- Email: [email protected]
Patient compliance and the conduct and interpretation of therapeutic trials
Patient Compliance and the Conduct and Interpretation of Therapeutic Trials R. Brian Haynes and Renato Dantes Department of Clinical Epidemiology and Biostatistics (R.B.H.;R.D.) and Department of Medicine, McMaster University, Hamilton, Ontario (R.B.H.)
ABSTRACT: Low patient compliance with prescribed treatments is a very common problem in clinical care and can seriously distort the generalizability and validity of controlled clinical trials. Aside from undermining the benefit of any treatment, noncompliance is often, but unpredictably, a marker for adverse patient outcomes independent of any treatment effect. The proper management of compliance in therapeutic trials depends in part on the objectives of the trial. If the purpose of the study is to determine whether a treatment does more good than harm to those who take it ("efficacy"), then noncompliers should be prevented from entering the trial. In a study of "effectiveness" (to determine whether a treatment does more good than harm to those to whom it is offered), quite the opposite approach is required. Regardless of the purpose of the trial, efforts should be made to balance the numbers of low compliers across the treatment groups, compliance should be monitored, and compliance improving strategies should be employed unless the main objective of the trial is to observe natural compliance. In all studies, because of the bias that noncompliance can have on results, the main analysis should include all those entered, whether or not compliant with the treatment regimen. KEY WORDS: patient compliance, random allocation, clinical trials, validity
INTRODUCTION Low patient compliance with prescribed regimens is a distressingly comm o n p r o b l e m that sharply limits the success of medical care a n d u n d e r m i n e s the execution a n d validity of clinical trials. Just as the practitioner m u s t be aware of a n d deal with this p r o b l e m in his or her practice, the clinical researcher m u s t take compliance into account in the design, execution, a n d analysis of clinical trials. F u r t h e r m o r e , the practitioner m u s t be able to anticipate the influence of compliance o n the o u t c o m e of clinical research if he or she is to interpret reports of therapeutic trials correctly. In this article, w e define compliance, describe the extent of noncompliance, explore the effect of n o n c o m p l i a n c e on. health outcomes, a n d p r o v i d e me-
Received January 31, 1986; revised October 14, 1986. Address reprint requests to: Dr. R.B. Haynes, Department of Clinical Epidemiology and Biostatistics, Rm 3V43, McMaster University Medical Center, 1200 Main St W, Hamilton, Ontario, Canada L8N 3Z5.
Controlled Clinical Trials 8:12-19 (1987) © Elsevier Science Publishing Co., Inc. 1987
12 0197-2456/87/$3.50
Patient Compliance and Therapeutic Trials
13
thodologic guidelines for the monitoring and management of compliance in controlled trials of medical treatments. THE DEFINITION OF COMPLIANCE AND THE MAGNITUDE OF NONCOMPLIANCE
In this article, compliance is used interchangeably with adherence and is defined as "the extent to which a person's behavior (in terms of taking medications, following diets, or executing lifestyle changes) coincides with medical or health advice" [1]. This definition is deceptively simple, however, for even the most straightforward regimens. For example, people who participate in hypertension screening projects and who are found to be hypertensive must have their blood pressure checked; must attend referral or follow-up appointments at a clinic for confirmation and at laboratory, and often x-ray and electrocardiography, facilities to assess possible underlying causes and target organ damage; must purchase and consume medications at the times and in the dose prescribed; may need to endure side effects; may be ordered to adhere to advice on weight reduction, salt restriction, exercise, and relaxation; must fit follow-up appointments into their schedules; and can look forward to these nuisances for the rest of their lives. Following treatment regimens, particularly for chronic illnesses, is not an easy task. In the research setting, the duration of demands on the participant may be shorter but the number and complexity of the tasks are frequently substantially greater, including questionnaires, additional assessment visits and special tests, often of an invasive nature. Added to this list is often the worry that any new treatments being tested may have as yet u n k n o w n adverse effects. The upshot of these difficulties in usual clinical settings has been reproducibly described [2]. Under usual circumstances of medical practice, up to 50% of patients on chronic medical regimens drop out of care entirely within 1 year of beginning it; only about 65% of those who continue to attend medical appointments consume enough of their medication to achieve therapeutic benefit; and adherance to dietary and other life-style changes is close to nil w h e n patients are followed for more than a few months. The additional demands of participation in a research project can result in even less cooperation. For example, Nessman et al. [3] could convince only 10% of otherwise eligible patients to enter a clinical trial testing alternative methods of improving the cooperation of patients with their antihypertensive regimens! No practitioner or researcher can afford to assume that patients or participants will follow prescribed treatments or follow-up procedures. Indeed, the opposite must be expected and planned for. DIFFERENCES BETWEEN COMPLIERS A N D NONCOMPLIERS
Separating individuals into compliant and noncompilant groups is usually based on arbitrary criteria, such as pill compliance. If a treatment is beneficial and its effect is dose dependent, it is obvious that high compilers and low compilers will differ quantitatively in the extent to which they exhibit a treatment response. In addition, it has been proposed
14
R.B. Haynes and R. Dantes that completely noncompliant patients can provide useful information concerning the natural history of disease or form a self-selected no-treatment control group in situations in which withholding treatment purposely would be unethical [4]. However, the validity of comparisons between compliers and noncompliers to determine the effect of treatments depends upon an important assumption: that noncompliers and compliers are similar to one another in all respects save that noncompliers receive less treatment than compliers. In fact, this is unlikely to be true as there is convincing evidence that noncompliers are substantively different from compliers in ways that are quite independent of the effects of the treatment prescribed. For example, men who were noncompliant with placebo in a randomized double-blind clinical trial of preventative therapies for coronary heart disease died at almost twice the rate of men who complied with placebo (28% mortality over 5 years compared with 15% mortality respectively, p < 0.0001) [5]. A similar effect has been observed in four additional placebo-controlled clinical trials [6]. However, the phenomenon is not entirely predictable. No such effect was observed in the Coronary Primary Prevention Trial [7], and epidemiologic studies have shown that volunteers are worse off [8], no different [9], and both worse and better off (depending on age) [10] than nonvolunteers.
INFLUENCE OF PATIENT NONCOMPLIANCE ON THE CONDUCT OF RESEARCH These findings have important implications for the design and interpretation of research. In the first place, it is clearly not reasonable to generalize the results of a study among compliant volunteers to all people with similar disorders since the majority of people will be noncompliant under usual conditions. Because most clinical trials are performed among volunteers (of necessity), and since so few people are naturally compliant, it must be concluded that the results of most clinical trials have at best a tenuous relationship to the "real world." That is not to say that trials with volunteers are valueless but rather to delimit the scope of their value: they answer questions concerning the "efficacy" of treatments (i.e., the extent to which they do more good than harm among people who actually take them) even though they may fall far short of answering the clinically relevant question of "effectiveness" (i.e. the extent to which treatments do more good than harm among people to w h o m they are offered). "Efficacy" or "explanatory" studies are important first steps in the testing of treatments, but producers and users of research must be aware of the gap between this form of initial testing and the much more difficult but pertinent "effectiveness" or "management" trials [111. Aside from the problem of generalizability of results, the differences between compliers and noncompliers can also distort the validity of trials, if not properly dealt with. Take, for example, the evaluation of anticoagulants following myocardial infarction, as reviewed by Chalmers et al. [12]. Initial reports of the use of anticoagulants compared case series of patients with "historical controls." Since patient consent is required for the use of a drug
Patient Compliance and Therapeutic Trials
15
in its investigational stages, members of a case series clearly must be volunteers, but not so historical controls who would generally comprise a mixture of nonvolunteers with a few potential volunteers. Anticoagulant-myocardial infarction studies with historical control groups showed substantially better results among anticoagulated patients. Nonrandomized control trials of anticoagulants gave similar though less spectacular results. Again, in such studies only volunteers can be included in the treatment group but the volunteer status of members of the control is seldom assessed. Worse still, in many of these studies, patients who refuse the treatment being evaluated are often assigned to the control group, making the comparison group all nonvolunteers or giving the comparison group a double load of nonvolunteers. Randomized clinical trials can neutralize the influence of compliance on the internal validity of studies by unbiased allocation of noncompliers to all treatment groups being compared and in the anticoagulant example such studies show virtually no benefit of treatment [12]. However, it does not follow that randomized trials are free of compliance problems. For example, the difference between treatment groups can be diminished by "contamination" of the control group by the active treatment in situations where control patients have access to the active treatment. This is a particular problem in "life-style" trials of diet or exercise, for example, and is a major shortcoming of the Multiple Risk Factor Intervention Trial [13]. This problem can also occur in drug trials of any over-the-counter drug such as acetylsalicylic acid for the prevention of stroke or myocardial infarction recurrence or of vitamin therapies. INVESTIGATOR MALCOMPLIANCE These acts of patient malcompliance are often aided and abetted by the well-intentioned but ill-advised malcompliance of researchers and/or their staffs. In the recruitment of patients for a study, patients may be entered into a particular study group because it is felt most appropriate for them. For example, a patient too ill to undergo surgery may be assigned to a nonsurgical control group, or a patient who is very ill might be assigned to the active treatment because the investigator believes it is unethical to withhold it, even though its value has not been formally established. These problems can occur even in otherwise well-executed randomized trials, though presumably to a far lesser extent than in nonrandomized trials. Furthermore, in the execution of studies, investigators may drop noncompliant patients from follow-up, ensuring that comparisons between treatment groups will become suspect. A logical extention of this fatal flaw is to analyze the results of studies excluding the noncompliers even when follow-up data on them are available. Last, though in some ways not worst, low patient cooperation can preclude the execution of some clinical trials through making it unfeasible to accumulate the sample size required to perform the study properly. Thus, under specified conditions, Goldsmith [14] has shown that it would require approximately five times as many participants to perform a trial at 50% mean compliance as
16
R.B. Haynes and R. Dantes at 100% mean compliance. Unfortunately, many investigations appear to begin on the assumption of much higher compliance, then fail to measure compliance during the execution of the study, and end concluding that there is no treatment benefit when, in fact, there has been no reasonable chance of observing even a sizable treatment benefit, due to the "noise" created by low compliance. A minor variant on this theme is to publish pharmacokinetic studies and dose-response curves for which compliance has not been taken into account. This must surely account for much of the intra- and interindividual variability in "drug absorption" seen in ambulatory studies and can also alter the shape of dose-response curves from such studies [14].
APPLICATION TO THE CONDUCT AND INTERPRETATION OF THERAPEUTIC TRIALS Those conducting clinical trials have a variety of strategies to monitor [15,16] and maximize [17,18] patient compliance, and clearly it is imperative that provision be made for doing so in the funding and logistical arrangements for trials. Furthermore, those reading reports of therapeutic trials should insist that compliance be managed properly, as documented in the methods section of the report, before they deem the results of trials worthy of consumption [19]. Beyond these generalities, there are some specific methodologic considerations that pertain to the compliance aspects of therapeutic trials. The first of these relates to the purpose of the trial. If the trial is to establish whether a treatment does more good than harm to people who take it as recommended, then every effort should be made to prevent noncompliers from entering the trial at all. This can be accomplished by asking for volunteers, then putting these volunteers through a "faintness of heart" period in which they are asked to comply with a regimen at least as demanding as the one that will be used in the trial. Only those patients who meet rigorous standards of compliance in the pretrial period should then be permitted to enter the study. The opposite approach is taken in an "effectiveness" or "management" trial, the object of which is to determine the value of a treatment for all patients who qualify on clinical grounds, regardless of their natural compliance. Such studies should, ideally, screen patient groups or ever, community populations for appropriate participants, being as persistent as is feasible within ethical and logistical constraints, and enter all such patients into the study. The Hypertension Detection and Follow-up Project provides an example of this approach, with participants being recruited from probability samples of the population adjacent to the study centers [20]. Regardless of whether the object is efficacy or effectiveness, once patients have agreed to enter the study, they should be allocated to treatment groups in such a fashion that the compliance characteristics of the groups are equivalent. To accomplish this, there is no substitute for formal random allocation of patients to treatment groups [21]. If there has been compliance screening before randomization, the precision of the statistical analysis can be aided by randomization within compliance strata, so that high and low compliers are equally distributed among the study groups.
Patient Compliance and Therapeutic Trials
17
During the execution of either efficacy or effectiveness studies, compliance with treatment (including the placebo or other control treatment) should be monitored by the most accurate methods that can be afforded [15,16]. If patients are to be divided into "compliant" and "noncompliant" groups, the division should ideally be made on the grounds of the relationship of the compliance level to the therapeutic response or outcome, if this is known [22]. The reasons for monitoring are twofold: to permit the efficient application of compliance-improving procedures, if the improvement of compliance is compatible with the objectives of the trial, and to determine the influence of compliance on the results of the trial [4]. With respect to the appropriateness of attempting to improve patient compliance during the course of the trial, this should obviously be done in all efficacy studies and also in all effectiveness or management studies save those in which the observation of natural compliance with an offered treatment is an intended objective of the trial. Of course, whenever compliance inducements are applied in a trial they should be described in the methods section of any research reports about the study: they constitute an important part of the intervention. Again, irrespective of whether the object is efficacy or effectiveness, once in the study, all participants should be accounted for in the final analysis in the study group to which they were originally assigned, whether or not they have complied with prescribed treatments. In the efficacy study this permits valid comparison between the treatment and control groups whereas in the effectiveness study this is to maintain both validity and generalizability to the population from which the participants were selected. Only w h e n these main, all-patient, analyses have been done can the investigators afford to indulge in subgroup analyses--
18
R.B. Haynes and R. Dantes ment trials for monitoring compliance and, in most studies, for applying effective compliance-improving strategies. Practitioners who wish to apply the results of clinical trials to the management of their own patients have a double task. First, they must base changes in their practices if possible on only those studies of otherwise adequate scientific merit in which compliance has been properly assessed and managed. Second, they must be prepared to prescribe not only the specific regimen that has been found of benefit but also to provide the ancilliary maneuvres required to achieve comparable levels of compliance with the regimen. Only through such an approach can we expect the results of scientifically valid research into therapy to be translated into effective treatment programs for our patients. Dr. Dantes is the recipient of an internationalClinicalEpidemiologyFellowshipfrom the Rockefeller Foundation.
REFERENCES
1. Haynes RB: Introduction. In: Compliance in Health Care, Haynes RB, Taylor DW, Sackett DL, eds. Baltimore: Johns Hopkins University Press, 1979, pp 1-7 2. Sackett DL, Snow JS: The magnitude of compliance and noncompliance. In: Compliance in Health Care, Haynes RB, Taylor DW, Sackett DL, eds. Baltimore: Johns Hopkins University Press, 1979, pp 11-22 3. Nessman DG, Carnahan JE, Nugent CA: Improving compliance: patient operated hypertension groups. Arch Intern Med 140:1427-1430,1980 4. Feinstein AR: "Compliance bias" and the interpretation of therapeutic trials. In: Compliance in Health Care, Haynes RB, Taylor DW, Sackett DL, eds. Baltimore: Johns Hopkins University Press, 1979, pp 309-335 5. The Coronary Drug Project Research Group: Influence of adherence to treatment and response of cholesterol on mortality in the Coronary Drug Project. N Engl J Med 303:1038-1041,1980 6. Epstein LH: The direct effect of compliance on health outcome. Health Psychol 3:385-393,1984 7. Lipid Research Clinics Program: The Lipid Research Clinics coronary primary prevention trial results. II. The relationship of reduction in incidence of coronary heart disease to cholesterol lowering. JAMA 251:365-374,1984 8. Stahl SM, Lawrie T, Neill P, Kelley C: Motivational interventions in community hypertension screening. Am J Public Health 67:345-350,1977 9. Silman AJ, Locke CM: Blood pressure distribution in responders and initial nonresponders in a population screening study. J Epidemiol Community Health 36:248-250,1982 10. Wing S, Tyroler HA, Manton KG: The participant effect: Mortality in a communitybased study compared to vital statistics. J Chron Dis 38:135-144,1985 11. Sackett DL, Gent M: Controversy in counting and attributing events in clinical trials. N Engl J Med 301:1410-1412,1979 12. Chalmers TC, Smith H, Ambrog A, Reitman D, Schroeder BJ: In defense of the VA randomized control trial of coronary artery surgery. Clin Res 26:229-235,1978 13. Multiple Risk Factor Intervention Trial Group: Multiple risk factor intervention trial. Risk factor changes and mortality results. JAMA 248:1465-1477,1982
Patient Compliance and Therapeutic Trials
19
14. Goldsmith CH: The effect of compliance distributions on therapeutic trials. In: Compliance in Health Care, Haynes RB, Taylor DW, Sackett DL, eds. Baltimore: Johns Hopkins University Press, 1979, pp. 297-308 15. Roth HP: Historical review: Comparison with other methods. Controlled Clin Trials 5:476--480,1984 16. Haynes RB, Taylor DW, Sackett DL, Gibson ES, Bernholz CD, Mukherjee J: Can simple clinical measurements detect patient noncompliance? Hypertension 2:757-764,1980 17. Haynes RB: A critical review of interventions to improve compliance with special reference to the role of physicians. Proceedings of the 1984 National Pharmaceutical Council Conference on Patient Compliance. Washington, D.C.: National Pharmaceutical Council, 1985, pp 45-57 18. Report of the NHLBI Working Group: Management of patient compliance in the treatment of hypertension. Hypertension 4:415-423,1982 19. Department of Clinical Epidemiology and Biostatistics, McMaster University: How to read clinical journals. V. To distinguish useful from useless or even harmful therapy. Can Med Assoc J 124:1156-1162,1981 20. Hypertension Detection and Follow-up Program Cooperative Group. Five-year findings of the Hypertension Detection and Follow-up Program. I. Reduction in mortality of persons with high blood pressure, including mild hypertension. JAMA 1979:242:2562-2571 21. Meinert CL, Tonascia S: Clinical Trials: Design, Conduct, and Analysis. New York: Oxford University Press, 1986 22. Gordis L: Conceptual and methodologic problems in measuring patient compliance. In: Compliance in Health Care, Haynes RB, Taylor DW, Sackett DL, eds. Baltimore: Johns Hopkins University Press, 1979, pp 23--48 23. The EC/IC Bypass Study Group: Failure of extracranial-intracranial arterial bypass to reduce the risk of ischemic stroke. Results of an international trial. N Engl J Med 313:1191-1200,1985
ABSTRACT: Low patient compliance with prescribed treatments is a very common problem in clinical care and can seriously distort the generalizability and validity of controlled clinical trials. Aside from undermining the benefit of any treatment, noncompliance is often, but unpredictably, a marker for adverse patient outcomes independent of any treatment effect. The proper management of compliance in therapeutic trials depends in part on the objectives of the trial. If the purpose of the study is to determine whether a treatment does more good than harm to those who take it ("efficacy"), then noncompliers should be prevented from entering the trial. In a study of "effectiveness" (to determine whether a treatment does more good than harm to those to whom it is offered), quite the opposite approach is required. Regardless of the purpose of the trial, efforts should be made to balance the numbers of low compliers across the treatment groups, compliance should be monitored, and compliance improving strategies should be employed unless the main objective of the trial is to observe natural compliance. In all studies, because of the bias that noncompliance can have on results, the main analysis should include all those entered, whether or not compliant with the treatment regimen. KEY WORDS: patient compliance, random allocation, clinical trials, validity
INTRODUCTION Low patient compliance with prescribed regimens is a distressingly comm o n p r o b l e m that sharply limits the success of medical care a n d u n d e r m i n e s the execution a n d validity of clinical trials. Just as the practitioner m u s t be aware of a n d deal with this p r o b l e m in his or her practice, the clinical researcher m u s t take compliance into account in the design, execution, a n d analysis of clinical trials. F u r t h e r m o r e , the practitioner m u s t be able to anticipate the influence of compliance o n the o u t c o m e of clinical research if he or she is to interpret reports of therapeutic trials correctly. In this article, w e define compliance, describe the extent of noncompliance, explore the effect of n o n c o m p l i a n c e on. health outcomes, a n d p r o v i d e me-
Received January 31, 1986; revised October 14, 1986. Address reprint requests to: Dr. R.B. Haynes, Department of Clinical Epidemiology and Biostatistics, Rm 3V43, McMaster University Medical Center, 1200 Main St W, Hamilton, Ontario, Canada L8N 3Z5.
Controlled Clinical Trials 8:12-19 (1987) © Elsevier Science Publishing Co., Inc. 1987
12 0197-2456/87/$3.50
Patient Compliance and Therapeutic Trials
13
thodologic guidelines for the monitoring and management of compliance in controlled trials of medical treatments. THE DEFINITION OF COMPLIANCE AND THE MAGNITUDE OF NONCOMPLIANCE
In this article, compliance is used interchangeably with adherence and is defined as "the extent to which a person's behavior (in terms of taking medications, following diets, or executing lifestyle changes) coincides with medical or health advice" [1]. This definition is deceptively simple, however, for even the most straightforward regimens. For example, people who participate in hypertension screening projects and who are found to be hypertensive must have their blood pressure checked; must attend referral or follow-up appointments at a clinic for confirmation and at laboratory, and often x-ray and electrocardiography, facilities to assess possible underlying causes and target organ damage; must purchase and consume medications at the times and in the dose prescribed; may need to endure side effects; may be ordered to adhere to advice on weight reduction, salt restriction, exercise, and relaxation; must fit follow-up appointments into their schedules; and can look forward to these nuisances for the rest of their lives. Following treatment regimens, particularly for chronic illnesses, is not an easy task. In the research setting, the duration of demands on the participant may be shorter but the number and complexity of the tasks are frequently substantially greater, including questionnaires, additional assessment visits and special tests, often of an invasive nature. Added to this list is often the worry that any new treatments being tested may have as yet u n k n o w n adverse effects. The upshot of these difficulties in usual clinical settings has been reproducibly described [2]. Under usual circumstances of medical practice, up to 50% of patients on chronic medical regimens drop out of care entirely within 1 year of beginning it; only about 65% of those who continue to attend medical appointments consume enough of their medication to achieve therapeutic benefit; and adherance to dietary and other life-style changes is close to nil w h e n patients are followed for more than a few months. The additional demands of participation in a research project can result in even less cooperation. For example, Nessman et al. [3] could convince only 10% of otherwise eligible patients to enter a clinical trial testing alternative methods of improving the cooperation of patients with their antihypertensive regimens! No practitioner or researcher can afford to assume that patients or participants will follow prescribed treatments or follow-up procedures. Indeed, the opposite must be expected and planned for. DIFFERENCES BETWEEN COMPLIERS A N D NONCOMPLIERS
Separating individuals into compliant and noncompilant groups is usually based on arbitrary criteria, such as pill compliance. If a treatment is beneficial and its effect is dose dependent, it is obvious that high compilers and low compilers will differ quantitatively in the extent to which they exhibit a treatment response. In addition, it has been proposed
14
R.B. Haynes and R. Dantes that completely noncompliant patients can provide useful information concerning the natural history of disease or form a self-selected no-treatment control group in situations in which withholding treatment purposely would be unethical [4]. However, the validity of comparisons between compliers and noncompliers to determine the effect of treatments depends upon an important assumption: that noncompliers and compliers are similar to one another in all respects save that noncompliers receive less treatment than compliers. In fact, this is unlikely to be true as there is convincing evidence that noncompliers are substantively different from compliers in ways that are quite independent of the effects of the treatment prescribed. For example, men who were noncompliant with placebo in a randomized double-blind clinical trial of preventative therapies for coronary heart disease died at almost twice the rate of men who complied with placebo (28% mortality over 5 years compared with 15% mortality respectively, p < 0.0001) [5]. A similar effect has been observed in four additional placebo-controlled clinical trials [6]. However, the phenomenon is not entirely predictable. No such effect was observed in the Coronary Primary Prevention Trial [7], and epidemiologic studies have shown that volunteers are worse off [8], no different [9], and both worse and better off (depending on age) [10] than nonvolunteers.
INFLUENCE OF PATIENT NONCOMPLIANCE ON THE CONDUCT OF RESEARCH These findings have important implications for the design and interpretation of research. In the first place, it is clearly not reasonable to generalize the results of a study among compliant volunteers to all people with similar disorders since the majority of people will be noncompliant under usual conditions. Because most clinical trials are performed among volunteers (of necessity), and since so few people are naturally compliant, it must be concluded that the results of most clinical trials have at best a tenuous relationship to the "real world." That is not to say that trials with volunteers are valueless but rather to delimit the scope of their value: they answer questions concerning the "efficacy" of treatments (i.e., the extent to which they do more good than harm among people who actually take them) even though they may fall far short of answering the clinically relevant question of "effectiveness" (i.e. the extent to which treatments do more good than harm among people to w h o m they are offered). "Efficacy" or "explanatory" studies are important first steps in the testing of treatments, but producers and users of research must be aware of the gap between this form of initial testing and the much more difficult but pertinent "effectiveness" or "management" trials [111. Aside from the problem of generalizability of results, the differences between compliers and noncompliers can also distort the validity of trials, if not properly dealt with. Take, for example, the evaluation of anticoagulants following myocardial infarction, as reviewed by Chalmers et al. [12]. Initial reports of the use of anticoagulants compared case series of patients with "historical controls." Since patient consent is required for the use of a drug
Patient Compliance and Therapeutic Trials
15
in its investigational stages, members of a case series clearly must be volunteers, but not so historical controls who would generally comprise a mixture of nonvolunteers with a few potential volunteers. Anticoagulant-myocardial infarction studies with historical control groups showed substantially better results among anticoagulated patients. Nonrandomized control trials of anticoagulants gave similar though less spectacular results. Again, in such studies only volunteers can be included in the treatment group but the volunteer status of members of the control is seldom assessed. Worse still, in many of these studies, patients who refuse the treatment being evaluated are often assigned to the control group, making the comparison group all nonvolunteers or giving the comparison group a double load of nonvolunteers. Randomized clinical trials can neutralize the influence of compliance on the internal validity of studies by unbiased allocation of noncompliers to all treatment groups being compared and in the anticoagulant example such studies show virtually no benefit of treatment [12]. However, it does not follow that randomized trials are free of compliance problems. For example, the difference between treatment groups can be diminished by "contamination" of the control group by the active treatment in situations where control patients have access to the active treatment. This is a particular problem in "life-style" trials of diet or exercise, for example, and is a major shortcoming of the Multiple Risk Factor Intervention Trial [13]. This problem can also occur in drug trials of any over-the-counter drug such as acetylsalicylic acid for the prevention of stroke or myocardial infarction recurrence or of vitamin therapies. INVESTIGATOR MALCOMPLIANCE These acts of patient malcompliance are often aided and abetted by the well-intentioned but ill-advised malcompliance of researchers and/or their staffs. In the recruitment of patients for a study, patients may be entered into a particular study group because it is felt most appropriate for them. For example, a patient too ill to undergo surgery may be assigned to a nonsurgical control group, or a patient who is very ill might be assigned to the active treatment because the investigator believes it is unethical to withhold it, even though its value has not been formally established. These problems can occur even in otherwise well-executed randomized trials, though presumably to a far lesser extent than in nonrandomized trials. Furthermore, in the execution of studies, investigators may drop noncompliant patients from follow-up, ensuring that comparisons between treatment groups will become suspect. A logical extention of this fatal flaw is to analyze the results of studies excluding the noncompliers even when follow-up data on them are available. Last, though in some ways not worst, low patient cooperation can preclude the execution of some clinical trials through making it unfeasible to accumulate the sample size required to perform the study properly. Thus, under specified conditions, Goldsmith [14] has shown that it would require approximately five times as many participants to perform a trial at 50% mean compliance as
16
R.B. Haynes and R. Dantes at 100% mean compliance. Unfortunately, many investigations appear to begin on the assumption of much higher compliance, then fail to measure compliance during the execution of the study, and end concluding that there is no treatment benefit when, in fact, there has been no reasonable chance of observing even a sizable treatment benefit, due to the "noise" created by low compliance. A minor variant on this theme is to publish pharmacokinetic studies and dose-response curves for which compliance has not been taken into account. This must surely account for much of the intra- and interindividual variability in "drug absorption" seen in ambulatory studies and can also alter the shape of dose-response curves from such studies [14].
APPLICATION TO THE CONDUCT AND INTERPRETATION OF THERAPEUTIC TRIALS Those conducting clinical trials have a variety of strategies to monitor [15,16] and maximize [17,18] patient compliance, and clearly it is imperative that provision be made for doing so in the funding and logistical arrangements for trials. Furthermore, those reading reports of therapeutic trials should insist that compliance be managed properly, as documented in the methods section of the report, before they deem the results of trials worthy of consumption [19]. Beyond these generalities, there are some specific methodologic considerations that pertain to the compliance aspects of therapeutic trials. The first of these relates to the purpose of the trial. If the trial is to establish whether a treatment does more good than harm to people who take it as recommended, then every effort should be made to prevent noncompliers from entering the trial at all. This can be accomplished by asking for volunteers, then putting these volunteers through a "faintness of heart" period in which they are asked to comply with a regimen at least as demanding as the one that will be used in the trial. Only those patients who meet rigorous standards of compliance in the pretrial period should then be permitted to enter the study. The opposite approach is taken in an "effectiveness" or "management" trial, the object of which is to determine the value of a treatment for all patients who qualify on clinical grounds, regardless of their natural compliance. Such studies should, ideally, screen patient groups or ever, community populations for appropriate participants, being as persistent as is feasible within ethical and logistical constraints, and enter all such patients into the study. The Hypertension Detection and Follow-up Project provides an example of this approach, with participants being recruited from probability samples of the population adjacent to the study centers [20]. Regardless of whether the object is efficacy or effectiveness, once patients have agreed to enter the study, they should be allocated to treatment groups in such a fashion that the compliance characteristics of the groups are equivalent. To accomplish this, there is no substitute for formal random allocation of patients to treatment groups [21]. If there has been compliance screening before randomization, the precision of the statistical analysis can be aided by randomization within compliance strata, so that high and low compliers are equally distributed among the study groups.
Patient Compliance and Therapeutic Trials
17
During the execution of either efficacy or effectiveness studies, compliance with treatment (including the placebo or other control treatment) should be monitored by the most accurate methods that can be afforded [15,16]. If patients are to be divided into "compliant" and "noncompliant" groups, the division should ideally be made on the grounds of the relationship of the compliance level to the therapeutic response or outcome, if this is known [22]. The reasons for monitoring are twofold: to permit the efficient application of compliance-improving procedures, if the improvement of compliance is compatible with the objectives of the trial, and to determine the influence of compliance on the results of the trial [4]. With respect to the appropriateness of attempting to improve patient compliance during the course of the trial, this should obviously be done in all efficacy studies and also in all effectiveness or management studies save those in which the observation of natural compliance with an offered treatment is an intended objective of the trial. Of course, whenever compliance inducements are applied in a trial they should be described in the methods section of any research reports about the study: they constitute an important part of the intervention. Again, irrespective of whether the object is efficacy or effectiveness, once in the study, all participants should be accounted for in the final analysis in the study group to which they were originally assigned, whether or not they have complied with prescribed treatments. In the efficacy study this permits valid comparison between the treatment and control groups whereas in the effectiveness study this is to maintain both validity and generalizability to the population from which the participants were selected. Only w h e n these main, all-patient, analyses have been done can the investigators afford to indulge in subgroup analyses--
18
R.B. Haynes and R. Dantes ment trials for monitoring compliance and, in most studies, for applying effective compliance-improving strategies. Practitioners who wish to apply the results of clinical trials to the management of their own patients have a double task. First, they must base changes in their practices if possible on only those studies of otherwise adequate scientific merit in which compliance has been properly assessed and managed. Second, they must be prepared to prescribe not only the specific regimen that has been found of benefit but also to provide the ancilliary maneuvres required to achieve comparable levels of compliance with the regimen. Only through such an approach can we expect the results of scientifically valid research into therapy to be translated into effective treatment programs for our patients. Dr. Dantes is the recipient of an internationalClinicalEpidemiologyFellowshipfrom the Rockefeller Foundation.
REFERENCES
1. Haynes RB: Introduction. In: Compliance in Health Care, Haynes RB, Taylor DW, Sackett DL, eds. Baltimore: Johns Hopkins University Press, 1979, pp 1-7 2. Sackett DL, Snow JS: The magnitude of compliance and noncompliance. In: Compliance in Health Care, Haynes RB, Taylor DW, Sackett DL, eds. Baltimore: Johns Hopkins University Press, 1979, pp 11-22 3. Nessman DG, Carnahan JE, Nugent CA: Improving compliance: patient operated hypertension groups. Arch Intern Med 140:1427-1430,1980 4. Feinstein AR: "Compliance bias" and the interpretation of therapeutic trials. In: Compliance in Health Care, Haynes RB, Taylor DW, Sackett DL, eds. Baltimore: Johns Hopkins University Press, 1979, pp 309-335 5. The Coronary Drug Project Research Group: Influence of adherence to treatment and response of cholesterol on mortality in the Coronary Drug Project. N Engl J Med 303:1038-1041,1980 6. Epstein LH: The direct effect of compliance on health outcome. Health Psychol 3:385-393,1984 7. Lipid Research Clinics Program: The Lipid Research Clinics coronary primary prevention trial results. II. The relationship of reduction in incidence of coronary heart disease to cholesterol lowering. JAMA 251:365-374,1984 8. Stahl SM, Lawrie T, Neill P, Kelley C: Motivational interventions in community hypertension screening. Am J Public Health 67:345-350,1977 9. Silman AJ, Locke CM: Blood pressure distribution in responders and initial nonresponders in a population screening study. J Epidemiol Community Health 36:248-250,1982 10. Wing S, Tyroler HA, Manton KG: The participant effect: Mortality in a communitybased study compared to vital statistics. J Chron Dis 38:135-144,1985 11. Sackett DL, Gent M: Controversy in counting and attributing events in clinical trials. N Engl J Med 301:1410-1412,1979 12. Chalmers TC, Smith H, Ambrog A, Reitman D, Schroeder BJ: In defense of the VA randomized control trial of coronary artery surgery. Clin Res 26:229-235,1978 13. Multiple Risk Factor Intervention Trial Group: Multiple risk factor intervention trial. Risk factor changes and mortality results. JAMA 248:1465-1477,1982
Patient Compliance and Therapeutic Trials
19
14. Goldsmith CH: The effect of compliance distributions on therapeutic trials. In: Compliance in Health Care, Haynes RB, Taylor DW, Sackett DL, eds. Baltimore: Johns Hopkins University Press, 1979, pp. 297-308 15. Roth HP: Historical review: Comparison with other methods. Controlled Clin Trials 5:476--480,1984 16. Haynes RB, Taylor DW, Sackett DL, Gibson ES, Bernholz CD, Mukherjee J: Can simple clinical measurements detect patient noncompliance? Hypertension 2:757-764,1980 17. Haynes RB: A critical review of interventions to improve compliance with special reference to the role of physicians. Proceedings of the 1984 National Pharmaceutical Council Conference on Patient Compliance. Washington, D.C.: National Pharmaceutical Council, 1985, pp 45-57 18. Report of the NHLBI Working Group: Management of patient compliance in the treatment of hypertension. Hypertension 4:415-423,1982 19. Department of Clinical Epidemiology and Biostatistics, McMaster University: How to read clinical journals. V. To distinguish useful from useless or even harmful therapy. Can Med Assoc J 124:1156-1162,1981 20. Hypertension Detection and Follow-up Program Cooperative Group. Five-year findings of the Hypertension Detection and Follow-up Program. I. Reduction in mortality of persons with high blood pressure, including mild hypertension. JAMA 1979:242:2562-2571 21. Meinert CL, Tonascia S: Clinical Trials: Design, Conduct, and Analysis. New York: Oxford University Press, 1986 22. Gordis L: Conceptual and methodologic problems in measuring patient compliance. In: Compliance in Health Care, Haynes RB, Taylor DW, Sackett DL, eds. Baltimore: Johns Hopkins University Press, 1979, pp 23--48 23. The EC/IC Bypass Study Group: Failure of extracranial-intracranial arterial bypass to reduce the risk of ischemic stroke. Results of an international trial. N Engl J Med 313:1191-1200,1985