicians who negotiated extensive and structured contact with drug representatives often received regular visits and generous supplies of both samples and patient education materials. Drug representative briefings, even though these needed to be viewed with healthy skepticism, also constituted a welcome form of CME for certain physicians.³¹

We found, however, that many offices do not have formal policies regarding pharmaceutical representatives and use questionable sample handling and storage practices. Interactions with pharmaceutical representatives can also be time consuming, potentially affecting the patient care schedule. Dispensing medications in the office setting may lead to increased responsibility and liability. We found that few clinicians provided extensive patient education about the samples they dispensed. Physicians need to be cognizant of the fact that they are bypassing the pharmacist. Directions to the patient need to be clear and comprehensive, addressing issues such as polypharmacy and drug-drug interactions. In addition, physicians should remain mindful of their own vulnerability with respect to the pharmaceutical industry’s influence over their prescribing habits. Drug detailing, even when insightful, may be biased.

Conclusions

The management of office visits with pharmaceutical representatives and the distribution of medication samples need to be addressed. Clinics with specific policies for interactions with drug companies appear to derive more satisfaction from their encounters. Practices should also maintain specific strategies when it comes to dispensing samples and should develop patterns of accountability for the handling of samples in the office.

Future research should include issues such as the impact of the pharmaceutical representative’s visit on patient finances, attitudes, disease management, and the physician-patient relationship. Further work is needed in the area of prescribing habits and the factors influencing physician decision making. More information on these topics may ultimately define and establish a relationship between the physician and the pharmaceutical industry that protects and promotes the interests of patients.

Acknowledgments

Our work was supported by a grant from the Agency for Health Care Policy and Research (1RO1 HS08776) to Benjamin Crabtree, PhD.

Being patient centered is a core value of medicine for many physicians. The principles of patient-centered medicine date back to the ancient Greek school of Cos, which was interested in the particulars of each patient.¹ More recently similar concepts have arisen in a variety of fields of human endeavor: the concept of physical diagnosis and deeper diagnosis of Balint,² the client-centered therapy of Rogers,³ the total-person approach to patient problems in nursing of Neuman and Young,⁴ the biopsychosocial model of Engel,⁵ and the disease- versus patient-centered medical practice of Byrne and Long.⁶ In the past decade the patient-centered concepts of Gerteis and colleagues⁷ have been applied to the hospital setting.

In the setting of primary care, and specifically family practice, patient-centered concepts incorporate 6 interactive components. The first component is the physician’s exploration of both the patients’ disease and 4 dimensions of the illness experience including: their feelings about being ill, their ideas about what is wrong with them, the impact of the problem on their daily functioning, and their expectations of what should be done. The second component is the physician’s understanding of the whole person. The third component is the patient and physician finding common ground regarding management. In the fourth component the physician incorporates prevention and health promotion into the visit. The fifth component is the enhancement of the patient-physician relationship. Finally, the sixth component requires that patient-centered practice be realistic. Our study addresses the first 3 of these components. Being patient centered does not mean that physicians abdicate control to the patient⁸ but rather that they find common ground in understanding the patients and more fully respond to their unique needs.⁹

What are the benefits of being patient centered? Previous research of specific communication variables indicates that patient-centered encounters result in: (1) the duration of the office visit remaining the same^10,11 (2) better patient satisfaction,¹² (3) higher physician satisfaction,¹⁰ and (4) fewer malpractice complaints.¹³ We focus on 2 other outcomes: patients’ health and efficiency of care.

Methods

Our study was designed to test the hypothesis that adult patients whose first visit in an episode of illness is patient centered will, by 2 months after the first visit: (1) more frequently demonstrate recovery from the symptom (and recovery from the concern about the symptom); (2) demonstrate better self-reported health; and (3) experience less subsequent medical care (ie, fewer visits, diagnostic tests, and referrals), compared with patients whose visit is not patient centered.

Data Collection and Participants

For our observational cohort study data were collected at 5 points: (1) the research assistant identified eligible patients in the physician’s office before the visit; (2) the office encounter was audiotaped and scored for patient-centered communication; (3) the research assistant held a postencounter interview with the patient; (4) we assessed, by chart review, the use of medical care during the 2-month follow-up; and (5) we conducted a follow up telephone interview with patients 2-months after the encounter.

Physician Selection. Physicians were recruited from the 250 family physicians practicing in London, Ontario, Canada, and the surrounding area. They were randomized within strata to ensure a representative sample in terms of year of graduation and geographic location and were selected using a modified version of the method of Borgiel and colleagues.¹⁴

Patients. We approached patients who were older than 18 years and had 1 or more recurring problems who presented to their physician’s office. Patients were excluded if they were too ill or disabled to answer questions, had no presenting problem, were in the office for counseling, were accompanied by another person, were not fluent in English, were hard of hearing, or were cognitively impaired. They were approached before they saw the physician and were blind to the study hypotheses.

Sample Size Estimation. The sample size required for correlations of 0.20 to be detected with an a set at 0.05 (2 tailed) and a b set at 0.10 was 259¹⁵ patients. Further inflation by 10% to account for the effect of clustering on multiple regression¹⁶ was thought to be reasonable (259÷0.9=288). Expecting 75% to cooperate, we aimed to approach 384 patients (288÷0.75).

Measures*

Measure of Patient-Centered Communication Score. The patient-centered communication score is based on 3 of the 6 components of the model of patient-centered medicine.^17-20 The first component (exploring the disease and the illness experience) received a high score when the physician explored the patients’ symptoms, prompts, feelings, ideas, function, and expectations. The second component (understanding the whole person) received a high score when the physician elicited and explored issues relating to life cycle, personality, or life context, including family. The third component (finding common ground) received a high score when the physician clearly described the problem and the management plan, answered questions about them, and discussed and agreed on them with the patient. Scoring sheets and procedures are described in detail elsewhere.²¹ Scores could range from 0 (not at all patient centered) to 100 (very patient centered).

Interrater reliability has been established in earlier versions of the measure and for the current version (r=0.69, 0.84, and 0.80 among 3 raters,²² 0.91 among 2 raters,²³ and 0.83 for n=19 for our study). Intrarater reliability was 0.73 (n=20).

Correlations with global scores encompassing the 3 components supported the validity of the score (0.63 in an earlier study²³ and 0.85 for our study, n=46).

Patient Perception of Patient-Centeredness. Based on the patient-centered model, a series of 14 items developed and validated in previous studies^24,25 were used to assess the patients’ postencounter perceptions of how patient centered the interaction with the physician had been.† Items were averaged into: total score, a subscore on exploring the disease and illness experience, and finding common ground. Low scores represented patient centeredness.

Patient Recovery from Discomfort and Concerns. The primary health outcome was the recovery measure based on the patients’ self-administered report on visual analogue scales (VAS) of the severity of the symptom they identified as the main presenting problem and their concern about that problem at 2 points: the postencounter interview and the follow-up 2 months later.^26,27 VAS have been tested for reliability and validity in studies of pain and nausea (correlation of 0.75 with an intensity score).²⁶ Each of the symptom recovery variables was continuous.

Patient Health Status. The Medical Outcomes Study Short Form-36 (SF-36) was used to assess self-reported secondary health outcomes. This valid and reliable measure¹⁸ is a multidimensional assessment of: physical health, mental health, perception of health, social health, pain, and role function. All were continuous variables except role function, for which the distribution of scores necessitated dichotomizing.

Medical Care. The care provided during the 2 months following the audiotaped encounter was assessed by chart review (adapted from Bass and coworkers²⁴) by 3 medical doctors (I.R.M., J.O., J.J.) blind to the identity of the family physician and the patient, and also to the patient-centered scores. Items abstracted were: the total number of visits during the 2 months (continuous variable); the number and kind of diagnostic tests ordered during the 2 months that were relevant to the problems presented at the audiotaped visit (dichotomous); and the number and kinds of referrals made during the 2 months that were relevant to the problems presented at the audiotaped visit (dichotomous).

Analysis. The hypotheses were tested using multiple regression for continuous outcomes and multiple logistic regression for dichotomous outcomes,²⁹ both adjusted for the effect of the clustering of patients by physician using “procedure mixed” in SAS for continuous outcomes and using both “procedure logistic” and “procedure IML” in SAS for dichotomous outcomes.³⁰ The unit of analysis was the patient.

The following confounding variables were included in preliminary multivariable analyses on the basis of their univariable relationships with outcomes at the level of P <.10: age, sex, number of family members at home, desire to share feelings, who initiated the visit, tense personality, coping skills, concomitant health problems, social support, marital status (married vs other), concomitant life problems, number of visits to the physician in the previous 12 months, and main problem (1 of 5 groups: digestive, musculoskeletal, respiratory, skin, and other).

Because of substantial sample attrition with so many covariates, and because only 2 variables were consistently associated with the outcome measures, each subsequent multivariable analysis was conducted with each of the primary independent variables and the 2 covariates (patients’ main presenting problem and marital status).

Results

Descriptive Results

Of the 102 randomly selected family physicians, 83 were eligible because they were still practicing in the area and had adequate office space to accommodate the research assistant. Of these, 39 (47%) agreed to participate and completed the data collection. The participants were similar to the refusers [Table 1] in year of graduation, practice location (rural or urban; high or low socioeconomic status) and sex; however, participants were significantly more likely to be certificants of the College of Family Physicians of Canada than refusers (59% and 27%, respectively; P=.007).

Of 464 eligible patients, 334 (72%) agreed to participate. Nineteen (~6%) were lost to the study. The final 315 participants represented an overall participation rate of 68%; their age was representative of the eligible patients, but there was a higher proportion of men than in the total group of eligible patients.

[Table 2] shows that the slim majority of final participants were women, and most were middle aged and married. Typical of the city, approximately 4 in 10 had more than a high school education. The most common presenting problems were respiratory in nature.

[Table 3] shows the descriptive results for key variables.

Hypothesis Testing Results

The patient-centered communication scores (based on the audiotape analysis) were not significantly related to any of the health outcomes after adjusting for the clustering of patients within practices and after controlling for the 2 confounding variables. Similarly, patient-centered communication scores were not related to any of the 3 medical care outcomes.

Patient-centered communication scores (based on the audiotape analysis) were significantly correlated in the expected direction, with patient perceptions that the patient and physician found common ground (r =-0.16; P=.01). High scores (indicating very patient-centered communication) were correlated with low patient perception scores (indicating patient-centeredness). The 2 other patient perception scores (total patient perception score and the subscore on patient perception that the illness experience was explored) were not significantly associated with patient-centered communication scores.

The total score of patients’ perceptions that the visit was patient centered was associated with positive health outcomes after adjusting for the clustering of patients within practices and after controlling for the 2 confounding variables [Table 4]. Patients’ postencounter levels of discomfort were lower when they perceived the visit to have been patient centered than not.

A similar result occurred for 2 other patient health outcomes: the patients’ postencounter level of concern (P=.02), and the mental health dimension of the SF-36 measure assessed 2 months after the study visit (P=.05). The subscore of patient perceptions that the patient and physician found common ground was associated with one of the health outcomes, the patients’ postencounter level of concern (P=.04). There were no significant associations of the subscore on patients’ perceptions that the illness experience had been explored with any of the patient health outcome measures.

Patients who perceived that their visit had been patient centered received fewer diagnostic tests [Table 5] and referrals [Table 6] in the subsequent 2 months. The proportion receiving diagnostic tests rose from 14.6% in the group who perceived that the visit had been patient centered (total score), to 24.3% in the group who perceived the visit was not. The proportion who were referred doubled from approximately 8% to 16%. These relationships were found even more strongly for the subscore on patient perceptions that the patient and the physician found common ground, but were not found for the subscore on patient perceptions that their illness experience had been explored. The proportion receiving diagnostic tests quadrupled from 4.1% in the group who perceived that the patient and the physician found common ground, to 25.4% in the group who perceived that common ground had not been attained. The proportion who were referred doubled from 6.1% to 14.9%. The number of visits by the patient to the family physician during the subsequent 2 months was not significantly related to the patient perceptions of patient centeredness, although there was a trend (P=.11) with the average number of visits in 2 months in the 4 quartiles of patient perceptions as follows: 1.0, 0.8, 1.2, and 1.3.

Discussion

Pathway to Improved Patient Outcomes

Patient-centered practice was associated with improved patients’ health status and increased efficiency of care (reduced diagnostic tests and referrals). However, only 1 of the 2 measures of patient-centered practice showed this result, the measure of patients’ perceptions of the patient centeredness of the visit. The measure that was based on ratings of audiotaped physician-patient interactions, while related to the patients’ perception, was not directly related to health status or efficiency.

The relationship of patients’ perceptions of patient centeredness with their health and efficiency of care was both statistically and clinically significant. Specifically, recovery was improved by 6 points on a 100-point scale; diagnostic tests and referrals were half as frequent if the visit was perceived to be patient centered.

The associations we found may imply a potentially important pathway (which could be tested in future trials), such as the one shown in the [Figure]. The pathway suggests a process through which patient-physician communication influences patients’ health, by first influencing the patients’ perceptions of being a full participant in the discussions during the encounter. Such a pathway has been noted by Sobel, whose review suggested a pathway to explain the lack of a direct relationship between patient education programs and patient health where there was a relationship between patient perceptions about their health and health outcomes. Sobel called this pathway “a biology of self-confidence.”³¹ He and others³² stress the critical role of patient perceptions in the healing process, which highlight that a person’s subjective experience influences biology.

How do we understand the results that show the ratings of the audiotape were not directly related to the outcomes, but the patient-centered perception measure was related to outcomes? One interpretation is that observable skills are not as important as patient perceptions. Although there is some evidence that skills training can improve both physicians’ behavior and patients’ health,³³ our findings and those of Bensing and Sluijs³⁴ indicate that differences in interviewing skills may not be associated with patient responses. Physicians may learn to go through the motions of patient-centered interviewing without understanding what it means to be a truly attentive and responsive listener. The implications of the current findings for educators are that education about communication should go well beyond skills training to a deeper understanding of what it means to be a responsive partner for the patient, during both that phase of the visit in which the problem is discussed and when the discussion of treatment options occurs. Two examples of such education approaches are: small group discussions between patients and physicians to illustrate the patients’ experiences and needs, and reviews of videotaped interviews with standardized patients participating in the review. Placing prime importance on the patients’ perceptions recognizes the influence of these perceptions on the patients’ subsequent health and epitomizes being truly patient centered.

Views that the visit was patient centered included perceptions about the discussion of the problem (exploring the illness experience) as well as discussion and agreement about treatment options (finding common ground). There is a substantial body of research supporting the importance of these discussions. The Headache Study found that patients’ perceptions that a full discussion of the problem had taken place predicted resolution of headaches after 1 year.³⁴ In keeping with our results, which found that finding common ground was more strongly associated with outcomes than exploring the illness experience, Riccardi and Kurtz³⁶ stressed that the physicians’ explanation to the patients was the crucial phase of the visit. Also, a key outcome study has found that patient agreement with the physician about the nature of the treatment and the need for follow-up were strongly associated with their recovery.³⁷

Efficiency of Medical Care

We found that patient-centered practice (assessed by patients’ perceptions) was associated with the efficiency of care by reducing subsequent diagnostic tests and referrals by half, after controlling for key confounding variables. These results were both statistically significant and clinically significant. Also, the number of subsequent visits to the family physician was lower (although not significantly) when the patient perceived the study visit to be patient centered. Efficiency in health service delivery was also found in a randomized trial of compassionate care in the emergency department setting with homeless patients.³⁸ In their study of continuity of care in Norwegian general practice, Hjortdahl and Borchgrevink³⁹ found that diagnostic tests were 10 times more likely to be ordered for patients about whom physicians reported the least previous knowledge compared with patients in whom they had reported fullest knowledge. Also, patients had only half the chance of being referred if their physicians knew them and their history.³⁸

One possible interpretation of the results of our study is that patient-centered physicians order fewer tests and refer less often. However, countering this interpretation is the fact that individual physicians in our study showed a range of patient-centered scores, as well as a range in test ordering and referral. In addition, the statistical analysis took account of the clustering of patients within a physician’s practice.

An alternative interpretation is that patients’ perceptions may influence resource use in several ways. For example, increased participation during the visit may reduce patients’ anxiety and their perceived need for investigations and referrals. Alternatively, patients’ perception that the physician has not understood their problem may provoke insecurities resulting in a request for further medical interventions. Also, if patients openly express their discontent with the encounter there may be an increase in physicians’ anxiety and a lowering of their threshold for diagnostic uncertainty, resulting in further investigations and referrals.

Certainly the finding that the failure to be patient centered (as perceived by the patient) was related to higher rates of referral and diagnostic tests should be a concern for medical education and health care policy. Perhaps of most importance is that the patients’ experience of being a participating member in the discussion of the problem and the treatment process may translate into the patients’ reduced need for further investigation or referral—simultaneously reducing the physicians’ need as well.

These findings counter a common misconception: that being patient centered means responding to every whim of the patient, thereby increasing expenses to the health care system.

Limitations

Approximately 30% of the patients refused to participate, and although the participants represented the age distribution of eligible patients, men were overrepresented in the study. Nonetheless, sex was not identified as a confounding variable for the associations studied.

Although no measure of severity was possible, the variables representing concurrent health problems and concurrent life problems were considered in the analysis strategy. They were not related to the outcome variables and were therefore not entered into the multivariable analyses.

One interpretation of the lack of association between patient-centered scores on the audiotaped interviews and subsequent health outcomes may be that the audiotape measure has failed to capture the important essence of the dynamic interaction between physicians and patients. The measure had a number of strengths, however; it had been tested for reliability and validity (compared with a global rating), and it was based on a theoretical framework. Also, it was correlated with one component of the patient perception measure of a patient-centered interview, a finding which indicates that future research should be directed toward determining physicians’ skills and behaviors that correlate with the patients’ positive perceptions, especially the perception that common ground has been reached. Such behaviors could then be emphasized in clinical teaching.

It should be noted that the utilization data were available only from the participating practices and not from care received elsewhere. Although this is a limitation, it would be expected that this lack of data would minimize the current relationship between patient-centered practice and utilization, because patients with less favorable perceptions would be potentially more likely to seek care elsewhere. Also, drug costs and hospital costs were not included and require further study. Future research could also build on these results about resource utilization and assess the specific kinds and actual costs of the diagnostic tests and referrals.

It could be argued that the results of our study demonstrated simply that people with positive perceptions and less severe problems achieved better health and more efficient services. We counter this interpretation with 2 thoughts. First, the preliminary step in our analysis included confounding variables to control for a variety of relevant variables (ie, personality and concomitant health problems). Only 2 confounding variables were influential enough to remain in the final analysis: marital status and diagnostic code of the main presenting problem. Second, patient perceptions were not independent of the physician-patient visit. They were influenced significantly by the communication score based on the audiotaped encounter, implying that the measure of perceptions was tapping not merely the patients’ general outlook on life, but also an important interactive component of visits between patients and physicians.

Conclusions

Patient-centered practice was associated with improved health status (less discomfort, less concern, and better mental health) and increased efficiency of care (fewer diagnostic tests and referrals).

Patients’ perceptions of the patient centeredness of the visit, but not the measure of audiotaped interactions, were directly associated with the positive outcomes. The subscore on patients’ perception of finding common ground was more strongly associated with the positive outcomes than the subscore on patients’ perception about exploring the illness experience.

Medical education should go beyond skills training to encourage physicians’ responsiveness to the patients’ unique experience. Therefore, involving real patients and standardized patients in teaching programs is recommended.

Health service organizations must recognize that efficiencies accrue from patient-centered practice and encourage such practice through structures that enhance continuity of the patient-physician relationship and through meaningful education programs.

Acknowledgments

Our project was supported by a grant from the Health Care Systems Research Program of the Ministry of Health of Ontario. The setting of the project was the Thames Valley Family Practice Research Unit (TVFPRU), a health system-linked research unit funded by the Ministry of Health of Ontario. The opinions contained are those of the authors, and no official endorsement by the Ministry is intended or should be inferred. The TVFPRU is part of the Centre for Studies in Family Medicine, Department of Family Medicine, The University of Western Ontario, London, Canada.

Being patient centered is a core value of medicine for many physicians. The principles of patient-centered medicine date back to the ancient Greek school of Cos, which was interested in the particulars of each patient.¹ More recently similar concepts have arisen in a variety of fields of human endeavor: the concept of physical diagnosis and deeper diagnosis of Balint,² the client-centered therapy of Rogers,³ the total-person approach to patient problems in nursing of Neuman and Young,⁴ the biopsychosocial model of Engel,⁵ and the disease- versus patient-centered medical practice of Byrne and Long.⁶ In the past decade the patient-centered concepts of Gerteis and colleagues⁷ have been applied to the hospital setting.

In the setting of primary care, and specifically family practice, patient-centered concepts incorporate 6 interactive components. The first component is the physician’s exploration of both the patients’ disease and 4 dimensions of the illness experience including: their feelings about being ill, their ideas about what is wrong with them, the impact of the problem on their daily functioning, and their expectations of what should be done. The second component is the physician’s understanding of the whole person. The third component is the patient and physician finding common ground regarding management. In the fourth component the physician incorporates prevention and health promotion into the visit. The fifth component is the enhancement of the patient-physician relationship. Finally, the sixth component requires that patient-centered practice be realistic. Our study addresses the first 3 of these components. Being patient centered does not mean that physicians abdicate control to the patient⁸ but rather that they find common ground in understanding the patients and more fully respond to their unique needs.⁹

What are the benefits of being patient centered? Previous research of specific communication variables indicates that patient-centered encounters result in: (1) the duration of the office visit remaining the same^10,11 (2) better patient satisfaction,¹² (3) higher physician satisfaction,¹⁰ and (4) fewer malpractice complaints.¹³ We focus on 2 other outcomes: patients’ health and efficiency of care.

Methods

Our study was designed to test the hypothesis that adult patients whose first visit in an episode of illness is patient centered will, by 2 months after the first visit: (1) more frequently demonstrate recovery from the symptom (and recovery from the concern about the symptom); (2) demonstrate better self-reported health; and (3) experience less subsequent medical care (ie, fewer visits, diagnostic tests, and referrals), compared with patients whose visit is not patient centered.

Data Collection and Participants

For our observational cohort study data were collected at 5 points: (1) the research assistant identified eligible patients in the physician’s office before the visit; (2) the office encounter was audiotaped and scored for patient-centered communication; (3) the research assistant held a postencounter interview with the patient; (4) we assessed, by chart review, the use of medical care during the 2-month follow-up; and (5) we conducted a follow up telephone interview with patients 2-months after the encounter.

Physician Selection. Physicians were recruited from the 250 family physicians practicing in London, Ontario, Canada, and the surrounding area. They were randomized within strata to ensure a representative sample in terms of year of graduation and geographic location and were selected using a modified version of the method of Borgiel and colleagues.¹⁴

Patients. We approached patients who were older than 18 years and had 1 or more recurring problems who presented to their physician’s office. Patients were excluded if they were too ill or disabled to answer questions, had no presenting problem, were in the office for counseling, were accompanied by another person, were not fluent in English, were hard of hearing, or were cognitively impaired. They were approached before they saw the physician and were blind to the study hypotheses.

Sample Size Estimation. The sample size required for correlations of 0.20 to be detected with an a set at 0.05 (2 tailed) and a b set at 0.10 was 259¹⁵ patients. Further inflation by 10% to account for the effect of clustering on multiple regression¹⁶ was thought to be reasonable (259÷0.9=288). Expecting 75% to cooperate, we aimed to approach 384 patients (288÷0.75).

Measures*

Measure of Patient-Centered Communication Score. The patient-centered communication score is based on 3 of the 6 components of the model of patient-centered medicine.^17-20 The first component (exploring the disease and the illness experience) received a high score when the physician explored the patients’ symptoms, prompts, feelings, ideas, function, and expectations. The second component (understanding the whole person) received a high score when the physician elicited and explored issues relating to life cycle, personality, or life context, including family. The third component (finding common ground) received a high score when the physician clearly described the problem and the management plan, answered questions about them, and discussed and agreed on them with the patient. Scoring sheets and procedures are described in detail elsewhere.²¹ Scores could range from 0 (not at all patient centered) to 100 (very patient centered).

Interrater reliability has been established in earlier versions of the measure and for the current version (r=0.69, 0.84, and 0.80 among 3 raters,²² 0.91 among 2 raters,²³ and 0.83 for n=19 for our study). Intrarater reliability was 0.73 (n=20).

Correlations with global scores encompassing the 3 components supported the validity of the score (0.63 in an earlier study²³ and 0.85 for our study, n=46).

Patient Perception of Patient-Centeredness. Based on the patient-centered model, a series of 14 items developed and validated in previous studies^24,25 were used to assess the patients’ postencounter perceptions of how patient centered the interaction with the physician had been.† Items were averaged into: total score, a subscore on exploring the disease and illness experience, and finding common ground. Low scores represented patient centeredness.

Patient Recovery from Discomfort and Concerns. The primary health outcome was the recovery measure based on the patients’ self-administered report on visual analogue scales (VAS) of the severity of the symptom they identified as the main presenting problem and their concern about that problem at 2 points: the postencounter interview and the follow-up 2 months later.^26,27 VAS have been tested for reliability and validity in studies of pain and nausea (correlation of 0.75 with an intensity score).²⁶ Each of the symptom recovery variables was continuous.

Patient Health Status. The Medical Outcomes Study Short Form-36 (SF-36) was used to assess self-reported secondary health outcomes. This valid and reliable measure¹⁸ is a multidimensional assessment of: physical health, mental health, perception of health, social health, pain, and role function. All were continuous variables except role function, for which the distribution of scores necessitated dichotomizing.

Medical Care. The care provided during the 2 months following the audiotaped encounter was assessed by chart review (adapted from Bass and coworkers²⁴) by 3 medical doctors (I.R.M., J.O., J.J.) blind to the identity of the family physician and the patient, and also to the patient-centered scores. Items abstracted were: the total number of visits during the 2 months (continuous variable); the number and kind of diagnostic tests ordered during the 2 months that were relevant to the problems presented at the audiotaped visit (dichotomous); and the number and kinds of referrals made during the 2 months that were relevant to the problems presented at the audiotaped visit (dichotomous).

Analysis. The hypotheses were tested using multiple regression for continuous outcomes and multiple logistic regression for dichotomous outcomes,²⁹ both adjusted for the effect of the clustering of patients by physician using “procedure mixed” in SAS for continuous outcomes and using both “procedure logistic” and “procedure IML” in SAS for dichotomous outcomes.³⁰ The unit of analysis was the patient.

The following confounding variables were included in preliminary multivariable analyses on the basis of their univariable relationships with outcomes at the level of P <.10: age, sex, number of family members at home, desire to share feelings, who initiated the visit, tense personality, coping skills, concomitant health problems, social support, marital status (married vs other), concomitant life problems, number of visits to the physician in the previous 12 months, and main problem (1 of 5 groups: digestive, musculoskeletal, respiratory, skin, and other).

Because of substantial sample attrition with so many covariates, and because only 2 variables were consistently associated with the outcome measures, each subsequent multivariable analysis was conducted with each of the primary independent variables and the 2 covariates (patients’ main presenting problem and marital status).

Results

Descriptive Results

Of the 102 randomly selected family physicians, 83 were eligible because they were still practicing in the area and had adequate office space to accommodate the research assistant. Of these, 39 (47%) agreed to participate and completed the data collection. The participants were similar to the refusers [Table 1] in year of graduation, practice location (rural or urban; high or low socioeconomic status) and sex; however, participants were significantly more likely to be certificants of the College of Family Physicians of Canada than refusers (59% and 27%, respectively; P=.007).

Of 464 eligible patients, 334 (72%) agreed to participate. Nineteen (~6%) were lost to the study. The final 315 participants represented an overall participation rate of 68%; their age was representative of the eligible patients, but there was a higher proportion of men than in the total group of eligible patients.

[Table 2] shows that the slim majority of final participants were women, and most were middle aged and married. Typical of the city, approximately 4 in 10 had more than a high school education. The most common presenting problems were respiratory in nature.

[Table 3] shows the descriptive results for key variables.

Hypothesis Testing Results

The patient-centered communication scores (based on the audiotape analysis) were not significantly related to any of the health outcomes after adjusting for the clustering of patients within practices and after controlling for the 2 confounding variables. Similarly, patient-centered communication scores were not related to any of the 3 medical care outcomes.

Patient-centered communication scores (based on the audiotape analysis) were significantly correlated in the expected direction, with patient perceptions that the patient and physician found common ground (r =-0.16; P=.01). High scores (indicating very patient-centered communication) were correlated with low patient perception scores (indicating patient-centeredness). The 2 other patient perception scores (total patient perception score and the subscore on patient perception that the illness experience was explored) were not significantly associated with patient-centered communication scores.

The total score of patients’ perceptions that the visit was patient centered was associated with positive health outcomes after adjusting for the clustering of patients within practices and after controlling for the 2 confounding variables [Table 4]. Patients’ postencounter levels of discomfort were lower when they perceived the visit to have been patient centered than not.

A similar result occurred for 2 other patient health outcomes: the patients’ postencounter level of concern (P=.02), and the mental health dimension of the SF-36 measure assessed 2 months after the study visit (P=.05). The subscore of patient perceptions that the patient and physician found common ground was associated with one of the health outcomes, the patients’ postencounter level of concern (P=.04). There were no significant associations of the subscore on patients’ perceptions that the illness experience had been explored with any of the patient health outcome measures.

Patients who perceived that their visit had been patient centered received fewer diagnostic tests [Table 5] and referrals [Table 6] in the subsequent 2 months. The proportion receiving diagnostic tests rose from 14.6% in the group who perceived that the visit had been patient centered (total score), to 24.3% in the group who perceived the visit was not. The proportion who were referred doubled from approximately 8% to 16%. These relationships were found even more strongly for the subscore on patient perceptions that the patient and the physician found common ground, but were not found for the subscore on patient perceptions that their illness experience had been explored. The proportion receiving diagnostic tests quadrupled from 4.1% in the group who perceived that the patient and the physician found common ground, to 25.4% in the group who perceived that common ground had not been attained. The proportion who were referred doubled from 6.1% to 14.9%. The number of visits by the patient to the family physician during the subsequent 2 months was not significantly related to the patient perceptions of patient centeredness, although there was a trend (P=.11) with the average number of visits in 2 months in the 4 quartiles of patient perceptions as follows: 1.0, 0.8, 1.2, and 1.3.

Discussion

Pathway to Improved Patient Outcomes

Patient-centered practice was associated with improved patients’ health status and increased efficiency of care (reduced diagnostic tests and referrals). However, only 1 of the 2 measures of patient-centered practice showed this result, the measure of patients’ perceptions of the patient centeredness of the visit. The measure that was based on ratings of audiotaped physician-patient interactions, while related to the patients’ perception, was not directly related to health status or efficiency.

The relationship of patients’ perceptions of patient centeredness with their health and efficiency of care was both statistically and clinically significant. Specifically, recovery was improved by 6 points on a 100-point scale; diagnostic tests and referrals were half as frequent if the visit was perceived to be patient centered.

The associations we found may imply a potentially important pathway (which could be tested in future trials), such as the one shown in the [Figure]. The pathway suggests a process through which patient-physician communication influences patients’ health, by first influencing the patients’ perceptions of being a full participant in the discussions during the encounter. Such a pathway has been noted by Sobel, whose review suggested a pathway to explain the lack of a direct relationship between patient education programs and patient health where there was a relationship between patient perceptions about their health and health outcomes. Sobel called this pathway “a biology of self-confidence.”³¹ He and others³² stress the critical role of patient perceptions in the healing process, which highlight that a person’s subjective experience influences biology.

How do we understand the results that show the ratings of the audiotape were not directly related to the outcomes, but the patient-centered perception measure was related to outcomes? One interpretation is that observable skills are not as important as patient perceptions. Although there is some evidence that skills training can improve both physicians’ behavior and patients’ health,³³ our findings and those of Bensing and Sluijs³⁴ indicate that differences in interviewing skills may not be associated with patient responses. Physicians may learn to go through the motions of patient-centered interviewing without understanding what it means to be a truly attentive and responsive listener. The implications of the current findings for educators are that education about communication should go well beyond skills training to a deeper understanding of what it means to be a responsive partner for the patient, during both that phase of the visit in which the problem is discussed and when the discussion of treatment options occurs. Two examples of such education approaches are: small group discussions between patients and physicians to illustrate the patients’ experiences and needs, and reviews of videotaped interviews with standardized patients participating in the review. Placing prime importance on the patients’ perceptions recognizes the influence of these perceptions on the patients’ subsequent health and epitomizes being truly patient centered.

Views that the visit was patient centered included perceptions about the discussion of the problem (exploring the illness experience) as well as discussion and agreement about treatment options (finding common ground). There is a substantial body of research supporting the importance of these discussions. The Headache Study found that patients’ perceptions that a full discussion of the problem had taken place predicted resolution of headaches after 1 year.³⁴ In keeping with our results, which found that finding common ground was more strongly associated with outcomes than exploring the illness experience, Riccardi and Kurtz³⁶ stressed that the physicians’ explanation to the patients was the crucial phase of the visit. Also, a key outcome study has found that patient agreement with the physician about the nature of the treatment and the need for follow-up were strongly associated with their recovery.³⁷

Efficiency of Medical Care

We found that patient-centered practice (assessed by patients’ perceptions) was associated with the efficiency of care by reducing subsequent diagnostic tests and referrals by half, after controlling for key confounding variables. These results were both statistically significant and clinically significant. Also, the number of subsequent visits to the family physician was lower (although not significantly) when the patient perceived the study visit to be patient centered. Efficiency in health service delivery was also found in a randomized trial of compassionate care in the emergency department setting with homeless patients.³⁸ In their study of continuity of care in Norwegian general practice, Hjortdahl and Borchgrevink³⁹ found that diagnostic tests were 10 times more likely to be ordered for patients about whom physicians reported the least previous knowledge compared with patients in whom they had reported fullest knowledge. Also, patients had only half the chance of being referred if their physicians knew them and their history.³⁸

One possible interpretation of the results of our study is that patient-centered physicians order fewer tests and refer less often. However, countering this interpretation is the fact that individual physicians in our study showed a range of patient-centered scores, as well as a range in test ordering and referral. In addition, the statistical analysis took account of the clustering of patients within a physician’s practice.

An alternative interpretation is that patients’ perceptions may influence resource use in several ways. For example, increased participation during the visit may reduce patients’ anxiety and their perceived need for investigations and referrals. Alternatively, patients’ perception that the physician has not understood their problem may provoke insecurities resulting in a request for further medical interventions. Also, if patients openly express their discontent with the encounter there may be an increase in physicians’ anxiety and a lowering of their threshold for diagnostic uncertainty, resulting in further investigations and referrals.

Certainly the finding that the failure to be patient centered (as perceived by the patient) was related to higher rates of referral and diagnostic tests should be a concern for medical education and health care policy. Perhaps of most importance is that the patients’ experience of being a participating member in the discussion of the problem and the treatment process may translate into the patients’ reduced need for further investigation or referral—simultaneously reducing the physicians’ need as well.

These findings counter a common misconception: that being patient centered means responding to every whim of the patient, thereby increasing expenses to the health care system.

Limitations

Approximately 30% of the patients refused to participate, and although the participants represented the age distribution of eligible patients, men were overrepresented in the study. Nonetheless, sex was not identified as a confounding variable for the associations studied.

Although no measure of severity was possible, the variables representing concurrent health problems and concurrent life problems were considered in the analysis strategy. They were not related to the outcome variables and were therefore not entered into the multivariable analyses.

One interpretation of the lack of association between patient-centered scores on the audiotaped interviews and subsequent health outcomes may be that the audiotape measure has failed to capture the important essence of the dynamic interaction between physicians and patients. The measure had a number of strengths, however; it had been tested for reliability and validity (compared with a global rating), and it was based on a theoretical framework. Also, it was correlated with one component of the patient perception measure of a patient-centered interview, a finding which indicates that future research should be directed toward determining physicians’ skills and behaviors that correlate with the patients’ positive perceptions, especially the perception that common ground has been reached. Such behaviors could then be emphasized in clinical teaching.

It should be noted that the utilization data were available only from the participating practices and not from care received elsewhere. Although this is a limitation, it would be expected that this lack of data would minimize the current relationship between patient-centered practice and utilization, because patients with less favorable perceptions would be potentially more likely to seek care elsewhere. Also, drug costs and hospital costs were not included and require further study. Future research could also build on these results about resource utilization and assess the specific kinds and actual costs of the diagnostic tests and referrals.

It could be argued that the results of our study demonstrated simply that people with positive perceptions and less severe problems achieved better health and more efficient services. We counter this interpretation with 2 thoughts. First, the preliminary step in our analysis included confounding variables to control for a variety of relevant variables (ie, personality and concomitant health problems). Only 2 confounding variables were influential enough to remain in the final analysis: marital status and diagnostic code of the main presenting problem. Second, patient perceptions were not independent of the physician-patient visit. They were influenced significantly by the communication score based on the audiotaped encounter, implying that the measure of perceptions was tapping not merely the patients’ general outlook on life, but also an important interactive component of visits between patients and physicians.

Conclusions

Patient-centered practice was associated with improved health status (less discomfort, less concern, and better mental health) and increased efficiency of care (fewer diagnostic tests and referrals).

Patients’ perceptions of the patient centeredness of the visit, but not the measure of audiotaped interactions, were directly associated with the positive outcomes. The subscore on patients’ perception of finding common ground was more strongly associated with the positive outcomes than the subscore on patients’ perception about exploring the illness experience.

Medical education should go beyond skills training to encourage physicians’ responsiveness to the patients’ unique experience. Therefore, involving real patients and standardized patients in teaching programs is recommended.

Health service organizations must recognize that efficiencies accrue from patient-centered practice and encourage such practice through structures that enhance continuity of the patient-physician relationship and through meaningful education programs.

Acknowledgments

Our project was supported by a grant from the Health Care Systems Research Program of the Ministry of Health of Ontario. The setting of the project was the Thames Valley Family Practice Research Unit (TVFPRU), a health system-linked research unit funded by the Ministry of Health of Ontario. The opinions contained are those of the authors, and no official endorsement by the Ministry is intended or should be inferred. The TVFPRU is part of the Centre for Studies in Family Medicine, Department of Family Medicine, The University of Western Ontario, London, Canada.

Being patient centered is a core value of medicine for many physicians. The principles of patient-centered medicine date back to the ancient Greek school of Cos, which was interested in the particulars of each patient.¹ More recently similar concepts have arisen in a variety of fields of human endeavor: the concept of physical diagnosis and deeper diagnosis of Balint,² the client-centered therapy of Rogers,³ the total-person approach to patient problems in nursing of Neuman and Young,⁴ the biopsychosocial model of Engel,⁵ and the disease- versus patient-centered medical practice of Byrne and Long.⁶ In the past decade the patient-centered concepts of Gerteis and colleagues⁷ have been applied to the hospital setting.

In the setting of primary care, and specifically family practice, patient-centered concepts incorporate 6 interactive components. The first component is the physician’s exploration of both the patients’ disease and 4 dimensions of the illness experience including: their feelings about being ill, their ideas about what is wrong with them, the impact of the problem on their daily functioning, and their expectations of what should be done. The second component is the physician’s understanding of the whole person. The third component is the patient and physician finding common ground regarding management. In the fourth component the physician incorporates prevention and health promotion into the visit. The fifth component is the enhancement of the patient-physician relationship. Finally, the sixth component requires that patient-centered practice be realistic. Our study addresses the first 3 of these components. Being patient centered does not mean that physicians abdicate control to the patient⁸ but rather that they find common ground in understanding the patients and more fully respond to their unique needs.⁹

What are the benefits of being patient centered? Previous research of specific communication variables indicates that patient-centered encounters result in: (1) the duration of the office visit remaining the same^10,11 (2) better patient satisfaction,¹² (3) higher physician satisfaction,¹⁰ and (4) fewer malpractice complaints.¹³ We focus on 2 other outcomes: patients’ health and efficiency of care.

Methods

Our study was designed to test the hypothesis that adult patients whose first visit in an episode of illness is patient centered will, by 2 months after the first visit: (1) more frequently demonstrate recovery from the symptom (and recovery from the concern about the symptom); (2) demonstrate better self-reported health; and (3) experience less subsequent medical care (ie, fewer visits, diagnostic tests, and referrals), compared with patients whose visit is not patient centered.

Data Collection and Participants

For our observational cohort study data were collected at 5 points: (1) the research assistant identified eligible patients in the physician’s office before the visit; (2) the office encounter was audiotaped and scored for patient-centered communication; (3) the research assistant held a postencounter interview with the patient; (4) we assessed, by chart review, the use of medical care during the 2-month follow-up; and (5) we conducted a follow up telephone interview with patients 2-months after the encounter.

Physician Selection. Physicians were recruited from the 250 family physicians practicing in London, Ontario, Canada, and the surrounding area. They were randomized within strata to ensure a representative sample in terms of year of graduation and geographic location and were selected using a modified version of the method of Borgiel and colleagues.¹⁴

Patients. We approached patients who were older than 18 years and had 1 or more recurring problems who presented to their physician’s office. Patients were excluded if they were too ill or disabled to answer questions, had no presenting problem, were in the office for counseling, were accompanied by another person, were not fluent in English, were hard of hearing, or were cognitively impaired. They were approached before they saw the physician and were blind to the study hypotheses.

Sample Size Estimation. The sample size required for correlations of 0.20 to be detected with an a set at 0.05 (2 tailed) and a b set at 0.10 was 259¹⁵ patients. Further inflation by 10% to account for the effect of clustering on multiple regression¹⁶ was thought to be reasonable (259÷0.9=288). Expecting 75% to cooperate, we aimed to approach 384 patients (288÷0.75).

Measures*

Measure of Patient-Centered Communication Score. The patient-centered communication score is based on 3 of the 6 components of the model of patient-centered medicine.^17-20 The first component (exploring the disease and the illness experience) received a high score when the physician explored the patients’ symptoms, prompts, feelings, ideas, function, and expectations. The second component (understanding the whole person) received a high score when the physician elicited and explored issues relating to life cycle, personality, or life context, including family. The third component (finding common ground) received a high score when the physician clearly described the problem and the management plan, answered questions about them, and discussed and agreed on them with the patient. Scoring sheets and procedures are described in detail elsewhere.²¹ Scores could range from 0 (not at all patient centered) to 100 (very patient centered).

Interrater reliability has been established in earlier versions of the measure and for the current version (r=0.69, 0.84, and 0.80 among 3 raters,²² 0.91 among 2 raters,²³ and 0.83 for n=19 for our study). Intrarater reliability was 0.73 (n=20).

Correlations with global scores encompassing the 3 components supported the validity of the score (0.63 in an earlier study²³ and 0.85 for our study, n=46).

Patient Perception of Patient-Centeredness. Based on the patient-centered model, a series of 14 items developed and validated in previous studies^24,25 were used to assess the patients’ postencounter perceptions of how patient centered the interaction with the physician had been.† Items were averaged into: total score, a subscore on exploring the disease and illness experience, and finding common ground. Low scores represented patient centeredness.

Patient Recovery from Discomfort and Concerns. The primary health outcome was the recovery measure based on the patients’ self-administered report on visual analogue scales (VAS) of the severity of the symptom they identified as the main presenting problem and their concern about that problem at 2 points: the postencounter interview and the follow-up 2 months later.^26,27 VAS have been tested for reliability and validity in studies of pain and nausea (correlation of 0.75 with an intensity score).²⁶ Each of the symptom recovery variables was continuous.

Patient Health Status. The Medical Outcomes Study Short Form-36 (SF-36) was used to assess self-reported secondary health outcomes. This valid and reliable measure¹⁸ is a multidimensional assessment of: physical health, mental health, perception of health, social health, pain, and role function. All were continuous variables except role function, for which the distribution of scores necessitated dichotomizing.

Medical Care. The care provided during the 2 months following the audiotaped encounter was assessed by chart review (adapted from Bass and coworkers²⁴) by 3 medical doctors (I.R.M., J.O., J.J.) blind to the identity of the family physician and the patient, and also to the patient-centered scores. Items abstracted were: the total number of visits during the 2 months (continuous variable); the number and kind of diagnostic tests ordered during the 2 months that were relevant to the problems presented at the audiotaped visit (dichotomous); and the number and kinds of referrals made during the 2 months that were relevant to the problems presented at the audiotaped visit (dichotomous).

Analysis. The hypotheses were tested using multiple regression for continuous outcomes and multiple logistic regression for dichotomous outcomes,²⁹ both adjusted for the effect of the clustering of patients by physician using “procedure mixed” in SAS for continuous outcomes and using both “procedure logistic” and “procedure IML” in SAS for dichotomous outcomes.³⁰ The unit of analysis was the patient.

The following confounding variables were included in preliminary multivariable analyses on the basis of their univariable relationships with outcomes at the level of P <.10: age, sex, number of family members at home, desire to share feelings, who initiated the visit, tense personality, coping skills, concomitant health problems, social support, marital status (married vs other), concomitant life problems, number of visits to the physician in the previous 12 months, and main problem (1 of 5 groups: digestive, musculoskeletal, respiratory, skin, and other).

Because of substantial sample attrition with so many covariates, and because only 2 variables were consistently associated with the outcome measures, each subsequent multivariable analysis was conducted with each of the primary independent variables and the 2 covariates (patients’ main presenting problem and marital status).

Results

Descriptive Results

Of the 102 randomly selected family physicians, 83 were eligible because they were still practicing in the area and had adequate office space to accommodate the research assistant. Of these, 39 (47%) agreed to participate and completed the data collection. The participants were similar to the refusers [Table 1] in year of graduation, practice location (rural or urban; high or low socioeconomic status) and sex; however, participants were significantly more likely to be certificants of the College of Family Physicians of Canada than refusers (59% and 27%, respectively; P=.007).

Of 464 eligible patients, 334 (72%) agreed to participate. Nineteen (~6%) were lost to the study. The final 315 participants represented an overall participation rate of 68%; their age was representative of the eligible patients, but there was a higher proportion of men than in the total group of eligible patients.

[Table 2] shows that the slim majority of final participants were women, and most were middle aged and married. Typical of the city, approximately 4 in 10 had more than a high school education. The most common presenting problems were respiratory in nature.

[Table 3] shows the descriptive results for key variables.

Hypothesis Testing Results

The patient-centered communication scores (based on the audiotape analysis) were not significantly related to any of the health outcomes after adjusting for the clustering of patients within practices and after controlling for the 2 confounding variables. Similarly, patient-centered communication scores were not related to any of the 3 medical care outcomes.

Patient-centered communication scores (based on the audiotape analysis) were significantly correlated in the expected direction, with patient perceptions that the patient and physician found common ground (r =-0.16; P=.01). High scores (indicating very patient-centered communication) were correlated with low patient perception scores (indicating patient-centeredness). The 2 other patient perception scores (total patient perception score and the subscore on patient perception that the illness experience was explored) were not significantly associated with patient-centered communication scores.

The total score of patients’ perceptions that the visit was patient centered was associated with positive health outcomes after adjusting for the clustering of patients within practices and after controlling for the 2 confounding variables [Table 4]. Patients’ postencounter levels of discomfort were lower when they perceived the visit to have been patient centered than not.

A similar result occurred for 2 other patient health outcomes: the patients’ postencounter level of concern (P=.02), and the mental health dimension of the SF-36 measure assessed 2 months after the study visit (P=.05). The subscore of patient perceptions that the patient and physician found common ground was associated with one of the health outcomes, the patients’ postencounter level of concern (P=.04). There were no significant associations of the subscore on patients’ perceptions that the illness experience had been explored with any of the patient health outcome measures.

Patients who perceived that their visit had been patient centered received fewer diagnostic tests [Table 5] and referrals [Table 6] in the subsequent 2 months. The proportion receiving diagnostic tests rose from 14.6% in the group who perceived that the visit had been patient centered (total score), to 24.3% in the group who perceived the visit was not. The proportion who were referred doubled from approximately 8% to 16%. These relationships were found even more strongly for the subscore on patient perceptions that the patient and the physician found common ground, but were not found for the subscore on patient perceptions that their illness experience had been explored. The proportion receiving diagnostic tests quadrupled from 4.1% in the group who perceived that the patient and the physician found common ground, to 25.4% in the group who perceived that common ground had not been attained. The proportion who were referred doubled from 6.1% to 14.9%. The number of visits by the patient to the family physician during the subsequent 2 months was not significantly related to the patient perceptions of patient centeredness, although there was a trend (P=.11) with the average number of visits in 2 months in the 4 quartiles of patient perceptions as follows: 1.0, 0.8, 1.2, and 1.3.

Discussion

Pathway to Improved Patient Outcomes

Patient-centered practice was associated with improved patients’ health status and increased efficiency of care (reduced diagnostic tests and referrals). However, only 1 of the 2 measures of patient-centered practice showed this result, the measure of patients’ perceptions of the patient centeredness of the visit. The measure that was based on ratings of audiotaped physician-patient interactions, while related to the patients’ perception, was not directly related to health status or efficiency.

The relationship of patients’ perceptions of patient centeredness with their health and efficiency of care was both statistically and clinically significant. Specifically, recovery was improved by 6 points on a 100-point scale; diagnostic tests and referrals were half as frequent if the visit was perceived to be patient centered.

The associations we found may imply a potentially important pathway (which could be tested in future trials), such as the one shown in the [Figure]. The pathway suggests a process through which patient-physician communication influences patients’ health, by first influencing the patients’ perceptions of being a full participant in the discussions during the encounter. Such a pathway has been noted by Sobel, whose review suggested a pathway to explain the lack of a direct relationship between patient education programs and patient health where there was a relationship between patient perceptions about their health and health outcomes. Sobel called this pathway “a biology of self-confidence.”³¹ He and others³² stress the critical role of patient perceptions in the healing process, which highlight that a person’s subjective experience influences biology.

How do we understand the results that show the ratings of the audiotape were not directly related to the outcomes, but the patient-centered perception measure was related to outcomes? One interpretation is that observable skills are not as important as patient perceptions. Although there is some evidence that skills training can improve both physicians’ behavior and patients’ health,³³ our findings and those of Bensing and Sluijs³⁴ indicate that differences in interviewing skills may not be associated with patient responses. Physicians may learn to go through the motions of patient-centered interviewing without understanding what it means to be a truly attentive and responsive listener. The implications of the current findings for educators are that education about communication should go well beyond skills training to a deeper understanding of what it means to be a responsive partner for the patient, during both that phase of the visit in which the problem is discussed and when the discussion of treatment options occurs. Two examples of such education approaches are: small group discussions between patients and physicians to illustrate the patients’ experiences and needs, and reviews of videotaped interviews with standardized patients participating in the review. Placing prime importance on the patients’ perceptions recognizes the influence of these perceptions on the patients’ subsequent health and epitomizes being truly patient centered.

Views that the visit was patient centered included perceptions about the discussion of the problem (exploring the illness experience) as well as discussion and agreement about treatment options (finding common ground). There is a substantial body of research supporting the importance of these discussions. The Headache Study found that patients’ perceptions that a full discussion of the problem had taken place predicted resolution of headaches after 1 year.³⁴ In keeping with our results, which found that finding common ground was more strongly associated with outcomes than exploring the illness experience, Riccardi and Kurtz³⁶ stressed that the physicians’ explanation to the patients was the crucial phase of the visit. Also, a key outcome study has found that patient agreement with the physician about the nature of the treatment and the need for follow-up were strongly associated with their recovery.³⁷

Efficiency of Medical Care

We found that patient-centered practice (assessed by patients’ perceptions) was associated with the efficiency of care by reducing subsequent diagnostic tests and referrals by half, after controlling for key confounding variables. These results were both statistically significant and clinically significant. Also, the number of subsequent visits to the family physician was lower (although not significantly) when the patient perceived the study visit to be patient centered. Efficiency in health service delivery was also found in a randomized trial of compassionate care in the emergency department setting with homeless patients.³⁸ In their study of continuity of care in Norwegian general practice, Hjortdahl and Borchgrevink³⁹ found that diagnostic tests were 10 times more likely to be ordered for patients about whom physicians reported the least previous knowledge compared with patients in whom they had reported fullest knowledge. Also, patients had only half the chance of being referred if their physicians knew them and their history.³⁸

One possible interpretation of the results of our study is that patient-centered physicians order fewer tests and refer less often. However, countering this interpretation is the fact that individual physicians in our study showed a range of patient-centered scores, as well as a range in test ordering and referral. In addition, the statistical analysis took account of the clustering of patients within a physician’s practice.

An alternative interpretation is that patients’ perceptions may influence resource use in several ways. For example, increased participation during the visit may reduce patients’ anxiety and their perceived need for investigations and referrals. Alternatively, patients’ perception that the physician has not understood their problem may provoke insecurities resulting in a request for further medical interventions. Also, if patients openly express their discontent with the encounter there may be an increase in physicians’ anxiety and a lowering of their threshold for diagnostic uncertainty, resulting in further investigations and referrals.

Certainly the finding that the failure to be patient centered (as perceived by the patient) was related to higher rates of referral and diagnostic tests should be a concern for medical education and health care policy. Perhaps of most importance is that the patients’ experience of being a participating member in the discussion of the problem and the treatment process may translate into the patients’ reduced need for further investigation or referral—simultaneously reducing the physicians’ need as well.

These findings counter a common misconception: that being patient centered means responding to every whim of the patient, thereby increasing expenses to the health care system.

Limitations

Approximately 30% of the patients refused to participate, and although the participants represented the age distribution of eligible patients, men were overrepresented in the study. Nonetheless, sex was not identified as a confounding variable for the associations studied.

Although no measure of severity was possible, the variables representing concurrent health problems and concurrent life problems were considered in the analysis strategy. They were not related to the outcome variables and were therefore not entered into the multivariable analyses.

One interpretation of the lack of association between patient-centered scores on the audiotaped interviews and subsequent health outcomes may be that the audiotape measure has failed to capture the important essence of the dynamic interaction between physicians and patients. The measure had a number of strengths, however; it had been tested for reliability and validity (compared with a global rating), and it was based on a theoretical framework. Also, it was correlated with one component of the patient perception measure of a patient-centered interview, a finding which indicates that future research should be directed toward determining physicians’ skills and behaviors that correlate with the patients’ positive perceptions, especially the perception that common ground has been reached. Such behaviors could then be emphasized in clinical teaching.

It should be noted that the utilization data were available only from the participating practices and not from care received elsewhere. Although this is a limitation, it would be expected that this lack of data would minimize the current relationship between patient-centered practice and utilization, because patients with less favorable perceptions would be potentially more likely to seek care elsewhere. Also, drug costs and hospital costs were not included and require further study. Future research could also build on these results about resource utilization and assess the specific kinds and actual costs of the diagnostic tests and referrals.

It could be argued that the results of our study demonstrated simply that people with positive perceptions and less severe problems achieved better health and more efficient services. We counter this interpretation with 2 thoughts. First, the preliminary step in our analysis included confounding variables to control for a variety of relevant variables (ie, personality and concomitant health problems). Only 2 confounding variables were influential enough to remain in the final analysis: marital status and diagnostic code of the main presenting problem. Second, patient perceptions were not independent of the physician-patient visit. They were influenced significantly by the communication score based on the audiotaped encounter, implying that the measure of perceptions was tapping not merely the patients’ general outlook on life, but also an important interactive component of visits between patients and physicians.

Conclusions

Patient-centered practice was associated with improved health status (less discomfort, less concern, and better mental health) and increased efficiency of care (fewer diagnostic tests and referrals).

Patients’ perceptions of the patient centeredness of the visit, but not the measure of audiotaped interactions, were directly associated with the positive outcomes. The subscore on patients’ perception of finding common ground was more strongly associated with the positive outcomes than the subscore on patients’ perception about exploring the illness experience.

Medical education should go beyond skills training to encourage physicians’ responsiveness to the patients’ unique experience. Therefore, involving real patients and standardized patients in teaching programs is recommended.

Health service organizations must recognize that efficiencies accrue from patient-centered practice and encourage such practice through structures that enhance continuity of the patient-physician relationship and through meaningful education programs.

Acknowledgments

Our project was supported by a grant from the Health Care Systems Research Program of the Ministry of Health of Ontario. The setting of the project was the Thames Valley Family Practice Research Unit (TVFPRU), a health system-linked research unit funded by the Ministry of Health of Ontario. The opinions contained are those of the authors, and no official endorsement by the Ministry is intended or should be inferred. The TVFPRU is part of the Centre for Studies in Family Medicine, Department of Family Medicine, The University of Western Ontario, London, Canada.

Alternative or complementary approaches to medical care are gaining loyalty from patients and increasing interest from the allopathic health care community.^1-4 In particular, professional organizations in the areas of acupuncture and manual therapy are offering and expanding continuing medical education (CME) programs in these fields for allopathic physicians, though there are few published data on their effectiveness.^5,6

The direct impact of CME on patients and clinical practice has been little studied, particularly in relation to the treatment of low back pain.⁷ Cherkin and colleagues^8,9 undertook an evaluation of a didactic CME program on low back pain in 1991 by studying patient satisfaction and provider attitudes. It appeared that the patients of providers who professed greater confidence in managing low back pain were more satisfied with their care, though negative attitudes previously expressed by clinicians toward low back pain did not change significantly after CME. In a discussion of this study it was suggested that patients might be seeking information and practical guidance rather than a cure or empathy.¹⁰ However, the investigators did not study the effects of modifying physical examination and manual skills in the care of these patients, factors that might play an important role in outcomes.

Although greater patient satisfaction has been associated with chiropractic care (which emphasizes manual skills) than that given by primary care physicians, there appears to be no association of satisfaction with practitioner self-confidence or days to functional recovery of the patient.^11,12

We developed a workshop for generalist clinicians in the skills of assessment, limited manual therapy, and a graded exercise program, and in a randomized controlled trial evaluated clinician self-efficacy and patient outcomes for acute low back pain.^13,14 We demonstrated that allopathic generalist physicians could be effectively trained in limited manual therapy with self-reported increased competence in managing low back pain. The patients receiving therapy showed a trend toward feeling completely better more quickly but reported no greater satisfaction or objective functional improvement in terms of activities than patients in the control group who were receiving only high-quality conventional care through workshop training.¹⁴

Of 31 physicians recruited into the manual therapy trial, 13 had previously been involved in a cohort study of utilization and back care therapy given by 208 practitioners (115 primary care generalists) to 1633 patients (644 patients of primary care generalists).¹¹ We examined the outcomes of patients with low back pain from the practices of these 13 physicians before (data from the cohort study) and after an intensive hands-on training workshop (data from the manual therapy trial), using similar patient recruitment methods and evaluation instruments ([Figure 1]. Our hypothesis was that this training in manual and assessment skills would improve patient outcomes and satisfaction.

Methods

The initial cohort study was undertaken to examine the prevalence, care seeking, and outcomes of acute low back pain in the state of North Carolina. No interventions were undertaken in this study. Methods and measures used in its implementation have been described previously by Carey and coworkers.¹¹ Clinicians were randomly selected from medical and chiropractic licensure files (primary care generalists, chiropractic physicians, and orthopedic surgeons). To be included in the cohort study clinicians had to see ambulatory patients at least 50% of the time and provide first contact care for acute low back pain. Of the 208 clinicians from different disciplines recruited into the study, 115 were primary care physicians. During a 10-month period consecutive patients with acute or subacute low back pain (<10 weeks) were enrolled unless they had received previous care for the episode, had received previous back surgery, had a history of cancer, were pregnant, had no telephone, or were unable to speak English. Patients were contacted by telephone after the visit by staff members of the University of North Carolina Survey Research Unit, and interviews were undertaken at baseline, 2, 4, 8, 12, and 24 weeks.

These interviews included questions about details of the back pain episode, medications and other therapies used, tests performed, work and compensation status, demographic data, and income level. Outcome data included responses to the 23-item Roland-Morris Back Disability Questionnaire.¹⁵ We also asked when the patients considered they were able to perform their usual daily activities after the back pain. Patient satisfaction was assessed on how well the physician communicated, listened, gave information and explained the cause of back pain, whether a detailed history was taken and the back examined carefully, and if advice was given on pain management, prevention, and activities of sleeping and sitting (yes/no responses). Other satisfaction items (overall treatment, pain relief provided, and patient abilities to walk, socialize, and work) were rated on a 5-point Likert scale (poor, fair, good, very good, or excellent). This was subsequently adapted to a dichotomous response. Clinical and utilization data were obtained from charts in the physicians’ offices to allow validation of survey variables.

The randomized trial of the effectiveness of limited manual therapy was started in 1995, 12 months after the closure of the cohort study. Patients were recruited by 31 generalist physicians (13 from the earlier cohort study and 18 volunteers from the 630 physicians on the North Carolina physician master file). The same inclusion and exclusion criteria from the cohort study were used except that the acceptable age range was 21 to 65 (compared with 75 years) and patients had no osteopenia, severe arthritis, morbid obesity, or neurological deficits and had not received previous manual therapy by the physician. These additional exclusion criteria were necessary to avoid possible adverse effects of manual therapy in the presence of disease and to eliminate patient bias of a preference for manual therapy based on previous experience.

For each arm of the study, after the first office visit the identical telephone interview questions and schedule used in the cohort study were implemented for up to 8 weeks. Chart abstraction methods and variables were also identical except that additional data were collected on the specifics of manual therapy given at each visit.

Two sequential weekend workshops with a refresher session for each of the 31 physicians (developed and given to 9 physicians per workshop by 3 family physicians skilled in manual therapy) were implemented before the start of the clinical trial. The purpose was to train these physicians in quality care for low back pain (explained to patients as enhanced care—the control arm) and in standardized limited manual therapy (the main component of the intervention arm).¹³ The term “enhanced care” was developed to minimize the impression for patients during randomization that they might either receive something special (manual therapy) or just routine care—both options needed to appear to be special to reduce placebo bias.

Training for the enhanced low back pain care arm included physician education in (1) the directed history and physical examination using Agency for Health Care Policy and Research (AHCPR) guidelines,¹⁶ (2) review of the efficacy of imaging and laboratory testing, (3) review of the efficacy of treatment modalities, and (4) use of specially designed patient handouts emphasizing progressive exercises, daily activities, and early return to function.

Training for the limited manual therapy arm included the enhanced low back pain skills plus:

• manual therapy, consisting of: (1) principles of manual therapy and explanatory models, (2) instruction in motion testing, and (3) instruction in limited manual therapy skills (soft-tissue, muscle energy, and high-velocity low-amplitude techniques involving psoas and piriformis muscles—lumbar spine, lumbosacral junction, and sacroiliac joints)
• workshop training and demonstrated competence in low back care on simulated patients
• guidance and practice in integrating limited manual therapy into the office visit
• education and practice in recruiting patients and in the procedures of random assignment of those patients to the control (enhanced care) arm or the manual therapy arm

After training, the physicians returned to their practices and worked on their newly learned skill for approximately 3 months before enrolling patients. At an agreed time they began to enroll patients and randomized them to enhanced care or enhanced care plus limited manual therapy using a blinded method.

Analysis

Using the patients of the 13 physicians active in both the original cohort study and the subsequent randomized trial of limited manual therapy, we compared outcomes between 3 groups of patients: (1) those whose usual care was only observed in the earlier cohort study, (2) those receiving enhanced care (control arm) in the randomized trial, and (3) those receiving enhanced care plus manual therapy (intervention arm) in the randomized trial. The major variables included in the analyses were age, sex, education, household income, duration of low back pain episode for more than 2 weeks, presence of sciatica, workers’ compensation status, and the Work Adaptation, Partnership, Growth, Affection, and Resolve Survey (a measure of job satisfaction).¹⁶ In each of the studies, outcome measures from repeated interviews included data on functional status over time and pain levels. Data on satisfaction with care, return to work, and time to functional and complete recovery were obtained either at 8 weeks or when the patients were better. The main outcome measures were the Roland-Morris adaptation of the Sickness Impact Profile (a 23-item scale with high scores indicating significant dysfunction), patients’ report of being all better or functionally better, and the date they were able to return to performing their usual daily activities.^10,14 Patient satisfaction measures were based on the scale developed by Cherkin and colleagues.⁸ The patient was the unit of analysis.

Pearson’s chi-square was used when comparing the 3 patient groups by physician performance and patient satisfaction, adjusting for baseline differences. Linear modeling was used to examine the relationship of manual therapy to functional status (Roland-Morris score). Cox proportional hazard modeling was undertaken to identify survival curves of time to functional recovery. In all analyses, standard errors were corrected for any intraclass correlation due to nonindependence of patients seeing the same physician.¹⁷

Results

The 13 physicians cared for 120 patients in the observational study and 191 patients in the randomized trial [Figure 1]. In each of the studies very similar numbers of patients were seen by each physician. Within the 2 arms of the randomized trial of manual therapy, baseline characteristics were essentially the same [Table 1].

There were some differences in baseline characteristics between patients in the cohort study and the randomized trial. More patients in the randomized trial (41.4% vs 25%, P=.01) had significantly higher severe baseline dysfunction (Roland-Morris score=16-23) than in the cohort study. The mean baseline Roland Morris score was 10.0 for the cohort study patients compared with 12.5 for the manual therapy study patients (P=.03). There were more patients on workers’ compensation in the cohort study (35%) than in the manual therapy study (16.3%, P=.002), and fewer had been employed in the previous 3 months (80.8% vs 90.0%, P=.02).

Patients in both arms of the randomized manual therapy trial were significantly more satisfied than their counterparts in the earlier cohort study in terms of how their physician gave care in the clinical encounter and whether they received effective advice [Table 2]. Patients participating in the manual therapy trial were also more satisfied with their pain relief, their physicians’ overall treatment of back pain, and their ability to perform activities of daily living. Specific items of clinical performance where differences were noted included: the physician took a detailed history; gave useful advice on pain, preventive measures, sleeping, and sitting strategies; and provided back exercises. There were no differences in patient ratings of the physician’s ability to listen effectively, perform a careful physical examination, and explain the cause of their back pain. There were no differences in general health status between the cohort and manual therapy study patients.

When adjusted for baseline function, the presence of sciatica, duration of pain, employment status, workers’ compensation, and income of more than $20,000, mean functional outcomes measured by Roland-Morris scores at 2 weeks were 8.2 for the patients in the cohort study and 6.7 in the manual therapy trial (P=.03); at 4 weeks mean scores were 7.2 for the cohort and 5.2 for patients in the manual therapy trial (P=.02), and at 8 weeks scores were 6.7 and 3.6, respectively (P=.002). These were clinically significant differences showing that all patients in the clinical trial had lower functional disability levels during an 8-week period than patients of the same physicians in the cohort study.

After controlling for baseline Roland-Morris score, duration of low back pain, sciatica, employment status, workers’ compensation, and income more than $20,000, time to functional recovery reported by the patient (using Cox proportional hazard modeling) between the 3 groups of patients was as follows: (1) manual therapy + enhanced care versus cohort: hazard ratio (HR)=1.16; 95% confidence interval (CI), 0.85-1.58; (2) enhanced care alone versus cohort: HR=1.13; 95% CI, 0.82-1.54; and (3) manual therapy + enhanced care versus enhanced care alone: HR=1.03; 95% CI, 0.75-1.40. None of these HRs were significant [Figure 2].

Discussion

Data from patients of 13 physicians involved in an observational study of back care, followed by a randomized trial of the effect of additional clinical skills developed by hands-on training, provide an unusual opportunity to study patient outcomes and physician performance over time. Similar eligibility and exclusion criteria, data collection methods, and outcome measures were used in both studies, allowing us to directly compare the variables of interest.

After controlling for confounding factors and physician clustering effect, patients in the clinical trial of manual therapy had significantly more baseline dysfunction at the index visit but still recovered more rapidly by self-report. After workshop training for the physicians, the proportion of patients who reported effective evaluation and management of their back problem increased substantially.

There were no differences in the 2 studies in the proportions of patients reporting on how effectively their physician listened to them, explained causation, and discussed occupational issues. These latter activities were not specifically addressed in the training workshop, which concentrated on clinical evaluation, manual techniques, and issues of physical rehabilitation. One would expect that a general placebo effect on patients of the training would increase all parameters of their perceptions of care, so the workshop may have produced specific learning effects on the clinicians. Despite a trend favoring the group receiving care from the “trained” physicians, there were no significant differences in how patients perceived the quality of the physician’s physical examination of their back—an item that was particularly emphasized in the workshops. This aspect of physician performance was already highly rated in the cohort group and may have been less likely to show improvement, given the need to know what constitutes a superior examination technique.

The improved reported performances of physicians and patient outcomes were noted in both arms of the manual therapy trial, compared with those of patients in the earlier cohort study, though there was no difference in time to full functional recovery between the 2 studies. The inconsistency between outcomes (Roland-Morris scores and time to functional recovery) is probably explainable by the fact that they measure different patient perceptions of back problems—specific versus global recovery. This inconsistency of measures has been noted in other studies of low back pain interventions.¹⁸

The improved outcomes of patients in the randomized trial compared with those in the cohort may be because of specific elements of the study applicable to all patients or may have been due to unmeasured baseline differences. The patients in the randomized trial were more impaired at baseline, which would tend toward worse rather than better outcomes if severity was not completely controlled for. There are several possible explanations for these findings: (1) workshop training did improve physician knowledge and skills; (2) involvement of patients in a clinical trial in which an intervention (enhanced care) was given in both arms of the study (one also receiving manual therapy) could heighten the overall positive effect; (3) the clinical trial involved more visits (average=3.6) than with patients in the first study (1.3)—this could add significantly to patient satisfaction and perceptions of good care; and (4) the recruited physicians were a special group of interested and motivated individuals who by their nature would produce better outcomes in a clinical trial.

Limitations

There are limitations to our findings and conclusions. There were differences in exclusion criteria that might have reduced the number of unhealthy people recruited into the manual therapy trial compared with the cohort study. This could have led to better outcomes, though we doubt that these differences made much impact given the average age of the patients (40 years).

Although we adjusted for baseline differences (income, workers’ compensation, and employment status), we are not sure why these differences occurred. One possibility is that the rapid population and commercial growth and high employment rates that occurred in the state of North Carolina in the time between the 2 studies improved the economic characteristics of patients presenting to the physicians over time. It is also possible that patients with higher incomes and fewer employment difficulties would be more likely to rate physician communication and management more highly and improve more rapidly.

The global effect on clinician knowledge, skills, and performance of a 3-year span of evolving clinical practice and involvement in low back pain studies (other than participating in a training workshop) cannot be quantified. For example, improved performance and outcomes might have been related to the publication of the AHCPR back pain guidelines in 1994, approximately 6 months before the start of the manual therapy study.¹⁶ However, at the low back pain workshop (where only the AHCPR “red flags” were presented) the participants seemed to be only minimally aware of these guidelines.

If the improved outcomes noted in this study were mainly because of the passage of time rather than workshop training, this provides an interesting insight into how practice changes for the better.

Patient satisfaction with care has been shown to correlate with outcomes.^19,20 Deyo and Diehl suggested that dissatisfaction with care for low back pain was related to failure to obtain an adequate explanation from the physician, while Cherkin and colleagues proposed that lack of confidence in management and negative attitudes of clinicians might be key issues to be addressed in achieving better outcomes.^8-10,21,22 However, didactic training to remedy these problems did not appear to be very effective in improving satisfaction. Following up on these suggestions, Smucker and coworkers,¹² using the large North Carolina cohort study, showed that clinician self-confidence (allopathic and chiropractic physicians) did not predict patient outcomes but commented that communication, time spent with the patient, and manual evaluation and treatment skills might be important variables affecting outcome.¹²

The margin of differences in outcomes shown in our preintervention and postintervention analysis leads us to suggest that allopathic physicians using a brief systematic evaluation and hands-on regional physical examination, sound advice on pain management and prevention, and an increasingly active exercise program can modestly improve early patient functioning and satisfaction in acute low back pain. Whether training in limited manual therapy adds to this benefit is unclear and must be taken into account in relation to the current expansion of CME in manipulative skills for allopathic physicians. The costs and benefits of providing this training—with the possibility of needing one extra office visit or more to fully implement enhanced care—would need to be assessed.²³

Alternative or complementary approaches to medical care are gaining loyalty from patients and increasing interest from the allopathic health care community.^1-4 In particular, professional organizations in the areas of acupuncture and manual therapy are offering and expanding continuing medical education (CME) programs in these fields for allopathic physicians, though there are few published data on their effectiveness.^5,6

The direct impact of CME on patients and clinical practice has been little studied, particularly in relation to the treatment of low back pain.⁷ Cherkin and colleagues^8,9 undertook an evaluation of a didactic CME program on low back pain in 1991 by studying patient satisfaction and provider attitudes. It appeared that the patients of providers who professed greater confidence in managing low back pain were more satisfied with their care, though negative attitudes previously expressed by clinicians toward low back pain did not change significantly after CME. In a discussion of this study it was suggested that patients might be seeking information and practical guidance rather than a cure or empathy.¹⁰ However, the investigators did not study the effects of modifying physical examination and manual skills in the care of these patients, factors that might play an important role in outcomes.

Although greater patient satisfaction has been associated with chiropractic care (which emphasizes manual skills) than that given by primary care physicians, there appears to be no association of satisfaction with practitioner self-confidence or days to functional recovery of the patient.^11,12

We developed a workshop for generalist clinicians in the skills of assessment, limited manual therapy, and a graded exercise program, and in a randomized controlled trial evaluated clinician self-efficacy and patient outcomes for acute low back pain.^13,14 We demonstrated that allopathic generalist physicians could be effectively trained in limited manual therapy with self-reported increased competence in managing low back pain. The patients receiving therapy showed a trend toward feeling completely better more quickly but reported no greater satisfaction or objective functional improvement in terms of activities than patients in the control group who were receiving only high-quality conventional care through workshop training.¹⁴

Of 31 physicians recruited into the manual therapy trial, 13 had previously been involved in a cohort study of utilization and back care therapy given by 208 practitioners (115 primary care generalists) to 1633 patients (644 patients of primary care generalists).¹¹ We examined the outcomes of patients with low back pain from the practices of these 13 physicians before (data from the cohort study) and after an intensive hands-on training workshop (data from the manual therapy trial), using similar patient recruitment methods and evaluation instruments ([Figure 1]. Our hypothesis was that this training in manual and assessment skills would improve patient outcomes and satisfaction.

Methods

The initial cohort study was undertaken to examine the prevalence, care seeking, and outcomes of acute low back pain in the state of North Carolina. No interventions were undertaken in this study. Methods and measures used in its implementation have been described previously by Carey and coworkers.¹¹ Clinicians were randomly selected from medical and chiropractic licensure files (primary care generalists, chiropractic physicians, and orthopedic surgeons). To be included in the cohort study clinicians had to see ambulatory patients at least 50% of the time and provide first contact care for acute low back pain. Of the 208 clinicians from different disciplines recruited into the study, 115 were primary care physicians. During a 10-month period consecutive patients with acute or subacute low back pain (<10 weeks) were enrolled unless they had received previous care for the episode, had received previous back surgery, had a history of cancer, were pregnant, had no telephone, or were unable to speak English. Patients were contacted by telephone after the visit by staff members of the University of North Carolina Survey Research Unit, and interviews were undertaken at baseline, 2, 4, 8, 12, and 24 weeks.

These interviews included questions about details of the back pain episode, medications and other therapies used, tests performed, work and compensation status, demographic data, and income level. Outcome data included responses to the 23-item Roland-Morris Back Disability Questionnaire.¹⁵ We also asked when the patients considered they were able to perform their usual daily activities after the back pain. Patient satisfaction was assessed on how well the physician communicated, listened, gave information and explained the cause of back pain, whether a detailed history was taken and the back examined carefully, and if advice was given on pain management, prevention, and activities of sleeping and sitting (yes/no responses). Other satisfaction items (overall treatment, pain relief provided, and patient abilities to walk, socialize, and work) were rated on a 5-point Likert scale (poor, fair, good, very good, or excellent). This was subsequently adapted to a dichotomous response. Clinical and utilization data were obtained from charts in the physicians’ offices to allow validation of survey variables.

The randomized trial of the effectiveness of limited manual therapy was started in 1995, 12 months after the closure of the cohort study. Patients were recruited by 31 generalist physicians (13 from the earlier cohort study and 18 volunteers from the 630 physicians on the North Carolina physician master file). The same inclusion and exclusion criteria from the cohort study were used except that the acceptable age range was 21 to 65 (compared with 75 years) and patients had no osteopenia, severe arthritis, morbid obesity, or neurological deficits and had not received previous manual therapy by the physician. These additional exclusion criteria were necessary to avoid possible adverse effects of manual therapy in the presence of disease and to eliminate patient bias of a preference for manual therapy based on previous experience.

For each arm of the study, after the first office visit the identical telephone interview questions and schedule used in the cohort study were implemented for up to 8 weeks. Chart abstraction methods and variables were also identical except that additional data were collected on the specifics of manual therapy given at each visit.

Two sequential weekend workshops with a refresher session for each of the 31 physicians (developed and given to 9 physicians per workshop by 3 family physicians skilled in manual therapy) were implemented before the start of the clinical trial. The purpose was to train these physicians in quality care for low back pain (explained to patients as enhanced care—the control arm) and in standardized limited manual therapy (the main component of the intervention arm).¹³ The term “enhanced care” was developed to minimize the impression for patients during randomization that they might either receive something special (manual therapy) or just routine care—both options needed to appear to be special to reduce placebo bias.

Training for the enhanced low back pain care arm included physician education in (1) the directed history and physical examination using Agency for Health Care Policy and Research (AHCPR) guidelines,¹⁶ (2) review of the efficacy of imaging and laboratory testing, (3) review of the efficacy of treatment modalities, and (4) use of specially designed patient handouts emphasizing progressive exercises, daily activities, and early return to function.

Training for the limited manual therapy arm included the enhanced low back pain skills plus:

• manual therapy, consisting of: (1) principles of manual therapy and explanatory models, (2) instruction in motion testing, and (3) instruction in limited manual therapy skills (soft-tissue, muscle energy, and high-velocity low-amplitude techniques involving psoas and piriformis muscles—lumbar spine, lumbosacral junction, and sacroiliac joints)
• workshop training and demonstrated competence in low back care on simulated patients
• guidance and practice in integrating limited manual therapy into the office visit
• education and practice in recruiting patients and in the procedures of random assignment of those patients to the control (enhanced care) arm or the manual therapy arm

After training, the physicians returned to their practices and worked on their newly learned skill for approximately 3 months before enrolling patients. At an agreed time they began to enroll patients and randomized them to enhanced care or enhanced care plus limited manual therapy using a blinded method.

Analysis

Using the patients of the 13 physicians active in both the original cohort study and the subsequent randomized trial of limited manual therapy, we compared outcomes between 3 groups of patients: (1) those whose usual care was only observed in the earlier cohort study, (2) those receiving enhanced care (control arm) in the randomized trial, and (3) those receiving enhanced care plus manual therapy (intervention arm) in the randomized trial. The major variables included in the analyses were age, sex, education, household income, duration of low back pain episode for more than 2 weeks, presence of sciatica, workers’ compensation status, and the Work Adaptation, Partnership, Growth, Affection, and Resolve Survey (a measure of job satisfaction).¹⁶ In each of the studies, outcome measures from repeated interviews included data on functional status over time and pain levels. Data on satisfaction with care, return to work, and time to functional and complete recovery were obtained either at 8 weeks or when the patients were better. The main outcome measures were the Roland-Morris adaptation of the Sickness Impact Profile (a 23-item scale with high scores indicating significant dysfunction), patients’ report of being all better or functionally better, and the date they were able to return to performing their usual daily activities.^10,14 Patient satisfaction measures were based on the scale developed by Cherkin and colleagues.⁸ The patient was the unit of analysis.

Pearson’s chi-square was used when comparing the 3 patient groups by physician performance and patient satisfaction, adjusting for baseline differences. Linear modeling was used to examine the relationship of manual therapy to functional status (Roland-Morris score). Cox proportional hazard modeling was undertaken to identify survival curves of time to functional recovery. In all analyses, standard errors were corrected for any intraclass correlation due to nonindependence of patients seeing the same physician.¹⁷

Results

The 13 physicians cared for 120 patients in the observational study and 191 patients in the randomized trial [Figure 1]. In each of the studies very similar numbers of patients were seen by each physician. Within the 2 arms of the randomized trial of manual therapy, baseline characteristics were essentially the same [Table 1].

There were some differences in baseline characteristics between patients in the cohort study and the randomized trial. More patients in the randomized trial (41.4% vs 25%, P=.01) had significantly higher severe baseline dysfunction (Roland-Morris score=16-23) than in the cohort study. The mean baseline Roland Morris score was 10.0 for the cohort study patients compared with 12.5 for the manual therapy study patients (P=.03). There were more patients on workers’ compensation in the cohort study (35%) than in the manual therapy study (16.3%, P=.002), and fewer had been employed in the previous 3 months (80.8% vs 90.0%, P=.02).

Patients in both arms of the randomized manual therapy trial were significantly more satisfied than their counterparts in the earlier cohort study in terms of how their physician gave care in the clinical encounter and whether they received effective advice [Table 2]. Patients participating in the manual therapy trial were also more satisfied with their pain relief, their physicians’ overall treatment of back pain, and their ability to perform activities of daily living. Specific items of clinical performance where differences were noted included: the physician took a detailed history; gave useful advice on pain, preventive measures, sleeping, and sitting strategies; and provided back exercises. There were no differences in patient ratings of the physician’s ability to listen effectively, perform a careful physical examination, and explain the cause of their back pain. There were no differences in general health status between the cohort and manual therapy study patients.

When adjusted for baseline function, the presence of sciatica, duration of pain, employment status, workers’ compensation, and income of more than $20,000, mean functional outcomes measured by Roland-Morris scores at 2 weeks were 8.2 for the patients in the cohort study and 6.7 in the manual therapy trial (P=.03); at 4 weeks mean scores were 7.2 for the cohort and 5.2 for patients in the manual therapy trial (P=.02), and at 8 weeks scores were 6.7 and 3.6, respectively (P=.002). These were clinically significant differences showing that all patients in the clinical trial had lower functional disability levels during an 8-week period than patients of the same physicians in the cohort study.

After controlling for baseline Roland-Morris score, duration of low back pain, sciatica, employment status, workers’ compensation, and income more than $20,000, time to functional recovery reported by the patient (using Cox proportional hazard modeling) between the 3 groups of patients was as follows: (1) manual therapy + enhanced care versus cohort: hazard ratio (HR)=1.16; 95% confidence interval (CI), 0.85-1.58; (2) enhanced care alone versus cohort: HR=1.13; 95% CI, 0.82-1.54; and (3) manual therapy + enhanced care versus enhanced care alone: HR=1.03; 95% CI, 0.75-1.40. None of these HRs were significant [Figure 2].

Discussion

Data from patients of 13 physicians involved in an observational study of back care, followed by a randomized trial of the effect of additional clinical skills developed by hands-on training, provide an unusual opportunity to study patient outcomes and physician performance over time. Similar eligibility and exclusion criteria, data collection methods, and outcome measures were used in both studies, allowing us to directly compare the variables of interest.

After controlling for confounding factors and physician clustering effect, patients in the clinical trial of manual therapy had significantly more baseline dysfunction at the index visit but still recovered more rapidly by self-report. After workshop training for the physicians, the proportion of patients who reported effective evaluation and management of their back problem increased substantially.

There were no differences in the 2 studies in the proportions of patients reporting on how effectively their physician listened to them, explained causation, and discussed occupational issues. These latter activities were not specifically addressed in the training workshop, which concentrated on clinical evaluation, manual techniques, and issues of physical rehabilitation. One would expect that a general placebo effect on patients of the training would increase all parameters of their perceptions of care, so the workshop may have produced specific learning effects on the clinicians. Despite a trend favoring the group receiving care from the “trained” physicians, there were no significant differences in how patients perceived the quality of the physician’s physical examination of their back—an item that was particularly emphasized in the workshops. This aspect of physician performance was already highly rated in the cohort group and may have been less likely to show improvement, given the need to know what constitutes a superior examination technique.

The improved reported performances of physicians and patient outcomes were noted in both arms of the manual therapy trial, compared with those of patients in the earlier cohort study, though there was no difference in time to full functional recovery between the 2 studies. The inconsistency between outcomes (Roland-Morris scores and time to functional recovery) is probably explainable by the fact that they measure different patient perceptions of back problems—specific versus global recovery. This inconsistency of measures has been noted in other studies of low back pain interventions.¹⁸

The improved outcomes of patients in the randomized trial compared with those in the cohort may be because of specific elements of the study applicable to all patients or may have been due to unmeasured baseline differences. The patients in the randomized trial were more impaired at baseline, which would tend toward worse rather than better outcomes if severity was not completely controlled for. There are several possible explanations for these findings: (1) workshop training did improve physician knowledge and skills; (2) involvement of patients in a clinical trial in which an intervention (enhanced care) was given in both arms of the study (one also receiving manual therapy) could heighten the overall positive effect; (3) the clinical trial involved more visits (average=3.6) than with patients in the first study (1.3)—this could add significantly to patient satisfaction and perceptions of good care; and (4) the recruited physicians were a special group of interested and motivated individuals who by their nature would produce better outcomes in a clinical trial.

Limitations

There are limitations to our findings and conclusions. There were differences in exclusion criteria that might have reduced the number of unhealthy people recruited into the manual therapy trial compared with the cohort study. This could have led to better outcomes, though we doubt that these differences made much impact given the average age of the patients (40 years).

Although we adjusted for baseline differences (income, workers’ compensation, and employment status), we are not sure why these differences occurred. One possibility is that the rapid population and commercial growth and high employment rates that occurred in the state of North Carolina in the time between the 2 studies improved the economic characteristics of patients presenting to the physicians over time. It is also possible that patients with higher incomes and fewer employment difficulties would be more likely to rate physician communication and management more highly and improve more rapidly.

The global effect on clinician knowledge, skills, and performance of a 3-year span of evolving clinical practice and involvement in low back pain studies (other than participating in a training workshop) cannot be quantified. For example, improved performance and outcomes might have been related to the publication of the AHCPR back pain guidelines in 1994, approximately 6 months before the start of the manual therapy study.¹⁶ However, at the low back pain workshop (where only the AHCPR “red flags” were presented) the participants seemed to be only minimally aware of these guidelines.

If the improved outcomes noted in this study were mainly because of the passage of time rather than workshop training, this provides an interesting insight into how practice changes for the better.

Patient satisfaction with care has been shown to correlate with outcomes.^19,20 Deyo and Diehl suggested that dissatisfaction with care for low back pain was related to failure to obtain an adequate explanation from the physician, while Cherkin and colleagues proposed that lack of confidence in management and negative attitudes of clinicians might be key issues to be addressed in achieving better outcomes.^8-10,21,22 However, didactic training to remedy these problems did not appear to be very effective in improving satisfaction. Following up on these suggestions, Smucker and coworkers,¹² using the large North Carolina cohort study, showed that clinician self-confidence (allopathic and chiropractic physicians) did not predict patient outcomes but commented that communication, time spent with the patient, and manual evaluation and treatment skills might be important variables affecting outcome.¹²

The margin of differences in outcomes shown in our preintervention and postintervention analysis leads us to suggest that allopathic physicians using a brief systematic evaluation and hands-on regional physical examination, sound advice on pain management and prevention, and an increasingly active exercise program can modestly improve early patient functioning and satisfaction in acute low back pain. Whether training in limited manual therapy adds to this benefit is unclear and must be taken into account in relation to the current expansion of CME in manipulative skills for allopathic physicians. The costs and benefits of providing this training—with the possibility of needing one extra office visit or more to fully implement enhanced care—would need to be assessed.²³

Alternative or complementary approaches to medical care are gaining loyalty from patients and increasing interest from the allopathic health care community.^1-4 In particular, professional organizations in the areas of acupuncture and manual therapy are offering and expanding continuing medical education (CME) programs in these fields for allopathic physicians, though there are few published data on their effectiveness.^5,6

The direct impact of CME on patients and clinical practice has been little studied, particularly in relation to the treatment of low back pain.⁷ Cherkin and colleagues^8,9 undertook an evaluation of a didactic CME program on low back pain in 1991 by studying patient satisfaction and provider attitudes. It appeared that the patients of providers who professed greater confidence in managing low back pain were more satisfied with their care, though negative attitudes previously expressed by clinicians toward low back pain did not change significantly after CME. In a discussion of this study it was suggested that patients might be seeking information and practical guidance rather than a cure or empathy.¹⁰ However, the investigators did not study the effects of modifying physical examination and manual skills in the care of these patients, factors that might play an important role in outcomes.

Although greater patient satisfaction has been associated with chiropractic care (which emphasizes manual skills) than that given by primary care physicians, there appears to be no association of satisfaction with practitioner self-confidence or days to functional recovery of the patient.^11,12

We developed a workshop for generalist clinicians in the skills of assessment, limited manual therapy, and a graded exercise program, and in a randomized controlled trial evaluated clinician self-efficacy and patient outcomes for acute low back pain.^13,14 We demonstrated that allopathic generalist physicians could be effectively trained in limited manual therapy with self-reported increased competence in managing low back pain. The patients receiving therapy showed a trend toward feeling completely better more quickly but reported no greater satisfaction or objective functional improvement in terms of activities than patients in the control group who were receiving only high-quality conventional care through workshop training.¹⁴

Of 31 physicians recruited into the manual therapy trial, 13 had previously been involved in a cohort study of utilization and back care therapy given by 208 practitioners (115 primary care generalists) to 1633 patients (644 patients of primary care generalists).¹¹ We examined the outcomes of patients with low back pain from the practices of these 13 physicians before (data from the cohort study) and after an intensive hands-on training workshop (data from the manual therapy trial), using similar patient recruitment methods and evaluation instruments ([Figure 1]. Our hypothesis was that this training in manual and assessment skills would improve patient outcomes and satisfaction.

Methods

The initial cohort study was undertaken to examine the prevalence, care seeking, and outcomes of acute low back pain in the state of North Carolina. No interventions were undertaken in this study. Methods and measures used in its implementation have been described previously by Carey and coworkers.¹¹ Clinicians were randomly selected from medical and chiropractic licensure files (primary care generalists, chiropractic physicians, and orthopedic surgeons). To be included in the cohort study clinicians had to see ambulatory patients at least 50% of the time and provide first contact care for acute low back pain. Of the 208 clinicians from different disciplines recruited into the study, 115 were primary care physicians. During a 10-month period consecutive patients with acute or subacute low back pain (<10 weeks) were enrolled unless they had received previous care for the episode, had received previous back surgery, had a history of cancer, were pregnant, had no telephone, or were unable to speak English. Patients were contacted by telephone after the visit by staff members of the University of North Carolina Survey Research Unit, and interviews were undertaken at baseline, 2, 4, 8, 12, and 24 weeks.

These interviews included questions about details of the back pain episode, medications and other therapies used, tests performed, work and compensation status, demographic data, and income level. Outcome data included responses to the 23-item Roland-Morris Back Disability Questionnaire.¹⁵ We also asked when the patients considered they were able to perform their usual daily activities after the back pain. Patient satisfaction was assessed on how well the physician communicated, listened, gave information and explained the cause of back pain, whether a detailed history was taken and the back examined carefully, and if advice was given on pain management, prevention, and activities of sleeping and sitting (yes/no responses). Other satisfaction items (overall treatment, pain relief provided, and patient abilities to walk, socialize, and work) were rated on a 5-point Likert scale (poor, fair, good, very good, or excellent). This was subsequently adapted to a dichotomous response. Clinical and utilization data were obtained from charts in the physicians’ offices to allow validation of survey variables.

The randomized trial of the effectiveness of limited manual therapy was started in 1995, 12 months after the closure of the cohort study. Patients were recruited by 31 generalist physicians (13 from the earlier cohort study and 18 volunteers from the 630 physicians on the North Carolina physician master file). The same inclusion and exclusion criteria from the cohort study were used except that the acceptable age range was 21 to 65 (compared with 75 years) and patients had no osteopenia, severe arthritis, morbid obesity, or neurological deficits and had not received previous manual therapy by the physician. These additional exclusion criteria were necessary to avoid possible adverse effects of manual therapy in the presence of disease and to eliminate patient bias of a preference for manual therapy based on previous experience.

For each arm of the study, after the first office visit the identical telephone interview questions and schedule used in the cohort study were implemented for up to 8 weeks. Chart abstraction methods and variables were also identical except that additional data were collected on the specifics of manual therapy given at each visit.

Two sequential weekend workshops with a refresher session for each of the 31 physicians (developed and given to 9 physicians per workshop by 3 family physicians skilled in manual therapy) were implemented before the start of the clinical trial. The purpose was to train these physicians in quality care for low back pain (explained to patients as enhanced care—the control arm) and in standardized limited manual therapy (the main component of the intervention arm).¹³ The term “enhanced care” was developed to minimize the impression for patients during randomization that they might either receive something special (manual therapy) or just routine care—both options needed to appear to be special to reduce placebo bias.

Training for the enhanced low back pain care arm included physician education in (1) the directed history and physical examination using Agency for Health Care Policy and Research (AHCPR) guidelines,¹⁶ (2) review of the efficacy of imaging and laboratory testing, (3) review of the efficacy of treatment modalities, and (4) use of specially designed patient handouts emphasizing progressive exercises, daily activities, and early return to function.

Training for the limited manual therapy arm included the enhanced low back pain skills plus:

• manual therapy, consisting of: (1) principles of manual therapy and explanatory models, (2) instruction in motion testing, and (3) instruction in limited manual therapy skills (soft-tissue, muscle energy, and high-velocity low-amplitude techniques involving psoas and piriformis muscles—lumbar spine, lumbosacral junction, and sacroiliac joints)
• workshop training and demonstrated competence in low back care on simulated patients
• guidance and practice in integrating limited manual therapy into the office visit
• education and practice in recruiting patients and in the procedures of random assignment of those patients to the control (enhanced care) arm or the manual therapy arm

After training, the physicians returned to their practices and worked on their newly learned skill for approximately 3 months before enrolling patients. At an agreed time they began to enroll patients and randomized them to enhanced care or enhanced care plus limited manual therapy using a blinded method.

Analysis

Using the patients of the 13 physicians active in both the original cohort study and the subsequent randomized trial of limited manual therapy, we compared outcomes between 3 groups of patients: (1) those whose usual care was only observed in the earlier cohort study, (2) those receiving enhanced care (control arm) in the randomized trial, and (3) those receiving enhanced care plus manual therapy (intervention arm) in the randomized trial. The major variables included in the analyses were age, sex, education, household income, duration of low back pain episode for more than 2 weeks, presence of sciatica, workers’ compensation status, and the Work Adaptation, Partnership, Growth, Affection, and Resolve Survey (a measure of job satisfaction).¹⁶ In each of the studies, outcome measures from repeated interviews included data on functional status over time and pain levels. Data on satisfaction with care, return to work, and time to functional and complete recovery were obtained either at 8 weeks or when the patients were better. The main outcome measures were the Roland-Morris adaptation of the Sickness Impact Profile (a 23-item scale with high scores indicating significant dysfunction), patients’ report of being all better or functionally better, and the date they were able to return to performing their usual daily activities.^10,14 Patient satisfaction measures were based on the scale developed by Cherkin and colleagues.⁸ The patient was the unit of analysis.

Pearson’s chi-square was used when comparing the 3 patient groups by physician performance and patient satisfaction, adjusting for baseline differences. Linear modeling was used to examine the relationship of manual therapy to functional status (Roland-Morris score). Cox proportional hazard modeling was undertaken to identify survival curves of time to functional recovery. In all analyses, standard errors were corrected for any intraclass correlation due to nonindependence of patients seeing the same physician.¹⁷

Results

The 13 physicians cared for 120 patients in the observational study and 191 patients in the randomized trial [Figure 1]. In each of the studies very similar numbers of patients were seen by each physician. Within the 2 arms of the randomized trial of manual therapy, baseline characteristics were essentially the same [Table 1].

There were some differences in baseline characteristics between patients in the cohort study and the randomized trial. More patients in the randomized trial (41.4% vs 25%, P=.01) had significantly higher severe baseline dysfunction (Roland-Morris score=16-23) than in the cohort study. The mean baseline Roland Morris score was 10.0 for the cohort study patients compared with 12.5 for the manual therapy study patients (P=.03). There were more patients on workers’ compensation in the cohort study (35%) than in the manual therapy study (16.3%, P=.002), and fewer had been employed in the previous 3 months (80.8% vs 90.0%, P=.02).

Patients in both arms of the randomized manual therapy trial were significantly more satisfied than their counterparts in the earlier cohort study in terms of how their physician gave care in the clinical encounter and whether they received effective advice [Table 2]. Patients participating in the manual therapy trial were also more satisfied with their pain relief, their physicians’ overall treatment of back pain, and their ability to perform activities of daily living. Specific items of clinical performance where differences were noted included: the physician took a detailed history; gave useful advice on pain, preventive measures, sleeping, and sitting strategies; and provided back exercises. There were no differences in patient ratings of the physician’s ability to listen effectively, perform a careful physical examination, and explain the cause of their back pain. There were no differences in general health status between the cohort and manual therapy study patients.

When adjusted for baseline function, the presence of sciatica, duration of pain, employment status, workers’ compensation, and income of more than $20,000, mean functional outcomes measured by Roland-Morris scores at 2 weeks were 8.2 for the patients in the cohort study and 6.7 in the manual therapy trial (P=.03); at 4 weeks mean scores were 7.2 for the cohort and 5.2 for patients in the manual therapy trial (P=.02), and at 8 weeks scores were 6.7 and 3.6, respectively (P=.002). These were clinically significant differences showing that all patients in the clinical trial had lower functional disability levels during an 8-week period than patients of the same physicians in the cohort study.

After controlling for baseline Roland-Morris score, duration of low back pain, sciatica, employment status, workers’ compensation, and income more than $20,000, time to functional recovery reported by the patient (using Cox proportional hazard modeling) between the 3 groups of patients was as follows: (1) manual therapy + enhanced care versus cohort: hazard ratio (HR)=1.16; 95% confidence interval (CI), 0.85-1.58; (2) enhanced care alone versus cohort: HR=1.13; 95% CI, 0.82-1.54; and (3) manual therapy + enhanced care versus enhanced care alone: HR=1.03; 95% CI, 0.75-1.40. None of these HRs were significant [Figure 2].

Discussion

Data from patients of 13 physicians involved in an observational study of back care, followed by a randomized trial of the effect of additional clinical skills developed by hands-on training, provide an unusual opportunity to study patient outcomes and physician performance over time. Similar eligibility and exclusion criteria, data collection methods, and outcome measures were used in both studies, allowing us to directly compare the variables of interest.

After controlling for confounding factors and physician clustering effect, patients in the clinical trial of manual therapy had significantly more baseline dysfunction at the index visit but still recovered more rapidly by self-report. After workshop training for the physicians, the proportion of patients who reported effective evaluation and management of their back problem increased substantially.

There were no differences in the 2 studies in the proportions of patients reporting on how effectively their physician listened to them, explained causation, and discussed occupational issues. These latter activities were not specifically addressed in the training workshop, which concentrated on clinical evaluation, manual techniques, and issues of physical rehabilitation. One would expect that a general placebo effect on patients of the training would increase all parameters of their perceptions of care, so the workshop may have produced specific learning effects on the clinicians. Despite a trend favoring the group receiving care from the “trained” physicians, there were no significant differences in how patients perceived the quality of the physician’s physical examination of their back—an item that was particularly emphasized in the workshops. This aspect of physician performance was already highly rated in the cohort group and may have been less likely to show improvement, given the need to know what constitutes a superior examination technique.

The improved reported performances of physicians and patient outcomes were noted in both arms of the manual therapy trial, compared with those of patients in the earlier cohort study, though there was no difference in time to full functional recovery between the 2 studies. The inconsistency between outcomes (Roland-Morris scores and time to functional recovery) is probably explainable by the fact that they measure different patient perceptions of back problems—specific versus global recovery. This inconsistency of measures has been noted in other studies of low back pain interventions.¹⁸

The improved outcomes of patients in the randomized trial compared with those in the cohort may be because of specific elements of the study applicable to all patients or may have been due to unmeasured baseline differences. The patients in the randomized trial were more impaired at baseline, which would tend toward worse rather than better outcomes if severity was not completely controlled for. There are several possible explanations for these findings: (1) workshop training did improve physician knowledge and skills; (2) involvement of patients in a clinical trial in which an intervention (enhanced care) was given in both arms of the study (one also receiving manual therapy) could heighten the overall positive effect; (3) the clinical trial involved more visits (average=3.6) than with patients in the first study (1.3)—this could add significantly to patient satisfaction and perceptions of good care; and (4) the recruited physicians were a special group of interested and motivated individuals who by their nature would produce better outcomes in a clinical trial.

Limitations

There are limitations to our findings and conclusions. There were differences in exclusion criteria that might have reduced the number of unhealthy people recruited into the manual therapy trial compared with the cohort study. This could have led to better outcomes, though we doubt that these differences made much impact given the average age of the patients (40 years).

Although we adjusted for baseline differences (income, workers’ compensation, and employment status), we are not sure why these differences occurred. One possibility is that the rapid population and commercial growth and high employment rates that occurred in the state of North Carolina in the time between the 2 studies improved the economic characteristics of patients presenting to the physicians over time. It is also possible that patients with higher incomes and fewer employment difficulties would be more likely to rate physician communication and management more highly and improve more rapidly.

The global effect on clinician knowledge, skills, and performance of a 3-year span of evolving clinical practice and involvement in low back pain studies (other than participating in a training workshop) cannot be quantified. For example, improved performance and outcomes might have been related to the publication of the AHCPR back pain guidelines in 1994, approximately 6 months before the start of the manual therapy study.¹⁶ However, at the low back pain workshop (where only the AHCPR “red flags” were presented) the participants seemed to be only minimally aware of these guidelines.

If the improved outcomes noted in this study were mainly because of the passage of time rather than workshop training, this provides an interesting insight into how practice changes for the better.

Patient satisfaction with care has been shown to correlate with outcomes.^19,20 Deyo and Diehl suggested that dissatisfaction with care for low back pain was related to failure to obtain an adequate explanation from the physician, while Cherkin and colleagues proposed that lack of confidence in management and negative attitudes of clinicians might be key issues to be addressed in achieving better outcomes.^8-10,21,22 However, didactic training to remedy these problems did not appear to be very effective in improving satisfaction. Following up on these suggestions, Smucker and coworkers,¹² using the large North Carolina cohort study, showed that clinician self-confidence (allopathic and chiropractic physicians) did not predict patient outcomes but commented that communication, time spent with the patient, and manual evaluation and treatment skills might be important variables affecting outcome.¹²

The margin of differences in outcomes shown in our preintervention and postintervention analysis leads us to suggest that allopathic physicians using a brief systematic evaluation and hands-on regional physical examination, sound advice on pain management and prevention, and an increasingly active exercise program can modestly improve early patient functioning and satisfaction in acute low back pain. Whether training in limited manual therapy adds to this benefit is unclear and must be taken into account in relation to the current expansion of CME in manipulative skills for allopathic physicians. The costs and benefits of providing this training—with the possibility of needing one extra office visit or more to fully implement enhanced care—would need to be assessed.²³

More than 2.5 million cases of pneumonia are diagnosed each year in the United States,¹ and the hospitalization rate ranges from fewer than 1 of every 1000 persons aged 35 to 44 years to more than 11 of every 1000 persons aged 75 years or older.^2,3 The annual direct cost of treating pneumonia is estimated to be $14 billion, of which $8 billion is for inpatient care. Pneumonia is also responsible for $9 billion in lost wages^4-6 and is the sixth leading cause of death in the United States⁶—and these numbers may be increasing.^7,8 It is the most frequently encountered life-threatening infectious disease in the United States.⁹ Accurate diagnosis, treatment, and follow-up are essential for treating this serious disease.

Physicians often obtain chest radiographs in the course of evaluating acutely ill febrile patients with suspected pneumonia. Many of these patients have some degree of fluid volume depletion secondary to the effects of the acute illness, as well as chronic illnesses and concomitant medication use. It is widely believed that the radiographic findings of community-acquired pneumonia (CAP) may be masked by volume depletion and that repletion may facilitate the expression of infiltrates on posthydration chest radiographs.⁴ This phenomenon has been noted in a recent publication about CAP,⁵ and we have encountered this clinical concept in discussions with colleagues.

Unfortunately, few data are available to support or refute the concept that fluid volume status affects the radiographic findings in CAP. A MEDLINE database search of the literature from 1966 to the present involving numerous combinations of key words yielded no human studies on this subject. One case study was reported of a dehydrated older person with a normal chest radiograph on admission for suspected pneumonia who developed a lobar infiltrate after rehydration.¹⁰

Our goal for this initial retrospective study was to determine if a relation exists between initial volume depletion, repletion, and subsequent radiographic findings of CAP. This information could be useful in determining the appropriate use of repeat radiographs in the evaluation of patients with clinical pneumonia.

Methods

The hospital records of 376 consecutive inpatient encounters with the discharge diagnosis of CAP between June 1996 and June 1998 were drawn from a 500-bed community teaching hospital. We reviewed them for the following inclusion criteria: 18 years of age or older; nonpregnant; chest radiographs obtained on admission and within 96 hours of admission; and at least 2 clinical indicators to support the radiographic diagnosis of pneumonia (fever >37.8 °C, leukocytosis, tachypnea, vomiting, pleuritic pain, productive cough, positive blood cultures.). A total of 125 records met the inclusion criteria. Each of these records was evaluated for reported chest radiograph results, fluids administered during the first 24 hours (intravenous and oral), blood urea nitrogen (BUN) level, repeat BUN level after 24 to 48 hours, admission creatinine level, repeat creatinine level after 24 to 48 hours, age, sex, race, weight, and admission serum sodium level.

We grouped the patients according to reported change in chest radiographs. The progression group (P) had chest radiograph reports indicating worsening appearance of infiltrates after fluid administration. The no-progression (NP) group had reports indicating either no change or improvement in the appearance of the infiltrates after fluid administration.

All continuous variables were analyzed with a 2-sample Student t test for equal means between the P and NP groups. We used stepwise logistic regression to determine associations between potential predictor variables and between the groups. Continuous variables were graphed against the logit to access linearity, and we examined the Hosmer-Lemeshaw statistic¹¹ to help determine adequacy of the model. We analyzed data using the PROC t test and the PROC logistic in the SAS 6.12 software.¹²

Results

A total of 125 records satisfied study admission criteria. Our record review revealed that 42 patients had radiographic worsening (P), and the remaining 83 had no change or improvement radiographically (NP). The cumulative data for each group is shown in [Table 1].

We found that admission BUN levels were significantly greater for the P group (mean=12.24 mmol/L [34.3 mg/dL]) than the NP group (mean=8.57 mmol/L [24.0 mg/dL]). Fluid intake during the first 48 hours after admission was also significantly greater for the P group (mean=5824.0 mL) than the NP group (mean=4764.3 mL). There was marginal significance (P=.053) between the 2 groups for age. There was no significant difference between the 2 groups for sodium levels on admission. Thus, those patients in the P group were more likely to have a higher admission BUN level, a higher fluid intake in the first 48 hours following admission, and to be younger than the NP patients.

Results of the logistic regression model showed that few factors were related to the change in radiographic findings, and no combination of factors seemed to be a good predictor for this change. The model poorly predicted which patients would show progressive radiographic infiltrates on the basis of age and admission BUN level. Although a significant difference was found between the 2 groups with respect to the first 48-hour fluid intake, this factor was dropped from the logistic regression model because it paralleled the admission BUN level [Table 2]. On the basis of a patient’s age and admission BUN level, the model could correctly predict only 28% of patients who would show progressive radiographic changes. In contrast, the same model correctly predicted 92% of patients who would show no change or improvement on subsequent radiographic studies. The c statistic (which is equivalent to the area under the curve for the receiver-operating characteristic) equals 0.68.

Discussion

The diagnosis of CAP is usually made by the presence of signs and symptoms of lower respiratory tract infection coupled with the finding of at least one acute opacity on the chest radiograph.^1,13 Some clinicians believe that such an opacity is necessary for the diagnosis of CAP, and they consider it the gold standard for the diagnosis in conjunction with supporting data from the history and physical examination.^2,4,5 However, radiographic and clinical findings are not always consistent. Heckerling¹⁴ found that no clinical findings other than auscultory abnormalities could be used to predict the presence of pneumonia on chest radiography. A review of the literature by Metlay and colleagues¹⁵ found that there is no reliable combination of findings that can rule in the diagnosis of pneumonia. Up to 25% of young adults may not have auscultory findings of pneumonia.⁴ It is well documented that elderly patients may initially manifest no fever, minimal respiratory symptoms, and minimal radiographic infiltrates despite laboratory documentation of pneumonia.^16,17 Patients with neutropenia or infections with atypical organisms or viruses may not manifest an infiltrate on chest radiograph until several days after clinical symptoms appear.^5,9,18 Dehydration or volume depletion has also been shown to produce this phenomenon.^4,5,10

Pneumonia has been defined as inflammation and consolidation of lung tissue due to an infectious agent. When sufficient fluid has collected in the normally air-filled spaces, X-ray beam attenuation occurs and an opacity appears on a chest radiograph. It has been postulated that decreased pulmonary hydrostatic pressure and increased pulmonary oncotic pressure secondary to dehydration or volume depletion may diminish the flux of fluid into the alveoli and interstitium and thus delay or alter the radiographic findings of pneumonia.¹⁹ Cooligan and coworkers²⁰ found that small increases in pulmonary capillary wedge pressure increased wet lung weight in euvolemic dogs with pneumococcal pneumonia, but no radiographic evaluation was performed. Caldwell and colleagues¹⁹ found that acute intravascular volume depletion did not affect the radiographic evolution of pneumococcal pneumonia in dogs, but they did not evaluate the effects of fluid volume repletion. Caldwell and coworkers also noted that on retrospective chart review of 20 consecutive human patients with pneumonia and the diagnosis of dehydration, all had an infiltrate apparent radiographically at the time of admission. The effects of rehydration were not addressed in these patients.

Our retrospective study appears to be the first to specifically address the relationship between pneumonia, hydration status, rehydration, and the effect on the radiographic appearance of pneumonia. Our results show an association between an elevated admission BUN level, volume of fluids given, and a worsening of the radiographic appearance of CAP. A future prospective study could teach us more about diagnostic strategies for CAP, including any situations in which repeated radiographs are helpful.

We were not able to address the question of whether the radiographic findings of pneumonia can be completely masked by dehydration and subsequently expressed by rehydration; our findings, however, suggest that there may be some basis for this premise.

Limitations

Our study has a number of limitations. There is an inherent selection bias in our retrospective process: Patients who received only one chest radiograph because of misdiagnosis, management decision, or other reasons were excluded from the study. Follow-up chest radiographs within a few days are generally not indicated and are not obtained in clinically stable patients who may only show worsening or improving patterns on radiograph.^6,13 Also, the retrospective nature of the study precluded the collection of a uniform database or uniform evaluation of possible etiologic agents. Viral and mycoplasma pneumonias are well documented to have a lag between onset of symptoms and evolution of radiographic findings,⁹ that may have led to inappropriate group assignment or complete exclusion from the study. The determination of the presence, progression, or nonprogression of radiographic findings is somewhat subjective and may affect our results. Our data are based on the reported radiograph interpretation from a randomly assigned radiologist. Interobserver agreement between radiologists for the determination of infiltrates on chest radiographs was found to be only 80% in a recent study.²¹ Finally, hydration status was difficult to determine. Although the BUN level was significantly higher in the P group than in the NP group, BUN level is not a specific marker for dehydration or intravascular volume status. The higher serum sodium levels of the P group, while not achieving statistical significance, may have been a better marker of total body water status. Although fluid volume depletion and dehydration are very different clinical problems,²² the determination of which condition was present, and the severity of that condition, was not possible in our study. Although statistically significant, the difference in fluid intake between the 2 groups may not be clinically significant. Furthermore, dehydration alone has been shown to adversely affect lung host defense in rats.²³ If applicable to humans, this could also unexpectedly affect our results.

Conclusions

To our knowledge, this study is the first to try to define the relationship of fluid volume status and radiographic progression of CAP in humans. The retrospective nature of our study leads to several limitations and potential sources of bias, but it appears to show that there is a correlation between markers of fluid volume status, hydration, and the evolution of the radiographic findings. A prospective clinical study is needed to further define this relationship.

Acknowledgments

Our work was funded by a grant from the Clinical Research Center of the Medical Center of Central Georgia. We wish to thank Jennifer K. Rayhill for her assistance in preparing this manuscript.

More than 2.5 million cases of pneumonia are diagnosed each year in the United States,¹ and the hospitalization rate ranges from fewer than 1 of every 1000 persons aged 35 to 44 years to more than 11 of every 1000 persons aged 75 years or older.^2,3 The annual direct cost of treating pneumonia is estimated to be $14 billion, of which $8 billion is for inpatient care. Pneumonia is also responsible for $9 billion in lost wages^4-6 and is the sixth leading cause of death in the United States⁶—and these numbers may be increasing.^7,8 It is the most frequently encountered life-threatening infectious disease in the United States.⁹ Accurate diagnosis, treatment, and follow-up are essential for treating this serious disease.

Physicians often obtain chest radiographs in the course of evaluating acutely ill febrile patients with suspected pneumonia. Many of these patients have some degree of fluid volume depletion secondary to the effects of the acute illness, as well as chronic illnesses and concomitant medication use. It is widely believed that the radiographic findings of community-acquired pneumonia (CAP) may be masked by volume depletion and that repletion may facilitate the expression of infiltrates on posthydration chest radiographs.⁴ This phenomenon has been noted in a recent publication about CAP,⁵ and we have encountered this clinical concept in discussions with colleagues.

Unfortunately, few data are available to support or refute the concept that fluid volume status affects the radiographic findings in CAP. A MEDLINE database search of the literature from 1966 to the present involving numerous combinations of key words yielded no human studies on this subject. One case study was reported of a dehydrated older person with a normal chest radiograph on admission for suspected pneumonia who developed a lobar infiltrate after rehydration.¹⁰

Our goal for this initial retrospective study was to determine if a relation exists between initial volume depletion, repletion, and subsequent radiographic findings of CAP. This information could be useful in determining the appropriate use of repeat radiographs in the evaluation of patients with clinical pneumonia.

Methods

The hospital records of 376 consecutive inpatient encounters with the discharge diagnosis of CAP between June 1996 and June 1998 were drawn from a 500-bed community teaching hospital. We reviewed them for the following inclusion criteria: 18 years of age or older; nonpregnant; chest radiographs obtained on admission and within 96 hours of admission; and at least 2 clinical indicators to support the radiographic diagnosis of pneumonia (fever >37.8 °C, leukocytosis, tachypnea, vomiting, pleuritic pain, productive cough, positive blood cultures.). A total of 125 records met the inclusion criteria. Each of these records was evaluated for reported chest radiograph results, fluids administered during the first 24 hours (intravenous and oral), blood urea nitrogen (BUN) level, repeat BUN level after 24 to 48 hours, admission creatinine level, repeat creatinine level after 24 to 48 hours, age, sex, race, weight, and admission serum sodium level.

We grouped the patients according to reported change in chest radiographs. The progression group (P) had chest radiograph reports indicating worsening appearance of infiltrates after fluid administration. The no-progression (NP) group had reports indicating either no change or improvement in the appearance of the infiltrates after fluid administration.

All continuous variables were analyzed with a 2-sample Student t test for equal means between the P and NP groups. We used stepwise logistic regression to determine associations between potential predictor variables and between the groups. Continuous variables were graphed against the logit to access linearity, and we examined the Hosmer-Lemeshaw statistic¹¹ to help determine adequacy of the model. We analyzed data using the PROC t test and the PROC logistic in the SAS 6.12 software.¹²

Results

A total of 125 records satisfied study admission criteria. Our record review revealed that 42 patients had radiographic worsening (P), and the remaining 83 had no change or improvement radiographically (NP). The cumulative data for each group is shown in [Table 1].

We found that admission BUN levels were significantly greater for the P group (mean=12.24 mmol/L [34.3 mg/dL]) than the NP group (mean=8.57 mmol/L [24.0 mg/dL]). Fluid intake during the first 48 hours after admission was also significantly greater for the P group (mean=5824.0 mL) than the NP group (mean=4764.3 mL). There was marginal significance (P=.053) between the 2 groups for age. There was no significant difference between the 2 groups for sodium levels on admission. Thus, those patients in the P group were more likely to have a higher admission BUN level, a higher fluid intake in the first 48 hours following admission, and to be younger than the NP patients.

Results of the logistic regression model showed that few factors were related to the change in radiographic findings, and no combination of factors seemed to be a good predictor for this change. The model poorly predicted which patients would show progressive radiographic infiltrates on the basis of age and admission BUN level. Although a significant difference was found between the 2 groups with respect to the first 48-hour fluid intake, this factor was dropped from the logistic regression model because it paralleled the admission BUN level [Table 2]. On the basis of a patient’s age and admission BUN level, the model could correctly predict only 28% of patients who would show progressive radiographic changes. In contrast, the same model correctly predicted 92% of patients who would show no change or improvement on subsequent radiographic studies. The c statistic (which is equivalent to the area under the curve for the receiver-operating characteristic) equals 0.68.

Discussion

The diagnosis of CAP is usually made by the presence of signs and symptoms of lower respiratory tract infection coupled with the finding of at least one acute opacity on the chest radiograph.^1,13 Some clinicians believe that such an opacity is necessary for the diagnosis of CAP, and they consider it the gold standard for the diagnosis in conjunction with supporting data from the history and physical examination.^2,4,5 However, radiographic and clinical findings are not always consistent. Heckerling¹⁴ found that no clinical findings other than auscultory abnormalities could be used to predict the presence of pneumonia on chest radiography. A review of the literature by Metlay and colleagues¹⁵ found that there is no reliable combination of findings that can rule in the diagnosis of pneumonia. Up to 25% of young adults may not have auscultory findings of pneumonia.⁴ It is well documented that elderly patients may initially manifest no fever, minimal respiratory symptoms, and minimal radiographic infiltrates despite laboratory documentation of pneumonia.^16,17 Patients with neutropenia or infections with atypical organisms or viruses may not manifest an infiltrate on chest radiograph until several days after clinical symptoms appear.^5,9,18 Dehydration or volume depletion has also been shown to produce this phenomenon.^4,5,10

Pneumonia has been defined as inflammation and consolidation of lung tissue due to an infectious agent. When sufficient fluid has collected in the normally air-filled spaces, X-ray beam attenuation occurs and an opacity appears on a chest radiograph. It has been postulated that decreased pulmonary hydrostatic pressure and increased pulmonary oncotic pressure secondary to dehydration or volume depletion may diminish the flux of fluid into the alveoli and interstitium and thus delay or alter the radiographic findings of pneumonia.¹⁹ Cooligan and coworkers²⁰ found that small increases in pulmonary capillary wedge pressure increased wet lung weight in euvolemic dogs with pneumococcal pneumonia, but no radiographic evaluation was performed. Caldwell and colleagues¹⁹ found that acute intravascular volume depletion did not affect the radiographic evolution of pneumococcal pneumonia in dogs, but they did not evaluate the effects of fluid volume repletion. Caldwell and coworkers also noted that on retrospective chart review of 20 consecutive human patients with pneumonia and the diagnosis of dehydration, all had an infiltrate apparent radiographically at the time of admission. The effects of rehydration were not addressed in these patients.

Our retrospective study appears to be the first to specifically address the relationship between pneumonia, hydration status, rehydration, and the effect on the radiographic appearance of pneumonia. Our results show an association between an elevated admission BUN level, volume of fluids given, and a worsening of the radiographic appearance of CAP. A future prospective study could teach us more about diagnostic strategies for CAP, including any situations in which repeated radiographs are helpful.

We were not able to address the question of whether the radiographic findings of pneumonia can be completely masked by dehydration and subsequently expressed by rehydration; our findings, however, suggest that there may be some basis for this premise.

Limitations

Our study has a number of limitations. There is an inherent selection bias in our retrospective process: Patients who received only one chest radiograph because of misdiagnosis, management decision, or other reasons were excluded from the study. Follow-up chest radiographs within a few days are generally not indicated and are not obtained in clinically stable patients who may only show worsening or improving patterns on radiograph.^6,13 Also, the retrospective nature of the study precluded the collection of a uniform database or uniform evaluation of possible etiologic agents. Viral and mycoplasma pneumonias are well documented to have a lag between onset of symptoms and evolution of radiographic findings,⁹ that may have led to inappropriate group assignment or complete exclusion from the study. The determination of the presence, progression, or nonprogression of radiographic findings is somewhat subjective and may affect our results. Our data are based on the reported radiograph interpretation from a randomly assigned radiologist. Interobserver agreement between radiologists for the determination of infiltrates on chest radiographs was found to be only 80% in a recent study.²¹ Finally, hydration status was difficult to determine. Although the BUN level was significantly higher in the P group than in the NP group, BUN level is not a specific marker for dehydration or intravascular volume status. The higher serum sodium levels of the P group, while not achieving statistical significance, may have been a better marker of total body water status. Although fluid volume depletion and dehydration are very different clinical problems,²² the determination of which condition was present, and the severity of that condition, was not possible in our study. Although statistically significant, the difference in fluid intake between the 2 groups may not be clinically significant. Furthermore, dehydration alone has been shown to adversely affect lung host defense in rats.²³ If applicable to humans, this could also unexpectedly affect our results.

Conclusions

To our knowledge, this study is the first to try to define the relationship of fluid volume status and radiographic progression of CAP in humans. The retrospective nature of our study leads to several limitations and potential sources of bias, but it appears to show that there is a correlation between markers of fluid volume status, hydration, and the evolution of the radiographic findings. A prospective clinical study is needed to further define this relationship.

Acknowledgments

Our work was funded by a grant from the Clinical Research Center of the Medical Center of Central Georgia. We wish to thank Jennifer K. Rayhill for her assistance in preparing this manuscript.

More than 2.5 million cases of pneumonia are diagnosed each year in the United States,¹ and the hospitalization rate ranges from fewer than 1 of every 1000 persons aged 35 to 44 years to more than 11 of every 1000 persons aged 75 years or older.^2,3 The annual direct cost of treating pneumonia is estimated to be $14 billion, of which $8 billion is for inpatient care. Pneumonia is also responsible for $9 billion in lost wages^4-6 and is the sixth leading cause of death in the United States⁶—and these numbers may be increasing.^7,8 It is the most frequently encountered life-threatening infectious disease in the United States.⁹ Accurate diagnosis, treatment, and follow-up are essential for treating this serious disease.

Physicians often obtain chest radiographs in the course of evaluating acutely ill febrile patients with suspected pneumonia. Many of these patients have some degree of fluid volume depletion secondary to the effects of the acute illness, as well as chronic illnesses and concomitant medication use. It is widely believed that the radiographic findings of community-acquired pneumonia (CAP) may be masked by volume depletion and that repletion may facilitate the expression of infiltrates on posthydration chest radiographs.⁴ This phenomenon has been noted in a recent publication about CAP,⁵ and we have encountered this clinical concept in discussions with colleagues.

Unfortunately, few data are available to support or refute the concept that fluid volume status affects the radiographic findings in CAP. A MEDLINE database search of the literature from 1966 to the present involving numerous combinations of key words yielded no human studies on this subject. One case study was reported of a dehydrated older person with a normal chest radiograph on admission for suspected pneumonia who developed a lobar infiltrate after rehydration.¹⁰

Our goal for this initial retrospective study was to determine if a relation exists between initial volume depletion, repletion, and subsequent radiographic findings of CAP. This information could be useful in determining the appropriate use of repeat radiographs in the evaluation of patients with clinical pneumonia.

Methods

The hospital records of 376 consecutive inpatient encounters with the discharge diagnosis of CAP between June 1996 and June 1998 were drawn from a 500-bed community teaching hospital. We reviewed them for the following inclusion criteria: 18 years of age or older; nonpregnant; chest radiographs obtained on admission and within 96 hours of admission; and at least 2 clinical indicators to support the radiographic diagnosis of pneumonia (fever >37.8 °C, leukocytosis, tachypnea, vomiting, pleuritic pain, productive cough, positive blood cultures.). A total of 125 records met the inclusion criteria. Each of these records was evaluated for reported chest radiograph results, fluids administered during the first 24 hours (intravenous and oral), blood urea nitrogen (BUN) level, repeat BUN level after 24 to 48 hours, admission creatinine level, repeat creatinine level after 24 to 48 hours, age, sex, race, weight, and admission serum sodium level.

We grouped the patients according to reported change in chest radiographs. The progression group (P) had chest radiograph reports indicating worsening appearance of infiltrates after fluid administration. The no-progression (NP) group had reports indicating either no change or improvement in the appearance of the infiltrates after fluid administration.

All continuous variables were analyzed with a 2-sample Student t test for equal means between the P and NP groups. We used stepwise logistic regression to determine associations between potential predictor variables and between the groups. Continuous variables were graphed against the logit to access linearity, and we examined the Hosmer-Lemeshaw statistic¹¹ to help determine adequacy of the model. We analyzed data using the PROC t test and the PROC logistic in the SAS 6.12 software.¹²

Results

A total of 125 records satisfied study admission criteria. Our record review revealed that 42 patients had radiographic worsening (P), and the remaining 83 had no change or improvement radiographically (NP). The cumulative data for each group is shown in [Table 1].

We found that admission BUN levels were significantly greater for the P group (mean=12.24 mmol/L [34.3 mg/dL]) than the NP group (mean=8.57 mmol/L [24.0 mg/dL]). Fluid intake during the first 48 hours after admission was also significantly greater for the P group (mean=5824.0 mL) than the NP group (mean=4764.3 mL). There was marginal significance (P=.053) between the 2 groups for age. There was no significant difference between the 2 groups for sodium levels on admission. Thus, those patients in the P group were more likely to have a higher admission BUN level, a higher fluid intake in the first 48 hours following admission, and to be younger than the NP patients.

Results of the logistic regression model showed that few factors were related to the change in radiographic findings, and no combination of factors seemed to be a good predictor for this change. The model poorly predicted which patients would show progressive radiographic infiltrates on the basis of age and admission BUN level. Although a significant difference was found between the 2 groups with respect to the first 48-hour fluid intake, this factor was dropped from the logistic regression model because it paralleled the admission BUN level [Table 2]. On the basis of a patient’s age and admission BUN level, the model could correctly predict only 28% of patients who would show progressive radiographic changes. In contrast, the same model correctly predicted 92% of patients who would show no change or improvement on subsequent radiographic studies. The c statistic (which is equivalent to the area under the curve for the receiver-operating characteristic) equals 0.68.

Discussion

The diagnosis of CAP is usually made by the presence of signs and symptoms of lower respiratory tract infection coupled with the finding of at least one acute opacity on the chest radiograph.^1,13 Some clinicians believe that such an opacity is necessary for the diagnosis of CAP, and they consider it the gold standard for the diagnosis in conjunction with supporting data from the history and physical examination.^2,4,5 However, radiographic and clinical findings are not always consistent. Heckerling¹⁴ found that no clinical findings other than auscultory abnormalities could be used to predict the presence of pneumonia on chest radiography. A review of the literature by Metlay and colleagues¹⁵ found that there is no reliable combination of findings that can rule in the diagnosis of pneumonia. Up to 25% of young adults may not have auscultory findings of pneumonia.⁴ It is well documented that elderly patients may initially manifest no fever, minimal respiratory symptoms, and minimal radiographic infiltrates despite laboratory documentation of pneumonia.^16,17 Patients with neutropenia or infections with atypical organisms or viruses may not manifest an infiltrate on chest radiograph until several days after clinical symptoms appear.^5,9,18 Dehydration or volume depletion has also been shown to produce this phenomenon.^4,5,10

Pneumonia has been defined as inflammation and consolidation of lung tissue due to an infectious agent. When sufficient fluid has collected in the normally air-filled spaces, X-ray beam attenuation occurs and an opacity appears on a chest radiograph. It has been postulated that decreased pulmonary hydrostatic pressure and increased pulmonary oncotic pressure secondary to dehydration or volume depletion may diminish the flux of fluid into the alveoli and interstitium and thus delay or alter the radiographic findings of pneumonia.¹⁹ Cooligan and coworkers²⁰ found that small increases in pulmonary capillary wedge pressure increased wet lung weight in euvolemic dogs with pneumococcal pneumonia, but no radiographic evaluation was performed. Caldwell and colleagues¹⁹ found that acute intravascular volume depletion did not affect the radiographic evolution of pneumococcal pneumonia in dogs, but they did not evaluate the effects of fluid volume repletion. Caldwell and coworkers also noted that on retrospective chart review of 20 consecutive human patients with pneumonia and the diagnosis of dehydration, all had an infiltrate apparent radiographically at the time of admission. The effects of rehydration were not addressed in these patients.

Our retrospective study appears to be the first to specifically address the relationship between pneumonia, hydration status, rehydration, and the effect on the radiographic appearance of pneumonia. Our results show an association between an elevated admission BUN level, volume of fluids given, and a worsening of the radiographic appearance of CAP. A future prospective study could teach us more about diagnostic strategies for CAP, including any situations in which repeated radiographs are helpful.

We were not able to address the question of whether the radiographic findings of pneumonia can be completely masked by dehydration and subsequently expressed by rehydration; our findings, however, suggest that there may be some basis for this premise.

Limitations

Our study has a number of limitations. There is an inherent selection bias in our retrospective process: Patients who received only one chest radiograph because of misdiagnosis, management decision, or other reasons were excluded from the study. Follow-up chest radiographs within a few days are generally not indicated and are not obtained in clinically stable patients who may only show worsening or improving patterns on radiograph.^6,13 Also, the retrospective nature of the study precluded the collection of a uniform database or uniform evaluation of possible etiologic agents. Viral and mycoplasma pneumonias are well documented to have a lag between onset of symptoms and evolution of radiographic findings,⁹ that may have led to inappropriate group assignment or complete exclusion from the study. The determination of the presence, progression, or nonprogression of radiographic findings is somewhat subjective and may affect our results. Our data are based on the reported radiograph interpretation from a randomly assigned radiologist. Interobserver agreement between radiologists for the determination of infiltrates on chest radiographs was found to be only 80% in a recent study.²¹ Finally, hydration status was difficult to determine. Although the BUN level was significantly higher in the P group than in the NP group, BUN level is not a specific marker for dehydration or intravascular volume status. The higher serum sodium levels of the P group, while not achieving statistical significance, may have been a better marker of total body water status. Although fluid volume depletion and dehydration are very different clinical problems,²² the determination of which condition was present, and the severity of that condition, was not possible in our study. Although statistically significant, the difference in fluid intake between the 2 groups may not be clinically significant. Furthermore, dehydration alone has been shown to adversely affect lung host defense in rats.²³ If applicable to humans, this could also unexpectedly affect our results.

Conclusions

To our knowledge, this study is the first to try to define the relationship of fluid volume status and radiographic progression of CAP in humans. The retrospective nature of our study leads to several limitations and potential sources of bias, but it appears to show that there is a correlation between markers of fluid volume status, hydration, and the evolution of the radiographic findings. A prospective clinical study is needed to further define this relationship.

Acknowledgments

Our work was funded by a grant from the Clinical Research Center of the Medical Center of Central Georgia. We wish to thank Jennifer K. Rayhill for her assistance in preparing this manuscript.

Standard obstetrical practice has included screening for anemia and the provision of iron supplements to anemic patients. This approach has been based on assumptions about anemia and iron deficiency that are not supported by the literature.

Anemia in pregnancy has been reported to be associated with preterm delivery.^1,2 However, this may not take into consideration the lower hemoglobin values normally present during the second trimester. The potentially spurious association between anemia and preterm delivery could be explained by the lower hemoglobin values that are expected during the second trimester. When measured at the time of preterm delivery, these lower hemoglobin values, which are normal in the second trimester (but not in the third trimester or in a nonpregnant woman), are often believed to have led to the preterm delivery.

Normal hemogloblin values from nonpregnant women cannot be assumed to apply to those who are pregnant. Average hemoglobin levels decrease to 11.6 g per dL at 20 to 24 weeks’ gestation, with the fifth percentile at 10.5 g per dL (hematocrit=32%). Anemia in pregnancy has been defined by criteria from the Centers for Disease Control and Prevention (CDC) as a hemoglobin level of less than 11 g per dL during the first and third trimesters and less than 10.5 g per dL during the second trimester.³

When anemia is present in pregnancy it cannot be assumed to be the result of iron deficiency, even though this type of anemia has been previously reported as the most common cause.⁴ The Camden study⁵ of 826 pregnant women showed preterm delivery and low birth weight associated with iron-deficiency anemia. Only 27.9% of the pregnant women had anemia, however, and only 12.5% of the patients with anemia had an iron deficiency. Thus, only 3.5% of the entire cohort had iron-deficiency anemia.⁶

Iron deficiency in pregnancy has been defined by the National Academy of Sciences panel on nutrition and pregnancy⁷ as ferritin levels lower than 12 ng per mL. A systematic overview⁸ of 55 studies relevant to laboratory tests for diagnosis of iron-deficiency anemia in variable patient populations found serum ferritin radioimmunoassay to be the most powerful test. Ferritin levels are considered the gold standard for the diagnosis of iron-deficiency anemia in pregnancy.⁹

We report a descriptive study of the use of ferritin levels to determine the need for iron supplementation among pregnant women with anemia.

Methods

Lebanon Family Health Services is a nonprofit federally subsidized community agency providing prenatal care and women’s health services to a diverse population without restriction on the basis of financial status. The prenatal patients cared for in this practice included women aged 15 to 40 years (23.6% were younger than 19 years; 70.8%, 19 to 30 years; and 5.6%, older than 30 years) of whom 65.3% were white, 29.2% Hispanic, 4.2% African American, and 1.4% Asian.

We evaluated all patients entering into prenatal care at Lebanon Family Health Services from April 1, 1997, through December 31, 1998, using prospective data collection and retrospective record review. Prenatal vitamins (including elemental iron 30 to 60 mg/day) were prescribed to all patients. Complete blood count was tested at the initial evaluation as part of a comprehensive screening. For patients who entered prenatal care at earlier than 28 weeks’ gestation complete blood count was checked again when they had reached that point.

The CDC criteria for defining anemia are hemoglobin levels less than 11 g per dL during the first and third trimesters, and less than 10.5 g per dL during the second trimester. A hemoglobin level of less than 11 g per dL at any time during the pregnancy was used as the cutoff point for anemia in this clinical practice, in consideration of patients with uncertain or inaccurate pregnancy dating.

Patients with anemia underwent blood testing for serum ferritin level, generally 1 week after a complete blood cell count. At the time, ferritin levels were determined using different test tubes than those used for other prenatal testing, thus resulting in a delay in obtaining blood for ferritin testing. If serum ferritin was 12 ng per dL or lower, iron-deficiency anemia was diagnosed and ferrous sulfate was prescribed for the remainder of the pregnancy and the postpartum period. If serum ferritin was greater than 12 ng per dL iron deficiency was excluded, anemia was generally considered to be pregnancy related, and further evaluation and treatment was at the discretion of the treating clinician.

Variables recorded for all patients with anemia were: estimated date of delivery, last menstrual period, date of any testing for complete blood count or ferritin levels, hemoglobin value, hematocrit value, red blood cell count (RBC), mean corpuscular volume (MCV), mean corpuscular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), red cell distribution width (RDW), any notation regarding microcytosis, and serum ferritin levels.

Results

A total of 182 patients were consecutively entered into prenatal care during the study period. Hemoglobin data were not obtained from 9 patients—4 transferred to another practice for care before initial blood testing, 2 did not carry the pregnancy until initial blood testing, and 3 were lost to follow-up before initial blood testing. Thus, hemoglobin data were available for 173 patients (95%).

Thirty-eight women (22%) had anemia defined as a hemoglobin level lower than 11 mg per dL at any time during the pregnancy. One patient was excluded from further evaluation because she entered prenatal care late, had an initial hemoglobin level of 9.6 mg per dL at 39 weeks’ gestation, and was given iron without serum ferritin determination.

Of the 37 patients with anemia who had serum ferritin level measurements, the values ranged from 3 to 91 ng per dL. Twenty of these 37 patients (54%) had iron-deficiency anemia (ferritin levels ranging from 3-10 mg/dL), and 17 patients (46%) had anemia not related to iron deficiency (range=13-91 mg/dL).

Of the 38 patients with anemia by our definition, 20 had hemoglobin levels between 10.5 and 10.9 mg per dL, of whom 10 were between 14 and 28 weeks’ gestation (second trimester). Thus only 28 patients (16%) had anemia according to the CDC criteria. Of the 27 patients with anemia according to the CDC criteria (one patient was excluded), 17 (63%) had iron-deficiency anemia, and 10 (37%) had anemia not related to iron deficiency.

Discussion

Screening for anemia in pregnancy has been based on an association with an increased risk for preterm delivery.^1,2 This association is likely due to the comparison of the natural nadir in hemoglobin values in the second trimester, measured at the time of preterm delivery, with the higher hemoglobin values, measured at the time of term deliveries; this concept, however, has not been fully evaluated.¹ Despite the lack of evidence that screening for anemia improves clinical maternal, fetal, or neonatal outcomes¹⁰ standard obstetrical practice has been to screen all pregnant women for anemia and empirically treat anemic patients with iron therapy.

Although there is no convincing data demonstrating clear harm from iron therapy during pregnancy, limiting such therapy to patients with strictly defined anemia and demonstrated iron deficiency may be prudent to minimize potential harms from a practice not shown to clearly provide benefits.

In actual practice, with the common misdating of pregnancies and the potential inefficiency of complex management rules, it is possible that cutoff hemoglobin levels of 11 mg per dL for defining anemia in pregnancy without respect to gestational age are being used instead of the CDC criteria. By doing so, more patients without iron-deficiency anemia will be labeled as anemic.

The use of hematologic indices provided with complete blood count determination would be more efficient and less costly than serum ferritin determination, if they are shown to differentiate iron deficiency from other causes of anemia in pregnant women. Scatterplots have been used to estimate the discriminatory ability of continuous variables for discerning between patient populations.¹¹ We made scatterplots for RBC, MCV, MCH, MCHC, and RDW. None of these scatterplots suggested discriminatory ability for any of these variables. The results were unchanged when limited to patients with CDC-defined anemia.

Costs

Ferritin determination may be cost-effective depending on its cost, the cost of iron therapy, the prevalence of iron-deficiency anemia (which is dependent on the criteria for defining anemia), and the nonfinancial burden of unnecessary iron therapy. The costs of sparing women unnecessary iron therapy on the basis of these variables are detailed in the [Table]. Serum ferritin level determination cost $30 at the laboratory we used for our study. Other methods of determining iron deficiency (such as iron and total iron-binding capacity levels, which cost $35) were not evaluated, since the evidence did not suggest that these values were accurate or well known in pregnant women. Using numbers derived from our study, checking the ferritin levels of 100 pregnant women with anemia would cost $3000 and spare 37 to 46 women from taking iron, at a cost of $37.24 to $73.56 to prevent 1 woman from taking an unnecessary course of iron therapy.

Caretakers prescribing iron therapy are familiar with its adverse effects and relatively low tolerability. In a dose-finding study¹² of 110 pregnant women randomized to 1 of 3 doses of ferrous sulfate daily, 32.4% of those taking 60 mg of elemental iron (equivalent to 325 mg of ferrous sulfate) and 40.5% of those taking 120 mg of elemental iron (equivalent to common twice a day dosing) had side effects. Dropout rates matched the side effect rates (32.4% and 38.8%, respectively). Thus, for every 5 women treated with iron, 2 will develop side effects and stop taking it.

The cost-effectiveness of ferritin determination is highly dependent on and inversely related to the prevalence of iron-deficiency anemia in the patient population. As seen in the Table, if the prevalence of iron-deficiency anemia is sufficiently low, ferritin determination may be very cost-effective.

Clinicians should consider the local costs of ferrous sulfate, ferritin determination, and the prevalence of iron-deficiency anemia in their patient population in the evaluation of the use of ferritin determination instead of empiric iron therapy. Alternately, clinicians may present some of the issues and uncertainties to their patients for combined decision making.

Conclusions

In our population of prenatal patients with anemia, only 54% had an iron deficiency. Diagnostic and therapeutic approaches to screening for anemia in pregnancy should be reconsidered and further evaluated to avoid unnecessary iron therapy.

Acknowledgments

We would like to acknowledge Siobhan Ulrich, Jan Balmer, Sue Gibson, and Tracy Chappel for their efforts in implementing the practice protocol and keeping track of patient records.

Standard obstetrical practice has included screening for anemia and the provision of iron supplements to anemic patients. This approach has been based on assumptions about anemia and iron deficiency that are not supported by the literature.

Anemia in pregnancy has been reported to be associated with preterm delivery.^1,2 However, this may not take into consideration the lower hemoglobin values normally present during the second trimester. The potentially spurious association between anemia and preterm delivery could be explained by the lower hemoglobin values that are expected during the second trimester. When measured at the time of preterm delivery, these lower hemoglobin values, which are normal in the second trimester (but not in the third trimester or in a nonpregnant woman), are often believed to have led to the preterm delivery.

Normal hemogloblin values from nonpregnant women cannot be assumed to apply to those who are pregnant. Average hemoglobin levels decrease to 11.6 g per dL at 20 to 24 weeks’ gestation, with the fifth percentile at 10.5 g per dL (hematocrit=32%). Anemia in pregnancy has been defined by criteria from the Centers for Disease Control and Prevention (CDC) as a hemoglobin level of less than 11 g per dL during the first and third trimesters and less than 10.5 g per dL during the second trimester.³

When anemia is present in pregnancy it cannot be assumed to be the result of iron deficiency, even though this type of anemia has been previously reported as the most common cause.⁴ The Camden study⁵ of 826 pregnant women showed preterm delivery and low birth weight associated with iron-deficiency anemia. Only 27.9% of the pregnant women had anemia, however, and only 12.5% of the patients with anemia had an iron deficiency. Thus, only 3.5% of the entire cohort had iron-deficiency anemia.⁶

Iron deficiency in pregnancy has been defined by the National Academy of Sciences panel on nutrition and pregnancy⁷ as ferritin levels lower than 12 ng per mL. A systematic overview⁸ of 55 studies relevant to laboratory tests for diagnosis of iron-deficiency anemia in variable patient populations found serum ferritin radioimmunoassay to be the most powerful test. Ferritin levels are considered the gold standard for the diagnosis of iron-deficiency anemia in pregnancy.⁹

We report a descriptive study of the use of ferritin levels to determine the need for iron supplementation among pregnant women with anemia.

Methods

Lebanon Family Health Services is a nonprofit federally subsidized community agency providing prenatal care and women’s health services to a diverse population without restriction on the basis of financial status. The prenatal patients cared for in this practice included women aged 15 to 40 years (23.6% were younger than 19 years; 70.8%, 19 to 30 years; and 5.6%, older than 30 years) of whom 65.3% were white, 29.2% Hispanic, 4.2% African American, and 1.4% Asian.

We evaluated all patients entering into prenatal care at Lebanon Family Health Services from April 1, 1997, through December 31, 1998, using prospective data collection and retrospective record review. Prenatal vitamins (including elemental iron 30 to 60 mg/day) were prescribed to all patients. Complete blood count was tested at the initial evaluation as part of a comprehensive screening. For patients who entered prenatal care at earlier than 28 weeks’ gestation complete blood count was checked again when they had reached that point.

The CDC criteria for defining anemia are hemoglobin levels less than 11 g per dL during the first and third trimesters, and less than 10.5 g per dL during the second trimester. A hemoglobin level of less than 11 g per dL at any time during the pregnancy was used as the cutoff point for anemia in this clinical practice, in consideration of patients with uncertain or inaccurate pregnancy dating.

Patients with anemia underwent blood testing for serum ferritin level, generally 1 week after a complete blood cell count. At the time, ferritin levels were determined using different test tubes than those used for other prenatal testing, thus resulting in a delay in obtaining blood for ferritin testing. If serum ferritin was 12 ng per dL or lower, iron-deficiency anemia was diagnosed and ferrous sulfate was prescribed for the remainder of the pregnancy and the postpartum period. If serum ferritin was greater than 12 ng per dL iron deficiency was excluded, anemia was generally considered to be pregnancy related, and further evaluation and treatment was at the discretion of the treating clinician.

Variables recorded for all patients with anemia were: estimated date of delivery, last menstrual period, date of any testing for complete blood count or ferritin levels, hemoglobin value, hematocrit value, red blood cell count (RBC), mean corpuscular volume (MCV), mean corpuscular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), red cell distribution width (RDW), any notation regarding microcytosis, and serum ferritin levels.

Results

A total of 182 patients were consecutively entered into prenatal care during the study period. Hemoglobin data were not obtained from 9 patients—4 transferred to another practice for care before initial blood testing, 2 did not carry the pregnancy until initial blood testing, and 3 were lost to follow-up before initial blood testing. Thus, hemoglobin data were available for 173 patients (95%).

Thirty-eight women (22%) had anemia defined as a hemoglobin level lower than 11 mg per dL at any time during the pregnancy. One patient was excluded from further evaluation because she entered prenatal care late, had an initial hemoglobin level of 9.6 mg per dL at 39 weeks’ gestation, and was given iron without serum ferritin determination.

Of the 37 patients with anemia who had serum ferritin level measurements, the values ranged from 3 to 91 ng per dL. Twenty of these 37 patients (54%) had iron-deficiency anemia (ferritin levels ranging from 3-10 mg/dL), and 17 patients (46%) had anemia not related to iron deficiency (range=13-91 mg/dL).

Of the 38 patients with anemia by our definition, 20 had hemoglobin levels between 10.5 and 10.9 mg per dL, of whom 10 were between 14 and 28 weeks’ gestation (second trimester). Thus only 28 patients (16%) had anemia according to the CDC criteria. Of the 27 patients with anemia according to the CDC criteria (one patient was excluded), 17 (63%) had iron-deficiency anemia, and 10 (37%) had anemia not related to iron deficiency.

Discussion

Screening for anemia in pregnancy has been based on an association with an increased risk for preterm delivery.^1,2 This association is likely due to the comparison of the natural nadir in hemoglobin values in the second trimester, measured at the time of preterm delivery, with the higher hemoglobin values, measured at the time of term deliveries; this concept, however, has not been fully evaluated.¹ Despite the lack of evidence that screening for anemia improves clinical maternal, fetal, or neonatal outcomes¹⁰ standard obstetrical practice has been to screen all pregnant women for anemia and empirically treat anemic patients with iron therapy.

Although there is no convincing data demonstrating clear harm from iron therapy during pregnancy, limiting such therapy to patients with strictly defined anemia and demonstrated iron deficiency may be prudent to minimize potential harms from a practice not shown to clearly provide benefits.

In actual practice, with the common misdating of pregnancies and the potential inefficiency of complex management rules, it is possible that cutoff hemoglobin levels of 11 mg per dL for defining anemia in pregnancy without respect to gestational age are being used instead of the CDC criteria. By doing so, more patients without iron-deficiency anemia will be labeled as anemic.

The use of hematologic indices provided with complete blood count determination would be more efficient and less costly than serum ferritin determination, if they are shown to differentiate iron deficiency from other causes of anemia in pregnant women. Scatterplots have been used to estimate the discriminatory ability of continuous variables for discerning between patient populations.¹¹ We made scatterplots for RBC, MCV, MCH, MCHC, and RDW. None of these scatterplots suggested discriminatory ability for any of these variables. The results were unchanged when limited to patients with CDC-defined anemia.

Costs

Ferritin determination may be cost-effective depending on its cost, the cost of iron therapy, the prevalence of iron-deficiency anemia (which is dependent on the criteria for defining anemia), and the nonfinancial burden of unnecessary iron therapy. The costs of sparing women unnecessary iron therapy on the basis of these variables are detailed in the [Table]. Serum ferritin level determination cost $30 at the laboratory we used for our study. Other methods of determining iron deficiency (such as iron and total iron-binding capacity levels, which cost $35) were not evaluated, since the evidence did not suggest that these values were accurate or well known in pregnant women. Using numbers derived from our study, checking the ferritin levels of 100 pregnant women with anemia would cost $3000 and spare 37 to 46 women from taking iron, at a cost of $37.24 to $73.56 to prevent 1 woman from taking an unnecessary course of iron therapy.

Caretakers prescribing iron therapy are familiar with its adverse effects and relatively low tolerability. In a dose-finding study¹² of 110 pregnant women randomized to 1 of 3 doses of ferrous sulfate daily, 32.4% of those taking 60 mg of elemental iron (equivalent to 325 mg of ferrous sulfate) and 40.5% of those taking 120 mg of elemental iron (equivalent to common twice a day dosing) had side effects. Dropout rates matched the side effect rates (32.4% and 38.8%, respectively). Thus, for every 5 women treated with iron, 2 will develop side effects and stop taking it.

The cost-effectiveness of ferritin determination is highly dependent on and inversely related to the prevalence of iron-deficiency anemia in the patient population. As seen in the Table, if the prevalence of iron-deficiency anemia is sufficiently low, ferritin determination may be very cost-effective.

Clinicians should consider the local costs of ferrous sulfate, ferritin determination, and the prevalence of iron-deficiency anemia in their patient population in the evaluation of the use of ferritin determination instead of empiric iron therapy. Alternately, clinicians may present some of the issues and uncertainties to their patients for combined decision making.

Conclusions

In our population of prenatal patients with anemia, only 54% had an iron deficiency. Diagnostic and therapeutic approaches to screening for anemia in pregnancy should be reconsidered and further evaluated to avoid unnecessary iron therapy.

Acknowledgments

We would like to acknowledge Siobhan Ulrich, Jan Balmer, Sue Gibson, and Tracy Chappel for their efforts in implementing the practice protocol and keeping track of patient records.

Standard obstetrical practice has included screening for anemia and the provision of iron supplements to anemic patients. This approach has been based on assumptions about anemia and iron deficiency that are not supported by the literature.

Anemia in pregnancy has been reported to be associated with preterm delivery.^1,2 However, this may not take into consideration the lower hemoglobin values normally present during the second trimester. The potentially spurious association between anemia and preterm delivery could be explained by the lower hemoglobin values that are expected during the second trimester. When measured at the time of preterm delivery, these lower hemoglobin values, which are normal in the second trimester (but not in the third trimester or in a nonpregnant woman), are often believed to have led to the preterm delivery.

Normal hemogloblin values from nonpregnant women cannot be assumed to apply to those who are pregnant. Average hemoglobin levels decrease to 11.6 g per dL at 20 to 24 weeks’ gestation, with the fifth percentile at 10.5 g per dL (hematocrit=32%). Anemia in pregnancy has been defined by criteria from the Centers for Disease Control and Prevention (CDC) as a hemoglobin level of less than 11 g per dL during the first and third trimesters and less than 10.5 g per dL during the second trimester.³

When anemia is present in pregnancy it cannot be assumed to be the result of iron deficiency, even though this type of anemia has been previously reported as the most common cause.⁴ The Camden study⁵ of 826 pregnant women showed preterm delivery and low birth weight associated with iron-deficiency anemia. Only 27.9% of the pregnant women had anemia, however, and only 12.5% of the patients with anemia had an iron deficiency. Thus, only 3.5% of the entire cohort had iron-deficiency anemia.⁶

Iron deficiency in pregnancy has been defined by the National Academy of Sciences panel on nutrition and pregnancy⁷ as ferritin levels lower than 12 ng per mL. A systematic overview⁸ of 55 studies relevant to laboratory tests for diagnosis of iron-deficiency anemia in variable patient populations found serum ferritin radioimmunoassay to be the most powerful test. Ferritin levels are considered the gold standard for the diagnosis of iron-deficiency anemia in pregnancy.⁹

We report a descriptive study of the use of ferritin levels to determine the need for iron supplementation among pregnant women with anemia.

Methods

Lebanon Family Health Services is a nonprofit federally subsidized community agency providing prenatal care and women’s health services to a diverse population without restriction on the basis of financial status. The prenatal patients cared for in this practice included women aged 15 to 40 years (23.6% were younger than 19 years; 70.8%, 19 to 30 years; and 5.6%, older than 30 years) of whom 65.3% were white, 29.2% Hispanic, 4.2% African American, and 1.4% Asian.

We evaluated all patients entering into prenatal care at Lebanon Family Health Services from April 1, 1997, through December 31, 1998, using prospective data collection and retrospective record review. Prenatal vitamins (including elemental iron 30 to 60 mg/day) were prescribed to all patients. Complete blood count was tested at the initial evaluation as part of a comprehensive screening. For patients who entered prenatal care at earlier than 28 weeks’ gestation complete blood count was checked again when they had reached that point.

The CDC criteria for defining anemia are hemoglobin levels less than 11 g per dL during the first and third trimesters, and less than 10.5 g per dL during the second trimester. A hemoglobin level of less than 11 g per dL at any time during the pregnancy was used as the cutoff point for anemia in this clinical practice, in consideration of patients with uncertain or inaccurate pregnancy dating.

Patients with anemia underwent blood testing for serum ferritin level, generally 1 week after a complete blood cell count. At the time, ferritin levels were determined using different test tubes than those used for other prenatal testing, thus resulting in a delay in obtaining blood for ferritin testing. If serum ferritin was 12 ng per dL or lower, iron-deficiency anemia was diagnosed and ferrous sulfate was prescribed for the remainder of the pregnancy and the postpartum period. If serum ferritin was greater than 12 ng per dL iron deficiency was excluded, anemia was generally considered to be pregnancy related, and further evaluation and treatment was at the discretion of the treating clinician.

Variables recorded for all patients with anemia were: estimated date of delivery, last menstrual period, date of any testing for complete blood count or ferritin levels, hemoglobin value, hematocrit value, red blood cell count (RBC), mean corpuscular volume (MCV), mean corpuscular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), red cell distribution width (RDW), any notation regarding microcytosis, and serum ferritin levels.

Results

A total of 182 patients were consecutively entered into prenatal care during the study period. Hemoglobin data were not obtained from 9 patients—4 transferred to another practice for care before initial blood testing, 2 did not carry the pregnancy until initial blood testing, and 3 were lost to follow-up before initial blood testing. Thus, hemoglobin data were available for 173 patients (95%).

Thirty-eight women (22%) had anemia defined as a hemoglobin level lower than 11 mg per dL at any time during the pregnancy. One patient was excluded from further evaluation because she entered prenatal care late, had an initial hemoglobin level of 9.6 mg per dL at 39 weeks’ gestation, and was given iron without serum ferritin determination.

Of the 37 patients with anemia who had serum ferritin level measurements, the values ranged from 3 to 91 ng per dL. Twenty of these 37 patients (54%) had iron-deficiency anemia (ferritin levels ranging from 3-10 mg/dL), and 17 patients (46%) had anemia not related to iron deficiency (range=13-91 mg/dL).

Of the 38 patients with anemia by our definition, 20 had hemoglobin levels between 10.5 and 10.9 mg per dL, of whom 10 were between 14 and 28 weeks’ gestation (second trimester). Thus only 28 patients (16%) had anemia according to the CDC criteria. Of the 27 patients with anemia according to the CDC criteria (one patient was excluded), 17 (63%) had iron-deficiency anemia, and 10 (37%) had anemia not related to iron deficiency.

Discussion

Screening for anemia in pregnancy has been based on an association with an increased risk for preterm delivery.^1,2 This association is likely due to the comparison of the natural nadir in hemoglobin values in the second trimester, measured at the time of preterm delivery, with the higher hemoglobin values, measured at the time of term deliveries; this concept, however, has not been fully evaluated.¹ Despite the lack of evidence that screening for anemia improves clinical maternal, fetal, or neonatal outcomes¹⁰ standard obstetrical practice has been to screen all pregnant women for anemia and empirically treat anemic patients with iron therapy.

Although there is no convincing data demonstrating clear harm from iron therapy during pregnancy, limiting such therapy to patients with strictly defined anemia and demonstrated iron deficiency may be prudent to minimize potential harms from a practice not shown to clearly provide benefits.

In actual practice, with the common misdating of pregnancies and the potential inefficiency of complex management rules, it is possible that cutoff hemoglobin levels of 11 mg per dL for defining anemia in pregnancy without respect to gestational age are being used instead of the CDC criteria. By doing so, more patients without iron-deficiency anemia will be labeled as anemic.

The use of hematologic indices provided with complete blood count determination would be more efficient and less costly than serum ferritin determination, if they are shown to differentiate iron deficiency from other causes of anemia in pregnant women. Scatterplots have been used to estimate the discriminatory ability of continuous variables for discerning between patient populations.¹¹ We made scatterplots for RBC, MCV, MCH, MCHC, and RDW. None of these scatterplots suggested discriminatory ability for any of these variables. The results were unchanged when limited to patients with CDC-defined anemia.

Costs

Ferritin determination may be cost-effective depending on its cost, the cost of iron therapy, the prevalence of iron-deficiency anemia (which is dependent on the criteria for defining anemia), and the nonfinancial burden of unnecessary iron therapy. The costs of sparing women unnecessary iron therapy on the basis of these variables are detailed in the [Table]. Serum ferritin level determination cost $30 at the laboratory we used for our study. Other methods of determining iron deficiency (such as iron and total iron-binding capacity levels, which cost $35) were not evaluated, since the evidence did not suggest that these values were accurate or well known in pregnant women. Using numbers derived from our study, checking the ferritin levels of 100 pregnant women with anemia would cost $3000 and spare 37 to 46 women from taking iron, at a cost of $37.24 to $73.56 to prevent 1 woman from taking an unnecessary course of iron therapy.

Caretakers prescribing iron therapy are familiar with its adverse effects and relatively low tolerability. In a dose-finding study¹² of 110 pregnant women randomized to 1 of 3 doses of ferrous sulfate daily, 32.4% of those taking 60 mg of elemental iron (equivalent to 325 mg of ferrous sulfate) and 40.5% of those taking 120 mg of elemental iron (equivalent to common twice a day dosing) had side effects. Dropout rates matched the side effect rates (32.4% and 38.8%, respectively). Thus, for every 5 women treated with iron, 2 will develop side effects and stop taking it.

The cost-effectiveness of ferritin determination is highly dependent on and inversely related to the prevalence of iron-deficiency anemia in the patient population. As seen in the Table, if the prevalence of iron-deficiency anemia is sufficiently low, ferritin determination may be very cost-effective.

Clinicians should consider the local costs of ferrous sulfate, ferritin determination, and the prevalence of iron-deficiency anemia in their patient population in the evaluation of the use of ferritin determination instead of empiric iron therapy. Alternately, clinicians may present some of the issues and uncertainties to their patients for combined decision making.

Conclusions

In our population of prenatal patients with anemia, only 54% had an iron deficiency. Diagnostic and therapeutic approaches to screening for anemia in pregnancy should be reconsidered and further evaluated to avoid unnecessary iron therapy.

Acknowledgments

We would like to acknowledge Siobhan Ulrich, Jan Balmer, Sue Gibson, and Tracy Chappel for their efforts in implementing the practice protocol and keeping track of patient records.

The detrimental effects of glucocorticoids on bone have long been recognized in the medical literature. An average of 5% of bone mass is lost during the first year of long-term therapy, and annualized rates of loss range from 0.3% to 3%.^1,2 This bone reduction occurs most rapidly during the first 6 to 12 months of glucocorticoid therapy and is dose-dependent and time-dependant.^3-5 Daily doses of oral prednisone greater than or equal to 7.5 mg or cumulative doses greater than 10 g produce the most significant effects, and alternate-day glucocorticoids do not decrease the risk.⁶ Although inhaled corticosteroids are generally considered safer, high inhaled doses can also reduce bone mass.¹ An inverse relation between inhaled corticosteroid dose and bone mineral density (BMD) has been reported; doubling of the inhaled dose led to a decrease in lumbar bone mineral density of 0.16 standard deviations points (SDs).⁷ Menopausal status, sex, and age are other important contributing factors for the development of glucocorticosteroid-induced osteoporosis (CIO); surprisingly, men younger than 50 years may lose a higher percentage of bone than postmenopausal women.^2,87,98

Of greater clinical significance is the increased incidence of fractures, which is 2 to 4 times higher than that of similar glucocorticoid-naive patients.^1,11 Overall estimates of fractures during long-term steroid therapy range from 30% to 50%.^1,4,5 Ultimately, all fracture types can result in skeletal deformities that cause extreme pain, exacerbate the primary autoimmune or inflammatory disease state, and represent a tremendous financial burden.⁶ Most alarming is the estimated incidence of mortality following a hip fracture in patients requiring long-term corticosteroids, which ranges from 5% to 9% in men older than 50 years and from 1% to 3% in age-matched women.¹

Pathophysiology Of CIO

Glucocorticoids decrease bone formation and increase bone resorption through a number of different mechanisms that are beyond the scope of our article. However, a brief overview is warranted to better understand the role bisphosphonates have in the treatment of steroid-induced osteoporosis. On a cellular level, glucocorticoids directly inhibit osteoblast function at the glucocorticoid receptor. This inhibition results in decreased replication, differentiation, proliferation, and life span of the osteoblasts.^2,4,11 Subsequently the total amount of bone restored during each remodeling cycle is decreased by 30%, leading to a reduced mean wall thickness.^1,4,5 Also, glucocorticoids enhance the activity and increase the number of osteoclasts, leading to a greater number of active resorption surfaces.^2,11

Additionally, steroid-treated patients demonstrate a dose-dependent malabsorption of calcium due to direct impairment of the intestinal cell calcium transport process.^3,8 This decreased calcium absorption is often evident within the first 2 weeks of therapy.^4,8 Then a secondary hyperparathyroidism ensues and urinary calcium excretion becomes double that of non-steroid-treated patients.^3,4 Gonadal hormones—potent regulators of bone metabolism—are secreted to a much lower extent in patients treated with glucocorticoids.³ Treated men in particular have circulating testosterone concentrations of only 50% of those in a control group.⁴

Corticosteroids also reduce levels of prostaglandin E2, insulin-like growth factors, phosphate, type I collagen, and noncollagens including osteocalcin.^4,5 Finally, muscle atrophy and subsequent wasting may be the most observable result of glucocorticoid treatment, resulting in less of the mechanical stimuli required to generate new bone formation.^1,5,11

BMD Assessment and Preventive Measures

Although the risk of CIO and increased fracture risk is well cited, only 62% of surveyed physicians in one study¹² rated osteoporosis as one of the 3 most significant side effects discussed with patients taking high-dose steroids, and 2 cohort studies^13,14 found only a 5.6% and a 14% prescription rate for preventative medications when patients were prescribed glucocorticoids. Perhaps this mediocre reaction from primary care providers results from a shortage of clinical evidence to support preventive measures. Ideally, clinical decisions should be made using data concerning fracture risk reduction. Unfortunately, studies have not been sufficiently powered to assess fracture risk reduction, making it difficult to base therapeutic decisions on patient-oriented outcomes. The primary outcome assessed is the difference in percentage change from baseline in the BMD of the lumbar spine, femoral neck, and femoral trochanter between the treatment and placebo groups. It is important to recognize that the percentage change in BMD is a surrogate marker and has not been shown to directly correlate with decreased fracture risk. Other factors, such as bone strength and rate of turnover, may also contribute to fragility.

A BMD taken at one site, such as vertebrae or hip, correlates with risk of fracture at other sites; the best predictor of fracture, however, seems to be a BMD at the site in question.¹⁵ Patients at risk for CIO may have a comparatively higher risk of fracture at a given bone density than patients at risk for osteoporosis from other causes.¹⁶ A decrease in BMD of 1 SD below the mean of that of healthy adults aged 35 years will lead to a 1.5-fold to 3-fold increase in fracture risk,¹⁵ but this same bone density measure in the same woman taking corticosteroids may underestimate her risk of fracture.¹⁷

Presently the gold standard for BMD measurement is dual energy X-ray absorptiometry (DEXA). At this time, anteroposterior DEXA is recommended of both the lumbar spine and femoral neck for patients at risk for CIO.^17,18 If only one site can be obtained, the recommended site depends on the age of the patient. The lumbar spine is the site of choice for men and women younger than 60 years, while the femoral neck should be evaluated in men and women 60 years and older.¹⁸

Practice Guidelines

Two current sets of guidelines on the management of CIO are presently available: the recommendations of the 1998 United Kingdom (UK) Consensus Group¹⁷ and the 1996 American College of Rheumatology (ACR) Task Force on Osteoporosis Guidelines.¹⁸ Prevention and treatment options include risk factor modification (smoking cessation, fall prevention, and so forth) adequate calcium (1500 mg/day) and vitamin D3 intake (800 IU/day), hormone replacement when appropriate, bisphosphonates, and calcitonin (Figure). Other less-proven therapies such as thiazide diuretics, anabolic steroids, and fluoride are considered.

The ACR and UK guidelines differ considerably with respect to use of bisphosphonates in patients receiving long-term high-dose glucocorticoid therapy. The ACR task force recommends bisphosphonates only in patients with contraindications to hormone replacement therapy (HRT) or established osteoporosis.¹⁸ This class is not recommended in premenopausal women or men age younger than 50 years because of the lack of long-term safety data. The newer UK publication emphasizes that of all therapeutic options bisphosphonates have the most compelling BMD data and should be considered first-line treatment.¹⁷

Since the UK guidelines were published, 8 trials showing beneficial effects of bisphosphonates in the treatment of CIO have been published or presented at major medical meetings. Encouraging clinical data have resulted in alendronate and risedronate receiving United States Food and Drug Administration (FDA) approval for the indication of CIO. Our objective was to provide an evidence-based perspective of the use of bisphosphonates in corticosteroid-treated patients.

Methods

Literature searches were conducted independently by 2 authors using the MEDLINE, CURRENT CONTENTS, and HEALTHSTAR electronic databases to identify studies of bisphosphonates in the prevention or treatment of corticosteroid-induced osteoporosis. Medical subject headings and search terms included “bisphosphonates,” “diphosphonates,” “glucocorticoids,” “steroids,” and “osteoporosis.” We examined bibliographies of selected citations and review articles to obtain additional references. Unpublished information was obtained by hand-searching abstracts from recent meetings (1998) of the American Society for Bone and Mineral Research and the ACR. Studies were included if the design was randomized and controlled and if they evaluated the use of oral bisphosphonates in adults. Obtaining BMD results using the gold standard of measurement—central DEXA—was also necessary for study inclusion. Two authors assessed the methodologic quality of the trials using the 5-point criteria developed by Jadad and colleagues.¹⁹

Results

Bisphosphonates have been studied for use in CIO for more than a decade. Our search yielded a total of 34 citations of studies researching bisphosphonates for this indication. Twenty-one of these^20-40 were excluded for not meeting the specified inclusion criteria. The 13 studies included in our analysis^41-53 addressed the use of bisphosphonates in the prevention and treatment of corticosteroid-induced osteoporosis. Each bisphosphonate had the following number of citations and total number of patients in all studies combined: risedronate, 3 (n=638); etidronate, 8 (n=522); alendronate, 1 (n=477); and clodronate, 1 (n=74). The demographics pertaining to each study are summarized in [Table 1].

The methodologic quality of the trials was agreed on by the 2 reviewers. Scores ranged from 1 to 4 with 1 trial scoring 1,⁴¹ 2 trials scoring 2,^48,53 2 trials scoring 3,^45,47 and 8 trials scoring 4.^{42-44,46,49-52} All studies were reported as randomized; only 1 trial, however, described the randomization process.⁴⁴ Three of the 13 (23%) did not report using double-blind methodology.^41,48,53

The primary outcome assessed in all selected clinical trials is the difference in percentage change from baseline in the BMD of the lumbar spine, femoral neck, and femoral trochanter between the treatment and placebo groups. This complies with the ACR guidelines that recommend measurement at the lumbar spine and femoral neck. In regard to the studied primary outcomes, changes in BMD at the lumbar spine in treatment groups ranged from means of -0.137% to 4.9%; the control group values ranged from 3.7% to 0.98%. Three of 13 studies (23%) did not show a significant benefit at the lumbar spine when compared with the control group^41,47,48; however, 2 of these studies^41,48 used unusual doses of bisphosphonate. BMD changes at the femoral neck on treatment ranged from 1.28% to 3.6%; control group changes from baseline ranged from 3.6% to 3.64%. Nine of the 13 studies (69%) were not able to show statistical significance between the treatment and control groups at the femoral neck.^41-43,48-53 Changes in the treatment groups at the femoral trochanter ranged from -1.35% to 2.7%; placebo changes ranged from 3.06% to 1.5%. More than one half of studies reporting data at the femoral trochanter (6/10) showed that the treatment group was not significantly better than control group.^41,42,45,46 Specific BMD results from each trial are available in [Table 2]. Results varied based on the population, the bisphosphonate being investigated, the dosing regimen, and other distinctions in the study design.

Six studies^{42,43,45,46,49,52} addressed the possible differences in treatment effect based on sex and menopausal status. The effect of bisphosphonate treatment on BMD in postmenopausal women was significant in 4 of the trials.^42,45,46,52 One study included postmenopausal women taking HRT and found alendronate 10 mg led to a higher increase in BMD in women not receiving estrogen.⁴⁵ Only one study, however, found a significant effect on BMD in premenopausal women receiving etidronate.⁴² Also, an increase in the BMD of men was found to be significantly increased in one study.⁴⁶

The percentage change from baseline of the BMD of the distal and midshaft radius was measured in 1 study of etidronate and 2 studies of risedronate^42,46,47; statistical significance was not obtained in any of the results. Other secondary outcomes measured in selected citations were: biochemical markers of bone resorption and formation including pyridinium crosslinks, osteocalcin, parathyroid hormone, ionized calcium, serum alkaline phosphatase, serum alanine aminotransferase, and bone-specific alkaline phosphatase. However, these are not discussed in detail because of the uncertainty of their relationship to fracture risk.

Fracture Risk

Although not a specified primary end point, the risk of fracture development was addressed in 9 of the studies.^{42,43,45-49,51-52} Since the rate of fractures was low in most studies and no studies were powered on the basis of fracture risk, some data were reported without statistical analysis. Six studies^{42,43,45-47,51} analyzed the difference between treatment and control groups with respect to fracture risk. Three of these studies reported at least a trend in reduced fracture rates^42,45,46; however, only 1 (in abstract) found a statistical difference in the overall population.⁴⁷ A 10.1% reduction in vertebral fractures was found in patients receiving risedronate (either 2.5 mg or 5 mg) at 12 months (P=.021).⁴⁷ According to an abstract, when these results were pooled with another trial⁴⁶ it was found that risedronate 5.0 mg statistically decreased the incidence of vertebral fractures at 1 year (16.2% vs 5.4%, P=.01).⁵⁴ Therefore, it would be necessary to treat 10 CIO patients for 1 year with risedronate 5 mg per day to prevent one vertebral fracture.

Although the fracture rate was not significant in most trials, the subpopulation of postmenopausal women experienced the greatest number of fractures and seemed to benefit the most from treatment. Two studies found borderline significance in the fracture rate when postmenopausal women were analyzed separately.^42,45 Etidronate and alendronate use led to an absolute risk reduction of 18.7% (P=.05) and 8.6% (P=.05), respectively. Men were the only other group to experience fractures in any of the studies. No fractures were reported in the premenopausal population.

When interpreting the results of these studies it is important to remember that trial results can vary according to a number of factors, including: definition of fracture, initial number of fractures, baseline BMD, and population differences. Therefore, the reported data should not be extrapolated to all populations. It should also be mentioned that all of the fracture data presented consisted of radiologically reported fractures, which does not necessarily correlate with clinical fractures.

Safety and Tolerability

The withdrawal rates of the included studies ranged from 0% to 62% ([Table 3]). Of the dropouts reported, 42% were secondary to protocol violations or administrative reasons, and 16% were because of noncompliance. Twenty percent of study withdrawals were because of adverse events, of which 78% (29/37) were deemed not related to the treatment regimen. No significant difference was reported in dropout rates between the treatment and control groups.

In most of the studies no statistical significance concerning adverse effects was found when the treatment group was compared with the control group. Nine studies^42-47,50,51 discussed the subgroup of gastrointestinal (GI) adverse events, since this has been the greatest historical concern about the use of bisphosphonates. Eight studies found no difference in overall GI effects, and one study⁴⁵ reported a statistical trend for increased GI side effects with dose escalation of alendronate. This trend was probably secondary to an increase in abdominal pain in the treatment group when compared with the placebo group. In one study,⁴⁶ diarrhea was more common in patients receiving 5 mg risedronate than those taking placebo (number needed to harm=15), although significance was not reported.

Discussion

The studies examining the use of bisphosphonates for the prevention or treatment of CIO are difficult to interpret because of the various bisphosphonates and regimens, the heterogenicity in populations, the effect of the underlying disease on bone, and concomitant therapeutic interventions. A review of the currently published literature suggests bisphosphonates effectively prevent vertebral bone loss in patients treated with long-term corticosteroids. Patients receiving steroids for more than 3 months (secondary prevention) gained bone mass when placed on a bisphosphonate, while patients naive to steroids (primary prevention) maintained more bone density than the control group.^{42-44,46,47,50,51,53} This reinforces the fact that patients taking steroids for more than 3 months have already lost bone that can be partially regained with bisphosphonate treatment. In contrast, patients given bisphosphonates who were naive to steroids did not have significant changes in BMD from baseline; this group, however, was able to maintain bone density while the control group lost bone density.

Data regarding the impact of bisphosphonates on the risk of CIO-induced fractures are sparse and inconclusive. This is not surprising since most of the trials have been of relatively short duration (<2 years), and have not been sufficiently powered to show fracture reduction. Postmenopausal women not taking estrogen seem to benefit most from using bisphosphonates for the prevention of bone loss and of vertebral fractures in CIO.

A meta-analysis of a similar set of data based on a Cochrane systematic review of published literature was reported in 1999.⁵⁵ The authors state that of the small number of controlled clinical trials examining the use of bone-sparing agents in patients at risk for CIO bisphosphonates have shown some of the best evidence for reducing bone loss, particularly at the lumbar spine. They also concluded that bone density changes correlate with fracture risk in patients with CIO, but there are insufficient data to make conclusions regarding fracture risk reduction and use of bisphosphonates.

Overall, adverse effects of the bisphosphonates were minimal, and no statistical significance was found in studied populations when compared with control groups. However, the incidence of GI adverse events with alendronate may be as high as 15% in clinical practice, despite low incidence rates in phase III trials, possibly because of administration errors.⁵⁶ Directions for use are the same for all oral bisphosphonates, and include staying upright and not eating for at least 30 minutes after administration. Because oral bioavailability is usually less than 5% even on an empty stomach, bisphosphonates should be taken with a full glass of water in the morning after an overnight fast.^11,56,57

Implications for further research

Although current evidence supporting bisphosphonate use documents efficacy in BMD changes and trends toward reduced vertebral fracture risk in the treatment of CIO, more research is clearly desirable. The recent FDA approval of alendronate and risedronate has highlighted the need for aggressive measures to prevent and treat CIO. More research is needed in large studies to assess vertebral and nonvertebral fractures. Head-to-head comparative trials of bisphosphonates with other pharmacologic options, such as hormonal therapy or calcitonin are essential to establish evidence-based clinical guidelines. Studies addressing combination therapy with bisphosphonates and HRT in postmenopausal women would also be useful. Finally, comparative studies of the various bisphosphonates in relation to one another are needed. These synthetic pyrophosphate analogs possess a broad range of potencies, selectivity, and adverse effect profiles. Because each bisphosphonate has unique biological, chemical, and physiochemical properties, the results of one bisphosphonate study cannot be extrapolated to other compounds within the same drug class.^3,56,58,59 Differences in study design, patient populations, and other confounding variables also prevent the assumption of a class effect with certain research findings. This distinction becomes especially important with the new bisphosphonates on the horizon (clodronate, tiludronate, ibandronate, and zoledronate).

Although the relationship of BMD to fracture risk is well documented in postmenopausal osteoporosis, the correlation is not as well established in glucocorticoid-induced bone loss.^2,60 The present data suggest a trend in the reduction of fracture risk observed with the use of bisphosphonates in the treatment of glucocorticoid-induced osteoporosis. This information is promising but requires definitive confirmation in larger trials. Because of the prolonged skeletal retention of bisphosphonates, more extended trials are also needed to clearly establish the safety of long-term therapy in younger patients.

Recommendations for clinical practice

All patients beginning high-dose (Ž7.5 mg/day prednisone) long-term (>6 months) glucocorticoid therapy should be evaluated for pharmacologic prophylaxis against osteoporosis. Because the majority of bone loss occurs within the first 6 months of therapy, clinicians must vigilantly develop a preventative plan in advance, preferably before the glucocorticoid prescription is given to the patient. Current guidelines recommend a baseline BMD measurement of all patients with DEXA to determine the risk of osteoporosis and the monitor the efficacy of the chosen preventative measures throughout the course of therapy. Initial BMD measurements expressed as a T score are essential tools for constructing an individualized strategy. Critical T scores of less than or equal to 1.0 are indicative of high-risk patients.¹⁸ In addition to the risk imposed on patients secondary to their underlying disease state, other patients with strong predictive factors include those who are elderly and at greater risk of falling and those who have a previous history of fractures.

BMD measurements may be repeated in 6 to 12 months, depending on initial bone mass. If BMD has decreased by more than 5% from baseline, the initial choice of therapy should be changed or expanded.^18,61

Both glucocorticoid doses and the length of therapy require continuous reevaluation to decrease the total cumulative dose. If possible, dose maximization of inhaled and topical corticosteroids is recommended before progression to the oral form. More obvious preventive lifestyle modifications include smoking cessation, maintenance of healthy body weight, regular weight-bearing exercise, decreased alcohol consumption, sodium restriction, and increased dietary calcium intake. Most patients will require additional calcium supplementation to meet the ACR recommendation of 1500 mg per day, and patients at risk for vitamin D deficiency may also require supplementation.¹⁸

However, even with appropriate supplementation 1 in 6 corticosteroid-treated patients will experience a radiographically detected vertebral fracture within 12 months.^46,47,54 Therefore, alternative treatment regimens are needed. Bisphosphonate therapy looks promising in the prevention and treatment of CIO with respect to positively affecting BMD. As a result of this encouraging data, alendronate and risedronate have received FDA approval for the indication of CIO. However, it remains to be seen whether the use of these agents can lead to a reduction in fractures, both vertebral and nonvertebral. Subpopulations at highest risk (postmenopausal women and men) may benefit most with regard to fracture risk reduction, but more data are needed before bisphosphonates should be routinely recommended as first-line preventive therapy. Once osteoporosis is established (regardless of the pathogenesis), treatment should be aggressive to prevent further loss of bone density. In this population bisphosphonates are an appropriate therapeutic option, and further data will clarify their impact on the risk of fractures.

The detrimental effects of glucocorticoids on bone have long been recognized in the medical literature. An average of 5% of bone mass is lost during the first year of long-term therapy, and annualized rates of loss range from 0.3% to 3%.^1,2 This bone reduction occurs most rapidly during the first 6 to 12 months of glucocorticoid therapy and is dose-dependent and time-dependant.^3-5 Daily doses of oral prednisone greater than or equal to 7.5 mg or cumulative doses greater than 10 g produce the most significant effects, and alternate-day glucocorticoids do not decrease the risk.⁶ Although inhaled corticosteroids are generally considered safer, high inhaled doses can also reduce bone mass.¹ An inverse relation between inhaled corticosteroid dose and bone mineral density (BMD) has been reported; doubling of the inhaled dose led to a decrease in lumbar bone mineral density of 0.16 standard deviations points (SDs).⁷ Menopausal status, sex, and age are other important contributing factors for the development of glucocorticosteroid-induced osteoporosis (CIO); surprisingly, men younger than 50 years may lose a higher percentage of bone than postmenopausal women.^2,87,98

Of greater clinical significance is the increased incidence of fractures, which is 2 to 4 times higher than that of similar glucocorticoid-naive patients.^1,11 Overall estimates of fractures during long-term steroid therapy range from 30% to 50%.^1,4,5 Ultimately, all fracture types can result in skeletal deformities that cause extreme pain, exacerbate the primary autoimmune or inflammatory disease state, and represent a tremendous financial burden.⁶ Most alarming is the estimated incidence of mortality following a hip fracture in patients requiring long-term corticosteroids, which ranges from 5% to 9% in men older than 50 years and from 1% to 3% in age-matched women.¹

Pathophysiology Of CIO

Glucocorticoids decrease bone formation and increase bone resorption through a number of different mechanisms that are beyond the scope of our article. However, a brief overview is warranted to better understand the role bisphosphonates have in the treatment of steroid-induced osteoporosis. On a cellular level, glucocorticoids directly inhibit osteoblast function at the glucocorticoid receptor. This inhibition results in decreased replication, differentiation, proliferation, and life span of the osteoblasts.^2,4,11 Subsequently the total amount of bone restored during each remodeling cycle is decreased by 30%, leading to a reduced mean wall thickness.^1,4,5 Also, glucocorticoids enhance the activity and increase the number of osteoclasts, leading to a greater number of active resorption surfaces.^2,11

Additionally, steroid-treated patients demonstrate a dose-dependent malabsorption of calcium due to direct impairment of the intestinal cell calcium transport process.^3,8 This decreased calcium absorption is often evident within the first 2 weeks of therapy.^4,8 Then a secondary hyperparathyroidism ensues and urinary calcium excretion becomes double that of non-steroid-treated patients.^3,4 Gonadal hormones—potent regulators of bone metabolism—are secreted to a much lower extent in patients treated with glucocorticoids.³ Treated men in particular have circulating testosterone concentrations of only 50% of those in a control group.⁴

Corticosteroids also reduce levels of prostaglandin E2, insulin-like growth factors, phosphate, type I collagen, and noncollagens including osteocalcin.^4,5 Finally, muscle atrophy and subsequent wasting may be the most observable result of glucocorticoid treatment, resulting in less of the mechanical stimuli required to generate new bone formation.^1,5,11

BMD Assessment and Preventive Measures

Although the risk of CIO and increased fracture risk is well cited, only 62% of surveyed physicians in one study¹² rated osteoporosis as one of the 3 most significant side effects discussed with patients taking high-dose steroids, and 2 cohort studies^13,14 found only a 5.6% and a 14% prescription rate for preventative medications when patients were prescribed glucocorticoids. Perhaps this mediocre reaction from primary care providers results from a shortage of clinical evidence to support preventive measures. Ideally, clinical decisions should be made using data concerning fracture risk reduction. Unfortunately, studies have not been sufficiently powered to assess fracture risk reduction, making it difficult to base therapeutic decisions on patient-oriented outcomes. The primary outcome assessed is the difference in percentage change from baseline in the BMD of the lumbar spine, femoral neck, and femoral trochanter between the treatment and placebo groups. It is important to recognize that the percentage change in BMD is a surrogate marker and has not been shown to directly correlate with decreased fracture risk. Other factors, such as bone strength and rate of turnover, may also contribute to fragility.

A BMD taken at one site, such as vertebrae or hip, correlates with risk of fracture at other sites; the best predictor of fracture, however, seems to be a BMD at the site in question.¹⁵ Patients at risk for CIO may have a comparatively higher risk of fracture at a given bone density than patients at risk for osteoporosis from other causes.¹⁶ A decrease in BMD of 1 SD below the mean of that of healthy adults aged 35 years will lead to a 1.5-fold to 3-fold increase in fracture risk,¹⁵ but this same bone density measure in the same woman taking corticosteroids may underestimate her risk of fracture.¹⁷

Presently the gold standard for BMD measurement is dual energy X-ray absorptiometry (DEXA). At this time, anteroposterior DEXA is recommended of both the lumbar spine and femoral neck for patients at risk for CIO.^17,18 If only one site can be obtained, the recommended site depends on the age of the patient. The lumbar spine is the site of choice for men and women younger than 60 years, while the femoral neck should be evaluated in men and women 60 years and older.¹⁸

Practice Guidelines

Two current sets of guidelines on the management of CIO are presently available: the recommendations of the 1998 United Kingdom (UK) Consensus Group¹⁷ and the 1996 American College of Rheumatology (ACR) Task Force on Osteoporosis Guidelines.¹⁸ Prevention and treatment options include risk factor modification (smoking cessation, fall prevention, and so forth) adequate calcium (1500 mg/day) and vitamin D3 intake (800 IU/day), hormone replacement when appropriate, bisphosphonates, and calcitonin (Figure). Other less-proven therapies such as thiazide diuretics, anabolic steroids, and fluoride are considered.

The ACR and UK guidelines differ considerably with respect to use of bisphosphonates in patients receiving long-term high-dose glucocorticoid therapy. The ACR task force recommends bisphosphonates only in patients with contraindications to hormone replacement therapy (HRT) or established osteoporosis.¹⁸ This class is not recommended in premenopausal women or men age younger than 50 years because of the lack of long-term safety data. The newer UK publication emphasizes that of all therapeutic options bisphosphonates have the most compelling BMD data and should be considered first-line treatment.¹⁷

Since the UK guidelines were published, 8 trials showing beneficial effects of bisphosphonates in the treatment of CIO have been published or presented at major medical meetings. Encouraging clinical data have resulted in alendronate and risedronate receiving United States Food and Drug Administration (FDA) approval for the indication of CIO. Our objective was to provide an evidence-based perspective of the use of bisphosphonates in corticosteroid-treated patients.

Methods

Literature searches were conducted independently by 2 authors using the MEDLINE, CURRENT CONTENTS, and HEALTHSTAR electronic databases to identify studies of bisphosphonates in the prevention or treatment of corticosteroid-induced osteoporosis. Medical subject headings and search terms included “bisphosphonates,” “diphosphonates,” “glucocorticoids,” “steroids,” and “osteoporosis.” We examined bibliographies of selected citations and review articles to obtain additional references. Unpublished information was obtained by hand-searching abstracts from recent meetings (1998) of the American Society for Bone and Mineral Research and the ACR. Studies were included if the design was randomized and controlled and if they evaluated the use of oral bisphosphonates in adults. Obtaining BMD results using the gold standard of measurement—central DEXA—was also necessary for study inclusion. Two authors assessed the methodologic quality of the trials using the 5-point criteria developed by Jadad and colleagues.¹⁹

Results

Bisphosphonates have been studied for use in CIO for more than a decade. Our search yielded a total of 34 citations of studies researching bisphosphonates for this indication. Twenty-one of these^20-40 were excluded for not meeting the specified inclusion criteria. The 13 studies included in our analysis^41-53 addressed the use of bisphosphonates in the prevention and treatment of corticosteroid-induced osteoporosis. Each bisphosphonate had the following number of citations and total number of patients in all studies combined: risedronate, 3 (n=638); etidronate, 8 (n=522); alendronate, 1 (n=477); and clodronate, 1 (n=74). The demographics pertaining to each study are summarized in [Table 1].

The methodologic quality of the trials was agreed on by the 2 reviewers. Scores ranged from 1 to 4 with 1 trial scoring 1,⁴¹ 2 trials scoring 2,^48,53 2 trials scoring 3,^45,47 and 8 trials scoring 4.^{42-44,46,49-52} All studies were reported as randomized; only 1 trial, however, described the randomization process.⁴⁴ Three of the 13 (23%) did not report using double-blind methodology.^41,48,53

The primary outcome assessed in all selected clinical trials is the difference in percentage change from baseline in the BMD of the lumbar spine, femoral neck, and femoral trochanter between the treatment and placebo groups. This complies with the ACR guidelines that recommend measurement at the lumbar spine and femoral neck. In regard to the studied primary outcomes, changes in BMD at the lumbar spine in treatment groups ranged from means of -0.137% to 4.9%; the control group values ranged from 3.7% to 0.98%. Three of 13 studies (23%) did not show a significant benefit at the lumbar spine when compared with the control group^41,47,48; however, 2 of these studies^41,48 used unusual doses of bisphosphonate. BMD changes at the femoral neck on treatment ranged from 1.28% to 3.6%; control group changes from baseline ranged from 3.6% to 3.64%. Nine of the 13 studies (69%) were not able to show statistical significance between the treatment and control groups at the femoral neck.^41-43,48-53 Changes in the treatment groups at the femoral trochanter ranged from -1.35% to 2.7%; placebo changes ranged from 3.06% to 1.5%. More than one half of studies reporting data at the femoral trochanter (6/10) showed that the treatment group was not significantly better than control group.^41,42,45,46 Specific BMD results from each trial are available in [Table 2]. Results varied based on the population, the bisphosphonate being investigated, the dosing regimen, and other distinctions in the study design.

Six studies^{42,43,45,46,49,52} addressed the possible differences in treatment effect based on sex and menopausal status. The effect of bisphosphonate treatment on BMD in postmenopausal women was significant in 4 of the trials.^42,45,46,52 One study included postmenopausal women taking HRT and found alendronate 10 mg led to a higher increase in BMD in women not receiving estrogen.⁴⁵ Only one study, however, found a significant effect on BMD in premenopausal women receiving etidronate.⁴² Also, an increase in the BMD of men was found to be significantly increased in one study.⁴⁶

The percentage change from baseline of the BMD of the distal and midshaft radius was measured in 1 study of etidronate and 2 studies of risedronate^42,46,47; statistical significance was not obtained in any of the results. Other secondary outcomes measured in selected citations were: biochemical markers of bone resorption and formation including pyridinium crosslinks, osteocalcin, parathyroid hormone, ionized calcium, serum alkaline phosphatase, serum alanine aminotransferase, and bone-specific alkaline phosphatase. However, these are not discussed in detail because of the uncertainty of their relationship to fracture risk.

Fracture Risk

Although not a specified primary end point, the risk of fracture development was addressed in 9 of the studies.^{42,43,45-49,51-52} Since the rate of fractures was low in most studies and no studies were powered on the basis of fracture risk, some data were reported without statistical analysis. Six studies^{42,43,45-47,51} analyzed the difference between treatment and control groups with respect to fracture risk. Three of these studies reported at least a trend in reduced fracture rates^42,45,46; however, only 1 (in abstract) found a statistical difference in the overall population.⁴⁷ A 10.1% reduction in vertebral fractures was found in patients receiving risedronate (either 2.5 mg or 5 mg) at 12 months (P=.021).⁴⁷ According to an abstract, when these results were pooled with another trial⁴⁶ it was found that risedronate 5.0 mg statistically decreased the incidence of vertebral fractures at 1 year (16.2% vs 5.4%, P=.01).⁵⁴ Therefore, it would be necessary to treat 10 CIO patients for 1 year with risedronate 5 mg per day to prevent one vertebral fracture.

Although the fracture rate was not significant in most trials, the subpopulation of postmenopausal women experienced the greatest number of fractures and seemed to benefit the most from treatment. Two studies found borderline significance in the fracture rate when postmenopausal women were analyzed separately.^42,45 Etidronate and alendronate use led to an absolute risk reduction of 18.7% (P=.05) and 8.6% (P=.05), respectively. Men were the only other group to experience fractures in any of the studies. No fractures were reported in the premenopausal population.

When interpreting the results of these studies it is important to remember that trial results can vary according to a number of factors, including: definition of fracture, initial number of fractures, baseline BMD, and population differences. Therefore, the reported data should not be extrapolated to all populations. It should also be mentioned that all of the fracture data presented consisted of radiologically reported fractures, which does not necessarily correlate with clinical fractures.

Safety and Tolerability

The withdrawal rates of the included studies ranged from 0% to 62% ([Table 3]). Of the dropouts reported, 42% were secondary to protocol violations or administrative reasons, and 16% were because of noncompliance. Twenty percent of study withdrawals were because of adverse events, of which 78% (29/37) were deemed not related to the treatment regimen. No significant difference was reported in dropout rates between the treatment and control groups.

In most of the studies no statistical significance concerning adverse effects was found when the treatment group was compared with the control group. Nine studies^42-47,50,51 discussed the subgroup of gastrointestinal (GI) adverse events, since this has been the greatest historical concern about the use of bisphosphonates. Eight studies found no difference in overall GI effects, and one study⁴⁵ reported a statistical trend for increased GI side effects with dose escalation of alendronate. This trend was probably secondary to an increase in abdominal pain in the treatment group when compared with the placebo group. In one study,⁴⁶ diarrhea was more common in patients receiving 5 mg risedronate than those taking placebo (number needed to harm=15), although significance was not reported.

Discussion

The studies examining the use of bisphosphonates for the prevention or treatment of CIO are difficult to interpret because of the various bisphosphonates and regimens, the heterogenicity in populations, the effect of the underlying disease on bone, and concomitant therapeutic interventions. A review of the currently published literature suggests bisphosphonates effectively prevent vertebral bone loss in patients treated with long-term corticosteroids. Patients receiving steroids for more than 3 months (secondary prevention) gained bone mass when placed on a bisphosphonate, while patients naive to steroids (primary prevention) maintained more bone density than the control group.^{42-44,46,47,50,51,53} This reinforces the fact that patients taking steroids for more than 3 months have already lost bone that can be partially regained with bisphosphonate treatment. In contrast, patients given bisphosphonates who were naive to steroids did not have significant changes in BMD from baseline; this group, however, was able to maintain bone density while the control group lost bone density.

Data regarding the impact of bisphosphonates on the risk of CIO-induced fractures are sparse and inconclusive. This is not surprising since most of the trials have been of relatively short duration (<2 years), and have not been sufficiently powered to show fracture reduction. Postmenopausal women not taking estrogen seem to benefit most from using bisphosphonates for the prevention of bone loss and of vertebral fractures in CIO.

A meta-analysis of a similar set of data based on a Cochrane systematic review of published literature was reported in 1999.⁵⁵ The authors state that of the small number of controlled clinical trials examining the use of bone-sparing agents in patients at risk for CIO bisphosphonates have shown some of the best evidence for reducing bone loss, particularly at the lumbar spine. They also concluded that bone density changes correlate with fracture risk in patients with CIO, but there are insufficient data to make conclusions regarding fracture risk reduction and use of bisphosphonates.

Overall, adverse effects of the bisphosphonates were minimal, and no statistical significance was found in studied populations when compared with control groups. However, the incidence of GI adverse events with alendronate may be as high as 15% in clinical practice, despite low incidence rates in phase III trials, possibly because of administration errors.⁵⁶ Directions for use are the same for all oral bisphosphonates, and include staying upright and not eating for at least 30 minutes after administration. Because oral bioavailability is usually less than 5% even on an empty stomach, bisphosphonates should be taken with a full glass of water in the morning after an overnight fast.^11,56,57

Implications for further research

Although current evidence supporting bisphosphonate use documents efficacy in BMD changes and trends toward reduced vertebral fracture risk in the treatment of CIO, more research is clearly desirable. The recent FDA approval of alendronate and risedronate has highlighted the need for aggressive measures to prevent and treat CIO. More research is needed in large studies to assess vertebral and nonvertebral fractures. Head-to-head comparative trials of bisphosphonates with other pharmacologic options, such as hormonal therapy or calcitonin are essential to establish evidence-based clinical guidelines. Studies addressing combination therapy with bisphosphonates and HRT in postmenopausal women would also be useful. Finally, comparative studies of the various bisphosphonates in relation to one another are needed. These synthetic pyrophosphate analogs possess a broad range of potencies, selectivity, and adverse effect profiles. Because each bisphosphonate has unique biological, chemical, and physiochemical properties, the results of one bisphosphonate study cannot be extrapolated to other compounds within the same drug class.^3,56,58,59 Differences in study design, patient populations, and other confounding variables also prevent the assumption of a class effect with certain research findings. This distinction becomes especially important with the new bisphosphonates on the horizon (clodronate, tiludronate, ibandronate, and zoledronate).

Although the relationship of BMD to fracture risk is well documented in postmenopausal osteoporosis, the correlation is not as well established in glucocorticoid-induced bone loss.^2,60 The present data suggest a trend in the reduction of fracture risk observed with the use of bisphosphonates in the treatment of glucocorticoid-induced osteoporosis. This information is promising but requires definitive confirmation in larger trials. Because of the prolonged skeletal retention of bisphosphonates, more extended trials are also needed to clearly establish the safety of long-term therapy in younger patients.

Recommendations for clinical practice

All patients beginning high-dose (Ž7.5 mg/day prednisone) long-term (>6 months) glucocorticoid therapy should be evaluated for pharmacologic prophylaxis against osteoporosis. Because the majority of bone loss occurs within the first 6 months of therapy, clinicians must vigilantly develop a preventative plan in advance, preferably before the glucocorticoid prescription is given to the patient. Current guidelines recommend a baseline BMD measurement of all patients with DEXA to determine the risk of osteoporosis and the monitor the efficacy of the chosen preventative measures throughout the course of therapy. Initial BMD measurements expressed as a T score are essential tools for constructing an individualized strategy. Critical T scores of less than or equal to 1.0 are indicative of high-risk patients.¹⁸ In addition to the risk imposed on patients secondary to their underlying disease state, other patients with strong predictive factors include those who are elderly and at greater risk of falling and those who have a previous history of fractures.

BMD measurements may be repeated in 6 to 12 months, depending on initial bone mass. If BMD has decreased by more than 5% from baseline, the initial choice of therapy should be changed or expanded.^18,61

Both glucocorticoid doses and the length of therapy require continuous reevaluation to decrease the total cumulative dose. If possible, dose maximization of inhaled and topical corticosteroids is recommended before progression to the oral form. More obvious preventive lifestyle modifications include smoking cessation, maintenance of healthy body weight, regular weight-bearing exercise, decreased alcohol consumption, sodium restriction, and increased dietary calcium intake. Most patients will require additional calcium supplementation to meet the ACR recommendation of 1500 mg per day, and patients at risk for vitamin D deficiency may also require supplementation.¹⁸

However, even with appropriate supplementation 1 in 6 corticosteroid-treated patients will experience a radiographically detected vertebral fracture within 12 months.^46,47,54 Therefore, alternative treatment regimens are needed. Bisphosphonate therapy looks promising in the prevention and treatment of CIO with respect to positively affecting BMD. As a result of this encouraging data, alendronate and risedronate have received FDA approval for the indication of CIO. However, it remains to be seen whether the use of these agents can lead to a reduction in fractures, both vertebral and nonvertebral. Subpopulations at highest risk (postmenopausal women and men) may benefit most with regard to fracture risk reduction, but more data are needed before bisphosphonates should be routinely recommended as first-line preventive therapy. Once osteoporosis is established (regardless of the pathogenesis), treatment should be aggressive to prevent further loss of bone density. In this population bisphosphonates are an appropriate therapeutic option, and further data will clarify their impact on the risk of fractures.

The detrimental effects of glucocorticoids on bone have long been recognized in the medical literature. An average of 5% of bone mass is lost during the first year of long-term therapy, and annualized rates of loss range from 0.3% to 3%.^1,2 This bone reduction occurs most rapidly during the first 6 to 12 months of glucocorticoid therapy and is dose-dependent and time-dependant.^3-5 Daily doses of oral prednisone greater than or equal to 7.5 mg or cumulative doses greater than 10 g produce the most significant effects, and alternate-day glucocorticoids do not decrease the risk.⁶ Although inhaled corticosteroids are generally considered safer, high inhaled doses can also reduce bone mass.¹ An inverse relation between inhaled corticosteroid dose and bone mineral density (BMD) has been reported; doubling of the inhaled dose led to a decrease in lumbar bone mineral density of 0.16 standard deviations points (SDs).⁷ Menopausal status, sex, and age are other important contributing factors for the development of glucocorticosteroid-induced osteoporosis (CIO); surprisingly, men younger than 50 years may lose a higher percentage of bone than postmenopausal women.^2,87,98

Of greater clinical significance is the increased incidence of fractures, which is 2 to 4 times higher than that of similar glucocorticoid-naive patients.^1,11 Overall estimates of fractures during long-term steroid therapy range from 30% to 50%.^1,4,5 Ultimately, all fracture types can result in skeletal deformities that cause extreme pain, exacerbate the primary autoimmune or inflammatory disease state, and represent a tremendous financial burden.⁶ Most alarming is the estimated incidence of mortality following a hip fracture in patients requiring long-term corticosteroids, which ranges from 5% to 9% in men older than 50 years and from 1% to 3% in age-matched women.¹

Pathophysiology Of CIO

Glucocorticoids decrease bone formation and increase bone resorption through a number of different mechanisms that are beyond the scope of our article. However, a brief overview is warranted to better understand the role bisphosphonates have in the treatment of steroid-induced osteoporosis. On a cellular level, glucocorticoids directly inhibit osteoblast function at the glucocorticoid receptor. This inhibition results in decreased replication, differentiation, proliferation, and life span of the osteoblasts.^2,4,11 Subsequently the total amount of bone restored during each remodeling cycle is decreased by 30%, leading to a reduced mean wall thickness.^1,4,5 Also, glucocorticoids enhance the activity and increase the number of osteoclasts, leading to a greater number of active resorption surfaces.^2,11

Additionally, steroid-treated patients demonstrate a dose-dependent malabsorption of calcium due to direct impairment of the intestinal cell calcium transport process.^3,8 This decreased calcium absorption is often evident within the first 2 weeks of therapy.^4,8 Then a secondary hyperparathyroidism ensues and urinary calcium excretion becomes double that of non-steroid-treated patients.^3,4 Gonadal hormones—potent regulators of bone metabolism—are secreted to a much lower extent in patients treated with glucocorticoids.³ Treated men in particular have circulating testosterone concentrations of only 50% of those in a control group.⁴

Corticosteroids also reduce levels of prostaglandin E2, insulin-like growth factors, phosphate, type I collagen, and noncollagens including osteocalcin.^4,5 Finally, muscle atrophy and subsequent wasting may be the most observable result of glucocorticoid treatment, resulting in less of the mechanical stimuli required to generate new bone formation.^1,5,11

BMD Assessment and Preventive Measures

Although the risk of CIO and increased fracture risk is well cited, only 62% of surveyed physicians in one study¹² rated osteoporosis as one of the 3 most significant side effects discussed with patients taking high-dose steroids, and 2 cohort studies^13,14 found only a 5.6% and a 14% prescription rate for preventative medications when patients were prescribed glucocorticoids. Perhaps this mediocre reaction from primary care providers results from a shortage of clinical evidence to support preventive measures. Ideally, clinical decisions should be made using data concerning fracture risk reduction. Unfortunately, studies have not been sufficiently powered to assess fracture risk reduction, making it difficult to base therapeutic decisions on patient-oriented outcomes. The primary outcome assessed is the difference in percentage change from baseline in the BMD of the lumbar spine, femoral neck, and femoral trochanter between the treatment and placebo groups. It is important to recognize that the percentage change in BMD is a surrogate marker and has not been shown to directly correlate with decreased fracture risk. Other factors, such as bone strength and rate of turnover, may also contribute to fragility.

A BMD taken at one site, such as vertebrae or hip, correlates with risk of fracture at other sites; the best predictor of fracture, however, seems to be a BMD at the site in question.¹⁵ Patients at risk for CIO may have a comparatively higher risk of fracture at a given bone density than patients at risk for osteoporosis from other causes.¹⁶ A decrease in BMD of 1 SD below the mean of that of healthy adults aged 35 years will lead to a 1.5-fold to 3-fold increase in fracture risk,¹⁵ but this same bone density measure in the same woman taking corticosteroids may underestimate her risk of fracture.¹⁷

Presently the gold standard for BMD measurement is dual energy X-ray absorptiometry (DEXA). At this time, anteroposterior DEXA is recommended of both the lumbar spine and femoral neck for patients at risk for CIO.^17,18 If only one site can be obtained, the recommended site depends on the age of the patient. The lumbar spine is the site of choice for men and women younger than 60 years, while the femoral neck should be evaluated in men and women 60 years and older.¹⁸

Practice Guidelines

Two current sets of guidelines on the management of CIO are presently available: the recommendations of the 1998 United Kingdom (UK) Consensus Group¹⁷ and the 1996 American College of Rheumatology (ACR) Task Force on Osteoporosis Guidelines.¹⁸ Prevention and treatment options include risk factor modification (smoking cessation, fall prevention, and so forth) adequate calcium (1500 mg/day) and vitamin D3 intake (800 IU/day), hormone replacement when appropriate, bisphosphonates, and calcitonin (Figure). Other less-proven therapies such as thiazide diuretics, anabolic steroids, and fluoride are considered.

The ACR and UK guidelines differ considerably with respect to use of bisphosphonates in patients receiving long-term high-dose glucocorticoid therapy. The ACR task force recommends bisphosphonates only in patients with contraindications to hormone replacement therapy (HRT) or established osteoporosis.¹⁸ This class is not recommended in premenopausal women or men age younger than 50 years because of the lack of long-term safety data. The newer UK publication emphasizes that of all therapeutic options bisphosphonates have the most compelling BMD data and should be considered first-line treatment.¹⁷

Since the UK guidelines were published, 8 trials showing beneficial effects of bisphosphonates in the treatment of CIO have been published or presented at major medical meetings. Encouraging clinical data have resulted in alendronate and risedronate receiving United States Food and Drug Administration (FDA) approval for the indication of CIO. Our objective was to provide an evidence-based perspective of the use of bisphosphonates in corticosteroid-treated patients.

Methods

Literature searches were conducted independently by 2 authors using the MEDLINE, CURRENT CONTENTS, and HEALTHSTAR electronic databases to identify studies of bisphosphonates in the prevention or treatment of corticosteroid-induced osteoporosis. Medical subject headings and search terms included “bisphosphonates,” “diphosphonates,” “glucocorticoids,” “steroids,” and “osteoporosis.” We examined bibliographies of selected citations and review articles to obtain additional references. Unpublished information was obtained by hand-searching abstracts from recent meetings (1998) of the American Society for Bone and Mineral Research and the ACR. Studies were included if the design was randomized and controlled and if they evaluated the use of oral bisphosphonates in adults. Obtaining BMD results using the gold standard of measurement—central DEXA—was also necessary for study inclusion. Two authors assessed the methodologic quality of the trials using the 5-point criteria developed by Jadad and colleagues.¹⁹

Results

Bisphosphonates have been studied for use in CIO for more than a decade. Our search yielded a total of 34 citations of studies researching bisphosphonates for this indication. Twenty-one of these^20-40 were excluded for not meeting the specified inclusion criteria. The 13 studies included in our analysis^41-53 addressed the use of bisphosphonates in the prevention and treatment of corticosteroid-induced osteoporosis. Each bisphosphonate had the following number of citations and total number of patients in all studies combined: risedronate, 3 (n=638); etidronate, 8 (n=522); alendronate, 1 (n=477); and clodronate, 1 (n=74). The demographics pertaining to each study are summarized in [Table 1].

The methodologic quality of the trials was agreed on by the 2 reviewers. Scores ranged from 1 to 4 with 1 trial scoring 1,⁴¹ 2 trials scoring 2,^48,53 2 trials scoring 3,^45,47 and 8 trials scoring 4.^{42-44,46,49-52} All studies were reported as randomized; only 1 trial, however, described the randomization process.⁴⁴ Three of the 13 (23%) did not report using double-blind methodology.^41,48,53

The primary outcome assessed in all selected clinical trials is the difference in percentage change from baseline in the BMD of the lumbar spine, femoral neck, and femoral trochanter between the treatment and placebo groups. This complies with the ACR guidelines that recommend measurement at the lumbar spine and femoral neck. In regard to the studied primary outcomes, changes in BMD at the lumbar spine in treatment groups ranged from means of -0.137% to 4.9%; the control group values ranged from 3.7% to 0.98%. Three of 13 studies (23%) did not show a significant benefit at the lumbar spine when compared with the control group^41,47,48; however, 2 of these studies^41,48 used unusual doses of bisphosphonate. BMD changes at the femoral neck on treatment ranged from 1.28% to 3.6%; control group changes from baseline ranged from 3.6% to 3.64%. Nine of the 13 studies (69%) were not able to show statistical significance between the treatment and control groups at the femoral neck.^41-43,48-53 Changes in the treatment groups at the femoral trochanter ranged from -1.35% to 2.7%; placebo changes ranged from 3.06% to 1.5%. More than one half of studies reporting data at the femoral trochanter (6/10) showed that the treatment group was not significantly better than control group.^41,42,45,46 Specific BMD results from each trial are available in [Table 2]. Results varied based on the population, the bisphosphonate being investigated, the dosing regimen, and other distinctions in the study design.

Six studies^{42,43,45,46,49,52} addressed the possible differences in treatment effect based on sex and menopausal status. The effect of bisphosphonate treatment on BMD in postmenopausal women was significant in 4 of the trials.^42,45,46,52 One study included postmenopausal women taking HRT and found alendronate 10 mg led to a higher increase in BMD in women not receiving estrogen.⁴⁵ Only one study, however, found a significant effect on BMD in premenopausal women receiving etidronate.⁴² Also, an increase in the BMD of men was found to be significantly increased in one study.⁴⁶

The percentage change from baseline of the BMD of the distal and midshaft radius was measured in 1 study of etidronate and 2 studies of risedronate^42,46,47; statistical significance was not obtained in any of the results. Other secondary outcomes measured in selected citations were: biochemical markers of bone resorption and formation including pyridinium crosslinks, osteocalcin, parathyroid hormone, ionized calcium, serum alkaline phosphatase, serum alanine aminotransferase, and bone-specific alkaline phosphatase. However, these are not discussed in detail because of the uncertainty of their relationship to fracture risk.

Fracture Risk

Although not a specified primary end point, the risk of fracture development was addressed in 9 of the studies.^{42,43,45-49,51-52} Since the rate of fractures was low in most studies and no studies were powered on the basis of fracture risk, some data were reported without statistical analysis. Six studies^{42,43,45-47,51} analyzed the difference between treatment and control groups with respect to fracture risk. Three of these studies reported at least a trend in reduced fracture rates^42,45,46; however, only 1 (in abstract) found a statistical difference in the overall population.⁴⁷ A 10.1% reduction in vertebral fractures was found in patients receiving risedronate (either 2.5 mg or 5 mg) at 12 months (P=.021).⁴⁷ According to an abstract, when these results were pooled with another trial⁴⁶ it was found that risedronate 5.0 mg statistically decreased the incidence of vertebral fractures at 1 year (16.2% vs 5.4%, P=.01).⁵⁴ Therefore, it would be necessary to treat 10 CIO patients for 1 year with risedronate 5 mg per day to prevent one vertebral fracture.

Although the fracture rate was not significant in most trials, the subpopulation of postmenopausal women experienced the greatest number of fractures and seemed to benefit the most from treatment. Two studies found borderline significance in the fracture rate when postmenopausal women were analyzed separately.^42,45 Etidronate and alendronate use led to an absolute risk reduction of 18.7% (P=.05) and 8.6% (P=.05), respectively. Men were the only other group to experience fractures in any of the studies. No fractures were reported in the premenopausal population.

When interpreting the results of these studies it is important to remember that trial results can vary according to a number of factors, including: definition of fracture, initial number of fractures, baseline BMD, and population differences. Therefore, the reported data should not be extrapolated to all populations. It should also be mentioned that all of the fracture data presented consisted of radiologically reported fractures, which does not necessarily correlate with clinical fractures.

Safety and Tolerability

The withdrawal rates of the included studies ranged from 0% to 62% ([Table 3]). Of the dropouts reported, 42% were secondary to protocol violations or administrative reasons, and 16% were because of noncompliance. Twenty percent of study withdrawals were because of adverse events, of which 78% (29/37) were deemed not related to the treatment regimen. No significant difference was reported in dropout rates between the treatment and control groups.

In most of the studies no statistical significance concerning adverse effects was found when the treatment group was compared with the control group. Nine studies^42-47,50,51 discussed the subgroup of gastrointestinal (GI) adverse events, since this has been the greatest historical concern about the use of bisphosphonates. Eight studies found no difference in overall GI effects, and one study⁴⁵ reported a statistical trend for increased GI side effects with dose escalation of alendronate. This trend was probably secondary to an increase in abdominal pain in the treatment group when compared with the placebo group. In one study,⁴⁶ diarrhea was more common in patients receiving 5 mg risedronate than those taking placebo (number needed to harm=15), although significance was not reported.

Discussion

The studies examining the use of bisphosphonates for the prevention or treatment of CIO are difficult to interpret because of the various bisphosphonates and regimens, the heterogenicity in populations, the effect of the underlying disease on bone, and concomitant therapeutic interventions. A review of the currently published literature suggests bisphosphonates effectively prevent vertebral bone loss in patients treated with long-term corticosteroids. Patients receiving steroids for more than 3 months (secondary prevention) gained bone mass when placed on a bisphosphonate, while patients naive to steroids (primary prevention) maintained more bone density than the control group.^{42-44,46,47,50,51,53} This reinforces the fact that patients taking steroids for more than 3 months have already lost bone that can be partially regained with bisphosphonate treatment. In contrast, patients given bisphosphonates who were naive to steroids did not have significant changes in BMD from baseline; this group, however, was able to maintain bone density while the control group lost bone density.

Data regarding the impact of bisphosphonates on the risk of CIO-induced fractures are sparse and inconclusive. This is not surprising since most of the trials have been of relatively short duration (<2 years), and have not been sufficiently powered to show fracture reduction. Postmenopausal women not taking estrogen seem to benefit most from using bisphosphonates for the prevention of bone loss and of vertebral fractures in CIO.

A meta-analysis of a similar set of data based on a Cochrane systematic review of published literature was reported in 1999.⁵⁵ The authors state that of the small number of controlled clinical trials examining the use of bone-sparing agents in patients at risk for CIO bisphosphonates have shown some of the best evidence for reducing bone loss, particularly at the lumbar spine. They also concluded that bone density changes correlate with fracture risk in patients with CIO, but there are insufficient data to make conclusions regarding fracture risk reduction and use of bisphosphonates.

Overall, adverse effects of the bisphosphonates were minimal, and no statistical significance was found in studied populations when compared with control groups. However, the incidence of GI adverse events with alendronate may be as high as 15% in clinical practice, despite low incidence rates in phase III trials, possibly because of administration errors.⁵⁶ Directions for use are the same for all oral bisphosphonates, and include staying upright and not eating for at least 30 minutes after administration. Because oral bioavailability is usually less than 5% even on an empty stomach, bisphosphonates should be taken with a full glass of water in the morning after an overnight fast.^11,56,57

Implications for further research

Although current evidence supporting bisphosphonate use documents efficacy in BMD changes and trends toward reduced vertebral fracture risk in the treatment of CIO, more research is clearly desirable. The recent FDA approval of alendronate and risedronate has highlighted the need for aggressive measures to prevent and treat CIO. More research is needed in large studies to assess vertebral and nonvertebral fractures. Head-to-head comparative trials of bisphosphonates with other pharmacologic options, such as hormonal therapy or calcitonin are essential to establish evidence-based clinical guidelines. Studies addressing combination therapy with bisphosphonates and HRT in postmenopausal women would also be useful. Finally, comparative studies of the various bisphosphonates in relation to one another are needed. These synthetic pyrophosphate analogs possess a broad range of potencies, selectivity, and adverse effect profiles. Because each bisphosphonate has unique biological, chemical, and physiochemical properties, the results of one bisphosphonate study cannot be extrapolated to other compounds within the same drug class.^3,56,58,59 Differences in study design, patient populations, and other confounding variables also prevent the assumption of a class effect with certain research findings. This distinction becomes especially important with the new bisphosphonates on the horizon (clodronate, tiludronate, ibandronate, and zoledronate).

Although the relationship of BMD to fracture risk is well documented in postmenopausal osteoporosis, the correlation is not as well established in glucocorticoid-induced bone loss.^2,60 The present data suggest a trend in the reduction of fracture risk observed with the use of bisphosphonates in the treatment of glucocorticoid-induced osteoporosis. This information is promising but requires definitive confirmation in larger trials. Because of the prolonged skeletal retention of bisphosphonates, more extended trials are also needed to clearly establish the safety of long-term therapy in younger patients.

Recommendations for clinical practice

All patients beginning high-dose (Ž7.5 mg/day prednisone) long-term (>6 months) glucocorticoid therapy should be evaluated for pharmacologic prophylaxis against osteoporosis. Because the majority of bone loss occurs within the first 6 months of therapy, clinicians must vigilantly develop a preventative plan in advance, preferably before the glucocorticoid prescription is given to the patient. Current guidelines recommend a baseline BMD measurement of all patients with DEXA to determine the risk of osteoporosis and the monitor the efficacy of the chosen preventative measures throughout the course of therapy. Initial BMD measurements expressed as a T score are essential tools for constructing an individualized strategy. Critical T scores of less than or equal to 1.0 are indicative of high-risk patients.¹⁸ In addition to the risk imposed on patients secondary to their underlying disease state, other patients with strong predictive factors include those who are elderly and at greater risk of falling and those who have a previous history of fractures.

BMD measurements may be repeated in 6 to 12 months, depending on initial bone mass. If BMD has decreased by more than 5% from baseline, the initial choice of therapy should be changed or expanded.^18,61

Both glucocorticoid doses and the length of therapy require continuous reevaluation to decrease the total cumulative dose. If possible, dose maximization of inhaled and topical corticosteroids is recommended before progression to the oral form. More obvious preventive lifestyle modifications include smoking cessation, maintenance of healthy body weight, regular weight-bearing exercise, decreased alcohol consumption, sodium restriction, and increased dietary calcium intake. Most patients will require additional calcium supplementation to meet the ACR recommendation of 1500 mg per day, and patients at risk for vitamin D deficiency may also require supplementation.¹⁸

However, even with appropriate supplementation 1 in 6 corticosteroid-treated patients will experience a radiographically detected vertebral fracture within 12 months.^46,47,54 Therefore, alternative treatment regimens are needed. Bisphosphonate therapy looks promising in the prevention and treatment of CIO with respect to positively affecting BMD. As a result of this encouraging data, alendronate and risedronate have received FDA approval for the indication of CIO. However, it remains to be seen whether the use of these agents can lead to a reduction in fractures, both vertebral and nonvertebral. Subpopulations at highest risk (postmenopausal women and men) may benefit most with regard to fracture risk reduction, but more data are needed before bisphosphonates should be routinely recommended as first-line preventive therapy. Once osteoporosis is established (regardless of the pathogenesis), treatment should be aggressive to prevent further loss of bone density. In this population bisphosphonates are an appropriate therapeutic option, and further data will clarify their impact on the risk of fractures.