Drug Compliance and the Neglected Concern For Validity

F. Robert Dwyer, University of Cincinnati
ABSTRACT - A brief review of the drug compliance literature is presented in order to frame this and perhaps other primary research efforts. A clinical study is outlined and two linear models are used to test the previously examined factors of compliance in conjunction with an untried variable, health locus of control. Third, the jackknife validation procedure is described and demonstrated to highlight its particular relevance in compliance and other sample-constrained research settings.
[ to cite ]:
F. Robert Dwyer (1979) ,"Drug Compliance and the Neglected Concern For Validity", in NA - Advances in Consumer Research Volume 06, eds. William L. Wilkie, Ann Abor, MI : Association for Consumer Research, Pages: 564-568.

Advances in Consumer Research Volume 6, 1979      Pages 564-568


F. Robert Dwyer, University of Cincinnati


A brief review of the drug compliance literature is presented in order to frame this and perhaps other primary research efforts. A clinical study is outlined and two linear models are used to test the previously examined factors of compliance in conjunction with an untried variable, health locus of control. Third, the jackknife validation procedure is described and demonstrated to highlight its particular relevance in compliance and other sample-constrained research settings.


There have been nearly 70 studies on patients who fail (intentionally or accidentally) to take their medications "as directed." In these studies, a variety of methods have been used to detect the drug defaulter: (1) interrogation, (2) pill counts, (3) stool and urine markers, (4) drug detection in blood or urine, and (5) failure to dispense or refill the prescription (Blackwell, 1972). Doctor intuition in detecting drug defaulters has been very unreliable; error rates are slightly better than chance, and is therefore seldom used in studies of compliance (Caron and Roth, 1968).

Boyd et al. (1974) have reviewed 35 patient compliance studies and have found defaulting rates from 4% to 90%, depending on the drug, patient group and method of assessment. On the latter point these authors computed weighted averages of noncompliance for the three primary methods of detection: urine tests, dosage counts, and patient self reports. While the mean rates of defaulting were comparable: 36%, 47%, and 38%, respectively, for each type of measure there is no indication of the convergent validity of the measures.

The consequences of overdoses, omissions, and improper mixing of prescription drugs involve varying health hazards for the defaulter. Patients who suffer from arthritis and default from anti-inflammatory drugs are at worst trading-off their own pain against the inconveniences of continuous medication. In a curative situation, a study by Leistyna and MacAuley (1966) showed over 75% of the patients defaulting from the prescribed number of doses of oral penicillin for their streptococcal pharyngitis had positive throat cultures. All compliant patients showed negative cultures.

If a physician is unaware that a hypertensive patient is not taking the medication according to directions and sees that the blood pressure is still elevated, the physician may prescribe larger doses of the same agent(s) or may prescribe a more potent anti-hypertensive medication (Hussar, 1975). On a similar note, patient noncompliance may play a part in distorting the controlled evaluation of drug therapy.

In addition to depriving the patient of the anticipated therapeutic benefits and possibly resulting in a progressive worsening of the condition being treated, the consequences of drug defaulting include the economic wastage involved and the hazard to health posed by cupboards stocked with unused or unidentified medications (Blackwell, 1972).

Factors in Noncompliance

The extent, nature, and consequences of drug defaulting have highlighted its significance from both a patient and societal perspective. Logically, numerous studies have attempted to uncover patient reasons and circumstantial factors associated with drug defaulting. While the findings have been somewhat inconclusive on the demographic and psycho-social correlates of compliance, conclusions on the impact of certain medication and treatment factors appear more consistent. Even on this latter point, caution is required since findings of no statistical significance on tested variables are often not reported and in no case to date have researchers validated their method or findings on a new set of data.

Treatment Factors

After an extensive literature review, Blackwell (1973) concludes:

Patients with prolonged conditions are clearly prone to lapses in compliance, especially when the treatment is prophylactic or suppressive (as with malaria), when the condition is mild or asymptomatic (in anemia of pregnancy) or when the consequences of stopping therapy may be delayed (for instance, as in epilepsy or schizophrenia). (p. 241)

The literature is replete with studies showing discontinuation of oral antibiotics because patients "feel better," yet the infection may not be under complete control.

Brook and associates (1971) showed that when noncompliance is likely to lead to an immediate and/or severe relapse, the patient is more likely to comply. The cardiac patients in their study were more adherent to their regimen of digoxin and diuretics than they were to their potassium supplements.

The effect of multiple medications on patient compliance has been widely studied. Clinite and Kabat (1969), Madden (1973), Kellaway and McCrae (1975), and Malahy (1966) found that compliance decreased among patients taking three or more medications. Noncompliance is also encouraged by multiple doses of the same drug.

The magnitude of this effect was demonstrated by Gatley (1968) when the number of defaulters (in a prospective trial) doubled as the number of tablets was increased from one to four per day.

It seems logical that a patient who experiences side effects is less likely to comply than the patient with no side effects. This suspicion is confirmed by a number of studies. Wynn-Williams and Artis (1958) found that PAS (Para-amino-salic acid) has produced gastrointestinal upset in many more defaulters than in non-defaulters. Hussar (1975) reports that a number of observers have noted that the development of impotence is a major reason for male patients to discontinue taking anti-psychotic medications.

A treatment dimension related to complexity, yet requiring special attention, is life style change mandates that can accompany prescription medications. For example, it is desirable for patients being treated with tranquilizers or sedatives to avoid excessive consumption of alcoholic beverages because of the increased likelihood of an excessive depressant effect. Many patients choose to cope with this mandate by simply not taking their drug (Hussar, 1975).

Patient Characteristics

Studies which have explored the psycho-social and demographic correlates of noncompliance have generally shown mixed results. However, no comprehensive studies of the entire patient population have ever been conducted. Therefore, research conclusions are constrained by the limited age range, economic status range, etc., of the accessible geographic, disease subpopulation studied.

Bakker and Dightman (1964) administered a battery of personality tests to women taking oral contraceptives. Those women who failed to take the Pill regularly were found to be more immature, irresponsible, and impulsive than their compliant counterparts. In addition, their personality profiles deviated more from their husband's than did those profiles of women who took the pill regularly. It is likely that some other personality dimensions would be associated with noncompliance for other medications.

Blackwell (1972) reports that some psychoanalytic writers have suggested that pills and capsules are symbolic representatives of the breast and penis. Little evidence in support of this notion is reported, however. More common in the psychiatrist's practice are patient paranoid delusions that cause the schizophrenic to equate drugs with poison. Also, Hussar (1975) reports an increasing number of patients concerned about becoming drug-dependent. To avoid such a possibility and to prove to themselves that they are not dependent, they may interrupt or by some other means modify their treatment regimen.

As might be expected, the patient-physician relationship can affect compliance. Francis et al. (1969) examined parent attitudes toward the physician in compliance of pediatric outpatients and found that compliance was reduced when the mother perceived the doctor as unfriendly and insensitive to the child's condition or illness.

A handful of studies have shown positive correlations between compliance and the patient's perceptions of susceptibility to disease, seriousness of the condition and efficacy of the treatment (Sharpe, 1977). In an attempt to explain preventive health behaviors, these factors, plus a patient's analysis of the "benefits and barriers" of treatment and a "cue to action," have been formulated by Rosenstock (1966) into what has been labeled the health belief model. Its performance has been mixed in the preventive setting (Becker et al., 1972; Oliver and Berger, 1978) while in a curative setting only its individual components have shown occasional significance (Sacket and Haynes, 1976).

Moving away from psychological correlates, Boyd et al. (1974) after collecting data on 380 prescriptions from 134 outpatients, made several conclusions on the effects of demographic variables -- sex, age, race, education, occupation, and social class -- on compliance. No differences in error rates were noted between races, sexes, education levels and the Hollingshead Index-defined social classes. Differences were identified by age groups, however. The 45-64 age group had the least number of errors per prescription. The authors' explanations of the age effects were:

1. Patients in the 25-44 age group are preoccupied in the working world and are not acutely health conscious.

2. The 45-64 age group is becoming more aware of their health while concurrently decreasing their other activities.

3. The 65 and over group is experiencing more complex health problems combined with decreasing ability to care for themselves.

Malahy's (1966) analysis of demographic effects was narrower than Boyd's. While the studies are consistent in their rejection of education effects, Malahy went on to conclude no age effects on compliance. The apparent discrepancy seems to be due to the limited range of Malahy's sample -- most patients were between the ages of 35 and 65.

Clinite and Kabat's (1969) study of 1,060 patients, however, suggests the question of age effects cannot be put aside. They noted that patients between 71 and 80 years had by far the lowest rate of noncompliance. It is possible that even Boyd's sample (n = 134) was not sufficiently broad to negate the variance impact of previously mentioned treatment factors associated with noncompliance and likely to covary with age. Examples include increased seriousness of disease and consequences of defaulting, multiple medications and complex treatments, and advancing senility.

Madden (1973) joins Malahy and Boyd et al. in finding an absence of education effects on compliance with antibiotic regimens. However, other studies have reported higher education levels for defaulters among tuberculosis patients and lower education standards for defaulters among pregnant women taking iron (Berry et al., 1963; Porter, 1969).

Income seems to be directly related to compliance, as several studies have shown that for some families the money is not always available to get the prescription filled or refilled (Gurwich and Emmanuel, 1974).

In their study of the compliance rates of mental outpatients, Lipman et al. (1965) found black patients less compliant than whites. A later study of psychiatric patients by Hare and Wilcox (1967) found no difference in compliance between blacks and whites. Adding to the confusion of the effects of subculture is a study by Kellaway and McCrae (1975) in New Zealand. They found Polynesian patients more likely to make simple errors -- unintentional mistakes -- than European patients. On the other hand, European patients showed a greater tendency for noncompliance involving a positive decision with at least some degree of volitional intent to alter or interfere with the prescribed regimen.

It is logical to expect patients in hospitals to show greater compliance rates than out-patients. This assumption was verified by Hare and Wilcox who checked urine marketers of in-patients, day-patients, and outpatients. Employing moderate criteria of interpretation of urine analysis, noncompliance rates were 19%, 39% and 48% for inpatients, day, and outpatients, respectively.

The effect of supervision of drug therapy may be reflected in the reported effects of marital status on compliance. A study of 220 chronically ill patients over 60 conducted by Schwartz et al. (1962) found married persons more compliant than patients living alone. Porter (1969) studied a broader population of patients on long term treatments, but also concluded that socially isolated patients tended to neglect their drugs.

The effects of a patient's previous experience with the same or a similar disease is unclear. In Ball's (1974) review of several studies he concludes experience has a positive impact on compliance. However, Lipman, et al. found experience to have a negative impact on compliance with meprobamate. In proper perspective, it seems experience would interact with the nature of the treatment and previous outcome on compliance.

Comprehension of Regimen

In addition to all these factors of compliance, Blackwell (1973) concluded on the basis of his own empirical evidence:

The most important contribution to compliance is the understanding a patient has of the illness, the need for treatment and the likely consequences of both. (p. 252)

Moreover, this point of view has been supported by recent research (Clinite and Kabat, 1976; Cole and Emmanuel, 1971; and Madden, 1973). The issue is not so clear-cut, though, as Malahy (1966) and Sackett et al. (1975) failed to find compliance effects from increased comprehension of treatment and illness.

Comprehension effects on compliance are of particular interest to policy makers as written, patient-oriented information may soon be dispensed with all or most prescription drugs.


Two basic administrative problems have confounded researchers in the compliance area. First, capturing an adequate sample size is constrained by the realm of experimenter's control and subject arrival rates. Due to the convenience of using the single researcher-affiliated health care site and/or coordination problems arising from multifacility studies, the former approach is most widely implemented. The single site constraint has a striking effect on the patient population studied. It means the researchers must study compliance in a patient population undergoing a variety of treatment programs or must study compliance in patients with a prevalent condition. The former thrust increases both the total variance and the proportion of variance from disease and treatment factors. The latter usually means a lengthy data collection period and a focus limited to such prevalent conditions as strep throat and hypertension.

The above-mentioned administrative problems tend to magnify the seriousness of analysis difficulties encountered. Typically computerized data searching techniques, stepwise regression or discriminant analysis, are utilized. These procedures are useful for exploratory research. However, because of their tendency to capitalize on spurious correlations in the data, they often uncover "significant" relationships (that often have intuitive appeal) that are, in fact, random. Before such findings can be labeled conclusive they must be tested on a new set of data.

This call for validation merely echoes the cries many consumer researchers have previously sounded (Sheth, 1977; Etgar, 1977). The most frequently recommended and utilized approach has been the holdout method. Here the data are randomly split; one half is used for model estimation, the other tests its stability or predictive validity. This procedure is infeasible, however, when costs and administrative constraints preclude an adequate sample size. Statistics from each data set may be biased in proportion to 1/n and therefore quite large when the sample size is small.

The following section describes a more efficient validation method, the jackknife, and demonstrates its viability in a study of compliance.


Nearly 20 years ago Quenouille (1956) and Tukey (1958) proposed a methodology for reducing estimation bias in small samples. The technique, labeled the jackknife by Tukey for its versatility as a statistical tool, offers ways to set sensible confidence limits in complex situations. An excellent detailed discussion of the technique is provided in Mosteller and Tukey (1977) and Miller (1974).

Briefly, the crux of the jackknife is to divide the data into g groups of size h, n = gh, and assess the effect of each of the groups by examining the effect on the sample statistics that results from omitting that group.

This is accomplished by making Y(j) the results of the complex calculation on the portion of the sample that omits the jth subgroup; estimates are made on a pool of (g - 1) subgroups. By subtracting Y(j) times (g - 1) from g times Yall, the corresponding result for the entire sample pseudo-values Yj* are obtained. Equivalently:

Yj* = g.Yall - (g-1) Y(j)     j=1,2,3...g   (1)

The pseudo-values can be used to set approximate confidence limits, using Students' t, as if they were independent, identically distributed random variables. The jackknife statistics, Y*, and an estimate of its variances, S*2 are given by:

EQUATIONS   (2)   -  (4)


Data were obtained from 49 newly diagnosed essential hypertensives, prescribed diuretics. Patients from a Midwest and two Southwest medical centers were interviewed three to five weeks after they received their prescriptions. This questionnaire obtained patients' self reports of compliance, experience with side effects, attitude toward the prescribing physician and demographics. In addition, a 20-item measure of comprehension of drug regimen and a health locus of control measure (Walleston et al., 1977) were administered. The health locus of control measure has intuitive appeal but is as yet untested in compliance studies. Finally, a behavioral measure of compliance, namely deviation from the hypothetical compliant refill date, was obtained from pharmacy archives.


The data were analyzed with two linear models. First, using variables suggested by the compliance literature, a discriminant function was calculated for classifying compliant and noncompliant patients as they were so indicated in self reports. [Variables in the discriminant function included: self report of side effects (0-1), comprehension, health locus of control, attitude toward physician, sex, some college (0-1), and children at home (0-1).] While the model was able to classify nearly 75% of the cases correctly, none of the variables had significant discriminatory power.

The second model was a regression analysis of the same variables on refill deviance. The dependent measure, number of days noncompliant per 100, was modified to prevent outliers. That is, a ceiling of 35 days was implemented since patients who discontinued early in the data collection period had a head start toward infinity deviance. The results of this analysis showed that only the health locus of control measure had significant (p < .1) variance reduction capability. Patients scored as internals were more compliant than their external counterparts.

Still, this apparently significant factor of compliance may be attributable to overdetermination or extensive searching of the data set. It represents a heretofore unnoticed factor of compliance that could add to (the confusion in) the literature. Before this insight receives acclaim it must meet the test of validation.

Validation via the Jackknife

The relationship between compliance and health locus of control was tested for stability by the approach previously outlined. More specifically, the data were divided into 41 groups [Only 41 measures of refill deviance were obtained as some patients still had had no need to refill their initial prescriptions.] of size h = 1 and the regression model was estimated 41 times. The n regression runs allowed calculations of pseudo-values and jackknife statistics for the regression coefficient of the health locus of control variable.

The results of the jackknife analysis are shown in Table 1. The coefficients' jackknife statistics, .045, divided by its corresponding standard deviation, .033, provides a t-statistic of 1.38. This is significant at the .2 level only. Hence, the health locus of control measure does not offer significant variance reduction in compliance.




The preceding pages have served two worthy ends. First, the review of the compliance literature has provided a springboard for other consumer researchers to enter the area. Product liability issues and the social costs of noncompliance are just two factors sparking the interests of pharmaceutical and policy makers in this field of consumer behavior (Weintraub, 1975).

Future efforts will be more meaningful if conclusions can be validated. Toward this end, the administrative and cost factors which restrict sample sizes in the compliance setting also render traditional validation methods infeasible. This paper, however, outlines and demonstrates a jackknife procedure for more efficiently validating small sample conclusions.

Although the usefulness of the jackknife was demonstrated only on simple regression coefficients, its applications are extensive. Straightforward extensions to discriminant functions (Crask and Perreault, 1977) and even analysis of variance (Dwyer, 1978) are possible.


C. B. Bakker and C. R. Dightman, "Psychological Factors in Fertility Control," Fertility and Sterility, 15 (1964), 559-67.

W. L. Ball, "Improving Patient Compliance with Therapeutic Regimens: Hamilton Symposium Examines the Problems and Solutions," Canadian Medical Association Journal, III (August 3, 1974), 268 +.

M. H. Becker, R. H. Drachman, and J. P. Kirscht, "Predicting Mothers Compliance with Pediatric Medical Regimens," Medical Care, 81 (1972), 843-54.

D. Berry, A. Ross, and K. Deuschle, "Tuberculosis Patients Treated at Home," American Review of Respiratory Diseases, 88 (1963), 769-72.

Barry Blackwell, "The Drug Defaulter," Clinical Pharmacology and Therapeutics, 13, 6 (November - December, 1972), 841-848.

Barry Blackwell, "Patient Compliance," The New England Journal of Medicine, 289 (August 2, 1973), 249-53.

James R. Boyd, Tim R. Covington, Walter F. Stanaszek and R. Timothy Coussons, "Drug Defaulting. Part I; Determinants of Compliance," American Journal of Hospital Pharmacy, 31 (April 1974), 362-67.

James R. Boyd, Tim R. Covington, Walter F. Stanaszek and R. Timothy Coussons, "Drug Defaulting. Part II: Analysis of Non-Compliance Patterns," American Journal of Hospital Pharmacy, 31 (May 1974), 486-91.

R. H. Brook, F. A. Appel, and C. Avery, "Effectiveness of Inpatient Followup Care," New England Journal of Medicine, 285 (1971), 1509-14.

H. S. Caron, and H. P. Roth, "Patients' Cooperation with a Medical Regimen: Difficulties in Identifying the NonCooperator," Journal of the American Medical Association, 203 (1968), 922-26.

James C. Clinite and Hugh F. Kabat, "Prescribed Drugs ... Errors During Self Administration," Journal of the American Pharmaceutical Association, NS9, 9(September 1969), 430-52.

Melvin R. Crask and William D. Perreault, Jr., "Validation of Discriminant Analysis in Marketing Research," Journal of Marketing Research, 14 (February 1977), 60-68.

F. Robert Dwyer, "An Experimental Study of Patient Use of Supplementary Drug Information," unpublished Ph.D. thesis, University of Minnesota, Minneapolis, 1978.

Michael Etgar, "Channel Environment and Channel Leadership,'' Journal of Marketing Research, 14 (February 1977), 69-76.

V. Francis , B. M. Korsch, and M. J. Morris, "Gaps in Doctor-Patient Communications," New England Journal of Medicine, 280 (1969), 535-40.

M. S. Gatley, "To Be Taken as Directed," Journal of the Royal College of General Practitioners, 16 (1968), 39-44.

Ernest Gurwich and Sister Emmanuel, "A Comprehensive Health Care Program," Journal of the American Pharmaceutical Association, NS14, 2 (February 1974) 71-79.

E. H. Hare and D. C. Wilcox, "Do Psychiatric In-Patients Take Their Pills," British Journal of Psychiatry, 113 (November 1967), 1435-39.

Daniel A. Hussar, "Patient Noncompliance," Journal of the American Pharmaceutical Association, NS15, 4 (April 1975), 183-90 +.

G. S. M. Kellaway and Ellen McCrae, "Non-Compliance and Errors of Drug Administration in Patients Discharged from Acute medical Wards," New Zealand Medical Journal, 81 (June 11, 1975), 508-12.

J. A. Leistyna and J. C. MacAuley, "Therapy of Streptococcal Infections: Do Pediatric Patients Receive Prescribed Oral Medication?" American Journal of Diseases and Children, 111 (1966).

R. S. Lipman, K. Richels, E. H. Ehlenhuth, L. C. Park and S. Fischer, "Neurotics Who Fail to Take Their Drugs," British Journal of Psychiatry, III (September 1965), 1043-49.

Edward E. Madden, Jr., "Evaluation of Outpatient Pharmacy Patient Counseling," Journal of the American Pharmaceutical Association, NS13, 8 (August 1973), 437-43.

Bernadine Malahy, "The Effect of Instruction and Labeling on the Number of Medication Errors Made by Patients at Home," American Journal of Hospital Pharmacy, 23 (June 1966), 283-92.

Rupert Miller, "The Jackknife -- A Review," Biometrika, 61 (1974), 1-15.

Frederick Mosteller and John W. Tukey, Data Analysis and Regression. Reading, Massachusetts: Addison-Wesley, 1977.

Richard L. Oliver and Philip K. Berger, "Testing Competing Models of Consumer Decision Making in the preventive Health Care Marketplace," in H. Keith Hunt, ed., Advances in Consumer Research, 5 (Ann Arbor: Association for Consumer Research, 1978), 277-82.

A. M. W. Porter, "Drug Defaulting in a General Practice," British Medical Journal, 1 (January 25, 1969), 218-22.

M. H. Quenouille, "Notes on Bias in Estimation," Biometrika, 43 (1956), 353-60.

I. M. Rosenstock, "Why People Use Health Services," Milbank Memorial Fund Quarterly, 44 (July 1966), 94-127.

D. L. Sackett and R. B. Haynes, Compliance with Therapeutic Regimens. (Baltimore: John Hopkins University Press, 1976).

D. L. Sackett, R. B. Haynes, E. S. Gibson, B. C. Hackett, D. W. Roberts, D. W. Taylor, A. L. Johnson, "Randomized Clinical Trial of Strategies for Improving Compliance in Primary Hypertension," Lancet, 1 (1975), 1205-1207.

D. Schwartz, M. Wang, I. Leitz, and M. E. W. Gross, "Medication Errors Made by Chronically Ill Patients," American Journal of Public Health, 52 (November 1962), 2015-19.

Thomas R. Sharpe, "Potential Effects on the Patient," Drug Information Journal, 11 (January 1977), 58-64.

Thomas R. Sharpe and Robert L. Mikeal, "Patient Compliance with Prescription Medication Regimens," Journal of the American Pharmaceutical Association, NS15, 4 (April 1975), 191-92 +.

Jagdish Sheth, "What is Multivariate Analysis," in Jagdish Sheth, ed., Multivariate Methods for Market and Survey Research, Chicago: American Marketing Association, 1977.

John W. Tukey, "Bias and Confidence in Not-Quite Large Samples," Annals of Mathematical Statistics, 29 (1958), 614.

Barbara S. Wallston, Kenneth A. Wallston, Gordon D. Kaplan, and Shirley A Maides, "Development and Validation of the Health Locus of Control (HLC) Scale," Nashville, Tennessee: Vanderbilt University, School of Nursing, 1975 (mimeo).

Michael Weintraub, "Promoting Patient Compliance: Role of Health Professionals, Government, and the Pharmaceutical Industry," New York State Journal of Medicine (October 1975), 2263-66.