Original Research

Unlike quantitative studies, qualitative studies proceed without formulating hypotheses, and they do not draw conclusions based on numerical data. The information they obtain can be used to generate ideas and hypotheses, which can then be tested with quantitative methods.

Qualitative methods—eg, focus groups, observation, in-depth individual interviews—emphasize the meaning and process of a clinical issue rather than the outcomes of interventions. Consider this simple, fictional example:

An educational intervention designed to improve compliance with asthma care was implemented among patients from indigent families. The intervention improved the rate of maintenance steroid inhaler use by a nonsignificant 3% of participants. Disappointed by this result, the researchers used qualitative methods to try to find out why the intervention did not have a more significant effect.

They met with several small groups of participants for loosely structured interviews known as “focus groups” and asked them specific questions about the educational program. One common theme from the focus group responses was that participants believed the facilitators of the educational intervention were condescending and did not understand the challenges faced by poor families. This finding led the researchers to redesign their intervention with facilitators who were better trained to support the needs of their targeted patient population.

When qualitative studies are helpful

In general, research topics that relate to subjects’ perceptions, attitudes, or opinions are well suited to qualitative research methods. Unlike clinical trials, no attempt is made in qualitative studies to compare identical groups or minimize confounding. In fact, differences among participants contribute positively to understanding of a topic.

However, like quantitative researchers, qualitative researchers begin with a clinical question. Qualitative methods, however, permit changing the question as new information is gathered. Rather than using a predetermined time frame or number of study subjects, qualitative studies continue until sufficient data have been gathered. Researchers may end the study when the responses of participants start falling into certain themes or patterns and additional responses do not add any new information.

CORRESPONDENCE
Sukanya Srinivasan, MD, MPH. E-mail: [email protected]

Unlike quantitative studies, qualitative studies proceed without formulating hypotheses, and they do not draw conclusions based on numerical data. The information they obtain can be used to generate ideas and hypotheses, which can then be tested with quantitative methods.

Qualitative methods—eg, focus groups, observation, in-depth individual interviews—emphasize the meaning and process of a clinical issue rather than the outcomes of interventions. Consider this simple, fictional example:

An educational intervention designed to improve compliance with asthma care was implemented among patients from indigent families. The intervention improved the rate of maintenance steroid inhaler use by a nonsignificant 3% of participants. Disappointed by this result, the researchers used qualitative methods to try to find out why the intervention did not have a more significant effect.

They met with several small groups of participants for loosely structured interviews known as “focus groups” and asked them specific questions about the educational program. One common theme from the focus group responses was that participants believed the facilitators of the educational intervention were condescending and did not understand the challenges faced by poor families. This finding led the researchers to redesign their intervention with facilitators who were better trained to support the needs of their targeted patient population.

When qualitative studies are helpful

In general, research topics that relate to subjects’ perceptions, attitudes, or opinions are well suited to qualitative research methods. Unlike clinical trials, no attempt is made in qualitative studies to compare identical groups or minimize confounding. In fact, differences among participants contribute positively to understanding of a topic.

However, like quantitative researchers, qualitative researchers begin with a clinical question. Qualitative methods, however, permit changing the question as new information is gathered. Rather than using a predetermined time frame or number of study subjects, qualitative studies continue until sufficient data have been gathered. Researchers may end the study when the responses of participants start falling into certain themes or patterns and additional responses do not add any new information.

CORRESPONDENCE
Sukanya Srinivasan, MD, MPH. E-mail: [email protected]

Unlike quantitative studies, qualitative studies proceed without formulating hypotheses, and they do not draw conclusions based on numerical data. The information they obtain can be used to generate ideas and hypotheses, which can then be tested with quantitative methods.

Qualitative methods—eg, focus groups, observation, in-depth individual interviews—emphasize the meaning and process of a clinical issue rather than the outcomes of interventions. Consider this simple, fictional example:

An educational intervention designed to improve compliance with asthma care was implemented among patients from indigent families. The intervention improved the rate of maintenance steroid inhaler use by a nonsignificant 3% of participants. Disappointed by this result, the researchers used qualitative methods to try to find out why the intervention did not have a more significant effect.

They met with several small groups of participants for loosely structured interviews known as “focus groups” and asked them specific questions about the educational program. One common theme from the focus group responses was that participants believed the facilitators of the educational intervention were condescending and did not understand the challenges faced by poor families. This finding led the researchers to redesign their intervention with facilitators who were better trained to support the needs of their targeted patient population.

When qualitative studies are helpful

In general, research topics that relate to subjects’ perceptions, attitudes, or opinions are well suited to qualitative research methods. Unlike clinical trials, no attempt is made in qualitative studies to compare identical groups or minimize confounding. In fact, differences among participants contribute positively to understanding of a topic.

However, like quantitative researchers, qualitative researchers begin with a clinical question. Qualitative methods, however, permit changing the question as new information is gathered. Rather than using a predetermined time frame or number of study subjects, qualitative studies continue until sufficient data have been gathered. Researchers may end the study when the responses of participants start falling into certain themes or patterns and additional responses do not add any new information.

CORRESPONDENCE
Sukanya Srinivasan, MD, MPH. E-mail: [email protected]

Arecent medical journal debate focuses on effects of direct-to-consumer advertising of prescription drugs (DTCA) on the physician-patient relationship.^1-11 Both sides contend that DTCA alters consumers’ communication behavior, and, ultimately, relationships with physicians, by encouraging greater patient participation and control. Increasingly, patients are asking physicians about advertised products and doctors do feel pressured to prescribe.^12-20 Thus, research to date has indicated that social norms for physician-patient communication are changing, but has not accounted for DTCA’s features that focus directly on physician-patient communication. This study examines DTCA’s references to physician-patient communication that may imply guidelines for consumers’ interaction behavior.

Pro and con opinions. Opinions vary regarding DTCA’s effects on health care and public health.²¹ Critics disagree about DTCA’s effects on cost (including time),^{1-3,8,10,22-27} consumers’ knowledge,^2,24-25 and health care quality.^1,3,4,24-25 Advocates view DTCA as empowering patients to partner with physicians,^4,24 initiate discussion,²⁵ show interest, and ask questions.^27,28 Opponents say DTCA undermines the relationship,^2,24,29,30 by overloading physicians with time-consuming questions they are unprepared to answer,^25,31 creating pressure to prescribe, and increasing patient demand that yields inappropriate prescribing.³²

The issue centers on who should be “in charge.” Proponents tend to value patients’ empowerment;^4,33 opponents generally advocate physicians’ authority.³⁴ However, both sides agree that DTCA influences patients to communicate more actively and take greater control.

Ultimate goal of DTCA.Because obtaining prescription drugs requires physicians’ cooperation, DTCA’s aims differ from traditional advertising. Successful ads must both attract consumers to products and facilitate consumers gaining physicians’ cooperation. Even “sold” consumers may not have the communication skills to interact appropriately and persuasively with physicians. Thus, to succeed commercially, DTCA must encourage particular consumer communication behaviors.

Establishing who is in control. Physician-patient relationships are developed and maintained largely via communication patterns. Communication patterns associated with physician-patient relationship models differ, largely, in terms of relational control.^35,36 Relational control, accomplished through communication, “refers to the process of establishing [who has] the right to direct, delimit, and define the actions of the dyad,” in this case, the physician-patient relationship.³⁷

Paternalism³⁶ casts the physician in control of information and decisions, and the patient as expected to cooperate.³⁸

Participatory models³⁵ reflect a partnership with relatively equal power evident in mutual information sharing and exploration of alternatives.³⁸

Consumerism places control in patients’ hands; consumers may bargain and engage actively in communication, but theoretically they control final decisions and may demand particular treatment regimens.³⁶

Are DTCAs “training” consumers? Previous content analyses of DTCA focus on marketing factors (eg, ad frequency, product type)^39,40 and on appeals, motivators, or inducements for consumers,^39-41 but do not address DTCA’s statements about physician-patient communication.⁴² When social cognitive theory is applied to DTCA, it suggests that DTCA may “train” consumers by providing models or examples from which to learn vicariously, while associating those models with positive outcomes or rewards, and the advertised drug, thus motivating consumers to seek the product.⁴³ Thus, DTCA may encourage specific communication behaviors as the means to acquire advertised products. If so, its influence may lie less in its educational function than in its social training function.

Although medical information may help consumers establish credibility and arm them with medical content for discussion, DTCA’s statements about communication may imply guidelines for interacting appropriately with physicians. An ad that reads, “Ask your doctor about drug X,” explicitly provides a model opening line and contains implicit messages about who should initiate interaction (the consumer, encouraged to ask), who should have control (the doctor, upon whom the patient depends for an answer), and appropriate interaction topics (drug X). This interpretation of DTCA’s messages is rooted in relational communication theory and research; a consumer urged to “ask” a physician is cast as “one-down”³⁷ or dependent on the physician for an answer. Alternatively, a consumer urged to “tell” a physician is portrayed as “in charge.” A message to “discuss” a matter with a physician indicates shared control.

The aim of our study. Our general question was: “What social norms regarding physician-patient communication does print DTCA suggest to consumers?” Specific research questions were:

How frequent are references to physicians in print DTCA?

How frequent are messages about physician-patient communication in print DTCA?

Within messages about physician-patient communication, what guidelines are implied, and with what frequencies, regarding: (a) who should initiate interaction, (b) who should have relational control, and (c) appropriate topics for interaction?

Methods

Sample

We examined all DTCA in 18 popular magazines (684 issues) from January 1998 to December 1999. We followed Bell, Kravitz, and Wilkes’s procedures to ensure a diverse sample of publications.³⁹ Thirteen magazines were selected to represent the highest-ranked lay magazines (based on advertising pages sold) in specified categories; 5 additional magazines were selected to represent diverse populations. They were business (Business Week), fishing/hunting/guns (Field & Stream), food/wine (Gourmet), home (Better Homes and Gardens), men (GQ), music (Rolling Stone), news and opinion (Time), parenting (Parents), personal finance (Money Magazine), sports (Sports Illustrated), tabloid/general editorial (Reader’s Digest), women (Vogue), and medicine/health (Prevention); and ethnicity (Ebony and Hispanic), age (Modern Maturity and New Choices for the Best Years), and sexual orientation (The Advocate). We identified 994 product-specific and reminder ads for 83 drugs addressing 15 types of medical conditions.²² (Product-specific ads identify products by name and use and are subject to FDA monitoring guidelines.²² Reminder ads simply identify products by name, without identifying use or related claims, risks, etc.) After eliminating duplicates, the sample of 225 advertisements was analyzed.

Coding systems

The unit of analysis for this investigation was a statement focusing on physician-patient communication. For each advertisement, we first identified statements referencing physicians. (Although we included the terms “health provider” and “health professional” as references to physicians, all but 4 ads used the terms “physician” or “doctor.”)

Next, among references to physicians, we identified statements focusing on physician-patient communication (eg, “ask your physician;” “your doctor will tell you”). For these statements, we developed a coding system to reflect types of messages implied regarding physician-patient communication by systematically reviewing 25% of the sample, while considering relational control theory.³⁷ Specific categories of messages, examples, and rules for coding were developed for 3 variables: (a) who should initiate communication, (b) who should take control, and (c) appropriate communication topic(s). Categories for each variable were mutually exclusive and exhaustive.

Upon completing development of the coding system, we applied it to the full sample of statements focusing on physician-patient communication. In addition, for each statement, the medical condition for which the drug was advertised was coded (14 disclosed conditions and a category for undisclosed conditions). Details of the coding system are available from the authors.

Initiating communication. Who should initiate communication was coded as (a) explicit directives to the consumer to initiate communication (eg, “ask your doctor,” “tell your doctor”), (b) implicit directives to the consumer to initiate communication (eg, “see your doctor about drug X,” “check with your doctor”), (c) references to doctor-initiated communication (eg, “your doctor will tell you,” “adhere to your doctor’s recommendations”), or (d) messages referencing both parties, implying either could initiate communication (eg, “my doctor and I agreed,” “you and your doctor must carefully discuss”).

Relational control. Consistent with relational control theory, 37 who should be in control was coded as (a) patient control (eg, “tell your doctor,” “let your professional know”), (b) physician control (eg, “ask your doctor,” “check with your doctor”), or (c) shared or unclear control (eg, “talk to your doctor,” “discuss with your doctor”).

Appropriate interaction topics. Suggested interaction topics were coded as (a) side effects, (b) risks of product use, (c) general product information, (d) clinical judgments (ie, determining appropriateness for the patient), or (e) topic unspecified or unclear (included multiple topics).

TABLE 1
References to physician(s) and to communication with physician(s) by medical condition

Coding procedures

A coder was trained, and initially acceptable inter- and intra-rater reliability levels were established. To eliminate effects due to particular magazines, products, etc, the 225 ads were placed in random order. The coder independently coded the randomlyordered sample of 225 advertisements. In addition, to assess reliability, the coder recoded (and the second author coded) a randomly selected subset of 25 ads. Final intra-rater reliabilities (percentage of agreement) and inter-rater reliabilities (Cohen’s kappa) were: initiator of interaction: 93.8%, κ=.93; relational control: 90.6%, κ=.89; and interaction topic: 92.2%, κ=.92.

TABLE 2
References regarding who initiates communication: Percentages by category

Results

References to physicians

The number of references to physicians per ad ranged from 0 to 12; the average exceeded 3 (TABLE 1). All but 4.4% of ads made reference to physicians. The major exception, mostly reminder ads for undisclosed conditions, contained little text. Numbers of references to physicians varied by disclosed medical condition, from lows of less than 2 (allergies, HIV/AIDS), to a high exceeding 6 (cardiovascular).

Physician-patient communication messages

The number of statements that focused on physician-patient communication ranged from 0 to 10 per ad. Most references to physicians (83.8%) focused on communication; typically ads contained more than 2 communication messages. Average numbers of communication messages varied by disclosed medical condition, from less than 2 (allergies, HIV/AIDS), to a high exceeding 5 (diabetes).

Cues regarding how to communicate with physicians

Who should initiate interaction. More than three quarters (76.1%) of communication references indicated that consumers should initiate communication; most did so explicitly (69.7%) (TABLE 2). The percent age of explicit directives to consumers to initiate communication varied widely by condition, from 37.5% (cancer) to 100% (non-HIV infection, undisclosed conditions). More than 50% of communication references, in all conditions except cancer, explicitly indicated the consumer as initiator. Implicit directives to consumers to initiate communication ranged from 0% (non-HIV infection) to 25% (respiratory).

Relatively few messages indicated the physician as initiator (14.1%), varying by medical condition from 0% (non-HIV infection, undisclosed) to 37.5% (cancer).

Messages indicating either party could initiate communication appeared in less than 10% of the statements (9.8%) and varied by medical condition from 0% (non-HIV infection, respiratory, undisclosed) to 34.8% (obstetric/gynecologic); this type of message appeared in less than 10% of communication messages in ads for 10 conditions.

Who should have relational control. The majority (54.5%) of communication messages placed physicians in control (TABLE 3). Nearly one third (30%) indicated shared (or unknown) control, while only about 15.5% cast consumers in control. However, relational control cues varied widely by medical condition. Physicians were cast in exclusive control in ads for undisclosed conditions (100%), although these numbers were small. For disclosed medical conditions, physician control ranged from 17.4% (obstetric/gynecologic) to 75% (dermatologic, respiratory). Consumer control ranged from 0% (gastrointestinal/nutritional, dermatologic, and undisclosed) to 38.2% (cardiovascular). One of the most striking differences due to medical condition occurred for obstetric/gynecologic ads, in which shared/unknown control dominated (80.4%).

Appropriate interaction topics. The most frequently suggested interaction topic was clinical appropriateness (41.8%), followed by general product information (32.1%) (TABLE 4). Fewer than 1-in-5 suggested topics focused on products’ negative aspects (8.5% each for side effects and risks). For 9.1% of the statements, no topic was suggested, or the suggested topic was unclear. Suggested topics varied by disclosed medical condition. Clinical judgment accounted for 30% or more of suggested topics in most disclosed medical conditions, ranging from 20% (dermatologic) to 67% (respiratory). General information accounted for 25% or more of suggested topics in most of the disclosed conditions, ranging from 12.5% (cancer) to 59% (allergies). The topic of side effects ranged from 0% (allergies, gastrointestinal/nutritional, tobacco/addiction, undisclosed conditions) to 19.1% (cardiovascular). Similarly, the topic of risks ranged from 0% (5 conditions) to 41.3% (obstetric/gynecologic). Follow-up analyses revealed that when the suggested topic was negative (risks or side effects), in only 10 of 98 cases (10.2%) was the physician indicated as initiating communication.

TABLE 3
References indicating relational control by medical condition: Percentages by category

TABLE 4
Suggested topics for physician-patient communication by medical condition: percentages by category

Discussion

Typical DTCA contains multiple messages about physician-patient communication. The primary way that DTCA may endorse a participatory model is via messages that encourage consumers to initiate conversations with physicians about products. About 70% of communication references explicitly direct consumers to do so. Otherwise, ads do not encourage consumers’ control. In fact, nearly 55% of communication references cast the physician in control, while only 15% placed the consumer in control. Thus, DTCA reinforces physicians’ relational control while encouraging consumers to initiate communication.

DTCA steers conversation topics toward products’ benefits and away from their deficits. Ads most often suggest products’ medical utility and appropriateness as topics (ie, general information, clinical judgments), while avoiding negative topics (ie, side effects, risks).

DTCA’s communication lessons for practice

Present results have implications for physician-patient interaction. First, to the extent that DTCA influences patients’ communication behavior, physicians increasingly may encounter patients who initiate communication by asking questions, often about advertised drugs. Some physicians may see such questions as requests or even demands for those drugs. Physicians report feeling pressure to prescribe products about which patients inquire;⁹ thus, patients merely asking more questions may be perceived as “demanding.”⁴⁴

However, physicians often perceive “patient demand” when patients have not specifically asked for a drug.⁴⁵ Physicians may want to check their perceptions before acting on them, recognizing that such questions may indicate patients’ preferences for a more participatory model, which, in turn, is associated with greater patient satisfaction.^46,47 Physicians desiring to avoid conflict when patients ask questions might encourage their participation rather than assuming “patient demand” or feeling pressure to alter prescribing behavior.

Second, despite some physicians’ concerns, DTCA’s communication messages do not encourage patients to take relational control, nor do they undermine physicians’ prescribing authority. Theoretically and ethically, physicians remain in control of decisions, including prescribing, by serving as learned intermediaries or “conduits of information between manufacturers and patients.”⁴⁸ Practically, physicians remain in control because their cooperation is necessary, even in cases where patients actively seek particular prescriptions.

Third, if DTCA influences patients’ choice of communication topics, patients may fail to inquire about drugs’ risks or side effects, a finding especially important in light of evidence indicating that consumers tend to not retain DTCA’s risk information.⁴⁹ Physicians need to be alerted to these trends so they ensure that conversations with patients include explicit discussion of drugs’ side effects and risks.

Limitations

This study has several limitations. First, we analyzed print DTCA only. Generalizing findings to television and Internet DTCA may not be possible.

Second, our sample, dated from 1998 to 1999, may differ systematically from current ads. However, our study does provide a theoretically-driven methodology for assessing, and understanding the implications of, changes in advertising strategies across time and media.

Third, we analyzed marketing efforts targeting consumers. Physicians are exposed to numerous pharmaceutical marketing efforts that may contain messages regarding physician-patient communication.

Fourth, we limited analysis of relational communication to relational control; communication theory and research considers additional relational dimensions (eg, affiliation, trust) that likely influence the physician-patient relationship. Finally, we identified DTCA messages that may influence consumers’ behavior; we did not investigate actual behavioral changes associated with exposure to DTCA.

ACKNOWLEDGMENTS

We wish to thank Katie M. Haynes, BS, for coding and technical assistance.

Arecent medical journal debate focuses on effects of direct-to-consumer advertising of prescription drugs (DTCA) on the physician-patient relationship.^1-11 Both sides contend that DTCA alters consumers’ communication behavior, and, ultimately, relationships with physicians, by encouraging greater patient participation and control. Increasingly, patients are asking physicians about advertised products and doctors do feel pressured to prescribe.^12-20 Thus, research to date has indicated that social norms for physician-patient communication are changing, but has not accounted for DTCA’s features that focus directly on physician-patient communication. This study examines DTCA’s references to physician-patient communication that may imply guidelines for consumers’ interaction behavior.

Pro and con opinions. Opinions vary regarding DTCA’s effects on health care and public health.²¹ Critics disagree about DTCA’s effects on cost (including time),^{1-3,8,10,22-27} consumers’ knowledge,^2,24-25 and health care quality.^1,3,4,24-25 Advocates view DTCA as empowering patients to partner with physicians,^4,24 initiate discussion,²⁵ show interest, and ask questions.^27,28 Opponents say DTCA undermines the relationship,^2,24,29,30 by overloading physicians with time-consuming questions they are unprepared to answer,^25,31 creating pressure to prescribe, and increasing patient demand that yields inappropriate prescribing.³²

The issue centers on who should be “in charge.” Proponents tend to value patients’ empowerment;^4,33 opponents generally advocate physicians’ authority.³⁴ However, both sides agree that DTCA influences patients to communicate more actively and take greater control.

Ultimate goal of DTCA.Because obtaining prescription drugs requires physicians’ cooperation, DTCA’s aims differ from traditional advertising. Successful ads must both attract consumers to products and facilitate consumers gaining physicians’ cooperation. Even “sold” consumers may not have the communication skills to interact appropriately and persuasively with physicians. Thus, to succeed commercially, DTCA must encourage particular consumer communication behaviors.

Establishing who is in control. Physician-patient relationships are developed and maintained largely via communication patterns. Communication patterns associated with physician-patient relationship models differ, largely, in terms of relational control.^35,36 Relational control, accomplished through communication, “refers to the process of establishing [who has] the right to direct, delimit, and define the actions of the dyad,” in this case, the physician-patient relationship.³⁷

Paternalism³⁶ casts the physician in control of information and decisions, and the patient as expected to cooperate.³⁸

Participatory models³⁵ reflect a partnership with relatively equal power evident in mutual information sharing and exploration of alternatives.³⁸

Consumerism places control in patients’ hands; consumers may bargain and engage actively in communication, but theoretically they control final decisions and may demand particular treatment regimens.³⁶

Are DTCAs “training” consumers? Previous content analyses of DTCA focus on marketing factors (eg, ad frequency, product type)^39,40 and on appeals, motivators, or inducements for consumers,^39-41 but do not address DTCA’s statements about physician-patient communication.⁴² When social cognitive theory is applied to DTCA, it suggests that DTCA may “train” consumers by providing models or examples from which to learn vicariously, while associating those models with positive outcomes or rewards, and the advertised drug, thus motivating consumers to seek the product.⁴³ Thus, DTCA may encourage specific communication behaviors as the means to acquire advertised products. If so, its influence may lie less in its educational function than in its social training function.

Although medical information may help consumers establish credibility and arm them with medical content for discussion, DTCA’s statements about communication may imply guidelines for interacting appropriately with physicians. An ad that reads, “Ask your doctor about drug X,” explicitly provides a model opening line and contains implicit messages about who should initiate interaction (the consumer, encouraged to ask), who should have control (the doctor, upon whom the patient depends for an answer), and appropriate interaction topics (drug X). This interpretation of DTCA’s messages is rooted in relational communication theory and research; a consumer urged to “ask” a physician is cast as “one-down”³⁷ or dependent on the physician for an answer. Alternatively, a consumer urged to “tell” a physician is portrayed as “in charge.” A message to “discuss” a matter with a physician indicates shared control.

The aim of our study. Our general question was: “What social norms regarding physician-patient communication does print DTCA suggest to consumers?” Specific research questions were:

How frequent are references to physicians in print DTCA?

How frequent are messages about physician-patient communication in print DTCA?

Within messages about physician-patient communication, what guidelines are implied, and with what frequencies, regarding: (a) who should initiate interaction, (b) who should have relational control, and (c) appropriate topics for interaction?

Methods

Sample

We examined all DTCA in 18 popular magazines (684 issues) from January 1998 to December 1999. We followed Bell, Kravitz, and Wilkes’s procedures to ensure a diverse sample of publications.³⁹ Thirteen magazines were selected to represent the highest-ranked lay magazines (based on advertising pages sold) in specified categories; 5 additional magazines were selected to represent diverse populations. They were business (Business Week), fishing/hunting/guns (Field & Stream), food/wine (Gourmet), home (Better Homes and Gardens), men (GQ), music (Rolling Stone), news and opinion (Time), parenting (Parents), personal finance (Money Magazine), sports (Sports Illustrated), tabloid/general editorial (Reader’s Digest), women (Vogue), and medicine/health (Prevention); and ethnicity (Ebony and Hispanic), age (Modern Maturity and New Choices for the Best Years), and sexual orientation (The Advocate). We identified 994 product-specific and reminder ads for 83 drugs addressing 15 types of medical conditions.²² (Product-specific ads identify products by name and use and are subject to FDA monitoring guidelines.²² Reminder ads simply identify products by name, without identifying use or related claims, risks, etc.) After eliminating duplicates, the sample of 225 advertisements was analyzed.

Coding systems

The unit of analysis for this investigation was a statement focusing on physician-patient communication. For each advertisement, we first identified statements referencing physicians. (Although we included the terms “health provider” and “health professional” as references to physicians, all but 4 ads used the terms “physician” or “doctor.”)

Next, among references to physicians, we identified statements focusing on physician-patient communication (eg, “ask your physician;” “your doctor will tell you”). For these statements, we developed a coding system to reflect types of messages implied regarding physician-patient communication by systematically reviewing 25% of the sample, while considering relational control theory.³⁷ Specific categories of messages, examples, and rules for coding were developed for 3 variables: (a) who should initiate communication, (b) who should take control, and (c) appropriate communication topic(s). Categories for each variable were mutually exclusive and exhaustive.

Upon completing development of the coding system, we applied it to the full sample of statements focusing on physician-patient communication. In addition, for each statement, the medical condition for which the drug was advertised was coded (14 disclosed conditions and a category for undisclosed conditions). Details of the coding system are available from the authors.

Initiating communication. Who should initiate communication was coded as (a) explicit directives to the consumer to initiate communication (eg, “ask your doctor,” “tell your doctor”), (b) implicit directives to the consumer to initiate communication (eg, “see your doctor about drug X,” “check with your doctor”), (c) references to doctor-initiated communication (eg, “your doctor will tell you,” “adhere to your doctor’s recommendations”), or (d) messages referencing both parties, implying either could initiate communication (eg, “my doctor and I agreed,” “you and your doctor must carefully discuss”).

Relational control. Consistent with relational control theory, 37 who should be in control was coded as (a) patient control (eg, “tell your doctor,” “let your professional know”), (b) physician control (eg, “ask your doctor,” “check with your doctor”), or (c) shared or unclear control (eg, “talk to your doctor,” “discuss with your doctor”).

Appropriate interaction topics. Suggested interaction topics were coded as (a) side effects, (b) risks of product use, (c) general product information, (d) clinical judgments (ie, determining appropriateness for the patient), or (e) topic unspecified or unclear (included multiple topics).

TABLE 1
References to physician(s) and to communication with physician(s) by medical condition

Coding procedures

A coder was trained, and initially acceptable inter- and intra-rater reliability levels were established. To eliminate effects due to particular magazines, products, etc, the 225 ads were placed in random order. The coder independently coded the randomlyordered sample of 225 advertisements. In addition, to assess reliability, the coder recoded (and the second author coded) a randomly selected subset of 25 ads. Final intra-rater reliabilities (percentage of agreement) and inter-rater reliabilities (Cohen’s kappa) were: initiator of interaction: 93.8%, κ=.93; relational control: 90.6%, κ=.89; and interaction topic: 92.2%, κ=.92.

TABLE 2
References regarding who initiates communication: Percentages by category

Results

References to physicians

The number of references to physicians per ad ranged from 0 to 12; the average exceeded 3 (TABLE 1). All but 4.4% of ads made reference to physicians. The major exception, mostly reminder ads for undisclosed conditions, contained little text. Numbers of references to physicians varied by disclosed medical condition, from lows of less than 2 (allergies, HIV/AIDS), to a high exceeding 6 (cardiovascular).

Physician-patient communication messages

The number of statements that focused on physician-patient communication ranged from 0 to 10 per ad. Most references to physicians (83.8%) focused on communication; typically ads contained more than 2 communication messages. Average numbers of communication messages varied by disclosed medical condition, from less than 2 (allergies, HIV/AIDS), to a high exceeding 5 (diabetes).

Cues regarding how to communicate with physicians

Who should initiate interaction. More than three quarters (76.1%) of communication references indicated that consumers should initiate communication; most did so explicitly (69.7%) (TABLE 2). The percent age of explicit directives to consumers to initiate communication varied widely by condition, from 37.5% (cancer) to 100% (non-HIV infection, undisclosed conditions). More than 50% of communication references, in all conditions except cancer, explicitly indicated the consumer as initiator. Implicit directives to consumers to initiate communication ranged from 0% (non-HIV infection) to 25% (respiratory).

Relatively few messages indicated the physician as initiator (14.1%), varying by medical condition from 0% (non-HIV infection, undisclosed) to 37.5% (cancer).

Messages indicating either party could initiate communication appeared in less than 10% of the statements (9.8%) and varied by medical condition from 0% (non-HIV infection, respiratory, undisclosed) to 34.8% (obstetric/gynecologic); this type of message appeared in less than 10% of communication messages in ads for 10 conditions.

Who should have relational control. The majority (54.5%) of communication messages placed physicians in control (TABLE 3). Nearly one third (30%) indicated shared (or unknown) control, while only about 15.5% cast consumers in control. However, relational control cues varied widely by medical condition. Physicians were cast in exclusive control in ads for undisclosed conditions (100%), although these numbers were small. For disclosed medical conditions, physician control ranged from 17.4% (obstetric/gynecologic) to 75% (dermatologic, respiratory). Consumer control ranged from 0% (gastrointestinal/nutritional, dermatologic, and undisclosed) to 38.2% (cardiovascular). One of the most striking differences due to medical condition occurred for obstetric/gynecologic ads, in which shared/unknown control dominated (80.4%).

Appropriate interaction topics. The most frequently suggested interaction topic was clinical appropriateness (41.8%), followed by general product information (32.1%) (TABLE 4). Fewer than 1-in-5 suggested topics focused on products’ negative aspects (8.5% each for side effects and risks). For 9.1% of the statements, no topic was suggested, or the suggested topic was unclear. Suggested topics varied by disclosed medical condition. Clinical judgment accounted for 30% or more of suggested topics in most disclosed medical conditions, ranging from 20% (dermatologic) to 67% (respiratory). General information accounted for 25% or more of suggested topics in most of the disclosed conditions, ranging from 12.5% (cancer) to 59% (allergies). The topic of side effects ranged from 0% (allergies, gastrointestinal/nutritional, tobacco/addiction, undisclosed conditions) to 19.1% (cardiovascular). Similarly, the topic of risks ranged from 0% (5 conditions) to 41.3% (obstetric/gynecologic). Follow-up analyses revealed that when the suggested topic was negative (risks or side effects), in only 10 of 98 cases (10.2%) was the physician indicated as initiating communication.

TABLE 3
References indicating relational control by medical condition: Percentages by category

TABLE 4
Suggested topics for physician-patient communication by medical condition: percentages by category

Discussion

Typical DTCA contains multiple messages about physician-patient communication. The primary way that DTCA may endorse a participatory model is via messages that encourage consumers to initiate conversations with physicians about products. About 70% of communication references explicitly direct consumers to do so. Otherwise, ads do not encourage consumers’ control. In fact, nearly 55% of communication references cast the physician in control, while only 15% placed the consumer in control. Thus, DTCA reinforces physicians’ relational control while encouraging consumers to initiate communication.

DTCA steers conversation topics toward products’ benefits and away from their deficits. Ads most often suggest products’ medical utility and appropriateness as topics (ie, general information, clinical judgments), while avoiding negative topics (ie, side effects, risks).

DTCA’s communication lessons for practice

Present results have implications for physician-patient interaction. First, to the extent that DTCA influences patients’ communication behavior, physicians increasingly may encounter patients who initiate communication by asking questions, often about advertised drugs. Some physicians may see such questions as requests or even demands for those drugs. Physicians report feeling pressure to prescribe products about which patients inquire;⁹ thus, patients merely asking more questions may be perceived as “demanding.”⁴⁴

However, physicians often perceive “patient demand” when patients have not specifically asked for a drug.⁴⁵ Physicians may want to check their perceptions before acting on them, recognizing that such questions may indicate patients’ preferences for a more participatory model, which, in turn, is associated with greater patient satisfaction.^46,47 Physicians desiring to avoid conflict when patients ask questions might encourage their participation rather than assuming “patient demand” or feeling pressure to alter prescribing behavior.

Second, despite some physicians’ concerns, DTCA’s communication messages do not encourage patients to take relational control, nor do they undermine physicians’ prescribing authority. Theoretically and ethically, physicians remain in control of decisions, including prescribing, by serving as learned intermediaries or “conduits of information between manufacturers and patients.”⁴⁸ Practically, physicians remain in control because their cooperation is necessary, even in cases where patients actively seek particular prescriptions.

Third, if DTCA influences patients’ choice of communication topics, patients may fail to inquire about drugs’ risks or side effects, a finding especially important in light of evidence indicating that consumers tend to not retain DTCA’s risk information.⁴⁹ Physicians need to be alerted to these trends so they ensure that conversations with patients include explicit discussion of drugs’ side effects and risks.

Limitations

This study has several limitations. First, we analyzed print DTCA only. Generalizing findings to television and Internet DTCA may not be possible.

Second, our sample, dated from 1998 to 1999, may differ systematically from current ads. However, our study does provide a theoretically-driven methodology for assessing, and understanding the implications of, changes in advertising strategies across time and media.

Third, we analyzed marketing efforts targeting consumers. Physicians are exposed to numerous pharmaceutical marketing efforts that may contain messages regarding physician-patient communication.

Fourth, we limited analysis of relational communication to relational control; communication theory and research considers additional relational dimensions (eg, affiliation, trust) that likely influence the physician-patient relationship. Finally, we identified DTCA messages that may influence consumers’ behavior; we did not investigate actual behavioral changes associated with exposure to DTCA.

ACKNOWLEDGMENTS

We wish to thank Katie M. Haynes, BS, for coding and technical assistance.

Arecent medical journal debate focuses on effects of direct-to-consumer advertising of prescription drugs (DTCA) on the physician-patient relationship.^1-11 Both sides contend that DTCA alters consumers’ communication behavior, and, ultimately, relationships with physicians, by encouraging greater patient participation and control. Increasingly, patients are asking physicians about advertised products and doctors do feel pressured to prescribe.^12-20 Thus, research to date has indicated that social norms for physician-patient communication are changing, but has not accounted for DTCA’s features that focus directly on physician-patient communication. This study examines DTCA’s references to physician-patient communication that may imply guidelines for consumers’ interaction behavior.

Pro and con opinions. Opinions vary regarding DTCA’s effects on health care and public health.²¹ Critics disagree about DTCA’s effects on cost (including time),^{1-3,8,10,22-27} consumers’ knowledge,^2,24-25 and health care quality.^1,3,4,24-25 Advocates view DTCA as empowering patients to partner with physicians,^4,24 initiate discussion,²⁵ show interest, and ask questions.^27,28 Opponents say DTCA undermines the relationship,^2,24,29,30 by overloading physicians with time-consuming questions they are unprepared to answer,^25,31 creating pressure to prescribe, and increasing patient demand that yields inappropriate prescribing.³²

The issue centers on who should be “in charge.” Proponents tend to value patients’ empowerment;^4,33 opponents generally advocate physicians’ authority.³⁴ However, both sides agree that DTCA influences patients to communicate more actively and take greater control.

Ultimate goal of DTCA.Because obtaining prescription drugs requires physicians’ cooperation, DTCA’s aims differ from traditional advertising. Successful ads must both attract consumers to products and facilitate consumers gaining physicians’ cooperation. Even “sold” consumers may not have the communication skills to interact appropriately and persuasively with physicians. Thus, to succeed commercially, DTCA must encourage particular consumer communication behaviors.

Establishing who is in control. Physician-patient relationships are developed and maintained largely via communication patterns. Communication patterns associated with physician-patient relationship models differ, largely, in terms of relational control.^35,36 Relational control, accomplished through communication, “refers to the process of establishing [who has] the right to direct, delimit, and define the actions of the dyad,” in this case, the physician-patient relationship.³⁷

Paternalism³⁶ casts the physician in control of information and decisions, and the patient as expected to cooperate.³⁸

Participatory models³⁵ reflect a partnership with relatively equal power evident in mutual information sharing and exploration of alternatives.³⁸

Consumerism places control in patients’ hands; consumers may bargain and engage actively in communication, but theoretically they control final decisions and may demand particular treatment regimens.³⁶

Are DTCAs “training” consumers? Previous content analyses of DTCA focus on marketing factors (eg, ad frequency, product type)^39,40 and on appeals, motivators, or inducements for consumers,^39-41 but do not address DTCA’s statements about physician-patient communication.⁴² When social cognitive theory is applied to DTCA, it suggests that DTCA may “train” consumers by providing models or examples from which to learn vicariously, while associating those models with positive outcomes or rewards, and the advertised drug, thus motivating consumers to seek the product.⁴³ Thus, DTCA may encourage specific communication behaviors as the means to acquire advertised products. If so, its influence may lie less in its educational function than in its social training function.

Although medical information may help consumers establish credibility and arm them with medical content for discussion, DTCA’s statements about communication may imply guidelines for interacting appropriately with physicians. An ad that reads, “Ask your doctor about drug X,” explicitly provides a model opening line and contains implicit messages about who should initiate interaction (the consumer, encouraged to ask), who should have control (the doctor, upon whom the patient depends for an answer), and appropriate interaction topics (drug X). This interpretation of DTCA’s messages is rooted in relational communication theory and research; a consumer urged to “ask” a physician is cast as “one-down”³⁷ or dependent on the physician for an answer. Alternatively, a consumer urged to “tell” a physician is portrayed as “in charge.” A message to “discuss” a matter with a physician indicates shared control.

The aim of our study. Our general question was: “What social norms regarding physician-patient communication does print DTCA suggest to consumers?” Specific research questions were:

How frequent are references to physicians in print DTCA?

How frequent are messages about physician-patient communication in print DTCA?

Within messages about physician-patient communication, what guidelines are implied, and with what frequencies, regarding: (a) who should initiate interaction, (b) who should have relational control, and (c) appropriate topics for interaction?

Methods

Sample

We examined all DTCA in 18 popular magazines (684 issues) from January 1998 to December 1999. We followed Bell, Kravitz, and Wilkes’s procedures to ensure a diverse sample of publications.³⁹ Thirteen magazines were selected to represent the highest-ranked lay magazines (based on advertising pages sold) in specified categories; 5 additional magazines were selected to represent diverse populations. They were business (Business Week), fishing/hunting/guns (Field & Stream), food/wine (Gourmet), home (Better Homes and Gardens), men (GQ), music (Rolling Stone), news and opinion (Time), parenting (Parents), personal finance (Money Magazine), sports (Sports Illustrated), tabloid/general editorial (Reader’s Digest), women (Vogue), and medicine/health (Prevention); and ethnicity (Ebony and Hispanic), age (Modern Maturity and New Choices for the Best Years), and sexual orientation (The Advocate). We identified 994 product-specific and reminder ads for 83 drugs addressing 15 types of medical conditions.²² (Product-specific ads identify products by name and use and are subject to FDA monitoring guidelines.²² Reminder ads simply identify products by name, without identifying use or related claims, risks, etc.) After eliminating duplicates, the sample of 225 advertisements was analyzed.

Coding systems

The unit of analysis for this investigation was a statement focusing on physician-patient communication. For each advertisement, we first identified statements referencing physicians. (Although we included the terms “health provider” and “health professional” as references to physicians, all but 4 ads used the terms “physician” or “doctor.”)

Next, among references to physicians, we identified statements focusing on physician-patient communication (eg, “ask your physician;” “your doctor will tell you”). For these statements, we developed a coding system to reflect types of messages implied regarding physician-patient communication by systematically reviewing 25% of the sample, while considering relational control theory.³⁷ Specific categories of messages, examples, and rules for coding were developed for 3 variables: (a) who should initiate communication, (b) who should take control, and (c) appropriate communication topic(s). Categories for each variable were mutually exclusive and exhaustive.

Upon completing development of the coding system, we applied it to the full sample of statements focusing on physician-patient communication. In addition, for each statement, the medical condition for which the drug was advertised was coded (14 disclosed conditions and a category for undisclosed conditions). Details of the coding system are available from the authors.

Initiating communication. Who should initiate communication was coded as (a) explicit directives to the consumer to initiate communication (eg, “ask your doctor,” “tell your doctor”), (b) implicit directives to the consumer to initiate communication (eg, “see your doctor about drug X,” “check with your doctor”), (c) references to doctor-initiated communication (eg, “your doctor will tell you,” “adhere to your doctor’s recommendations”), or (d) messages referencing both parties, implying either could initiate communication (eg, “my doctor and I agreed,” “you and your doctor must carefully discuss”).

Relational control. Consistent with relational control theory, 37 who should be in control was coded as (a) patient control (eg, “tell your doctor,” “let your professional know”), (b) physician control (eg, “ask your doctor,” “check with your doctor”), or (c) shared or unclear control (eg, “talk to your doctor,” “discuss with your doctor”).

Appropriate interaction topics. Suggested interaction topics were coded as (a) side effects, (b) risks of product use, (c) general product information, (d) clinical judgments (ie, determining appropriateness for the patient), or (e) topic unspecified or unclear (included multiple topics).

TABLE 1
References to physician(s) and to communication with physician(s) by medical condition

Coding procedures

A coder was trained, and initially acceptable inter- and intra-rater reliability levels were established. To eliminate effects due to particular magazines, products, etc, the 225 ads were placed in random order. The coder independently coded the randomlyordered sample of 225 advertisements. In addition, to assess reliability, the coder recoded (and the second author coded) a randomly selected subset of 25 ads. Final intra-rater reliabilities (percentage of agreement) and inter-rater reliabilities (Cohen’s kappa) were: initiator of interaction: 93.8%, κ=.93; relational control: 90.6%, κ=.89; and interaction topic: 92.2%, κ=.92.

TABLE 2
References regarding who initiates communication: Percentages by category

Results

References to physicians

The number of references to physicians per ad ranged from 0 to 12; the average exceeded 3 (TABLE 1). All but 4.4% of ads made reference to physicians. The major exception, mostly reminder ads for undisclosed conditions, contained little text. Numbers of references to physicians varied by disclosed medical condition, from lows of less than 2 (allergies, HIV/AIDS), to a high exceeding 6 (cardiovascular).

Physician-patient communication messages

The number of statements that focused on physician-patient communication ranged from 0 to 10 per ad. Most references to physicians (83.8%) focused on communication; typically ads contained more than 2 communication messages. Average numbers of communication messages varied by disclosed medical condition, from less than 2 (allergies, HIV/AIDS), to a high exceeding 5 (diabetes).

Cues regarding how to communicate with physicians

Who should initiate interaction. More than three quarters (76.1%) of communication references indicated that consumers should initiate communication; most did so explicitly (69.7%) (TABLE 2). The percent age of explicit directives to consumers to initiate communication varied widely by condition, from 37.5% (cancer) to 100% (non-HIV infection, undisclosed conditions). More than 50% of communication references, in all conditions except cancer, explicitly indicated the consumer as initiator. Implicit directives to consumers to initiate communication ranged from 0% (non-HIV infection) to 25% (respiratory).

Relatively few messages indicated the physician as initiator (14.1%), varying by medical condition from 0% (non-HIV infection, undisclosed) to 37.5% (cancer).

Messages indicating either party could initiate communication appeared in less than 10% of the statements (9.8%) and varied by medical condition from 0% (non-HIV infection, respiratory, undisclosed) to 34.8% (obstetric/gynecologic); this type of message appeared in less than 10% of communication messages in ads for 10 conditions.

Who should have relational control. The majority (54.5%) of communication messages placed physicians in control (TABLE 3). Nearly one third (30%) indicated shared (or unknown) control, while only about 15.5% cast consumers in control. However, relational control cues varied widely by medical condition. Physicians were cast in exclusive control in ads for undisclosed conditions (100%), although these numbers were small. For disclosed medical conditions, physician control ranged from 17.4% (obstetric/gynecologic) to 75% (dermatologic, respiratory). Consumer control ranged from 0% (gastrointestinal/nutritional, dermatologic, and undisclosed) to 38.2% (cardiovascular). One of the most striking differences due to medical condition occurred for obstetric/gynecologic ads, in which shared/unknown control dominated (80.4%).

Appropriate interaction topics. The most frequently suggested interaction topic was clinical appropriateness (41.8%), followed by general product information (32.1%) (TABLE 4). Fewer than 1-in-5 suggested topics focused on products’ negative aspects (8.5% each for side effects and risks). For 9.1% of the statements, no topic was suggested, or the suggested topic was unclear. Suggested topics varied by disclosed medical condition. Clinical judgment accounted for 30% or more of suggested topics in most disclosed medical conditions, ranging from 20% (dermatologic) to 67% (respiratory). General information accounted for 25% or more of suggested topics in most of the disclosed conditions, ranging from 12.5% (cancer) to 59% (allergies). The topic of side effects ranged from 0% (allergies, gastrointestinal/nutritional, tobacco/addiction, undisclosed conditions) to 19.1% (cardiovascular). Similarly, the topic of risks ranged from 0% (5 conditions) to 41.3% (obstetric/gynecologic). Follow-up analyses revealed that when the suggested topic was negative (risks or side effects), in only 10 of 98 cases (10.2%) was the physician indicated as initiating communication.

TABLE 3
References indicating relational control by medical condition: Percentages by category

TABLE 4
Suggested topics for physician-patient communication by medical condition: percentages by category

Discussion

Typical DTCA contains multiple messages about physician-patient communication. The primary way that DTCA may endorse a participatory model is via messages that encourage consumers to initiate conversations with physicians about products. About 70% of communication references explicitly direct consumers to do so. Otherwise, ads do not encourage consumers’ control. In fact, nearly 55% of communication references cast the physician in control, while only 15% placed the consumer in control. Thus, DTCA reinforces physicians’ relational control while encouraging consumers to initiate communication.

DTCA steers conversation topics toward products’ benefits and away from their deficits. Ads most often suggest products’ medical utility and appropriateness as topics (ie, general information, clinical judgments), while avoiding negative topics (ie, side effects, risks).

DTCA’s communication lessons for practice

Present results have implications for physician-patient interaction. First, to the extent that DTCA influences patients’ communication behavior, physicians increasingly may encounter patients who initiate communication by asking questions, often about advertised drugs. Some physicians may see such questions as requests or even demands for those drugs. Physicians report feeling pressure to prescribe products about which patients inquire;⁹ thus, patients merely asking more questions may be perceived as “demanding.”⁴⁴

However, physicians often perceive “patient demand” when patients have not specifically asked for a drug.⁴⁵ Physicians may want to check their perceptions before acting on them, recognizing that such questions may indicate patients’ preferences for a more participatory model, which, in turn, is associated with greater patient satisfaction.^46,47 Physicians desiring to avoid conflict when patients ask questions might encourage their participation rather than assuming “patient demand” or feeling pressure to alter prescribing behavior.

Second, despite some physicians’ concerns, DTCA’s communication messages do not encourage patients to take relational control, nor do they undermine physicians’ prescribing authority. Theoretically and ethically, physicians remain in control of decisions, including prescribing, by serving as learned intermediaries or “conduits of information between manufacturers and patients.”⁴⁸ Practically, physicians remain in control because their cooperation is necessary, even in cases where patients actively seek particular prescriptions.

Third, if DTCA influences patients’ choice of communication topics, patients may fail to inquire about drugs’ risks or side effects, a finding especially important in light of evidence indicating that consumers tend to not retain DTCA’s risk information.⁴⁹ Physicians need to be alerted to these trends so they ensure that conversations with patients include explicit discussion of drugs’ side effects and risks.

Limitations

This study has several limitations. First, we analyzed print DTCA only. Generalizing findings to television and Internet DTCA may not be possible.

Second, our sample, dated from 1998 to 1999, may differ systematically from current ads. However, our study does provide a theoretically-driven methodology for assessing, and understanding the implications of, changes in advertising strategies across time and media.

Third, we analyzed marketing efforts targeting consumers. Physicians are exposed to numerous pharmaceutical marketing efforts that may contain messages regarding physician-patient communication.

Fourth, we limited analysis of relational communication to relational control; communication theory and research considers additional relational dimensions (eg, affiliation, trust) that likely influence the physician-patient relationship. Finally, we identified DTCA messages that may influence consumers’ behavior; we did not investigate actual behavioral changes associated with exposure to DTCA.

ACKNOWLEDGMENTS

We wish to thank Katie M. Haynes, BS, for coding and technical assistance.

Levels of evidence can make it easier for busy physicians to apply the results of clinical research to their practice and to incorporate evidence-based medicine into patient care.

Levels of evidence are assigned to studies based on the quality of their design, validity, and applicability to patient care. The Agency for Health Care Quality and Research (AHRQ) has proposed that any system assigning levels of evidence should incorporate quality, quantity, and consistency of the evidence. Leading family medicine journals have adopted a uniform grading system known as the Strength of Recommendation Taxonomy¹ (SORT), which includes these key elements and 3 levels of evidence. SORT is one among several different methods of grading levels of evidence that make use of similar principles. SORT’s primary advantage is its simplicity.

The randomized controlled trial (RCT) is the most rigorous study design. According to SORT, RCTs that deal with patient-oriented outcomes and include concealment, double-blinding, intention-to-treat analysis, and complete follow-up (and meta-analyses or systematic reviews of such randomized trials) provide a level of evidence (LOE) of 1. Observational studies, such as cohort and case-control studies (and systematic reviews that include them), are less rigorous in their design, and they are given an LOE of 2. Level 3 evidence, the lowest level, is assigned to consensus guidelines, expert opinion, usual practice, etc, or to studies that look at intermediate or diseaseoriented outcomes.

Although the Nurses Health Study,² a large cohort trial involving nearly 88,000 women, and other observational studies (LOE: 2) suggested a cardiovascular benefit from vitamin E, the Finnish Alpha- Tocopherol, Beta-Carotene Cancer Prevention study,³ a well-designed RCT (LOE: 1) proved the opposite. A recently published Italian study⁴ provided Level 3 evidence, demonstrating that vitamin E prevented an oxidation-induced reduction in coronary blood flow. Therefore, based on the highest level of evidence available, vitamin E to prevent cardiovascular disease is not recommended.

Levels of evidence can make it easier for busy physicians to apply the results of clinical research to their practice and to incorporate evidence-based medicine into patient care.

Levels of evidence are assigned to studies based on the quality of their design, validity, and applicability to patient care. The Agency for Health Care Quality and Research (AHRQ) has proposed that any system assigning levels of evidence should incorporate quality, quantity, and consistency of the evidence. Leading family medicine journals have adopted a uniform grading system known as the Strength of Recommendation Taxonomy¹ (SORT), which includes these key elements and 3 levels of evidence. SORT is one among several different methods of grading levels of evidence that make use of similar principles. SORT’s primary advantage is its simplicity.

The randomized controlled trial (RCT) is the most rigorous study design. According to SORT, RCTs that deal with patient-oriented outcomes and include concealment, double-blinding, intention-to-treat analysis, and complete follow-up (and meta-analyses or systematic reviews of such randomized trials) provide a level of evidence (LOE) of 1. Observational studies, such as cohort and case-control studies (and systematic reviews that include them), are less rigorous in their design, and they are given an LOE of 2. Level 3 evidence, the lowest level, is assigned to consensus guidelines, expert opinion, usual practice, etc, or to studies that look at intermediate or diseaseoriented outcomes.

Although the Nurses Health Study,² a large cohort trial involving nearly 88,000 women, and other observational studies (LOE: 2) suggested a cardiovascular benefit from vitamin E, the Finnish Alpha- Tocopherol, Beta-Carotene Cancer Prevention study,³ a well-designed RCT (LOE: 1) proved the opposite. A recently published Italian study⁴ provided Level 3 evidence, demonstrating that vitamin E prevented an oxidation-induced reduction in coronary blood flow. Therefore, based on the highest level of evidence available, vitamin E to prevent cardiovascular disease is not recommended.

Levels of evidence can make it easier for busy physicians to apply the results of clinical research to their practice and to incorporate evidence-based medicine into patient care.

Levels of evidence are assigned to studies based on the quality of their design, validity, and applicability to patient care. The Agency for Health Care Quality and Research (AHRQ) has proposed that any system assigning levels of evidence should incorporate quality, quantity, and consistency of the evidence. Leading family medicine journals have adopted a uniform grading system known as the Strength of Recommendation Taxonomy¹ (SORT), which includes these key elements and 3 levels of evidence. SORT is one among several different methods of grading levels of evidence that make use of similar principles. SORT’s primary advantage is its simplicity.

The randomized controlled trial (RCT) is the most rigorous study design. According to SORT, RCTs that deal with patient-oriented outcomes and include concealment, double-blinding, intention-to-treat analysis, and complete follow-up (and meta-analyses or systematic reviews of such randomized trials) provide a level of evidence (LOE) of 1. Observational studies, such as cohort and case-control studies (and systematic reviews that include them), are less rigorous in their design, and they are given an LOE of 2. Level 3 evidence, the lowest level, is assigned to consensus guidelines, expert opinion, usual practice, etc, or to studies that look at intermediate or diseaseoriented outcomes.

Although the Nurses Health Study,² a large cohort trial involving nearly 88,000 women, and other observational studies (LOE: 2) suggested a cardiovascular benefit from vitamin E, the Finnish Alpha- Tocopherol, Beta-Carotene Cancer Prevention study,³ a well-designed RCT (LOE: 1) proved the opposite. A recently published Italian study⁴ provided Level 3 evidence, demonstrating that vitamin E prevented an oxidation-induced reduction in coronary blood flow. Therefore, based on the highest level of evidence available, vitamin E to prevent cardiovascular disease is not recommended.

Routine dipstick testing is time-consuming and expensive, especially when carried out over multiple visits. False-positive test results are frequent and often lead to further laboratory examinations. Today, when our care of patients is squeezed by both time and monetary constraints, we have a rare opportunity to make office visits more productive and to save patients the burden of unnecessary work-ups.

Review methods

We searched Medline from 1966 to September 2004 for English language articles using keyword searching for “proteinuria” or “glycosuria” and “prenatal” or “pregnancy.” We explored the Cochrane review, AHRQ National Guideline Clearinghouse, the Institute for Clinical Systems Improvement, and Google. The reference list of each article reviewed was examined for additional studies, but none were identified.

All 6 identified studies that investigated glycosuria as a predictor for gestational diabetes mellitus or proteinuria as a predictor for preeclampsia are reviewed in this analysis. One study examined both. Because every study used different dipstick methods of determining results, or definitions of abnormal, each was evaluated separately.

What the evidence shows

Found at some point in about 50% of women, glycosuria is believed to be due to an increased glomerular filtration rate.³ The renal threshold for glucose is highly variable and may lead to a positive test result for glycosuria despite a normal blood sugar. High intake of ascorbic acid or high urinary ketone levels may result in false-positive results. There have been 4 published studies designed to assess the value of glycosuria as a screen for gestational diabetes mellitus.^4-7 All used urine dipsticks (TABLE 1).

Watson: Urine test a poor screening instrument

In an observational prospective study of 500 women, Watson evaluated glycosuria (trace, ≥100 mg/dL) detected on 2 separate prenatal visits (17% of women) as a predictor of gestational diabetes.⁴ Gestational diabetes was defined as an abnormal 50-g glucose screen at 28 weeks gestation confirmed by an abnormal 100-mg 3-hour oral glucose tolerance test (OGTT).

He reported that glycosuria used as a screening test for gestational diabetes had a sensitivity of 27% and a specificity of 83% with a negative predictive value (PV–) of 96% and a positive predictive value (PV+) of 7% in a population with an unusually high prevalence of gestational diabetes of 4.4%. The high prevalence of gestational diabetes in this cohort increased the PV+ of urine screening for glycosuria. Women with severe glycosuria (>250 mg/dL, 2+) on 2 determinations during the first 2 trimesters had a 21% chance (PV+) of being diagnosed as having gestational diabetes.

The author concluded that urine testing for glucose was a poor screening test and was not worthwhile after the 28-week blood glucose challenge. He believed urine testing during the first 2 trimesters was indicated to early identify those 3.8% of women with severe (2+) glycosuria (sensitivity 18%, PV–96%). However, the incidence of glycosuria was not increased in those women with gestational diabetes when compared with those with normal glucose screening values.

Gribble: No evidence supports improved outcomes from earlier identification of gestational diabetes

Gribble et al retrospectively examined 2745 charts of women at low risk for gestational diabetes in their first 2 trimesters of pregnancy.⁵ Two urine dipstick screening determinations positive for glycosuria (≥250 mg/dL) during the first 2 trimesters before a blood glucose screening test were 7% sensitive and 98% specific with a PV–of 97% and a PV+ of 13% in a population with a prevalence of gestational diabetes of 3.1%.⁵

Less than 1% had glycosuria on their first prenatal visit and were excluded from the study. Only 7% of women (6/85) who were subsequently diagnosed with gestational diabetes had glycosuria during the first 2 trimesters of their pregnancy. There was no statistically significant association (P<.05) between glycosuria and maternal body mass index, age, history, multiparity, or birth weight of an infant greater than 4 kg. Many of these are considered risk factors for gestational diabetes. Over 8% of women with a normal 1-hour screen had glycosuria in the third trimester. Requiring 2 positive urine tests and analyzing data collected before the third trimester lowered sensitivity and the PV+.

The authors recommended continuing glycosuria testing in the first two trimesters and then stop testing after the blood screen for gestational diabetes at 24 to 28 weeks although they noted that there was no evidence to support an improved pregnancy outcome because of earlier identification in gestational diabetes.

Hooper and Buhling: Urine glucose screening should be abandoned

In a retrospective study by Hooper of 610 patients who did not have glycosuria at the first prenatal visit, I calculated a sensitivity of 36%, specificity of 98%, a NPV of 99%, and a PPV of 27% using a single glycosuria value of ≥100 mg/dL in a population with a prevalence of gestational diabetes of 1.8%.⁶ The author advised that urine screening for gestational diabetes and preeclampsia be abandoned.

In a prospective German study, 1001 women were followed throughout their pregnancy.⁷ Glycosuria was detected in 8.2% of patients. Twenty-seven percent (267/1001) had an abnormal 50-g (>140 mg/dL) glucose screening test result, 178 (67% of them) completed a 3-hour 75-g glucose diagnostic test and 37 (4.1%) had gestational diabetes.

Of the 729 patients with a normal 50-g screening test, 52 (7%) had glycosuria while of the 37 with gestational diabetes, 4 (11%) had glycosuria. Sensitivity was 11% with a PV–of 95%. The 50-gram glucose screening test was done at 33.8±3 weeks gestation, later than the 28 weeks recommended in this country. Also the cutoff values for the diagnosis of gestational diabetes were lower than those of the American Diabetes Association. Both changes would increase the incidence of gestational diabetes and the sensitivity of urine glucose screening. The authors recommended against screening for glycosuria.

Summary of the studies

Three of the 4 studies most likely overestimate the sensitivity of glycosuria for predicting gestational diabetes. All but Gribble et al included urine testing results collected in the third trimester, after the gold standard oral glucose screening test and diagnostic test were completed. Furthermore, most urine tests were probably done in the third trimester when prenatal visits occur more frequently and when glycosuria is more prevalent.⁷ Both of these factors would tend to falsely elevate the sensitivity of testing for glycosuria in the first and second trimesters, when it is theoretically most useful. Gribble et al reported that including third-trimester data did not change the predictive values of glycosuria for gestational diabetes; the other investigators did not.

TABLE 1
Accuracy of glycosuria for predicting gestational diabetes mellitus

Recommendations from professional societies

The American Diabetes Association recommends blood glucose testing as soon as possible in high-risk women and routinely at 24 to 28 weeks gestation in those at lower risk.⁸ The American College of Obstetricians and Gynecologists (ACOG) does not address urine testing for glucose.⁹ The Institute for Clinical Systems Improvement (ICSI) considers urine dipsticks for glycosuria unreliable.¹⁰

Abandoning all but the initial urinalysis may miss a few women with true but unrecognized diabetes mellitus. None of the studies presented above address this problem although screening for diabetes mellitus using urine test strips is not an ideal screening test, identifying only between 30% and 59% of a predominately middle-aged nonpregnant group.¹¹

There is no evidence that testing for gestational diabetes before 28 weeks, as might be prompted by urine testing, changes pregnancy outcome. Screening for gestational diabetes by glycosuria is not effective with low sensitivities and low positive predictive values. False-positive tests outnumber true positives 11:1, leading to unnecessary further testing. Based on the information available, it appears safe to abandon routine urine testing for glucose at every prenatal visit. This recommendation stands regardless of the debate over the value of screening for gestational diabetes by 50-g glucose challenge followed by an OGTT if indicated.¹²

Proteinuria as a predictor for preeclampsia

Proteinuria in pregnancy is common. One study of 913 women reported that 3.8% of them had proteinuria by automated dipstick testing on their first antenatal visit and 40.8% had dipstick-positive (≥1+) proteinuria at least once during the course of their pregnancy.¹³ In another study of 3122 otherwise healthy women with a single gestation, 9.8% of the women had at least 1 episode of dipstick proteinuria ≥30 mg/dL (≥1+).⁵

Detection of proteinuria in hospitalized hypertensive pregnant women by visual reading of dipsticks, as is the usual office practice, has a high false-positive rate for true proteinuria (≥300 mg/L) with a PV+ (true positives/true plus false positives) of 24% for 1+, 53% for 2+, and 93% for 3+ or 4+.¹⁴ Another study reported a PV+ of 38% for ≥1+ proteinuria.¹⁵ A recent literature review concluded that the accuracy of 1+ proteinuria in pregnant women by dipstick was “poor and therefore of limited usefulness.”¹⁶ In a busy office with a number of healthy nonhypertensive women, the false-positive rate is high due to contamination with vaginal secretions, previous exercise, high specific gravity of urine, or other benign causes.^17-19 In contrast to the high false-positive rates noted in the previous studies, Meyer et al reported a negative predictive value of only 34% for trace or negative proteinuria in hospitalized women with hypertension in pregnancy.²⁰ Proteinuria detected by dipstick using visual or automated testing alone is a poor indicator for true proteinuria although the automated method is the more accurate of the 2.¹⁴ When the measurement of proteinuria is indicated for the early identification of preeclampsia, then a random protein:creatinine ratio is a better test choice.^14,15,21

Three studies have addressed the question: Is proteinuria an accurate predictor for preeclampsia?^6,15,16 Preeclampsia is defined as an elevated blood pressure with either proteinuria or edema or both.¹⁵

In a prospective observational study carried out in Australia, 866 non-hypertensive women were tested using an automated dipstick method for proteinuria on their first prenatal visit and 35 were ≥1+ positive.¹³ Twenty-five (71%) of these women had proteinuria detected during subsequent visits, and 2 (6%) of them developed preeclampsia. Of the 833 women who did not have proteinuria on the first visit, 316 had it on sub-sequent dipstick testing, and 15 of these women developed preeclampsia. Of the 512 who never had proteinuria, 9 developed preeclampsia (sensitivity=63%, PV–=98%). Proteinuria at the first visit may be a risk factor for subsequent preeclampsia (relative risk=2.2; 95% CI, 0.49–9.6]). Of the 8 women who developed proteinuria before hypertension developed, 5 could be considered at high risk: 2 had proteinuria at their first prenatal visit, 2 had multiple gestations, and 1 had a history of preeclampsia. Pregnancy outcomes were similar in the proteinuria and no proteinuria groups. The authors recommended discontinuing urine protein testing except in high-risk women (TABLE 2).

A retrospective study of 3104 low-risk American women which excluded those at high risk (multiple gestations, diabetes mellitus, preexisting hypertension, renal disease, or ≥30 mg/dL [1+] proteinuria at the first prenatal visit) found routine visually evaluated dipstick determination for proteinuria of no value in the prediction of preeclampsia.²² In this study for the 6.1% of woman who had a blood pressure of greater than 140/90 mm Hg, a weight gain of 3 pounds a week or more, or greater than 1+ edema, testing for proteinuria was considered to be for diagnostic reasons. When the remaining 2802 patients were evaluated throughout their pregnancy, 90.3% had no proteinuria, 7.6 % were 1+, and 2.2% were ≥2+. The sensitivity and PV+ of proteinuria for preeclampsia in routine patients were 5% and 96% respectively.

The presence of proteinuria was increased in younger women and those with a greater prepregnant body mass index but not with pregnancy-associated hypertension—preeclampsia, fetal distress, abruption, low birth weight, prematurity, stillbirth, or Apgar scores less than 7 at 5 minutes. The authors concluded that there is no evidence supporting routine urine dipstick protein determinations during uncomplicated prenatal visits.

In another retrospective study of 610 women, 18 % had ≥1+ proteinuria during at least one prenatal visit and 17 (3%) developed preeclampsia.⁶ Three women with preeclampsia (17%) developed proteinuria before hypertension. But the timing of the appearance of proteinuria was not otherwise specified, and it may have been remote from the hypertension. The sensitivity of proteinuria detected prior to the onset of hypertension for preeclampsia was 71% with a PV–of 99%. The author advised against routine dipstick testing.

TABLE 2
Accuracy of proteinuria for predicting preeclampsia

Recommendations and practices of others

Routine testing at antenatal visits for proteinuria is not helpful in predicting preeclampsia and should be targeted at women with an increased blood pressure or acute weight gain. ACOG advises that there is no reliable predictive test for preeclampsia.²³ The US Preventive Services Task Force advises urine testing for protein only after abnormalities in blood pressure appear.²⁴ The Canadian Task Force on the Periodic Health Examination²⁵ and other groups in Australia²⁶ advise against testing, as does a standard textbook of obstetrics.²⁷ ICSI suggests that prenatal care would be improved by discontinuing routine urine dipstick testing.¹⁰

Most groups support further evaluation of proteinuria^21,26,30 or glycosuria²⁶ found on the initial urinalysis at the first prenatal visit although there is little evidence to support this course of action.³¹ Based on the results of these studies and the recommendations of other groups, it is reasonable to reserve urine protein testing (using a more accurate method than a dipstick) for women with an elevated blood pressure.

Acknowledgments

The author wishes to thank Thomas L. Mead and Cora Damon for assisting in the library research of the topic and Colleen Flewelling for technical assistance.

CORRESPONDING AUTHOR
William A. Alto, MD, MPH, 4 Sheridan Drive, Fairfield, ME 04937. E-mail: [email protected]

Routine dipstick testing is time-consuming and expensive, especially when carried out over multiple visits. False-positive test results are frequent and often lead to further laboratory examinations. Today, when our care of patients is squeezed by both time and monetary constraints, we have a rare opportunity to make office visits more productive and to save patients the burden of unnecessary work-ups.

Review methods

We searched Medline from 1966 to September 2004 for English language articles using keyword searching for “proteinuria” or “glycosuria” and “prenatal” or “pregnancy.” We explored the Cochrane review, AHRQ National Guideline Clearinghouse, the Institute for Clinical Systems Improvement, and Google. The reference list of each article reviewed was examined for additional studies, but none were identified.

All 6 identified studies that investigated glycosuria as a predictor for gestational diabetes mellitus or proteinuria as a predictor for preeclampsia are reviewed in this analysis. One study examined both. Because every study used different dipstick methods of determining results, or definitions of abnormal, each was evaluated separately.

What the evidence shows

Found at some point in about 50% of women, glycosuria is believed to be due to an increased glomerular filtration rate.³ The renal threshold for glucose is highly variable and may lead to a positive test result for glycosuria despite a normal blood sugar. High intake of ascorbic acid or high urinary ketone levels may result in false-positive results. There have been 4 published studies designed to assess the value of glycosuria as a screen for gestational diabetes mellitus.^4-7 All used urine dipsticks (TABLE 1).

Watson: Urine test a poor screening instrument

In an observational prospective study of 500 women, Watson evaluated glycosuria (trace, ≥100 mg/dL) detected on 2 separate prenatal visits (17% of women) as a predictor of gestational diabetes.⁴ Gestational diabetes was defined as an abnormal 50-g glucose screen at 28 weeks gestation confirmed by an abnormal 100-mg 3-hour oral glucose tolerance test (OGTT).

He reported that glycosuria used as a screening test for gestational diabetes had a sensitivity of 27% and a specificity of 83% with a negative predictive value (PV–) of 96% and a positive predictive value (PV+) of 7% in a population with an unusually high prevalence of gestational diabetes of 4.4%. The high prevalence of gestational diabetes in this cohort increased the PV+ of urine screening for glycosuria. Women with severe glycosuria (>250 mg/dL, 2+) on 2 determinations during the first 2 trimesters had a 21% chance (PV+) of being diagnosed as having gestational diabetes.

The author concluded that urine testing for glucose was a poor screening test and was not worthwhile after the 28-week blood glucose challenge. He believed urine testing during the first 2 trimesters was indicated to early identify those 3.8% of women with severe (2+) glycosuria (sensitivity 18%, PV–96%). However, the incidence of glycosuria was not increased in those women with gestational diabetes when compared with those with normal glucose screening values.

Gribble: No evidence supports improved outcomes from earlier identification of gestational diabetes

Gribble et al retrospectively examined 2745 charts of women at low risk for gestational diabetes in their first 2 trimesters of pregnancy.⁵ Two urine dipstick screening determinations positive for glycosuria (≥250 mg/dL) during the first 2 trimesters before a blood glucose screening test were 7% sensitive and 98% specific with a PV–of 97% and a PV+ of 13% in a population with a prevalence of gestational diabetes of 3.1%.⁵

Less than 1% had glycosuria on their first prenatal visit and were excluded from the study. Only 7% of women (6/85) who were subsequently diagnosed with gestational diabetes had glycosuria during the first 2 trimesters of their pregnancy. There was no statistically significant association (P<.05) between glycosuria and maternal body mass index, age, history, multiparity, or birth weight of an infant greater than 4 kg. Many of these are considered risk factors for gestational diabetes. Over 8% of women with a normal 1-hour screen had glycosuria in the third trimester. Requiring 2 positive urine tests and analyzing data collected before the third trimester lowered sensitivity and the PV+.

The authors recommended continuing glycosuria testing in the first two trimesters and then stop testing after the blood screen for gestational diabetes at 24 to 28 weeks although they noted that there was no evidence to support an improved pregnancy outcome because of earlier identification in gestational diabetes.

Hooper and Buhling: Urine glucose screening should be abandoned

In a retrospective study by Hooper of 610 patients who did not have glycosuria at the first prenatal visit, I calculated a sensitivity of 36%, specificity of 98%, a NPV of 99%, and a PPV of 27% using a single glycosuria value of ≥100 mg/dL in a population with a prevalence of gestational diabetes of 1.8%.⁶ The author advised that urine screening for gestational diabetes and preeclampsia be abandoned.

In a prospective German study, 1001 women were followed throughout their pregnancy.⁷ Glycosuria was detected in 8.2% of patients. Twenty-seven percent (267/1001) had an abnormal 50-g (>140 mg/dL) glucose screening test result, 178 (67% of them) completed a 3-hour 75-g glucose diagnostic test and 37 (4.1%) had gestational diabetes.

Of the 729 patients with a normal 50-g screening test, 52 (7%) had glycosuria while of the 37 with gestational diabetes, 4 (11%) had glycosuria. Sensitivity was 11% with a PV–of 95%. The 50-gram glucose screening test was done at 33.8±3 weeks gestation, later than the 28 weeks recommended in this country. Also the cutoff values for the diagnosis of gestational diabetes were lower than those of the American Diabetes Association. Both changes would increase the incidence of gestational diabetes and the sensitivity of urine glucose screening. The authors recommended against screening for glycosuria.

Summary of the studies

Three of the 4 studies most likely overestimate the sensitivity of glycosuria for predicting gestational diabetes. All but Gribble et al included urine testing results collected in the third trimester, after the gold standard oral glucose screening test and diagnostic test were completed. Furthermore, most urine tests were probably done in the third trimester when prenatal visits occur more frequently and when glycosuria is more prevalent.⁷ Both of these factors would tend to falsely elevate the sensitivity of testing for glycosuria in the first and second trimesters, when it is theoretically most useful. Gribble et al reported that including third-trimester data did not change the predictive values of glycosuria for gestational diabetes; the other investigators did not.

TABLE 1
Accuracy of glycosuria for predicting gestational diabetes mellitus

Recommendations from professional societies

The American Diabetes Association recommends blood glucose testing as soon as possible in high-risk women and routinely at 24 to 28 weeks gestation in those at lower risk.⁸ The American College of Obstetricians and Gynecologists (ACOG) does not address urine testing for glucose.⁹ The Institute for Clinical Systems Improvement (ICSI) considers urine dipsticks for glycosuria unreliable.¹⁰

Abandoning all but the initial urinalysis may miss a few women with true but unrecognized diabetes mellitus. None of the studies presented above address this problem although screening for diabetes mellitus using urine test strips is not an ideal screening test, identifying only between 30% and 59% of a predominately middle-aged nonpregnant group.¹¹

There is no evidence that testing for gestational diabetes before 28 weeks, as might be prompted by urine testing, changes pregnancy outcome. Screening for gestational diabetes by glycosuria is not effective with low sensitivities and low positive predictive values. False-positive tests outnumber true positives 11:1, leading to unnecessary further testing. Based on the information available, it appears safe to abandon routine urine testing for glucose at every prenatal visit. This recommendation stands regardless of the debate over the value of screening for gestational diabetes by 50-g glucose challenge followed by an OGTT if indicated.¹²

Proteinuria as a predictor for preeclampsia

Proteinuria in pregnancy is common. One study of 913 women reported that 3.8% of them had proteinuria by automated dipstick testing on their first antenatal visit and 40.8% had dipstick-positive (≥1+) proteinuria at least once during the course of their pregnancy.¹³ In another study of 3122 otherwise healthy women with a single gestation, 9.8% of the women had at least 1 episode of dipstick proteinuria ≥30 mg/dL (≥1+).⁵

Detection of proteinuria in hospitalized hypertensive pregnant women by visual reading of dipsticks, as is the usual office practice, has a high false-positive rate for true proteinuria (≥300 mg/L) with a PV+ (true positives/true plus false positives) of 24% for 1+, 53% for 2+, and 93% for 3+ or 4+.¹⁴ Another study reported a PV+ of 38% for ≥1+ proteinuria.¹⁵ A recent literature review concluded that the accuracy of 1+ proteinuria in pregnant women by dipstick was “poor and therefore of limited usefulness.”¹⁶ In a busy office with a number of healthy nonhypertensive women, the false-positive rate is high due to contamination with vaginal secretions, previous exercise, high specific gravity of urine, or other benign causes.^17-19 In contrast to the high false-positive rates noted in the previous studies, Meyer et al reported a negative predictive value of only 34% for trace or negative proteinuria in hospitalized women with hypertension in pregnancy.²⁰ Proteinuria detected by dipstick using visual or automated testing alone is a poor indicator for true proteinuria although the automated method is the more accurate of the 2.¹⁴ When the measurement of proteinuria is indicated for the early identification of preeclampsia, then a random protein:creatinine ratio is a better test choice.^14,15,21

Three studies have addressed the question: Is proteinuria an accurate predictor for preeclampsia?^6,15,16 Preeclampsia is defined as an elevated blood pressure with either proteinuria or edema or both.¹⁵

In a prospective observational study carried out in Australia, 866 non-hypertensive women were tested using an automated dipstick method for proteinuria on their first prenatal visit and 35 were ≥1+ positive.¹³ Twenty-five (71%) of these women had proteinuria detected during subsequent visits, and 2 (6%) of them developed preeclampsia. Of the 833 women who did not have proteinuria on the first visit, 316 had it on sub-sequent dipstick testing, and 15 of these women developed preeclampsia. Of the 512 who never had proteinuria, 9 developed preeclampsia (sensitivity=63%, PV–=98%). Proteinuria at the first visit may be a risk factor for subsequent preeclampsia (relative risk=2.2; 95% CI, 0.49–9.6]). Of the 8 women who developed proteinuria before hypertension developed, 5 could be considered at high risk: 2 had proteinuria at their first prenatal visit, 2 had multiple gestations, and 1 had a history of preeclampsia. Pregnancy outcomes were similar in the proteinuria and no proteinuria groups. The authors recommended discontinuing urine protein testing except in high-risk women (TABLE 2).

A retrospective study of 3104 low-risk American women which excluded those at high risk (multiple gestations, diabetes mellitus, preexisting hypertension, renal disease, or ≥30 mg/dL [1+] proteinuria at the first prenatal visit) found routine visually evaluated dipstick determination for proteinuria of no value in the prediction of preeclampsia.²² In this study for the 6.1% of woman who had a blood pressure of greater than 140/90 mm Hg, a weight gain of 3 pounds a week or more, or greater than 1+ edema, testing for proteinuria was considered to be for diagnostic reasons. When the remaining 2802 patients were evaluated throughout their pregnancy, 90.3% had no proteinuria, 7.6 % were 1+, and 2.2% were ≥2+. The sensitivity and PV+ of proteinuria for preeclampsia in routine patients were 5% and 96% respectively.

The presence of proteinuria was increased in younger women and those with a greater prepregnant body mass index but not with pregnancy-associated hypertension—preeclampsia, fetal distress, abruption, low birth weight, prematurity, stillbirth, or Apgar scores less than 7 at 5 minutes. The authors concluded that there is no evidence supporting routine urine dipstick protein determinations during uncomplicated prenatal visits.

In another retrospective study of 610 women, 18 % had ≥1+ proteinuria during at least one prenatal visit and 17 (3%) developed preeclampsia.⁶ Three women with preeclampsia (17%) developed proteinuria before hypertension. But the timing of the appearance of proteinuria was not otherwise specified, and it may have been remote from the hypertension. The sensitivity of proteinuria detected prior to the onset of hypertension for preeclampsia was 71% with a PV–of 99%. The author advised against routine dipstick testing.

TABLE 2
Accuracy of proteinuria for predicting preeclampsia

Recommendations and practices of others

Routine testing at antenatal visits for proteinuria is not helpful in predicting preeclampsia and should be targeted at women with an increased blood pressure or acute weight gain. ACOG advises that there is no reliable predictive test for preeclampsia.²³ The US Preventive Services Task Force advises urine testing for protein only after abnormalities in blood pressure appear.²⁴ The Canadian Task Force on the Periodic Health Examination²⁵ and other groups in Australia²⁶ advise against testing, as does a standard textbook of obstetrics.²⁷ ICSI suggests that prenatal care would be improved by discontinuing routine urine dipstick testing.¹⁰

Most groups support further evaluation of proteinuria^21,26,30 or glycosuria²⁶ found on the initial urinalysis at the first prenatal visit although there is little evidence to support this course of action.³¹ Based on the results of these studies and the recommendations of other groups, it is reasonable to reserve urine protein testing (using a more accurate method than a dipstick) for women with an elevated blood pressure.

Acknowledgments

The author wishes to thank Thomas L. Mead and Cora Damon for assisting in the library research of the topic and Colleen Flewelling for technical assistance.

CORRESPONDING AUTHOR
William A. Alto, MD, MPH, 4 Sheridan Drive, Fairfield, ME 04937. E-mail: [email protected]

Routine dipstick testing is time-consuming and expensive, especially when carried out over multiple visits. False-positive test results are frequent and often lead to further laboratory examinations. Today, when our care of patients is squeezed by both time and monetary constraints, we have a rare opportunity to make office visits more productive and to save patients the burden of unnecessary work-ups.

Review methods

We searched Medline from 1966 to September 2004 for English language articles using keyword searching for “proteinuria” or “glycosuria” and “prenatal” or “pregnancy.” We explored the Cochrane review, AHRQ National Guideline Clearinghouse, the Institute for Clinical Systems Improvement, and Google. The reference list of each article reviewed was examined for additional studies, but none were identified.

All 6 identified studies that investigated glycosuria as a predictor for gestational diabetes mellitus or proteinuria as a predictor for preeclampsia are reviewed in this analysis. One study examined both. Because every study used different dipstick methods of determining results, or definitions of abnormal, each was evaluated separately.

What the evidence shows

Found at some point in about 50% of women, glycosuria is believed to be due to an increased glomerular filtration rate.³ The renal threshold for glucose is highly variable and may lead to a positive test result for glycosuria despite a normal blood sugar. High intake of ascorbic acid or high urinary ketone levels may result in false-positive results. There have been 4 published studies designed to assess the value of glycosuria as a screen for gestational diabetes mellitus.^4-7 All used urine dipsticks (TABLE 1).

Watson: Urine test a poor screening instrument

In an observational prospective study of 500 women, Watson evaluated glycosuria (trace, ≥100 mg/dL) detected on 2 separate prenatal visits (17% of women) as a predictor of gestational diabetes.⁴ Gestational diabetes was defined as an abnormal 50-g glucose screen at 28 weeks gestation confirmed by an abnormal 100-mg 3-hour oral glucose tolerance test (OGTT).

He reported that glycosuria used as a screening test for gestational diabetes had a sensitivity of 27% and a specificity of 83% with a negative predictive value (PV–) of 96% and a positive predictive value (PV+) of 7% in a population with an unusually high prevalence of gestational diabetes of 4.4%. The high prevalence of gestational diabetes in this cohort increased the PV+ of urine screening for glycosuria. Women with severe glycosuria (>250 mg/dL, 2+) on 2 determinations during the first 2 trimesters had a 21% chance (PV+) of being diagnosed as having gestational diabetes.

The author concluded that urine testing for glucose was a poor screening test and was not worthwhile after the 28-week blood glucose challenge. He believed urine testing during the first 2 trimesters was indicated to early identify those 3.8% of women with severe (2+) glycosuria (sensitivity 18%, PV–96%). However, the incidence of glycosuria was not increased in those women with gestational diabetes when compared with those with normal glucose screening values.

Gribble: No evidence supports improved outcomes from earlier identification of gestational diabetes

Gribble et al retrospectively examined 2745 charts of women at low risk for gestational diabetes in their first 2 trimesters of pregnancy.⁵ Two urine dipstick screening determinations positive for glycosuria (≥250 mg/dL) during the first 2 trimesters before a blood glucose screening test were 7% sensitive and 98% specific with a PV–of 97% and a PV+ of 13% in a population with a prevalence of gestational diabetes of 3.1%.⁵

Less than 1% had glycosuria on their first prenatal visit and were excluded from the study. Only 7% of women (6/85) who were subsequently diagnosed with gestational diabetes had glycosuria during the first 2 trimesters of their pregnancy. There was no statistically significant association (P<.05) between glycosuria and maternal body mass index, age, history, multiparity, or birth weight of an infant greater than 4 kg. Many of these are considered risk factors for gestational diabetes. Over 8% of women with a normal 1-hour screen had glycosuria in the third trimester. Requiring 2 positive urine tests and analyzing data collected before the third trimester lowered sensitivity and the PV+.

The authors recommended continuing glycosuria testing in the first two trimesters and then stop testing after the blood screen for gestational diabetes at 24 to 28 weeks although they noted that there was no evidence to support an improved pregnancy outcome because of earlier identification in gestational diabetes.

Hooper and Buhling: Urine glucose screening should be abandoned

In a retrospective study by Hooper of 610 patients who did not have glycosuria at the first prenatal visit, I calculated a sensitivity of 36%, specificity of 98%, a NPV of 99%, and a PPV of 27% using a single glycosuria value of ≥100 mg/dL in a population with a prevalence of gestational diabetes of 1.8%.⁶ The author advised that urine screening for gestational diabetes and preeclampsia be abandoned.

In a prospective German study, 1001 women were followed throughout their pregnancy.⁷ Glycosuria was detected in 8.2% of patients. Twenty-seven percent (267/1001) had an abnormal 50-g (>140 mg/dL) glucose screening test result, 178 (67% of them) completed a 3-hour 75-g glucose diagnostic test and 37 (4.1%) had gestational diabetes.

Of the 729 patients with a normal 50-g screening test, 52 (7%) had glycosuria while of the 37 with gestational diabetes, 4 (11%) had glycosuria. Sensitivity was 11% with a PV–of 95%. The 50-gram glucose screening test was done at 33.8±3 weeks gestation, later than the 28 weeks recommended in this country. Also the cutoff values for the diagnosis of gestational diabetes were lower than those of the American Diabetes Association. Both changes would increase the incidence of gestational diabetes and the sensitivity of urine glucose screening. The authors recommended against screening for glycosuria.

Summary of the studies

Three of the 4 studies most likely overestimate the sensitivity of glycosuria for predicting gestational diabetes. All but Gribble et al included urine testing results collected in the third trimester, after the gold standard oral glucose screening test and diagnostic test were completed. Furthermore, most urine tests were probably done in the third trimester when prenatal visits occur more frequently and when glycosuria is more prevalent.⁷ Both of these factors would tend to falsely elevate the sensitivity of testing for glycosuria in the first and second trimesters, when it is theoretically most useful. Gribble et al reported that including third-trimester data did not change the predictive values of glycosuria for gestational diabetes; the other investigators did not.

TABLE 1
Accuracy of glycosuria for predicting gestational diabetes mellitus

Recommendations from professional societies

The American Diabetes Association recommends blood glucose testing as soon as possible in high-risk women and routinely at 24 to 28 weeks gestation in those at lower risk.⁸ The American College of Obstetricians and Gynecologists (ACOG) does not address urine testing for glucose.⁹ The Institute for Clinical Systems Improvement (ICSI) considers urine dipsticks for glycosuria unreliable.¹⁰

Abandoning all but the initial urinalysis may miss a few women with true but unrecognized diabetes mellitus. None of the studies presented above address this problem although screening for diabetes mellitus using urine test strips is not an ideal screening test, identifying only between 30% and 59% of a predominately middle-aged nonpregnant group.¹¹

There is no evidence that testing for gestational diabetes before 28 weeks, as might be prompted by urine testing, changes pregnancy outcome. Screening for gestational diabetes by glycosuria is not effective with low sensitivities and low positive predictive values. False-positive tests outnumber true positives 11:1, leading to unnecessary further testing. Based on the information available, it appears safe to abandon routine urine testing for glucose at every prenatal visit. This recommendation stands regardless of the debate over the value of screening for gestational diabetes by 50-g glucose challenge followed by an OGTT if indicated.¹²

Proteinuria as a predictor for preeclampsia

Proteinuria in pregnancy is common. One study of 913 women reported that 3.8% of them had proteinuria by automated dipstick testing on their first antenatal visit and 40.8% had dipstick-positive (≥1+) proteinuria at least once during the course of their pregnancy.¹³ In another study of 3122 otherwise healthy women with a single gestation, 9.8% of the women had at least 1 episode of dipstick proteinuria ≥30 mg/dL (≥1+).⁵

Detection of proteinuria in hospitalized hypertensive pregnant women by visual reading of dipsticks, as is the usual office practice, has a high false-positive rate for true proteinuria (≥300 mg/L) with a PV+ (true positives/true plus false positives) of 24% for 1+, 53% for 2+, and 93% for 3+ or 4+.¹⁴ Another study reported a PV+ of 38% for ≥1+ proteinuria.¹⁵ A recent literature review concluded that the accuracy of 1+ proteinuria in pregnant women by dipstick was “poor and therefore of limited usefulness.”¹⁶ In a busy office with a number of healthy nonhypertensive women, the false-positive rate is high due to contamination with vaginal secretions, previous exercise, high specific gravity of urine, or other benign causes.^17-19 In contrast to the high false-positive rates noted in the previous studies, Meyer et al reported a negative predictive value of only 34% for trace or negative proteinuria in hospitalized women with hypertension in pregnancy.²⁰ Proteinuria detected by dipstick using visual or automated testing alone is a poor indicator for true proteinuria although the automated method is the more accurate of the 2.¹⁴ When the measurement of proteinuria is indicated for the early identification of preeclampsia, then a random protein:creatinine ratio is a better test choice.^14,15,21

Three studies have addressed the question: Is proteinuria an accurate predictor for preeclampsia?^6,15,16 Preeclampsia is defined as an elevated blood pressure with either proteinuria or edema or both.¹⁵

In a prospective observational study carried out in Australia, 866 non-hypertensive women were tested using an automated dipstick method for proteinuria on their first prenatal visit and 35 were ≥1+ positive.¹³ Twenty-five (71%) of these women had proteinuria detected during subsequent visits, and 2 (6%) of them developed preeclampsia. Of the 833 women who did not have proteinuria on the first visit, 316 had it on sub-sequent dipstick testing, and 15 of these women developed preeclampsia. Of the 512 who never had proteinuria, 9 developed preeclampsia (sensitivity=63%, PV–=98%). Proteinuria at the first visit may be a risk factor for subsequent preeclampsia (relative risk=2.2; 95% CI, 0.49–9.6]). Of the 8 women who developed proteinuria before hypertension developed, 5 could be considered at high risk: 2 had proteinuria at their first prenatal visit, 2 had multiple gestations, and 1 had a history of preeclampsia. Pregnancy outcomes were similar in the proteinuria and no proteinuria groups. The authors recommended discontinuing urine protein testing except in high-risk women (TABLE 2).

A retrospective study of 3104 low-risk American women which excluded those at high risk (multiple gestations, diabetes mellitus, preexisting hypertension, renal disease, or ≥30 mg/dL [1+] proteinuria at the first prenatal visit) found routine visually evaluated dipstick determination for proteinuria of no value in the prediction of preeclampsia.²² In this study for the 6.1% of woman who had a blood pressure of greater than 140/90 mm Hg, a weight gain of 3 pounds a week or more, or greater than 1+ edema, testing for proteinuria was considered to be for diagnostic reasons. When the remaining 2802 patients were evaluated throughout their pregnancy, 90.3% had no proteinuria, 7.6 % were 1+, and 2.2% were ≥2+. The sensitivity and PV+ of proteinuria for preeclampsia in routine patients were 5% and 96% respectively.

The presence of proteinuria was increased in younger women and those with a greater prepregnant body mass index but not with pregnancy-associated hypertension—preeclampsia, fetal distress, abruption, low birth weight, prematurity, stillbirth, or Apgar scores less than 7 at 5 minutes. The authors concluded that there is no evidence supporting routine urine dipstick protein determinations during uncomplicated prenatal visits.

In another retrospective study of 610 women, 18 % had ≥1+ proteinuria during at least one prenatal visit and 17 (3%) developed preeclampsia.⁶ Three women with preeclampsia (17%) developed proteinuria before hypertension. But the timing of the appearance of proteinuria was not otherwise specified, and it may have been remote from the hypertension. The sensitivity of proteinuria detected prior to the onset of hypertension for preeclampsia was 71% with a PV–of 99%. The author advised against routine dipstick testing.

TABLE 2
Accuracy of proteinuria for predicting preeclampsia

Recommendations and practices of others

Routine testing at antenatal visits for proteinuria is not helpful in predicting preeclampsia and should be targeted at women with an increased blood pressure or acute weight gain. ACOG advises that there is no reliable predictive test for preeclampsia.²³ The US Preventive Services Task Force advises urine testing for protein only after abnormalities in blood pressure appear.²⁴ The Canadian Task Force on the Periodic Health Examination²⁵ and other groups in Australia²⁶ advise against testing, as does a standard textbook of obstetrics.²⁷ ICSI suggests that prenatal care would be improved by discontinuing routine urine dipstick testing.¹⁰

Most groups support further evaluation of proteinuria^21,26,30 or glycosuria²⁶ found on the initial urinalysis at the first prenatal visit although there is little evidence to support this course of action.³¹ Based on the results of these studies and the recommendations of other groups, it is reasonable to reserve urine protein testing (using a more accurate method than a dipstick) for women with an elevated blood pressure.

Acknowledgments

The author wishes to thank Thomas L. Mead and Cora Damon for assisting in the library research of the topic and Colleen Flewelling for technical assistance.

CORRESPONDING AUTHOR
William A. Alto, MD, MPH, 4 Sheridan Drive, Fairfield, ME 04937. E-mail: [email protected]

Even when depression is properly diagnosed and treatment is prescribed, the rate of patient adherence to regimens can drop to as low as 33% within the first 3 months of therapy¹—far short of the universally recommended 4 to 9 months of treatment (see Minimum duration of treatment). The rate is even lower when lifestyle and other more behaviorally demanding regimens are instituted.⁹

This study demonstrated that use of a management flow sheet, in conjunction with suitable instructions to physicians and education of patients, overcame the usual causes of discontinuance and enabled far more patients to adhere to a prescribed medical regimen than is reported by other current research, ultimately alleviating depressive symptoms regardless of cause.

Methods

Setting

The study was conducted during 2003 to 2004 in a private suburban/urban family medicine group in the Midwestern United States. Fifteen family physicians practice in the group, which cares for about 55,000 patients, most of whom are insured.

Subjects

One-hundred three patients at the clinic were newly diagnosed with varying degrees of depression by 3 doctors in the practice. All were included in the study.

Diagnoses were confirmed by patient history, physical examination, interview, and responses to a 9-item diagnostic survey (Patient Health Questionnaire [PHQ-9]—APPENDIX 1, available online at www.jfponline.com, and in our February 2003 issue [J Fam Pract 2003; 52:126]). The survey has a sensitivity of 73% and a specificity of 98% when compared with a Structured Clinical Interview administered by a mental health clinician.^10,11

No exclusion criteria were applied. Subjects were included regardless of age, gender, race, severity of depression, associated medical conditions, or insurance status. No patients refused to participate. However, of the 103 enrolled patients, 1 was later imprisoned, 2 died, and 3 transferred from the practice. Of the remaining 97, 36 were identified too late in the study to meet the 9-month protocol at the time of final analysis. Therefore, though their comorbidity and depression level data are included in this research, final conclusions relative to “measurement of adherence” were not.

The database for this study, therefore, is 97 subjects for whom data were secured, and 61 for whom adherence or nonadherence was measured. The practice continues to monitor all enrolled patients, and other enrollees for the purposes described in this project.

Experimental design

The point of this study was to determine whether a flow sheet (FIGURE 1) incorporating a checklist for comorbid disorders, medication reference guide, and a major depression reference guide (FIGURE 2), combined with patient education, would improve patient adherence with a pharmacologic regimen and reduce or eliminate depression symptoms without a subsequent relapse.

Doctors in the practice were informed of the project and educated by the author regarding its purpose, protocol, intended outcomes, and methodology.

Though a substantial number of illnesses could be considered comorbid with depression, it would be unrealistic and unwieldy to include them all. Nine conditions were included as sample characteristics, for 2 reasons. First, experience has shown that these particular comorbidities are prevalent among patients presenting to the family physicians. Second, a set of symptoms associated with each of these selected comorbidities often overlaps those of depression, and may therefore cloud the final diagnosis. The prevalence of diagnosed and documented comorbidities, which may interfere with a diagnosis of depression, is summarized in TABLE 1.

All patients who were thought to be depressed or who exhibited depressive symptoms were asked to complete a PHQ-9. None declined. All were educated by the attending physician during the initial office appointment, and given informational material to explain the disease and the necessity of adhering to a prescribed regimen for a period of no less than 9 months. A flow sheet, containing information relative to office calls, follow-up PHQ-9s, and other summaries of medication, comorbidities, and treatment regimens was inserted into their respective charts.

Following the initial appointment, patients were encouraged to schedule other visits at 4 weeks, within 4 to 9 months, and at one year. During these follow-up appointments, physicians stressed the need for continuing medication for no less than 9 months. Every patient who did not return for a follow-up appointment after 6 months, as indicated by a systematic chart review, was contacted by phone by a registered nurse employed by the practice. All of these patients subsequently scheduled an appointment, confirmed they were still following the regimen, or informed the nurse that they had discontinued their medication(s).

TABLE 1
Comorbidity summary of depression patients (n=91)

Data collection and analysis

Periodically throughout the study period of 1½ years, patient charts were audited to collect data on demographics and comorbidities, to quantify the number of patients adhering to prescribed medications for a minimum 9 months, and to compile results of the 2 PHQ-9 surveys. These data were then contrasted with existing clinical research data to demonstrate that the procedure significantly improved patient adherence to a prescribed regimen.

Results

Data from this study indicate that 61 of the 103 patients enrolled in the study completed at least 9 months’ follow-up. Based on patients’ verbal input, a second PHQ-9, notations in charts, subsequent appointments, phone follow-ups, and chart medication reviews, 40 of these 61 patients (66%) adhered to prescribed daily drug therapy for depression for at least 9 months—double the 33% adherence rate described in clinical literature.¹

Seventy-one (78%) of the patients followed in this study had 1 or more significant comorbid illnesses; 54 (76%) had 2 or more. The most common comorbidities included anxiety, migraine, and irritable bowel syndrome, with rates of 54%, 48%, and 32%, respectively (TABLE 1).

TABLE 2 summarizes the comparison of initial and follow-up PHQ-9 data after medication was begun and after an interval of at least 4 weeks. Based on the initial PHQ-9 score, 80% of patients presented with moderate, moderate-severe, or severe depressive symptoms. The average initial PHQ-9 score was 14.2±5.1 (SD).

On follow-up, only 40% of patients were documented to have the same range of severity of symptoms. The average follow-up PHQ-9 score was 8.3±6.2 (SD) (P<.001) vs initial score. Thirty-six of these 40 patients (90%) remained on their initially prescribed medications.

TABLE 2
Distribution of PHQ-9 scores

Discussion

Patients discontinue their medications many reasons (TABLE 3).^1,6,13-16 These obstacles to drug therapy often result in therapeutic failure. Given we now have better-tolerated medications, nonadherence may result more from poor patient commitment to treatment than from adverse drug effects.¹⁴

Communicate with patients. The literature also provides insight into persuasions likely to increase patient compliance. TABLE 4 lists indicative factors.^1,6,9,17

Explicit communication with patients regarding the expected duration of antidepressant therapy may reduce premature discontinuation of medication use.¹⁷

Better communication between patients and physicians about antidepressant treatment, both before and during treatment, may promote adherence.¹

Another study showed that a strong alliance between physicians and patients that involves discussions about adverse drug effects may alleviate patients’ concerns and help them continue treatment.¹

Moreover, intolerance to one antidepressant is not necessarily indicative of intolerance to another, even within the same drug class. Therefore, patients who respond poorly to one drug or who experience adverse effects may benefit by switching to another antidepressant medication.¹ This medication shift, however, necessitates good communication between patients and clinicians about treatment experiences.¹

Adherence can be improved. This study showed that patient adherence to a prescribed medical regimen significantly improved over the life of the study. The 9-month medication adherence rate of 66% dramatically exceeds the 33% rate chronicled in the literature. Over time, the use of the process outlined was associated with significant reductions in the severity of depressive symptoms.

A few caveats. One limitation of this study is its small number of subjects, and the deficiency of data for subjects who had died, transferred out of the practice, or were otherwise lost to contact.

The lack of a control group is also acknowledged. However, comparisons were made between this study and the adherence rates documented in other studies.

Though the PHQ-9 diagnostic tool is reliable and valid, it is self-administered. Likewise, data collection—ie, whether they discontinued medications, and, if so, for what reason—depended on patients’ responses.

Even though the project stressed patient adherence, the use of the flow sheet may very well have contributed to increased physician awareness and physician education, which therefore, in itself, may have resulted in improved patient compliance.

The results of this project can be generalized only to practices similar to its setting. Other practices with different methods or types of information systems may not achieve the same results when using a flow sheet. Further research in a wider area using a larger number of subjects with broader demographics is necessary to corroborate these findings.

TABLE 3
Reasons for discontinuing medications

TABLE 4
Factors conducive to regimen compliance

CORRESPONDING AUTHOR
Gary Ruoff, MD, Westside Family Medical Center, 6565 West Main Street, Kalamazooo, MI 49009. E-mail: [email protected]

Even when depression is properly diagnosed and treatment is prescribed, the rate of patient adherence to regimens can drop to as low as 33% within the first 3 months of therapy¹—far short of the universally recommended 4 to 9 months of treatment (see Minimum duration of treatment). The rate is even lower when lifestyle and other more behaviorally demanding regimens are instituted.⁹

This study demonstrated that use of a management flow sheet, in conjunction with suitable instructions to physicians and education of patients, overcame the usual causes of discontinuance and enabled far more patients to adhere to a prescribed medical regimen than is reported by other current research, ultimately alleviating depressive symptoms regardless of cause.

Methods

Setting

The study was conducted during 2003 to 2004 in a private suburban/urban family medicine group in the Midwestern United States. Fifteen family physicians practice in the group, which cares for about 55,000 patients, most of whom are insured.

Subjects

One-hundred three patients at the clinic were newly diagnosed with varying degrees of depression by 3 doctors in the practice. All were included in the study.

Diagnoses were confirmed by patient history, physical examination, interview, and responses to a 9-item diagnostic survey (Patient Health Questionnaire [PHQ-9]—APPENDIX 1, available online at www.jfponline.com, and in our February 2003 issue [J Fam Pract 2003; 52:126]). The survey has a sensitivity of 73% and a specificity of 98% when compared with a Structured Clinical Interview administered by a mental health clinician.^10,11

No exclusion criteria were applied. Subjects were included regardless of age, gender, race, severity of depression, associated medical conditions, or insurance status. No patients refused to participate. However, of the 103 enrolled patients, 1 was later imprisoned, 2 died, and 3 transferred from the practice. Of the remaining 97, 36 were identified too late in the study to meet the 9-month protocol at the time of final analysis. Therefore, though their comorbidity and depression level data are included in this research, final conclusions relative to “measurement of adherence” were not.

The database for this study, therefore, is 97 subjects for whom data were secured, and 61 for whom adherence or nonadherence was measured. The practice continues to monitor all enrolled patients, and other enrollees for the purposes described in this project.

Experimental design

The point of this study was to determine whether a flow sheet (FIGURE 1) incorporating a checklist for comorbid disorders, medication reference guide, and a major depression reference guide (FIGURE 2), combined with patient education, would improve patient adherence with a pharmacologic regimen and reduce or eliminate depression symptoms without a subsequent relapse.

Doctors in the practice were informed of the project and educated by the author regarding its purpose, protocol, intended outcomes, and methodology.

Though a substantial number of illnesses could be considered comorbid with depression, it would be unrealistic and unwieldy to include them all. Nine conditions were included as sample characteristics, for 2 reasons. First, experience has shown that these particular comorbidities are prevalent among patients presenting to the family physicians. Second, a set of symptoms associated with each of these selected comorbidities often overlaps those of depression, and may therefore cloud the final diagnosis. The prevalence of diagnosed and documented comorbidities, which may interfere with a diagnosis of depression, is summarized in TABLE 1.

All patients who were thought to be depressed or who exhibited depressive symptoms were asked to complete a PHQ-9. None declined. All were educated by the attending physician during the initial office appointment, and given informational material to explain the disease and the necessity of adhering to a prescribed regimen for a period of no less than 9 months. A flow sheet, containing information relative to office calls, follow-up PHQ-9s, and other summaries of medication, comorbidities, and treatment regimens was inserted into their respective charts.

Following the initial appointment, patients were encouraged to schedule other visits at 4 weeks, within 4 to 9 months, and at one year. During these follow-up appointments, physicians stressed the need for continuing medication for no less than 9 months. Every patient who did not return for a follow-up appointment after 6 months, as indicated by a systematic chart review, was contacted by phone by a registered nurse employed by the practice. All of these patients subsequently scheduled an appointment, confirmed they were still following the regimen, or informed the nurse that they had discontinued their medication(s).

TABLE 1
Comorbidity summary of depression patients (n=91)

Data collection and analysis

Periodically throughout the study period of 1½ years, patient charts were audited to collect data on demographics and comorbidities, to quantify the number of patients adhering to prescribed medications for a minimum 9 months, and to compile results of the 2 PHQ-9 surveys. These data were then contrasted with existing clinical research data to demonstrate that the procedure significantly improved patient adherence to a prescribed regimen.

Results

Data from this study indicate that 61 of the 103 patients enrolled in the study completed at least 9 months’ follow-up. Based on patients’ verbal input, a second PHQ-9, notations in charts, subsequent appointments, phone follow-ups, and chart medication reviews, 40 of these 61 patients (66%) adhered to prescribed daily drug therapy for depression for at least 9 months—double the 33% adherence rate described in clinical literature.¹

Seventy-one (78%) of the patients followed in this study had 1 or more significant comorbid illnesses; 54 (76%) had 2 or more. The most common comorbidities included anxiety, migraine, and irritable bowel syndrome, with rates of 54%, 48%, and 32%, respectively (TABLE 1).

TABLE 2 summarizes the comparison of initial and follow-up PHQ-9 data after medication was begun and after an interval of at least 4 weeks. Based on the initial PHQ-9 score, 80% of patients presented with moderate, moderate-severe, or severe depressive symptoms. The average initial PHQ-9 score was 14.2±5.1 (SD).

On follow-up, only 40% of patients were documented to have the same range of severity of symptoms. The average follow-up PHQ-9 score was 8.3±6.2 (SD) (P<.001) vs initial score. Thirty-six of these 40 patients (90%) remained on their initially prescribed medications.

TABLE 2
Distribution of PHQ-9 scores

Discussion

Patients discontinue their medications many reasons (TABLE 3).^1,6,13-16 These obstacles to drug therapy often result in therapeutic failure. Given we now have better-tolerated medications, nonadherence may result more from poor patient commitment to treatment than from adverse drug effects.¹⁴

Communicate with patients. The literature also provides insight into persuasions likely to increase patient compliance. TABLE 4 lists indicative factors.^1,6,9,17

Explicit communication with patients regarding the expected duration of antidepressant therapy may reduce premature discontinuation of medication use.¹⁷

Better communication between patients and physicians about antidepressant treatment, both before and during treatment, may promote adherence.¹

Another study showed that a strong alliance between physicians and patients that involves discussions about adverse drug effects may alleviate patients’ concerns and help them continue treatment.¹

Moreover, intolerance to one antidepressant is not necessarily indicative of intolerance to another, even within the same drug class. Therefore, patients who respond poorly to one drug or who experience adverse effects may benefit by switching to another antidepressant medication.¹ This medication shift, however, necessitates good communication between patients and clinicians about treatment experiences.¹

Adherence can be improved. This study showed that patient adherence to a prescribed medical regimen significantly improved over the life of the study. The 9-month medication adherence rate of 66% dramatically exceeds the 33% rate chronicled in the literature. Over time, the use of the process outlined was associated with significant reductions in the severity of depressive symptoms.

A few caveats. One limitation of this study is its small number of subjects, and the deficiency of data for subjects who had died, transferred out of the practice, or were otherwise lost to contact.

The lack of a control group is also acknowledged. However, comparisons were made between this study and the adherence rates documented in other studies.

Though the PHQ-9 diagnostic tool is reliable and valid, it is self-administered. Likewise, data collection—ie, whether they discontinued medications, and, if so, for what reason—depended on patients’ responses.

Even though the project stressed patient adherence, the use of the flow sheet may very well have contributed to increased physician awareness and physician education, which therefore, in itself, may have resulted in improved patient compliance.

The results of this project can be generalized only to practices similar to its setting. Other practices with different methods or types of information systems may not achieve the same results when using a flow sheet. Further research in a wider area using a larger number of subjects with broader demographics is necessary to corroborate these findings.

TABLE 3
Reasons for discontinuing medications

TABLE 4
Factors conducive to regimen compliance

CORRESPONDING AUTHOR
Gary Ruoff, MD, Westside Family Medical Center, 6565 West Main Street, Kalamazooo, MI 49009. E-mail: [email protected]

Even when depression is properly diagnosed and treatment is prescribed, the rate of patient adherence to regimens can drop to as low as 33% within the first 3 months of therapy¹—far short of the universally recommended 4 to 9 months of treatment (see Minimum duration of treatment). The rate is even lower when lifestyle and other more behaviorally demanding regimens are instituted.⁹

This study demonstrated that use of a management flow sheet, in conjunction with suitable instructions to physicians and education of patients, overcame the usual causes of discontinuance and enabled far more patients to adhere to a prescribed medical regimen than is reported by other current research, ultimately alleviating depressive symptoms regardless of cause.

Methods

Setting

The study was conducted during 2003 to 2004 in a private suburban/urban family medicine group in the Midwestern United States. Fifteen family physicians practice in the group, which cares for about 55,000 patients, most of whom are insured.

Subjects

One-hundred three patients at the clinic were newly diagnosed with varying degrees of depression by 3 doctors in the practice. All were included in the study.

Diagnoses were confirmed by patient history, physical examination, interview, and responses to a 9-item diagnostic survey (Patient Health Questionnaire [PHQ-9]—APPENDIX 1, available online at www.jfponline.com, and in our February 2003 issue [J Fam Pract 2003; 52:126]). The survey has a sensitivity of 73% and a specificity of 98% when compared with a Structured Clinical Interview administered by a mental health clinician.^10,11

No exclusion criteria were applied. Subjects were included regardless of age, gender, race, severity of depression, associated medical conditions, or insurance status. No patients refused to participate. However, of the 103 enrolled patients, 1 was later imprisoned, 2 died, and 3 transferred from the practice. Of the remaining 97, 36 were identified too late in the study to meet the 9-month protocol at the time of final analysis. Therefore, though their comorbidity and depression level data are included in this research, final conclusions relative to “measurement of adherence” were not.

The database for this study, therefore, is 97 subjects for whom data were secured, and 61 for whom adherence or nonadherence was measured. The practice continues to monitor all enrolled patients, and other enrollees for the purposes described in this project.

Experimental design

The point of this study was to determine whether a flow sheet (FIGURE 1) incorporating a checklist for comorbid disorders, medication reference guide, and a major depression reference guide (FIGURE 2), combined with patient education, would improve patient adherence with a pharmacologic regimen and reduce or eliminate depression symptoms without a subsequent relapse.

Doctors in the practice were informed of the project and educated by the author regarding its purpose, protocol, intended outcomes, and methodology.

Though a substantial number of illnesses could be considered comorbid with depression, it would be unrealistic and unwieldy to include them all. Nine conditions were included as sample characteristics, for 2 reasons. First, experience has shown that these particular comorbidities are prevalent among patients presenting to the family physicians. Second, a set of symptoms associated with each of these selected comorbidities often overlaps those of depression, and may therefore cloud the final diagnosis. The prevalence of diagnosed and documented comorbidities, which may interfere with a diagnosis of depression, is summarized in TABLE 1.

All patients who were thought to be depressed or who exhibited depressive symptoms were asked to complete a PHQ-9. None declined. All were educated by the attending physician during the initial office appointment, and given informational material to explain the disease and the necessity of adhering to a prescribed regimen for a period of no less than 9 months. A flow sheet, containing information relative to office calls, follow-up PHQ-9s, and other summaries of medication, comorbidities, and treatment regimens was inserted into their respective charts.

Following the initial appointment, patients were encouraged to schedule other visits at 4 weeks, within 4 to 9 months, and at one year. During these follow-up appointments, physicians stressed the need for continuing medication for no less than 9 months. Every patient who did not return for a follow-up appointment after 6 months, as indicated by a systematic chart review, was contacted by phone by a registered nurse employed by the practice. All of these patients subsequently scheduled an appointment, confirmed they were still following the regimen, or informed the nurse that they had discontinued their medication(s).

TABLE 1
Comorbidity summary of depression patients (n=91)

Data collection and analysis

Periodically throughout the study period of 1½ years, patient charts were audited to collect data on demographics and comorbidities, to quantify the number of patients adhering to prescribed medications for a minimum 9 months, and to compile results of the 2 PHQ-9 surveys. These data were then contrasted with existing clinical research data to demonstrate that the procedure significantly improved patient adherence to a prescribed regimen.

Results

Data from this study indicate that 61 of the 103 patients enrolled in the study completed at least 9 months’ follow-up. Based on patients’ verbal input, a second PHQ-9, notations in charts, subsequent appointments, phone follow-ups, and chart medication reviews, 40 of these 61 patients (66%) adhered to prescribed daily drug therapy for depression for at least 9 months—double the 33% adherence rate described in clinical literature.¹

Seventy-one (78%) of the patients followed in this study had 1 or more significant comorbid illnesses; 54 (76%) had 2 or more. The most common comorbidities included anxiety, migraine, and irritable bowel syndrome, with rates of 54%, 48%, and 32%, respectively (TABLE 1).

TABLE 2 summarizes the comparison of initial and follow-up PHQ-9 data after medication was begun and after an interval of at least 4 weeks. Based on the initial PHQ-9 score, 80% of patients presented with moderate, moderate-severe, or severe depressive symptoms. The average initial PHQ-9 score was 14.2±5.1 (SD).

On follow-up, only 40% of patients were documented to have the same range of severity of symptoms. The average follow-up PHQ-9 score was 8.3±6.2 (SD) (P<.001) vs initial score. Thirty-six of these 40 patients (90%) remained on their initially prescribed medications.

TABLE 2
Distribution of PHQ-9 scores

Discussion

Patients discontinue their medications many reasons (TABLE 3).^1,6,13-16 These obstacles to drug therapy often result in therapeutic failure. Given we now have better-tolerated medications, nonadherence may result more from poor patient commitment to treatment than from adverse drug effects.¹⁴

Communicate with patients. The literature also provides insight into persuasions likely to increase patient compliance. TABLE 4 lists indicative factors.^1,6,9,17

Explicit communication with patients regarding the expected duration of antidepressant therapy may reduce premature discontinuation of medication use.¹⁷

Better communication between patients and physicians about antidepressant treatment, both before and during treatment, may promote adherence.¹

Another study showed that a strong alliance between physicians and patients that involves discussions about adverse drug effects may alleviate patients’ concerns and help them continue treatment.¹

Moreover, intolerance to one antidepressant is not necessarily indicative of intolerance to another, even within the same drug class. Therefore, patients who respond poorly to one drug or who experience adverse effects may benefit by switching to another antidepressant medication.¹ This medication shift, however, necessitates good communication between patients and clinicians about treatment experiences.¹

Adherence can be improved. This study showed that patient adherence to a prescribed medical regimen significantly improved over the life of the study. The 9-month medication adherence rate of 66% dramatically exceeds the 33% rate chronicled in the literature. Over time, the use of the process outlined was associated with significant reductions in the severity of depressive symptoms.

A few caveats. One limitation of this study is its small number of subjects, and the deficiency of data for subjects who had died, transferred out of the practice, or were otherwise lost to contact.

The lack of a control group is also acknowledged. However, comparisons were made between this study and the adherence rates documented in other studies.

Though the PHQ-9 diagnostic tool is reliable and valid, it is self-administered. Likewise, data collection—ie, whether they discontinued medications, and, if so, for what reason—depended on patients’ responses.

Even though the project stressed patient adherence, the use of the flow sheet may very well have contributed to increased physician awareness and physician education, which therefore, in itself, may have resulted in improved patient compliance.

The results of this project can be generalized only to practices similar to its setting. Other practices with different methods or types of information systems may not achieve the same results when using a flow sheet. Further research in a wider area using a larger number of subjects with broader demographics is necessary to corroborate these findings.

TABLE 3
Reasons for discontinuing medications

TABLE 4
Factors conducive to regimen compliance

CORRESPONDING AUTHOR
Gary Ruoff, MD, Westside Family Medical Center, 6565 West Main Street, Kalamazooo, MI 49009. E-mail: [email protected]

Despite its potential importance, the problem of wasted medication has been studied little. Some previous research has concerned inpatient hospital and nursing facility drug discards.^1-5 However, a pharmacy-based initiative for collecting wasted medications in Alberta, Canada, accumulated 204 tons of medicines over a 7-year period, suggesting the need for further research on outpatient drug waste.⁶

The specific aim of our study was to assess the occurrence, costs, and reasons for medication waste in a population of older adults by doing in-home surveys and counts of leftover medications. We also sought to determine why patients do not always finish their full prescriptions.

Methods

Our study, conducted from May 1999 to November 1999, was a cross-sectional survey describing medication use and nonuse in older adults in a retirement community. To minimize recall bias, researchers both used questionnaires and visited participants’ homes to sort pills according to active use or waste. Any medication prescribed within the past year that the study participant did not intend to use before its expiration date was considered wasted. If subjects recalled medications that had been thrown away, these were recorded. Questionnaire data included name, date of birth, sex, length of time in the residence, current medications, type and amount of medication discarded in the past year, and reasons for nonuse of medications.

All community residents in the study population had full prescription drug benefits without co-payments. In general, residents of this community have relatively high levels of education (78% had a bachelor’s degree or higher) and yearly income (only 4% were receiving less than $20,000).⁷ Primary care was largely obtained from an academic family physician and an internist, both with certificates of added qualification in geriatrics, and 2 geriatric nurse practitioners. To promote full disclosure of waste, data were kept confidential from the providers who cared for the study subjects. As a result, no clinical consequences of waste could be determined. The criteria for entry were age 65 years or older, voluntary response to study recruitment advertisements, residence in the facility for at least 1 year, and contact with a licensed health care provider within the past year. No volunteer subject was found ineligible. The Committee for Protection of Human Subjects at Dartmouth Medical School approved the study protocol.

The researchers coded medications by pharmaceutical class and calculated totals for each drug, including the costs of current medication use and total annual costs due to waste, using 1999 Red Book⁸ median wholesale drug cost estimates for a 3 months’ supply when exact prescription quantities were unknown.

Results

A total of 73 subjects received in-home pharmacy evaluations and completed questionnaires. Of these, 49 were women (67%), and 24 were men (33%). All were white, and all were older than 65 years. The mean age was 81.2 ± 6.0 (standard deviation [SD]) years. The mean number of years in residence was 5.4 ± 2.5 (SD).

The sum of all costs of wasted medication was $2011 in the study group (n=66). Mean per-person annual cost of wasted medication was $30.47 (range = $0-$131.56) based on the 66 subjects for whom complete data on pill counts were available. Pill counts were missing or incomplete on 7 questionnaires, which were not counted. Median annual waste was only $12.32, because 32 of the subjects (48%) wasted no medications. A total of 2078 wasted pills were found for the 66 subjects, yielding a mean of 31.5 pills wasted per subject (range = 0 to 208).

Mean waste represented 2.3% of total annual medication costs, which were $1302.78 per subject (interquartile range = $584.61-$1773.90). Increasing age was correlated with a higher number of pills wasted (r=0.35; P=.03) and a higher total cost of waste (r=0.20; P=.10). Total waste did not represent a fixed percentage of total annual medication costs. As yearly medication costs rose, waste as a percentage of yearly costs (ie, inefficient medication use) decreased significantly (r = -0.32; P=.02).

The most frequently wasted medication classes were antibiotics, benzodiazepines, and antihypertensives Table 1. Many of the medications listed as frequently wasted are taken episodically rather than in a stable daily pattern. Table 1 also shows that the most frequently wasted medications (eg, antibiotics) are not necessarily the most prescribed or the costliest.

Table 1 also shows the total annual relative costs of wasted medication by pharmaceutical class. Benzodiazepines, antidepressants, and antihypertensive medications combined accounted for a third of the total annual costs due to waste. The reasons for waste of medications and the relative contribution of each to total waste are presented in Table 2. The perception by subjects that a medical condition had resolved or that a medication was ineffective accounted for more than half of the cost due to waste. Physician and geriatric nurse practitioner perspectives are not captured by these data.

Discussion

On the basis of comprehensive home assessments, our study provides an estimate of wasted medication and the reasons for it in an outpatient population of older adults.

Most waste derived from 2 factors: the resolution of the condition for which the medication was prescribed and perceived ineffectiveness of a medication for its purpose. Together these 2 reasons accounted for more than half of the costs. This finding implies that acute conditions are central to waste, especially when medications for such conditions have high unit costs. Further support for the importance of acute conditions was that higher yearly drug expenditures were associated with lower percentage waste. Thus, high annual drug costs reflected stable, efficient patterns of medication use.

Although it may be difficult for clinicians to estimate how many pills to dispense, efforts should be made to determine the effectiveness and tolerability of medications before prescribing full quantities. Judicious use of samples is a possible remedy for this problem. Small prescriptions requiring multiple pharmacy visits would not help, but research on optimal prescribing quantities might lend some insight. Physicians should encourage patients to finish prescribed antibiotics if tolerated and not needlessly change prescriptions when previous pills remain. Further suggestions for promoting medication compliance in older adults are available in the medical literature.⁹

Limitations

We emphasize that our pilot study was small and not necessarily generalizable, yet it has made progress in a neglected area of research. As employees of the retirement community, the physicians caring for these patients are motivated to be fiscally responsible, and for this reason the mean annual waste detected in our study may, if anything, have been a substantial underestimate. Also, some subjects may have been overly optimistic in concluding that they intended to use all of a prescription medication on an as-needed basis. The lack of drug co-payments in this population, however, may predispose to more waste.

Based on our interviews, we found it uncharacteristic of most subjects to throw any pills away, but further underestimation could have occurred because of forgotten disposal of medicines. Researchers were motivated to find waste but could find none for 48% of subjects. Those subjects seemed sure that waste was absent.

Even though we counted more than 2000 wasted pills, numbers of specific medications were small. Thus, our analysis was confined to broad pharmaceutical classes and overall reasons for waste. This limitation could be overcome in more focused studies of specific wasted medications.

Conclusions

If, as we found in our study, average medication waste of $30 per person-year represents a conservative estimate, given that there are nearly 35 million individuals older than 65 years in the United States,¹⁰ the total national costs due to medication waste would not be less than $1 billion per year. Clearly, further studies in varied populations are required to confirm our waste estimate, and more research is needed to find effective waste reduction strategies. Despite the limitations of our study, physicians should begin to take note of what happens to prescribed medicines. That may serve as the most immediate basis for waste reduction.

Related resources

The Drugs and Devices Information Line Contains links to pharmacoepidemiology resources, maintained by the Pharmacoepidemiology Program, Harvard School of Public Health. http://www.hsph.harvard.edu/Organizations/DDIL/ddilhpge.html

Topics of Pharmacoepidemiology and Pharmacoeconomics A listserv for discussions related to these areas. http://www.findmail.com/list/pharmacoepidemiology/

ISPE—International Society for Pharmacoepidemiology A non-profit international professional organization dedicated to promoting pharmacoepidemiology. http://www.pharmacoepi.org/index.htm

Pharmacoepidemiology and Drug Safety The official journal of the International Society for Pharmacoepidemiology http://www.interscience.wiley.com/jpages/1053-8569/

Acknowledgment

The author acknowledges the assistance of the following undergraduate research assistants: Allison Robbins, Barbara Jones, Eva Liu, Karen Walp, Cynthia Oberto, Amanda Cook, John Raser, Sarah Hamilton, Anjali Godambe, and Michelle Anatone. Allen Dietrich, MD, provided guidance in the planning and execution of the study. Harlan Krumholz, MD, and Jerome Kassirer, MD, of the Yale University Robert Wood Johnson Clinical Scholars Program provided critical review of the manuscript.

Despite its potential importance, the problem of wasted medication has been studied little. Some previous research has concerned inpatient hospital and nursing facility drug discards.^1-5 However, a pharmacy-based initiative for collecting wasted medications in Alberta, Canada, accumulated 204 tons of medicines over a 7-year period, suggesting the need for further research on outpatient drug waste.⁶

The specific aim of our study was to assess the occurrence, costs, and reasons for medication waste in a population of older adults by doing in-home surveys and counts of leftover medications. We also sought to determine why patients do not always finish their full prescriptions.

Methods

Our study, conducted from May 1999 to November 1999, was a cross-sectional survey describing medication use and nonuse in older adults in a retirement community. To minimize recall bias, researchers both used questionnaires and visited participants’ homes to sort pills according to active use or waste. Any medication prescribed within the past year that the study participant did not intend to use before its expiration date was considered wasted. If subjects recalled medications that had been thrown away, these were recorded. Questionnaire data included name, date of birth, sex, length of time in the residence, current medications, type and amount of medication discarded in the past year, and reasons for nonuse of medications.

All community residents in the study population had full prescription drug benefits without co-payments. In general, residents of this community have relatively high levels of education (78% had a bachelor’s degree or higher) and yearly income (only 4% were receiving less than $20,000).⁷ Primary care was largely obtained from an academic family physician and an internist, both with certificates of added qualification in geriatrics, and 2 geriatric nurse practitioners. To promote full disclosure of waste, data were kept confidential from the providers who cared for the study subjects. As a result, no clinical consequences of waste could be determined. The criteria for entry were age 65 years or older, voluntary response to study recruitment advertisements, residence in the facility for at least 1 year, and contact with a licensed health care provider within the past year. No volunteer subject was found ineligible. The Committee for Protection of Human Subjects at Dartmouth Medical School approved the study protocol.

The researchers coded medications by pharmaceutical class and calculated totals for each drug, including the costs of current medication use and total annual costs due to waste, using 1999 Red Book⁸ median wholesale drug cost estimates for a 3 months’ supply when exact prescription quantities were unknown.

Results

A total of 73 subjects received in-home pharmacy evaluations and completed questionnaires. Of these, 49 were women (67%), and 24 were men (33%). All were white, and all were older than 65 years. The mean age was 81.2 ± 6.0 (standard deviation [SD]) years. The mean number of years in residence was 5.4 ± 2.5 (SD).

The sum of all costs of wasted medication was $2011 in the study group (n=66). Mean per-person annual cost of wasted medication was $30.47 (range = $0-$131.56) based on the 66 subjects for whom complete data on pill counts were available. Pill counts were missing or incomplete on 7 questionnaires, which were not counted. Median annual waste was only $12.32, because 32 of the subjects (48%) wasted no medications. A total of 2078 wasted pills were found for the 66 subjects, yielding a mean of 31.5 pills wasted per subject (range = 0 to 208).

Mean waste represented 2.3% of total annual medication costs, which were $1302.78 per subject (interquartile range = $584.61-$1773.90). Increasing age was correlated with a higher number of pills wasted (r=0.35; P=.03) and a higher total cost of waste (r=0.20; P=.10). Total waste did not represent a fixed percentage of total annual medication costs. As yearly medication costs rose, waste as a percentage of yearly costs (ie, inefficient medication use) decreased significantly (r = -0.32; P=.02).

The most frequently wasted medication classes were antibiotics, benzodiazepines, and antihypertensives Table 1. Many of the medications listed as frequently wasted are taken episodically rather than in a stable daily pattern. Table 1 also shows that the most frequently wasted medications (eg, antibiotics) are not necessarily the most prescribed or the costliest.

Table 1 also shows the total annual relative costs of wasted medication by pharmaceutical class. Benzodiazepines, antidepressants, and antihypertensive medications combined accounted for a third of the total annual costs due to waste. The reasons for waste of medications and the relative contribution of each to total waste are presented in Table 2. The perception by subjects that a medical condition had resolved or that a medication was ineffective accounted for more than half of the cost due to waste. Physician and geriatric nurse practitioner perspectives are not captured by these data.

Discussion

On the basis of comprehensive home assessments, our study provides an estimate of wasted medication and the reasons for it in an outpatient population of older adults.

Most waste derived from 2 factors: the resolution of the condition for which the medication was prescribed and perceived ineffectiveness of a medication for its purpose. Together these 2 reasons accounted for more than half of the costs. This finding implies that acute conditions are central to waste, especially when medications for such conditions have high unit costs. Further support for the importance of acute conditions was that higher yearly drug expenditures were associated with lower percentage waste. Thus, high annual drug costs reflected stable, efficient patterns of medication use.

Although it may be difficult for clinicians to estimate how many pills to dispense, efforts should be made to determine the effectiveness and tolerability of medications before prescribing full quantities. Judicious use of samples is a possible remedy for this problem. Small prescriptions requiring multiple pharmacy visits would not help, but research on optimal prescribing quantities might lend some insight. Physicians should encourage patients to finish prescribed antibiotics if tolerated and not needlessly change prescriptions when previous pills remain. Further suggestions for promoting medication compliance in older adults are available in the medical literature.⁹

Limitations

We emphasize that our pilot study was small and not necessarily generalizable, yet it has made progress in a neglected area of research. As employees of the retirement community, the physicians caring for these patients are motivated to be fiscally responsible, and for this reason the mean annual waste detected in our study may, if anything, have been a substantial underestimate. Also, some subjects may have been overly optimistic in concluding that they intended to use all of a prescription medication on an as-needed basis. The lack of drug co-payments in this population, however, may predispose to more waste.

Based on our interviews, we found it uncharacteristic of most subjects to throw any pills away, but further underestimation could have occurred because of forgotten disposal of medicines. Researchers were motivated to find waste but could find none for 48% of subjects. Those subjects seemed sure that waste was absent.

Even though we counted more than 2000 wasted pills, numbers of specific medications were small. Thus, our analysis was confined to broad pharmaceutical classes and overall reasons for waste. This limitation could be overcome in more focused studies of specific wasted medications.

Conclusions

If, as we found in our study, average medication waste of $30 per person-year represents a conservative estimate, given that there are nearly 35 million individuals older than 65 years in the United States,¹⁰ the total national costs due to medication waste would not be less than $1 billion per year. Clearly, further studies in varied populations are required to confirm our waste estimate, and more research is needed to find effective waste reduction strategies. Despite the limitations of our study, physicians should begin to take note of what happens to prescribed medicines. That may serve as the most immediate basis for waste reduction.

Related resources

The Drugs and Devices Information Line Contains links to pharmacoepidemiology resources, maintained by the Pharmacoepidemiology Program, Harvard School of Public Health. http://www.hsph.harvard.edu/Organizations/DDIL/ddilhpge.html

Topics of Pharmacoepidemiology and Pharmacoeconomics A listserv for discussions related to these areas. http://www.findmail.com/list/pharmacoepidemiology/

ISPE—International Society for Pharmacoepidemiology A non-profit international professional organization dedicated to promoting pharmacoepidemiology. http://www.pharmacoepi.org/index.htm

Pharmacoepidemiology and Drug Safety The official journal of the International Society for Pharmacoepidemiology http://www.interscience.wiley.com/jpages/1053-8569/

Acknowledgment

The author acknowledges the assistance of the following undergraduate research assistants: Allison Robbins, Barbara Jones, Eva Liu, Karen Walp, Cynthia Oberto, Amanda Cook, John Raser, Sarah Hamilton, Anjali Godambe, and Michelle Anatone. Allen Dietrich, MD, provided guidance in the planning and execution of the study. Harlan Krumholz, MD, and Jerome Kassirer, MD, of the Yale University Robert Wood Johnson Clinical Scholars Program provided critical review of the manuscript.

Despite its potential importance, the problem of wasted medication has been studied little. Some previous research has concerned inpatient hospital and nursing facility drug discards.^1-5 However, a pharmacy-based initiative for collecting wasted medications in Alberta, Canada, accumulated 204 tons of medicines over a 7-year period, suggesting the need for further research on outpatient drug waste.⁶

The specific aim of our study was to assess the occurrence, costs, and reasons for medication waste in a population of older adults by doing in-home surveys and counts of leftover medications. We also sought to determine why patients do not always finish their full prescriptions.

Methods

Our study, conducted from May 1999 to November 1999, was a cross-sectional survey describing medication use and nonuse in older adults in a retirement community. To minimize recall bias, researchers both used questionnaires and visited participants’ homes to sort pills according to active use or waste. Any medication prescribed within the past year that the study participant did not intend to use before its expiration date was considered wasted. If subjects recalled medications that had been thrown away, these were recorded. Questionnaire data included name, date of birth, sex, length of time in the residence, current medications, type and amount of medication discarded in the past year, and reasons for nonuse of medications.

All community residents in the study population had full prescription drug benefits without co-payments. In general, residents of this community have relatively high levels of education (78% had a bachelor’s degree or higher) and yearly income (only 4% were receiving less than $20,000).⁷ Primary care was largely obtained from an academic family physician and an internist, both with certificates of added qualification in geriatrics, and 2 geriatric nurse practitioners. To promote full disclosure of waste, data were kept confidential from the providers who cared for the study subjects. As a result, no clinical consequences of waste could be determined. The criteria for entry were age 65 years or older, voluntary response to study recruitment advertisements, residence in the facility for at least 1 year, and contact with a licensed health care provider within the past year. No volunteer subject was found ineligible. The Committee for Protection of Human Subjects at Dartmouth Medical School approved the study protocol.

The researchers coded medications by pharmaceutical class and calculated totals for each drug, including the costs of current medication use and total annual costs due to waste, using 1999 Red Book⁸ median wholesale drug cost estimates for a 3 months’ supply when exact prescription quantities were unknown.

Results

A total of 73 subjects received in-home pharmacy evaluations and completed questionnaires. Of these, 49 were women (67%), and 24 were men (33%). All were white, and all were older than 65 years. The mean age was 81.2 ± 6.0 (standard deviation [SD]) years. The mean number of years in residence was 5.4 ± 2.5 (SD).

The sum of all costs of wasted medication was $2011 in the study group (n=66). Mean per-person annual cost of wasted medication was $30.47 (range = $0-$131.56) based on the 66 subjects for whom complete data on pill counts were available. Pill counts were missing or incomplete on 7 questionnaires, which were not counted. Median annual waste was only $12.32, because 32 of the subjects (48%) wasted no medications. A total of 2078 wasted pills were found for the 66 subjects, yielding a mean of 31.5 pills wasted per subject (range = 0 to 208).

Mean waste represented 2.3% of total annual medication costs, which were $1302.78 per subject (interquartile range = $584.61-$1773.90). Increasing age was correlated with a higher number of pills wasted (r=0.35; P=.03) and a higher total cost of waste (r=0.20; P=.10). Total waste did not represent a fixed percentage of total annual medication costs. As yearly medication costs rose, waste as a percentage of yearly costs (ie, inefficient medication use) decreased significantly (r = -0.32; P=.02).

The most frequently wasted medication classes were antibiotics, benzodiazepines, and antihypertensives Table 1. Many of the medications listed as frequently wasted are taken episodically rather than in a stable daily pattern. Table 1 also shows that the most frequently wasted medications (eg, antibiotics) are not necessarily the most prescribed or the costliest.

Table 1 also shows the total annual relative costs of wasted medication by pharmaceutical class. Benzodiazepines, antidepressants, and antihypertensive medications combined accounted for a third of the total annual costs due to waste. The reasons for waste of medications and the relative contribution of each to total waste are presented in Table 2. The perception by subjects that a medical condition had resolved or that a medication was ineffective accounted for more than half of the cost due to waste. Physician and geriatric nurse practitioner perspectives are not captured by these data.

Discussion

On the basis of comprehensive home assessments, our study provides an estimate of wasted medication and the reasons for it in an outpatient population of older adults.

Most waste derived from 2 factors: the resolution of the condition for which the medication was prescribed and perceived ineffectiveness of a medication for its purpose. Together these 2 reasons accounted for more than half of the costs. This finding implies that acute conditions are central to waste, especially when medications for such conditions have high unit costs. Further support for the importance of acute conditions was that higher yearly drug expenditures were associated with lower percentage waste. Thus, high annual drug costs reflected stable, efficient patterns of medication use.

Although it may be difficult for clinicians to estimate how many pills to dispense, efforts should be made to determine the effectiveness and tolerability of medications before prescribing full quantities. Judicious use of samples is a possible remedy for this problem. Small prescriptions requiring multiple pharmacy visits would not help, but research on optimal prescribing quantities might lend some insight. Physicians should encourage patients to finish prescribed antibiotics if tolerated and not needlessly change prescriptions when previous pills remain. Further suggestions for promoting medication compliance in older adults are available in the medical literature.⁹

Limitations

We emphasize that our pilot study was small and not necessarily generalizable, yet it has made progress in a neglected area of research. As employees of the retirement community, the physicians caring for these patients are motivated to be fiscally responsible, and for this reason the mean annual waste detected in our study may, if anything, have been a substantial underestimate. Also, some subjects may have been overly optimistic in concluding that they intended to use all of a prescription medication on an as-needed basis. The lack of drug co-payments in this population, however, may predispose to more waste.

Based on our interviews, we found it uncharacteristic of most subjects to throw any pills away, but further underestimation could have occurred because of forgotten disposal of medicines. Researchers were motivated to find waste but could find none for 48% of subjects. Those subjects seemed sure that waste was absent.

Even though we counted more than 2000 wasted pills, numbers of specific medications were small. Thus, our analysis was confined to broad pharmaceutical classes and overall reasons for waste. This limitation could be overcome in more focused studies of specific wasted medications.

Conclusions

If, as we found in our study, average medication waste of $30 per person-year represents a conservative estimate, given that there are nearly 35 million individuals older than 65 years in the United States,¹⁰ the total national costs due to medication waste would not be less than $1 billion per year. Clearly, further studies in varied populations are required to confirm our waste estimate, and more research is needed to find effective waste reduction strategies. Despite the limitations of our study, physicians should begin to take note of what happens to prescribed medicines. That may serve as the most immediate basis for waste reduction.

Related resources

The Drugs and Devices Information Line Contains links to pharmacoepidemiology resources, maintained by the Pharmacoepidemiology Program, Harvard School of Public Health. http://www.hsph.harvard.edu/Organizations/DDIL/ddilhpge.html

Topics of Pharmacoepidemiology and Pharmacoeconomics A listserv for discussions related to these areas. http://www.findmail.com/list/pharmacoepidemiology/

ISPE—International Society for Pharmacoepidemiology A non-profit international professional organization dedicated to promoting pharmacoepidemiology. http://www.pharmacoepi.org/index.htm

Pharmacoepidemiology and Drug Safety The official journal of the International Society for Pharmacoepidemiology http://www.interscience.wiley.com/jpages/1053-8569/

Acknowledgment

The author acknowledges the assistance of the following undergraduate research assistants: Allison Robbins, Barbara Jones, Eva Liu, Karen Walp, Cynthia Oberto, Amanda Cook, John Raser, Sarah Hamilton, Anjali Godambe, and Michelle Anatone. Allen Dietrich, MD, provided guidance in the planning and execution of the study. Harlan Krumholz, MD, and Jerome Kassirer, MD, of the Yale University Robert Wood Johnson Clinical Scholars Program provided critical review of the manuscript.