Original Research

Continuity of care has been shown to be associated with a variety of positive outcomes including patient satisfaction,¹ compliance with medication regimens,² and health services utilization.³ High continuity of physician care is also associated with a decreased likelihood of future hospitalization.^4,5 One explanation for this finding is that continuity leads to increased knowledge and trust between the patient and the physician.⁶ This increased knowledge and trust may make it easier for the physician to manage medical problems in the office or over the phone and thereby avoid hospitalizing the patient. Similarly, having care continuity with a specific physician is significantly associated with a decreased likelihood of emergency department (ED) use.⁷

Although data are continually accumulating indicating that continuity between a patient and a clinician has positive benefits, it is clear that it is not always easy to achieve high continuity in practice.⁸ This is increasingly the case as physicians work in larger groups where there is likely to be less continuity with an individual provider.⁹ According to data from the 1987 National Medical Expenditure Survey, only half of all patients have high continuity with a physician.⁸ This situation may be worsening as more Americans are enrolled in managed health care plans that frequently change their panel of approved physicians.¹⁰ As a result, patients may be forced to change their physician on multiple occasions, which could have negative consequences on the management of their medical problems.

Since continuity between a patient and a physician has positive health benefits in the general population, it makes intuitive sense that these benefits would be greatest for persons with chronic conditions. Asthma is a common chronic disease affecting 14 to 15 million people in the United States and is the most common chronic disease of childhood, affecting nearly 5 million children in the United States.¹¹ Asthma accounts for more than 470,000 hospitalizations and more than 5000 deaths annually.¹² Rates of hospitalization for asthma have been increasing and reflect that it is the most common discharge diagnosis among children.¹³ In addition to the morbidity and mortality associated with this disease, some treatments have possible adverse effects.¹⁴ Thus, the benefits that could accrue from improved health care delivery are considerable.

Patients with asthma have a greater desire than the general population to maintain continuity with a physician, even when the visit is not for asthma.¹⁵ A possible reason continuity of care may be important is that the management of chronic conditions requires ongoing monitoring and decisions about when changes in therapy are appropriate. When a patient is seen by the same physician, that provider is more likely to know when tests and treatment changes are indicated. Effective physician-patient interaction is an important component of health care delivery. For patients with chronic illnesses, interaction processes that include the physician giving more information and the patient having greater control during the visit are associated with better health.¹⁵ Physician and patient roles in managing asthma differ across physician-patient relationships,¹⁶ but the relation between continuity of care and the physician-patient relationship has not been specified for patients with asthma.

It is important to determine the role and importance of continuity in the delivery of health care to patients with asthma. The purpose of our study was to examine the relationship between continuity of care and physician-patient interaction among patients with and without asthma.

Methods

The data for our study came from an omnibus survey of the patient satisfaction of recipients of the Kentucky Medicaid program. Our study is a secondary analysis of that large data set. In 1997 a survey was mailed to a stratified random sample of adult participants (aged Ž18 years) in the Kentucky Medicaid fee-for-service program. The design followed the Dillman method, using an initial wave of surveys with reminder postcards and 2 additional waves of surveys to nonresponders.¹⁷ The response rate to the survey was 60%, with a total sample of 2308.

The survey items were based on the Health Employer Data and Information Set (HEDIS 3.0) customer satisfaction survey,¹⁸ the Consumer Assessments of Health Plans Study (CAHPS)¹⁹ completed by the Agency for Health Care Policy and Research (AHCPR), and satisfaction surveys used in Kentucky during previous assessments in 1987, 1989, and 1991 (KENPAC program). The internal review board of the University of Kentucky approved the survey.

Our sample was limited to individuals who visited primary care physicians often enough to assess continuity. They reported utilization on a single survey item assessing outpatient visits. A total of 1726 respondents reported making 2 or more visits to a physician’s office, clinic, or ED during the previous 12 months. In this group 404 reported having asthma. The prevalence of asthma in this population is higher than in the general population but not exceptionally higher than that found in Medicaid programs.²⁰ Continuity was measured by the question “Over the past 12 months, when you went for medical care, how often did you see the same doctor or provider?” The 4 response categories were “always,” “most of the time,” “sometimes,” and “rarely or never” (reverse scored from 1=”rarely or never” to 4=”always”). Although continuity can be measured in a variety of ways, patient self-reports have commonly been used.^1,21 In the present context, perceptions of continuity may be no less important than actual continuity in predicting patients’ evaluative ratings of physician-patient interaction.

The outcome measures were patient assessments of the health care they had received in Medicaid programs during the past 12 months. They are consistent with other self-report measures and were created for the CAHPS by the AHCPR (now the Agency for Healthcare Research and Quality). The present survey reference to 12 months differed from the original CAHPS survey reference to 6 months. The measures included an item about provider communication (“Doctor or provider listened to you and talked with you about your care”) and an item about patient influence (“Your ability to influence the treatment you received from a doctor or provider for your health problems”). These were measured on 5-point scales (reverse scored from 1=poor to 5=excellent).

Analysis

We computed bivariate analyses comparing characteristics of the groups of patients with and without asthma and assessing the relation between continuity of care and physician-patient interaction for each group (chi-square, Student t test). Then the relation between continuity of care and physician-patient interaction was evaluated in multivariate linear regression analyses in the presence of the following control variables: age, sex, education, race, number of visits, general health, health improvement, and life satisfaction. Among the variables available for analysis, these were identified as most likely to confound the relation between continuity of care and patient perceptions of the physician-patient relationship. General health (“In general, would you say your health is:”), health improvement (“Compared to one year ago, how would you rate your health in general now?”), and life satisfaction (“Overall, how satisfied or dissatisfied are you with how your life is going?”) were rated using 5-point scales coded so that higher scores mean better health, more improved health, and greater overall life satisfaction, respectively. We performed separate linear regression models for the patients with and without asthma and examined the contribution of continuity to each model. Linear regression models with all respondents combined were also performed, and interaction terms were entered for asthma status interacting with the other independent variables. Only respondents who had complete data on all items could be included in the regression analyses. We conducted all analyses with SAS statistical programming software release 6.09 (SAS Institute, Inc; Cary, NC), using complete data for each item.

Results

The characteristics of the respondents appear in Table 1. Their demographic characteristics were typical for the Medicaid population in Kentucky. Although most of the respondents were white, more of the patients with asthma were white than were those without the condition. In addition, the asthma patients reported higher numbers of health care visits and poorer health than those without asthma. Reported continuity of care and respondents’ perceptions of provider communication and patient influence are shown in Table 2. The respondents with and without asthma did not differ on these variables of interest. More than half of the respondents (58.8%) reported always seeing the same health care provider in the past 12 months (scale mean ± standard deviation=3.5 ± 0.7).

Bivariate Analyses

In bivariate analyses, ratings of the physician-patient relationship were compared across continuity of care categories. Individuals who “always” saw the same doctor or provider were compared with a category of “less than always,” which had been collapsed across “most of the time,” “sometimes,” and “rarely or never.” On average, respondents rated provider communication and patient influence between “good” and “very good.” Both persons who did and did not have asthma who saw the same doctor or provider for all their health care visits rated provider communication and patient influence significantly higher than did individuals who had less continuity (P <.01). This is shown in Table 3.

Regression Analyses

We computed separate linear regression models for individuals who did and did not have asthma to assess the contribution of continuity of care in the presence of control variables in predicting ratings of provider communication and patient influence for these groups. All 4 response levels were included in the continuity of care variable. The correlation matrix for the independent variables produced no correlations between independent variables that exceeded 0.5, suggesting that these variables could be included in the same analysis. The P values and standardized regression coefficients for the independent variables in the asthmatic and nonasthmatic models are presented for the provider communication models in Table 4 and for the patient influence models in Table 5. All models were significant at P <.05.

For persons with asthma, continuity of care was the only variable (P=.01) that significantly contributed to the provider communication model (Table 4, Model 1) and the only variable (P=.02) other than life satisfaction (P=.04) that contributed to the patient influence model (Table 5, Model 1). In the provider communication and patient influence models, the standardized estimates for the continuity parameter were 0.15 and 0.14, respectively, higher than any other estimates in the models. The nonstandardized parameter estimates for continuity were 0.26 and 0.25, respectively.

For persons who did not have asthma, continuity of care significantly contributed (P=.001) to both the provider communication (Table 4, Model 2) and patient influence models (Table 5, Model 2). Unlike the models for persons with asthma, 5 additional variables significantly contributed to these models (P≤.01): age, number of visits, general health, health improvement, and life satisfaction. The standardized parameter estimates for continuity were similar to those in the asthmatic models (0.14 for each). Continuity ranked only third among the estimates in the provider communication model and second among the estimates in the patient influence model. The nonstandardized parameter estimates for continuity were 0.24 and 0.23, respectively.

The linear regression models combining all respondents were significant in predicting provider communication (P=.001) and patient influence (P=.001). Continuity of care, age, number of visits, general health, health improvement, and life satisfaction significantly contributed to the models (P≤.01 for each). Asthma status and the interaction terms between that status and the other independent variables were not significant, with the exception of the interaction term between that status and number of visits, which predicted provider communication (P=.03). With the asthma interaction terms largely nonsignificant, subsequent discussion will address only the separate asthmatic and nonasthmatic models.

Discussion

Our results confirm earlier findings that continuity of care is important in health care delivery. For both respondents with and without asthma, continuity of care with an individual provider significantly predicted their ratings of provider communication and patient influence alone and in the presence of control variables. Also, the results suggest that continuity of care may be particularly important in certain populations. Differences in the regression models for the respondents with and without asthma suggest a particularly important role of continuity of care in the physician-patient relationship for patients with this disease. Among those persons, continuity of care was the only variable predicting patient perceptions of physician-patient communication after controlling for many other relevant variables; it was 1 of only 2 variables predicting perceptions of patient influence. Among persons who did not have asthma, continuity shared its importance with several other variables.

Our results do not suggest that continuity is important only to patients with asthma. For both patients with and without asthma, continuity of care was an important predictor of provider communication and patient influence. However, it is the unique prominence of continuity of care in the asthma models that is interesting, in the context of several likely predictive variables that were nonsignificant.

If a sample did not have sufficient size and power to detect significant effects, such differences could have been an artifact of differences in sample size. However, the results of a power analysis show that in the group of patients with asthma (the smaller sample) there was 80% power to detect with 95% confidence a correlation as small as 0.06, which is a miniscule effect. Thus the sample with asthma offered sufficient power to detect the effects of all the independent variables, but only the effect of continuity of care emerged as significant for that group.

What is special about the physician-patient relationship over time for patients with asthma? Our findings suggest that for these patients, understanding of their disease and treatment and a feeling of comanagement with the physician are crucial and seem to be directly related to continuity of care. The patterns we saw in the patients with asthma are consistent with previous work that suggests the importance of continuity of care to patients with chronic illnesses. These patients have reported valuing continuity more than do patients with acute problems,²² and persons with asthma have reported willingness to wait more days for care from their regular physician for moderately serious acute illnesses.¹⁴ Studies from the 1980s found that patients with chronic illnesses maintained greater continuity with individual physicians.^23,24 The respondents with asthma in this 1997 survey did not report greater continuity than those who did not have it. This makes the prominent role of continuity in their evaluations of the physician-patient relationship more interesting, because differences in the level of continuity do not explain the importance of continuity. The respondents with asthma did report significantly more health care visits than those without asthma. Thus, they maintained high levels of continuity across a more challenging number of visits.

When patients concentrate their care with a single physician, those physicians are more likely to develop an accumulated knowledge about their patients’ medical conditions. This knowledge goes beyond simply knowing the patient’s diagnoses and medications. It includes a finer understanding of the severity of each medical problem and how multiple problems interact. More important, it includes the development of a relationship between the patient and the physician and awareness of the patient’s knowledge of the disease and personal preferences for medical treatment.

It has been argued that the importance of continuity of care cannot be conceptualized simply as the frequency of seeing one physician versus another.²⁵ However, personal continuity suggests an ongoing therapeutic relationship between the patient and the physician. In this case, the nature and quality of the contacts are more important than the number. The current data suggest that perhaps the importance of continuity to the ongoing therapeutic relationship is heightened for patients with asthma. It may be that the immediately perceptible morbidity of an asthma exacerbation and the relief provided by the physician contribute to the patient’s evaluation of the relationship. Positive qualities of physician-patient interaction have been linked to satisfaction for patients with ongoing medical problems²⁶ and to “better health” in chronically ill patients.¹⁵ Policies and practices that encourage continuity and an effective therapeutic relationship between the patient and the physician should be investigated and implemented. In the current environment of proposals for disease management treatment teams for diseases such as asthma, it is important that continuity of care between the patient and the physician is not completely eliminated through the use of multiple providers.

Limitations

Our study has several limitations. First, the data were based on self-reports. We did not independently validate either the diagnosis of asthma or reported utilization of care. However, chronic diseases have been successfully assessed through self-reports in a variety of large-scale surveys, such as the National Health Interview Survey. Moreover, the utilization questions are based on reliable and valid items from the CAHPS. Continuity of care was also assessed using a single item asking respondents about the level of continuity they experienced with an individual physician in the past year. The merits of this item as an assessment of perceived continuity include its distinct reference to continuity rather than asking whether the respondent has a regular or usual physician. Single-item reports of having a regular or usual physician have been interpreted as continuity of care.^21,27,28 However, having a regular or usual physician is not the same as maintaining continuity of care with an individual provider over time. For example, a patient who reports having a regular physician may see other physicians in that practice for a majority of visits because that regular provider is frequently unavailable when the patient needs acute care. A second limitation of our study is that the data came from a survey of participants in a Medicaid program, thereby potentially affecting the generalizability of the results to a more affluent population. In the Kentucky Medicaid fee-for-service program, recipients may have greater choice of physicians than participants in more restrictive health care plans.

Conclusions

The results of our study of an existing data set suggest several directions for future research. One direction would be to look more specifically at the patient-physician relationship and its impact on outcomes. Patients’ trust in their physicians may be particularly important in understanding continuity of care for chronically ill patients.²⁹ Further research into the mechanisms linking continuity of care and characteristics of the physician-patient relationship could begin to discern the direction of influence between them and their combined impact on health outcomes.

Continuity of care matters. Particularly for patients with asthma, continuity of care with an individual provider is linked to important aspects of health care delivery, specifically physician-patient interaction. Changes in health care systems that increasingly promote discontinuity with individual physicians may be especially disruptive for patients with chronic conditions.

Acknowledgments

Our study was funded in part by the Kentucky Department of Medicaid Services.

Continuity of care has been shown to be associated with a variety of positive outcomes including patient satisfaction,¹ compliance with medication regimens,² and health services utilization.³ High continuity of physician care is also associated with a decreased likelihood of future hospitalization.^4,5 One explanation for this finding is that continuity leads to increased knowledge and trust between the patient and the physician.⁶ This increased knowledge and trust may make it easier for the physician to manage medical problems in the office or over the phone and thereby avoid hospitalizing the patient. Similarly, having care continuity with a specific physician is significantly associated with a decreased likelihood of emergency department (ED) use.⁷

Although data are continually accumulating indicating that continuity between a patient and a clinician has positive benefits, it is clear that it is not always easy to achieve high continuity in practice.⁸ This is increasingly the case as physicians work in larger groups where there is likely to be less continuity with an individual provider.⁹ According to data from the 1987 National Medical Expenditure Survey, only half of all patients have high continuity with a physician.⁸ This situation may be worsening as more Americans are enrolled in managed health care plans that frequently change their panel of approved physicians.¹⁰ As a result, patients may be forced to change their physician on multiple occasions, which could have negative consequences on the management of their medical problems.

Since continuity between a patient and a physician has positive health benefits in the general population, it makes intuitive sense that these benefits would be greatest for persons with chronic conditions. Asthma is a common chronic disease affecting 14 to 15 million people in the United States and is the most common chronic disease of childhood, affecting nearly 5 million children in the United States.¹¹ Asthma accounts for more than 470,000 hospitalizations and more than 5000 deaths annually.¹² Rates of hospitalization for asthma have been increasing and reflect that it is the most common discharge diagnosis among children.¹³ In addition to the morbidity and mortality associated with this disease, some treatments have possible adverse effects.¹⁴ Thus, the benefits that could accrue from improved health care delivery are considerable.

Patients with asthma have a greater desire than the general population to maintain continuity with a physician, even when the visit is not for asthma.¹⁵ A possible reason continuity of care may be important is that the management of chronic conditions requires ongoing monitoring and decisions about when changes in therapy are appropriate. When a patient is seen by the same physician, that provider is more likely to know when tests and treatment changes are indicated. Effective physician-patient interaction is an important component of health care delivery. For patients with chronic illnesses, interaction processes that include the physician giving more information and the patient having greater control during the visit are associated with better health.¹⁵ Physician and patient roles in managing asthma differ across physician-patient relationships,¹⁶ but the relation between continuity of care and the physician-patient relationship has not been specified for patients with asthma.

It is important to determine the role and importance of continuity in the delivery of health care to patients with asthma. The purpose of our study was to examine the relationship between continuity of care and physician-patient interaction among patients with and without asthma.

Methods

The data for our study came from an omnibus survey of the patient satisfaction of recipients of the Kentucky Medicaid program. Our study is a secondary analysis of that large data set. In 1997 a survey was mailed to a stratified random sample of adult participants (aged Ž18 years) in the Kentucky Medicaid fee-for-service program. The design followed the Dillman method, using an initial wave of surveys with reminder postcards and 2 additional waves of surveys to nonresponders.¹⁷ The response rate to the survey was 60%, with a total sample of 2308.

The survey items were based on the Health Employer Data and Information Set (HEDIS 3.0) customer satisfaction survey,¹⁸ the Consumer Assessments of Health Plans Study (CAHPS)¹⁹ completed by the Agency for Health Care Policy and Research (AHCPR), and satisfaction surveys used in Kentucky during previous assessments in 1987, 1989, and 1991 (KENPAC program). The internal review board of the University of Kentucky approved the survey.

Our sample was limited to individuals who visited primary care physicians often enough to assess continuity. They reported utilization on a single survey item assessing outpatient visits. A total of 1726 respondents reported making 2 or more visits to a physician’s office, clinic, or ED during the previous 12 months. In this group 404 reported having asthma. The prevalence of asthma in this population is higher than in the general population but not exceptionally higher than that found in Medicaid programs.²⁰ Continuity was measured by the question “Over the past 12 months, when you went for medical care, how often did you see the same doctor or provider?” The 4 response categories were “always,” “most of the time,” “sometimes,” and “rarely or never” (reverse scored from 1=”rarely or never” to 4=”always”). Although continuity can be measured in a variety of ways, patient self-reports have commonly been used.^1,21 In the present context, perceptions of continuity may be no less important than actual continuity in predicting patients’ evaluative ratings of physician-patient interaction.

The outcome measures were patient assessments of the health care they had received in Medicaid programs during the past 12 months. They are consistent with other self-report measures and were created for the CAHPS by the AHCPR (now the Agency for Healthcare Research and Quality). The present survey reference to 12 months differed from the original CAHPS survey reference to 6 months. The measures included an item about provider communication (“Doctor or provider listened to you and talked with you about your care”) and an item about patient influence (“Your ability to influence the treatment you received from a doctor or provider for your health problems”). These were measured on 5-point scales (reverse scored from 1=poor to 5=excellent).

Analysis

We computed bivariate analyses comparing characteristics of the groups of patients with and without asthma and assessing the relation between continuity of care and physician-patient interaction for each group (chi-square, Student t test). Then the relation between continuity of care and physician-patient interaction was evaluated in multivariate linear regression analyses in the presence of the following control variables: age, sex, education, race, number of visits, general health, health improvement, and life satisfaction. Among the variables available for analysis, these were identified as most likely to confound the relation between continuity of care and patient perceptions of the physician-patient relationship. General health (“In general, would you say your health is:”), health improvement (“Compared to one year ago, how would you rate your health in general now?”), and life satisfaction (“Overall, how satisfied or dissatisfied are you with how your life is going?”) were rated using 5-point scales coded so that higher scores mean better health, more improved health, and greater overall life satisfaction, respectively. We performed separate linear regression models for the patients with and without asthma and examined the contribution of continuity to each model. Linear regression models with all respondents combined were also performed, and interaction terms were entered for asthma status interacting with the other independent variables. Only respondents who had complete data on all items could be included in the regression analyses. We conducted all analyses with SAS statistical programming software release 6.09 (SAS Institute, Inc; Cary, NC), using complete data for each item.

Results

The characteristics of the respondents appear in Table 1. Their demographic characteristics were typical for the Medicaid population in Kentucky. Although most of the respondents were white, more of the patients with asthma were white than were those without the condition. In addition, the asthma patients reported higher numbers of health care visits and poorer health than those without asthma. Reported continuity of care and respondents’ perceptions of provider communication and patient influence are shown in Table 2. The respondents with and without asthma did not differ on these variables of interest. More than half of the respondents (58.8%) reported always seeing the same health care provider in the past 12 months (scale mean ± standard deviation=3.5 ± 0.7).

Bivariate Analyses

In bivariate analyses, ratings of the physician-patient relationship were compared across continuity of care categories. Individuals who “always” saw the same doctor or provider were compared with a category of “less than always,” which had been collapsed across “most of the time,” “sometimes,” and “rarely or never.” On average, respondents rated provider communication and patient influence between “good” and “very good.” Both persons who did and did not have asthma who saw the same doctor or provider for all their health care visits rated provider communication and patient influence significantly higher than did individuals who had less continuity (P <.01). This is shown in Table 3.

Regression Analyses

We computed separate linear regression models for individuals who did and did not have asthma to assess the contribution of continuity of care in the presence of control variables in predicting ratings of provider communication and patient influence for these groups. All 4 response levels were included in the continuity of care variable. The correlation matrix for the independent variables produced no correlations between independent variables that exceeded 0.5, suggesting that these variables could be included in the same analysis. The P values and standardized regression coefficients for the independent variables in the asthmatic and nonasthmatic models are presented for the provider communication models in Table 4 and for the patient influence models in Table 5. All models were significant at P <.05.

For persons with asthma, continuity of care was the only variable (P=.01) that significantly contributed to the provider communication model (Table 4, Model 1) and the only variable (P=.02) other than life satisfaction (P=.04) that contributed to the patient influence model (Table 5, Model 1). In the provider communication and patient influence models, the standardized estimates for the continuity parameter were 0.15 and 0.14, respectively, higher than any other estimates in the models. The nonstandardized parameter estimates for continuity were 0.26 and 0.25, respectively.

For persons who did not have asthma, continuity of care significantly contributed (P=.001) to both the provider communication (Table 4, Model 2) and patient influence models (Table 5, Model 2). Unlike the models for persons with asthma, 5 additional variables significantly contributed to these models (P≤.01): age, number of visits, general health, health improvement, and life satisfaction. The standardized parameter estimates for continuity were similar to those in the asthmatic models (0.14 for each). Continuity ranked only third among the estimates in the provider communication model and second among the estimates in the patient influence model. The nonstandardized parameter estimates for continuity were 0.24 and 0.23, respectively.

The linear regression models combining all respondents were significant in predicting provider communication (P=.001) and patient influence (P=.001). Continuity of care, age, number of visits, general health, health improvement, and life satisfaction significantly contributed to the models (P≤.01 for each). Asthma status and the interaction terms between that status and the other independent variables were not significant, with the exception of the interaction term between that status and number of visits, which predicted provider communication (P=.03). With the asthma interaction terms largely nonsignificant, subsequent discussion will address only the separate asthmatic and nonasthmatic models.

Discussion

Our results confirm earlier findings that continuity of care is important in health care delivery. For both respondents with and without asthma, continuity of care with an individual provider significantly predicted their ratings of provider communication and patient influence alone and in the presence of control variables. Also, the results suggest that continuity of care may be particularly important in certain populations. Differences in the regression models for the respondents with and without asthma suggest a particularly important role of continuity of care in the physician-patient relationship for patients with this disease. Among those persons, continuity of care was the only variable predicting patient perceptions of physician-patient communication after controlling for many other relevant variables; it was 1 of only 2 variables predicting perceptions of patient influence. Among persons who did not have asthma, continuity shared its importance with several other variables.

Our results do not suggest that continuity is important only to patients with asthma. For both patients with and without asthma, continuity of care was an important predictor of provider communication and patient influence. However, it is the unique prominence of continuity of care in the asthma models that is interesting, in the context of several likely predictive variables that were nonsignificant.

If a sample did not have sufficient size and power to detect significant effects, such differences could have been an artifact of differences in sample size. However, the results of a power analysis show that in the group of patients with asthma (the smaller sample) there was 80% power to detect with 95% confidence a correlation as small as 0.06, which is a miniscule effect. Thus the sample with asthma offered sufficient power to detect the effects of all the independent variables, but only the effect of continuity of care emerged as significant for that group.

What is special about the physician-patient relationship over time for patients with asthma? Our findings suggest that for these patients, understanding of their disease and treatment and a feeling of comanagement with the physician are crucial and seem to be directly related to continuity of care. The patterns we saw in the patients with asthma are consistent with previous work that suggests the importance of continuity of care to patients with chronic illnesses. These patients have reported valuing continuity more than do patients with acute problems,²² and persons with asthma have reported willingness to wait more days for care from their regular physician for moderately serious acute illnesses.¹⁴ Studies from the 1980s found that patients with chronic illnesses maintained greater continuity with individual physicians.^23,24 The respondents with asthma in this 1997 survey did not report greater continuity than those who did not have it. This makes the prominent role of continuity in their evaluations of the physician-patient relationship more interesting, because differences in the level of continuity do not explain the importance of continuity. The respondents with asthma did report significantly more health care visits than those without asthma. Thus, they maintained high levels of continuity across a more challenging number of visits.

When patients concentrate their care with a single physician, those physicians are more likely to develop an accumulated knowledge about their patients’ medical conditions. This knowledge goes beyond simply knowing the patient’s diagnoses and medications. It includes a finer understanding of the severity of each medical problem and how multiple problems interact. More important, it includes the development of a relationship between the patient and the physician and awareness of the patient’s knowledge of the disease and personal preferences for medical treatment.

It has been argued that the importance of continuity of care cannot be conceptualized simply as the frequency of seeing one physician versus another.²⁵ However, personal continuity suggests an ongoing therapeutic relationship between the patient and the physician. In this case, the nature and quality of the contacts are more important than the number. The current data suggest that perhaps the importance of continuity to the ongoing therapeutic relationship is heightened for patients with asthma. It may be that the immediately perceptible morbidity of an asthma exacerbation and the relief provided by the physician contribute to the patient’s evaluation of the relationship. Positive qualities of physician-patient interaction have been linked to satisfaction for patients with ongoing medical problems²⁶ and to “better health” in chronically ill patients.¹⁵ Policies and practices that encourage continuity and an effective therapeutic relationship between the patient and the physician should be investigated and implemented. In the current environment of proposals for disease management treatment teams for diseases such as asthma, it is important that continuity of care between the patient and the physician is not completely eliminated through the use of multiple providers.

Limitations

Our study has several limitations. First, the data were based on self-reports. We did not independently validate either the diagnosis of asthma or reported utilization of care. However, chronic diseases have been successfully assessed through self-reports in a variety of large-scale surveys, such as the National Health Interview Survey. Moreover, the utilization questions are based on reliable and valid items from the CAHPS. Continuity of care was also assessed using a single item asking respondents about the level of continuity they experienced with an individual physician in the past year. The merits of this item as an assessment of perceived continuity include its distinct reference to continuity rather than asking whether the respondent has a regular or usual physician. Single-item reports of having a regular or usual physician have been interpreted as continuity of care.^21,27,28 However, having a regular or usual physician is not the same as maintaining continuity of care with an individual provider over time. For example, a patient who reports having a regular physician may see other physicians in that practice for a majority of visits because that regular provider is frequently unavailable when the patient needs acute care. A second limitation of our study is that the data came from a survey of participants in a Medicaid program, thereby potentially affecting the generalizability of the results to a more affluent population. In the Kentucky Medicaid fee-for-service program, recipients may have greater choice of physicians than participants in more restrictive health care plans.

Conclusions

The results of our study of an existing data set suggest several directions for future research. One direction would be to look more specifically at the patient-physician relationship and its impact on outcomes. Patients’ trust in their physicians may be particularly important in understanding continuity of care for chronically ill patients.²⁹ Further research into the mechanisms linking continuity of care and characteristics of the physician-patient relationship could begin to discern the direction of influence between them and their combined impact on health outcomes.

Continuity of care matters. Particularly for patients with asthma, continuity of care with an individual provider is linked to important aspects of health care delivery, specifically physician-patient interaction. Changes in health care systems that increasingly promote discontinuity with individual physicians may be especially disruptive for patients with chronic conditions.

Acknowledgments

Our study was funded in part by the Kentucky Department of Medicaid Services.

Continuity of care has been shown to be associated with a variety of positive outcomes including patient satisfaction,¹ compliance with medication regimens,² and health services utilization.³ High continuity of physician care is also associated with a decreased likelihood of future hospitalization.^4,5 One explanation for this finding is that continuity leads to increased knowledge and trust between the patient and the physician.⁶ This increased knowledge and trust may make it easier for the physician to manage medical problems in the office or over the phone and thereby avoid hospitalizing the patient. Similarly, having care continuity with a specific physician is significantly associated with a decreased likelihood of emergency department (ED) use.⁷

Although data are continually accumulating indicating that continuity between a patient and a clinician has positive benefits, it is clear that it is not always easy to achieve high continuity in practice.⁸ This is increasingly the case as physicians work in larger groups where there is likely to be less continuity with an individual provider.⁹ According to data from the 1987 National Medical Expenditure Survey, only half of all patients have high continuity with a physician.⁸ This situation may be worsening as more Americans are enrolled in managed health care plans that frequently change their panel of approved physicians.¹⁰ As a result, patients may be forced to change their physician on multiple occasions, which could have negative consequences on the management of their medical problems.

Since continuity between a patient and a physician has positive health benefits in the general population, it makes intuitive sense that these benefits would be greatest for persons with chronic conditions. Asthma is a common chronic disease affecting 14 to 15 million people in the United States and is the most common chronic disease of childhood, affecting nearly 5 million children in the United States.¹¹ Asthma accounts for more than 470,000 hospitalizations and more than 5000 deaths annually.¹² Rates of hospitalization for asthma have been increasing and reflect that it is the most common discharge diagnosis among children.¹³ In addition to the morbidity and mortality associated with this disease, some treatments have possible adverse effects.¹⁴ Thus, the benefits that could accrue from improved health care delivery are considerable.

Patients with asthma have a greater desire than the general population to maintain continuity with a physician, even when the visit is not for asthma.¹⁵ A possible reason continuity of care may be important is that the management of chronic conditions requires ongoing monitoring and decisions about when changes in therapy are appropriate. When a patient is seen by the same physician, that provider is more likely to know when tests and treatment changes are indicated. Effective physician-patient interaction is an important component of health care delivery. For patients with chronic illnesses, interaction processes that include the physician giving more information and the patient having greater control during the visit are associated with better health.¹⁵ Physician and patient roles in managing asthma differ across physician-patient relationships,¹⁶ but the relation between continuity of care and the physician-patient relationship has not been specified for patients with asthma.

It is important to determine the role and importance of continuity in the delivery of health care to patients with asthma. The purpose of our study was to examine the relationship between continuity of care and physician-patient interaction among patients with and without asthma.

Methods

The data for our study came from an omnibus survey of the patient satisfaction of recipients of the Kentucky Medicaid program. Our study is a secondary analysis of that large data set. In 1997 a survey was mailed to a stratified random sample of adult participants (aged Ž18 years) in the Kentucky Medicaid fee-for-service program. The design followed the Dillman method, using an initial wave of surveys with reminder postcards and 2 additional waves of surveys to nonresponders.¹⁷ The response rate to the survey was 60%, with a total sample of 2308.

The survey items were based on the Health Employer Data and Information Set (HEDIS 3.0) customer satisfaction survey,¹⁸ the Consumer Assessments of Health Plans Study (CAHPS)¹⁹ completed by the Agency for Health Care Policy and Research (AHCPR), and satisfaction surveys used in Kentucky during previous assessments in 1987, 1989, and 1991 (KENPAC program). The internal review board of the University of Kentucky approved the survey.

Our sample was limited to individuals who visited primary care physicians often enough to assess continuity. They reported utilization on a single survey item assessing outpatient visits. A total of 1726 respondents reported making 2 or more visits to a physician’s office, clinic, or ED during the previous 12 months. In this group 404 reported having asthma. The prevalence of asthma in this population is higher than in the general population but not exceptionally higher than that found in Medicaid programs.²⁰ Continuity was measured by the question “Over the past 12 months, when you went for medical care, how often did you see the same doctor or provider?” The 4 response categories were “always,” “most of the time,” “sometimes,” and “rarely or never” (reverse scored from 1=”rarely or never” to 4=”always”). Although continuity can be measured in a variety of ways, patient self-reports have commonly been used.^1,21 In the present context, perceptions of continuity may be no less important than actual continuity in predicting patients’ evaluative ratings of physician-patient interaction.

The outcome measures were patient assessments of the health care they had received in Medicaid programs during the past 12 months. They are consistent with other self-report measures and were created for the CAHPS by the AHCPR (now the Agency for Healthcare Research and Quality). The present survey reference to 12 months differed from the original CAHPS survey reference to 6 months. The measures included an item about provider communication (“Doctor or provider listened to you and talked with you about your care”) and an item about patient influence (“Your ability to influence the treatment you received from a doctor or provider for your health problems”). These were measured on 5-point scales (reverse scored from 1=poor to 5=excellent).

Analysis

We computed bivariate analyses comparing characteristics of the groups of patients with and without asthma and assessing the relation between continuity of care and physician-patient interaction for each group (chi-square, Student t test). Then the relation between continuity of care and physician-patient interaction was evaluated in multivariate linear regression analyses in the presence of the following control variables: age, sex, education, race, number of visits, general health, health improvement, and life satisfaction. Among the variables available for analysis, these were identified as most likely to confound the relation between continuity of care and patient perceptions of the physician-patient relationship. General health (“In general, would you say your health is:”), health improvement (“Compared to one year ago, how would you rate your health in general now?”), and life satisfaction (“Overall, how satisfied or dissatisfied are you with how your life is going?”) were rated using 5-point scales coded so that higher scores mean better health, more improved health, and greater overall life satisfaction, respectively. We performed separate linear regression models for the patients with and without asthma and examined the contribution of continuity to each model. Linear regression models with all respondents combined were also performed, and interaction terms were entered for asthma status interacting with the other independent variables. Only respondents who had complete data on all items could be included in the regression analyses. We conducted all analyses with SAS statistical programming software release 6.09 (SAS Institute, Inc; Cary, NC), using complete data for each item.

Results

The characteristics of the respondents appear in Table 1. Their demographic characteristics were typical for the Medicaid population in Kentucky. Although most of the respondents were white, more of the patients with asthma were white than were those without the condition. In addition, the asthma patients reported higher numbers of health care visits and poorer health than those without asthma. Reported continuity of care and respondents’ perceptions of provider communication and patient influence are shown in Table 2. The respondents with and without asthma did not differ on these variables of interest. More than half of the respondents (58.8%) reported always seeing the same health care provider in the past 12 months (scale mean ± standard deviation=3.5 ± 0.7).

Bivariate Analyses

In bivariate analyses, ratings of the physician-patient relationship were compared across continuity of care categories. Individuals who “always” saw the same doctor or provider were compared with a category of “less than always,” which had been collapsed across “most of the time,” “sometimes,” and “rarely or never.” On average, respondents rated provider communication and patient influence between “good” and “very good.” Both persons who did and did not have asthma who saw the same doctor or provider for all their health care visits rated provider communication and patient influence significantly higher than did individuals who had less continuity (P <.01). This is shown in Table 3.

Regression Analyses

We computed separate linear regression models for individuals who did and did not have asthma to assess the contribution of continuity of care in the presence of control variables in predicting ratings of provider communication and patient influence for these groups. All 4 response levels were included in the continuity of care variable. The correlation matrix for the independent variables produced no correlations between independent variables that exceeded 0.5, suggesting that these variables could be included in the same analysis. The P values and standardized regression coefficients for the independent variables in the asthmatic and nonasthmatic models are presented for the provider communication models in Table 4 and for the patient influence models in Table 5. All models were significant at P <.05.

For persons with asthma, continuity of care was the only variable (P=.01) that significantly contributed to the provider communication model (Table 4, Model 1) and the only variable (P=.02) other than life satisfaction (P=.04) that contributed to the patient influence model (Table 5, Model 1). In the provider communication and patient influence models, the standardized estimates for the continuity parameter were 0.15 and 0.14, respectively, higher than any other estimates in the models. The nonstandardized parameter estimates for continuity were 0.26 and 0.25, respectively.

For persons who did not have asthma, continuity of care significantly contributed (P=.001) to both the provider communication (Table 4, Model 2) and patient influence models (Table 5, Model 2). Unlike the models for persons with asthma, 5 additional variables significantly contributed to these models (P≤.01): age, number of visits, general health, health improvement, and life satisfaction. The standardized parameter estimates for continuity were similar to those in the asthmatic models (0.14 for each). Continuity ranked only third among the estimates in the provider communication model and second among the estimates in the patient influence model. The nonstandardized parameter estimates for continuity were 0.24 and 0.23, respectively.

The linear regression models combining all respondents were significant in predicting provider communication (P=.001) and patient influence (P=.001). Continuity of care, age, number of visits, general health, health improvement, and life satisfaction significantly contributed to the models (P≤.01 for each). Asthma status and the interaction terms between that status and the other independent variables were not significant, with the exception of the interaction term between that status and number of visits, which predicted provider communication (P=.03). With the asthma interaction terms largely nonsignificant, subsequent discussion will address only the separate asthmatic and nonasthmatic models.

Discussion

Our results confirm earlier findings that continuity of care is important in health care delivery. For both respondents with and without asthma, continuity of care with an individual provider significantly predicted their ratings of provider communication and patient influence alone and in the presence of control variables. Also, the results suggest that continuity of care may be particularly important in certain populations. Differences in the regression models for the respondents with and without asthma suggest a particularly important role of continuity of care in the physician-patient relationship for patients with this disease. Among those persons, continuity of care was the only variable predicting patient perceptions of physician-patient communication after controlling for many other relevant variables; it was 1 of only 2 variables predicting perceptions of patient influence. Among persons who did not have asthma, continuity shared its importance with several other variables.

Our results do not suggest that continuity is important only to patients with asthma. For both patients with and without asthma, continuity of care was an important predictor of provider communication and patient influence. However, it is the unique prominence of continuity of care in the asthma models that is interesting, in the context of several likely predictive variables that were nonsignificant.

If a sample did not have sufficient size and power to detect significant effects, such differences could have been an artifact of differences in sample size. However, the results of a power analysis show that in the group of patients with asthma (the smaller sample) there was 80% power to detect with 95% confidence a correlation as small as 0.06, which is a miniscule effect. Thus the sample with asthma offered sufficient power to detect the effects of all the independent variables, but only the effect of continuity of care emerged as significant for that group.

What is special about the physician-patient relationship over time for patients with asthma? Our findings suggest that for these patients, understanding of their disease and treatment and a feeling of comanagement with the physician are crucial and seem to be directly related to continuity of care. The patterns we saw in the patients with asthma are consistent with previous work that suggests the importance of continuity of care to patients with chronic illnesses. These patients have reported valuing continuity more than do patients with acute problems,²² and persons with asthma have reported willingness to wait more days for care from their regular physician for moderately serious acute illnesses.¹⁴ Studies from the 1980s found that patients with chronic illnesses maintained greater continuity with individual physicians.^23,24 The respondents with asthma in this 1997 survey did not report greater continuity than those who did not have it. This makes the prominent role of continuity in their evaluations of the physician-patient relationship more interesting, because differences in the level of continuity do not explain the importance of continuity. The respondents with asthma did report significantly more health care visits than those without asthma. Thus, they maintained high levels of continuity across a more challenging number of visits.

When patients concentrate their care with a single physician, those physicians are more likely to develop an accumulated knowledge about their patients’ medical conditions. This knowledge goes beyond simply knowing the patient’s diagnoses and medications. It includes a finer understanding of the severity of each medical problem and how multiple problems interact. More important, it includes the development of a relationship between the patient and the physician and awareness of the patient’s knowledge of the disease and personal preferences for medical treatment.

It has been argued that the importance of continuity of care cannot be conceptualized simply as the frequency of seeing one physician versus another.²⁵ However, personal continuity suggests an ongoing therapeutic relationship between the patient and the physician. In this case, the nature and quality of the contacts are more important than the number. The current data suggest that perhaps the importance of continuity to the ongoing therapeutic relationship is heightened for patients with asthma. It may be that the immediately perceptible morbidity of an asthma exacerbation and the relief provided by the physician contribute to the patient’s evaluation of the relationship. Positive qualities of physician-patient interaction have been linked to satisfaction for patients with ongoing medical problems²⁶ and to “better health” in chronically ill patients.¹⁵ Policies and practices that encourage continuity and an effective therapeutic relationship between the patient and the physician should be investigated and implemented. In the current environment of proposals for disease management treatment teams for diseases such as asthma, it is important that continuity of care between the patient and the physician is not completely eliminated through the use of multiple providers.

Limitations

Our study has several limitations. First, the data were based on self-reports. We did not independently validate either the diagnosis of asthma or reported utilization of care. However, chronic diseases have been successfully assessed through self-reports in a variety of large-scale surveys, such as the National Health Interview Survey. Moreover, the utilization questions are based on reliable and valid items from the CAHPS. Continuity of care was also assessed using a single item asking respondents about the level of continuity they experienced with an individual physician in the past year. The merits of this item as an assessment of perceived continuity include its distinct reference to continuity rather than asking whether the respondent has a regular or usual physician. Single-item reports of having a regular or usual physician have been interpreted as continuity of care.^21,27,28 However, having a regular or usual physician is not the same as maintaining continuity of care with an individual provider over time. For example, a patient who reports having a regular physician may see other physicians in that practice for a majority of visits because that regular provider is frequently unavailable when the patient needs acute care. A second limitation of our study is that the data came from a survey of participants in a Medicaid program, thereby potentially affecting the generalizability of the results to a more affluent population. In the Kentucky Medicaid fee-for-service program, recipients may have greater choice of physicians than participants in more restrictive health care plans.

Conclusions

The results of our study of an existing data set suggest several directions for future research. One direction would be to look more specifically at the patient-physician relationship and its impact on outcomes. Patients’ trust in their physicians may be particularly important in understanding continuity of care for chronically ill patients.²⁹ Further research into the mechanisms linking continuity of care and characteristics of the physician-patient relationship could begin to discern the direction of influence between them and their combined impact on health outcomes.

Continuity of care matters. Particularly for patients with asthma, continuity of care with an individual provider is linked to important aspects of health care delivery, specifically physician-patient interaction. Changes in health care systems that increasingly promote discontinuity with individual physicians may be especially disruptive for patients with chronic conditions.

Acknowledgments

Our study was funded in part by the Kentucky Department of Medicaid Services.

The birth of a first child is a time of major transition for a couple, marked by significant changes in the roles and responsibilities of both parents. The nature of their responsibilities change, and the volume of their work increases markedly. Comparative data from a national longitudinal study indicate that adults of childbearing age (25-44 years) invest more combined total work hours into the home and workplace than other adults, with workloads of 82 to 84 hours per week for employed women and 70 to 71 hours per week for employed men compared with less than 67 and 61 hours per week for older women and men, respectively.¹ The way partners respond to these workload issues can have an impact on marital happiness, as shown by several studies demonstrating an association between wives’ marital satisfaction and their husbands’ participation in household work.^2-7 The sharing of work responsibility is particularly important in the first few months after childbirth, with research demonstrating a significant association between the mother’s mental health and the degree of her partner’s emotional and practical support.⁸ However, many new mothers perceive a decline in their husbands’ participation in household chores and expressions of caring over the first postpartum year.⁹

Is this move to a more traditional division of labor intentional? Would it persist if couples were given the opportunity to learn about and plan for their postpartum responsibilities? Would a more equitable division of family and household work enhance the mental and physical health of women and men and increase their marital happiness in these early childbearing years? These questions form the basis of this research in postpartum health and family systems and are particularly relevant to family physicians who work within a broad definition of health that considers the context of community and family. Though it is common for family physicians to refer expectant parents to prenatal classes on childbirth and infant care, couples typically have few or no opportunities to formally prepare themselves for their postpartum lifestyle changes, including their new and increasing work responsibilities. Parenting classes address the needs of the child and the parent-child relationship¹⁰ while couples’ groups look at partners’ communication and conflict management,¹¹ but few structured opportunities exist for couples to plan for family and household work needs after their child is born.

There were 3 goals for this study: (1) describe the work patterns of a group of employed couples living together and expecting their first child; (2) provide a formal opportunity for couples to establish a plan for sharing postpartum work; and (3) evaluate gender differences in anticipated changes in workload from before to after childbirth. Of special interest was learning how couples intended to share postpartum work responsibilities (eg, equally vs unequally) when given a formal opportunity to actively plan for postpartum work distribution. This descriptive study represents the initial phase of a randomized controlled trial testing the impact of a prenatal work planning session (conducted in the context of childbirth education classes) on the partners’ postpartum work distribution, selected mental and physical health outcomes, and marital satisfaction.

Methods

English-speaking couples living together and expecting their first child were eligible for this study. Participants were recruited from prenatal classes offered through 2 St. Paul HealthEast hospitals from November 1998 through August 1999. Research assistants visited 30 of 34 HealthEast prenatal classes during the third class session to describe the study, enroll participants, and distribute prenatal surveys. Potential subjects were told that couples randomized to the intervention would attend 2 breakout sessions that would address emotional and practical support around the time of childbirth. Data for this study were derived from prenatal surveys and from worksheets indicating partners’ anticipated postpartum work time commitments.

The prenatal survey completed independently by all subjects during the third prenatal class session was used to gather demographic data (age, sex, education, race, marital status, and employment status); information on work responsibilities, including the number of hours per week devoted to employment and various household chores; perceptions of equity of household responsibilities among partners (measured on a 1 to 7 scale where 1=partner does everything, 4=we share equally, and 7=I do everything); and satisfaction with the partner’s contribution to household responsibilities (1=very dissatisfied and 7=very satisfied).

Approximately half of the enrolled couples were then randomized to the intervention, which consisted of 2 30-minute breakout sessions held during the fourth and fifth prenatal classes. In the first session partners were asked to tell each other what the other person did to make them feel loved and cared for. During the second breakout session each couple completed a worksheet together that asked them to list the amount of time each partner planned to spend at 6 months postpartum doing various tasks, caring for their child, and participating in paid employment. Suggested time estimates for the work tasks (in hours per week) were provided (Table 1) based on the results of a previous pilot study that asked first-time parents to estimate their actual workloads at 6 months postpartum. Parents were also asked to indicate any time contributed to these areas by outside sources (eg, babysitter, daycare provider, relative, housekeeper).

Student t tests were used to investigate gender differences in the amount of time invested in work prenatally, perceived degree of sharing household tasks with the partner, satisfaction with the partner’s contribution to household work, and projected prenatal to postpartum work changes. Paired t tests were used to examine subjects’ anticipated prenatal to postpartum changes in workload. Congruency in partners’ perceptions about how they currently share household responsibilities was determined by summing their responses to the question, “Please circle the number that best describes how you and your partner currently share household responsibilities.” Responses were given on a 1 to 7 scale where 1=“partner does everything,” 4=“We share equally,” and 7=“I do everything.” Thus, a summed score of 8 indicated perfect congruency.

Results

Of the 722 expectant parents informed of the study, 76 were ineligible to participate (usually because they were not living with a partner), 346 refused to participate (the most common reason for refusal was concern about leaving the large classroom for a breakout session), and 300 (149 men and 151 women) agreed to participate, for a response rate of 46% (300/646). The mean age of the participants was 29.3 years (standard deviation=4.6); 93.3% were white; 88.6% were married; 97.9% were employed; and 62% had a 4-year college or advanced degree. A total of 132 people (66 men and 66 women) participated in the breakout sessions.

The amount of time expectant fathers and mothers devoted to various household tasks and employment prenatally is shown in Table 1. Although men and women contributed similar amounts of time to household work—18.2 and 20.0 hours, respectively—they tended to divide these tasks according to traditional work patterns, with men investing more time in household repairs, lawn care, and snow removal, and women spending more time with cooking, cleaning, laundry, and shopping. Compared with women, men worked longer hours at their jobs, and this resulted in a heavier mean total prenatal workload for men by 8.4 hours per week (P=.000).

In response to the question of how they shared household responsibilities, women reported a belief that they contributed more to household chores than their partners (mean=4.4 and 3.8 for women and men, respectively; P=.000). This finding is consistent with women’s slightly higher estimated contributions to household tasks (P=NS). Partners’ perceptions about how they shared household work were congruent: 90% of couples had summed congruency scores in the 7 to 9 range, which allows for no more than a 1-point deviation from a perfect summed congruency score of 8. On average men were more satisfied than women with their partner’s contribution to household work (mean=6.0 and 5.4 for men and women, respectively; P=.000). The projected prenatal to postpartum changes in workload were considerable for both men and women (Table 2). Women predicted an 85% increase, and men anticipated a 53% expansion of total workload, with a net result of women planning to work 9 hours per week more than men at 6 months postpartum (P <.001). Both men and women predicted significant increases in time spent on household tasks, child care, and total work; the projected changes in effort related to child care and total work were significantly greater for women than men (P=.000), as shown in Table 3. Both men and women planned to reduce their paid work commitments after childbirth, women to a greater degree than men (P=.000).

Discussion

The results indicate that while men shouldered heavier workloads prenatally, women anticipated working longer hours than men at 6 months postpartum by 9 hours per week. This amounts to an 85% (48.7 hours/week) increase in workload for women, compared with a 53% (33.3 hours/week) increase for men. Although both are astounding increases, women clearly anticipated a larger expansion of work than men. Such dramatic changes in work responsibilities realized by new mothers might be at least partially responsible for the mental and physical problems that often plague women after childbirth.^8,12

The projected postpartum difference in workload between men and women was not unexpected, given the findings of Kahn¹ that on average, adult women of all ages in the United States bear heavier total workloads than men. It is noteworthy, however, that this gender difference in workloads was anticipated even by a group of couples who had had an opportunity to systematically study and preplan their postpartum work distribution.

It appears that to some degree these planned gender discrepancies in postpartum work responsibilities might be explained on the basis of traditional sex role assumptions. Women planned to take on more of the child care and household responsibilities after childbirth than men: 79 versus 52 hours per week. To help compensate for this considerable expansion of unpaid work, expectant mothers also planned to trim an average of 11.7 hours per week from their paid jobs compared with expectant fathers’ anticipated drop of 2.2 hours per week. For many women this change would likely result in part-time work. Previous studies have documented that both men and women tend to work part-time more in the childbearing years than at any other time in their adult lives, and women’s use of part-time work hours during this period of life tends to be much greater than men’s, often 2 to 3 times more.¹³ These data reinforce the need for couples to consider many complex issues in their postpartum workload planning, such as whether their dissimilar reductions in paid work will have a differential impact on their career satisfaction and opportunities and whether their joint plans for curtailing employment hours could ultimately benefit the family unit by improving child and family development.

Although expectant fathers and mothers in this study tended to follow traditional patterns in their qualitative division of various household responsibilities, they devoted similar amounts of total time to household tasks (18.2 and 20.0 hours/week, respectively). This finding contrasts with that of previous studies showing a much greater share of household work being performed by women,^1,14,15 often twice as much or more.^1,14,16 The results could be related to several factors. First, the observed prenatal work patterns may be somewhat atypical for these couples: Several women indicated that they had cut back on their housework or employment hours because of pregnancy-associated fatigue or other health problems. Second, this was a very homogenous population of employed young couples without children, in contrast to the more diverse samples (which included adults with children and more unemployed wives) used in many previous studies. Alternatively, these findings might represent a societal trend toward men and women sharing housework more equitably.

Importantly for many of these parents, the changes in work that they anticipated after giving birth likely represent the largest and most abrupt increase in work responsibilities that they will face in their adult lives. Unfortunately, it is a change for which many parents are ill-prepared. This lack of preparation is likely due, at least in part, to the paucity of information available on new parents’ actual workloads (no specific information was found in the medical, sociological, and psychological searches), society’s tendency to focus on the health needs of newborns and children more than those of their parents, and the absence of a consistent method for educating adolescents and young adults about the responsibilities of supporting and nurturing a family.

Each of these needs will be addressed. First, additional research is needed on changes in work responsibilities for new parents from more diverse populations, and we need a greater understanding of how these work responsibilities affect health and marital well-being (the goal of the ongoing randomized controlled trial). Second, we need a broader view of postpartum care, such that the physical, mental, and social needs of both the parents and newborn are considered in an ongoing manner. This is a perspective that family physicians are uniquely positioned to adopt and foster within the context of prenatal and postpartum care for the family unit. Moreover, family physicians could also be part of the solution to the third need, that of teaching would-be parents about postpartum work and family responsibilities. These efforts may pay important dividends in strengthening the fiber of the family and improving the well-being of its individual members.

Limitations

The limitations of this study include the modest response rate, the potential for selection bias, and the relatively homogeneous sample. In addition, parents’ estimations of work time may not be completely accurate, and these simple estimations do not account for such factors as the intensity of work at any given time (as when one juggles numerous responsibilities concurrently) or the issue of who holds the ultimate responsibility for a given task. Though future workload projections may be even more inaccurate than current estimations, the results shown here are more than individuals’ guesses about their future work; they represent couples’ intentional plans for sharing postpartum work responsibilities.

Conclusions

These expectant first-time parents anticipated considerable expansions in their work activities after childbirth, with women planning a greater share of the total postpartum workload. This information is important for new parents and for the health care providers who attend them as they resume their household, family, and paid work responsibilities after childbirth.

Acknowledgments

This study was funded by the University of Minnesota graduate school.

The author would like to thank Anne Marie Weber-Main for her editing assistance and Bruce Center for his help with data analysis.

The birth of a first child is a time of major transition for a couple, marked by significant changes in the roles and responsibilities of both parents. The nature of their responsibilities change, and the volume of their work increases markedly. Comparative data from a national longitudinal study indicate that adults of childbearing age (25-44 years) invest more combined total work hours into the home and workplace than other adults, with workloads of 82 to 84 hours per week for employed women and 70 to 71 hours per week for employed men compared with less than 67 and 61 hours per week for older women and men, respectively.¹ The way partners respond to these workload issues can have an impact on marital happiness, as shown by several studies demonstrating an association between wives’ marital satisfaction and their husbands’ participation in household work.^2-7 The sharing of work responsibility is particularly important in the first few months after childbirth, with research demonstrating a significant association between the mother’s mental health and the degree of her partner’s emotional and practical support.⁸ However, many new mothers perceive a decline in their husbands’ participation in household chores and expressions of caring over the first postpartum year.⁹

Is this move to a more traditional division of labor intentional? Would it persist if couples were given the opportunity to learn about and plan for their postpartum responsibilities? Would a more equitable division of family and household work enhance the mental and physical health of women and men and increase their marital happiness in these early childbearing years? These questions form the basis of this research in postpartum health and family systems and are particularly relevant to family physicians who work within a broad definition of health that considers the context of community and family. Though it is common for family physicians to refer expectant parents to prenatal classes on childbirth and infant care, couples typically have few or no opportunities to formally prepare themselves for their postpartum lifestyle changes, including their new and increasing work responsibilities. Parenting classes address the needs of the child and the parent-child relationship¹⁰ while couples’ groups look at partners’ communication and conflict management,¹¹ but few structured opportunities exist for couples to plan for family and household work needs after their child is born.

There were 3 goals for this study: (1) describe the work patterns of a group of employed couples living together and expecting their first child; (2) provide a formal opportunity for couples to establish a plan for sharing postpartum work; and (3) evaluate gender differences in anticipated changes in workload from before to after childbirth. Of special interest was learning how couples intended to share postpartum work responsibilities (eg, equally vs unequally) when given a formal opportunity to actively plan for postpartum work distribution. This descriptive study represents the initial phase of a randomized controlled trial testing the impact of a prenatal work planning session (conducted in the context of childbirth education classes) on the partners’ postpartum work distribution, selected mental and physical health outcomes, and marital satisfaction.

Methods

English-speaking couples living together and expecting their first child were eligible for this study. Participants were recruited from prenatal classes offered through 2 St. Paul HealthEast hospitals from November 1998 through August 1999. Research assistants visited 30 of 34 HealthEast prenatal classes during the third class session to describe the study, enroll participants, and distribute prenatal surveys. Potential subjects were told that couples randomized to the intervention would attend 2 breakout sessions that would address emotional and practical support around the time of childbirth. Data for this study were derived from prenatal surveys and from worksheets indicating partners’ anticipated postpartum work time commitments.

The prenatal survey completed independently by all subjects during the third prenatal class session was used to gather demographic data (age, sex, education, race, marital status, and employment status); information on work responsibilities, including the number of hours per week devoted to employment and various household chores; perceptions of equity of household responsibilities among partners (measured on a 1 to 7 scale where 1=partner does everything, 4=we share equally, and 7=I do everything); and satisfaction with the partner’s contribution to household responsibilities (1=very dissatisfied and 7=very satisfied).

Approximately half of the enrolled couples were then randomized to the intervention, which consisted of 2 30-minute breakout sessions held during the fourth and fifth prenatal classes. In the first session partners were asked to tell each other what the other person did to make them feel loved and cared for. During the second breakout session each couple completed a worksheet together that asked them to list the amount of time each partner planned to spend at 6 months postpartum doing various tasks, caring for their child, and participating in paid employment. Suggested time estimates for the work tasks (in hours per week) were provided (Table 1) based on the results of a previous pilot study that asked first-time parents to estimate their actual workloads at 6 months postpartum. Parents were also asked to indicate any time contributed to these areas by outside sources (eg, babysitter, daycare provider, relative, housekeeper).

Student t tests were used to investigate gender differences in the amount of time invested in work prenatally, perceived degree of sharing household tasks with the partner, satisfaction with the partner’s contribution to household work, and projected prenatal to postpartum work changes. Paired t tests were used to examine subjects’ anticipated prenatal to postpartum changes in workload. Congruency in partners’ perceptions about how they currently share household responsibilities was determined by summing their responses to the question, “Please circle the number that best describes how you and your partner currently share household responsibilities.” Responses were given on a 1 to 7 scale where 1=“partner does everything,” 4=“We share equally,” and 7=“I do everything.” Thus, a summed score of 8 indicated perfect congruency.

Results

Of the 722 expectant parents informed of the study, 76 were ineligible to participate (usually because they were not living with a partner), 346 refused to participate (the most common reason for refusal was concern about leaving the large classroom for a breakout session), and 300 (149 men and 151 women) agreed to participate, for a response rate of 46% (300/646). The mean age of the participants was 29.3 years (standard deviation=4.6); 93.3% were white; 88.6% were married; 97.9% were employed; and 62% had a 4-year college or advanced degree. A total of 132 people (66 men and 66 women) participated in the breakout sessions.

The amount of time expectant fathers and mothers devoted to various household tasks and employment prenatally is shown in Table 1. Although men and women contributed similar amounts of time to household work—18.2 and 20.0 hours, respectively—they tended to divide these tasks according to traditional work patterns, with men investing more time in household repairs, lawn care, and snow removal, and women spending more time with cooking, cleaning, laundry, and shopping. Compared with women, men worked longer hours at their jobs, and this resulted in a heavier mean total prenatal workload for men by 8.4 hours per week (P=.000).

In response to the question of how they shared household responsibilities, women reported a belief that they contributed more to household chores than their partners (mean=4.4 and 3.8 for women and men, respectively; P=.000). This finding is consistent with women’s slightly higher estimated contributions to household tasks (P=NS). Partners’ perceptions about how they shared household work were congruent: 90% of couples had summed congruency scores in the 7 to 9 range, which allows for no more than a 1-point deviation from a perfect summed congruency score of 8. On average men were more satisfied than women with their partner’s contribution to household work (mean=6.0 and 5.4 for men and women, respectively; P=.000). The projected prenatal to postpartum changes in workload were considerable for both men and women (Table 2). Women predicted an 85% increase, and men anticipated a 53% expansion of total workload, with a net result of women planning to work 9 hours per week more than men at 6 months postpartum (P <.001). Both men and women predicted significant increases in time spent on household tasks, child care, and total work; the projected changes in effort related to child care and total work were significantly greater for women than men (P=.000), as shown in Table 3. Both men and women planned to reduce their paid work commitments after childbirth, women to a greater degree than men (P=.000).

Discussion

The results indicate that while men shouldered heavier workloads prenatally, women anticipated working longer hours than men at 6 months postpartum by 9 hours per week. This amounts to an 85% (48.7 hours/week) increase in workload for women, compared with a 53% (33.3 hours/week) increase for men. Although both are astounding increases, women clearly anticipated a larger expansion of work than men. Such dramatic changes in work responsibilities realized by new mothers might be at least partially responsible for the mental and physical problems that often plague women after childbirth.^8,12

The projected postpartum difference in workload between men and women was not unexpected, given the findings of Kahn¹ that on average, adult women of all ages in the United States bear heavier total workloads than men. It is noteworthy, however, that this gender difference in workloads was anticipated even by a group of couples who had had an opportunity to systematically study and preplan their postpartum work distribution.

It appears that to some degree these planned gender discrepancies in postpartum work responsibilities might be explained on the basis of traditional sex role assumptions. Women planned to take on more of the child care and household responsibilities after childbirth than men: 79 versus 52 hours per week. To help compensate for this considerable expansion of unpaid work, expectant mothers also planned to trim an average of 11.7 hours per week from their paid jobs compared with expectant fathers’ anticipated drop of 2.2 hours per week. For many women this change would likely result in part-time work. Previous studies have documented that both men and women tend to work part-time more in the childbearing years than at any other time in their adult lives, and women’s use of part-time work hours during this period of life tends to be much greater than men’s, often 2 to 3 times more.¹³ These data reinforce the need for couples to consider many complex issues in their postpartum workload planning, such as whether their dissimilar reductions in paid work will have a differential impact on their career satisfaction and opportunities and whether their joint plans for curtailing employment hours could ultimately benefit the family unit by improving child and family development.

Although expectant fathers and mothers in this study tended to follow traditional patterns in their qualitative division of various household responsibilities, they devoted similar amounts of total time to household tasks (18.2 and 20.0 hours/week, respectively). This finding contrasts with that of previous studies showing a much greater share of household work being performed by women,^1,14,15 often twice as much or more.^1,14,16 The results could be related to several factors. First, the observed prenatal work patterns may be somewhat atypical for these couples: Several women indicated that they had cut back on their housework or employment hours because of pregnancy-associated fatigue or other health problems. Second, this was a very homogenous population of employed young couples without children, in contrast to the more diverse samples (which included adults with children and more unemployed wives) used in many previous studies. Alternatively, these findings might represent a societal trend toward men and women sharing housework more equitably.

Importantly for many of these parents, the changes in work that they anticipated after giving birth likely represent the largest and most abrupt increase in work responsibilities that they will face in their adult lives. Unfortunately, it is a change for which many parents are ill-prepared. This lack of preparation is likely due, at least in part, to the paucity of information available on new parents’ actual workloads (no specific information was found in the medical, sociological, and psychological searches), society’s tendency to focus on the health needs of newborns and children more than those of their parents, and the absence of a consistent method for educating adolescents and young adults about the responsibilities of supporting and nurturing a family.

Each of these needs will be addressed. First, additional research is needed on changes in work responsibilities for new parents from more diverse populations, and we need a greater understanding of how these work responsibilities affect health and marital well-being (the goal of the ongoing randomized controlled trial). Second, we need a broader view of postpartum care, such that the physical, mental, and social needs of both the parents and newborn are considered in an ongoing manner. This is a perspective that family physicians are uniquely positioned to adopt and foster within the context of prenatal and postpartum care for the family unit. Moreover, family physicians could also be part of the solution to the third need, that of teaching would-be parents about postpartum work and family responsibilities. These efforts may pay important dividends in strengthening the fiber of the family and improving the well-being of its individual members.

Limitations

The limitations of this study include the modest response rate, the potential for selection bias, and the relatively homogeneous sample. In addition, parents’ estimations of work time may not be completely accurate, and these simple estimations do not account for such factors as the intensity of work at any given time (as when one juggles numerous responsibilities concurrently) or the issue of who holds the ultimate responsibility for a given task. Though future workload projections may be even more inaccurate than current estimations, the results shown here are more than individuals’ guesses about their future work; they represent couples’ intentional plans for sharing postpartum work responsibilities.

Conclusions

These expectant first-time parents anticipated considerable expansions in their work activities after childbirth, with women planning a greater share of the total postpartum workload. This information is important for new parents and for the health care providers who attend them as they resume their household, family, and paid work responsibilities after childbirth.

Acknowledgments

This study was funded by the University of Minnesota graduate school.

The author would like to thank Anne Marie Weber-Main for her editing assistance and Bruce Center for his help with data analysis.

The birth of a first child is a time of major transition for a couple, marked by significant changes in the roles and responsibilities of both parents. The nature of their responsibilities change, and the volume of their work increases markedly. Comparative data from a national longitudinal study indicate that adults of childbearing age (25-44 years) invest more combined total work hours into the home and workplace than other adults, with workloads of 82 to 84 hours per week for employed women and 70 to 71 hours per week for employed men compared with less than 67 and 61 hours per week for older women and men, respectively.¹ The way partners respond to these workload issues can have an impact on marital happiness, as shown by several studies demonstrating an association between wives’ marital satisfaction and their husbands’ participation in household work.^2-7 The sharing of work responsibility is particularly important in the first few months after childbirth, with research demonstrating a significant association between the mother’s mental health and the degree of her partner’s emotional and practical support.⁸ However, many new mothers perceive a decline in their husbands’ participation in household chores and expressions of caring over the first postpartum year.⁹

Is this move to a more traditional division of labor intentional? Would it persist if couples were given the opportunity to learn about and plan for their postpartum responsibilities? Would a more equitable division of family and household work enhance the mental and physical health of women and men and increase their marital happiness in these early childbearing years? These questions form the basis of this research in postpartum health and family systems and are particularly relevant to family physicians who work within a broad definition of health that considers the context of community and family. Though it is common for family physicians to refer expectant parents to prenatal classes on childbirth and infant care, couples typically have few or no opportunities to formally prepare themselves for their postpartum lifestyle changes, including their new and increasing work responsibilities. Parenting classes address the needs of the child and the parent-child relationship¹⁰ while couples’ groups look at partners’ communication and conflict management,¹¹ but few structured opportunities exist for couples to plan for family and household work needs after their child is born.

There were 3 goals for this study: (1) describe the work patterns of a group of employed couples living together and expecting their first child; (2) provide a formal opportunity for couples to establish a plan for sharing postpartum work; and (3) evaluate gender differences in anticipated changes in workload from before to after childbirth. Of special interest was learning how couples intended to share postpartum work responsibilities (eg, equally vs unequally) when given a formal opportunity to actively plan for postpartum work distribution. This descriptive study represents the initial phase of a randomized controlled trial testing the impact of a prenatal work planning session (conducted in the context of childbirth education classes) on the partners’ postpartum work distribution, selected mental and physical health outcomes, and marital satisfaction.

Methods

English-speaking couples living together and expecting their first child were eligible for this study. Participants were recruited from prenatal classes offered through 2 St. Paul HealthEast hospitals from November 1998 through August 1999. Research assistants visited 30 of 34 HealthEast prenatal classes during the third class session to describe the study, enroll participants, and distribute prenatal surveys. Potential subjects were told that couples randomized to the intervention would attend 2 breakout sessions that would address emotional and practical support around the time of childbirth. Data for this study were derived from prenatal surveys and from worksheets indicating partners’ anticipated postpartum work time commitments.

The prenatal survey completed independently by all subjects during the third prenatal class session was used to gather demographic data (age, sex, education, race, marital status, and employment status); information on work responsibilities, including the number of hours per week devoted to employment and various household chores; perceptions of equity of household responsibilities among partners (measured on a 1 to 7 scale where 1=partner does everything, 4=we share equally, and 7=I do everything); and satisfaction with the partner’s contribution to household responsibilities (1=very dissatisfied and 7=very satisfied).

Approximately half of the enrolled couples were then randomized to the intervention, which consisted of 2 30-minute breakout sessions held during the fourth and fifth prenatal classes. In the first session partners were asked to tell each other what the other person did to make them feel loved and cared for. During the second breakout session each couple completed a worksheet together that asked them to list the amount of time each partner planned to spend at 6 months postpartum doing various tasks, caring for their child, and participating in paid employment. Suggested time estimates for the work tasks (in hours per week) were provided (Table 1) based on the results of a previous pilot study that asked first-time parents to estimate their actual workloads at 6 months postpartum. Parents were also asked to indicate any time contributed to these areas by outside sources (eg, babysitter, daycare provider, relative, housekeeper).

Student t tests were used to investigate gender differences in the amount of time invested in work prenatally, perceived degree of sharing household tasks with the partner, satisfaction with the partner’s contribution to household work, and projected prenatal to postpartum work changes. Paired t tests were used to examine subjects’ anticipated prenatal to postpartum changes in workload. Congruency in partners’ perceptions about how they currently share household responsibilities was determined by summing their responses to the question, “Please circle the number that best describes how you and your partner currently share household responsibilities.” Responses were given on a 1 to 7 scale where 1=“partner does everything,” 4=“We share equally,” and 7=“I do everything.” Thus, a summed score of 8 indicated perfect congruency.

Results

Of the 722 expectant parents informed of the study, 76 were ineligible to participate (usually because they were not living with a partner), 346 refused to participate (the most common reason for refusal was concern about leaving the large classroom for a breakout session), and 300 (149 men and 151 women) agreed to participate, for a response rate of 46% (300/646). The mean age of the participants was 29.3 years (standard deviation=4.6); 93.3% were white; 88.6% were married; 97.9% were employed; and 62% had a 4-year college or advanced degree. A total of 132 people (66 men and 66 women) participated in the breakout sessions.

The amount of time expectant fathers and mothers devoted to various household tasks and employment prenatally is shown in Table 1. Although men and women contributed similar amounts of time to household work—18.2 and 20.0 hours, respectively—they tended to divide these tasks according to traditional work patterns, with men investing more time in household repairs, lawn care, and snow removal, and women spending more time with cooking, cleaning, laundry, and shopping. Compared with women, men worked longer hours at their jobs, and this resulted in a heavier mean total prenatal workload for men by 8.4 hours per week (P=.000).

In response to the question of how they shared household responsibilities, women reported a belief that they contributed more to household chores than their partners (mean=4.4 and 3.8 for women and men, respectively; P=.000). This finding is consistent with women’s slightly higher estimated contributions to household tasks (P=NS). Partners’ perceptions about how they shared household work were congruent: 90% of couples had summed congruency scores in the 7 to 9 range, which allows for no more than a 1-point deviation from a perfect summed congruency score of 8. On average men were more satisfied than women with their partner’s contribution to household work (mean=6.0 and 5.4 for men and women, respectively; P=.000). The projected prenatal to postpartum changes in workload were considerable for both men and women (Table 2). Women predicted an 85% increase, and men anticipated a 53% expansion of total workload, with a net result of women planning to work 9 hours per week more than men at 6 months postpartum (P <.001). Both men and women predicted significant increases in time spent on household tasks, child care, and total work; the projected changes in effort related to child care and total work were significantly greater for women than men (P=.000), as shown in Table 3. Both men and women planned to reduce their paid work commitments after childbirth, women to a greater degree than men (P=.000).

Discussion

The results indicate that while men shouldered heavier workloads prenatally, women anticipated working longer hours than men at 6 months postpartum by 9 hours per week. This amounts to an 85% (48.7 hours/week) increase in workload for women, compared with a 53% (33.3 hours/week) increase for men. Although both are astounding increases, women clearly anticipated a larger expansion of work than men. Such dramatic changes in work responsibilities realized by new mothers might be at least partially responsible for the mental and physical problems that often plague women after childbirth.^8,12

The projected postpartum difference in workload between men and women was not unexpected, given the findings of Kahn¹ that on average, adult women of all ages in the United States bear heavier total workloads than men. It is noteworthy, however, that this gender difference in workloads was anticipated even by a group of couples who had had an opportunity to systematically study and preplan their postpartum work distribution.

It appears that to some degree these planned gender discrepancies in postpartum work responsibilities might be explained on the basis of traditional sex role assumptions. Women planned to take on more of the child care and household responsibilities after childbirth than men: 79 versus 52 hours per week. To help compensate for this considerable expansion of unpaid work, expectant mothers also planned to trim an average of 11.7 hours per week from their paid jobs compared with expectant fathers’ anticipated drop of 2.2 hours per week. For many women this change would likely result in part-time work. Previous studies have documented that both men and women tend to work part-time more in the childbearing years than at any other time in their adult lives, and women’s use of part-time work hours during this period of life tends to be much greater than men’s, often 2 to 3 times more.¹³ These data reinforce the need for couples to consider many complex issues in their postpartum workload planning, such as whether their dissimilar reductions in paid work will have a differential impact on their career satisfaction and opportunities and whether their joint plans for curtailing employment hours could ultimately benefit the family unit by improving child and family development.

Although expectant fathers and mothers in this study tended to follow traditional patterns in their qualitative division of various household responsibilities, they devoted similar amounts of total time to household tasks (18.2 and 20.0 hours/week, respectively). This finding contrasts with that of previous studies showing a much greater share of household work being performed by women,^1,14,15 often twice as much or more.^1,14,16 The results could be related to several factors. First, the observed prenatal work patterns may be somewhat atypical for these couples: Several women indicated that they had cut back on their housework or employment hours because of pregnancy-associated fatigue or other health problems. Second, this was a very homogenous population of employed young couples without children, in contrast to the more diverse samples (which included adults with children and more unemployed wives) used in many previous studies. Alternatively, these findings might represent a societal trend toward men and women sharing housework more equitably.

Importantly for many of these parents, the changes in work that they anticipated after giving birth likely represent the largest and most abrupt increase in work responsibilities that they will face in their adult lives. Unfortunately, it is a change for which many parents are ill-prepared. This lack of preparation is likely due, at least in part, to the paucity of information available on new parents’ actual workloads (no specific information was found in the medical, sociological, and psychological searches), society’s tendency to focus on the health needs of newborns and children more than those of their parents, and the absence of a consistent method for educating adolescents and young adults about the responsibilities of supporting and nurturing a family.

Each of these needs will be addressed. First, additional research is needed on changes in work responsibilities for new parents from more diverse populations, and we need a greater understanding of how these work responsibilities affect health and marital well-being (the goal of the ongoing randomized controlled trial). Second, we need a broader view of postpartum care, such that the physical, mental, and social needs of both the parents and newborn are considered in an ongoing manner. This is a perspective that family physicians are uniquely positioned to adopt and foster within the context of prenatal and postpartum care for the family unit. Moreover, family physicians could also be part of the solution to the third need, that of teaching would-be parents about postpartum work and family responsibilities. These efforts may pay important dividends in strengthening the fiber of the family and improving the well-being of its individual members.

Limitations

The limitations of this study include the modest response rate, the potential for selection bias, and the relatively homogeneous sample. In addition, parents’ estimations of work time may not be completely accurate, and these simple estimations do not account for such factors as the intensity of work at any given time (as when one juggles numerous responsibilities concurrently) or the issue of who holds the ultimate responsibility for a given task. Though future workload projections may be even more inaccurate than current estimations, the results shown here are more than individuals’ guesses about their future work; they represent couples’ intentional plans for sharing postpartum work responsibilities.

Conclusions

These expectant first-time parents anticipated considerable expansions in their work activities after childbirth, with women planning a greater share of the total postpartum workload. This information is important for new parents and for the health care providers who attend them as they resume their household, family, and paid work responsibilities after childbirth.

Acknowledgments

This study was funded by the University of Minnesota graduate school.

The author would like to thank Anne Marie Weber-Main for her editing assistance and Bruce Center for his help with data analysis.

Clinical question

Which antiarrhythmic agents are efficacious for conversion of nonpostoperative atrial fibrillation and for subsequent maintenance of sinus rhythm?

Atrial fibrillation (AF) is the most common sustained tachyarrhythmia faced by all physicians. The prevalence of AF, estimated at 0.4% in the general population,¹ increases with age to almost 10% among those aged 80 to 89 years.^2,3 The age-adjusted incidence of AF has increased over the last 30 years.⁴ AF accounts for more days of hospitalization for either acute hemodynamic compromise or treatment of the arrhythmia than all ventricular arrhythmias combined.⁵ All admissions for the complications of stroke and chronic heart failure are not reflected in these data. Overall, patients with AF have twice the mortality of a control population without AF and an attributable risk of stroke of 24% in those aged 80 to 89 years.²

One of the most important issues for management of AF is the need for conversion to sinus rhythm and subsequent maintenance of sinus rhythm (MSR), particularly for symptomatic patients. Although conversion can be accomplished by electrical cardioversion, it is frequently accomplished with pharmacologic agents because of patient or physician preference and anesthesia risks. These agents may also be used for subsequent MSR. In addition to the numerous relatively new or investigational agents such as ibutilide and dofetilide there are at least 7 agents commonly used for either conversion or MSR: quinidine, disopyramide, procainamide, flecainide, propafenone, amiodarone, and sotalol.⁶ This plethora of antiarrhythmic agents for either conversion of AF or MSR makes it difficult for physicians to know which are best for their patients. We reviewed the evidence on pharmacologic management of AF as part of the Johns Hopkins Evidence-Based Practice Center sponsored by the Agency for Health Care Policy and Research (now the Agency for Healthcare Research and Quality).

Methods

Search Strategy

We used the Medical Subject Heading terms “atrial fibrillation,” “atrial flutter,” “random allocation,” “double-blind method,” and “single-blind method.” Publication types of “randomized controlled trial” and “controlled clinical trial” were included. Although the search was not restricted to citations in the English-language literature, subsequent article review involved only English-language publications because of budgetary constraints.

The primary literature source was the CENTRAL database, The Cochrane Library 1998 issues 1 and 2, produced by the Cochrane Collaboration from EMBASE and MEDLINE and encompassing 1948 through the present.^7,8 Second, MEDLINE was searched using both OVID and PubMed from 1966 to May 1998. Third, we used the PubMed feature of “related articles” for primary articles identified in the CENTRAL database. Fourth, a review of recent hand search results submitted to the Baltimore Cochrane Center from the Cardiovascular Randomized Controlled Trial Registry was used. Finally, to capture newly published studies the core study team scanned the contents of the journals most frequently cited in the search results database.

To address the issue of publication bias we asked investigators in the field and search coordinators of relevant Cochrane Collaborative Review Groups to identify any trials they were aware of that had been completed but not published. We decided that construction of funnel plots was practical because of the relatively small number of trials for any specific pharmacologic agent.

Study Inclusion

Articles had to report original data on pharmacologic management of nonpostoperative AF in adults in the context of a randomized clinical trial to be eligible for inclusion in our review. Pairs of independent investigators reviewed all identified abstracts according to these inclusion criteria. All discrepancies about inclusion were resolved by consensus.

Study Quality Assessment

The Evidence-Based Practice Center team developed a data form for extracting information on study quality based on a review of forms used in other meta-analytic studies by study investigators,^9-11 a literature review of the topic,^12,13 and with the assistance of the Cochrane Collaboration. The form contained 22 questions assessing study quality in 5 areas: clarity of description of the study population; potential for bias and confounding; description of therapy, outcomes and follow-up; and statistical quality and interpretation. Each question included a 4 to 5–level subjective ranking of study quality with the resultant score for each of the 5 areas comprising the total points accumulated out of the maximum possible points for all relevant questions in that area. The overall study quality score consisted of the mean score of these 5 areas.

Teams of independent reviewers assessed the quality of each study with differences resolved by consensus. Given the difficult nature of assessing study quality based on article review, the team decided to collectively review and discuss any articles receiving an overall score less than 50% to reach decisions regarding study inclusion.

Data Extraction

Because of the large volume of articles for review, quantitative data were extracted by one reviewer and then checked for accuracy by a second reviewer with consensus resolution of differences. The reviewers were not blinded to the author, institution, and journal, because recent work has indicated that such masking makes little difference in the results.¹⁴ In trials involving both AF and atrial flutter patients, data were only extracted for the AF patients whenever possible.

Data Synthesis and Analysis

Before doing the meta-analysis we first performed both qualitative and quantitative assessments of heterogeneity between the trials to ensure appropriateness of subsequent data combination. The reviewers subjectively assessed qualitative heterogeneity on the basis of similarity between studies on age of subjects, type and duration of AF, comorbidities, therapeutic regimens, and follow-up times. We performed quantitative analysis of heterogeneity using the statistical test of data heterogeneity included in Review Manager (RevMan) version 3.1 (Cochrane Collaboration, Oxford, England).

For data synthesis we defined control treatment to include placebo, verapamil, diltiazem, or digoxin. An analysis of identified trials evaluating verapamil, diltiazem, or digoxin compared with placebo supported this definition, since all of these agents were found to have no efficacy compared with placebo for either conversion or MSR.¹⁵ We also combined treatment arms within a given study that used the same antiarrhythmic agent at different dosages. Analysis of these arms individually supported their consideration as one arm.¹⁵ When life table analysis was used, we extracted the resultant cumulative percentages of successful outcomes and applied them to the initial overall subject number in each trial arm to derive a proportion for meta-analysis inclusion.

We constructed evidence tables to present the data separately for the 2 main outcomes of conversion of AF and MSR and created scatter plots of the absolute rates of conversion and MSR.

For meta-analysis the primary effect measure chosen was the odds ratio (OR) with studies weighted based on the precision of the estimate within each study. A fixed-effects model was used. In cases of significant quantitative data heterogeneity, we explored the etiology of the heterogeneity and used random-effects modeling when appropriate.

We chose the following categorization of strength of evidence by noting the placement of the point estimate of the OR and the width of the confidence interval (CI) surrounding it: (1) strong evidence of efficacy: OR >1.0, 99% CI does not include 1.0 (P <.01); (2) moderate evidence of efficacy: OR >1.0, 95% CI does not include 1.0, but 99% CI includes 1.0 (.01 P .05); (3) suggestive evidence of efficacy: 95% CI includes 1.0 in the lower tail (.05< P <.25), and the OR is in a clinically meaningful range; (4) inconclusive evidence of efficacy: 95% CI is widely distributed around 1.0; and (5) strong evidence of lack of efficacy: OR near 1.0, 95% CI is narrow and does not include a clinically meaningful difference from an OR of 1.0. When the point estimate was less than 1.0, we called this negative efficacy and used the same categorization of strong, moderate, and suggestive evidence on the basis of the point estimate OR and CI. For clarity our reported CIs are at the 95% level.

We also estimated the number needed to treat (NNT) from the resultant OR. The NNT provides an estimate of the number of subjects needed to treat with a therapy to have one more subject experience a desired outcome relative to the comparison group. To do these calculations for the conversion data we assumed a 30% spontaneous conversion rate for the control treatment group, which was consistent with the data. Similarly, to calculate the NNT for MSR we assumed a 30% recurrence rate of AF by 6 months in the control treatment group, which was also consistent with the data. The upper and lower 95% CI estimates for each OR were used to estimate the NNT.

All analyses were completed using RevMan.

Results

Search Strategy and Study Inclusion

Our review of 521 abstracts identified 130 articles for review.¹⁵ After article review, 36 studies^16-51 were eligible for inclusion in our meta-analysis,²⁵ relevant to the conversion of AF outcome and 15 to MSR outcome. All 36 studies used control treatment comparison groups. Our inquiry of experts did not identify any trials for inclusion that had been completed but not published.

In addition to these 36 trials our search identified 16 trials involving unique comparisons between antiarrhythmic agents that precluded meta-analysis. For completeness the results of these trials are discussed and the data presented in Tables 1E and 2E.* We also identified 15 trials using new or uncommon agents. Discussion of these results was published previously.¹⁵

Study Quality Assessment

Based on study quality scores we concluded that all 36 identified trials were of sufficient quality for inclusion. The overall quality scores ranged from 36% to 84% with only 2 studies^17,22 having an overall score less than 50%. Team review of these articles deemed them acceptable for inclusion. Details on the study quality scores was published previously.¹⁵

Study Characteristics and Qualitative Synthesis

Table 1 and Table 2 show important design elements and results of the trials (ie, subject characteristics, sample size, treatment regimens, follow-up times, and reported treatment effects).

The important subject characteristics reported involve age, type of AF, and duration of AF. These areas have an impact on the responsiveness to conversion and ability to avoid recurrent AF.

The mean ages for the trials were generally comparable, ranging between 47 and 71 years with only 7 trials (19%) having mean ages greater than 65 years.

Overall, the studies provided sparse and varying terminology regarding the type and duration of AF; because of this we were unable to reliably segregate studies accordingly. Thus, we relied on the verbatim descriptive terminology used by each study with the understanding that this represented differing definitions between the studies. This difficulty in assessing the type of AF was primarily relevant to conversion studies involving propafenone, amiodarone, and quinidine.^{16,29,32,35,36} These 5 trials all reported control treatment conversion rates greater than 70%, suggesting that the enrolled subjects had predominantly paroxysmal AF. To examine the potential effect of this, we evaluated the quantitative change in the meta-analysis data when excluding these 5 articles with high outlier spontaneous conversion rates.

The therapeutic regimens were generally comparable for any given antiarrhythmic agent for both conversion of AF and MSR. Notably, of the 12 trials evaluating propafenone for conversion of AF, half used oral regimens, and half used intravenous regimens. Separate quantitative analysis comparing these routes showed no significant differences in treatment effects.¹⁵

Regarding follow-up times, the 25 trials of conversion of AF were all comparable and were typically less than 24 hours. There was variability in follow-up time among the 15 trials involving MSR, with a range of 1 to 15 months. For any given antiarrhythmic agent, there was at least one trial with a minimum of 6 months follow-up time.

Overall, our subjective qualitative synthesis of the 36 trials regarding trial inclusion/exclusion criteria, trial size, subject age, subject sex, comorbidities, therapeutic regimens, follow-up times, and reported treatment effects suggested that quantitative synthesis was reasonable because of relatively minor qualitative differences among the studies.

Quantitative Synthesis: Evidence on Pharmacologic Conversion of AF Figure 1 shows the scatter plot of absolute conversion rates for these 25 studies. Two trials involved 2 antiarrhythmic agent arms compared with a third placebo arm thus providing 27 data points.^6,20

All of the antiarrhythmic agents except sotalol had point estimates of conversion rates consistent with efficacy compared with control treatment, though many were not statistically significant. The evidence for sotalol was consistent with negative efficacy for conversion of AF.

The results of the mathematical pooling of these 25 trials are shown in Table 3. The strongest evidence of efficacy of conversion of AF compared with control treatment existed for ibutilide/dofetilide (OR=29.1; 95% CI, 9.8-86.1)^38-40 and flecainide (OR=24.7; 95% CI, 9.0-68.3).^20-23 The range of estimated NNT to have one more subject convert relative to control treatment is 1.5 to 2.0 for both ibutilide/dofetilide and flecainide.

With respect to propafenone there was some modest quantitative heterogeneity of the data for conversion of AF presumably related to issues regarding type and duration of AF. Since we were unable to definitively clarify these issues, we felt a more conservative random-effects model was appropriate for this meta-analysis since that type of modeling assumes variability in the estimated population treatment effects between the studies. Thus, although the magnitude of treatment effect compared with control treatment was less for propafenone (OR=4.6; 95% CI, 2.6-8.2)^16,20,24-33 than for ibutilide/dofetilide or flecainide, the results gave strong evidence of propafenone efficacy for conversion of AF. The estimated range of NNT to have one more subject convert relative to control treatment is 2.0 to 4.5.

We analyzed the impact of the 5 trials with exceptionally high spontaneous conversion rates for AF, 3 of which involved propafenone. Exclusion of these 3 trials^16,29,32 did not substantially alter the pooled treatment effect of the remaining 9 trials (OR=6.6; 95% CI, 3.6-12.0).

The data on quinidine (OR=2.9; 95% CI, 1.2-7.0)^16-18 were consistent with moderate evidence of efficacy for conversion of AF. The summary data for quinidine versus control treatment remained consistent with moderate evidence of efficacy for conversion of AF (OR=7.2; 955 CI, 1.7-30.4) when we performed outlier analysis by excluding the trial by Capucci and colleagues¹⁶ that had a high spontaneous conversion rate.

Comparable with the situation with propafenone, the data on amiodarone had modest quantitative heterogeneity, likely because of issues regarding type and duration of AF and prevalence of coronary artery disease. Given this, we again chose to perform more conservative random-effects modeling for this data synthesis. As such, the data on amiodarone (OR=5.7; 95% CI, 1.0-33.4)^34-36 were consistent with suggestive evidence of efficacy for conversion of AF compared with control treatment. Outlier analysis involving exclusion of 2 trials with high spontaneous conversion rates^35,36 left only one small trial³⁴ as evidence of amiodarone efficacy versus control treatment for conversion of AF. This trial had a sample size of only 24 subjects with resultant extremely wide CIs that made interpretation of this data difficult (OR=69.0; 95% CI, 3.2-1500.0).

The summary data for both disopyramide and sotalol each reflected only one relatively small trial. For disopyramide the data (OR=7.0; 95% CI, 0.3-153.0)¹⁹ were consistent with suggestive evidence of efficacy compared with control treatment. For sotalol the data (OR=0.4; CI, 0.0-3.0)³⁷ were consistent with suggestive evidence of negative efficacy compared with control treatment.

As part of the overall project evaluating management of atrial fibrillation by the Johns Hopkins Evidence-Based Practice Center, we also reviewed the data on 8 trials that had direct comparisons between the major antiarrhythmic agents for conversion of AF.¹⁵ Because of the overall paucity of data on these direct comparisons, mathematical data pooling was not feasible. The one trial evaluating procainamide compared with flecainide reported lower conversion rates for procainamide. In general, these results were consistent with our meta-analysis results.

Quantitative Synthesis: Evidence of Pharmacologic MSR

Figure 2 shows the scatter plot of absolute rates for MSR of the identified trials. Two trials reported their results in a manner not conducive for our data extraction.^17,48 The results of these 2 trials are included in Table 2. Two other trials involved 2 pharmacologic arms compared with one control treatment arm, resulting in 15 data points on Figure 2.^41,49 Notably, none of these trials examined the efficacy of amiodarone or procainamide compared with control treatment for MSR.

All of the major antiarrhythmic agents had evidence of efficacy for MSR compared with control treatment, although some were not statistically significant.

The results of mathematical data pooling for MSR are shown in Table 4. All of the antiarrhythmic agents had strong and relatively comparable evidence of efficacy compared with control treatment, and the point estimates were all consistent with fairly large treatment effect sizes: quinidine (OR=4.1; 95% CI, 2.5-6.7)^18,41-43; disopyramide (OR=3.4; 95% CI, 1.6-7.1)^44-45; flecainide (OR=3.1; 95% CI, 1.5-6.2)^46-48; propafenone (OR=3.7; 95% CI, 2.4-5.7)^27,49-51; and sotalol (OR=7.1; 95% CI, 3.8-13.4).^37,49

The estimated range of NNT to have one less subject experience AF recurrence relative to control treatment is as follows: quinidine 2.3 to 4.6, disopyramide 2.2 to 9.4, flecainide 2.3 to 10.9, propafenone 2.4 to 4.8, and sotalol 1.8 to 3.1.

Although we identified no clinical trials comparing amiodarone with a control treatment, 2 trials did compare amiodarone to other antiarrhythmic agents (Table 2) and should at least be noted given the overall paucity of data on amiodarone for MSR. One small trial compared amiodarone with quinidine (OR=1.1; 95% CI, 0.1-20.0) and was inconclusive. However, a second trial⁶⁵ compared amiodarone with disopyramide (OR=3.2; 95% CI, 1.0-9.6) and was consistent with moderate evidence of amiodarone efficacy compared with disopyramide for MSR. This study reported only interim results, and our searches did not identify the final results of the trial. One could infer from this study that there is indirect strong evidence of amiodarone efficacy for MSR compared with control treatment, since disopyramide had strong evidence of efficacy compared with control treatment.

As another part of the project evaluating management of atrial fibrillation by the Johns Hopkins Evidence-Based Practice Center, we reviewed the data on 10 trials that had direct comparisons between the major antiarrhythmic agents regarding MSR in AF.¹⁵ Because of the overall paucity of data on these direct comparisons, mathematical data pooling was not feasible and definitive ranking of the agents for MSR efficacy was not possible. Overall, these results were consistent with our meta-analysis showing no one agent as clearly superior over other agents.

Evidence on Adverse Events

During our data extraction we only noted where trials specifically mentioned various events such as ventricular arrhythmias or other nontransient arrhythmias (Table 5). We did not perform formal data synthesis regarding adverse events because the data were too sporadically reported.

In addition, caution must be used in interpreting rates of adverse events that resulted in study withdrawal or dosage decreases, since there was no uniformity regarding the indications for withdrawals of dosage decreases among the studies. Also with respect to conversion trials, many studies involved one-time study drug administration that limited the applicability of this adverse event definition.

Discussion

Pharmacologic conversion of AF is frequently the therapy of choice compared with electrical cardioversion, especially in cases of short-duration AF, significant anesthesia risk, or recent postprandial status of a patient. Little guidance based on scientific evidence has existed regarding the best pharmacologic agents to achieve conversion of AF. On the basis of this formal data review, we are unable to state definitively the relative efficacy of the agents compared with each other because of the inability to ensure comparable subjects within the control treatment groups for the evaluated trials. However, this data synthesis did find that the strongest evidence of efficacy compared with control treatment for conversion of AF existed for ibutilide/dofetilide and flecainide. Less strong but still conclusive evidence existed for propafenone. Quinidine had moderate evidence of efficacy, while only suggestive evidence of efficacy existed for disopyramide and amiodarone. Finally, sotalol had suggestive evidence of negative efficacy compared with control treatment for conversion of AF. Notably, there was no randomized trial on the use of procainamide compared with control treatment for conversion of AF.

The clinical implications of these findings need to be viewed in the light of previous reports regarding adverse events, since our ability to synthesize the adverse event data from these trials was limited.

Ibutilide and dofetilide are new class III antiarrhythmic agents currently undergoing extensive clinical trials. Although limited primarily to clinical trial data, our data and other reports conclude that these drugs have a rate of ventricular arrhythmias (particularly torsade de pointes) of 3% to 9%.⁵² However, there were no reported deaths or prolonged resuscitations among the trials examined.^38-40 Data from long-term use in everyday clinical practice evaluating these agents in less controlled circumstances are not available.

There have been reports of increased mortality with flecainide, although this occurred for prevention of ventricular ectopic activity in subjects with coronary artery disease in the Cardiac Arrhythmia Suppression Trial.⁵³ However, patients with atrial fibrillation may frequently also have ventricular ectopic activity and coronary artery disease. A recent review of flecainide safety for treatment of supraventricular arrhythmias using both randomized clinical trials and uncontrolled trials concluded that the risk of clinically significant adverse cardiac effects was small but not negligible.⁵⁴ From 1794 reviewed treatment courses 2% had atrial proarrhythmic events with some requiring urgent electrical cardioversion because of hemodynamic compromise, and 2% had pre-excitation worsening or new ventricular arrhythmias including 9 cases of sustained ventricular tachycardia or fibrillation and 4 cases of sudden cardiac death. Another report retrospectively compared the mortality rates of patients with atrial arrhythmias in completed pharmaceutical company–sponsored trials treated with flecainide with a population seen at the research arrhythmia clinic.⁵⁵ The researchers concluded that there appeared to be no excess mortality in patients treated with flecainide for supraventricular arrhythmias. If the main concern among patients with atrial fibrillation is coronary artery disease and resultant ventricular dysfunction, our data synthesis was unable to address this because of poor documentation of definitions regarding presence of coronary artery disease, presence of abnormal left ventricular function, and lack of result stratification by these conditions.

Since these 2 agents (ibutilide/dofetilide and flecainide) had the largest treatment effect sizes for conversion of AF, additional research directly comparing them, comparing them with electrical cardioversion, and better quantifying adverse event rates stratified by the presence of coronary artery disease, structural heart disease, left ventricular hypertrophy, and long QT intervals would help solidify their efficacy and safety for conversion of AF.

Similarly, more research on the efficacy of amiodarone is warranted given the paucity of data, a general perception of relatively minor side effects, and a high prevalence of clinical use for AF.

Pharmacologic MSR for AF is a therapeutic option for patients with high recurrence rates and patients with symptomatic AF. Comparable with conversion of AF therapy, no consensus exists on the best pharmacologic agents to achieve MSR in AF. Our formal data synthesis was unable to show definitively the relative efficacy of the agents for MSR compared with each other because of the inability to ensure comparable subjects within the control treatment groups for the evaluated trials. However, this data synthesis did find strong and comparable efficacy evidence for quinidine, disopyramide, flecainide, propafenone, and sotalol. Notably, the data for amiodarone use for MSR are sparse with no trials comparing amiodarone with control treatment, and no trial evaluated procainamide either compared with control treatment or another agent.

The clinical implications of these data also need to be viewed in light of previous reports regarding adverse events, since our ability to synthesize the adverse event data was limited. The issues regarding flecainide have already been discussed. The Class Ia agents quinidine, disopyramide, and procainamide have classically been associated with torsade de pointes because of their prolongation of the QT interval, but cases of torsade de pointes have also been reported with propafenone, flecainide, amiodarone, and sotalol. The reported risk factors for proarrhythmic events with each of the agents vary from hypokalemia and bradycardia for quinidine to serum concentration for sotalol. A recent review concluded that all of the antiarrhythmic agents have potential for uncommon but serious proarrhythmic effects.⁵⁶ Unfortunately, this does not help the clinician sort through all of the available agents.

More research involving direct comparisons between all these agents for MSR in AF would help to definitively rank the efficacy of the agents and to compare their adverse event profiles. Stratification of patients on the basis of the presence of coronary artery disease, structural heart disease, left ventricular hypertrophy, and long QT intervals would permit better assessment of adverse event risks. Both the ongoing AF Follow-up Investigation of Rhythm Management (AFFIRM)⁵⁷ sponsored by the National Heart, Lung, and Blood Institute and the ongoing Prognosis in Afib (PAIF)⁵⁸ study may help provide more information directly comparing agents for MSR.

Limitations

Overall with respect to our data synthesis for both conversion of AF and MSR, we cannot exclude a publication bias despite our best efforts to minimize this known limitation of evidence reviews.

In terms of the actual trials reviewed, we do not believe that subject-specific factors significantly influenced the accumulated evidence based on examination of the inclusion/exclusion criteria and baseline subject characteristics of all the reviewed trials. However, 4 points about this should be noted. First, the age range of the subjects in these trials was somewhat younger than might be seen in a population-based sample of AF. Since it is possible that response to pharmacologic therapy may differ with age, this needs to be kept in mind. Second, our target population consisted of nonpostoperative AF. The accumulated evidence, therefore, may not be applicable to subjects with postoperative AF. In addition, it is difficult to assure the generalizability of our results based on randomized clinical trials to everyday clinical practice. Third, given the relatively small number of trials for any given comparison, we were unable to perform sensitivity analysis on estimated treatment effects on the basis of our assessments of study quality. Finally, our results regarding quinidine may partially reflect time-dependent improvements in medical care. The majority of trials evaluating quinidine were older. However, for both conversion and MSR at least one trial of quinidine was contemporary, and in both conditions found quinidine less efficacious than the older trials.

It is important to note areas of missing evidence that limit more definitive statements for selection of antiarrhythmic agents for management of AF. First, there are few direct comparisons between antiarrhythmic agents for either conversion of AF or MSR. Since control treatment groups vary between trials, direct comparisons between antiarrhythmic agents are instrumental in assessing relative efficacy. Second, there are particularly sparse data for amiodarone and procainamide, especially with respect to MSR. Although several published reviews^6,59 report efficacy of these agents for conversion of AF or MSR, our data from randomized clinical trials (particularly for MSR) do not support this. The AFFIRM and PIAF trials may help address this issue. Third, almost no data were found in this review for the effects of the various antiarrhythmic agents on quality of life. Since patient experiences may significantly influence treatment compliance, quality of life effects need to be better defined. Finally, the follow-up times for all trials on MSR were relatively short. Since the ability to remain free of recurrence has an impact on a patient’s preference for continuing therapy, it would be informative to test the antiarrhythmic agents over a longer period of time for efficacy. These last 2 points may also be addressed in the AFFIRM trial.

Conclusions

Our formal data synthesis of 36 randomized clinical trials of pharmacologic AF conversion and MSR found evidence consistent with superior efficacy relative to control treatment for AF conversion with ibutilide/dofetilide and flecainide. The strength of evidence for MSR relative to control treatment was strong and comparable for quinidine, disopyramide, flecainide, propafenone, and sotalol. Most important, despite the high prevalence of AF the data for the relative efficacy of the antiarrhythmic agents for both conversion and MSR are sparse and inconclusive. Defining these relative efficacies should be a research priority.

Recommendations for clinical practice

On the basis of data from randomized clinical trials, ibutilide, dofetilide, and flecainide have superior efficacy for conversion of AF. However, the data are sparse for ibutilide and dofetilide, and use of flecainide needs to be considered in the context of other comorbidities, such as ventricular ectopy and coronary artery disease. For maintenance of sinus rhythm, no one agent has been shown to have superior efficacy. Clinical practices need to focus on upcoming trial results that involve direct comparisons among agents to better understand relative efficacies of the antiarrhythmic agents for both aspects of AF management.

Acknowledgments

Our study was conducted by the Johns Hopkins Evidence-Based Practice Center through contract No. 290-97-0006 from the Agency for Health Care Policy and Research, Rockville, Maryland. We are responsible for its contents including any clinical or treatment recommendations. No statement in this article should be construed as an official position of the Agency for Healthcare Research and Quality or the United States Department of Health and Human Services. Dr Miller was supported by the Hayden Whitney Smith Research Scholarship. We thank Hanan S. Bell, PhD; Ronald D. Berger, MD; Gary Gerstenblith, MD; David E. Haines, MD; Michael L. Lefevre, MD, MSPH; Andrew Epstein, MD; John A. Kastor, MD; Chris Burton, MD; Jerome A. Osheroff, MD; Barbara J. Drew, RN, PhD; and Kathleen McCauley, RN, PhD, for their assistance as expert advisers for this study. We also thank David Yu, MD, and Paul Abboud for their assistance with this study.

We are especially grateful to Donna Lea for her secretarial support.

Clinical question

Which antiarrhythmic agents are efficacious for conversion of nonpostoperative atrial fibrillation and for subsequent maintenance of sinus rhythm?

Atrial fibrillation (AF) is the most common sustained tachyarrhythmia faced by all physicians. The prevalence of AF, estimated at 0.4% in the general population,¹ increases with age to almost 10% among those aged 80 to 89 years.^2,3 The age-adjusted incidence of AF has increased over the last 30 years.⁴ AF accounts for more days of hospitalization for either acute hemodynamic compromise or treatment of the arrhythmia than all ventricular arrhythmias combined.⁵ All admissions for the complications of stroke and chronic heart failure are not reflected in these data. Overall, patients with AF have twice the mortality of a control population without AF and an attributable risk of stroke of 24% in those aged 80 to 89 years.²

One of the most important issues for management of AF is the need for conversion to sinus rhythm and subsequent maintenance of sinus rhythm (MSR), particularly for symptomatic patients. Although conversion can be accomplished by electrical cardioversion, it is frequently accomplished with pharmacologic agents because of patient or physician preference and anesthesia risks. These agents may also be used for subsequent MSR. In addition to the numerous relatively new or investigational agents such as ibutilide and dofetilide there are at least 7 agents commonly used for either conversion or MSR: quinidine, disopyramide, procainamide, flecainide, propafenone, amiodarone, and sotalol.⁶ This plethora of antiarrhythmic agents for either conversion of AF or MSR makes it difficult for physicians to know which are best for their patients. We reviewed the evidence on pharmacologic management of AF as part of the Johns Hopkins Evidence-Based Practice Center sponsored by the Agency for Health Care Policy and Research (now the Agency for Healthcare Research and Quality).

Methods

Search Strategy

We used the Medical Subject Heading terms “atrial fibrillation,” “atrial flutter,” “random allocation,” “double-blind method,” and “single-blind method.” Publication types of “randomized controlled trial” and “controlled clinical trial” were included. Although the search was not restricted to citations in the English-language literature, subsequent article review involved only English-language publications because of budgetary constraints.

The primary literature source was the CENTRAL database, The Cochrane Library 1998 issues 1 and 2, produced by the Cochrane Collaboration from EMBASE and MEDLINE and encompassing 1948 through the present.^7,8 Second, MEDLINE was searched using both OVID and PubMed from 1966 to May 1998. Third, we used the PubMed feature of “related articles” for primary articles identified in the CENTRAL database. Fourth, a review of recent hand search results submitted to the Baltimore Cochrane Center from the Cardiovascular Randomized Controlled Trial Registry was used. Finally, to capture newly published studies the core study team scanned the contents of the journals most frequently cited in the search results database.

To address the issue of publication bias we asked investigators in the field and search coordinators of relevant Cochrane Collaborative Review Groups to identify any trials they were aware of that had been completed but not published. We decided that construction of funnel plots was practical because of the relatively small number of trials for any specific pharmacologic agent.

Study Inclusion

Articles had to report original data on pharmacologic management of nonpostoperative AF in adults in the context of a randomized clinical trial to be eligible for inclusion in our review. Pairs of independent investigators reviewed all identified abstracts according to these inclusion criteria. All discrepancies about inclusion were resolved by consensus.

Study Quality Assessment

The Evidence-Based Practice Center team developed a data form for extracting information on study quality based on a review of forms used in other meta-analytic studies by study investigators,^9-11 a literature review of the topic,^12,13 and with the assistance of the Cochrane Collaboration. The form contained 22 questions assessing study quality in 5 areas: clarity of description of the study population; potential for bias and confounding; description of therapy, outcomes and follow-up; and statistical quality and interpretation. Each question included a 4 to 5–level subjective ranking of study quality with the resultant score for each of the 5 areas comprising the total points accumulated out of the maximum possible points for all relevant questions in that area. The overall study quality score consisted of the mean score of these 5 areas.

Teams of independent reviewers assessed the quality of each study with differences resolved by consensus. Given the difficult nature of assessing study quality based on article review, the team decided to collectively review and discuss any articles receiving an overall score less than 50% to reach decisions regarding study inclusion.

Data Extraction

Because of the large volume of articles for review, quantitative data were extracted by one reviewer and then checked for accuracy by a second reviewer with consensus resolution of differences. The reviewers were not blinded to the author, institution, and journal, because recent work has indicated that such masking makes little difference in the results.¹⁴ In trials involving both AF and atrial flutter patients, data were only extracted for the AF patients whenever possible.

Data Synthesis and Analysis

Before doing the meta-analysis we first performed both qualitative and quantitative assessments of heterogeneity between the trials to ensure appropriateness of subsequent data combination. The reviewers subjectively assessed qualitative heterogeneity on the basis of similarity between studies on age of subjects, type and duration of AF, comorbidities, therapeutic regimens, and follow-up times. We performed quantitative analysis of heterogeneity using the statistical test of data heterogeneity included in Review Manager (RevMan) version 3.1 (Cochrane Collaboration, Oxford, England).

For data synthesis we defined control treatment to include placebo, verapamil, diltiazem, or digoxin. An analysis of identified trials evaluating verapamil, diltiazem, or digoxin compared with placebo supported this definition, since all of these agents were found to have no efficacy compared with placebo for either conversion or MSR.¹⁵ We also combined treatment arms within a given study that used the same antiarrhythmic agent at different dosages. Analysis of these arms individually supported their consideration as one arm.¹⁵ When life table analysis was used, we extracted the resultant cumulative percentages of successful outcomes and applied them to the initial overall subject number in each trial arm to derive a proportion for meta-analysis inclusion.

We constructed evidence tables to present the data separately for the 2 main outcomes of conversion of AF and MSR and created scatter plots of the absolute rates of conversion and MSR.

For meta-analysis the primary effect measure chosen was the odds ratio (OR) with studies weighted based on the precision of the estimate within each study. A fixed-effects model was used. In cases of significant quantitative data heterogeneity, we explored the etiology of the heterogeneity and used random-effects modeling when appropriate.

We chose the following categorization of strength of evidence by noting the placement of the point estimate of the OR and the width of the confidence interval (CI) surrounding it: (1) strong evidence of efficacy: OR >1.0, 99% CI does not include 1.0 (P <.01); (2) moderate evidence of efficacy: OR >1.0, 95% CI does not include 1.0, but 99% CI includes 1.0 (.01 P .05); (3) suggestive evidence of efficacy: 95% CI includes 1.0 in the lower tail (.05< P <.25), and the OR is in a clinically meaningful range; (4) inconclusive evidence of efficacy: 95% CI is widely distributed around 1.0; and (5) strong evidence of lack of efficacy: OR near 1.0, 95% CI is narrow and does not include a clinically meaningful difference from an OR of 1.0. When the point estimate was less than 1.0, we called this negative efficacy and used the same categorization of strong, moderate, and suggestive evidence on the basis of the point estimate OR and CI. For clarity our reported CIs are at the 95% level.

We also estimated the number needed to treat (NNT) from the resultant OR. The NNT provides an estimate of the number of subjects needed to treat with a therapy to have one more subject experience a desired outcome relative to the comparison group. To do these calculations for the conversion data we assumed a 30% spontaneous conversion rate for the control treatment group, which was consistent with the data. Similarly, to calculate the NNT for MSR we assumed a 30% recurrence rate of AF by 6 months in the control treatment group, which was also consistent with the data. The upper and lower 95% CI estimates for each OR were used to estimate the NNT.

All analyses were completed using RevMan.

Results

Search Strategy and Study Inclusion

Our review of 521 abstracts identified 130 articles for review.¹⁵ After article review, 36 studies^16-51 were eligible for inclusion in our meta-analysis,²⁵ relevant to the conversion of AF outcome and 15 to MSR outcome. All 36 studies used control treatment comparison groups. Our inquiry of experts did not identify any trials for inclusion that had been completed but not published.

In addition to these 36 trials our search identified 16 trials involving unique comparisons between antiarrhythmic agents that precluded meta-analysis. For completeness the results of these trials are discussed and the data presented in Tables 1E and 2E.* We also identified 15 trials using new or uncommon agents. Discussion of these results was published previously.¹⁵

Study Quality Assessment

Based on study quality scores we concluded that all 36 identified trials were of sufficient quality for inclusion. The overall quality scores ranged from 36% to 84% with only 2 studies^17,22 having an overall score less than 50%. Team review of these articles deemed them acceptable for inclusion. Details on the study quality scores was published previously.¹⁵

Study Characteristics and Qualitative Synthesis

Table 1 and Table 2 show important design elements and results of the trials (ie, subject characteristics, sample size, treatment regimens, follow-up times, and reported treatment effects).

The important subject characteristics reported involve age, type of AF, and duration of AF. These areas have an impact on the responsiveness to conversion and ability to avoid recurrent AF.

The mean ages for the trials were generally comparable, ranging between 47 and 71 years with only 7 trials (19%) having mean ages greater than 65 years.

Overall, the studies provided sparse and varying terminology regarding the type and duration of AF; because of this we were unable to reliably segregate studies accordingly. Thus, we relied on the verbatim descriptive terminology used by each study with the understanding that this represented differing definitions between the studies. This difficulty in assessing the type of AF was primarily relevant to conversion studies involving propafenone, amiodarone, and quinidine.^{16,29,32,35,36} These 5 trials all reported control treatment conversion rates greater than 70%, suggesting that the enrolled subjects had predominantly paroxysmal AF. To examine the potential effect of this, we evaluated the quantitative change in the meta-analysis data when excluding these 5 articles with high outlier spontaneous conversion rates.

The therapeutic regimens were generally comparable for any given antiarrhythmic agent for both conversion of AF and MSR. Notably, of the 12 trials evaluating propafenone for conversion of AF, half used oral regimens, and half used intravenous regimens. Separate quantitative analysis comparing these routes showed no significant differences in treatment effects.¹⁵

Regarding follow-up times, the 25 trials of conversion of AF were all comparable and were typically less than 24 hours. There was variability in follow-up time among the 15 trials involving MSR, with a range of 1 to 15 months. For any given antiarrhythmic agent, there was at least one trial with a minimum of 6 months follow-up time.

Overall, our subjective qualitative synthesis of the 36 trials regarding trial inclusion/exclusion criteria, trial size, subject age, subject sex, comorbidities, therapeutic regimens, follow-up times, and reported treatment effects suggested that quantitative synthesis was reasonable because of relatively minor qualitative differences among the studies.

Quantitative Synthesis: Evidence on Pharmacologic Conversion of AF Figure 1 shows the scatter plot of absolute conversion rates for these 25 studies. Two trials involved 2 antiarrhythmic agent arms compared with a third placebo arm thus providing 27 data points.^6,20

All of the antiarrhythmic agents except sotalol had point estimates of conversion rates consistent with efficacy compared with control treatment, though many were not statistically significant. The evidence for sotalol was consistent with negative efficacy for conversion of AF.

The results of the mathematical pooling of these 25 trials are shown in Table 3. The strongest evidence of efficacy of conversion of AF compared with control treatment existed for ibutilide/dofetilide (OR=29.1; 95% CI, 9.8-86.1)^38-40 and flecainide (OR=24.7; 95% CI, 9.0-68.3).^20-23 The range of estimated NNT to have one more subject convert relative to control treatment is 1.5 to 2.0 for both ibutilide/dofetilide and flecainide.

With respect to propafenone there was some modest quantitative heterogeneity of the data for conversion of AF presumably related to issues regarding type and duration of AF. Since we were unable to definitively clarify these issues, we felt a more conservative random-effects model was appropriate for this meta-analysis since that type of modeling assumes variability in the estimated population treatment effects between the studies. Thus, although the magnitude of treatment effect compared with control treatment was less for propafenone (OR=4.6; 95% CI, 2.6-8.2)^16,20,24-33 than for ibutilide/dofetilide or flecainide, the results gave strong evidence of propafenone efficacy for conversion of AF. The estimated range of NNT to have one more subject convert relative to control treatment is 2.0 to 4.5.

We analyzed the impact of the 5 trials with exceptionally high spontaneous conversion rates for AF, 3 of which involved propafenone. Exclusion of these 3 trials^16,29,32 did not substantially alter the pooled treatment effect of the remaining 9 trials (OR=6.6; 95% CI, 3.6-12.0).

The data on quinidine (OR=2.9; 95% CI, 1.2-7.0)^16-18 were consistent with moderate evidence of efficacy for conversion of AF. The summary data for quinidine versus control treatment remained consistent with moderate evidence of efficacy for conversion of AF (OR=7.2; 955 CI, 1.7-30.4) when we performed outlier analysis by excluding the trial by Capucci and colleagues¹⁶ that had a high spontaneous conversion rate.

Comparable with the situation with propafenone, the data on amiodarone had modest quantitative heterogeneity, likely because of issues regarding type and duration of AF and prevalence of coronary artery disease. Given this, we again chose to perform more conservative random-effects modeling for this data synthesis. As such, the data on amiodarone (OR=5.7; 95% CI, 1.0-33.4)^34-36 were consistent with suggestive evidence of efficacy for conversion of AF compared with control treatment. Outlier analysis involving exclusion of 2 trials with high spontaneous conversion rates^35,36 left only one small trial³⁴ as evidence of amiodarone efficacy versus control treatment for conversion of AF. This trial had a sample size of only 24 subjects with resultant extremely wide CIs that made interpretation of this data difficult (OR=69.0; 95% CI, 3.2-1500.0).

The summary data for both disopyramide and sotalol each reflected only one relatively small trial. For disopyramide the data (OR=7.0; 95% CI, 0.3-153.0)¹⁹ were consistent with suggestive evidence of efficacy compared with control treatment. For sotalol the data (OR=0.4; CI, 0.0-3.0)³⁷ were consistent with suggestive evidence of negative efficacy compared with control treatment.

As part of the overall project evaluating management of atrial fibrillation by the Johns Hopkins Evidence-Based Practice Center, we also reviewed the data on 8 trials that had direct comparisons between the major antiarrhythmic agents for conversion of AF.¹⁵ Because of the overall paucity of data on these direct comparisons, mathematical data pooling was not feasible. The one trial evaluating procainamide compared with flecainide reported lower conversion rates for procainamide. In general, these results were consistent with our meta-analysis results.

Quantitative Synthesis: Evidence of Pharmacologic MSR

Figure 2 shows the scatter plot of absolute rates for MSR of the identified trials. Two trials reported their results in a manner not conducive for our data extraction.^17,48 The results of these 2 trials are included in Table 2. Two other trials involved 2 pharmacologic arms compared with one control treatment arm, resulting in 15 data points on Figure 2.^41,49 Notably, none of these trials examined the efficacy of amiodarone or procainamide compared with control treatment for MSR.

All of the major antiarrhythmic agents had evidence of efficacy for MSR compared with control treatment, although some were not statistically significant.

The results of mathematical data pooling for MSR are shown in Table 4. All of the antiarrhythmic agents had strong and relatively comparable evidence of efficacy compared with control treatment, and the point estimates were all consistent with fairly large treatment effect sizes: quinidine (OR=4.1; 95% CI, 2.5-6.7)^18,41-43; disopyramide (OR=3.4; 95% CI, 1.6-7.1)^44-45; flecainide (OR=3.1; 95% CI, 1.5-6.2)^46-48; propafenone (OR=3.7; 95% CI, 2.4-5.7)^27,49-51; and sotalol (OR=7.1; 95% CI, 3.8-13.4).^37,49

The estimated range of NNT to have one less subject experience AF recurrence relative to control treatment is as follows: quinidine 2.3 to 4.6, disopyramide 2.2 to 9.4, flecainide 2.3 to 10.9, propafenone 2.4 to 4.8, and sotalol 1.8 to 3.1.

Although we identified no clinical trials comparing amiodarone with a control treatment, 2 trials did compare amiodarone to other antiarrhythmic agents (Table 2) and should at least be noted given the overall paucity of data on amiodarone for MSR. One small trial compared amiodarone with quinidine (OR=1.1; 95% CI, 0.1-20.0) and was inconclusive. However, a second trial⁶⁵ compared amiodarone with disopyramide (OR=3.2; 95% CI, 1.0-9.6) and was consistent with moderate evidence of amiodarone efficacy compared with disopyramide for MSR. This study reported only interim results, and our searches did not identify the final results of the trial. One could infer from this study that there is indirect strong evidence of amiodarone efficacy for MSR compared with control treatment, since disopyramide had strong evidence of efficacy compared with control treatment.

As another part of the project evaluating management of atrial fibrillation by the Johns Hopkins Evidence-Based Practice Center, we reviewed the data on 10 trials that had direct comparisons between the major antiarrhythmic agents regarding MSR in AF.¹⁵ Because of the overall paucity of data on these direct comparisons, mathematical data pooling was not feasible and definitive ranking of the agents for MSR efficacy was not possible. Overall, these results were consistent with our meta-analysis showing no one agent as clearly superior over other agents.

Evidence on Adverse Events

During our data extraction we only noted where trials specifically mentioned various events such as ventricular arrhythmias or other nontransient arrhythmias (Table 5). We did not perform formal data synthesis regarding adverse events because the data were too sporadically reported.

In addition, caution must be used in interpreting rates of adverse events that resulted in study withdrawal or dosage decreases, since there was no uniformity regarding the indications for withdrawals of dosage decreases among the studies. Also with respect to conversion trials, many studies involved one-time study drug administration that limited the applicability of this adverse event definition.

Discussion

Pharmacologic conversion of AF is frequently the therapy of choice compared with electrical cardioversion, especially in cases of short-duration AF, significant anesthesia risk, or recent postprandial status of a patient. Little guidance based on scientific evidence has existed regarding the best pharmacologic agents to achieve conversion of AF. On the basis of this formal data review, we are unable to state definitively the relative efficacy of the agents compared with each other because of the inability to ensure comparable subjects within the control treatment groups for the evaluated trials. However, this data synthesis did find that the strongest evidence of efficacy compared with control treatment for conversion of AF existed for ibutilide/dofetilide and flecainide. Less strong but still conclusive evidence existed for propafenone. Quinidine had moderate evidence of efficacy, while only suggestive evidence of efficacy existed for disopyramide and amiodarone. Finally, sotalol had suggestive evidence of negative efficacy compared with control treatment for conversion of AF. Notably, there was no randomized trial on the use of procainamide compared with control treatment for conversion of AF.

The clinical implications of these findings need to be viewed in the light of previous reports regarding adverse events, since our ability to synthesize the adverse event data from these trials was limited.

Ibutilide and dofetilide are new class III antiarrhythmic agents currently undergoing extensive clinical trials. Although limited primarily to clinical trial data, our data and other reports conclude that these drugs have a rate of ventricular arrhythmias (particularly torsade de pointes) of 3% to 9%.⁵² However, there were no reported deaths or prolonged resuscitations among the trials examined.^38-40 Data from long-term use in everyday clinical practice evaluating these agents in less controlled circumstances are not available.

There have been reports of increased mortality with flecainide, although this occurred for prevention of ventricular ectopic activity in subjects with coronary artery disease in the Cardiac Arrhythmia Suppression Trial.⁵³ However, patients with atrial fibrillation may frequently also have ventricular ectopic activity and coronary artery disease. A recent review of flecainide safety for treatment of supraventricular arrhythmias using both randomized clinical trials and uncontrolled trials concluded that the risk of clinically significant adverse cardiac effects was small but not negligible.⁵⁴ From 1794 reviewed treatment courses 2% had atrial proarrhythmic events with some requiring urgent electrical cardioversion because of hemodynamic compromise, and 2% had pre-excitation worsening or new ventricular arrhythmias including 9 cases of sustained ventricular tachycardia or fibrillation and 4 cases of sudden cardiac death. Another report retrospectively compared the mortality rates of patients with atrial arrhythmias in completed pharmaceutical company–sponsored trials treated with flecainide with a population seen at the research arrhythmia clinic.⁵⁵ The researchers concluded that there appeared to be no excess mortality in patients treated with flecainide for supraventricular arrhythmias. If the main concern among patients with atrial fibrillation is coronary artery disease and resultant ventricular dysfunction, our data synthesis was unable to address this because of poor documentation of definitions regarding presence of coronary artery disease, presence of abnormal left ventricular function, and lack of result stratification by these conditions.

Since these 2 agents (ibutilide/dofetilide and flecainide) had the largest treatment effect sizes for conversion of AF, additional research directly comparing them, comparing them with electrical cardioversion, and better quantifying adverse event rates stratified by the presence of coronary artery disease, structural heart disease, left ventricular hypertrophy, and long QT intervals would help solidify their efficacy and safety for conversion of AF.

Similarly, more research on the efficacy of amiodarone is warranted given the paucity of data, a general perception of relatively minor side effects, and a high prevalence of clinical use for AF.

Pharmacologic MSR for AF is a therapeutic option for patients with high recurrence rates and patients with symptomatic AF. Comparable with conversion of AF therapy, no consensus exists on the best pharmacologic agents to achieve MSR in AF. Our formal data synthesis was unable to show definitively the relative efficacy of the agents for MSR compared with each other because of the inability to ensure comparable subjects within the control treatment groups for the evaluated trials. However, this data synthesis did find strong and comparable efficacy evidence for quinidine, disopyramide, flecainide, propafenone, and sotalol. Notably, the data for amiodarone use for MSR are sparse with no trials comparing amiodarone with control treatment, and no trial evaluated procainamide either compared with control treatment or another agent.

The clinical implications of these data also need to be viewed in light of previous reports regarding adverse events, since our ability to synthesize the adverse event data was limited. The issues regarding flecainide have already been discussed. The Class Ia agents quinidine, disopyramide, and procainamide have classically been associated with torsade de pointes because of their prolongation of the QT interval, but cases of torsade de pointes have also been reported with propafenone, flecainide, amiodarone, and sotalol. The reported risk factors for proarrhythmic events with each of the agents vary from hypokalemia and bradycardia for quinidine to serum concentration for sotalol. A recent review concluded that all of the antiarrhythmic agents have potential for uncommon but serious proarrhythmic effects.⁵⁶ Unfortunately, this does not help the clinician sort through all of the available agents.

More research involving direct comparisons between all these agents for MSR in AF would help to definitively rank the efficacy of the agents and to compare their adverse event profiles. Stratification of patients on the basis of the presence of coronary artery disease, structural heart disease, left ventricular hypertrophy, and long QT intervals would permit better assessment of adverse event risks. Both the ongoing AF Follow-up Investigation of Rhythm Management (AFFIRM)⁵⁷ sponsored by the National Heart, Lung, and Blood Institute and the ongoing Prognosis in Afib (PAIF)⁵⁸ study may help provide more information directly comparing agents for MSR.

Limitations

Overall with respect to our data synthesis for both conversion of AF and MSR, we cannot exclude a publication bias despite our best efforts to minimize this known limitation of evidence reviews.

In terms of the actual trials reviewed, we do not believe that subject-specific factors significantly influenced the accumulated evidence based on examination of the inclusion/exclusion criteria and baseline subject characteristics of all the reviewed trials. However, 4 points about this should be noted. First, the age range of the subjects in these trials was somewhat younger than might be seen in a population-based sample of AF. Since it is possible that response to pharmacologic therapy may differ with age, this needs to be kept in mind. Second, our target population consisted of nonpostoperative AF. The accumulated evidence, therefore, may not be applicable to subjects with postoperative AF. In addition, it is difficult to assure the generalizability of our results based on randomized clinical trials to everyday clinical practice. Third, given the relatively small number of trials for any given comparison, we were unable to perform sensitivity analysis on estimated treatment effects on the basis of our assessments of study quality. Finally, our results regarding quinidine may partially reflect time-dependent improvements in medical care. The majority of trials evaluating quinidine were older. However, for both conversion and MSR at least one trial of quinidine was contemporary, and in both conditions found quinidine less efficacious than the older trials.

It is important to note areas of missing evidence that limit more definitive statements for selection of antiarrhythmic agents for management of AF. First, there are few direct comparisons between antiarrhythmic agents for either conversion of AF or MSR. Since control treatment groups vary between trials, direct comparisons between antiarrhythmic agents are instrumental in assessing relative efficacy. Second, there are particularly sparse data for amiodarone and procainamide, especially with respect to MSR. Although several published reviews^6,59 report efficacy of these agents for conversion of AF or MSR, our data from randomized clinical trials (particularly for MSR) do not support this. The AFFIRM and PIAF trials may help address this issue. Third, almost no data were found in this review for the effects of the various antiarrhythmic agents on quality of life. Since patient experiences may significantly influence treatment compliance, quality of life effects need to be better defined. Finally, the follow-up times for all trials on MSR were relatively short. Since the ability to remain free of recurrence has an impact on a patient’s preference for continuing therapy, it would be informative to test the antiarrhythmic agents over a longer period of time for efficacy. These last 2 points may also be addressed in the AFFIRM trial.

Conclusions

Our formal data synthesis of 36 randomized clinical trials of pharmacologic AF conversion and MSR found evidence consistent with superior efficacy relative to control treatment for AF conversion with ibutilide/dofetilide and flecainide. The strength of evidence for MSR relative to control treatment was strong and comparable for quinidine, disopyramide, flecainide, propafenone, and sotalol. Most important, despite the high prevalence of AF the data for the relative efficacy of the antiarrhythmic agents for both conversion and MSR are sparse and inconclusive. Defining these relative efficacies should be a research priority.

Recommendations for clinical practice

On the basis of data from randomized clinical trials, ibutilide, dofetilide, and flecainide have superior efficacy for conversion of AF. However, the data are sparse for ibutilide and dofetilide, and use of flecainide needs to be considered in the context of other comorbidities, such as ventricular ectopy and coronary artery disease. For maintenance of sinus rhythm, no one agent has been shown to have superior efficacy. Clinical practices need to focus on upcoming trial results that involve direct comparisons among agents to better understand relative efficacies of the antiarrhythmic agents for both aspects of AF management.

Acknowledgments

Our study was conducted by the Johns Hopkins Evidence-Based Practice Center through contract No. 290-97-0006 from the Agency for Health Care Policy and Research, Rockville, Maryland. We are responsible for its contents including any clinical or treatment recommendations. No statement in this article should be construed as an official position of the Agency for Healthcare Research and Quality or the United States Department of Health and Human Services. Dr Miller was supported by the Hayden Whitney Smith Research Scholarship. We thank Hanan S. Bell, PhD; Ronald D. Berger, MD; Gary Gerstenblith, MD; David E. Haines, MD; Michael L. Lefevre, MD, MSPH; Andrew Epstein, MD; John A. Kastor, MD; Chris Burton, MD; Jerome A. Osheroff, MD; Barbara J. Drew, RN, PhD; and Kathleen McCauley, RN, PhD, for their assistance as expert advisers for this study. We also thank David Yu, MD, and Paul Abboud for their assistance with this study.

We are especially grateful to Donna Lea for her secretarial support.

Clinical question

Which antiarrhythmic agents are efficacious for conversion of nonpostoperative atrial fibrillation and for subsequent maintenance of sinus rhythm?

Atrial fibrillation (AF) is the most common sustained tachyarrhythmia faced by all physicians. The prevalence of AF, estimated at 0.4% in the general population,¹ increases with age to almost 10% among those aged 80 to 89 years.^2,3 The age-adjusted incidence of AF has increased over the last 30 years.⁴ AF accounts for more days of hospitalization for either acute hemodynamic compromise or treatment of the arrhythmia than all ventricular arrhythmias combined.⁵ All admissions for the complications of stroke and chronic heart failure are not reflected in these data. Overall, patients with AF have twice the mortality of a control population without AF and an attributable risk of stroke of 24% in those aged 80 to 89 years.²

One of the most important issues for management of AF is the need for conversion to sinus rhythm and subsequent maintenance of sinus rhythm (MSR), particularly for symptomatic patients. Although conversion can be accomplished by electrical cardioversion, it is frequently accomplished with pharmacologic agents because of patient or physician preference and anesthesia risks. These agents may also be used for subsequent MSR. In addition to the numerous relatively new or investigational agents such as ibutilide and dofetilide there are at least 7 agents commonly used for either conversion or MSR: quinidine, disopyramide, procainamide, flecainide, propafenone, amiodarone, and sotalol.⁶ This plethora of antiarrhythmic agents for either conversion of AF or MSR makes it difficult for physicians to know which are best for their patients. We reviewed the evidence on pharmacologic management of AF as part of the Johns Hopkins Evidence-Based Practice Center sponsored by the Agency for Health Care Policy and Research (now the Agency for Healthcare Research and Quality).

Methods

Search Strategy

We used the Medical Subject Heading terms “atrial fibrillation,” “atrial flutter,” “random allocation,” “double-blind method,” and “single-blind method.” Publication types of “randomized controlled trial” and “controlled clinical trial” were included. Although the search was not restricted to citations in the English-language literature, subsequent article review involved only English-language publications because of budgetary constraints.

The primary literature source was the CENTRAL database, The Cochrane Library 1998 issues 1 and 2, produced by the Cochrane Collaboration from EMBASE and MEDLINE and encompassing 1948 through the present.^7,8 Second, MEDLINE was searched using both OVID and PubMed from 1966 to May 1998. Third, we used the PubMed feature of “related articles” for primary articles identified in the CENTRAL database. Fourth, a review of recent hand search results submitted to the Baltimore Cochrane Center from the Cardiovascular Randomized Controlled Trial Registry was used. Finally, to capture newly published studies the core study team scanned the contents of the journals most frequently cited in the search results database.

To address the issue of publication bias we asked investigators in the field and search coordinators of relevant Cochrane Collaborative Review Groups to identify any trials they were aware of that had been completed but not published. We decided that construction of funnel plots was practical because of the relatively small number of trials for any specific pharmacologic agent.

Study Inclusion

Articles had to report original data on pharmacologic management of nonpostoperative AF in adults in the context of a randomized clinical trial to be eligible for inclusion in our review. Pairs of independent investigators reviewed all identified abstracts according to these inclusion criteria. All discrepancies about inclusion were resolved by consensus.

Study Quality Assessment

The Evidence-Based Practice Center team developed a data form for extracting information on study quality based on a review of forms used in other meta-analytic studies by study investigators,^9-11 a literature review of the topic,^12,13 and with the assistance of the Cochrane Collaboration. The form contained 22 questions assessing study quality in 5 areas: clarity of description of the study population; potential for bias and confounding; description of therapy, outcomes and follow-up; and statistical quality and interpretation. Each question included a 4 to 5–level subjective ranking of study quality with the resultant score for each of the 5 areas comprising the total points accumulated out of the maximum possible points for all relevant questions in that area. The overall study quality score consisted of the mean score of these 5 areas.

Teams of independent reviewers assessed the quality of each study with differences resolved by consensus. Given the difficult nature of assessing study quality based on article review, the team decided to collectively review and discuss any articles receiving an overall score less than 50% to reach decisions regarding study inclusion.

Data Extraction

Because of the large volume of articles for review, quantitative data were extracted by one reviewer and then checked for accuracy by a second reviewer with consensus resolution of differences. The reviewers were not blinded to the author, institution, and journal, because recent work has indicated that such masking makes little difference in the results.¹⁴ In trials involving both AF and atrial flutter patients, data were only extracted for the AF patients whenever possible.

Data Synthesis and Analysis

Before doing the meta-analysis we first performed both qualitative and quantitative assessments of heterogeneity between the trials to ensure appropriateness of subsequent data combination. The reviewers subjectively assessed qualitative heterogeneity on the basis of similarity between studies on age of subjects, type and duration of AF, comorbidities, therapeutic regimens, and follow-up times. We performed quantitative analysis of heterogeneity using the statistical test of data heterogeneity included in Review Manager (RevMan) version 3.1 (Cochrane Collaboration, Oxford, England).

For data synthesis we defined control treatment to include placebo, verapamil, diltiazem, or digoxin. An analysis of identified trials evaluating verapamil, diltiazem, or digoxin compared with placebo supported this definition, since all of these agents were found to have no efficacy compared with placebo for either conversion or MSR.¹⁵ We also combined treatment arms within a given study that used the same antiarrhythmic agent at different dosages. Analysis of these arms individually supported their consideration as one arm.¹⁵ When life table analysis was used, we extracted the resultant cumulative percentages of successful outcomes and applied them to the initial overall subject number in each trial arm to derive a proportion for meta-analysis inclusion.

We constructed evidence tables to present the data separately for the 2 main outcomes of conversion of AF and MSR and created scatter plots of the absolute rates of conversion and MSR.

For meta-analysis the primary effect measure chosen was the odds ratio (OR) with studies weighted based on the precision of the estimate within each study. A fixed-effects model was used. In cases of significant quantitative data heterogeneity, we explored the etiology of the heterogeneity and used random-effects modeling when appropriate.

We chose the following categorization of strength of evidence by noting the placement of the point estimate of the OR and the width of the confidence interval (CI) surrounding it: (1) strong evidence of efficacy: OR >1.0, 99% CI does not include 1.0 (P <.01); (2) moderate evidence of efficacy: OR >1.0, 95% CI does not include 1.0, but 99% CI includes 1.0 (.01 P .05); (3) suggestive evidence of efficacy: 95% CI includes 1.0 in the lower tail (.05< P <.25), and the OR is in a clinically meaningful range; (4) inconclusive evidence of efficacy: 95% CI is widely distributed around 1.0; and (5) strong evidence of lack of efficacy: OR near 1.0, 95% CI is narrow and does not include a clinically meaningful difference from an OR of 1.0. When the point estimate was less than 1.0, we called this negative efficacy and used the same categorization of strong, moderate, and suggestive evidence on the basis of the point estimate OR and CI. For clarity our reported CIs are at the 95% level.

We also estimated the number needed to treat (NNT) from the resultant OR. The NNT provides an estimate of the number of subjects needed to treat with a therapy to have one more subject experience a desired outcome relative to the comparison group. To do these calculations for the conversion data we assumed a 30% spontaneous conversion rate for the control treatment group, which was consistent with the data. Similarly, to calculate the NNT for MSR we assumed a 30% recurrence rate of AF by 6 months in the control treatment group, which was also consistent with the data. The upper and lower 95% CI estimates for each OR were used to estimate the NNT.

All analyses were completed using RevMan.

Results

Search Strategy and Study Inclusion

Our review of 521 abstracts identified 130 articles for review.¹⁵ After article review, 36 studies^16-51 were eligible for inclusion in our meta-analysis,²⁵ relevant to the conversion of AF outcome and 15 to MSR outcome. All 36 studies used control treatment comparison groups. Our inquiry of experts did not identify any trials for inclusion that had been completed but not published.

In addition to these 36 trials our search identified 16 trials involving unique comparisons between antiarrhythmic agents that precluded meta-analysis. For completeness the results of these trials are discussed and the data presented in Tables 1E and 2E.* We also identified 15 trials using new or uncommon agents. Discussion of these results was published previously.¹⁵

Study Quality Assessment

Based on study quality scores we concluded that all 36 identified trials were of sufficient quality for inclusion. The overall quality scores ranged from 36% to 84% with only 2 studies^17,22 having an overall score less than 50%. Team review of these articles deemed them acceptable for inclusion. Details on the study quality scores was published previously.¹⁵

Study Characteristics and Qualitative Synthesis

Table 1 and Table 2 show important design elements and results of the trials (ie, subject characteristics, sample size, treatment regimens, follow-up times, and reported treatment effects).

The important subject characteristics reported involve age, type of AF, and duration of AF. These areas have an impact on the responsiveness to conversion and ability to avoid recurrent AF.

The mean ages for the trials were generally comparable, ranging between 47 and 71 years with only 7 trials (19%) having mean ages greater than 65 years.

Overall, the studies provided sparse and varying terminology regarding the type and duration of AF; because of this we were unable to reliably segregate studies accordingly. Thus, we relied on the verbatim descriptive terminology used by each study with the understanding that this represented differing definitions between the studies. This difficulty in assessing the type of AF was primarily relevant to conversion studies involving propafenone, amiodarone, and quinidine.^{16,29,32,35,36} These 5 trials all reported control treatment conversion rates greater than 70%, suggesting that the enrolled subjects had predominantly paroxysmal AF. To examine the potential effect of this, we evaluated the quantitative change in the meta-analysis data when excluding these 5 articles with high outlier spontaneous conversion rates.

The therapeutic regimens were generally comparable for any given antiarrhythmic agent for both conversion of AF and MSR. Notably, of the 12 trials evaluating propafenone for conversion of AF, half used oral regimens, and half used intravenous regimens. Separate quantitative analysis comparing these routes showed no significant differences in treatment effects.¹⁵

Regarding follow-up times, the 25 trials of conversion of AF were all comparable and were typically less than 24 hours. There was variability in follow-up time among the 15 trials involving MSR, with a range of 1 to 15 months. For any given antiarrhythmic agent, there was at least one trial with a minimum of 6 months follow-up time.

Overall, our subjective qualitative synthesis of the 36 trials regarding trial inclusion/exclusion criteria, trial size, subject age, subject sex, comorbidities, therapeutic regimens, follow-up times, and reported treatment effects suggested that quantitative synthesis was reasonable because of relatively minor qualitative differences among the studies.

Quantitative Synthesis: Evidence on Pharmacologic Conversion of AF Figure 1 shows the scatter plot of absolute conversion rates for these 25 studies. Two trials involved 2 antiarrhythmic agent arms compared with a third placebo arm thus providing 27 data points.^6,20

All of the antiarrhythmic agents except sotalol had point estimates of conversion rates consistent with efficacy compared with control treatment, though many were not statistically significant. The evidence for sotalol was consistent with negative efficacy for conversion of AF.

The results of the mathematical pooling of these 25 trials are shown in Table 3. The strongest evidence of efficacy of conversion of AF compared with control treatment existed for ibutilide/dofetilide (OR=29.1; 95% CI, 9.8-86.1)^38-40 and flecainide (OR=24.7; 95% CI, 9.0-68.3).^20-23 The range of estimated NNT to have one more subject convert relative to control treatment is 1.5 to 2.0 for both ibutilide/dofetilide and flecainide.

With respect to propafenone there was some modest quantitative heterogeneity of the data for conversion of AF presumably related to issues regarding type and duration of AF. Since we were unable to definitively clarify these issues, we felt a more conservative random-effects model was appropriate for this meta-analysis since that type of modeling assumes variability in the estimated population treatment effects between the studies. Thus, although the magnitude of treatment effect compared with control treatment was less for propafenone (OR=4.6; 95% CI, 2.6-8.2)^16,20,24-33 than for ibutilide/dofetilide or flecainide, the results gave strong evidence of propafenone efficacy for conversion of AF. The estimated range of NNT to have one more subject convert relative to control treatment is 2.0 to 4.5.

We analyzed the impact of the 5 trials with exceptionally high spontaneous conversion rates for AF, 3 of which involved propafenone. Exclusion of these 3 trials^16,29,32 did not substantially alter the pooled treatment effect of the remaining 9 trials (OR=6.6; 95% CI, 3.6-12.0).

The data on quinidine (OR=2.9; 95% CI, 1.2-7.0)^16-18 were consistent with moderate evidence of efficacy for conversion of AF. The summary data for quinidine versus control treatment remained consistent with moderate evidence of efficacy for conversion of AF (OR=7.2; 955 CI, 1.7-30.4) when we performed outlier analysis by excluding the trial by Capucci and colleagues¹⁶ that had a high spontaneous conversion rate.

Comparable with the situation with propafenone, the data on amiodarone had modest quantitative heterogeneity, likely because of issues regarding type and duration of AF and prevalence of coronary artery disease. Given this, we again chose to perform more conservative random-effects modeling for this data synthesis. As such, the data on amiodarone (OR=5.7; 95% CI, 1.0-33.4)^34-36 were consistent with suggestive evidence of efficacy for conversion of AF compared with control treatment. Outlier analysis involving exclusion of 2 trials with high spontaneous conversion rates^35,36 left only one small trial³⁴ as evidence of amiodarone efficacy versus control treatment for conversion of AF. This trial had a sample size of only 24 subjects with resultant extremely wide CIs that made interpretation of this data difficult (OR=69.0; 95% CI, 3.2-1500.0).

The summary data for both disopyramide and sotalol each reflected only one relatively small trial. For disopyramide the data (OR=7.0; 95% CI, 0.3-153.0)¹⁹ were consistent with suggestive evidence of efficacy compared with control treatment. For sotalol the data (OR=0.4; CI, 0.0-3.0)³⁷ were consistent with suggestive evidence of negative efficacy compared with control treatment.

As part of the overall project evaluating management of atrial fibrillation by the Johns Hopkins Evidence-Based Practice Center, we also reviewed the data on 8 trials that had direct comparisons between the major antiarrhythmic agents for conversion of AF.¹⁵ Because of the overall paucity of data on these direct comparisons, mathematical data pooling was not feasible. The one trial evaluating procainamide compared with flecainide reported lower conversion rates for procainamide. In general, these results were consistent with our meta-analysis results.

Quantitative Synthesis: Evidence of Pharmacologic MSR

Figure 2 shows the scatter plot of absolute rates for MSR of the identified trials. Two trials reported their results in a manner not conducive for our data extraction.^17,48 The results of these 2 trials are included in Table 2. Two other trials involved 2 pharmacologic arms compared with one control treatment arm, resulting in 15 data points on Figure 2.^41,49 Notably, none of these trials examined the efficacy of amiodarone or procainamide compared with control treatment for MSR.

All of the major antiarrhythmic agents had evidence of efficacy for MSR compared with control treatment, although some were not statistically significant.

The results of mathematical data pooling for MSR are shown in Table 4. All of the antiarrhythmic agents had strong and relatively comparable evidence of efficacy compared with control treatment, and the point estimates were all consistent with fairly large treatment effect sizes: quinidine (OR=4.1; 95% CI, 2.5-6.7)^18,41-43; disopyramide (OR=3.4; 95% CI, 1.6-7.1)^44-45; flecainide (OR=3.1; 95% CI, 1.5-6.2)^46-48; propafenone (OR=3.7; 95% CI, 2.4-5.7)^27,49-51; and sotalol (OR=7.1; 95% CI, 3.8-13.4).^37,49

The estimated range of NNT to have one less subject experience AF recurrence relative to control treatment is as follows: quinidine 2.3 to 4.6, disopyramide 2.2 to 9.4, flecainide 2.3 to 10.9, propafenone 2.4 to 4.8, and sotalol 1.8 to 3.1.

Although we identified no clinical trials comparing amiodarone with a control treatment, 2 trials did compare amiodarone to other antiarrhythmic agents (Table 2) and should at least be noted given the overall paucity of data on amiodarone for MSR. One small trial compared amiodarone with quinidine (OR=1.1; 95% CI, 0.1-20.0) and was inconclusive. However, a second trial⁶⁵ compared amiodarone with disopyramide (OR=3.2; 95% CI, 1.0-9.6) and was consistent with moderate evidence of amiodarone efficacy compared with disopyramide for MSR. This study reported only interim results, and our searches did not identify the final results of the trial. One could infer from this study that there is indirect strong evidence of amiodarone efficacy for MSR compared with control treatment, since disopyramide had strong evidence of efficacy compared with control treatment.

As another part of the project evaluating management of atrial fibrillation by the Johns Hopkins Evidence-Based Practice Center, we reviewed the data on 10 trials that had direct comparisons between the major antiarrhythmic agents regarding MSR in AF.¹⁵ Because of the overall paucity of data on these direct comparisons, mathematical data pooling was not feasible and definitive ranking of the agents for MSR efficacy was not possible. Overall, these results were consistent with our meta-analysis showing no one agent as clearly superior over other agents.

Evidence on Adverse Events

During our data extraction we only noted where trials specifically mentioned various events such as ventricular arrhythmias or other nontransient arrhythmias (Table 5). We did not perform formal data synthesis regarding adverse events because the data were too sporadically reported.

In addition, caution must be used in interpreting rates of adverse events that resulted in study withdrawal or dosage decreases, since there was no uniformity regarding the indications for withdrawals of dosage decreases among the studies. Also with respect to conversion trials, many studies involved one-time study drug administration that limited the applicability of this adverse event definition.

Discussion

Pharmacologic conversion of AF is frequently the therapy of choice compared with electrical cardioversion, especially in cases of short-duration AF, significant anesthesia risk, or recent postprandial status of a patient. Little guidance based on scientific evidence has existed regarding the best pharmacologic agents to achieve conversion of AF. On the basis of this formal data review, we are unable to state definitively the relative efficacy of the agents compared with each other because of the inability to ensure comparable subjects within the control treatment groups for the evaluated trials. However, this data synthesis did find that the strongest evidence of efficacy compared with control treatment for conversion of AF existed for ibutilide/dofetilide and flecainide. Less strong but still conclusive evidence existed for propafenone. Quinidine had moderate evidence of efficacy, while only suggestive evidence of efficacy existed for disopyramide and amiodarone. Finally, sotalol had suggestive evidence of negative efficacy compared with control treatment for conversion of AF. Notably, there was no randomized trial on the use of procainamide compared with control treatment for conversion of AF.

The clinical implications of these findings need to be viewed in the light of previous reports regarding adverse events, since our ability to synthesize the adverse event data from these trials was limited.

Ibutilide and dofetilide are new class III antiarrhythmic agents currently undergoing extensive clinical trials. Although limited primarily to clinical trial data, our data and other reports conclude that these drugs have a rate of ventricular arrhythmias (particularly torsade de pointes) of 3% to 9%.⁵² However, there were no reported deaths or prolonged resuscitations among the trials examined.^38-40 Data from long-term use in everyday clinical practice evaluating these agents in less controlled circumstances are not available.

There have been reports of increased mortality with flecainide, although this occurred for prevention of ventricular ectopic activity in subjects with coronary artery disease in the Cardiac Arrhythmia Suppression Trial.⁵³ However, patients with atrial fibrillation may frequently also have ventricular ectopic activity and coronary artery disease. A recent review of flecainide safety for treatment of supraventricular arrhythmias using both randomized clinical trials and uncontrolled trials concluded that the risk of clinically significant adverse cardiac effects was small but not negligible.⁵⁴ From 1794 reviewed treatment courses 2% had atrial proarrhythmic events with some requiring urgent electrical cardioversion because of hemodynamic compromise, and 2% had pre-excitation worsening or new ventricular arrhythmias including 9 cases of sustained ventricular tachycardia or fibrillation and 4 cases of sudden cardiac death. Another report retrospectively compared the mortality rates of patients with atrial arrhythmias in completed pharmaceutical company–sponsored trials treated with flecainide with a population seen at the research arrhythmia clinic.⁵⁵ The researchers concluded that there appeared to be no excess mortality in patients treated with flecainide for supraventricular arrhythmias. If the main concern among patients with atrial fibrillation is coronary artery disease and resultant ventricular dysfunction, our data synthesis was unable to address this because of poor documentation of definitions regarding presence of coronary artery disease, presence of abnormal left ventricular function, and lack of result stratification by these conditions.

Since these 2 agents (ibutilide/dofetilide and flecainide) had the largest treatment effect sizes for conversion of AF, additional research directly comparing them, comparing them with electrical cardioversion, and better quantifying adverse event rates stratified by the presence of coronary artery disease, structural heart disease, left ventricular hypertrophy, and long QT intervals would help solidify their efficacy and safety for conversion of AF.

Similarly, more research on the efficacy of amiodarone is warranted given the paucity of data, a general perception of relatively minor side effects, and a high prevalence of clinical use for AF.

Pharmacologic MSR for AF is a therapeutic option for patients with high recurrence rates and patients with symptomatic AF. Comparable with conversion of AF therapy, no consensus exists on the best pharmacologic agents to achieve MSR in AF. Our formal data synthesis was unable to show definitively the relative efficacy of the agents for MSR compared with each other because of the inability to ensure comparable subjects within the control treatment groups for the evaluated trials. However, this data synthesis did find strong and comparable efficacy evidence for quinidine, disopyramide, flecainide, propafenone, and sotalol. Notably, the data for amiodarone use for MSR are sparse with no trials comparing amiodarone with control treatment, and no trial evaluated procainamide either compared with control treatment or another agent.

The clinical implications of these data also need to be viewed in light of previous reports regarding adverse events, since our ability to synthesize the adverse event data was limited. The issues regarding flecainide have already been discussed. The Class Ia agents quinidine, disopyramide, and procainamide have classically been associated with torsade de pointes because of their prolongation of the QT interval, but cases of torsade de pointes have also been reported with propafenone, flecainide, amiodarone, and sotalol. The reported risk factors for proarrhythmic events with each of the agents vary from hypokalemia and bradycardia for quinidine to serum concentration for sotalol. A recent review concluded that all of the antiarrhythmic agents have potential for uncommon but serious proarrhythmic effects.⁵⁶ Unfortunately, this does not help the clinician sort through all of the available agents.

More research involving direct comparisons between all these agents for MSR in AF would help to definitively rank the efficacy of the agents and to compare their adverse event profiles. Stratification of patients on the basis of the presence of coronary artery disease, structural heart disease, left ventricular hypertrophy, and long QT intervals would permit better assessment of adverse event risks. Both the ongoing AF Follow-up Investigation of Rhythm Management (AFFIRM)⁵⁷ sponsored by the National Heart, Lung, and Blood Institute and the ongoing Prognosis in Afib (PAIF)⁵⁸ study may help provide more information directly comparing agents for MSR.

Limitations

Overall with respect to our data synthesis for both conversion of AF and MSR, we cannot exclude a publication bias despite our best efforts to minimize this known limitation of evidence reviews.

In terms of the actual trials reviewed, we do not believe that subject-specific factors significantly influenced the accumulated evidence based on examination of the inclusion/exclusion criteria and baseline subject characteristics of all the reviewed trials. However, 4 points about this should be noted. First, the age range of the subjects in these trials was somewhat younger than might be seen in a population-based sample of AF. Since it is possible that response to pharmacologic therapy may differ with age, this needs to be kept in mind. Second, our target population consisted of nonpostoperative AF. The accumulated evidence, therefore, may not be applicable to subjects with postoperative AF. In addition, it is difficult to assure the generalizability of our results based on randomized clinical trials to everyday clinical practice. Third, given the relatively small number of trials for any given comparison, we were unable to perform sensitivity analysis on estimated treatment effects on the basis of our assessments of study quality. Finally, our results regarding quinidine may partially reflect time-dependent improvements in medical care. The majority of trials evaluating quinidine were older. However, for both conversion and MSR at least one trial of quinidine was contemporary, and in both conditions found quinidine less efficacious than the older trials.

It is important to note areas of missing evidence that limit more definitive statements for selection of antiarrhythmic agents for management of AF. First, there are few direct comparisons between antiarrhythmic agents for either conversion of AF or MSR. Since control treatment groups vary between trials, direct comparisons between antiarrhythmic agents are instrumental in assessing relative efficacy. Second, there are particularly sparse data for amiodarone and procainamide, especially with respect to MSR. Although several published reviews^6,59 report efficacy of these agents for conversion of AF or MSR, our data from randomized clinical trials (particularly for MSR) do not support this. The AFFIRM and PIAF trials may help address this issue. Third, almost no data were found in this review for the effects of the various antiarrhythmic agents on quality of life. Since patient experiences may significantly influence treatment compliance, quality of life effects need to be better defined. Finally, the follow-up times for all trials on MSR were relatively short. Since the ability to remain free of recurrence has an impact on a patient’s preference for continuing therapy, it would be informative to test the antiarrhythmic agents over a longer period of time for efficacy. These last 2 points may also be addressed in the AFFIRM trial.

Conclusions

Our formal data synthesis of 36 randomized clinical trials of pharmacologic AF conversion and MSR found evidence consistent with superior efficacy relative to control treatment for AF conversion with ibutilide/dofetilide and flecainide. The strength of evidence for MSR relative to control treatment was strong and comparable for quinidine, disopyramide, flecainide, propafenone, and sotalol. Most important, despite the high prevalence of AF the data for the relative efficacy of the antiarrhythmic agents for both conversion and MSR are sparse and inconclusive. Defining these relative efficacies should be a research priority.

Recommendations for clinical practice

On the basis of data from randomized clinical trials, ibutilide, dofetilide, and flecainide have superior efficacy for conversion of AF. However, the data are sparse for ibutilide and dofetilide, and use of flecainide needs to be considered in the context of other comorbidities, such as ventricular ectopy and coronary artery disease. For maintenance of sinus rhythm, no one agent has been shown to have superior efficacy. Clinical practices need to focus on upcoming trial results that involve direct comparisons among agents to better understand relative efficacies of the antiarrhythmic agents for both aspects of AF management.

Acknowledgments

Our study was conducted by the Johns Hopkins Evidence-Based Practice Center through contract No. 290-97-0006 from the Agency for Health Care Policy and Research, Rockville, Maryland. We are responsible for its contents including any clinical or treatment recommendations. No statement in this article should be construed as an official position of the Agency for Healthcare Research and Quality or the United States Department of Health and Human Services. Dr Miller was supported by the Hayden Whitney Smith Research Scholarship. We thank Hanan S. Bell, PhD; Ronald D. Berger, MD; Gary Gerstenblith, MD; David E. Haines, MD; Michael L. Lefevre, MD, MSPH; Andrew Epstein, MD; John A. Kastor, MD; Chris Burton, MD; Jerome A. Osheroff, MD; Barbara J. Drew, RN, PhD; and Kathleen McCauley, RN, PhD, for their assistance as expert advisers for this study. We also thank David Yu, MD, and Paul Abboud for their assistance with this study.

We are especially grateful to Donna Lea for her secretarial support.

Despite efforts by the public health community and individual clinicians, tobacco use remains a significant health problem.¹ After funding a series of research projects to develop more effective intervention methods for use by physicians and dentists, the National Cancer Institute (NCI)² published a monograph highlighting findings in this area. One of their recommendations was that physicians and dentists initiate office-based activities to enhance and support their tobacco cessation messages. This comprehensive organized effort would increase patient exposure to consistent environmental cues and facilitate patients’ movement along the stages of the readiness to change mode^1,3 resulting in additional cessation attempts and lower smoking rates. Recommended office-based strategies included creating an environment that encouraged cessation, training clinicians in cessation skills, using nurses and other staff in identification and counseling of smokers, and systematically identifying and tracking smokers. Two years after the publication of that monograph and during the time of our study, the Agency for Health Care Policy and Research (AHCPR)⁴ released a clinical practice guideline on smoking cessation that included recommendations for the use of office-based activities. The primary objective of our study was to describe the extent to which family physician practices have implemented 15 office-based activities (Table) for smoking cessation abstracted from the NCI monograph.

Methods

The study consisted of family physician practices recruited from the membership list of the Nebraska Academy of Family Physicians (NAFP). Approximately 95% of Nebraska family physicians are members of NAFP. Individual physicians on the membership list were grouped according to practice address, resulting in a pool of 209 practices. We stratified these practices as urban or rural based on county population density and sampled them proportionally to ensure a representative sample. To include a sufficient number of practices with a minimum of unnecessary contact, practices were recruited in successive waves. We randomly generated a short list of practices from each density stratum. A letter explaining the general outline of our study was sent randomly to one physician in each practice and followed a week later by a telephone call to that physician. Three recruitment waves (for a total of 155 practices) were conducted in an attempt to recruit 100 practices.

We employed a cross-sectional integrated research design.^5-8 A research nurse collected data in 1995 and 1996 during a 1-day visit to each practice. Data included a practice environment checklist identifying smoking cessation activities. Focused observation of the practice environment⁹ and key informant interviews¹⁰ were used to identify environmental cues for smoking cessation, physician and staff use of tobacco, type and placement of smoking cessation patient education materials, and office staff roles relative to smoking cessation activities. We identified physician attitude and beliefs about smoking cessation through audiotaped semistructured in-depth interviews.¹¹ Medical records were reviewed to identify methods of documentation of smoking status and smoking cessation efforts.

Successful adoption of office-based cessation activities was determined using the summed score of 15 items (Table) abstracted from the NCI monograph (coded as 0 if not implemented and 1 if implemented). We then used the summed scores to identify and describe general findings across practices. Data from the interviews and field notes were analyzed qualitatively using the template organizing style.¹² We generated common themes and compared them with the descriptive statistics.

Results

Ninety-one practices participated in the study (57%). Data from 1 site was lost when the research nurse’s briefcase was stolen. Another site did not allow us to audit charts, so their data were not included. Responding sites did not differ from nonrespondents in terms of physician sex, rural or urban locale, and group versus solo practice. The resultant sample was 43% urban and 30% solo practices (urban vs rural locale: c2=1.14, P=.286).

All of the physicians voiced the belief that smoking poses a significant health problem in their patient population and agreed that they needed to address this problem with their patients. Additionally, the majority (74.4%) felt confident about their smoking cessation counseling skills.

These attitudes, however, did not translate into the use of office-based activities for the majority of practices. The office-based activity score had a mean of 5.93 (standard deviation=2.47) with actual scores ranging from 0 to 13 out of a possible maximum 15. The Figure shows the distribution of these scores.

Sixty-six percent of the sites were either posted as nonsmoking or did not provide receptacles for smoking (Table). All of the sites had either an official (written) policy (51%) or an informal policy restricting staff smoking on the premises. Twenty-eight percent of practices had no physicians or staff who were tobacco users. Only 10% provided waiting room reading materials that were free of smoking advertisements. Twenty-one percent advertised in the waiting or reception area that help was available to stop smoking.

The majority of sites (78%) had patient education materials on smoking cessation, and 52% of these had materials placed so they were directly accessible to patients. More than half of the sites relied on pharmaceutical companies to supply these materials, the majority of which suggested the use of a nicotine replacement system. Almost two thirds of the practices used printed materials as their sole educational avenue, as opposed to including audiotapes and videotapes.

Involvement of support staff in office-based smoking cessation activities was limited. Designating a staff member to maintain patient education materials (24%) was the most common. Only 3 practices used support staff to assess tobacco use by asking about it while taking vital signs; one of these also had a person on site to counsel. In 5 additional practices a support person was involved in tobacco counseling or follow-up with patients attempting to quit.

Most physicians were not able to readily use the patient’s chart as an effective cue for identifying smokers at each visit. Smoking status was documented in 51% of the 1951 medical records reviewed but was seldom documented on the face sheet (13% of all records). In the majority of cases, documentation was generally located in the back of the chart on a health history questionnaire.

Discussion

In a meta-analysis of 39 controlled smoking cessation trials, Kottke and colleagues¹³ identified having the patient receive multiple cessation messages from both physicians and nonphysicians as an important common attribute of successful interventions in medical practice. Hollis and coworkers¹⁴ have shown that nurse involvement in smoking cessation counseling reduces physician burden, makes counseling more likely to occur, and increases cessation rates compared with brief physician advice only. Fiore and colleagues¹⁵ and Robinson and coworkers¹⁶ have reported that adding a question about smoking status to the vital signs portion of the progress note increased the likelihood of smoking-related discussions between patients and their physicians. These are only a few of the multitude of articles similar to those reported in the NCI monograph that support the effectiveness of including staff and support activities in a comprehensive office-based approach rather than solely a physician-based approach to smoking cessation. Most of the activities recommended are simple and do not involve considerable costs or additional staff time.

In our study, however, most physicians did not use office-based activities to support what they did individually, to increase avenues to provide cessation messages, or to create a cessation-friendly environment. Although most of the practices had an office environment that reduced cues to smoke by restricting smoking of patients and staff in the clinic, almost all provided reading materials promoting smoking in their advertisements. Few practices proactively supported the importance of cessation by means of visual cues or information on available help in the waiting room. Most did have easily accessible smoking cessation patient education materials; however, the majority of these were from pharmaceutical companies and were designed to promote the use of nicotine replacement therapy. Although these materials could be helpful in promoting cessation for those patients ready to quit, they are of little use for motivating the 80% to 90% of smokers who are not currently interested in quitting.¹⁷ In a significant majority of these practices, physician time was used for both the most mundane (identification) and the most important (motivation and counseling) aspects of smoking cessation. All of the physicians in our study indicated that smoking was a significant health problem in their patient population and believed it was their responsibility to address cessation with smoking patients. Our knowledge of effective cessation techniques has clearly outpaced these practices’ ability or desire to implement them.

There are a number of strengths that make our study unique. The most important of these is that our results are based on direct observation of activities in 89 practices, almost half of all the family practices in Nebraska. It is very likely that our results reflect actual typical practice in our area. Additionally, our use of a multimethod approach enhances the validity of the results by triangulating data (eg, comparing our key informant information about documentation with audits of the medical record).

Limitations

There are some factors that affect the generalizability of our results. Our sample was composed of Nebraska family practices and may not represent other states that differ in smoking rates, taxes on tobacco, or other factors affecting smoking rates. Our assessment of practices focused primarily on the occurrence of office-based activities, so our office-based activity score does not reflect the activities of the individual target physician in the practice. It is possible that a practice could have a low score and a very proactive physician. However, since office-based activities are designed to support the physician’s message, not eliminate it, this practice would still be missing opportunities to reinforce and support the physician’s ability to provide effective cessation messages. Finally, although the NCI monograph and the AHCPR practice guideline outline the suggested best practice for office-based activities, it may not be one that all physicians embrace. Some may consciously choose to limit these activities in their practice.

Conclusions

The NCI monograph suggests that practices are like patients in their stage of readiness¹⁸ to implement office-based strategies. We speculate that movement of a practice out of the first basic level (where the majority of our practices fell) would require acceptance of the use of office-based systems as the standard of care. Examples would include seeing and hearing their peers use office-based activities effectively, being reimbursed for having these activities in place, training residents in a system that uses office-based activities, and providing help to practices interested in implementing and maintaining activities suited to their needs. Our efforts must now focus on helping practices implement the knowledge we have gained.

Acknowledgments

Our study was supported by a grant from the Nebraska Department of Health and Human Services, Cancer and Smoking Disease Program (96-05B). We wish to express our thanks to Connie Gibb, RN, for her invaluable assistance in data collection, Naomi Lacy, PhD, for her editing, and all of the family physicians in Nebraska who were willing to open their practices to us.

Despite efforts by the public health community and individual clinicians, tobacco use remains a significant health problem.¹ After funding a series of research projects to develop more effective intervention methods for use by physicians and dentists, the National Cancer Institute (NCI)² published a monograph highlighting findings in this area. One of their recommendations was that physicians and dentists initiate office-based activities to enhance and support their tobacco cessation messages. This comprehensive organized effort would increase patient exposure to consistent environmental cues and facilitate patients’ movement along the stages of the readiness to change mode^1,3 resulting in additional cessation attempts and lower smoking rates. Recommended office-based strategies included creating an environment that encouraged cessation, training clinicians in cessation skills, using nurses and other staff in identification and counseling of smokers, and systematically identifying and tracking smokers. Two years after the publication of that monograph and during the time of our study, the Agency for Health Care Policy and Research (AHCPR)⁴ released a clinical practice guideline on smoking cessation that included recommendations for the use of office-based activities. The primary objective of our study was to describe the extent to which family physician practices have implemented 15 office-based activities (Table) for smoking cessation abstracted from the NCI monograph.

Methods

The study consisted of family physician practices recruited from the membership list of the Nebraska Academy of Family Physicians (NAFP). Approximately 95% of Nebraska family physicians are members of NAFP. Individual physicians on the membership list were grouped according to practice address, resulting in a pool of 209 practices. We stratified these practices as urban or rural based on county population density and sampled them proportionally to ensure a representative sample. To include a sufficient number of practices with a minimum of unnecessary contact, practices were recruited in successive waves. We randomly generated a short list of practices from each density stratum. A letter explaining the general outline of our study was sent randomly to one physician in each practice and followed a week later by a telephone call to that physician. Three recruitment waves (for a total of 155 practices) were conducted in an attempt to recruit 100 practices.

We employed a cross-sectional integrated research design.^5-8 A research nurse collected data in 1995 and 1996 during a 1-day visit to each practice. Data included a practice environment checklist identifying smoking cessation activities. Focused observation of the practice environment⁹ and key informant interviews¹⁰ were used to identify environmental cues for smoking cessation, physician and staff use of tobacco, type and placement of smoking cessation patient education materials, and office staff roles relative to smoking cessation activities. We identified physician attitude and beliefs about smoking cessation through audiotaped semistructured in-depth interviews.¹¹ Medical records were reviewed to identify methods of documentation of smoking status and smoking cessation efforts.

Successful adoption of office-based cessation activities was determined using the summed score of 15 items (Table) abstracted from the NCI monograph (coded as 0 if not implemented and 1 if implemented). We then used the summed scores to identify and describe general findings across practices. Data from the interviews and field notes were analyzed qualitatively using the template organizing style.¹² We generated common themes and compared them with the descriptive statistics.

Results

Ninety-one practices participated in the study (57%). Data from 1 site was lost when the research nurse’s briefcase was stolen. Another site did not allow us to audit charts, so their data were not included. Responding sites did not differ from nonrespondents in terms of physician sex, rural or urban locale, and group versus solo practice. The resultant sample was 43% urban and 30% solo practices (urban vs rural locale: c2=1.14, P=.286).

All of the physicians voiced the belief that smoking poses a significant health problem in their patient population and agreed that they needed to address this problem with their patients. Additionally, the majority (74.4%) felt confident about their smoking cessation counseling skills.

These attitudes, however, did not translate into the use of office-based activities for the majority of practices. The office-based activity score had a mean of 5.93 (standard deviation=2.47) with actual scores ranging from 0 to 13 out of a possible maximum 15. The Figure shows the distribution of these scores.

Sixty-six percent of the sites were either posted as nonsmoking or did not provide receptacles for smoking (Table). All of the sites had either an official (written) policy (51%) or an informal policy restricting staff smoking on the premises. Twenty-eight percent of practices had no physicians or staff who were tobacco users. Only 10% provided waiting room reading materials that were free of smoking advertisements. Twenty-one percent advertised in the waiting or reception area that help was available to stop smoking.

The majority of sites (78%) had patient education materials on smoking cessation, and 52% of these had materials placed so they were directly accessible to patients. More than half of the sites relied on pharmaceutical companies to supply these materials, the majority of which suggested the use of a nicotine replacement system. Almost two thirds of the practices used printed materials as their sole educational avenue, as opposed to including audiotapes and videotapes.

Involvement of support staff in office-based smoking cessation activities was limited. Designating a staff member to maintain patient education materials (24%) was the most common. Only 3 practices used support staff to assess tobacco use by asking about it while taking vital signs; one of these also had a person on site to counsel. In 5 additional practices a support person was involved in tobacco counseling or follow-up with patients attempting to quit.

Most physicians were not able to readily use the patient’s chart as an effective cue for identifying smokers at each visit. Smoking status was documented in 51% of the 1951 medical records reviewed but was seldom documented on the face sheet (13% of all records). In the majority of cases, documentation was generally located in the back of the chart on a health history questionnaire.

Discussion

In a meta-analysis of 39 controlled smoking cessation trials, Kottke and colleagues¹³ identified having the patient receive multiple cessation messages from both physicians and nonphysicians as an important common attribute of successful interventions in medical practice. Hollis and coworkers¹⁴ have shown that nurse involvement in smoking cessation counseling reduces physician burden, makes counseling more likely to occur, and increases cessation rates compared with brief physician advice only. Fiore and colleagues¹⁵ and Robinson and coworkers¹⁶ have reported that adding a question about smoking status to the vital signs portion of the progress note increased the likelihood of smoking-related discussions between patients and their physicians. These are only a few of the multitude of articles similar to those reported in the NCI monograph that support the effectiveness of including staff and support activities in a comprehensive office-based approach rather than solely a physician-based approach to smoking cessation. Most of the activities recommended are simple and do not involve considerable costs or additional staff time.

In our study, however, most physicians did not use office-based activities to support what they did individually, to increase avenues to provide cessation messages, or to create a cessation-friendly environment. Although most of the practices had an office environment that reduced cues to smoke by restricting smoking of patients and staff in the clinic, almost all provided reading materials promoting smoking in their advertisements. Few practices proactively supported the importance of cessation by means of visual cues or information on available help in the waiting room. Most did have easily accessible smoking cessation patient education materials; however, the majority of these were from pharmaceutical companies and were designed to promote the use of nicotine replacement therapy. Although these materials could be helpful in promoting cessation for those patients ready to quit, they are of little use for motivating the 80% to 90% of smokers who are not currently interested in quitting.¹⁷ In a significant majority of these practices, physician time was used for both the most mundane (identification) and the most important (motivation and counseling) aspects of smoking cessation. All of the physicians in our study indicated that smoking was a significant health problem in their patient population and believed it was their responsibility to address cessation with smoking patients. Our knowledge of effective cessation techniques has clearly outpaced these practices’ ability or desire to implement them.

There are a number of strengths that make our study unique. The most important of these is that our results are based on direct observation of activities in 89 practices, almost half of all the family practices in Nebraska. It is very likely that our results reflect actual typical practice in our area. Additionally, our use of a multimethod approach enhances the validity of the results by triangulating data (eg, comparing our key informant information about documentation with audits of the medical record).

Limitations

There are some factors that affect the generalizability of our results. Our sample was composed of Nebraska family practices and may not represent other states that differ in smoking rates, taxes on tobacco, or other factors affecting smoking rates. Our assessment of practices focused primarily on the occurrence of office-based activities, so our office-based activity score does not reflect the activities of the individual target physician in the practice. It is possible that a practice could have a low score and a very proactive physician. However, since office-based activities are designed to support the physician’s message, not eliminate it, this practice would still be missing opportunities to reinforce and support the physician’s ability to provide effective cessation messages. Finally, although the NCI monograph and the AHCPR practice guideline outline the suggested best practice for office-based activities, it may not be one that all physicians embrace. Some may consciously choose to limit these activities in their practice.

Conclusions

The NCI monograph suggests that practices are like patients in their stage of readiness¹⁸ to implement office-based strategies. We speculate that movement of a practice out of the first basic level (where the majority of our practices fell) would require acceptance of the use of office-based systems as the standard of care. Examples would include seeing and hearing their peers use office-based activities effectively, being reimbursed for having these activities in place, training residents in a system that uses office-based activities, and providing help to practices interested in implementing and maintaining activities suited to their needs. Our efforts must now focus on helping practices implement the knowledge we have gained.

Acknowledgments

Our study was supported by a grant from the Nebraska Department of Health and Human Services, Cancer and Smoking Disease Program (96-05B). We wish to express our thanks to Connie Gibb, RN, for her invaluable assistance in data collection, Naomi Lacy, PhD, for her editing, and all of the family physicians in Nebraska who were willing to open their practices to us.

Despite efforts by the public health community and individual clinicians, tobacco use remains a significant health problem.¹ After funding a series of research projects to develop more effective intervention methods for use by physicians and dentists, the National Cancer Institute (NCI)² published a monograph highlighting findings in this area. One of their recommendations was that physicians and dentists initiate office-based activities to enhance and support their tobacco cessation messages. This comprehensive organized effort would increase patient exposure to consistent environmental cues and facilitate patients’ movement along the stages of the readiness to change mode^1,3 resulting in additional cessation attempts and lower smoking rates. Recommended office-based strategies included creating an environment that encouraged cessation, training clinicians in cessation skills, using nurses and other staff in identification and counseling of smokers, and systematically identifying and tracking smokers. Two years after the publication of that monograph and during the time of our study, the Agency for Health Care Policy and Research (AHCPR)⁴ released a clinical practice guideline on smoking cessation that included recommendations for the use of office-based activities. The primary objective of our study was to describe the extent to which family physician practices have implemented 15 office-based activities (Table) for smoking cessation abstracted from the NCI monograph.

Methods

The study consisted of family physician practices recruited from the membership list of the Nebraska Academy of Family Physicians (NAFP). Approximately 95% of Nebraska family physicians are members of NAFP. Individual physicians on the membership list were grouped according to practice address, resulting in a pool of 209 practices. We stratified these practices as urban or rural based on county population density and sampled them proportionally to ensure a representative sample. To include a sufficient number of practices with a minimum of unnecessary contact, practices were recruited in successive waves. We randomly generated a short list of practices from each density stratum. A letter explaining the general outline of our study was sent randomly to one physician in each practice and followed a week later by a telephone call to that physician. Three recruitment waves (for a total of 155 practices) were conducted in an attempt to recruit 100 practices.

We employed a cross-sectional integrated research design.^5-8 A research nurse collected data in 1995 and 1996 during a 1-day visit to each practice. Data included a practice environment checklist identifying smoking cessation activities. Focused observation of the practice environment⁹ and key informant interviews¹⁰ were used to identify environmental cues for smoking cessation, physician and staff use of tobacco, type and placement of smoking cessation patient education materials, and office staff roles relative to smoking cessation activities. We identified physician attitude and beliefs about smoking cessation through audiotaped semistructured in-depth interviews.¹¹ Medical records were reviewed to identify methods of documentation of smoking status and smoking cessation efforts.

Successful adoption of office-based cessation activities was determined using the summed score of 15 items (Table) abstracted from the NCI monograph (coded as 0 if not implemented and 1 if implemented). We then used the summed scores to identify and describe general findings across practices. Data from the interviews and field notes were analyzed qualitatively using the template organizing style.¹² We generated common themes and compared them with the descriptive statistics.

Results

Ninety-one practices participated in the study (57%). Data from 1 site was lost when the research nurse’s briefcase was stolen. Another site did not allow us to audit charts, so their data were not included. Responding sites did not differ from nonrespondents in terms of physician sex, rural or urban locale, and group versus solo practice. The resultant sample was 43% urban and 30% solo practices (urban vs rural locale: c2=1.14, P=.286).

All of the physicians voiced the belief that smoking poses a significant health problem in their patient population and agreed that they needed to address this problem with their patients. Additionally, the majority (74.4%) felt confident about their smoking cessation counseling skills.

These attitudes, however, did not translate into the use of office-based activities for the majority of practices. The office-based activity score had a mean of 5.93 (standard deviation=2.47) with actual scores ranging from 0 to 13 out of a possible maximum 15. The Figure shows the distribution of these scores.

Sixty-six percent of the sites were either posted as nonsmoking or did not provide receptacles for smoking (Table). All of the sites had either an official (written) policy (51%) or an informal policy restricting staff smoking on the premises. Twenty-eight percent of practices had no physicians or staff who were tobacco users. Only 10% provided waiting room reading materials that were free of smoking advertisements. Twenty-one percent advertised in the waiting or reception area that help was available to stop smoking.

The majority of sites (78%) had patient education materials on smoking cessation, and 52% of these had materials placed so they were directly accessible to patients. More than half of the sites relied on pharmaceutical companies to supply these materials, the majority of which suggested the use of a nicotine replacement system. Almost two thirds of the practices used printed materials as their sole educational avenue, as opposed to including audiotapes and videotapes.

Involvement of support staff in office-based smoking cessation activities was limited. Designating a staff member to maintain patient education materials (24%) was the most common. Only 3 practices used support staff to assess tobacco use by asking about it while taking vital signs; one of these also had a person on site to counsel. In 5 additional practices a support person was involved in tobacco counseling or follow-up with patients attempting to quit.

Most physicians were not able to readily use the patient’s chart as an effective cue for identifying smokers at each visit. Smoking status was documented in 51% of the 1951 medical records reviewed but was seldom documented on the face sheet (13% of all records). In the majority of cases, documentation was generally located in the back of the chart on a health history questionnaire.

Discussion

In a meta-analysis of 39 controlled smoking cessation trials, Kottke and colleagues¹³ identified having the patient receive multiple cessation messages from both physicians and nonphysicians as an important common attribute of successful interventions in medical practice. Hollis and coworkers¹⁴ have shown that nurse involvement in smoking cessation counseling reduces physician burden, makes counseling more likely to occur, and increases cessation rates compared with brief physician advice only. Fiore and colleagues¹⁵ and Robinson and coworkers¹⁶ have reported that adding a question about smoking status to the vital signs portion of the progress note increased the likelihood of smoking-related discussions between patients and their physicians. These are only a few of the multitude of articles similar to those reported in the NCI monograph that support the effectiveness of including staff and support activities in a comprehensive office-based approach rather than solely a physician-based approach to smoking cessation. Most of the activities recommended are simple and do not involve considerable costs or additional staff time.

In our study, however, most physicians did not use office-based activities to support what they did individually, to increase avenues to provide cessation messages, or to create a cessation-friendly environment. Although most of the practices had an office environment that reduced cues to smoke by restricting smoking of patients and staff in the clinic, almost all provided reading materials promoting smoking in their advertisements. Few practices proactively supported the importance of cessation by means of visual cues or information on available help in the waiting room. Most did have easily accessible smoking cessation patient education materials; however, the majority of these were from pharmaceutical companies and were designed to promote the use of nicotine replacement therapy. Although these materials could be helpful in promoting cessation for those patients ready to quit, they are of little use for motivating the 80% to 90% of smokers who are not currently interested in quitting.¹⁷ In a significant majority of these practices, physician time was used for both the most mundane (identification) and the most important (motivation and counseling) aspects of smoking cessation. All of the physicians in our study indicated that smoking was a significant health problem in their patient population and believed it was their responsibility to address cessation with smoking patients. Our knowledge of effective cessation techniques has clearly outpaced these practices’ ability or desire to implement them.

There are a number of strengths that make our study unique. The most important of these is that our results are based on direct observation of activities in 89 practices, almost half of all the family practices in Nebraska. It is very likely that our results reflect actual typical practice in our area. Additionally, our use of a multimethod approach enhances the validity of the results by triangulating data (eg, comparing our key informant information about documentation with audits of the medical record).

Limitations

There are some factors that affect the generalizability of our results. Our sample was composed of Nebraska family practices and may not represent other states that differ in smoking rates, taxes on tobacco, or other factors affecting smoking rates. Our assessment of practices focused primarily on the occurrence of office-based activities, so our office-based activity score does not reflect the activities of the individual target physician in the practice. It is possible that a practice could have a low score and a very proactive physician. However, since office-based activities are designed to support the physician’s message, not eliminate it, this practice would still be missing opportunities to reinforce and support the physician’s ability to provide effective cessation messages. Finally, although the NCI monograph and the AHCPR practice guideline outline the suggested best practice for office-based activities, it may not be one that all physicians embrace. Some may consciously choose to limit these activities in their practice.

Conclusions

The NCI monograph suggests that practices are like patients in their stage of readiness¹⁸ to implement office-based strategies. We speculate that movement of a practice out of the first basic level (where the majority of our practices fell) would require acceptance of the use of office-based systems as the standard of care. Examples would include seeing and hearing their peers use office-based activities effectively, being reimbursed for having these activities in place, training residents in a system that uses office-based activities, and providing help to practices interested in implementing and maintaining activities suited to their needs. Our efforts must now focus on helping practices implement the knowledge we have gained.

Acknowledgments

Our study was supported by a grant from the Nebraska Department of Health and Human Services, Cancer and Smoking Disease Program (96-05B). We wish to express our thanks to Connie Gibb, RN, for her invaluable assistance in data collection, Naomi Lacy, PhD, for her editing, and all of the family physicians in Nebraska who were willing to open their practices to us.

The prevalence of domestic abuse in 3 primary care studies ranged from 6% to 22%. In these studies domestic abuse was measured as physical abuse^1-4 or a combination of both physical and verbal abuse.^1,2,4 In these and many other studies, an abusive relationship was associated with psychological problems,^3-14 drug and alcohol abuse,^3,6,7,9-11 and many other different symptoms.^{3,5,7-13,15-19} Detection of abusive relationships in clinical settings is assumed to improve outcomes.^3,20,21 Guidelines for detection and management are well described.²²

Despite its prevalence and importance, a woman’s abusive relationship often goes unnoticed in primary care practice.^1-3,23 We used the results of a comprehensive written health survey of 1526 women aged 19 to 69 from 31 practices across the United States to reemphasize why specific inquiry about an abusive relationship is important and how it might be accomplished in a busy office setting.

Methods

Settings and Patients

We used data from 31 sites participating in a 3-year quality improvement project.²⁴ In 1999, at 2 times separated by 6 months, all sites asked 2 sequential samples of adults aged 19 to 69 years to complete an anonymous health survey. Each site determined when and how the sample would be obtained. A total of 2528 patients responded. The number of responses by practice ranged from 19 to 156, and the median was 64. Of these patients, 1584 were women.

Measures

The Screening Tool for an Abusive Relationship. To develop the single-item screening tool,⁹ primary care physicians and nurses examined several methods for detection of abuse identified in the published literature.^25-27 They developed 2 word-and-picture charts based on previously tested methods.^28,29 Two groups of women who had recently experienced domestic abuse (total N=13) reviewed these charts, and their suggestions were incorporated.

For our investigation, the word-and-picture Relationships Chart was used (Figure 1). To validate the Relationships Chart we administered it to 51 women volunteers in urban and rural domestic abuse support groups. The control group consisted of 48 randomly selected patients in 2 obstetric and gynecology clinics. These women were also asked to complete 41 items about spouse or partner abuse.²⁵ The items were scored on a 5-point scale so the maximum score would be 205 and the minimum score 41.

Women seeking help from a support group because of their current or previous involvement in an abusive relationship scored much worse than women in the control group on both the single-item chart and the multi-item score (P <.001). Based on the chart, when women reported potential abuse “none or a little of the time,” their average score on the multi-item questionnaire was 62; when their response was “some of the time” their average score was 86; and when they reported “often or always” experiencing potential abuse during the previous 4 weeks their average score was 114.

Eighty-eight of the women completed the Relationship Chart a second time within 10 days after the initial administration. The average test-retest correlation was 0.60; on the 5-point scale 88.4% of the responses stayed the same or shifted by only 1 scale point.

Thus, the Relationship Chart had reasonable face and criterion validity and more than met minimum standards for reliability.³⁰

The Patient Health Survey. The Relationships Chart was included as part of a patient health survey called “Improve Your Medical Care.” This survey was used as part of a quality improvement system that has been shown to improve care for different populations.^31,32 However, in our study it was used only to sample patients’ needs and characteristics in the participant practices. The patient health survey contains questions about demographics and diagnoses (8 items), function (6 items), symptoms and bothers (22 items), habits and prevention (13 items), and utilization (4 items). A complete copy of this survey is available on request.

All practices received computer-scored reports for each respondent within 1 week of completion.

Analysis

To compare scores on the Relationships Chart with measures of functional status using the Dartmouth Primary Care Cooperative Information and Research Network (COOP) charts, we used a score of 4 or 5 on a 5-point Likert scale as the threshold for significant limitation in function.³³ For bothersome symptoms we used a response of “often” or “always” in the previous 4 weeks. We used the chi-square test to assess the statistical significance of the comparisons.

We used prevalence ratios to compare the prevalence of abusive relationships for women with and without functional limits and symptoms.

Results

Clinical and Functional Impacts of an Abusive Relationship Ninety-six percent of the women completed the chart (1526/1584). Possible domestic abuse (as assessed by this chart) was reported by 13% (201) of the women visiting these 31 outpatient practices. The range of positive responses was 0% to 50%; the median was 11%.

The Table lists several factors highly associated with an abusive relationship. Women in an abusive relationship were more likely to be young and poor. They more often had multiple problems with function and many bothersome symptoms. They less often engaged in good personal health habits. More of these women smoked cigarettes: 24% vs 15%, P=.002; more of them drank more than 6 alcoholic beverages a week: 12% vs 7%, P=.001). They more often reported that they had been confined to bed in the previous 3 months than women who were not abused.

Despite a higher frequency of clinic use, women in an abusive relationship more often reported problems with access and slightly less often completed necessary preventive actions. However, they had the same number of chronic diagnoses and used the hospital and chronic medications at the same rate as other women (data not shown).

The most common functional limitations reported by all women were pain (24%), physical (21%), and feelings (18%). Figure 2 illustrates the probability of an abusive relationship for women with or without a particular functional limitation. Pain and feelings were most often associated with an abusive relationship. Almost a third of the women who are limited by feelings are at risk for domestic abuse. The prevalence ratio for feelings was 3.0.

The most common symptoms reported were eating/weight (26%), dizziness/fatigue (24%), joint pains (23%), back pains (22%), sleep problems (22%), and headaches (21%). Figure 3 illustrates the probability of an abusive relationship for women with or without a particular symptom. Among the 6 most prevalent symptoms, headache was most strongly associated with domestic abuse (24%). Among women who reported abusive relationships “most or all of the time” the chance of significant problems with feelings, dizziness/fatigue, headache, or sleep was greater than 50% (data not shown).

Identifying Women at Risk

Although a combination of demographic characteristics and clinical problems can be used to identify a group of women at high risk for abuse,³ such an approach makes little practical sense. For example, among women who have inadequate money and significant problems with feelings, 41% (36/89) were likely to have an abusive relationship. However, among the 1025 women with neither financial difficulties nor emotional problems, 64 were at risk for abuse (6%).

Discussion

We found that approximately 13% of women aged 19 to 69 years in 31 practices were identified as being in an abusive relationship by a single written measure. These women had significant psychosocial issues, poor health habits, and many somatic complaints.

Our descriptive study of screening for an abusive relationship in women has 4 important implications for clinical practice:

1. It reaffirms the prevalence of abuse in office practice settings. In these 31 practices, the median rate of possible abuse was 11%.
2. It provides a detailed illustration of the burden of clinical and social illness borne by women in abusive relationships. One other study³ has provided a similar level of detail. Where measures of symptoms were comparable, both studies found similar prevalence ratios for abuse.
3. It reminds clinicians that direct inquiry about possible abuse is likely to be much more effective than using a combination of demographic and clinical factors.^3,15 Clinical intuition used to select an at-risk woman from a group will be less effective than asking her directly about possible abuse.
4. These results add to the growing body of literature that single-item word-and-picture charts that query patients about complex issues are very useful in clinical practice. Such charts have been shown to be valid and reliable for identifying important functional limitations in adults,^28,29 depression,³³ health and social problems in adolescents,³⁴ and spirituality.³⁵ Because they are single-item they are good for screening, can be easily adapted to other languages, and serve as the foundation for deeper inquiry about physician-patient interaction.^33,36 For example, more than one third of the women in these practices felt that their clinician was unaware of important functional limitations or provided poor education about them.

This particular measure for an abusive relationship has face and criterion validity and reasonable reliability. It seems acceptable for use in practice, since 96% of the women who completed a health survey also completed this item. However, it is intentionally not as specific as direct questions about physical abuse such as: “Has your husband, boyfriend, partner/lover hit, kicked, threatened, or otherwise frightened you?”²² The clinicians and women who assisted in the design of the Relationships Chart felt that a screening question would appear less intimidating if it were less specific and did not require a yes or no response. We previously received the same advice from adolescents for inquiring about antisocial behavior and drug abuse.³⁴ The pictures add to the appeal of the Dartmouth COOP Charts but do not seem to influence responses.³⁷

The validation study (comparing response categories to another instrument) and the results illustrated in the Table (comparing functional and symptomatic impacts by response categories) indicate that a woman who says that she has experienced an abusive relationship at least some of the time in the past 4 weeks probably is experiencing abuse. For those women who screen positive, more direct inquiry is indicated about the nature of the abuse.

If clinicians do not actively screen but choose to wait for a woman in an abusive relationship to declare herself, the data suggest that she may enter a labyrinth of nonspecific complaints and increased utilization of services. If clinicians actively inquire about abuse, short-term demands for accurate documentation and effective referral will have to be met,^20-22 and the longer-term benefit to the patient and clinician may be considerable.

Conclusions

Domestic abuse is a prevalent and important problem for women, and simple measures can detect domestic abuse in community practice. It is necessary to think of abuse when some functional and symptomatic issues are present, but that alone is not sufficient. Direct inquiry will be more effective. The symptoms and functional problems of these women may be erroneously evaluated unless the domestic abuse is detected.

Acknowledgments

We thank and acknowledge the physician practice sites that participated in our project: Baylor College of Medicine, Houston, Tex; Berlin Health System, Green Bay, Wis; Beth Israel Deaconess Medical Center, Boston, Mass; Cambridge Health Alliance, Cambridge, Mass; Dana Farber Cancer Institute, Boston, Mass; Federation of Swedish County Councils/Sweden Anderstop Health Center, Greater Lawrence Family Health Center, Lawrence, Mass; Harvard University Health Services, Cambridge, Mass; Independent Health: Buffalo Medical Group, Tonawanda Medical Associates, Buffalo, NY; Joslin Diabetes Center, Boston, Mass; Lathem Medical Group, Lathem, NY; Magic Valley Health Center, Twin Falls, Idaho; Mayo Health System: Luther Midelfort, Eau Clair, Wis, and Mayo Clinic Scottsdale, Ariz; Medical University of South Carolina, Charleston, SC; MeritCare, Fargo, ND; PeaceHealth, Eugene, Ore; PennState Geisinger Health System, Pa; Scripps Clinic, La Jolla, Calif; SSM Health System, St. Louis, Ohio; Dean Health System, Madison, Wis; St. Mary’s Health System, Jefferson City, Mo; Strong Health, University of Rochester, Rochester NY; and ThedaCare, Appleton, Wis.

The prevalence of domestic abuse in 3 primary care studies ranged from 6% to 22%. In these studies domestic abuse was measured as physical abuse^1-4 or a combination of both physical and verbal abuse.^1,2,4 In these and many other studies, an abusive relationship was associated with psychological problems,^3-14 drug and alcohol abuse,^3,6,7,9-11 and many other different symptoms.^{3,5,7-13,15-19} Detection of abusive relationships in clinical settings is assumed to improve outcomes.^3,20,21 Guidelines for detection and management are well described.²²

Despite its prevalence and importance, a woman’s abusive relationship often goes unnoticed in primary care practice.^1-3,23 We used the results of a comprehensive written health survey of 1526 women aged 19 to 69 from 31 practices across the United States to reemphasize why specific inquiry about an abusive relationship is important and how it might be accomplished in a busy office setting.

Methods

Settings and Patients

We used data from 31 sites participating in a 3-year quality improvement project.²⁴ In 1999, at 2 times separated by 6 months, all sites asked 2 sequential samples of adults aged 19 to 69 years to complete an anonymous health survey. Each site determined when and how the sample would be obtained. A total of 2528 patients responded. The number of responses by practice ranged from 19 to 156, and the median was 64. Of these patients, 1584 were women.

Measures

The Screening Tool for an Abusive Relationship. To develop the single-item screening tool,⁹ primary care physicians and nurses examined several methods for detection of abuse identified in the published literature.^25-27 They developed 2 word-and-picture charts based on previously tested methods.^28,29 Two groups of women who had recently experienced domestic abuse (total N=13) reviewed these charts, and their suggestions were incorporated.

For our investigation, the word-and-picture Relationships Chart was used (Figure 1). To validate the Relationships Chart we administered it to 51 women volunteers in urban and rural domestic abuse support groups. The control group consisted of 48 randomly selected patients in 2 obstetric and gynecology clinics. These women were also asked to complete 41 items about spouse or partner abuse.²⁵ The items were scored on a 5-point scale so the maximum score would be 205 and the minimum score 41.

Women seeking help from a support group because of their current or previous involvement in an abusive relationship scored much worse than women in the control group on both the single-item chart and the multi-item score (P <.001). Based on the chart, when women reported potential abuse “none or a little of the time,” their average score on the multi-item questionnaire was 62; when their response was “some of the time” their average score was 86; and when they reported “often or always” experiencing potential abuse during the previous 4 weeks their average score was 114.

Eighty-eight of the women completed the Relationship Chart a second time within 10 days after the initial administration. The average test-retest correlation was 0.60; on the 5-point scale 88.4% of the responses stayed the same or shifted by only 1 scale point.

Thus, the Relationship Chart had reasonable face and criterion validity and more than met minimum standards for reliability.³⁰

The Patient Health Survey. The Relationships Chart was included as part of a patient health survey called “Improve Your Medical Care.” This survey was used as part of a quality improvement system that has been shown to improve care for different populations.^31,32 However, in our study it was used only to sample patients’ needs and characteristics in the participant practices. The patient health survey contains questions about demographics and diagnoses (8 items), function (6 items), symptoms and bothers (22 items), habits and prevention (13 items), and utilization (4 items). A complete copy of this survey is available on request.

All practices received computer-scored reports for each respondent within 1 week of completion.

Analysis

To compare scores on the Relationships Chart with measures of functional status using the Dartmouth Primary Care Cooperative Information and Research Network (COOP) charts, we used a score of 4 or 5 on a 5-point Likert scale as the threshold for significant limitation in function.³³ For bothersome symptoms we used a response of “often” or “always” in the previous 4 weeks. We used the chi-square test to assess the statistical significance of the comparisons.

We used prevalence ratios to compare the prevalence of abusive relationships for women with and without functional limits and symptoms.

Results

Clinical and Functional Impacts of an Abusive Relationship Ninety-six percent of the women completed the chart (1526/1584). Possible domestic abuse (as assessed by this chart) was reported by 13% (201) of the women visiting these 31 outpatient practices. The range of positive responses was 0% to 50%; the median was 11%.

The Table lists several factors highly associated with an abusive relationship. Women in an abusive relationship were more likely to be young and poor. They more often had multiple problems with function and many bothersome symptoms. They less often engaged in good personal health habits. More of these women smoked cigarettes: 24% vs 15%, P=.002; more of them drank more than 6 alcoholic beverages a week: 12% vs 7%, P=.001). They more often reported that they had been confined to bed in the previous 3 months than women who were not abused.

Despite a higher frequency of clinic use, women in an abusive relationship more often reported problems with access and slightly less often completed necessary preventive actions. However, they had the same number of chronic diagnoses and used the hospital and chronic medications at the same rate as other women (data not shown).

The most common functional limitations reported by all women were pain (24%), physical (21%), and feelings (18%). Figure 2 illustrates the probability of an abusive relationship for women with or without a particular functional limitation. Pain and feelings were most often associated with an abusive relationship. Almost a third of the women who are limited by feelings are at risk for domestic abuse. The prevalence ratio for feelings was 3.0.

The most common symptoms reported were eating/weight (26%), dizziness/fatigue (24%), joint pains (23%), back pains (22%), sleep problems (22%), and headaches (21%). Figure 3 illustrates the probability of an abusive relationship for women with or without a particular symptom. Among the 6 most prevalent symptoms, headache was most strongly associated with domestic abuse (24%). Among women who reported abusive relationships “most or all of the time” the chance of significant problems with feelings, dizziness/fatigue, headache, or sleep was greater than 50% (data not shown).

Identifying Women at Risk

Although a combination of demographic characteristics and clinical problems can be used to identify a group of women at high risk for abuse,³ such an approach makes little practical sense. For example, among women who have inadequate money and significant problems with feelings, 41% (36/89) were likely to have an abusive relationship. However, among the 1025 women with neither financial difficulties nor emotional problems, 64 were at risk for abuse (6%).

Discussion

We found that approximately 13% of women aged 19 to 69 years in 31 practices were identified as being in an abusive relationship by a single written measure. These women had significant psychosocial issues, poor health habits, and many somatic complaints.

Our descriptive study of screening for an abusive relationship in women has 4 important implications for clinical practice:

1. It reaffirms the prevalence of abuse in office practice settings. In these 31 practices, the median rate of possible abuse was 11%.
2. It provides a detailed illustration of the burden of clinical and social illness borne by women in abusive relationships. One other study³ has provided a similar level of detail. Where measures of symptoms were comparable, both studies found similar prevalence ratios for abuse.
3. It reminds clinicians that direct inquiry about possible abuse is likely to be much more effective than using a combination of demographic and clinical factors.^3,15 Clinical intuition used to select an at-risk woman from a group will be less effective than asking her directly about possible abuse.
4. These results add to the growing body of literature that single-item word-and-picture charts that query patients about complex issues are very useful in clinical practice. Such charts have been shown to be valid and reliable for identifying important functional limitations in adults,^28,29 depression,³³ health and social problems in adolescents,³⁴ and spirituality.³⁵ Because they are single-item they are good for screening, can be easily adapted to other languages, and serve as the foundation for deeper inquiry about physician-patient interaction.^33,36 For example, more than one third of the women in these practices felt that their clinician was unaware of important functional limitations or provided poor education about them.

This particular measure for an abusive relationship has face and criterion validity and reasonable reliability. It seems acceptable for use in practice, since 96% of the women who completed a health survey also completed this item. However, it is intentionally not as specific as direct questions about physical abuse such as: “Has your husband, boyfriend, partner/lover hit, kicked, threatened, or otherwise frightened you?”²² The clinicians and women who assisted in the design of the Relationships Chart felt that a screening question would appear less intimidating if it were less specific and did not require a yes or no response. We previously received the same advice from adolescents for inquiring about antisocial behavior and drug abuse.³⁴ The pictures add to the appeal of the Dartmouth COOP Charts but do not seem to influence responses.³⁷

The validation study (comparing response categories to another instrument) and the results illustrated in the Table (comparing functional and symptomatic impacts by response categories) indicate that a woman who says that she has experienced an abusive relationship at least some of the time in the past 4 weeks probably is experiencing abuse. For those women who screen positive, more direct inquiry is indicated about the nature of the abuse.

If clinicians do not actively screen but choose to wait for a woman in an abusive relationship to declare herself, the data suggest that she may enter a labyrinth of nonspecific complaints and increased utilization of services. If clinicians actively inquire about abuse, short-term demands for accurate documentation and effective referral will have to be met,^20-22 and the longer-term benefit to the patient and clinician may be considerable.

Conclusions

Domestic abuse is a prevalent and important problem for women, and simple measures can detect domestic abuse in community practice. It is necessary to think of abuse when some functional and symptomatic issues are present, but that alone is not sufficient. Direct inquiry will be more effective. The symptoms and functional problems of these women may be erroneously evaluated unless the domestic abuse is detected.

Acknowledgments

We thank and acknowledge the physician practice sites that participated in our project: Baylor College of Medicine, Houston, Tex; Berlin Health System, Green Bay, Wis; Beth Israel Deaconess Medical Center, Boston, Mass; Cambridge Health Alliance, Cambridge, Mass; Dana Farber Cancer Institute, Boston, Mass; Federation of Swedish County Councils/Sweden Anderstop Health Center, Greater Lawrence Family Health Center, Lawrence, Mass; Harvard University Health Services, Cambridge, Mass; Independent Health: Buffalo Medical Group, Tonawanda Medical Associates, Buffalo, NY; Joslin Diabetes Center, Boston, Mass; Lathem Medical Group, Lathem, NY; Magic Valley Health Center, Twin Falls, Idaho; Mayo Health System: Luther Midelfort, Eau Clair, Wis, and Mayo Clinic Scottsdale, Ariz; Medical University of South Carolina, Charleston, SC; MeritCare, Fargo, ND; PeaceHealth, Eugene, Ore; PennState Geisinger Health System, Pa; Scripps Clinic, La Jolla, Calif; SSM Health System, St. Louis, Ohio; Dean Health System, Madison, Wis; St. Mary’s Health System, Jefferson City, Mo; Strong Health, University of Rochester, Rochester NY; and ThedaCare, Appleton, Wis.

The prevalence of domestic abuse in 3 primary care studies ranged from 6% to 22%. In these studies domestic abuse was measured as physical abuse^1-4 or a combination of both physical and verbal abuse.^1,2,4 In these and many other studies, an abusive relationship was associated with psychological problems,^3-14 drug and alcohol abuse,^3,6,7,9-11 and many other different symptoms.^{3,5,7-13,15-19} Detection of abusive relationships in clinical settings is assumed to improve outcomes.^3,20,21 Guidelines for detection and management are well described.²²

Despite its prevalence and importance, a woman’s abusive relationship often goes unnoticed in primary care practice.^1-3,23 We used the results of a comprehensive written health survey of 1526 women aged 19 to 69 from 31 practices across the United States to reemphasize why specific inquiry about an abusive relationship is important and how it might be accomplished in a busy office setting.

Methods

Settings and Patients

We used data from 31 sites participating in a 3-year quality improvement project.²⁴ In 1999, at 2 times separated by 6 months, all sites asked 2 sequential samples of adults aged 19 to 69 years to complete an anonymous health survey. Each site determined when and how the sample would be obtained. A total of 2528 patients responded. The number of responses by practice ranged from 19 to 156, and the median was 64. Of these patients, 1584 were women.

Measures

The Screening Tool for an Abusive Relationship. To develop the single-item screening tool,⁹ primary care physicians and nurses examined several methods for detection of abuse identified in the published literature.^25-27 They developed 2 word-and-picture charts based on previously tested methods.^28,29 Two groups of women who had recently experienced domestic abuse (total N=13) reviewed these charts, and their suggestions were incorporated.

For our investigation, the word-and-picture Relationships Chart was used (Figure 1). To validate the Relationships Chart we administered it to 51 women volunteers in urban and rural domestic abuse support groups. The control group consisted of 48 randomly selected patients in 2 obstetric and gynecology clinics. These women were also asked to complete 41 items about spouse or partner abuse.²⁵ The items were scored on a 5-point scale so the maximum score would be 205 and the minimum score 41.

Women seeking help from a support group because of their current or previous involvement in an abusive relationship scored much worse than women in the control group on both the single-item chart and the multi-item score (P <.001). Based on the chart, when women reported potential abuse “none or a little of the time,” their average score on the multi-item questionnaire was 62; when their response was “some of the time” their average score was 86; and when they reported “often or always” experiencing potential abuse during the previous 4 weeks their average score was 114.

Eighty-eight of the women completed the Relationship Chart a second time within 10 days after the initial administration. The average test-retest correlation was 0.60; on the 5-point scale 88.4% of the responses stayed the same or shifted by only 1 scale point.

Thus, the Relationship Chart had reasonable face and criterion validity and more than met minimum standards for reliability.³⁰

The Patient Health Survey. The Relationships Chart was included as part of a patient health survey called “Improve Your Medical Care.” This survey was used as part of a quality improvement system that has been shown to improve care for different populations.^31,32 However, in our study it was used only to sample patients’ needs and characteristics in the participant practices. The patient health survey contains questions about demographics and diagnoses (8 items), function (6 items), symptoms and bothers (22 items), habits and prevention (13 items), and utilization (4 items). A complete copy of this survey is available on request.

All practices received computer-scored reports for each respondent within 1 week of completion.

Analysis

To compare scores on the Relationships Chart with measures of functional status using the Dartmouth Primary Care Cooperative Information and Research Network (COOP) charts, we used a score of 4 or 5 on a 5-point Likert scale as the threshold for significant limitation in function.³³ For bothersome symptoms we used a response of “often” or “always” in the previous 4 weeks. We used the chi-square test to assess the statistical significance of the comparisons.

We used prevalence ratios to compare the prevalence of abusive relationships for women with and without functional limits and symptoms.

Results

Clinical and Functional Impacts of an Abusive Relationship Ninety-six percent of the women completed the chart (1526/1584). Possible domestic abuse (as assessed by this chart) was reported by 13% (201) of the women visiting these 31 outpatient practices. The range of positive responses was 0% to 50%; the median was 11%.

The Table lists several factors highly associated with an abusive relationship. Women in an abusive relationship were more likely to be young and poor. They more often had multiple problems with function and many bothersome symptoms. They less often engaged in good personal health habits. More of these women smoked cigarettes: 24% vs 15%, P=.002; more of them drank more than 6 alcoholic beverages a week: 12% vs 7%, P=.001). They more often reported that they had been confined to bed in the previous 3 months than women who were not abused.

Despite a higher frequency of clinic use, women in an abusive relationship more often reported problems with access and slightly less often completed necessary preventive actions. However, they had the same number of chronic diagnoses and used the hospital and chronic medications at the same rate as other women (data not shown).

The most common functional limitations reported by all women were pain (24%), physical (21%), and feelings (18%). Figure 2 illustrates the probability of an abusive relationship for women with or without a particular functional limitation. Pain and feelings were most often associated with an abusive relationship. Almost a third of the women who are limited by feelings are at risk for domestic abuse. The prevalence ratio for feelings was 3.0.

The most common symptoms reported were eating/weight (26%), dizziness/fatigue (24%), joint pains (23%), back pains (22%), sleep problems (22%), and headaches (21%). Figure 3 illustrates the probability of an abusive relationship for women with or without a particular symptom. Among the 6 most prevalent symptoms, headache was most strongly associated with domestic abuse (24%). Among women who reported abusive relationships “most or all of the time” the chance of significant problems with feelings, dizziness/fatigue, headache, or sleep was greater than 50% (data not shown).

Identifying Women at Risk

Although a combination of demographic characteristics and clinical problems can be used to identify a group of women at high risk for abuse,³ such an approach makes little practical sense. For example, among women who have inadequate money and significant problems with feelings, 41% (36/89) were likely to have an abusive relationship. However, among the 1025 women with neither financial difficulties nor emotional problems, 64 were at risk for abuse (6%).

Discussion

We found that approximately 13% of women aged 19 to 69 years in 31 practices were identified as being in an abusive relationship by a single written measure. These women had significant psychosocial issues, poor health habits, and many somatic complaints.

Our descriptive study of screening for an abusive relationship in women has 4 important implications for clinical practice:

1. It reaffirms the prevalence of abuse in office practice settings. In these 31 practices, the median rate of possible abuse was 11%.
2. It provides a detailed illustration of the burden of clinical and social illness borne by women in abusive relationships. One other study³ has provided a similar level of detail. Where measures of symptoms were comparable, both studies found similar prevalence ratios for abuse.
3. It reminds clinicians that direct inquiry about possible abuse is likely to be much more effective than using a combination of demographic and clinical factors.^3,15 Clinical intuition used to select an at-risk woman from a group will be less effective than asking her directly about possible abuse.
4. These results add to the growing body of literature that single-item word-and-picture charts that query patients about complex issues are very useful in clinical practice. Such charts have been shown to be valid and reliable for identifying important functional limitations in adults,^28,29 depression,³³ health and social problems in adolescents,³⁴ and spirituality.³⁵ Because they are single-item they are good for screening, can be easily adapted to other languages, and serve as the foundation for deeper inquiry about physician-patient interaction.^33,36 For example, more than one third of the women in these practices felt that their clinician was unaware of important functional limitations or provided poor education about them.

This particular measure for an abusive relationship has face and criterion validity and reasonable reliability. It seems acceptable for use in practice, since 96% of the women who completed a health survey also completed this item. However, it is intentionally not as specific as direct questions about physical abuse such as: “Has your husband, boyfriend, partner/lover hit, kicked, threatened, or otherwise frightened you?”²² The clinicians and women who assisted in the design of the Relationships Chart felt that a screening question would appear less intimidating if it were less specific and did not require a yes or no response. We previously received the same advice from adolescents for inquiring about antisocial behavior and drug abuse.³⁴ The pictures add to the appeal of the Dartmouth COOP Charts but do not seem to influence responses.³⁷

The validation study (comparing response categories to another instrument) and the results illustrated in the Table (comparing functional and symptomatic impacts by response categories) indicate that a woman who says that she has experienced an abusive relationship at least some of the time in the past 4 weeks probably is experiencing abuse. For those women who screen positive, more direct inquiry is indicated about the nature of the abuse.

If clinicians do not actively screen but choose to wait for a woman in an abusive relationship to declare herself, the data suggest that she may enter a labyrinth of nonspecific complaints and increased utilization of services. If clinicians actively inquire about abuse, short-term demands for accurate documentation and effective referral will have to be met,^20-22 and the longer-term benefit to the patient and clinician may be considerable.

Conclusions

Domestic abuse is a prevalent and important problem for women, and simple measures can detect domestic abuse in community practice. It is necessary to think of abuse when some functional and symptomatic issues are present, but that alone is not sufficient. Direct inquiry will be more effective. The symptoms and functional problems of these women may be erroneously evaluated unless the domestic abuse is detected.

Acknowledgments

We thank and acknowledge the physician practice sites that participated in our project: Baylor College of Medicine, Houston, Tex; Berlin Health System, Green Bay, Wis; Beth Israel Deaconess Medical Center, Boston, Mass; Cambridge Health Alliance, Cambridge, Mass; Dana Farber Cancer Institute, Boston, Mass; Federation of Swedish County Councils/Sweden Anderstop Health Center, Greater Lawrence Family Health Center, Lawrence, Mass; Harvard University Health Services, Cambridge, Mass; Independent Health: Buffalo Medical Group, Tonawanda Medical Associates, Buffalo, NY; Joslin Diabetes Center, Boston, Mass; Lathem Medical Group, Lathem, NY; Magic Valley Health Center, Twin Falls, Idaho; Mayo Health System: Luther Midelfort, Eau Clair, Wis, and Mayo Clinic Scottsdale, Ariz; Medical University of South Carolina, Charleston, SC; MeritCare, Fargo, ND; PeaceHealth, Eugene, Ore; PennState Geisinger Health System, Pa; Scripps Clinic, La Jolla, Calif; SSM Health System, St. Louis, Ohio; Dean Health System, Madison, Wis; St. Mary’s Health System, Jefferson City, Mo; Strong Health, University of Rochester, Rochester NY; and ThedaCare, Appleton, Wis.

In many countries the structure of health care is under review, and strengthening the delivery of primary health care is a common concern.^1,2 Primary care implies medical care in the context of the individual’s psychosocial and family structure—the “contextual complexity” of medical care³—and this primary care orientation⁴ improves cost-effectiveness.⁵ The medical discipline most directly involved in developing primary care is family practice; in Europe and the United Kingdom it is also referred to as general practice. We will use the term family practice, since these terms are used interchangeably in the international literature.³

For the continued development of the discipline of family practice, it is essential to evaluate the needs of patients and the effectiveness of primary care, and to develop evidence to guide practice. To achieve this, routine patient care in the community has to be subjected to systematic research. Practice-based research networks (PBRNs) provide the primary care disciplines with the research laboratories needed to promote scientifically rigorous collection of data. These laboratories reflect the social setting of practice and the personal relations between physicians and patients over time. They provide the opportunities to study unselected health problems, the effect of continuity of care, individual disease prevention strategies, care of families, and the implications of providing care with respect to individual and sociocultural norms and values.^6,7

This personal dimension of family practice can cause tension with the need for research to be representative of and applicable to family practice in general. This is a limitation of single-practice research. Thus, researchers usually aim for a representative mix of family practices in primary care research. Involving more practices also opens the possibility of recruiting larger numbers of patients and physicians. The networking of practices for research is a way of investing in the research culture of family medicine and creating favorable conditions to collect data, test interventions, and study the outcomes of care. PBRNs have become an important element in family practice and have been increasing in number and scope.^8-10 Their development can benefit from a better understanding of their strengths and weaknesses. In particular, the sharing of different experiences can help identify some key aspects of network operations.

Methods

The 1998 Wisconsin Research Network Conference provided an opportunity for the authors to study PBRN experiences. Network experiences from the United Kingdom and the Netherlands were presented, and this resulted in a comparison of 3 networks in 3 countries: the Wisconsin Research Network (United States), the Wessex Primary Care Research Network (England),¹¹ and the Nijmegen Family Practice Academic Network (the Netherlands).^12-14 Although this is an arbitrary selection of networks, each has a record of family practice research and operates in its prevailing health care setting and within its national culture of clinical research. This suggested that the comparisons would yield insights that are relevant for family practice research networks in general.

In the presentations and discussions various aspects of PBRN structure and operation were addressed. We condensed these into 4 key areas: (1) the missions of the networks; (2) the contribution of the networks to the evidence base of family medicine;^11-31 (3) the management of the networks, relationship to members, and data collection; and (4) the financing of the network infrastructure and studies.

Results

Information about the 3 networks is summarized in the Tables: In Table 1 the key characteristics are outlined, and in Table 2 the results of 5 characteristic studies are listed, including funding source and main publications. The Wisconsin network lists more than 700 members, including 467 family physicians in 207 practices (approximately one third of the family physicians in the state), who cover a population of more than 900,000. The network also includes specialists, nurses, and researchers. The Wessex network covers a patient population of more than 1.7 million people and involves 234 family physicians in 125 practices, as well as practice nurses, dentists, pharmacists, and optometrists. The Nijmegen network is based on a stable group of 10 practices with 25 family physicians. It covers a population of 45,000 that reflects the composition of the Dutch population in age and social class. Each network has a direct link to academic family medicine.

Missions of the Networks

The goal of all 3 networks is to increase the evidence base of primary care. However, there are important differences in the way this mission is put into operation. The Nijmegen network collects patient-related data on an ongoing basis. Since 1971 it has collected all presented morbidity, and since 1986 it has accumulated a core set of process and outcome data from patients with hypertension and heart disease, diabetes mellitus, asthma, and chronic obstructive pulmonary disease. Thus, a database of long-term individual morbidity and outcome of care has been built that forms the index for further clinical research. Data collection and associated research are centrally structured, and all practices and physicians are fully committed to the data collection.

In contrast, Wisconsin and Wessex formed networks for family physicians and other primary care professionals with an interest in research. The networks provide support for research initiatives by offering opportunities to join studies (including research training) and by supporting their own research initiatives. Involvement is optional and reflects the interests and ambitions of the individual practices. Approximately half of the clinicians (350) in the Wisconsin network have been actively involved in 1 or more studies. To date, 70% of the practices in the Wessex network have been involved in 1 or more projects.

Contributions to the Evidence Base

These networks indicated characteristic studies done by (or in the case of Wisconsin, supported by) their network (Table 2). Two main areas of research are represented in these studies: (1) the quality of the care during the projects on the identification of disease risk (Wisconsin), asthma information,²³ and venepuncture²⁵ (Wessex); and (2) prevention and treatment of common morbidity. This covers a wide range of illnesses and addresses an interesting mixture of objectives. The asthma studies in Wisconsin²¹ and Nijmegen²⁹ and the Wisconsin alcoholism study²⁰ focus on the pathophysiological aspects of the disease and potential intervention effects (efficacy), while the Wessex osteoporosis study and the Nijmegen preventive cardiology study³⁰ assessed the effectiveness of interventions under primary care conditions. Documenting the natural course of the disease and preventive actions are featured in studies on screening^15-19 (Wisconsin), hay fever²² (Wessex), childhood morbidity,²⁸ diabetes mellitus,²⁶ and depression³¹ (Nijmegen). The role of the networks in the conduct of these studies ranges from total responsibility (Wessex, Nijmegen) to a more varied role ranging from total responsibility to recruiting practices only (Wisconsin). Summarizing the research output is difficult given the range of clinical primary care topics covered. The summaries in Table 2 suggest that the research is aimed at addressing essential clinical decisions family physicians are facing in their routine daily care.

Management of the Networks and Relationship to Members

The Wessex and Nijmegen networks are university based. The Wisconsin network was initiated by the Wisconsin Academy of Family Physicians (WAFP) and is managed by and receives support from the University of Wisconsin. All 3 coordinating academic centers promote ownership of network activities by the participating practices. This includes regular exchange of information on new and ongoing research and study-specific instructions. The consent of the participating practices is mandatory before a study can be performed, irrespective of the source of funding or the director of the study. In Wisconsin, the WAFP must approve major commitments of the members’ time and energy. Each network involves participating physicians as principal or co-principal investigators when possible. There are differences in the way regular contacts are maintained, depending on the size of the network and the geography. The small Nijmegen network holds monthly meetings; the larger Wisconsin and Wessex networks organize an annual conference and apply distance communication technology: newsletters, Web sites, E-mail, and an electronic discussion list. They 2 larger networks maintain closer contact with a core group of active researchers through project team meetings.

Data Collection

Data collection methods are varied. The Nijmegen network collects a standard set of patient-related data for every practice on a routine basis. The other networks collect only project-specific data derived from medical records, laboratory tests, physician surveys, and patient interviews. In addition, Wisconsin uses multiple methods ranging from qualitative methods to chart review to review clinical databases. Currently there are projects piloting the application of new technology including interactive voice recording and Internet-based data acquisition.

The Financing of the Networks

The funding of the networks has to cover 2 areas: the network infrastructure and specific research projects. The latter involves the main national funding bodies for biomedical research including scientific foundations and industry (Table 2). The infrastructure costs include co-ordination, methodologic support, and administrative support. Recently limited structural support has been provided by the university (Nijmegen) and the National Health Service Research and Development Program (Wessex). In Wisconsin basic financial support comes from the WAFP, with the university providing senior staff time and office space. This funding, however, is insufficient to provide comprehensive support for the multidisciplinary research that characterises PBRNs. In Nijmegen a contract between the university and the practices determines financial and scientific duties, given their intensive cooperation.

Also, obtaining grant support for projects has proved difficult for all networks. Review committees are often mainly accustomed to reductionistic research about unitary and well-defined problems and have been concerned about pragmatic designs.

Discussion

We used information from 3 networks in 3 countries to describe PBRNs in the context of primary care developments in their country of origin. This enhanced the richness of our data but is also the major limitation of our study because it did not necessarily represent a complete view of a majority of the existing experience. However, our comparison illustrates the common elements of each of the 3 PBRNs: They have each successfully recruited large numbers of unselected patients from different practices for epidemiologic and clinical research, efficacy (and, to a reasonable extent, effectiveness) studies, and studies aimed at improving our understanding of the process of care in family practice.

The recruitment of members into networks and specific research studies depends on their interest and willingness to contribute to research. Each network allows clinicians to make their own decision about whether to be involved in studies. Networks consist of self-selected practices, and in the United States and the United Kingdom not all practices agree to participate in all studies. This has implications for selection bias, particularly in studies where the clinician or the practice is the subject of study. However, given the fact that care is provided for unselected patients, practice self-selection will have less impact on patient-directed research. The Nijmegen experience illustrates that there may be a time in the development of the network when membership becomes more demanding, and members may be even less representative of clinicians in general. However, none of the network directors mentioned the retention of practices and physicians as a problem. Apparently this is related to the flexible approach in recruitment.

The Wisconsin and Wessex networks hope to stimulate physicians’ personal involvement in research to create a questioning environment. This is an additional bottom-up development in creating a research culture. In this respect the Nijmegen network seems at first to be very different, functioning more as a top-down university-centered research program. This is not true, however, as the Nijmegen network has provided family practice input to the medical school for more than 20 years, and 8 of the 25 physicians to date have received the highest academic degree of MD, PhD. The Nijmegen situation represents the full circle of changed research culture, with networked family physicians in charge of an independent family practice research program.

Ownership of the research is a particularly sensitive issue that can make or break the success of a study. Only when family physicians are confident that the data collection is relevant to and compatible with the demands of routine practice will it be possible to pursue a study.³² A key function of PBRN management is close communication and negotiation between researchers and physicians. An important outcome of this communication is the role of the network in providing direct input from primary care clinicians about the relevance of proposed studies for the development of family medicine. There is a need for direct research into the interests of clinicians who have to cope with the full complexity of patient care in the community setting.

A second key function of PBRN management is to develop research methodology in the network: Better research methodology will facilitate physician involvement, assist funding, and assure the obtained data are valid. Networks develop their own momentum. Initially, simple descriptive studies are conducted, but with increasing experience PBRNs can address larger and more complex collaborative projects. This process in the Nijmegen network has already been analyzed.³³

PBRN studies may support current care practices and have a quality assurance focus in improving interventions, or cast doubt on current care practices and contribute to the development of new ones. A variety of primary care settings need to be used for these studies, given the impact of environment on the outcome of care. Evidence from small specialty settings can only be introduced directly into routine family practice to a limited degree.

The relationship between practice and research

The 3 networks expressed the need to support routine practice, do research, and, at the same time, raise the quality of care in the network. An integral relationship between practice and research is apparent in each of the networks. Questioning and supporting primary care simultaneously, however, is more ambivalent than it may seem at first sight. For family medicine it is particularly important to demonstrate that interventions work under most prevailing primary care conditions (evidence-based practice).³⁴ But because a substantial number of interventions are used mainly in primary care settings, the potential of these interventions (efficacy) must also be studied in primary care. The studies of the pathophysiology of asthma fall into this category; these require ideal rather than prevailing practice conditions. This requires a choice from networks about how to perform these types of studies. The Nijmegen network represents an academic primary care setting tuned to the requirements of efficacy research. The Wisconsin and Wessex networks reflect more the existing variations in actual care, and this provides excellent opportunities for studying effectiveness of care under primary care conditions. A number of studies in the networks were descriptive, detailing the natural course of illness and disease under primary care conditions. These are important as they demonstrate to what extent evidence from other studies can be directly translated to practice.

The recent financial support for networks in the Netherlands⁸ and the United Kingdom⁹ heralds the increasing awareness of the importance of primary care evidence. Within the last year in the US Agency for Healthcare Research and Quality (AHRQ) has for the first time offered direct support for building PBRN infrastructure. But because the structure of research networks depend on the research agenda, there is a need for more appropriate financing of their infrastructures. The increasing awareness of the importance of practice-based research is further highlighted by the formation of the Federation of Practice-Based Research Networks (FPBRN), which is working to help networks communicate and collaborate on projects across national and international boundaries.

Conclusions

Family practice research networks are an important way of facilitating research in primary care and of assuring sufficient primary care emphasis in clinical studies. The scientific products of these networks, as judged from their publications, make valuable contributions to the evidence base of primary care.

In many countries the structure of health care is under review, and strengthening the delivery of primary health care is a common concern.^1,2 Primary care implies medical care in the context of the individual’s psychosocial and family structure—the “contextual complexity” of medical care³—and this primary care orientation⁴ improves cost-effectiveness.⁵ The medical discipline most directly involved in developing primary care is family practice; in Europe and the United Kingdom it is also referred to as general practice. We will use the term family practice, since these terms are used interchangeably in the international literature.³

For the continued development of the discipline of family practice, it is essential to evaluate the needs of patients and the effectiveness of primary care, and to develop evidence to guide practice. To achieve this, routine patient care in the community has to be subjected to systematic research. Practice-based research networks (PBRNs) provide the primary care disciplines with the research laboratories needed to promote scientifically rigorous collection of data. These laboratories reflect the social setting of practice and the personal relations between physicians and patients over time. They provide the opportunities to study unselected health problems, the effect of continuity of care, individual disease prevention strategies, care of families, and the implications of providing care with respect to individual and sociocultural norms and values.^6,7

This personal dimension of family practice can cause tension with the need for research to be representative of and applicable to family practice in general. This is a limitation of single-practice research. Thus, researchers usually aim for a representative mix of family practices in primary care research. Involving more practices also opens the possibility of recruiting larger numbers of patients and physicians. The networking of practices for research is a way of investing in the research culture of family medicine and creating favorable conditions to collect data, test interventions, and study the outcomes of care. PBRNs have become an important element in family practice and have been increasing in number and scope.^8-10 Their development can benefit from a better understanding of their strengths and weaknesses. In particular, the sharing of different experiences can help identify some key aspects of network operations.

Methods

The 1998 Wisconsin Research Network Conference provided an opportunity for the authors to study PBRN experiences. Network experiences from the United Kingdom and the Netherlands were presented, and this resulted in a comparison of 3 networks in 3 countries: the Wisconsin Research Network (United States), the Wessex Primary Care Research Network (England),¹¹ and the Nijmegen Family Practice Academic Network (the Netherlands).^12-14 Although this is an arbitrary selection of networks, each has a record of family practice research and operates in its prevailing health care setting and within its national culture of clinical research. This suggested that the comparisons would yield insights that are relevant for family practice research networks in general.

In the presentations and discussions various aspects of PBRN structure and operation were addressed. We condensed these into 4 key areas: (1) the missions of the networks; (2) the contribution of the networks to the evidence base of family medicine;^11-31 (3) the management of the networks, relationship to members, and data collection; and (4) the financing of the network infrastructure and studies.

Results

Information about the 3 networks is summarized in the Tables: In Table 1 the key characteristics are outlined, and in Table 2 the results of 5 characteristic studies are listed, including funding source and main publications. The Wisconsin network lists more than 700 members, including 467 family physicians in 207 practices (approximately one third of the family physicians in the state), who cover a population of more than 900,000. The network also includes specialists, nurses, and researchers. The Wessex network covers a patient population of more than 1.7 million people and involves 234 family physicians in 125 practices, as well as practice nurses, dentists, pharmacists, and optometrists. The Nijmegen network is based on a stable group of 10 practices with 25 family physicians. It covers a population of 45,000 that reflects the composition of the Dutch population in age and social class. Each network has a direct link to academic family medicine.

Missions of the Networks

The goal of all 3 networks is to increase the evidence base of primary care. However, there are important differences in the way this mission is put into operation. The Nijmegen network collects patient-related data on an ongoing basis. Since 1971 it has collected all presented morbidity, and since 1986 it has accumulated a core set of process and outcome data from patients with hypertension and heart disease, diabetes mellitus, asthma, and chronic obstructive pulmonary disease. Thus, a database of long-term individual morbidity and outcome of care has been built that forms the index for further clinical research. Data collection and associated research are centrally structured, and all practices and physicians are fully committed to the data collection.

In contrast, Wisconsin and Wessex formed networks for family physicians and other primary care professionals with an interest in research. The networks provide support for research initiatives by offering opportunities to join studies (including research training) and by supporting their own research initiatives. Involvement is optional and reflects the interests and ambitions of the individual practices. Approximately half of the clinicians (350) in the Wisconsin network have been actively involved in 1 or more studies. To date, 70% of the practices in the Wessex network have been involved in 1 or more projects.

Contributions to the Evidence Base

These networks indicated characteristic studies done by (or in the case of Wisconsin, supported by) their network (Table 2). Two main areas of research are represented in these studies: (1) the quality of the care during the projects on the identification of disease risk (Wisconsin), asthma information,²³ and venepuncture²⁵ (Wessex); and (2) prevention and treatment of common morbidity. This covers a wide range of illnesses and addresses an interesting mixture of objectives. The asthma studies in Wisconsin²¹ and Nijmegen²⁹ and the Wisconsin alcoholism study²⁰ focus on the pathophysiological aspects of the disease and potential intervention effects (efficacy), while the Wessex osteoporosis study and the Nijmegen preventive cardiology study³⁰ assessed the effectiveness of interventions under primary care conditions. Documenting the natural course of the disease and preventive actions are featured in studies on screening^15-19 (Wisconsin), hay fever²² (Wessex), childhood morbidity,²⁸ diabetes mellitus,²⁶ and depression³¹ (Nijmegen). The role of the networks in the conduct of these studies ranges from total responsibility (Wessex, Nijmegen) to a more varied role ranging from total responsibility to recruiting practices only (Wisconsin). Summarizing the research output is difficult given the range of clinical primary care topics covered. The summaries in Table 2 suggest that the research is aimed at addressing essential clinical decisions family physicians are facing in their routine daily care.

Management of the Networks and Relationship to Members

The Wessex and Nijmegen networks are university based. The Wisconsin network was initiated by the Wisconsin Academy of Family Physicians (WAFP) and is managed by and receives support from the University of Wisconsin. All 3 coordinating academic centers promote ownership of network activities by the participating practices. This includes regular exchange of information on new and ongoing research and study-specific instructions. The consent of the participating practices is mandatory before a study can be performed, irrespective of the source of funding or the director of the study. In Wisconsin, the WAFP must approve major commitments of the members’ time and energy. Each network involves participating physicians as principal or co-principal investigators when possible. There are differences in the way regular contacts are maintained, depending on the size of the network and the geography. The small Nijmegen network holds monthly meetings; the larger Wisconsin and Wessex networks organize an annual conference and apply distance communication technology: newsletters, Web sites, E-mail, and an electronic discussion list. They 2 larger networks maintain closer contact with a core group of active researchers through project team meetings.

Data Collection

Data collection methods are varied. The Nijmegen network collects a standard set of patient-related data for every practice on a routine basis. The other networks collect only project-specific data derived from medical records, laboratory tests, physician surveys, and patient interviews. In addition, Wisconsin uses multiple methods ranging from qualitative methods to chart review to review clinical databases. Currently there are projects piloting the application of new technology including interactive voice recording and Internet-based data acquisition.

The Financing of the Networks

The funding of the networks has to cover 2 areas: the network infrastructure and specific research projects. The latter involves the main national funding bodies for biomedical research including scientific foundations and industry (Table 2). The infrastructure costs include co-ordination, methodologic support, and administrative support. Recently limited structural support has been provided by the university (Nijmegen) and the National Health Service Research and Development Program (Wessex). In Wisconsin basic financial support comes from the WAFP, with the university providing senior staff time and office space. This funding, however, is insufficient to provide comprehensive support for the multidisciplinary research that characterises PBRNs. In Nijmegen a contract between the university and the practices determines financial and scientific duties, given their intensive cooperation.

Also, obtaining grant support for projects has proved difficult for all networks. Review committees are often mainly accustomed to reductionistic research about unitary and well-defined problems and have been concerned about pragmatic designs.

Discussion

We used information from 3 networks in 3 countries to describe PBRNs in the context of primary care developments in their country of origin. This enhanced the richness of our data but is also the major limitation of our study because it did not necessarily represent a complete view of a majority of the existing experience. However, our comparison illustrates the common elements of each of the 3 PBRNs: They have each successfully recruited large numbers of unselected patients from different practices for epidemiologic and clinical research, efficacy (and, to a reasonable extent, effectiveness) studies, and studies aimed at improving our understanding of the process of care in family practice.

The recruitment of members into networks and specific research studies depends on their interest and willingness to contribute to research. Each network allows clinicians to make their own decision about whether to be involved in studies. Networks consist of self-selected practices, and in the United States and the United Kingdom not all practices agree to participate in all studies. This has implications for selection bias, particularly in studies where the clinician or the practice is the subject of study. However, given the fact that care is provided for unselected patients, practice self-selection will have less impact on patient-directed research. The Nijmegen experience illustrates that there may be a time in the development of the network when membership becomes more demanding, and members may be even less representative of clinicians in general. However, none of the network directors mentioned the retention of practices and physicians as a problem. Apparently this is related to the flexible approach in recruitment.

The Wisconsin and Wessex networks hope to stimulate physicians’ personal involvement in research to create a questioning environment. This is an additional bottom-up development in creating a research culture. In this respect the Nijmegen network seems at first to be very different, functioning more as a top-down university-centered research program. This is not true, however, as the Nijmegen network has provided family practice input to the medical school for more than 20 years, and 8 of the 25 physicians to date have received the highest academic degree of MD, PhD. The Nijmegen situation represents the full circle of changed research culture, with networked family physicians in charge of an independent family practice research program.

Ownership of the research is a particularly sensitive issue that can make or break the success of a study. Only when family physicians are confident that the data collection is relevant to and compatible with the demands of routine practice will it be possible to pursue a study.³² A key function of PBRN management is close communication and negotiation between researchers and physicians. An important outcome of this communication is the role of the network in providing direct input from primary care clinicians about the relevance of proposed studies for the development of family medicine. There is a need for direct research into the interests of clinicians who have to cope with the full complexity of patient care in the community setting.

A second key function of PBRN management is to develop research methodology in the network: Better research methodology will facilitate physician involvement, assist funding, and assure the obtained data are valid. Networks develop their own momentum. Initially, simple descriptive studies are conducted, but with increasing experience PBRNs can address larger and more complex collaborative projects. This process in the Nijmegen network has already been analyzed.³³

PBRN studies may support current care practices and have a quality assurance focus in improving interventions, or cast doubt on current care practices and contribute to the development of new ones. A variety of primary care settings need to be used for these studies, given the impact of environment on the outcome of care. Evidence from small specialty settings can only be introduced directly into routine family practice to a limited degree.

The relationship between practice and research

The 3 networks expressed the need to support routine practice, do research, and, at the same time, raise the quality of care in the network. An integral relationship between practice and research is apparent in each of the networks. Questioning and supporting primary care simultaneously, however, is more ambivalent than it may seem at first sight. For family medicine it is particularly important to demonstrate that interventions work under most prevailing primary care conditions (evidence-based practice).³⁴ But because a substantial number of interventions are used mainly in primary care settings, the potential of these interventions (efficacy) must also be studied in primary care. The studies of the pathophysiology of asthma fall into this category; these require ideal rather than prevailing practice conditions. This requires a choice from networks about how to perform these types of studies. The Nijmegen network represents an academic primary care setting tuned to the requirements of efficacy research. The Wisconsin and Wessex networks reflect more the existing variations in actual care, and this provides excellent opportunities for studying effectiveness of care under primary care conditions. A number of studies in the networks were descriptive, detailing the natural course of illness and disease under primary care conditions. These are important as they demonstrate to what extent evidence from other studies can be directly translated to practice.

The recent financial support for networks in the Netherlands⁸ and the United Kingdom⁹ heralds the increasing awareness of the importance of primary care evidence. Within the last year in the US Agency for Healthcare Research and Quality (AHRQ) has for the first time offered direct support for building PBRN infrastructure. But because the structure of research networks depend on the research agenda, there is a need for more appropriate financing of their infrastructures. The increasing awareness of the importance of practice-based research is further highlighted by the formation of the Federation of Practice-Based Research Networks (FPBRN), which is working to help networks communicate and collaborate on projects across national and international boundaries.

Conclusions

Family practice research networks are an important way of facilitating research in primary care and of assuring sufficient primary care emphasis in clinical studies. The scientific products of these networks, as judged from their publications, make valuable contributions to the evidence base of primary care.

In many countries the structure of health care is under review, and strengthening the delivery of primary health care is a common concern.^1,2 Primary care implies medical care in the context of the individual’s psychosocial and family structure—the “contextual complexity” of medical care³—and this primary care orientation⁴ improves cost-effectiveness.⁵ The medical discipline most directly involved in developing primary care is family practice; in Europe and the United Kingdom it is also referred to as general practice. We will use the term family practice, since these terms are used interchangeably in the international literature.³

For the continued development of the discipline of family practice, it is essential to evaluate the needs of patients and the effectiveness of primary care, and to develop evidence to guide practice. To achieve this, routine patient care in the community has to be subjected to systematic research. Practice-based research networks (PBRNs) provide the primary care disciplines with the research laboratories needed to promote scientifically rigorous collection of data. These laboratories reflect the social setting of practice and the personal relations between physicians and patients over time. They provide the opportunities to study unselected health problems, the effect of continuity of care, individual disease prevention strategies, care of families, and the implications of providing care with respect to individual and sociocultural norms and values.^6,7

This personal dimension of family practice can cause tension with the need for research to be representative of and applicable to family practice in general. This is a limitation of single-practice research. Thus, researchers usually aim for a representative mix of family practices in primary care research. Involving more practices also opens the possibility of recruiting larger numbers of patients and physicians. The networking of practices for research is a way of investing in the research culture of family medicine and creating favorable conditions to collect data, test interventions, and study the outcomes of care. PBRNs have become an important element in family practice and have been increasing in number and scope.^8-10 Their development can benefit from a better understanding of their strengths and weaknesses. In particular, the sharing of different experiences can help identify some key aspects of network operations.

Methods

The 1998 Wisconsin Research Network Conference provided an opportunity for the authors to study PBRN experiences. Network experiences from the United Kingdom and the Netherlands were presented, and this resulted in a comparison of 3 networks in 3 countries: the Wisconsin Research Network (United States), the Wessex Primary Care Research Network (England),¹¹ and the Nijmegen Family Practice Academic Network (the Netherlands).^12-14 Although this is an arbitrary selection of networks, each has a record of family practice research and operates in its prevailing health care setting and within its national culture of clinical research. This suggested that the comparisons would yield insights that are relevant for family practice research networks in general.

In the presentations and discussions various aspects of PBRN structure and operation were addressed. We condensed these into 4 key areas: (1) the missions of the networks; (2) the contribution of the networks to the evidence base of family medicine;^11-31 (3) the management of the networks, relationship to members, and data collection; and (4) the financing of the network infrastructure and studies.

Results

Information about the 3 networks is summarized in the Tables: In Table 1 the key characteristics are outlined, and in Table 2 the results of 5 characteristic studies are listed, including funding source and main publications. The Wisconsin network lists more than 700 members, including 467 family physicians in 207 practices (approximately one third of the family physicians in the state), who cover a population of more than 900,000. The network also includes specialists, nurses, and researchers. The Wessex network covers a patient population of more than 1.7 million people and involves 234 family physicians in 125 practices, as well as practice nurses, dentists, pharmacists, and optometrists. The Nijmegen network is based on a stable group of 10 practices with 25 family physicians. It covers a population of 45,000 that reflects the composition of the Dutch population in age and social class. Each network has a direct link to academic family medicine.

Missions of the Networks

The goal of all 3 networks is to increase the evidence base of primary care. However, there are important differences in the way this mission is put into operation. The Nijmegen network collects patient-related data on an ongoing basis. Since 1971 it has collected all presented morbidity, and since 1986 it has accumulated a core set of process and outcome data from patients with hypertension and heart disease, diabetes mellitus, asthma, and chronic obstructive pulmonary disease. Thus, a database of long-term individual morbidity and outcome of care has been built that forms the index for further clinical research. Data collection and associated research are centrally structured, and all practices and physicians are fully committed to the data collection.

In contrast, Wisconsin and Wessex formed networks for family physicians and other primary care professionals with an interest in research. The networks provide support for research initiatives by offering opportunities to join studies (including research training) and by supporting their own research initiatives. Involvement is optional and reflects the interests and ambitions of the individual practices. Approximately half of the clinicians (350) in the Wisconsin network have been actively involved in 1 or more studies. To date, 70% of the practices in the Wessex network have been involved in 1 or more projects.

Contributions to the Evidence Base

These networks indicated characteristic studies done by (or in the case of Wisconsin, supported by) their network (Table 2). Two main areas of research are represented in these studies: (1) the quality of the care during the projects on the identification of disease risk (Wisconsin), asthma information,²³ and venepuncture²⁵ (Wessex); and (2) prevention and treatment of common morbidity. This covers a wide range of illnesses and addresses an interesting mixture of objectives. The asthma studies in Wisconsin²¹ and Nijmegen²⁹ and the Wisconsin alcoholism study²⁰ focus on the pathophysiological aspects of the disease and potential intervention effects (efficacy), while the Wessex osteoporosis study and the Nijmegen preventive cardiology study³⁰ assessed the effectiveness of interventions under primary care conditions. Documenting the natural course of the disease and preventive actions are featured in studies on screening^15-19 (Wisconsin), hay fever²² (Wessex), childhood morbidity,²⁸ diabetes mellitus,²⁶ and depression³¹ (Nijmegen). The role of the networks in the conduct of these studies ranges from total responsibility (Wessex, Nijmegen) to a more varied role ranging from total responsibility to recruiting practices only (Wisconsin). Summarizing the research output is difficult given the range of clinical primary care topics covered. The summaries in Table 2 suggest that the research is aimed at addressing essential clinical decisions family physicians are facing in their routine daily care.

Management of the Networks and Relationship to Members

The Wessex and Nijmegen networks are university based. The Wisconsin network was initiated by the Wisconsin Academy of Family Physicians (WAFP) and is managed by and receives support from the University of Wisconsin. All 3 coordinating academic centers promote ownership of network activities by the participating practices. This includes regular exchange of information on new and ongoing research and study-specific instructions. The consent of the participating practices is mandatory before a study can be performed, irrespective of the source of funding or the director of the study. In Wisconsin, the WAFP must approve major commitments of the members’ time and energy. Each network involves participating physicians as principal or co-principal investigators when possible. There are differences in the way regular contacts are maintained, depending on the size of the network and the geography. The small Nijmegen network holds monthly meetings; the larger Wisconsin and Wessex networks organize an annual conference and apply distance communication technology: newsletters, Web sites, E-mail, and an electronic discussion list. They 2 larger networks maintain closer contact with a core group of active researchers through project team meetings.

Data Collection

Data collection methods are varied. The Nijmegen network collects a standard set of patient-related data for every practice on a routine basis. The other networks collect only project-specific data derived from medical records, laboratory tests, physician surveys, and patient interviews. In addition, Wisconsin uses multiple methods ranging from qualitative methods to chart review to review clinical databases. Currently there are projects piloting the application of new technology including interactive voice recording and Internet-based data acquisition.

The Financing of the Networks

The funding of the networks has to cover 2 areas: the network infrastructure and specific research projects. The latter involves the main national funding bodies for biomedical research including scientific foundations and industry (Table 2). The infrastructure costs include co-ordination, methodologic support, and administrative support. Recently limited structural support has been provided by the university (Nijmegen) and the National Health Service Research and Development Program (Wessex). In Wisconsin basic financial support comes from the WAFP, with the university providing senior staff time and office space. This funding, however, is insufficient to provide comprehensive support for the multidisciplinary research that characterises PBRNs. In Nijmegen a contract between the university and the practices determines financial and scientific duties, given their intensive cooperation.

Also, obtaining grant support for projects has proved difficult for all networks. Review committees are often mainly accustomed to reductionistic research about unitary and well-defined problems and have been concerned about pragmatic designs.

Discussion

We used information from 3 networks in 3 countries to describe PBRNs in the context of primary care developments in their country of origin. This enhanced the richness of our data but is also the major limitation of our study because it did not necessarily represent a complete view of a majority of the existing experience. However, our comparison illustrates the common elements of each of the 3 PBRNs: They have each successfully recruited large numbers of unselected patients from different practices for epidemiologic and clinical research, efficacy (and, to a reasonable extent, effectiveness) studies, and studies aimed at improving our understanding of the process of care in family practice.

The recruitment of members into networks and specific research studies depends on their interest and willingness to contribute to research. Each network allows clinicians to make their own decision about whether to be involved in studies. Networks consist of self-selected practices, and in the United States and the United Kingdom not all practices agree to participate in all studies. This has implications for selection bias, particularly in studies where the clinician or the practice is the subject of study. However, given the fact that care is provided for unselected patients, practice self-selection will have less impact on patient-directed research. The Nijmegen experience illustrates that there may be a time in the development of the network when membership becomes more demanding, and members may be even less representative of clinicians in general. However, none of the network directors mentioned the retention of practices and physicians as a problem. Apparently this is related to the flexible approach in recruitment.

The Wisconsin and Wessex networks hope to stimulate physicians’ personal involvement in research to create a questioning environment. This is an additional bottom-up development in creating a research culture. In this respect the Nijmegen network seems at first to be very different, functioning more as a top-down university-centered research program. This is not true, however, as the Nijmegen network has provided family practice input to the medical school for more than 20 years, and 8 of the 25 physicians to date have received the highest academic degree of MD, PhD. The Nijmegen situation represents the full circle of changed research culture, with networked family physicians in charge of an independent family practice research program.

Ownership of the research is a particularly sensitive issue that can make or break the success of a study. Only when family physicians are confident that the data collection is relevant to and compatible with the demands of routine practice will it be possible to pursue a study.³² A key function of PBRN management is close communication and negotiation between researchers and physicians. An important outcome of this communication is the role of the network in providing direct input from primary care clinicians about the relevance of proposed studies for the development of family medicine. There is a need for direct research into the interests of clinicians who have to cope with the full complexity of patient care in the community setting.

A second key function of PBRN management is to develop research methodology in the network: Better research methodology will facilitate physician involvement, assist funding, and assure the obtained data are valid. Networks develop their own momentum. Initially, simple descriptive studies are conducted, but with increasing experience PBRNs can address larger and more complex collaborative projects. This process in the Nijmegen network has already been analyzed.³³

PBRN studies may support current care practices and have a quality assurance focus in improving interventions, or cast doubt on current care practices and contribute to the development of new ones. A variety of primary care settings need to be used for these studies, given the impact of environment on the outcome of care. Evidence from small specialty settings can only be introduced directly into routine family practice to a limited degree.

The relationship between practice and research

The 3 networks expressed the need to support routine practice, do research, and, at the same time, raise the quality of care in the network. An integral relationship between practice and research is apparent in each of the networks. Questioning and supporting primary care simultaneously, however, is more ambivalent than it may seem at first sight. For family medicine it is particularly important to demonstrate that interventions work under most prevailing primary care conditions (evidence-based practice).³⁴ But because a substantial number of interventions are used mainly in primary care settings, the potential of these interventions (efficacy) must also be studied in primary care. The studies of the pathophysiology of asthma fall into this category; these require ideal rather than prevailing practice conditions. This requires a choice from networks about how to perform these types of studies. The Nijmegen network represents an academic primary care setting tuned to the requirements of efficacy research. The Wisconsin and Wessex networks reflect more the existing variations in actual care, and this provides excellent opportunities for studying effectiveness of care under primary care conditions. A number of studies in the networks were descriptive, detailing the natural course of illness and disease under primary care conditions. These are important as they demonstrate to what extent evidence from other studies can be directly translated to practice.

The recent financial support for networks in the Netherlands⁸ and the United Kingdom⁹ heralds the increasing awareness of the importance of primary care evidence. Within the last year in the US Agency for Healthcare Research and Quality (AHRQ) has for the first time offered direct support for building PBRN infrastructure. But because the structure of research networks depend on the research agenda, there is a need for more appropriate financing of their infrastructures. The increasing awareness of the importance of practice-based research is further highlighted by the formation of the Federation of Practice-Based Research Networks (FPBRN), which is working to help networks communicate and collaborate on projects across national and international boundaries.

Conclusions

Family practice research networks are an important way of facilitating research in primary care and of assuring sufficient primary care emphasis in clinical studies. The scientific products of these networks, as judged from their publications, make valuable contributions to the evidence base of primary care.

Rising health care costs (especially for medications) and a growing concern that unnecessary antibiotic use is creating troublesome bacterial resistance patterns have triggered increasing attention to upper respiratory infection (URI) care.¹ URIs are the most frequent reason for office visits and antibiotic use, and at least 50% of patient visits with the diagnosis of URI, cold, or bronchitis include the writing of antibiotic prescriptions.^1-2 Also, there is reason to believe that antibiotic use for URI symptoms has increased during the past 2 decades, despite increasing public health efforts to the contrary.³

Much of the expanding literature on URI patients is descriptive, attempting to provide understanding of their expectations and the relationship of various characteristics to antibiotic use for this illness.^4-12 Because of the belief that it is patients who create most of the visits and antibiotic use, most studies of URI care address patients’ perceptions, attitudes, and satisfaction.^2,4-12 Some studies, however, have shown that the physician’s perception that a patient expects antibiotics is the strongest predictor of prescriptions, even though that perception is not always correct.^4-5 This literature suggests that once patients with URIs appear in the clinic it is likely that they will leave with an antibiotic prescription, either because the clinician feels that the illness might be helped by antibiotics or because he or she believes that the patient will be unhappy if antibiotics are not provided. Miller and colleagues¹³ obtained physician questionnaires for patients with a suspected infection as the reason for the visit. They found that physicians responded to their perception that patients wanted an antibiotic only when the physicians were uncertain about the need for it. When they were fairly certain that an antibiotic was either needed or not needed, perceived patient demand was rarely a factor. Patient satisfaction does not appear to be related to the receipt of antibiotics even when they are expected.^5-6

Patients who call or come in within the first few days of the onset of their illness constitute a particularly troubling subgroup of patients for most clinicians. Because it is usually difficult for a clinician to predict whether patients with early URI symptoms are going to encounter complications, it is hard for the clinician to avoid assuming that these patients are either sicker or unusually desirous of a test or treatment. In the absence of any studies of this subgroup, it is possible that those making early contact are particularly likely to receive antibiotics. They will continue to frustrate clinicians and to confound health care system efforts to reduce unnecessary visits and treatments.

To understand the local failure of the implementation of a URI clinical guideline targeted at reducing unnecessary patient visits and antibiotic prescriptions, we conducted a study of the characteristics of patients who seek care for URI symptoms.¹⁴ A study of that guideline’s implementation had demonstrated that only 13% of primary care patients with respiratory symptoms were eligible for guideline care, and that subgroup did not show a decrease in clinic visits, antibiotic use, or care costs after guideline implementation.¹⁵

Our study provided an opportunity to better understand the adult patient or parent who seeks care particularly early in the course of the illness. Our research question was: Are those patients or parents who make contact with their care providers particularly early in the course of the illness different from those who make contact later in any way that should affect the way in which they are approached by clinicians or nurses? We hypothesized that the illnesses of these patients are different or that they have different reasons for seeking care from those who present later in the course of the illness. We felt that if clinicians understood these differences, perhaps they could help these patients more effectively, and care systems could reduce unnecessary services and antibiotic usage.

Methods

To test these hypotheses we conducted a telephone survey of cross-sectional samples of adults or parents of children seeking care for URI symptoms at one of 3 diverse primary care clinics in the Minneapolis/St. Paul region of Minnesota: a primary care site for a large multispecialty group practice in an affluent suburb, a staff-model health maintenance organization clinic in a blue-collar suburb, and a medium-sized primary care medical group in a nearby town.

To be included in the sample, adults had to be from 18 to 64 years old or parents of children aged 3 months to 17 years. The patients needed to have a runny nose, cough, or sore throat and to have called or visited the clinic within 14 days (adults) or 10 days (parents) of the onset of symptoms during the study period (March 1, 1997, to May 1, 1997). The subjects were excluded if they reported poor general health, ear pain or infection, or did not speak English. Eligible people were identified from each clinic’s logs of walk-in visits, scheduled appointments, or calls to get advice.

A telephone interview was conducted between 48 and 96 hours after the initial contact with the clinic. Patient selection continued until at least 80 adults and 80 parents of children had participated from each site. The interview consisted of 88 questions devised by study investigators and revised after interviews with 20 patients. Additional pretesting was done with the first 15 patients identified in one clinic. A second 15-item follow-up interview was conducted with the subset of participants who had initially made contact with the clinic within 7 days (adults) or 5 days (parents) of the onset of symptoms. This follow-up interview was made on day 14 (adults) or day 10 (parents) of symptom onset to assess the episode of illness more completely.

Because the dependent variable of interest was the number of days between the onset of illness and the patient’s contact with the clinic, subjects were divided into 3 approximately equal groups. Because contact in the first 2 days of illness may be the most difficult for clinicians to understand and because McIsaac and coworkers16 showed that adults with URI symptoms for more than 2 days were 2.7 times more likely to visit the physician than those sick for 1 to 2 days, the members of this group were designated as the early callers. These callers were compared with those who waited to contact their clinic until between 3 and 5 days or after 5 days of symptoms (who constitute the other 2 groups). Because nearly all subjects had made both a call and a visit (usually within a very short period of time), it was not possible to clearly separate the response into these 2 groups in the analysis.

Descriptive statistics were examined separately for each of the study variables for adult patients and parents. We used the chi-square test of association to compare categorical variables for the 3 illness duration subgroups within each age group.

Results

Interviews were completed with 257 adult patients and 249 parents of child patients (with completion rates of those who met the selection criteria of 94% and 90%, respectively). The sample size for each subgroup by duration of illness and the demographic characteristics of these individuals are listed in Table 1.

Twenty-eight percent of the adult patients and 41% of the child patients had been ill only 1 to 2 days at the time of their first contact with the health care system. None of the characteristics listed in Table 1 differed significantly among these groups, except that parents calling within 2 days of their child’s illness were somewhat less likely to have a college education than parents calling later.

Table 2 illustrates that early callers (in the first 2 days) did not have symptoms that were more frequent or severe than the other groups. In fact, children in this group were less likely to have a cough or green nasal drainage than those in the groups that made contact later (after 3 or more days). Fever, however, was less frequent in both adults and children of parents who sought care after 5 days.

Adult patients and parents who contacted the clinic early were not more likely to report being unsure that the symptoms represented a cold or being unsure how to treat the symptoms than those coming in after the first 2 days. There was also no greater likelihood that either adults or parents in the early calling group would report a history of any of the following: colds lasting longer; colds being complicated by sinusitis, otitis, or streptococcal infections; seeing a physician for a cold or being told by a physician to be seen; or receiving antibiotics for cold symptoms. Parents (but not adult patients) who made contact early, however, were significantly more likely to report a history of having recovered faster with an antibiotic prescription in the past (42% vs 22% and 23%, P=.01).

The only other differences among these 3 groups were in what they were seeking from the contact and whether they felt that the cold had lasted too long. Adult patients (but not parents) who sought care early were more likely to be worried about complications, while those who appeared later were more likely to want antibiotics and relief (Table 3).

Both adult patients and parents who made contact later were more likely to feel that the cold had lasted too long. For adult patients, 87% of groups that made contact later—but only 46% of those who made contact early—felt that the illness had been going on too long (P=.001). This finding was similar for parents of sick children but to a lesser degree (78% vs 58%, P=.01).

There were no differences in knowledge or beliefs about colds among these 3 groups for either adult patients or parents. Although more than 80% of all groups believed viruses cause colds, 50% believed bacteria also can cause them, and 80% agreed that getting tired and rundown causes colds. Eighty-five percent thought that colds resolve on their own, but 97% thought that rest helps, 66% felt that steam or Vitamin C helps, and nearly half believed in the value of chicken soup.

Finally, early callers were no more likely than those who came in later to report attitudes toward medical care or satisfaction with the visit that might affect their timing of medical care use. Thus, the duration of illness had no effect on respondent reactions to any of the following statements: (1) I will do just about anything to avoid going to the doctor; (2) when I am sick, I try not to let others know; (3) I usually go to the doctor as soon as I start to feel bad; (4) I have paid for my health insurance, so I might as well use it; (5) overall, I liked the manner in which my/my child’s problem was handled; (6) I trust the advice I was given for this illness; and (7) I would recommend this clinic to family or friends for an illness like this.

What about antibiotic use? Although 65% of patients received antibiotics within the 14-day observation period of our study, there was no relationship between receipt of antibiotics and the duration of symptoms at the time of first contact.

Discussion

Contrary to our original hypotheses and the expectations of many clinicians, symptomatic adults and parents of symptomatic children who seek medical assistance soon after the development of URI symptoms do not appear to be different in any important way from those making contact later. They do not have different demographic characteristics, beliefs about colds, or past experiences that might lead them to seek this very early contact. They are also not more likely to have different health status or to have different or more severe symptoms than those seeking care later in their illness (except for a somewhat greater frequency of fever). Although they are more likely to be concerned about complications, they are not any more likely to be unsure about the diagnosis or treatment and do not seem to want anything from the health care system that is different from those making contact later. Specifically, they are not any more likely to want antibiotics for their illness and are not more likely to receive them, despite the fact that parents who call or visit in the first 2 days are twice as likely to report that their child recovered more quickly with antibiotics in the past. Finally, they do not report attitudes toward medical care or satisfaction with the care of this illness that are any different from those who make later contact.

Because nearly one third of adult patients and two fifths of the parents of child patients seek care too early for the clinician to be very confident about the course of the illness, what is it that leads them to this action? There may be some reason that was not addressed in our survey, or the respondents may not feel comfortable admitting to their reasons in an interview sponsored by the health system. It seems more likely, however, that they are simply somewhat more eager to bring health problems to medical attention. The evidence-based guideline that led to our study suggests that almost all the patients in any of the groups we compared are seeking care too early.¹⁷

The overall rate at which antibiotics were prescribed for these patients (65%) is similar to that reported by most other studies.^1,4-5 One of the principal aims of the guideline that led to our study was reduction of unnecessary antibiotic use, along with encouragement for initial telephone care instead of office visits for uncomplicated cases.¹⁷ The guideline also recommended contact only under these conditions: worsening or new symptoms after the first 3 to 5 days and lack of improvement after 7 to 10 days (children) or 14 days (adults).

However, an impact study of that guideline’s implementation by O’Connor and colleagues¹⁵ found that patients who received telephone advice were 50% more likely to come in for an office visit after guideline implementation than before. Also, although antibiotic use at initial contact for the diagnosis of a URI fell from 24% to 16%, the likelihood of receiving an antibiotic during a 21-day follow-up period actually rose after the implementation of the guideline. O’Connor and coworkers concluded that the failure of the guideline implementation to achieve its aims was the result of conflict with patient expectations and desires. Because only those adult patients with illnesses of more than 5 days’ duration reported desires for antibiotic prescriptions at a rate as high as that for actual use, patient desires do not seem to be the only driver of antibiotic use.

Limitations

Our analysis is limited by its inability to separate those who called from those who visited, although the usual concurrence of those events makes any such study difficult. It is also limited in that it is focused entirely on patient responses to a telephone survey, with no additional information about the illness diagnoses, care, or outcomes from chart audits.

Conclusions

There were no obvious explanations in our survey results for the decision by a sizable minority of those seeking medical attention for URI symptoms to do so very soon after the onset of their symptoms. Clinicians will need to elicit and address the chief concerns and needs of each patient they see, regardless of the timing of that contact. It is possible that providing more general information to potential patients about the nature and care of URIs will help health care systems delay or reduce the perceived need of these patients for such contacts, but that remains to be proved. It is time for researchers to move into trials of various intervention strategies that may reduce both unnecessary visits and the associated antibiotic use.

Acknowledgments

Our project was supported by a grant from the Institute for Clinical Systems Improvement, Minneapolis, Minnesota. We are also grateful to the clinic staff who cooperated with our need to identify patients making contact for their URI symptoms.

Rising health care costs (especially for medications) and a growing concern that unnecessary antibiotic use is creating troublesome bacterial resistance patterns have triggered increasing attention to upper respiratory infection (URI) care.¹ URIs are the most frequent reason for office visits and antibiotic use, and at least 50% of patient visits with the diagnosis of URI, cold, or bronchitis include the writing of antibiotic prescriptions.^1-2 Also, there is reason to believe that antibiotic use for URI symptoms has increased during the past 2 decades, despite increasing public health efforts to the contrary.³

Much of the expanding literature on URI patients is descriptive, attempting to provide understanding of their expectations and the relationship of various characteristics to antibiotic use for this illness.^4-12 Because of the belief that it is patients who create most of the visits and antibiotic use, most studies of URI care address patients’ perceptions, attitudes, and satisfaction.^2,4-12 Some studies, however, have shown that the physician’s perception that a patient expects antibiotics is the strongest predictor of prescriptions, even though that perception is not always correct.^4-5 This literature suggests that once patients with URIs appear in the clinic it is likely that they will leave with an antibiotic prescription, either because the clinician feels that the illness might be helped by antibiotics or because he or she believes that the patient will be unhappy if antibiotics are not provided. Miller and colleagues¹³ obtained physician questionnaires for patients with a suspected infection as the reason for the visit. They found that physicians responded to their perception that patients wanted an antibiotic only when the physicians were uncertain about the need for it. When they were fairly certain that an antibiotic was either needed or not needed, perceived patient demand was rarely a factor. Patient satisfaction does not appear to be related to the receipt of antibiotics even when they are expected.^5-6

Patients who call or come in within the first few days of the onset of their illness constitute a particularly troubling subgroup of patients for most clinicians. Because it is usually difficult for a clinician to predict whether patients with early URI symptoms are going to encounter complications, it is hard for the clinician to avoid assuming that these patients are either sicker or unusually desirous of a test or treatment. In the absence of any studies of this subgroup, it is possible that those making early contact are particularly likely to receive antibiotics. They will continue to frustrate clinicians and to confound health care system efforts to reduce unnecessary visits and treatments.

To understand the local failure of the implementation of a URI clinical guideline targeted at reducing unnecessary patient visits and antibiotic prescriptions, we conducted a study of the characteristics of patients who seek care for URI symptoms.¹⁴ A study of that guideline’s implementation had demonstrated that only 13% of primary care patients with respiratory symptoms were eligible for guideline care, and that subgroup did not show a decrease in clinic visits, antibiotic use, or care costs after guideline implementation.¹⁵

Our study provided an opportunity to better understand the adult patient or parent who seeks care particularly early in the course of the illness. Our research question was: Are those patients or parents who make contact with their care providers particularly early in the course of the illness different from those who make contact later in any way that should affect the way in which they are approached by clinicians or nurses? We hypothesized that the illnesses of these patients are different or that they have different reasons for seeking care from those who present later in the course of the illness. We felt that if clinicians understood these differences, perhaps they could help these patients more effectively, and care systems could reduce unnecessary services and antibiotic usage.

Methods

To test these hypotheses we conducted a telephone survey of cross-sectional samples of adults or parents of children seeking care for URI symptoms at one of 3 diverse primary care clinics in the Minneapolis/St. Paul region of Minnesota: a primary care site for a large multispecialty group practice in an affluent suburb, a staff-model health maintenance organization clinic in a blue-collar suburb, and a medium-sized primary care medical group in a nearby town.

To be included in the sample, adults had to be from 18 to 64 years old or parents of children aged 3 months to 17 years. The patients needed to have a runny nose, cough, or sore throat and to have called or visited the clinic within 14 days (adults) or 10 days (parents) of the onset of symptoms during the study period (March 1, 1997, to May 1, 1997). The subjects were excluded if they reported poor general health, ear pain or infection, or did not speak English. Eligible people were identified from each clinic’s logs of walk-in visits, scheduled appointments, or calls to get advice.

A telephone interview was conducted between 48 and 96 hours after the initial contact with the clinic. Patient selection continued until at least 80 adults and 80 parents of children had participated from each site. The interview consisted of 88 questions devised by study investigators and revised after interviews with 20 patients. Additional pretesting was done with the first 15 patients identified in one clinic. A second 15-item follow-up interview was conducted with the subset of participants who had initially made contact with the clinic within 7 days (adults) or 5 days (parents) of the onset of symptoms. This follow-up interview was made on day 14 (adults) or day 10 (parents) of symptom onset to assess the episode of illness more completely.

Because the dependent variable of interest was the number of days between the onset of illness and the patient’s contact with the clinic, subjects were divided into 3 approximately equal groups. Because contact in the first 2 days of illness may be the most difficult for clinicians to understand and because McIsaac and coworkers16 showed that adults with URI symptoms for more than 2 days were 2.7 times more likely to visit the physician than those sick for 1 to 2 days, the members of this group were designated as the early callers. These callers were compared with those who waited to contact their clinic until between 3 and 5 days or after 5 days of symptoms (who constitute the other 2 groups). Because nearly all subjects had made both a call and a visit (usually within a very short period of time), it was not possible to clearly separate the response into these 2 groups in the analysis.

Descriptive statistics were examined separately for each of the study variables for adult patients and parents. We used the chi-square test of association to compare categorical variables for the 3 illness duration subgroups within each age group.

Results

Interviews were completed with 257 adult patients and 249 parents of child patients (with completion rates of those who met the selection criteria of 94% and 90%, respectively). The sample size for each subgroup by duration of illness and the demographic characteristics of these individuals are listed in Table 1.

Twenty-eight percent of the adult patients and 41% of the child patients had been ill only 1 to 2 days at the time of their first contact with the health care system. None of the characteristics listed in Table 1 differed significantly among these groups, except that parents calling within 2 days of their child’s illness were somewhat less likely to have a college education than parents calling later.

Table 2 illustrates that early callers (in the first 2 days) did not have symptoms that were more frequent or severe than the other groups. In fact, children in this group were less likely to have a cough or green nasal drainage than those in the groups that made contact later (after 3 or more days). Fever, however, was less frequent in both adults and children of parents who sought care after 5 days.

Adult patients and parents who contacted the clinic early were not more likely to report being unsure that the symptoms represented a cold or being unsure how to treat the symptoms than those coming in after the first 2 days. There was also no greater likelihood that either adults or parents in the early calling group would report a history of any of the following: colds lasting longer; colds being complicated by sinusitis, otitis, or streptococcal infections; seeing a physician for a cold or being told by a physician to be seen; or receiving antibiotics for cold symptoms. Parents (but not adult patients) who made contact early, however, were significantly more likely to report a history of having recovered faster with an antibiotic prescription in the past (42% vs 22% and 23%, P=.01).

The only other differences among these 3 groups were in what they were seeking from the contact and whether they felt that the cold had lasted too long. Adult patients (but not parents) who sought care early were more likely to be worried about complications, while those who appeared later were more likely to want antibiotics and relief (Table 3).

Both adult patients and parents who made contact later were more likely to feel that the cold had lasted too long. For adult patients, 87% of groups that made contact later—but only 46% of those who made contact early—felt that the illness had been going on too long (P=.001). This finding was similar for parents of sick children but to a lesser degree (78% vs 58%, P=.01).

There were no differences in knowledge or beliefs about colds among these 3 groups for either adult patients or parents. Although more than 80% of all groups believed viruses cause colds, 50% believed bacteria also can cause them, and 80% agreed that getting tired and rundown causes colds. Eighty-five percent thought that colds resolve on their own, but 97% thought that rest helps, 66% felt that steam or Vitamin C helps, and nearly half believed in the value of chicken soup.

Finally, early callers were no more likely than those who came in later to report attitudes toward medical care or satisfaction with the visit that might affect their timing of medical care use. Thus, the duration of illness had no effect on respondent reactions to any of the following statements: (1) I will do just about anything to avoid going to the doctor; (2) when I am sick, I try not to let others know; (3) I usually go to the doctor as soon as I start to feel bad; (4) I have paid for my health insurance, so I might as well use it; (5) overall, I liked the manner in which my/my child’s problem was handled; (6) I trust the advice I was given for this illness; and (7) I would recommend this clinic to family or friends for an illness like this.

What about antibiotic use? Although 65% of patients received antibiotics within the 14-day observation period of our study, there was no relationship between receipt of antibiotics and the duration of symptoms at the time of first contact.

Discussion

Contrary to our original hypotheses and the expectations of many clinicians, symptomatic adults and parents of symptomatic children who seek medical assistance soon after the development of URI symptoms do not appear to be different in any important way from those making contact later. They do not have different demographic characteristics, beliefs about colds, or past experiences that might lead them to seek this very early contact. They are also not more likely to have different health status or to have different or more severe symptoms than those seeking care later in their illness (except for a somewhat greater frequency of fever). Although they are more likely to be concerned about complications, they are not any more likely to be unsure about the diagnosis or treatment and do not seem to want anything from the health care system that is different from those making contact later. Specifically, they are not any more likely to want antibiotics for their illness and are not more likely to receive them, despite the fact that parents who call or visit in the first 2 days are twice as likely to report that their child recovered more quickly with antibiotics in the past. Finally, they do not report attitudes toward medical care or satisfaction with the care of this illness that are any different from those who make later contact.

Because nearly one third of adult patients and two fifths of the parents of child patients seek care too early for the clinician to be very confident about the course of the illness, what is it that leads them to this action? There may be some reason that was not addressed in our survey, or the respondents may not feel comfortable admitting to their reasons in an interview sponsored by the health system. It seems more likely, however, that they are simply somewhat more eager to bring health problems to medical attention. The evidence-based guideline that led to our study suggests that almost all the patients in any of the groups we compared are seeking care too early.¹⁷

The overall rate at which antibiotics were prescribed for these patients (65%) is similar to that reported by most other studies.^1,4-5 One of the principal aims of the guideline that led to our study was reduction of unnecessary antibiotic use, along with encouragement for initial telephone care instead of office visits for uncomplicated cases.¹⁷ The guideline also recommended contact only under these conditions: worsening or new symptoms after the first 3 to 5 days and lack of improvement after 7 to 10 days (children) or 14 days (adults).

However, an impact study of that guideline’s implementation by O’Connor and colleagues¹⁵ found that patients who received telephone advice were 50% more likely to come in for an office visit after guideline implementation than before. Also, although antibiotic use at initial contact for the diagnosis of a URI fell from 24% to 16%, the likelihood of receiving an antibiotic during a 21-day follow-up period actually rose after the implementation of the guideline. O’Connor and coworkers concluded that the failure of the guideline implementation to achieve its aims was the result of conflict with patient expectations and desires. Because only those adult patients with illnesses of more than 5 days’ duration reported desires for antibiotic prescriptions at a rate as high as that for actual use, patient desires do not seem to be the only driver of antibiotic use.

Limitations

Our analysis is limited by its inability to separate those who called from those who visited, although the usual concurrence of those events makes any such study difficult. It is also limited in that it is focused entirely on patient responses to a telephone survey, with no additional information about the illness diagnoses, care, or outcomes from chart audits.

Conclusions

There were no obvious explanations in our survey results for the decision by a sizable minority of those seeking medical attention for URI symptoms to do so very soon after the onset of their symptoms. Clinicians will need to elicit and address the chief concerns and needs of each patient they see, regardless of the timing of that contact. It is possible that providing more general information to potential patients about the nature and care of URIs will help health care systems delay or reduce the perceived need of these patients for such contacts, but that remains to be proved. It is time for researchers to move into trials of various intervention strategies that may reduce both unnecessary visits and the associated antibiotic use.

Acknowledgments

Our project was supported by a grant from the Institute for Clinical Systems Improvement, Minneapolis, Minnesota. We are also grateful to the clinic staff who cooperated with our need to identify patients making contact for their URI symptoms.

Rising health care costs (especially for medications) and a growing concern that unnecessary antibiotic use is creating troublesome bacterial resistance patterns have triggered increasing attention to upper respiratory infection (URI) care.¹ URIs are the most frequent reason for office visits and antibiotic use, and at least 50% of patient visits with the diagnosis of URI, cold, or bronchitis include the writing of antibiotic prescriptions.^1-2 Also, there is reason to believe that antibiotic use for URI symptoms has increased during the past 2 decades, despite increasing public health efforts to the contrary.³

Much of the expanding literature on URI patients is descriptive, attempting to provide understanding of their expectations and the relationship of various characteristics to antibiotic use for this illness.^4-12 Because of the belief that it is patients who create most of the visits and antibiotic use, most studies of URI care address patients’ perceptions, attitudes, and satisfaction.^2,4-12 Some studies, however, have shown that the physician’s perception that a patient expects antibiotics is the strongest predictor of prescriptions, even though that perception is not always correct.^4-5 This literature suggests that once patients with URIs appear in the clinic it is likely that they will leave with an antibiotic prescription, either because the clinician feels that the illness might be helped by antibiotics or because he or she believes that the patient will be unhappy if antibiotics are not provided. Miller and colleagues¹³ obtained physician questionnaires for patients with a suspected infection as the reason for the visit. They found that physicians responded to their perception that patients wanted an antibiotic only when the physicians were uncertain about the need for it. When they were fairly certain that an antibiotic was either needed or not needed, perceived patient demand was rarely a factor. Patient satisfaction does not appear to be related to the receipt of antibiotics even when they are expected.^5-6

Patients who call or come in within the first few days of the onset of their illness constitute a particularly troubling subgroup of patients for most clinicians. Because it is usually difficult for a clinician to predict whether patients with early URI symptoms are going to encounter complications, it is hard for the clinician to avoid assuming that these patients are either sicker or unusually desirous of a test or treatment. In the absence of any studies of this subgroup, it is possible that those making early contact are particularly likely to receive antibiotics. They will continue to frustrate clinicians and to confound health care system efforts to reduce unnecessary visits and treatments.

To understand the local failure of the implementation of a URI clinical guideline targeted at reducing unnecessary patient visits and antibiotic prescriptions, we conducted a study of the characteristics of patients who seek care for URI symptoms.¹⁴ A study of that guideline’s implementation had demonstrated that only 13% of primary care patients with respiratory symptoms were eligible for guideline care, and that subgroup did not show a decrease in clinic visits, antibiotic use, or care costs after guideline implementation.¹⁵

Our study provided an opportunity to better understand the adult patient or parent who seeks care particularly early in the course of the illness. Our research question was: Are those patients or parents who make contact with their care providers particularly early in the course of the illness different from those who make contact later in any way that should affect the way in which they are approached by clinicians or nurses? We hypothesized that the illnesses of these patients are different or that they have different reasons for seeking care from those who present later in the course of the illness. We felt that if clinicians understood these differences, perhaps they could help these patients more effectively, and care systems could reduce unnecessary services and antibiotic usage.

Methods

To test these hypotheses we conducted a telephone survey of cross-sectional samples of adults or parents of children seeking care for URI symptoms at one of 3 diverse primary care clinics in the Minneapolis/St. Paul region of Minnesota: a primary care site for a large multispecialty group practice in an affluent suburb, a staff-model health maintenance organization clinic in a blue-collar suburb, and a medium-sized primary care medical group in a nearby town.

To be included in the sample, adults had to be from 18 to 64 years old or parents of children aged 3 months to 17 years. The patients needed to have a runny nose, cough, or sore throat and to have called or visited the clinic within 14 days (adults) or 10 days (parents) of the onset of symptoms during the study period (March 1, 1997, to May 1, 1997). The subjects were excluded if they reported poor general health, ear pain or infection, or did not speak English. Eligible people were identified from each clinic’s logs of walk-in visits, scheduled appointments, or calls to get advice.

A telephone interview was conducted between 48 and 96 hours after the initial contact with the clinic. Patient selection continued until at least 80 adults and 80 parents of children had participated from each site. The interview consisted of 88 questions devised by study investigators and revised after interviews with 20 patients. Additional pretesting was done with the first 15 patients identified in one clinic. A second 15-item follow-up interview was conducted with the subset of participants who had initially made contact with the clinic within 7 days (adults) or 5 days (parents) of the onset of symptoms. This follow-up interview was made on day 14 (adults) or day 10 (parents) of symptom onset to assess the episode of illness more completely.

Because the dependent variable of interest was the number of days between the onset of illness and the patient’s contact with the clinic, subjects were divided into 3 approximately equal groups. Because contact in the first 2 days of illness may be the most difficult for clinicians to understand and because McIsaac and coworkers16 showed that adults with URI symptoms for more than 2 days were 2.7 times more likely to visit the physician than those sick for 1 to 2 days, the members of this group were designated as the early callers. These callers were compared with those who waited to contact their clinic until between 3 and 5 days or after 5 days of symptoms (who constitute the other 2 groups). Because nearly all subjects had made both a call and a visit (usually within a very short period of time), it was not possible to clearly separate the response into these 2 groups in the analysis.

Descriptive statistics were examined separately for each of the study variables for adult patients and parents. We used the chi-square test of association to compare categorical variables for the 3 illness duration subgroups within each age group.

Results

Interviews were completed with 257 adult patients and 249 parents of child patients (with completion rates of those who met the selection criteria of 94% and 90%, respectively). The sample size for each subgroup by duration of illness and the demographic characteristics of these individuals are listed in Table 1.

Twenty-eight percent of the adult patients and 41% of the child patients had been ill only 1 to 2 days at the time of their first contact with the health care system. None of the characteristics listed in Table 1 differed significantly among these groups, except that parents calling within 2 days of their child’s illness were somewhat less likely to have a college education than parents calling later.

Table 2 illustrates that early callers (in the first 2 days) did not have symptoms that were more frequent or severe than the other groups. In fact, children in this group were less likely to have a cough or green nasal drainage than those in the groups that made contact later (after 3 or more days). Fever, however, was less frequent in both adults and children of parents who sought care after 5 days.

Adult patients and parents who contacted the clinic early were not more likely to report being unsure that the symptoms represented a cold or being unsure how to treat the symptoms than those coming in after the first 2 days. There was also no greater likelihood that either adults or parents in the early calling group would report a history of any of the following: colds lasting longer; colds being complicated by sinusitis, otitis, or streptococcal infections; seeing a physician for a cold or being told by a physician to be seen; or receiving antibiotics for cold symptoms. Parents (but not adult patients) who made contact early, however, were significantly more likely to report a history of having recovered faster with an antibiotic prescription in the past (42% vs 22% and 23%, P=.01).

The only other differences among these 3 groups were in what they were seeking from the contact and whether they felt that the cold had lasted too long. Adult patients (but not parents) who sought care early were more likely to be worried about complications, while those who appeared later were more likely to want antibiotics and relief (Table 3).

Both adult patients and parents who made contact later were more likely to feel that the cold had lasted too long. For adult patients, 87% of groups that made contact later—but only 46% of those who made contact early—felt that the illness had been going on too long (P=.001). This finding was similar for parents of sick children but to a lesser degree (78% vs 58%, P=.01).

There were no differences in knowledge or beliefs about colds among these 3 groups for either adult patients or parents. Although more than 80% of all groups believed viruses cause colds, 50% believed bacteria also can cause them, and 80% agreed that getting tired and rundown causes colds. Eighty-five percent thought that colds resolve on their own, but 97% thought that rest helps, 66% felt that steam or Vitamin C helps, and nearly half believed in the value of chicken soup.

Finally, early callers were no more likely than those who came in later to report attitudes toward medical care or satisfaction with the visit that might affect their timing of medical care use. Thus, the duration of illness had no effect on respondent reactions to any of the following statements: (1) I will do just about anything to avoid going to the doctor; (2) when I am sick, I try not to let others know; (3) I usually go to the doctor as soon as I start to feel bad; (4) I have paid for my health insurance, so I might as well use it; (5) overall, I liked the manner in which my/my child’s problem was handled; (6) I trust the advice I was given for this illness; and (7) I would recommend this clinic to family or friends for an illness like this.

What about antibiotic use? Although 65% of patients received antibiotics within the 14-day observation period of our study, there was no relationship between receipt of antibiotics and the duration of symptoms at the time of first contact.

Discussion

Contrary to our original hypotheses and the expectations of many clinicians, symptomatic adults and parents of symptomatic children who seek medical assistance soon after the development of URI symptoms do not appear to be different in any important way from those making contact later. They do not have different demographic characteristics, beliefs about colds, or past experiences that might lead them to seek this very early contact. They are also not more likely to have different health status or to have different or more severe symptoms than those seeking care later in their illness (except for a somewhat greater frequency of fever). Although they are more likely to be concerned about complications, they are not any more likely to be unsure about the diagnosis or treatment and do not seem to want anything from the health care system that is different from those making contact later. Specifically, they are not any more likely to want antibiotics for their illness and are not more likely to receive them, despite the fact that parents who call or visit in the first 2 days are twice as likely to report that their child recovered more quickly with antibiotics in the past. Finally, they do not report attitudes toward medical care or satisfaction with the care of this illness that are any different from those who make later contact.

Because nearly one third of adult patients and two fifths of the parents of child patients seek care too early for the clinician to be very confident about the course of the illness, what is it that leads them to this action? There may be some reason that was not addressed in our survey, or the respondents may not feel comfortable admitting to their reasons in an interview sponsored by the health system. It seems more likely, however, that they are simply somewhat more eager to bring health problems to medical attention. The evidence-based guideline that led to our study suggests that almost all the patients in any of the groups we compared are seeking care too early.¹⁷

The overall rate at which antibiotics were prescribed for these patients (65%) is similar to that reported by most other studies.^1,4-5 One of the principal aims of the guideline that led to our study was reduction of unnecessary antibiotic use, along with encouragement for initial telephone care instead of office visits for uncomplicated cases.¹⁷ The guideline also recommended contact only under these conditions: worsening or new symptoms after the first 3 to 5 days and lack of improvement after 7 to 10 days (children) or 14 days (adults).

However, an impact study of that guideline’s implementation by O’Connor and colleagues¹⁵ found that patients who received telephone advice were 50% more likely to come in for an office visit after guideline implementation than before. Also, although antibiotic use at initial contact for the diagnosis of a URI fell from 24% to 16%, the likelihood of receiving an antibiotic during a 21-day follow-up period actually rose after the implementation of the guideline. O’Connor and coworkers concluded that the failure of the guideline implementation to achieve its aims was the result of conflict with patient expectations and desires. Because only those adult patients with illnesses of more than 5 days’ duration reported desires for antibiotic prescriptions at a rate as high as that for actual use, patient desires do not seem to be the only driver of antibiotic use.

Limitations

Our analysis is limited by its inability to separate those who called from those who visited, although the usual concurrence of those events makes any such study difficult. It is also limited in that it is focused entirely on patient responses to a telephone survey, with no additional information about the illness diagnoses, care, or outcomes from chart audits.

Conclusions

There were no obvious explanations in our survey results for the decision by a sizable minority of those seeking medical attention for URI symptoms to do so very soon after the onset of their symptoms. Clinicians will need to elicit and address the chief concerns and needs of each patient they see, regardless of the timing of that contact. It is possible that providing more general information to potential patients about the nature and care of URIs will help health care systems delay or reduce the perceived need of these patients for such contacts, but that remains to be proved. It is time for researchers to move into trials of various intervention strategies that may reduce both unnecessary visits and the associated antibiotic use.

Acknowledgments

Our project was supported by a grant from the Institute for Clinical Systems Improvement, Minneapolis, Minnesota. We are also grateful to the clinic staff who cooperated with our need to identify patients making contact for their URI symptoms.

A large body of literature has addressed the frequent use of antibiotics for common upper and lower respiratory tract infections in the outpatient setting. The many dangers of this practice, including the development of bacterial resistance,^1,2 adverse drug reactions,^3-5 and negative financial implications,⁶ have been discussed, but very few methods to resolve the problem have been tested. Guidelines and educational strategies have been touted by some advocates as an essential part of the solution process and have been shown to have some degree of success in specific settings.^7,8 Recently, the American Academy of Family Physicians, the American Academy of Pediatrics, and the Centers for Disease Control and Prevention have collaborated to develop a set of recommendations to help clinicians use antibiotics more appropriately when treating patients with common respiratory illnesses.^9-13 The results of these educational efforts have not been evaluated.

Although many aspects of the antibiotic overprescribing issue may start with physician beliefs, training, and practice setting, other factors have been postulated, including meeting patient expectations, economic issues, and time constraints. A practice that has not been reported in the literature but has been employed by many physicians (anecdotally) is the use of a back-up prescription strategy. This approach addresses the complex problem of satisfying patients in a timely manner while re-educating them in a nonconfrontational way.

The term “back-up prescription” applies to the writing of a prescription that is to be filled at a later time, and only if the patient’s condition deteriorates or fails to improve. At the time the prescription is written the physician explains to the patient (or the family) the reasons for not giving an immediate-fill antibiotic prescription and gives advice on symptomatic treatment for the current problem. Additionally, specific guidance is given on clinical parameters and the timing of when to fill the prescription if the condition progresses. To our knowledge, this practice has not been studied adequately in family medicine settings although similar strategies have been used for other health-related conditions.^14-16 The use of back-up treatment for malaria (also referred to as reserve treatment) has been mentioned in the literature.¹⁵ Davy and colleagues¹⁶ have also reported the use of back-up antibiotics for the treatment of undifferentiated acute respiratory tract infection with cough among primary care family physicians and pediatricians. This study, which was based on a self-reported survey, primarily sought to identify the frequency with which reserve antibiotics were prescribed to this group of children. It did not address the actual practice of using a back-up prescription.

In exploring the use of the back-up antibiotic prescription strategy it is essential to assess the degree of patient satisfaction and the fill rates for back-up prescriptions and their predictors. In their study on patient satisfaction and antibiotic prescriptions for respiratory infections, Hamm and coworkers¹⁷ elicited patient satisfaction levels immediately following a physician encounter. However, seeking the opinions of patients about their satisfaction immediately after an encounter may not yield accurate responses because more time may be required to assess other factors such as the effect (or lack of effect) of the treatment suggested, including antibiotic prescriptions given.

Methods

Study Design and Setting

We performed an observational study on the current prescribing habits of our physicians. The study included prospective data collection on the use of a back-up antibiotic prescription strategy among patients presenting with complaints of common respiratory symptoms to 28 physicians and 2 physician extenders in 3 family practice clinics between January and April 1999. These clinics are part of the Scott & White Healthcare System and are located in Temple (Santa Fe Clinic), Waco, and Killeen, Texas.

The practice of providing back-up antibiotic prescriptions was regularly used by many of the physicians as part of their routine management options, while others were unfamiliar with the concept or used it only rarely. All physicians were advised of the study objectives and were encouraged to enroll patients who received back-up antibiotic prescriptions. However, the physicians were not asked to change their usual prescribing habits. The study protocol was approved by the Institutional Review Board of the Scott & White Memorial Hospital and Clinic.

Study Participants and Data Collection

A concerted effort was made to enroll all patients who presented with complaints of a head cold or respiratory symptoms during this 4-month period. Our inclusion criteria were strict but broad. Patients were enrolled in the study if they had head congestion, sinus congestion, fever, headache, cough, chest congestion, or sore throat. Patients were only excluded if they had one dominant symptom and physical finding, such as earache. In addition to the front desk personnel, physicians and nurses could also enroll patients in the study if the patient brought up the need for treatment of respiratory complaints that were not mentioned to the appointment clerks (eg, “Oh, by the way, while I am here for my blood pressure follow-up, would you check out my head cold. I think I may be coming down with something and thought maybe I should get some antibiotics.”).

When patients reported for their appointments, a physician survey was attached to the front of their chart by front desk office personnel. This survey could also be added to the chart when the patient was put into an examination room if the nurse was made aware of the patient’s expectation for evaluation of respiratory symptoms. The physician survey which was filled out at the conclusion of the office visit elicited information regarding: (1) physician and patient demographic information; (2) the patient’s primary complaints; (3) whether the physician was the patient’s primary care physician; (4) type of prescription given to the patient (an immediate-fill antibiotic prescription, a back-up antibiotic prescription, or no antibiotic prescription); and (5) physician subjective rating, on a 5-point scale, of the clinical necessity for prescribing antibiotics for the patient.

The patients who were given back-up antibiotic prescriptions were each given a patient survey to complete with instructions to return the form in a provided preaddressed envelope 7 days after their initial appointment. Patients who did not return their surveys were called by the research coordinator, and the surveys were completed over the phone.

The patient survey included questions about: (1) patient satisfaction with the care received; (2) their perceived need before the office visit for an antibiotic prescription; (3) whether they received a written back-up antibiotic prescription; (4) whether they filled the back-up prescription; and (5) whether they required any subsequent medical care for the same illness.

Definition of the Back-up Strategy

A back-up antibiotic prescription strategy was defined in our study as a prescription given to a patient along with instructions to fill the prescription only if the condition deteriorated or failed to improve within a predefined number of days. The exact number of days was not standardized by the study protocol, allowing each physician to customize this aspect of care.

Statistical Analysis

Data management and analysis were performed using SAS¹⁸ on a mainframe and the Statistical Package for the Social Sciences¹⁹ on a personal computer. We determined physicians’ use of the back-up antibiotic prescription strategy using selected variables by comparing study subjects who received back-up antibiotic prescriptions with those given immediate-fill prescriptions. We computed crude odds ratios (ORs) and 95% confidence intervals (CIs) for use of the back-up prescription strategy. Variables that were statistically significant in their bivariate relationship with use of the back-up antibiotic prescription strategy and those with some biological plausibility (eg, patient age) were entered into a multivariate logistic regression modeling to compute adjusted ORs.

We also computed patient satisfaction and fill rates of back-up antibiotic prescriptions. Agreement between patients’ perceived need for antibiotics before the office visit and physicians’ subjective rating of the clinical necessity to prescribe antibiotics for patients was assessed using the k statistic. To determine correlates of patient satisfaction with the back-up prescription strategy, we compared satisfaction rates of study subjects by patient and physician characteristics, the presenting respiratory complaints, and several selected characteristics. Finally, correlates of back-up prescription filling were similarly determined by comparing filling rates by the same characteristics.

Group differences were assessed for significance using the chi-square statistic or Fisher’s exact test for categorical variables and analysis of variance for continuous variables. All tests were 2-tailed and were considered significant at P <.05.

Results

A total of 947 patients were evaluated for common respiratory symptoms by 19 family physicians, 2 physician extenders (a nurse practitioner and a physician assistant), and 9 family medicine residents.

Rates and Correlates of Back-up Antibiotic Prescriptions

From the 947 enrolled patients with common respiratory symptoms, 441 (46.6%) were not given antibiotics: 286 (30.2%) were given back-up antibiotic prescriptions, and 220 (23.2%) were given immediate-fill antibiotic prescriptions. Patients younger than 35 years and those with complaints of cough were twice as likely to be given back-up antibiotic prescriptions. Female sex and health care provider role as a physician extender were the only physician characteristics that were positively associated with the use of back-up prescriptions. Neither the role as primary care physician nor the physician’s number of years in practice were related to the type of prescription given.

Of 286 patients given back-up antibiotic prescriptions, we obtained completed follow-up surveys from 255 (89.2%). There were no significant differences between respondents and nonrespondents regarding demographic variables.

Rate and Correlates of Patient Satisfaction

Of the 255 patients who responded, 245 (96.1%) reported that they were satisfied with the care they received at their visit. The majority of the patients (76.1%) felt that their illness would require an antibiotic when their appointment was scheduled. However, only 36.9% of their physicians felt that their illness warranted the use of antibiotics. There was no significant agreement (P=.08) between patients’ perceived need for antibiotics before the office visit and physicians’ subjective rating of the clinical necessity to prescribe antibiotics (Table 1).

Patient and physicians characteristics were not associated with patient self-reported satisfaction rate with the care they received. Satisfaction rates were, however, significantly associated with patient complaints of sinus congestion (Table 2) and a patient’s requirement for additional care at a later time for the same illness (Table 3). Patients with complaints of sinus congestion and those who required additional care at a later time reported significantly less satisfaction.

Fill Rate and Correlates of Back-up Antibiotic Prescription

The overall back-up antibiotic prescription fill rate was 50.2%. Fill rates did not differ significantly by patient characteristics or their self-reported satisfaction with the care received, physician characteristics, or whether the physician was the patient’s primary care physician.

Additional Care

Additional care (defined as any subsequent contact with a health care provider) was required for 9.0% (n=23) of the patients in our study who received back-up antibiotic prescriptions. Of these, 10 consulted by telephone about their illness. Another 12 made repeat office visits, and 1 made an emergency room visit for an exacerbation of asthma; that patient was subsequently admitted overnight for management of her asthma. Of the 23 patients who sought additional care, 17 (74%) filled their back-up antibiotic prescriptions.

Discussion

Several factors are associated with the overprescription of antimicrobials for common respiratory symptoms, including physician specialty, physician knowledge base of the natural history of viral respiratory infections, clinician and patient experiences, patient expectations, and economic pressures related to time and reimbursement. Mainous and colleagues²⁰ and Nyquist and coworkers²¹ have reported that family physicians and general practitioners have prescribed antibiotics significantly more than pediatricians for children with upper respiratory infections (URIs). Schwartz and colleagues²² also conducted a survey based on a written case scenario that highlighted the significant discrepancy between the prescribing habits of family physicians and pediatricians. Compared with 53% of pediatricians, 71% of family physicians would immediately prescribe an antibiotic for a child who had a single day of scant light green and yellow nasal discharge and low-grade fever (P=.001).

Both clinician and patient experiences may also promote antibiotic overusage. If a patient has received an antibiotic for a URI in the past and had a good outcome, that positive experience creates an impression that antibiotic therapy is required and proper.²³ Similarly when clinicians prescribe antibiotics and patients get better, the clinician may incorrectly assume a cause and effect relationship that reinforces the behavior. The negative experiences that a physician has with patients are also worth considering. Clearly, there are still patients who are adamant about getting an antibiotic for every minor cold they catch. These patient encounters are frequently frustrating and time-consuming for physicians, and the emotions they evoke are very powerful. Studies have shown that strong emotions may actually facilitate the memory process,²⁴ and these emotionally charged encounters are more memorable than the routine office visits. This situation may lead physicians into believing that many more patients will demand antibiotics than really would, and some physicians may be writing these questionable prescriptions to avoid conflict.

The expectations of patients also play a large role in perpetuating the overprescription of antibiotics. Vinson and Lutz²⁵ have shown that parental expectations have a large impact on decisions of physicians to prescribe antibiotics for children with cough. There is no doubt that many patients expect antibiotics for URIs. In our study 76% of the patients felt their illness would require an antibiotic before the office visit. Not meeting that expectation makes clinicians uncomfortable and fearful that patients will be dissatisfied, despite studies that show differently.¹⁷

In our study, half of the patients given a back-up antibiotic prescription filled it by the seventh day. What is the significance of this? Critics would say that we enabled many patients to get unnecessary antibiotics. We prefer to interpret the 50% fill rate as an overall reduction from the usual practice. We know from unpublished chart reviews of our physicians in acute care clinics that patients presenting with URIs receive antibiotics approximately 60% of the time. This rate is similar to what is quoted in the literature for antibiotic usage for URIs.²⁶ In our study, we found that approximately 23% of patients got an immediate-fill antibiotic, 30% got a back-up prescription, and the rest received advice on symptomatic management but no antibiotic treatment. The finding that only half of the back-up group filled their prescriptions is a significant reduction (approximately 15%) in overall antibiotic usage. Such a reduction has an immediate positive effect on all the problems caused by the overusage of antibiotics, and may have an impact on the expectations and behavior of these patients with future URIs.

We found that patients were generally very satisfied when a back-up antibiotic prescription strategy was used. Although 96% of respondents reported that they were satisfied with their care, we believe that there are multiple factors involved in patient satisfaction, but our study methodology did not allow us to isolate those that were attributed to the back-up antibiotic prescription strategy. However, in general, using this approach did not appear to affect overall satisfaction with the physician-patient encounter.

Limitations

There are many limitations to our study. First, during the study period there may have been an artificially high use of the back-up strategy compared with what normally occurs in our physician practices. All of the physicians involved were advised of the objectives of our study. The concept of a back-up prescription was not new to them, but those who were not familiar were encouraged to be open to the opportunity to use it. Other physicians who routinely used this strategy discussed their success with it and may have influenced some of their peers to use it more frequently. We suspect that the 30% rate of the back-up concept with URI patients may be an overestimate from the usual practice of these physicians. Also, the data were collected during the peak of the influenza season, and we suspect many of the physicians were more confident that much of what they were treating in the office was of viral etiology. Consequently, they would be more likely to use a back-up than an immediate-fill prescription. Also, simply knowing that the data were being collected may have changed some of the prescribing habits of the physicians in terms of their overall use of antibiotics (Hawthorne effect). Although no precise baseline use of antibiotics was established in this group of patients with these physicians, chart reviews of patients with similar complaints before the study indicated an antibiotic usage rate of 55%. (National figures derived from Medicare claims data indicate a rough estimate as high as 60%). Future studies should consider randomizing groups of physicians into users and nonusers of the back-up prescription strategy to more accurately measure the effects of this practice.

Another limitation to our study was that physicians were allowed to enroll patients even if they were not identified by the front office personnel as meeting the enrollment criteria. This may have introduced a selection bias in the study, although we know that the actual number of patients enrolled by physicians was only a fraction of the total. The use of a uniform standard protocol should be adhered to in future studies.

Finally, satisfaction rates were based on self-reported data. Because these patients were seen in their usual site of medical outpatient care they may have given socially desirable responses and been reluctant to report negative experiences fearing that the information would influence their future care.

Conclusions

The back-up antibiotic prescription strategy appears to be a reasonable option for treating patients with common respiratory symptoms in the ambulatory setting. It was associated with a high degree of patient satisfaction and may be useful as a method of re-educating patients and decreasing the use of antibiotics. The finding that half of the patients chose not to fill these prescriptions also suggests a potential health care cost savings opportunity.

A large body of literature has addressed the frequent use of antibiotics for common upper and lower respiratory tract infections in the outpatient setting. The many dangers of this practice, including the development of bacterial resistance,^1,2 adverse drug reactions,^3-5 and negative financial implications,⁶ have been discussed, but very few methods to resolve the problem have been tested. Guidelines and educational strategies have been touted by some advocates as an essential part of the solution process and have been shown to have some degree of success in specific settings.^7,8 Recently, the American Academy of Family Physicians, the American Academy of Pediatrics, and the Centers for Disease Control and Prevention have collaborated to develop a set of recommendations to help clinicians use antibiotics more appropriately when treating patients with common respiratory illnesses.^9-13 The results of these educational efforts have not been evaluated.

Although many aspects of the antibiotic overprescribing issue may start with physician beliefs, training, and practice setting, other factors have been postulated, including meeting patient expectations, economic issues, and time constraints. A practice that has not been reported in the literature but has been employed by many physicians (anecdotally) is the use of a back-up prescription strategy. This approach addresses the complex problem of satisfying patients in a timely manner while re-educating them in a nonconfrontational way.

The term “back-up prescription” applies to the writing of a prescription that is to be filled at a later time, and only if the patient’s condition deteriorates or fails to improve. At the time the prescription is written the physician explains to the patient (or the family) the reasons for not giving an immediate-fill antibiotic prescription and gives advice on symptomatic treatment for the current problem. Additionally, specific guidance is given on clinical parameters and the timing of when to fill the prescription if the condition progresses. To our knowledge, this practice has not been studied adequately in family medicine settings although similar strategies have been used for other health-related conditions.^14-16 The use of back-up treatment for malaria (also referred to as reserve treatment) has been mentioned in the literature.¹⁵ Davy and colleagues¹⁶ have also reported the use of back-up antibiotics for the treatment of undifferentiated acute respiratory tract infection with cough among primary care family physicians and pediatricians. This study, which was based on a self-reported survey, primarily sought to identify the frequency with which reserve antibiotics were prescribed to this group of children. It did not address the actual practice of using a back-up prescription.

In exploring the use of the back-up antibiotic prescription strategy it is essential to assess the degree of patient satisfaction and the fill rates for back-up prescriptions and their predictors. In their study on patient satisfaction and antibiotic prescriptions for respiratory infections, Hamm and coworkers¹⁷ elicited patient satisfaction levels immediately following a physician encounter. However, seeking the opinions of patients about their satisfaction immediately after an encounter may not yield accurate responses because more time may be required to assess other factors such as the effect (or lack of effect) of the treatment suggested, including antibiotic prescriptions given.

Methods

Study Design and Setting

We performed an observational study on the current prescribing habits of our physicians. The study included prospective data collection on the use of a back-up antibiotic prescription strategy among patients presenting with complaints of common respiratory symptoms to 28 physicians and 2 physician extenders in 3 family practice clinics between January and April 1999. These clinics are part of the Scott & White Healthcare System and are located in Temple (Santa Fe Clinic), Waco, and Killeen, Texas.

The practice of providing back-up antibiotic prescriptions was regularly used by many of the physicians as part of their routine management options, while others were unfamiliar with the concept or used it only rarely. All physicians were advised of the study objectives and were encouraged to enroll patients who received back-up antibiotic prescriptions. However, the physicians were not asked to change their usual prescribing habits. The study protocol was approved by the Institutional Review Board of the Scott & White Memorial Hospital and Clinic.

Study Participants and Data Collection

A concerted effort was made to enroll all patients who presented with complaints of a head cold or respiratory symptoms during this 4-month period. Our inclusion criteria were strict but broad. Patients were enrolled in the study if they had head congestion, sinus congestion, fever, headache, cough, chest congestion, or sore throat. Patients were only excluded if they had one dominant symptom and physical finding, such as earache. In addition to the front desk personnel, physicians and nurses could also enroll patients in the study if the patient brought up the need for treatment of respiratory complaints that were not mentioned to the appointment clerks (eg, “Oh, by the way, while I am here for my blood pressure follow-up, would you check out my head cold. I think I may be coming down with something and thought maybe I should get some antibiotics.”).

When patients reported for their appointments, a physician survey was attached to the front of their chart by front desk office personnel. This survey could also be added to the chart when the patient was put into an examination room if the nurse was made aware of the patient’s expectation for evaluation of respiratory symptoms. The physician survey which was filled out at the conclusion of the office visit elicited information regarding: (1) physician and patient demographic information; (2) the patient’s primary complaints; (3) whether the physician was the patient’s primary care physician; (4) type of prescription given to the patient (an immediate-fill antibiotic prescription, a back-up antibiotic prescription, or no antibiotic prescription); and (5) physician subjective rating, on a 5-point scale, of the clinical necessity for prescribing antibiotics for the patient.

The patients who were given back-up antibiotic prescriptions were each given a patient survey to complete with instructions to return the form in a provided preaddressed envelope 7 days after their initial appointment. Patients who did not return their surveys were called by the research coordinator, and the surveys were completed over the phone.

The patient survey included questions about: (1) patient satisfaction with the care received; (2) their perceived need before the office visit for an antibiotic prescription; (3) whether they received a written back-up antibiotic prescription; (4) whether they filled the back-up prescription; and (5) whether they required any subsequent medical care for the same illness.

Definition of the Back-up Strategy

A back-up antibiotic prescription strategy was defined in our study as a prescription given to a patient along with instructions to fill the prescription only if the condition deteriorated or failed to improve within a predefined number of days. The exact number of days was not standardized by the study protocol, allowing each physician to customize this aspect of care.

Statistical Analysis

Data management and analysis were performed using SAS¹⁸ on a mainframe and the Statistical Package for the Social Sciences¹⁹ on a personal computer. We determined physicians’ use of the back-up antibiotic prescription strategy using selected variables by comparing study subjects who received back-up antibiotic prescriptions with those given immediate-fill prescriptions. We computed crude odds ratios (ORs) and 95% confidence intervals (CIs) for use of the back-up prescription strategy. Variables that were statistically significant in their bivariate relationship with use of the back-up antibiotic prescription strategy and those with some biological plausibility (eg, patient age) were entered into a multivariate logistic regression modeling to compute adjusted ORs.

We also computed patient satisfaction and fill rates of back-up antibiotic prescriptions. Agreement between patients’ perceived need for antibiotics before the office visit and physicians’ subjective rating of the clinical necessity to prescribe antibiotics for patients was assessed using the k statistic. To determine correlates of patient satisfaction with the back-up prescription strategy, we compared satisfaction rates of study subjects by patient and physician characteristics, the presenting respiratory complaints, and several selected characteristics. Finally, correlates of back-up prescription filling were similarly determined by comparing filling rates by the same characteristics.

Group differences were assessed for significance using the chi-square statistic or Fisher’s exact test for categorical variables and analysis of variance for continuous variables. All tests were 2-tailed and were considered significant at P <.05.

Results

A total of 947 patients were evaluated for common respiratory symptoms by 19 family physicians, 2 physician extenders (a nurse practitioner and a physician assistant), and 9 family medicine residents.

Rates and Correlates of Back-up Antibiotic Prescriptions

From the 947 enrolled patients with common respiratory symptoms, 441 (46.6%) were not given antibiotics: 286 (30.2%) were given back-up antibiotic prescriptions, and 220 (23.2%) were given immediate-fill antibiotic prescriptions. Patients younger than 35 years and those with complaints of cough were twice as likely to be given back-up antibiotic prescriptions. Female sex and health care provider role as a physician extender were the only physician characteristics that were positively associated with the use of back-up prescriptions. Neither the role as primary care physician nor the physician’s number of years in practice were related to the type of prescription given.

Of 286 patients given back-up antibiotic prescriptions, we obtained completed follow-up surveys from 255 (89.2%). There were no significant differences between respondents and nonrespondents regarding demographic variables.

Rate and Correlates of Patient Satisfaction

Of the 255 patients who responded, 245 (96.1%) reported that they were satisfied with the care they received at their visit. The majority of the patients (76.1%) felt that their illness would require an antibiotic when their appointment was scheduled. However, only 36.9% of their physicians felt that their illness warranted the use of antibiotics. There was no significant agreement (P=.08) between patients’ perceived need for antibiotics before the office visit and physicians’ subjective rating of the clinical necessity to prescribe antibiotics (Table 1).

Patient and physicians characteristics were not associated with patient self-reported satisfaction rate with the care they received. Satisfaction rates were, however, significantly associated with patient complaints of sinus congestion (Table 2) and a patient’s requirement for additional care at a later time for the same illness (Table 3). Patients with complaints of sinus congestion and those who required additional care at a later time reported significantly less satisfaction.

Fill Rate and Correlates of Back-up Antibiotic Prescription

The overall back-up antibiotic prescription fill rate was 50.2%. Fill rates did not differ significantly by patient characteristics or their self-reported satisfaction with the care received, physician characteristics, or whether the physician was the patient’s primary care physician.

Additional Care

Additional care (defined as any subsequent contact with a health care provider) was required for 9.0% (n=23) of the patients in our study who received back-up antibiotic prescriptions. Of these, 10 consulted by telephone about their illness. Another 12 made repeat office visits, and 1 made an emergency room visit for an exacerbation of asthma; that patient was subsequently admitted overnight for management of her asthma. Of the 23 patients who sought additional care, 17 (74%) filled their back-up antibiotic prescriptions.

Discussion

Several factors are associated with the overprescription of antimicrobials for common respiratory symptoms, including physician specialty, physician knowledge base of the natural history of viral respiratory infections, clinician and patient experiences, patient expectations, and economic pressures related to time and reimbursement. Mainous and colleagues²⁰ and Nyquist and coworkers²¹ have reported that family physicians and general practitioners have prescribed antibiotics significantly more than pediatricians for children with upper respiratory infections (URIs). Schwartz and colleagues²² also conducted a survey based on a written case scenario that highlighted the significant discrepancy between the prescribing habits of family physicians and pediatricians. Compared with 53% of pediatricians, 71% of family physicians would immediately prescribe an antibiotic for a child who had a single day of scant light green and yellow nasal discharge and low-grade fever (P=.001).

Both clinician and patient experiences may also promote antibiotic overusage. If a patient has received an antibiotic for a URI in the past and had a good outcome, that positive experience creates an impression that antibiotic therapy is required and proper.²³ Similarly when clinicians prescribe antibiotics and patients get better, the clinician may incorrectly assume a cause and effect relationship that reinforces the behavior. The negative experiences that a physician has with patients are also worth considering. Clearly, there are still patients who are adamant about getting an antibiotic for every minor cold they catch. These patient encounters are frequently frustrating and time-consuming for physicians, and the emotions they evoke are very powerful. Studies have shown that strong emotions may actually facilitate the memory process,²⁴ and these emotionally charged encounters are more memorable than the routine office visits. This situation may lead physicians into believing that many more patients will demand antibiotics than really would, and some physicians may be writing these questionable prescriptions to avoid conflict.

The expectations of patients also play a large role in perpetuating the overprescription of antibiotics. Vinson and Lutz²⁵ have shown that parental expectations have a large impact on decisions of physicians to prescribe antibiotics for children with cough. There is no doubt that many patients expect antibiotics for URIs. In our study 76% of the patients felt their illness would require an antibiotic before the office visit. Not meeting that expectation makes clinicians uncomfortable and fearful that patients will be dissatisfied, despite studies that show differently.¹⁷

In our study, half of the patients given a back-up antibiotic prescription filled it by the seventh day. What is the significance of this? Critics would say that we enabled many patients to get unnecessary antibiotics. We prefer to interpret the 50% fill rate as an overall reduction from the usual practice. We know from unpublished chart reviews of our physicians in acute care clinics that patients presenting with URIs receive antibiotics approximately 60% of the time. This rate is similar to what is quoted in the literature for antibiotic usage for URIs.²⁶ In our study, we found that approximately 23% of patients got an immediate-fill antibiotic, 30% got a back-up prescription, and the rest received advice on symptomatic management but no antibiotic treatment. The finding that only half of the back-up group filled their prescriptions is a significant reduction (approximately 15%) in overall antibiotic usage. Such a reduction has an immediate positive effect on all the problems caused by the overusage of antibiotics, and may have an impact on the expectations and behavior of these patients with future URIs.

We found that patients were generally very satisfied when a back-up antibiotic prescription strategy was used. Although 96% of respondents reported that they were satisfied with their care, we believe that there are multiple factors involved in patient satisfaction, but our study methodology did not allow us to isolate those that were attributed to the back-up antibiotic prescription strategy. However, in general, using this approach did not appear to affect overall satisfaction with the physician-patient encounter.

Limitations

There are many limitations to our study. First, during the study period there may have been an artificially high use of the back-up strategy compared with what normally occurs in our physician practices. All of the physicians involved were advised of the objectives of our study. The concept of a back-up prescription was not new to them, but those who were not familiar were encouraged to be open to the opportunity to use it. Other physicians who routinely used this strategy discussed their success with it and may have influenced some of their peers to use it more frequently. We suspect that the 30% rate of the back-up concept with URI patients may be an overestimate from the usual practice of these physicians. Also, the data were collected during the peak of the influenza season, and we suspect many of the physicians were more confident that much of what they were treating in the office was of viral etiology. Consequently, they would be more likely to use a back-up than an immediate-fill prescription. Also, simply knowing that the data were being collected may have changed some of the prescribing habits of the physicians in terms of their overall use of antibiotics (Hawthorne effect). Although no precise baseline use of antibiotics was established in this group of patients with these physicians, chart reviews of patients with similar complaints before the study indicated an antibiotic usage rate of 55%. (National figures derived from Medicare claims data indicate a rough estimate as high as 60%). Future studies should consider randomizing groups of physicians into users and nonusers of the back-up prescription strategy to more accurately measure the effects of this practice.

Another limitation to our study was that physicians were allowed to enroll patients even if they were not identified by the front office personnel as meeting the enrollment criteria. This may have introduced a selection bias in the study, although we know that the actual number of patients enrolled by physicians was only a fraction of the total. The use of a uniform standard protocol should be adhered to in future studies.

Finally, satisfaction rates were based on self-reported data. Because these patients were seen in their usual site of medical outpatient care they may have given socially desirable responses and been reluctant to report negative experiences fearing that the information would influence their future care.

Conclusions

The back-up antibiotic prescription strategy appears to be a reasonable option for treating patients with common respiratory symptoms in the ambulatory setting. It was associated with a high degree of patient satisfaction and may be useful as a method of re-educating patients and decreasing the use of antibiotics. The finding that half of the patients chose not to fill these prescriptions also suggests a potential health care cost savings opportunity.

A large body of literature has addressed the frequent use of antibiotics for common upper and lower respiratory tract infections in the outpatient setting. The many dangers of this practice, including the development of bacterial resistance,^1,2 adverse drug reactions,^3-5 and negative financial implications,⁶ have been discussed, but very few methods to resolve the problem have been tested. Guidelines and educational strategies have been touted by some advocates as an essential part of the solution process and have been shown to have some degree of success in specific settings.^7,8 Recently, the American Academy of Family Physicians, the American Academy of Pediatrics, and the Centers for Disease Control and Prevention have collaborated to develop a set of recommendations to help clinicians use antibiotics more appropriately when treating patients with common respiratory illnesses.^9-13 The results of these educational efforts have not been evaluated.

Although many aspects of the antibiotic overprescribing issue may start with physician beliefs, training, and practice setting, other factors have been postulated, including meeting patient expectations, economic issues, and time constraints. A practice that has not been reported in the literature but has been employed by many physicians (anecdotally) is the use of a back-up prescription strategy. This approach addresses the complex problem of satisfying patients in a timely manner while re-educating them in a nonconfrontational way.

The term “back-up prescription” applies to the writing of a prescription that is to be filled at a later time, and only if the patient’s condition deteriorates or fails to improve. At the time the prescription is written the physician explains to the patient (or the family) the reasons for not giving an immediate-fill antibiotic prescription and gives advice on symptomatic treatment for the current problem. Additionally, specific guidance is given on clinical parameters and the timing of when to fill the prescription if the condition progresses. To our knowledge, this practice has not been studied adequately in family medicine settings although similar strategies have been used for other health-related conditions.^14-16 The use of back-up treatment for malaria (also referred to as reserve treatment) has been mentioned in the literature.¹⁵ Davy and colleagues¹⁶ have also reported the use of back-up antibiotics for the treatment of undifferentiated acute respiratory tract infection with cough among primary care family physicians and pediatricians. This study, which was based on a self-reported survey, primarily sought to identify the frequency with which reserve antibiotics were prescribed to this group of children. It did not address the actual practice of using a back-up prescription.

In exploring the use of the back-up antibiotic prescription strategy it is essential to assess the degree of patient satisfaction and the fill rates for back-up prescriptions and their predictors. In their study on patient satisfaction and antibiotic prescriptions for respiratory infections, Hamm and coworkers¹⁷ elicited patient satisfaction levels immediately following a physician encounter. However, seeking the opinions of patients about their satisfaction immediately after an encounter may not yield accurate responses because more time may be required to assess other factors such as the effect (or lack of effect) of the treatment suggested, including antibiotic prescriptions given.

Methods

Study Design and Setting

We performed an observational study on the current prescribing habits of our physicians. The study included prospective data collection on the use of a back-up antibiotic prescription strategy among patients presenting with complaints of common respiratory symptoms to 28 physicians and 2 physician extenders in 3 family practice clinics between January and April 1999. These clinics are part of the Scott & White Healthcare System and are located in Temple (Santa Fe Clinic), Waco, and Killeen, Texas.

The practice of providing back-up antibiotic prescriptions was regularly used by many of the physicians as part of their routine management options, while others were unfamiliar with the concept or used it only rarely. All physicians were advised of the study objectives and were encouraged to enroll patients who received back-up antibiotic prescriptions. However, the physicians were not asked to change their usual prescribing habits. The study protocol was approved by the Institutional Review Board of the Scott & White Memorial Hospital and Clinic.

Study Participants and Data Collection

A concerted effort was made to enroll all patients who presented with complaints of a head cold or respiratory symptoms during this 4-month period. Our inclusion criteria were strict but broad. Patients were enrolled in the study if they had head congestion, sinus congestion, fever, headache, cough, chest congestion, or sore throat. Patients were only excluded if they had one dominant symptom and physical finding, such as earache. In addition to the front desk personnel, physicians and nurses could also enroll patients in the study if the patient brought up the need for treatment of respiratory complaints that were not mentioned to the appointment clerks (eg, “Oh, by the way, while I am here for my blood pressure follow-up, would you check out my head cold. I think I may be coming down with something and thought maybe I should get some antibiotics.”).

When patients reported for their appointments, a physician survey was attached to the front of their chart by front desk office personnel. This survey could also be added to the chart when the patient was put into an examination room if the nurse was made aware of the patient’s expectation for evaluation of respiratory symptoms. The physician survey which was filled out at the conclusion of the office visit elicited information regarding: (1) physician and patient demographic information; (2) the patient’s primary complaints; (3) whether the physician was the patient’s primary care physician; (4) type of prescription given to the patient (an immediate-fill antibiotic prescription, a back-up antibiotic prescription, or no antibiotic prescription); and (5) physician subjective rating, on a 5-point scale, of the clinical necessity for prescribing antibiotics for the patient.

The patients who were given back-up antibiotic prescriptions were each given a patient survey to complete with instructions to return the form in a provided preaddressed envelope 7 days after their initial appointment. Patients who did not return their surveys were called by the research coordinator, and the surveys were completed over the phone.

The patient survey included questions about: (1) patient satisfaction with the care received; (2) their perceived need before the office visit for an antibiotic prescription; (3) whether they received a written back-up antibiotic prescription; (4) whether they filled the back-up prescription; and (5) whether they required any subsequent medical care for the same illness.

Definition of the Back-up Strategy

A back-up antibiotic prescription strategy was defined in our study as a prescription given to a patient along with instructions to fill the prescription only if the condition deteriorated or failed to improve within a predefined number of days. The exact number of days was not standardized by the study protocol, allowing each physician to customize this aspect of care.

Statistical Analysis

Data management and analysis were performed using SAS¹⁸ on a mainframe and the Statistical Package for the Social Sciences¹⁹ on a personal computer. We determined physicians’ use of the back-up antibiotic prescription strategy using selected variables by comparing study subjects who received back-up antibiotic prescriptions with those given immediate-fill prescriptions. We computed crude odds ratios (ORs) and 95% confidence intervals (CIs) for use of the back-up prescription strategy. Variables that were statistically significant in their bivariate relationship with use of the back-up antibiotic prescription strategy and those with some biological plausibility (eg, patient age) were entered into a multivariate logistic regression modeling to compute adjusted ORs.

We also computed patient satisfaction and fill rates of back-up antibiotic prescriptions. Agreement between patients’ perceived need for antibiotics before the office visit and physicians’ subjective rating of the clinical necessity to prescribe antibiotics for patients was assessed using the k statistic. To determine correlates of patient satisfaction with the back-up prescription strategy, we compared satisfaction rates of study subjects by patient and physician characteristics, the presenting respiratory complaints, and several selected characteristics. Finally, correlates of back-up prescription filling were similarly determined by comparing filling rates by the same characteristics.

Group differences were assessed for significance using the chi-square statistic or Fisher’s exact test for categorical variables and analysis of variance for continuous variables. All tests were 2-tailed and were considered significant at P <.05.

Results

A total of 947 patients were evaluated for common respiratory symptoms by 19 family physicians, 2 physician extenders (a nurse practitioner and a physician assistant), and 9 family medicine residents.

Rates and Correlates of Back-up Antibiotic Prescriptions

From the 947 enrolled patients with common respiratory symptoms, 441 (46.6%) were not given antibiotics: 286 (30.2%) were given back-up antibiotic prescriptions, and 220 (23.2%) were given immediate-fill antibiotic prescriptions. Patients younger than 35 years and those with complaints of cough were twice as likely to be given back-up antibiotic prescriptions. Female sex and health care provider role as a physician extender were the only physician characteristics that were positively associated with the use of back-up prescriptions. Neither the role as primary care physician nor the physician’s number of years in practice were related to the type of prescription given.

Of 286 patients given back-up antibiotic prescriptions, we obtained completed follow-up surveys from 255 (89.2%). There were no significant differences between respondents and nonrespondents regarding demographic variables.

Rate and Correlates of Patient Satisfaction

Of the 255 patients who responded, 245 (96.1%) reported that they were satisfied with the care they received at their visit. The majority of the patients (76.1%) felt that their illness would require an antibiotic when their appointment was scheduled. However, only 36.9% of their physicians felt that their illness warranted the use of antibiotics. There was no significant agreement (P=.08) between patients’ perceived need for antibiotics before the office visit and physicians’ subjective rating of the clinical necessity to prescribe antibiotics (Table 1).

Patient and physicians characteristics were not associated with patient self-reported satisfaction rate with the care they received. Satisfaction rates were, however, significantly associated with patient complaints of sinus congestion (Table 2) and a patient’s requirement for additional care at a later time for the same illness (Table 3). Patients with complaints of sinus congestion and those who required additional care at a later time reported significantly less satisfaction.

Fill Rate and Correlates of Back-up Antibiotic Prescription

The overall back-up antibiotic prescription fill rate was 50.2%. Fill rates did not differ significantly by patient characteristics or their self-reported satisfaction with the care received, physician characteristics, or whether the physician was the patient’s primary care physician.

Additional Care

Additional care (defined as any subsequent contact with a health care provider) was required for 9.0% (n=23) of the patients in our study who received back-up antibiotic prescriptions. Of these, 10 consulted by telephone about their illness. Another 12 made repeat office visits, and 1 made an emergency room visit for an exacerbation of asthma; that patient was subsequently admitted overnight for management of her asthma. Of the 23 patients who sought additional care, 17 (74%) filled their back-up antibiotic prescriptions.

Discussion

Several factors are associated with the overprescription of antimicrobials for common respiratory symptoms, including physician specialty, physician knowledge base of the natural history of viral respiratory infections, clinician and patient experiences, patient expectations, and economic pressures related to time and reimbursement. Mainous and colleagues²⁰ and Nyquist and coworkers²¹ have reported that family physicians and general practitioners have prescribed antibiotics significantly more than pediatricians for children with upper respiratory infections (URIs). Schwartz and colleagues²² also conducted a survey based on a written case scenario that highlighted the significant discrepancy between the prescribing habits of family physicians and pediatricians. Compared with 53% of pediatricians, 71% of family physicians would immediately prescribe an antibiotic for a child who had a single day of scant light green and yellow nasal discharge and low-grade fever (P=.001).

Both clinician and patient experiences may also promote antibiotic overusage. If a patient has received an antibiotic for a URI in the past and had a good outcome, that positive experience creates an impression that antibiotic therapy is required and proper.²³ Similarly when clinicians prescribe antibiotics and patients get better, the clinician may incorrectly assume a cause and effect relationship that reinforces the behavior. The negative experiences that a physician has with patients are also worth considering. Clearly, there are still patients who are adamant about getting an antibiotic for every minor cold they catch. These patient encounters are frequently frustrating and time-consuming for physicians, and the emotions they evoke are very powerful. Studies have shown that strong emotions may actually facilitate the memory process,²⁴ and these emotionally charged encounters are more memorable than the routine office visits. This situation may lead physicians into believing that many more patients will demand antibiotics than really would, and some physicians may be writing these questionable prescriptions to avoid conflict.

The expectations of patients also play a large role in perpetuating the overprescription of antibiotics. Vinson and Lutz²⁵ have shown that parental expectations have a large impact on decisions of physicians to prescribe antibiotics for children with cough. There is no doubt that many patients expect antibiotics for URIs. In our study 76% of the patients felt their illness would require an antibiotic before the office visit. Not meeting that expectation makes clinicians uncomfortable and fearful that patients will be dissatisfied, despite studies that show differently.¹⁷

In our study, half of the patients given a back-up antibiotic prescription filled it by the seventh day. What is the significance of this? Critics would say that we enabled many patients to get unnecessary antibiotics. We prefer to interpret the 50% fill rate as an overall reduction from the usual practice. We know from unpublished chart reviews of our physicians in acute care clinics that patients presenting with URIs receive antibiotics approximately 60% of the time. This rate is similar to what is quoted in the literature for antibiotic usage for URIs.²⁶ In our study, we found that approximately 23% of patients got an immediate-fill antibiotic, 30% got a back-up prescription, and the rest received advice on symptomatic management but no antibiotic treatment. The finding that only half of the back-up group filled their prescriptions is a significant reduction (approximately 15%) in overall antibiotic usage. Such a reduction has an immediate positive effect on all the problems caused by the overusage of antibiotics, and may have an impact on the expectations and behavior of these patients with future URIs.

We found that patients were generally very satisfied when a back-up antibiotic prescription strategy was used. Although 96% of respondents reported that they were satisfied with their care, we believe that there are multiple factors involved in patient satisfaction, but our study methodology did not allow us to isolate those that were attributed to the back-up antibiotic prescription strategy. However, in general, using this approach did not appear to affect overall satisfaction with the physician-patient encounter.

Limitations

There are many limitations to our study. First, during the study period there may have been an artificially high use of the back-up strategy compared with what normally occurs in our physician practices. All of the physicians involved were advised of the objectives of our study. The concept of a back-up prescription was not new to them, but those who were not familiar were encouraged to be open to the opportunity to use it. Other physicians who routinely used this strategy discussed their success with it and may have influenced some of their peers to use it more frequently. We suspect that the 30% rate of the back-up concept with URI patients may be an overestimate from the usual practice of these physicians. Also, the data were collected during the peak of the influenza season, and we suspect many of the physicians were more confident that much of what they were treating in the office was of viral etiology. Consequently, they would be more likely to use a back-up than an immediate-fill prescription. Also, simply knowing that the data were being collected may have changed some of the prescribing habits of the physicians in terms of their overall use of antibiotics (Hawthorne effect). Although no precise baseline use of antibiotics was established in this group of patients with these physicians, chart reviews of patients with similar complaints before the study indicated an antibiotic usage rate of 55%. (National figures derived from Medicare claims data indicate a rough estimate as high as 60%). Future studies should consider randomizing groups of physicians into users and nonusers of the back-up prescription strategy to more accurately measure the effects of this practice.

Another limitation to our study was that physicians were allowed to enroll patients even if they were not identified by the front office personnel as meeting the enrollment criteria. This may have introduced a selection bias in the study, although we know that the actual number of patients enrolled by physicians was only a fraction of the total. The use of a uniform standard protocol should be adhered to in future studies.

Finally, satisfaction rates were based on self-reported data. Because these patients were seen in their usual site of medical outpatient care they may have given socially desirable responses and been reluctant to report negative experiences fearing that the information would influence their future care.

Conclusions

The back-up antibiotic prescription strategy appears to be a reasonable option for treating patients with common respiratory symptoms in the ambulatory setting. It was associated with a high degree of patient satisfaction and may be useful as a method of re-educating patients and decreasing the use of antibiotics. The finding that half of the patients chose not to fill these prescriptions also suggests a potential health care cost savings opportunity.

Family physicians are in an optimal position to identify women who are victims of abuse, because they are often the first point of contact in the medical arena. However, recent studies indicate that family physicians continue to be reticent in accepting this responsibility, thus contributing to the underdetection of woman abuse.^1,2 For almost 2 decades family medicine educators and researchers have made a concerted effort to understand and increase identification and treatment of woman abuse by family physicians.^1-17 As part of this initiative, our focus has been on the development of a screening tool for family physicians to use in the context of a routine office visit or a well-woman examination to identify and assess women who are experiencing emotional, physical, or sexual abuse by their partners.^8,18

The Woman Abuse Screening Tool (WAST), which consists of 7 questions, was developed and pilot tested using purposive samples of abused and nonabused women.¹⁸ It was found to have high internal consistency among this sample ({a} =0.95). It also demonstrated construct validity, with total scores correlating highly (r=0.96) with scores on the Abuse Risk Inventory (ARI).¹⁸ The validation study also provided evidence of discriminant validity, finding significant differences in the scores of abused and nonabused women both on individual items and on the overall scores.¹⁸

The first 2 questions of the WAST (“In general, how would you describe your relationship: a lot of tension, some tension, no tension?” and “Do you and your partner work out arguments: with great difficulty, some difficulty, no difficulty?” constitute the WAST-Short, which has been an effective tool for initially screening for the presence of abuse.¹⁸ The screening tool correctly classified 91.7% of the abused women and 100% of the nonabused women in the validation study.¹⁸ These 2 questions were also identified by the abused women in the validation study as those with which they would be most comfortable if asked by their family physicians. The remaining questions on the WAST were used to gain a more complete assessment of the abuse. In the validation study there were significant differences found between the abused and nonabused women on the mean overall WAST scores (18 vs 8.8, respectively; P <.001).

To establish the generalizability of the WAST, we field-tested it by having family physicians ask the questions of adult women in the general population who were presenting for routine visits (complete physical examination or prenatal care) as well as acute complaints.¹⁹ Although reported interest of family physicians in having a brief screening tool had been the genesis of this program of study, their comfort in using the WAST during a clinical encounter had not been assessed.⁸ Also, determining the level of comfort of women patients being asked the WAST questions by a family physician during an actual office visit versus a hypothetical encounter (as was the case in the validation study) was viewed as important.¹⁸

Inquiring about abuse has been found to cause discomfort for both physicians and women patients. It has been noted previously that family physicians remain reluctant to delve into the issue of woman abuse in spite of the fact that educating physicians about this abuse (including the use of a screening protocol) has been shown to significantly increase the detection rates of abused women in emergency departments.^20,21 Also, both patients and physicians have indicated that the discomfort of physicians with issues of abuse may deter them from inquiring about this topic.^7,8,22,23 Data from previous studies showing a decline in detection once a formal assessment protocol is discontinued emphasize the importance of maintaining a continuous screening approach if woman abuse is to be detected.²¹ Thus knowledge of the level of comfort physicians have in using the WAST and whether it aided in their identification of woman abuse and determining their ongoing commitment to use it required investigation.

Women are often reluctant to disclose abuse to their family physicians for numerous reasons, including shame, denial, fear of reprisal by their partner, a tendency to minimize or normalize the abuse, fear of a negative or punitive response by their physician, or assignment of power and control to the physician.^6,24-26 However, studies have shown that when women feel understood, listened to, and validated by their physicians they are more inclined to discuss the abuse.^27-29 Also, previous studies with abused women^22,23,27-29 have found that they want their physicians to take responsibility for asking questions about abuse and to do so in a manner that is caring, respectful, and supportive. Thus, determining the comfort of women being asked the WAST questions by their family physicians was viewed as essential to our study.

Therefore, the objectives of field testing the WAST were to assess its validity and reliability in the general population within the context of the family practice setting; to determine the comfort levels of family physicians administering the WAST, their perceptions of its ability to help them identify abused women, and their willingness to continue using it in their practices; and to determine the self-reported comfort of patients with being asked the WAST questions by their family physicians.

Our study was approved by the Review Board for Health Sciences Research Involving Human Subjects at the University of Western Ontario.

Methods

Setting

Our study was conducted in the offices of family physicians located in London, Ontario, Canada, and the surrounding area. The recruitment and data collection took place from March 1997 to August 1998.

Instruments

The WAST. Although the original version of the WAST consisted of 7 questions, an eighth question (“Has your partner ever abused you sexually?”) was added for our study (Figure). This question was thought to be clinically important when assessing women who screen positive on the WAST-Short. The 2 questions that make up the WAST-Short assess the degree of relationship tension and the amount of difficulty that the woman and her partner have in working out arguments on a scale of 1 to 3.

Scores on the WAST-Short are computed on the basis of a criterion cutoff score of 1, which involves assigning a score of 1 to the most extreme positive responses for each of the 2 items (ie, “a lot of tension” and “great difficulty”) and a score of 0 to the other response options.¹⁸ The remaining 6 questions are used to gain a more complete assessment of the abuse by asking the respondent to rate the frequency of various feelings and experiences on a scale from 1 (often) to 3 (never). The WAST items are recoded and summed to calculate the overall score.

The Abuse Risk Inventory. The Abuse Risk Inventory (ARI) is a 25-item self-report measure used in the identification of woman abuse and is also described as being useful in the assessment process.³⁰ Respondents rate 25 items on the basis of frequency of occurrence using a 4-point scale ranging from “rarely or never” to “always.” A score of 50 or higher suggests that the respondent may be in an abusive situation or at risk for abuse.³⁰ The ARI has demonstrated reliability (a=.91).30

Physician and Patient Comfort with the WAST Questionnaires. These self-report questionnaires were used to determine the level of comfort of physicians and patients with asking or being asked each of the WAST questions. Responses were given using a 4-point scale ranging from 1 (not at all comfortable) to 4 (very comfortable).

Prior Knowledge Questionnaire. This questionnaire assessed a physician’s previous or concurrent relationships with the patient and her partner by identifying various contexts (eg, workplace, leisure) through which the physician is connected with the patient and her partner in the role other than as the family physician. This questionnaire was included because of the potential influence of the physician’s personal relationship with the patient and her partner on both the patient’s willingness to disclose abuse and the physician’s comfort in inquiring about it.

The Perceived Usefulness Questionnaire. This questionnaire asked physicians to respond to the following statements using a 5-point Likert scale ranging from 1 (strongly agree) to 5 (strongly disagree): “The wording of the WAST was clear”; “The WAST helped me to identify women who are abused”; “I feel better able to identify women who are abused using the WAST”; and “I felt comfortable asking questions on the WAST.” Physicians were also asked to indicate whether they would continue to use the WAST in their practice using the same 5-point scale.

Physician Participants

Our goal was to achieve a stratified random sample of 20 physicians practicing in urban and rural settings from a sampling frame of 400 family physicians in London, Ontario, Canada, and the surrounding area. The family physician investigators telephoned a total of 44 physicians who were selected from the sampling frame using a random numbers table. This followed the recruitment process reported by Borgeil and colleagues.³¹ Physicians who agreed to participate in our study were mailed a letter of information, a consent form, and directions for the study protocol, including how to administer the WAST and a list of community resources for women who were abused.

Patient Participants

For patients, we followed the recommendations of DeVellis, who has outlined a sample size range with a minimum of 200 and a maximum of 1000 respondents to explore the factorial validity of a new measure.³² To ensure that sufficient variability would exist across responses, we aimed for a moderate sample size of approximately 300 subjects.

To be included in our study the women patients were required to be older than 18 years; attending for a periodic health examination, for prenatal care, or with acute symptoms of illness; English speaking; unaccompanied by another person; currently involved in an intimate relationship (married or common law); and they had to consider the attending physician their primary care physician.

Instrument Administration

The 20 participating physicians were asked to administer the WAST to 15 to 20 consecutive women patients who met the inclusion criteria and consented to participate in the study. At the conclusion of each patient visit the physicians were requested to complete the WAST comfort questionnaire and the prior knowledge questionnaire. When the data collection was completed they were asked to report their perceptions of the WAST.

Each woman was approached by the research assistant in one of the physician’s examining rooms before her visit with the family physician. The research assistant explained the study, provided the patient with a letter of information, and if she agreed to participate supplied a consent form for signature. During the patient recruitment process, the research assistant maintained a written log describing eligible and ineligible patients, reasons for refusal, and other pertinent data, such as the physician’s knowledge of whether a patient was in an abusive relationship. At the conclusion of the physician-patient encounter, the research assistant met with the patient in a private area and asked her to complete the ARI and the measure assessing her comfort with the WAST questions asked.

Data Analysis

To determine the reliability and validity of the WAST, we calculated Cronbach a and Pearson correlation coefficients for the WAST and the ARI. Differences in both the nominal-level demographic information of patients and the responses of physicians and patients to the study measures on the basis of selected variables (family practice certification status for physicians, positive versus negative screen for patients) were analyzed using cross-tabulations and chi-square calculations. Differences in interval and ratio level measures (including demographic information and scale totals) were analyzed with independent samples Student t tests. Analyses involving the length of time physicians had been in practice were conducted using a computed variable (1997 minus year of graduation), which was then recoded into the decade of graduation. Scoring of the WAST involved recoding the responses to reflect a higher score for higher reported frequency of experiences and then summing the WAST scores for individuals who answered all 8 items. ARI scores were calculated for respondents who had answered all 25 items using the procedure outlined by Yegidis.³⁰

Results

Validity and Reliability of the WAST in the Family Practice Context Overall WAST and ARI scores were correlated (r=0.69, P=.01). The WAST was found to be a reliable measure in the family practice context, achieving a coefficient a of 0.75, indicating good internal consistency.

Physician Characteristics

To secure the 20 family physicians required for the study, we had to contact 44 physicians randomly selected from the sampling frame, yielding an acceptance rate of 45.5%. The final sample of physicians consisted of 7 women and 13 men. The average number of years since graduation was 22.9 (range=6-46 years). There were 8 physicians in rural practice and 12 from the city of London, Ontario. Fourteen were in a group practice arrangement, and 14 were certificants of the College of Family Physicians of Canada (CFPC). There were no significant differences between the physicians who agreed to participate and those who declined, on the basis of sex, certification status, years since graduation, practice type (solo vs group), and practice location (urban vs rural).

Patient Characteristics

A total of 456 patients were asked to participate in our study. Fifty-seven women were deemed ineligible on the basis of the inclusion criteria, resulting in 399 eligible patients. Ninety-two (23.1%) of these refused, giving lack of time, degree of sickness, and discomfort in discussing personal issues as their reasons. Thus the final sample included 307 women.

The average age of these patients was 46.2 years (range=18-86 years). The majority (87.6%) were married or in a common-law relationship. The patients were primarily white (97.6%), and 44.7% reported having postsecondary education. More than half of the subjects (58.9%) were employed, and 58.7% reported an annual household income of more than $30,000 (Table 1).

Of the 307 patients screened, 26 (8.5%) were identified by the WAST-Short as experiencing abuse. The demographics of the sample for those who screened positive and negative for abuse are provided in Table 1. No significant differences were found. However, the 26 women who screened positive for abuse reported a wide range of income levels, with 9 women (34.6%) indicating an annual income of more than $50,000.

Table 2 shows the individual WAST item responses and overall scores for the total sample divided into 2 groups: those who screened positive for experiencing abuse and those who screened negative. Significant differences were found between the 2 groups for each item and for the overall WAST scores.

Physician Perceptions of and Comfort with the WAST

The majority of the physicians (85%) thought the wording of the WAST was clear. Sixty-five percent indicated that it assisted them in identifying women who were abused, and 70% felt more confident in identifying abused women when using the WAST. Also, 75% of physicians reported that they would continue to use the WAST in their practice. We did not systematically inquire about a physician’s previous knowledge of a patient’s experience with abuse. However, this information was often reported to the research assistant anecdotally, who then recorded these conversations in her logbook. According to the logbook entries, 6 of the physicians had been aware of previous abuse experienced by some of the women participating in the study.

All the physicians were comfortable with the items on the WAST, as indicated by a mean score of 3.6 on the question “How comfortable were you in asking your patients the WAST questions?” (1=not at all comfortable; 4=very comfortable).

There was a significant association between the number of years since graduation and the reported comfort level of physicians with asking each of the WAST questions; those who had been in practice for a greater length of time were more comfortable than more recent graduates. For example, 85.7% and 100% of physicians who graduated in the 1950s and 1960s, respectively, reported feeling very comfortable asking question 8, compared with 62.1% and 0% of graduates from the 1980s and 1990s, respectively (P <.001). This trend was consistent for each of the WAST items. No significant differences were found in the level of comfort of the physicians on 6 of the WAST questions on the basis of certification status. However, this was not the case when asking the 2 items related to physical abuse, which had smaller proportions of physicians with CFPC certification feeling very comfortable compared with the noncertificants (57.4% vs 76.7% and 60.6% vs 78.1% on questions 4 and 6, respectively; P <.05). Higher proportions of women physicians than men reported being very comfortable when asking the WAST questions addressing physical, emotional, and sexual abuse (77.9% vs 54.9%; 74.8% vs 52.0%; and 77.9% vs 53.8%, respectively; P <.001). There was no association found between the comfort level of physicians and their previous knowledge of their patients.

Patient Comfort with the WAST

For all the WAST items, a minimum of 91% of the women reported being comfortable or very comfortable when asked the questions by their family physician. The average comfort level score across all items was 3.6 (Table 3). However, the abused women were significantly less comfortable than the nonabused women with the questions that addressed physical and sexual abuse (including the question asking whether arguments resulted in a violent outcome) with all 3 questions achieving a significance level of P <.05.

Discussion

The 8-item WAST was found to be a reliable and valid measure in the family practice context among the general population. The WAST-Short identified 26 women (8.5% of the sample) as experiencing abuse, and there was a significant difference between the abused and nonabused women on their total WAST scores. Although not directly transferable, these findings are noteworthy when compared with a 1993 survey of 12,300 Canadian women older than 18 years reporting that 10% of women had experienced violence in the 12 months before the survey.³³

There were no differences in the demographic characteristics between the women who screened positive for abuse and those who screened negative. However, there was a wider range of income reported by the 26 women who screened positive. This finding supports the literature, which indicates that woman abuse is present at all economic levels and in all social classes.^34,35

Both the patients and their family physicians reported they were comfortable with the WAST, and the comfort level scores of the physicians remained high despite the increasingly sensitive nature of the questions. This strong endorsement suggests that the WAST should be applied in the family practice setting. The majority of physicians perceived the WAST to be helpful for identifying women experiencing abuse and indicated their intentions to continue using it.

Physicians who had been practicing longer expressed more comfort with asking the WAST questions than did their colleagues with less experience. This may reflect their greater awareness of the important role played by psychosocial factors in the lives and health of their patients.

Tudiver and Permaul-Woods³⁶ found no difference in the perceived diagnostic skills for identifying woman abuse between certificants and noncertificants of the CFPC. Our study findings indicate that certificants were less comfortable in asking the 2 questions about physical abuse. Despite their reluctance to ask these questions, the majority of physicians with CFPC certification indicated their commitment to continue using the WAST. The ultimate test will be to see if family physicians persist in the application of the WAST despite fears of opening a “Pandora’s box”⁷ or “a can of worms”.⁸

Some authors have considered the influence of physician sex on the level of comfort of physicians inquiring about abuse.^37,38 In our study the women physicians reported more comfort than the men in asking about emotional, physical, and sexual abuse.

The vast majority of women patients were comfortable in being asked the WAST questions. However, those who screened positive for abuse did express less comfort with questions related to physical and sexual abuse. These findings suggest that for some patients discussing abuse with their family physician may be problematic. They may view physical violence as socially unacceptable behavior and thus a taboo subject for discussion. It may also reflect the patient’s feelings of shame, fear, guilt, and self blame.^11,22,24,25 An environment promoting safety, confidentiality, respect, trust, caring, validation, and a nonjudgemental atmosphere is necessary when screening for abuse.^{22,23,27,29,39}

Compared with a decade ago, several reliable and valid screening tools for detecting woman abuse are now available for use by primary care physicians.^18,40-42 The WAST joins the menu of screening tools from which physicians can choose. Its future use is supported by the reported physician and patient comfort levels with its questions being asked during the clinical encounter.

Limitations

Our study was based on a sample of family physicians drawn from a single geographic area, which limits the generalizability of the findings to physicians in other regions. Also, because of the recruitment method physicians may have agreed to participate because of their previous knowledge of the recruiter’s expertise in the field of abuse, resulting in a biased sample. Although the majority of physicians indicated that they would continue to use the WAST in the future we did not ask them how this would occur. Our recommendation would be that at minimum the WAST-Short be administered to women presenting for routine visits, including complete physical examinations and prenatal care as well as acute complaints.

As reported, we did not systematically inquire about the physician’s previous knowledge of the past abuse of a participant. Furthermore, we did not document if a specific intervention transpired with the women identified as abused. These issues are paramount if screening tools for woman abuse are to be viewed as useful and effective in addressing this serious problem. Future studies should include ways to assess and evaluate both interventions and patient outcomes.

The occurrence of abuse in this group of patients may have been underestimated. The information spontaneously offered by some patients at the time of their refusal to participate in our study suggests that they were in an abusive relationship. This reflects the reality of conducting research on a sensitive issue. Also, the preponderance of white English-speaking middle-class women in our study may limit the generalizability to more diverse populations.

However, these limitations do not detract from the important findings of our study, which demonstrates that the WAST-Short questionnaire identifies women experiencing abuse, and the full 8-item WAST helps family physicians explore the extent of that abuse. Finally, and perhaps of most clinical significance, both patients and family physicians were comfortable with the incorporation of WAST into the clinical encounter.

Acknowledgments

Our study was supported by a grant from Searle Canada. The conclusions are those of the authors, and no endorsement by Searle Canada is intended or should be inferred.