- Email: [email protected]
General concepts for use of markers in clinical trials
General Concepts for Use of Markers in Clinical Trials Leon Gordis Department of Epidemiology, Johns Hopkins University School of Hygiene and Public Health, Baltimore, Maryland
ABSTRACT: Both indirect and direct methods have been widely employed for measuring medication compliance. Indirect methods include therapeutic or preventive outcome, assessment by the physician, interview with the patient, whether or not the prescription was filled, and a count of remaining pills. In many situations, direct methods may be feasible, including measurement of blood or serum levels or testing urine for excretion of the medication itself, a metabolic by-product, or a marker or tracer that has been added to the medication for detection purposes. For example, detection of penicillin or of salicylates can be used in measuring compliance, while in other circumstances, it may be necessary to add a detectable label to the medication. Ideally, such a marker should be nontoxic and pharmacologically and chemically inert [Porter AM: Br Med J 1:218-222, 1969]. The marker should be unaffected by physical and chemical properties of the urine, such as pH and temperature, quickly and freely excreted, and noncumulative. A simple, sensitive, and specific detection method should be available, and the marker should be such that the patient is unaware that it has been added. In direct measures of compliance, it is necessary to consider pharmacokinetic variations among individuals in absorption, distribution, metabolism, and excretion of drugs. In addition, temporal aspect~ of the sampling scheme assume great importance. Finally, the definition of compliance and noncompliance for a given study in relation to the specific question being tested in the investigation, as well as in regard to use of the marker itself, is an important issue for consideration. Since indirect methods are often insufficient for measuring compliance, direct measures of compliance are usually preferable in clinical trials.
M u c h of the literature dealing w i t h c o m p l i a n c e focuses o n the d y n a m i c s of c o m p l i a n c e [1]. Investigators often try to characterize g r o u p s of c o m p l i e r s a n d g r o u p s of n o n c o m p l i e r s to u n d e r s t a n d the d y n a m i c s of n o n c o m p l i a n c e a n d to m o d i f y this behavior. H o w e v e r , this is not identical to the issue b e i n g a d d r e s s e d in this w o r k s h o p . W e are talking a b o u t a s s e s s i n g c o m p l i a n c e or n o n c o m p l i a n c e for p u r p o s e s of d e t e r m i n i n g the effectiveness of a d r u g in a clinical trial. T h e r e is little p o i n t discussing use of a m a r k e r for m e a s u r i n g c o m p l i a n c e in a clinical trial unless w e k n o w w h y w e are s t u d y i n g compliance. W e m i g h t
Address reprint requests to: Leon Gordis, M.D., Dr. P.H., Professorand Chairman, Department of Epidemiology, Johns Hopkins University School of Hygiene and PublicHealth, 615 North Wolfe Street, Baltimore, Maryland 21205. Controlled ClinicalTrials5:481-487 (1984) © ElsevierSciencePublishing Co., Inc. 1984 52 VanderbiltAve., New York, New York10017
481 0197-2456/84/$03.00
482
Leon Gordis exclude potential noncompliers before the study begins. For example, the Veterans Administration hypertension study assessed the patients initially and excluded potential noncompliers from the trial. A second possibility might be to identify noncompliers during the trial itself, while data are being gathered, and exclude them from the analysis. Third, we might assist adherence counselors, people who specialize in enhancing patient compliance, by identifying for them people who are noncompliers. Fourth, compliance testing could be used to enhance compliance, either by warning people that their compliance is being tested or by telling them the results of the testing. This might have some motivational effect in enhancing compliance. This point is particularly important in distinguishing between studies focusing on the dynamics of compliance and studies focusing on the adequacy of compliance for proper interpretation of the results of clinical trials. In the former, both compliers and noncompliers are studied to characterize them. Intervention is not desirable unless an intervention study is being conducted. On the other hand, if a clinical trial aims to maximize the proportion of the patients w h o take the medication that is being tested, it might be reasonable to test compliance and give patients feedback of their results. We therefore need to address the issue of whether the patient should be unaware of the marker if one is used in a clinical trial. Noncompliance must also be considered in analyzing the data. Canner et al. discussed the issue of stratification for compliance and the implications of postrandomization stratification [2]. Identifying and excluding noncompliers during a trial or stratifying by compliance raises the issues that noncompliance can be determined by social and other characteristics of the individual involved or by the doctor-patient relationshi p or may be related to characteristics of the actual agent being tested. Before one excludes noncompliers, we should ask whether we want to exclude noncompliers or whether noncompliance is an important endpoint that may be of interest in assessing the agent under study. Dr. Insull drew a distinction between the "best case" study, where a limited clinical trial is done under the best circumstances, and a "real-world" study, where it is carried out in a community that includes both compliers and noncompliers. It seems that we are talking about an attributable benefit concept: What is the benefit to those who actually comply, and what is the benefit to the total population of people for w h o m the drug was prescribed? I would like to raise the question of whether we are using a marker as a measure of patient behavior or as a surrogate for a cumulative index of serum drug levels over time. Are we interested in whether the patient has actually put the pills in his or her mouth at the appropriate time? Or do we want to know whether the patient achieved and maintained an adequate blood level for therapeutic effect? Both are legitimate endpoints. However, when we consider the issue of markers, we should address this question because the answer may make a great difference. A person may be a poor complier but obtain an adequate blood level if he takes the medication at all. Conversely, a person may be a very good complier but not achieve a good serum level for a variety of reasons. For example, we have issues of the absorption of the agent: Was the agent taken with meals or not with meals? What is the metabolism of the agent? Was it taken with other drugs? Was the subject a smoker?
483
Use of Markers in Clinical Trials SMOKING: r-~
5O
None
Light Heovy
~
30
T T,
r~
¢,1 ~o
i
AGE, yr :
18- 39
40-59
60-92
Figure 1 Effectof smoking and age on the metabolic clearance of antipyrine. Bars show mean _+ S.E. Data are from 307 healthy subjects [3]. Figure 1 shows the effects of smoking and age on the clearance of antipyrine. In the two younger age groups, the heavy smokers have a more rapid rate of clearance than the light smokers. Therefore, to assess compliance, do we have to monitor smoking in some way, using nicotine or cotinine, as much as we might have to monitor the agent itself? We must also consider genetic differences. Figure 2 shows the decay of ethyl biscoumacetate in the plasma of eight normal volunteers after a single intravenous dose of 20 mg/kg [4,5]. The slopes differ in these eight normal individuals, demonstrating the tremendous heterogeneity among individuals that can affect the rates at which drugs are metabolized. We know that alcohol intake can affect the absorption and metabolism of drugs. We know that other conditions, such as fever or concomitant disease, may affect absorption and metabolism. The question is: What other diseases must we know about in patients who are participating in a clinical trial, in view of the fact that the underlying or associated disease they have may affect the ultimate serum levels or therapeutic effects of the drugs they have taken? One might also ask: How many patients have saved themselves from iatrogenic disease or death by being noncompliers at appropriate times? But that is not the issue to be addressed here. Compliance among controls is less a marker question than it is a detection of medication: How do we get evidence that the controls have not taken the agent? In the Aspirin Myocardial Infarction Study, an attempt was made to ensure that controls did not take aspirin. However, when one compiles a list of aspirin-containing preparations on the market, it is so large that it may be impossible for a patient to avoid taking aspirin. This is a difficult issue. I would like to turn to the definition of "noncompliance." "Compliance" is a spectrum and not an absolute point. There are various ways for categorizing patients in terms of compliance. Many investigators dichotomize the
484
Leon Gordis
|
I
I
I
I
i I
I Z
1 3
1 4
1 5
1
20Or-
I00
--J L.O
~E
50 40
30
20
tOi
0
TIME Figure
(HOURS)
2 Decay of ethyl biscoumacetate in the plasma of eight normal volunteers after a single intravenous dose of 20 mg/kg [4].
p o p u l a t i o n into compliers a n d noncompliers, sometimes on a biologic basis; that is, does the d r u g work? For example, if a p e r s o n ' s blood pressure drops, he is a complier; if it does not drop, he is a noncomplier. The reasoning is s o m e w h a t circular, because y o u are defining compliance in terms of the positive o u t c o m e that y o u are seeking to evaluate in testing an agent. Compliance has also been defined on a statistical basis. Investigators m a y take the m e d i a n level of compliance and consider people w h o are below it as noncompliers. Or, an arbitrary level m a y be chosen for a cutoff, such as 30 or 75 percent. Usually, we do not have a biologic basis for selecting a clear cutoff point for compliance and noncompliance. If we are going to use a marker and w e w a n t to k n o w what the marker is for, and if we are going to d e t e r m i n e compliance or noncompliance, we must k n o w precisely w h a t we m e a n by "compliance" and "noncompliance." We also can look at compliance as a continuous variable, which is w h a t it is. H o w e v e r , in clinical trials, this approach generally does not help m u c h because we usually w a n t to characterize a given individual as a compiler or a noncomplier. A n o t h e r problem is that t h e r e can be considerable intraindividual variation over time. Table 1 shows the coefficients of variation for a n u m b e r of a g e n t s
485
Use of Markers in Clinical Trials Table 1
Summary of Interindividual and Intraindividual Coefficients Variation for Drug Metabolism in Man Interindividual Intraindividual coefficient of coefficient of Parameter variation (a) (%) variation (b) (%) Phenylbutazone t 1/2 17.29 8.66 Antipyrine t 1/2 24.29 13.49 Phenacetin t 1/2 4.40 19.50 Phenacetin AUC 54.93 58.81
of
b/a
0.50 0.56 4.43 1.07
[6]. For phenacetin's half-life, the intraindividual coefficient of variation is very high. This raises serious questions when you take samples at different points in time. What kind of intraindividual variations would you anticipate might occur even if the patient is complying? Another important issue is "percent compliance." For example, Dr. Insull mentioned 65% compliance. We must be clear about what we mean by 65% compliance or 50% compliance or whatever. Figure 3 shows four hypothetical patients w h o are 50% compliers. Each patient had 10 urine tests done for a drug, and all are 50% compliers. One person was a noncomplier and something happened to make that person comply. Perhaps the adherence counselor got to him, or something changed in this person's life relating to illness behavior. The second subject changed from compliance to noncompliance. The third is a hypothetical example of a patient who alternates doses. The fourth subject is more of a haphazard distribution of pluses and minuses. The therapeutic and preventive implications are quite different here. If the first or second patient is on penicillin prophylaxis, that patient may be unprotected for half the time. On the other hand, the third patient may be adequately protected by having alternated doses. Usually we do not have the biologic information in terms of h o w much the patient can miss and still be protected and get a therapeutic effect. Therefore, it is not sufficient to say that a patient is a 50% complier. The pattern of noncompliance must be described. Figure 3 Hypothetical results of sequential tests in four 50% compliers. .
.
.
.
+
+
+
l
.
Patient
2
+
+
+
+
+
.
Patient
3
+
-
+
-
+
-
+
-
+
-
Patient
4
+
+
-
-
-
+
+
-
+
-
.
+
+
Patient
.
.
.
486
Leon Gordis When we measure compliance we are essentially sampling. Let us say these tests represent urine tests. Are we sampling every day of the week? Are we selecting days at random, from Sunday through Saturday, or from Monday through Friday? What is the sampling frame from which we are drawing those tests? Several other issues should be looked at regarding compliance testing: Where were the urine specimens obtained (home, school, work, etc.)? Did the patient know the reason for the tests? Was the patient characterized on the basis of one test, or was testing done periodically? Was the schedule fixed, variable, or random, and did the patient know the schedule? Aside from the issue of sampling, what about medications that are prescribed tWice daily or thrice daily? What should be the sampling schema for measuring compliance? In our experience with school children on twice daily medication, for example, we found we could get a morning specimen at school, but it was very hard to come into a home and ask for the evening specimen. This is a tremendous logistic problem in measuring compliance in clinical trials that involve multiple doses per day. Another important issue is whether reducing the frequency of medication will improve patient compliance. Is one accomplishing something positive for the patient by reducing the frequency? Consider the situation where a patient is on tWice-a-day medication and then gets changed to once-a-day medication. Let us assume that the patient is a 50% complier. If he remains a 50% complier after the frequency of the medication is reduced, we have a potential for serious trouble. Since he is only getting medication once a day and is a 50% complier, he only takes the medication every alternate day. Therefore, if we reduce the frequency of doses to enhance the patient's welfare, there is an assumption not only that compliance is not getting worse, but that it is also not staying the same, but is getting at least twice as good to break even. This issue has not been investigated adequately, partly because the problem of monitoring compliance in multiple-dose regimens, such as getting the evening specimen. I would like to raise the question of whether we should be concerned with the predictive value of the test in clinical trials. What is the probability that a patient has taken the medication given a certain test result when compliance is monitored? This raises a host of other methodologic issues: In a clinical trial, is the central question regarding markers one of sensitivity and specificity, or is it the predictive value of the marker? Finally, ! would like to ask whether it is important that the patient remain unaware of the marker. If it is important that the patient remain unaware, it raises ethical issues in carrying out the study because of the problem of informed consent. Certain human volunteers committees do not want patients' urines examined to see whether they have taken the medication, unless the patients have been adequately informed of the reason for the urine test being done. This is a potentially serious problem.
Use of Markers in Clinical Trials
487
REFERENCES 1. 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 2. 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 3. Vestal RE, Norris AH, Tobin JD, et al: Antipyrine metabolism in man: Influence of age, alcohol, caffeine, and smoking. Clin Pharmacol Ther 18:425--432, 1975 4. Vesell ES: Factors causing interindividual variations of drug concentrations in blood. Clin Pharmacol Ther 16:135-148, 1974 5. Brodie BB, Weiner M, Burns JJ, et al: The physiological disposition of ethyl biscoumacetate (tromexan) in man and a method for its estimation in biological material. J Pharmacol Exp Ther 106:453463, 1952 6. Alvares AP, Kappas A, Eiseman JL, et al: Intraindividual variation in drug disposition. Clin Pharmacol Ther 26:407--419, 1979
ABSTRACT: Both indirect and direct methods have been widely employed for measuring medication compliance. Indirect methods include therapeutic or preventive outcome, assessment by the physician, interview with the patient, whether or not the prescription was filled, and a count of remaining pills. In many situations, direct methods may be feasible, including measurement of blood or serum levels or testing urine for excretion of the medication itself, a metabolic by-product, or a marker or tracer that has been added to the medication for detection purposes. For example, detection of penicillin or of salicylates can be used in measuring compliance, while in other circumstances, it may be necessary to add a detectable label to the medication. Ideally, such a marker should be nontoxic and pharmacologically and chemically inert [Porter AM: Br Med J 1:218-222, 1969]. The marker should be unaffected by physical and chemical properties of the urine, such as pH and temperature, quickly and freely excreted, and noncumulative. A simple, sensitive, and specific detection method should be available, and the marker should be such that the patient is unaware that it has been added. In direct measures of compliance, it is necessary to consider pharmacokinetic variations among individuals in absorption, distribution, metabolism, and excretion of drugs. In addition, temporal aspect~ of the sampling scheme assume great importance. Finally, the definition of compliance and noncompliance for a given study in relation to the specific question being tested in the investigation, as well as in regard to use of the marker itself, is an important issue for consideration. Since indirect methods are often insufficient for measuring compliance, direct measures of compliance are usually preferable in clinical trials.
M u c h of the literature dealing w i t h c o m p l i a n c e focuses o n the d y n a m i c s of c o m p l i a n c e [1]. Investigators often try to characterize g r o u p s of c o m p l i e r s a n d g r o u p s of n o n c o m p l i e r s to u n d e r s t a n d the d y n a m i c s of n o n c o m p l i a n c e a n d to m o d i f y this behavior. H o w e v e r , this is not identical to the issue b e i n g a d d r e s s e d in this w o r k s h o p . W e are talking a b o u t a s s e s s i n g c o m p l i a n c e or n o n c o m p l i a n c e for p u r p o s e s of d e t e r m i n i n g the effectiveness of a d r u g in a clinical trial. T h e r e is little p o i n t discussing use of a m a r k e r for m e a s u r i n g c o m p l i a n c e in a clinical trial unless w e k n o w w h y w e are s t u d y i n g compliance. W e m i g h t
Address reprint requests to: Leon Gordis, M.D., Dr. P.H., Professorand Chairman, Department of Epidemiology, Johns Hopkins University School of Hygiene and PublicHealth, 615 North Wolfe Street, Baltimore, Maryland 21205. Controlled ClinicalTrials5:481-487 (1984) © ElsevierSciencePublishing Co., Inc. 1984 52 VanderbiltAve., New York, New York10017
481 0197-2456/84/$03.00
482
Leon Gordis exclude potential noncompliers before the study begins. For example, the Veterans Administration hypertension study assessed the patients initially and excluded potential noncompliers from the trial. A second possibility might be to identify noncompliers during the trial itself, while data are being gathered, and exclude them from the analysis. Third, we might assist adherence counselors, people who specialize in enhancing patient compliance, by identifying for them people who are noncompliers. Fourth, compliance testing could be used to enhance compliance, either by warning people that their compliance is being tested or by telling them the results of the testing. This might have some motivational effect in enhancing compliance. This point is particularly important in distinguishing between studies focusing on the dynamics of compliance and studies focusing on the adequacy of compliance for proper interpretation of the results of clinical trials. In the former, both compliers and noncompliers are studied to characterize them. Intervention is not desirable unless an intervention study is being conducted. On the other hand, if a clinical trial aims to maximize the proportion of the patients w h o take the medication that is being tested, it might be reasonable to test compliance and give patients feedback of their results. We therefore need to address the issue of whether the patient should be unaware of the marker if one is used in a clinical trial. Noncompliance must also be considered in analyzing the data. Canner et al. discussed the issue of stratification for compliance and the implications of postrandomization stratification [2]. Identifying and excluding noncompliers during a trial or stratifying by compliance raises the issues that noncompliance can be determined by social and other characteristics of the individual involved or by the doctor-patient relationshi p or may be related to characteristics of the actual agent being tested. Before one excludes noncompliers, we should ask whether we want to exclude noncompliers or whether noncompliance is an important endpoint that may be of interest in assessing the agent under study. Dr. Insull drew a distinction between the "best case" study, where a limited clinical trial is done under the best circumstances, and a "real-world" study, where it is carried out in a community that includes both compliers and noncompliers. It seems that we are talking about an attributable benefit concept: What is the benefit to those who actually comply, and what is the benefit to the total population of people for w h o m the drug was prescribed? I would like to raise the question of whether we are using a marker as a measure of patient behavior or as a surrogate for a cumulative index of serum drug levels over time. Are we interested in whether the patient has actually put the pills in his or her mouth at the appropriate time? Or do we want to know whether the patient achieved and maintained an adequate blood level for therapeutic effect? Both are legitimate endpoints. However, when we consider the issue of markers, we should address this question because the answer may make a great difference. A person may be a poor complier but obtain an adequate blood level if he takes the medication at all. Conversely, a person may be a very good complier but not achieve a good serum level for a variety of reasons. For example, we have issues of the absorption of the agent: Was the agent taken with meals or not with meals? What is the metabolism of the agent? Was it taken with other drugs? Was the subject a smoker?
483
Use of Markers in Clinical Trials SMOKING: r-~
5O
None
Light Heovy
~
30
T T,
r~
¢,1 ~o
i
AGE, yr :
18- 39
40-59
60-92
Figure 1 Effectof smoking and age on the metabolic clearance of antipyrine. Bars show mean _+ S.E. Data are from 307 healthy subjects [3]. Figure 1 shows the effects of smoking and age on the clearance of antipyrine. In the two younger age groups, the heavy smokers have a more rapid rate of clearance than the light smokers. Therefore, to assess compliance, do we have to monitor smoking in some way, using nicotine or cotinine, as much as we might have to monitor the agent itself? We must also consider genetic differences. Figure 2 shows the decay of ethyl biscoumacetate in the plasma of eight normal volunteers after a single intravenous dose of 20 mg/kg [4,5]. The slopes differ in these eight normal individuals, demonstrating the tremendous heterogeneity among individuals that can affect the rates at which drugs are metabolized. We know that alcohol intake can affect the absorption and metabolism of drugs. We know that other conditions, such as fever or concomitant disease, may affect absorption and metabolism. The question is: What other diseases must we know about in patients who are participating in a clinical trial, in view of the fact that the underlying or associated disease they have may affect the ultimate serum levels or therapeutic effects of the drugs they have taken? One might also ask: How many patients have saved themselves from iatrogenic disease or death by being noncompliers at appropriate times? But that is not the issue to be addressed here. Compliance among controls is less a marker question than it is a detection of medication: How do we get evidence that the controls have not taken the agent? In the Aspirin Myocardial Infarction Study, an attempt was made to ensure that controls did not take aspirin. However, when one compiles a list of aspirin-containing preparations on the market, it is so large that it may be impossible for a patient to avoid taking aspirin. This is a difficult issue. I would like to turn to the definition of "noncompliance." "Compliance" is a spectrum and not an absolute point. There are various ways for categorizing patients in terms of compliance. Many investigators dichotomize the
484
Leon Gordis
|
I
I
I
I
i I
I Z
1 3
1 4
1 5
1
20Or-
I00
--J L.O
~E
50 40
30
20
tOi
0
TIME Figure
(HOURS)
2 Decay of ethyl biscoumacetate in the plasma of eight normal volunteers after a single intravenous dose of 20 mg/kg [4].
p o p u l a t i o n into compliers a n d noncompliers, sometimes on a biologic basis; that is, does the d r u g work? For example, if a p e r s o n ' s blood pressure drops, he is a complier; if it does not drop, he is a noncomplier. The reasoning is s o m e w h a t circular, because y o u are defining compliance in terms of the positive o u t c o m e that y o u are seeking to evaluate in testing an agent. Compliance has also been defined on a statistical basis. Investigators m a y take the m e d i a n level of compliance and consider people w h o are below it as noncompliers. Or, an arbitrary level m a y be chosen for a cutoff, such as 30 or 75 percent. Usually, we do not have a biologic basis for selecting a clear cutoff point for compliance and noncompliance. If we are going to use a marker and w e w a n t to k n o w what the marker is for, and if we are going to d e t e r m i n e compliance or noncompliance, we must k n o w precisely w h a t we m e a n by "compliance" and "noncompliance." We also can look at compliance as a continuous variable, which is w h a t it is. H o w e v e r , in clinical trials, this approach generally does not help m u c h because we usually w a n t to characterize a given individual as a compiler or a noncomplier. A n o t h e r problem is that t h e r e can be considerable intraindividual variation over time. Table 1 shows the coefficients of variation for a n u m b e r of a g e n t s
485
Use of Markers in Clinical Trials Table 1
Summary of Interindividual and Intraindividual Coefficients Variation for Drug Metabolism in Man Interindividual Intraindividual coefficient of coefficient of Parameter variation (a) (%) variation (b) (%) Phenylbutazone t 1/2 17.29 8.66 Antipyrine t 1/2 24.29 13.49 Phenacetin t 1/2 4.40 19.50 Phenacetin AUC 54.93 58.81
of
b/a
0.50 0.56 4.43 1.07
[6]. For phenacetin's half-life, the intraindividual coefficient of variation is very high. This raises serious questions when you take samples at different points in time. What kind of intraindividual variations would you anticipate might occur even if the patient is complying? Another important issue is "percent compliance." For example, Dr. Insull mentioned 65% compliance. We must be clear about what we mean by 65% compliance or 50% compliance or whatever. Figure 3 shows four hypothetical patients w h o are 50% compliers. Each patient had 10 urine tests done for a drug, and all are 50% compliers. One person was a noncomplier and something happened to make that person comply. Perhaps the adherence counselor got to him, or something changed in this person's life relating to illness behavior. The second subject changed from compliance to noncompliance. The third is a hypothetical example of a patient who alternates doses. The fourth subject is more of a haphazard distribution of pluses and minuses. The therapeutic and preventive implications are quite different here. If the first or second patient is on penicillin prophylaxis, that patient may be unprotected for half the time. On the other hand, the third patient may be adequately protected by having alternated doses. Usually we do not have the biologic information in terms of h o w much the patient can miss and still be protected and get a therapeutic effect. Therefore, it is not sufficient to say that a patient is a 50% complier. The pattern of noncompliance must be described. Figure 3 Hypothetical results of sequential tests in four 50% compliers. .
.
.
.
+
+
+
l
.
Patient
2
+
+
+
+
+
.
Patient
3
+
-
+
-
+
-
+
-
+
-
Patient
4
+
+
-
-
-
+
+
-
+
-
.
+
+
Patient
.
.
.
486
Leon Gordis When we measure compliance we are essentially sampling. Let us say these tests represent urine tests. Are we sampling every day of the week? Are we selecting days at random, from Sunday through Saturday, or from Monday through Friday? What is the sampling frame from which we are drawing those tests? Several other issues should be looked at regarding compliance testing: Where were the urine specimens obtained (home, school, work, etc.)? Did the patient know the reason for the tests? Was the patient characterized on the basis of one test, or was testing done periodically? Was the schedule fixed, variable, or random, and did the patient know the schedule? Aside from the issue of sampling, what about medications that are prescribed tWice daily or thrice daily? What should be the sampling schema for measuring compliance? In our experience with school children on twice daily medication, for example, we found we could get a morning specimen at school, but it was very hard to come into a home and ask for the evening specimen. This is a tremendous logistic problem in measuring compliance in clinical trials that involve multiple doses per day. Another important issue is whether reducing the frequency of medication will improve patient compliance. Is one accomplishing something positive for the patient by reducing the frequency? Consider the situation where a patient is on tWice-a-day medication and then gets changed to once-a-day medication. Let us assume that the patient is a 50% complier. If he remains a 50% complier after the frequency of the medication is reduced, we have a potential for serious trouble. Since he is only getting medication once a day and is a 50% complier, he only takes the medication every alternate day. Therefore, if we reduce the frequency of doses to enhance the patient's welfare, there is an assumption not only that compliance is not getting worse, but that it is also not staying the same, but is getting at least twice as good to break even. This issue has not been investigated adequately, partly because the problem of monitoring compliance in multiple-dose regimens, such as getting the evening specimen. I would like to raise the question of whether we should be concerned with the predictive value of the test in clinical trials. What is the probability that a patient has taken the medication given a certain test result when compliance is monitored? This raises a host of other methodologic issues: In a clinical trial, is the central question regarding markers one of sensitivity and specificity, or is it the predictive value of the marker? Finally, ! would like to ask whether it is important that the patient remain unaware of the marker. If it is important that the patient remain unaware, it raises ethical issues in carrying out the study because of the problem of informed consent. Certain human volunteers committees do not want patients' urines examined to see whether they have taken the medication, unless the patients have been adequately informed of the reason for the urine test being done. This is a potentially serious problem.
Use of Markers in Clinical Trials
487
REFERENCES 1. 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 2. 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 3. Vestal RE, Norris AH, Tobin JD, et al: Antipyrine metabolism in man: Influence of age, alcohol, caffeine, and smoking. Clin Pharmacol Ther 18:425--432, 1975 4. Vesell ES: Factors causing interindividual variations of drug concentrations in blood. Clin Pharmacol Ther 16:135-148, 1974 5. Brodie BB, Weiner M, Burns JJ, et al: The physiological disposition of ethyl biscoumacetate (tromexan) in man and a method for its estimation in biological material. J Pharmacol Exp Ther 106:453463, 1952 6. Alvares AP, Kappas A, Eiseman JL, et al: Intraindividual variation in drug disposition. Clin Pharmacol Ther 26:407--419, 1979