A 28-year-old woman presents to the emergency department in acute distress, with a 3-day history of worsening asthma. Her prescribed medications include an inhaled β2 agonist and an inhaled steroid. When questioned, she breathlessly admits to “occasionally” missing her medications “maybe only once or twice.”
A 55-year-old man with posttraumatic seizure disorder has been taking phenytoin since his injury. His seizures were initially adequately controlled but he recently has been having weekly seizures. In an office visit he resentfully denies missing any of his medication.
The Importance of Clinical Examination
Physicians should measure compliance for patients prescribed a self-administered treatment because noncompliance is common and physicians can help patients improve their compliance1, 2 and increase the benefit they derive from therapy. Compliance with long-term self-administered medication therapy is approximately 50% for those who remain in care.3 There is a wide range of compliance among patients, from 0% to 100%. This average compliance rate of 50% provides only the most limited picture of compliance; in general, there is substantial undercompliance. Furthermore, compliance by individuals can vary considerably over time. Compliance rates for short-term self-administered therapies average about 75% initially but decrease to less than 25% for the completion of antibiotic therapy for acute infections. Aside from its potential for undermining the effectiveness of any treatment, noncompliance is associated with poorer prognosis.4
Table 15-1 depicts combinations of treatment outcome and compliance that present in clinical practice and need to be distinguished from one another to initiate the appropriate intervention. The bottom right cell, D, represents the most desirable state: high compliance with achievement of the treatment goal. Patients who fall into cell A (low compliance and suboptimal achievement of the treatment goal) are in need of efforts to promote compliance. Patients in cell B (high compliance without achieving the treatment goal) require more or better treatment, whereas those in cell C (achievement of the treatment goal despite low compliance) need less treatment prescribed or may actually have been misdiagnosed or mistreated and do not merit intervention to increase compliance, at least until the need for treatment is reassessed. The aim of compliance assessment, along with other diagnostic tests, is to categorize patients into the appropriate cells. When it has been determined that patients occupy cell A, B, or C, physicians may then alter treatment to attempt to move patients into cell D.
Table 15-1Complications and Treatment Responsea |Favorite Table|Download (.pdf) Table 15-1 Complications and Treatment Responsea
|Achievement of the Treatment Goal ||Compliance Rate |
| ||Low ||High |
|No ||A: The target group ||B: Inadequate therapy? |
|Yes ||C: Unnecessary therapy? ||D: Ideal |
The 2 cases illustrate the importance of distinguishing between noncompliance and lack of therapeutic efficacy. On closer questioning, the first patient revealed that she had been using her inhaled steroid sporadically and gradually lost control of her asthma without change in extrinsic allergic stimuli. She initially provided 2 useful clinical clues to important deviation from her prescribed regimen: worsening symptoms while prescribed usually adequate therapy and admission of “occasional” noncompliance. Reinstituting her usual regimen and reinforcing the need for compliance, particularly with the inhaled steroid, improved her treatment results.
Case 2 required a different solution: the phenytoin levels eventually proved to be in the therapeutic range, confirming the patient's compliance, and a second medication was required to improve long-term control. Accurate assessment of compliance by questioning could prevent overdosing the patient with additional therapy on the assumption of noncompliance and permit timely addition of the second therapy.
Although maintaining an adequate level of compliance is central to deriving benefit from any efficacious therapy, the degree of compliance necessary to achieve a measurable benefit from specific medications is variable. Haynes et al5 found that 80% compliance was necessary to achieve a reduction in blood pressure from antihypertensive therapy with the types and doses of medication that were prescribed by primary care physicians, whereas Markowitz6 reported that children receiving only a third of their prescribed penicillin had substantial protection from recurrences of rheumatic fever. The thresholds of compliance for acceptable therapeutic effects are not known for most regimens.
The Nature of Noncompliance
On a practical level, it is not difficult to imagine why noncompliant behavior occurs. Patients often find medical regimens complicated, inconvenient, embarrassing, or expensive. Particularly for chronic disorders, the short-term disadvantages frequently outweigh the long-term advantages.
At a theoretic level, the nature and determinants of noncompliant behavior are more complex and not well understood, although there are interesting models.7 Numerous studies of the “determinants” of compliance have led to the following generalizations.8 Sociodemographic factors such as age, sex, race, intelligence, and education have little to do with compliance. Low compliance is a problem with self-administered treatments for all disorders, but patients with psychiatric problems are less likely to comply, and those with (other) disabilities caused by disease are more likely to comply. Long waiting times at clinics and long gaps between appointments lead to patients’ missing appointments and dropping out of care. The more complex or costly the regimen and the longer its duration, the less the compliance.
Most studies determining the limitations and strengths of clinical information about compliance include pill counts and measurement of serum levels of drugs or tracers. Special medication monitors can also reveal patterns of medication consumption that cannot be obtained by other means. None of these more accurate methods is likely to be handy to practitioners for most therapeutic regimens.
No clinical measurement of compliance approaches perfection, but clinical information can be used to narrow down the situations in which compliance measurement is most likely to be important for the care of the patient. A 3-step sequence will identify most noncompliers.
Nonattendance at appointments. Dropout rates are high with many treatments, and nonattendance at a scheduled appointment is the first step astray.
Lack (or loss) of responsiveness to a usually (or previously) adequate dose of treatment. These patients are most in need of further assessment of their compliance to separate problems of therapy from those of compliance. Cases 1 and 2 would qualify for this route.
For patients whose compliance is in doubt, particularly those who come to attention through steps 1 and 2, use the most appropriate method(s) from Table 15-2. Direct measures of medication consumption are most accurate, but they are available for only a small number of medications, can indicate spuriously high compliance if the patient takes the prescribed dose only during the time leading up to assessments of drug levels, and do not apply to most nonmedication regimens, such as weight-loss diets. Even when available, they take time and money to obtain and are unlikely to help in such situations as the acute care of a patient in the midst of a crisis.
Table 15-2Methods to Measure Compliance |Favorite Table|Download (.pdf) Table 15-2 Methods to Measure Compliance
|Drug or metabolite levels |
|Tracer compounds |
|Appointment keeping |
|Therapeutic response |
|Pill count |
|Pharmacy records |
|Medication event monitors |
Questioning of patients is the most widely applicable method of measuring compliance. Careful questioning will identify more than half of those who are noncompliant without falsely labeling many of the compliers. Patients should be asked to indicate, without prompting, exactly what medications they are taking and when they are taking them. This may reveal a different understanding of and adherence to the regimen than was prescribed. For patients who report a generally correct understanding of their prescription, the details of any noncompliance should be sought. The method of asking likely affects the accuracy of the response. Studies assessing the value of patient self-report have used a nonjudgmental, nonthreatening approach, prefacing the question with a remark such as the following9: “People often have difficulty taking their pills for one reason or another.” The question also must be asked in a particular way: “Have you ever missed any of your pills?” If the answer is affirmative, then ask the patient to estimate how many pills he or she has missed during the previous day and week. The interview can also provide insight into the possible reasons for noncompliance. This valuable clinical information can allow prompt reevaluation of the current regimen if the information is interpreted appropriately. It is essential to take into account that even if the patient admits to missing any medication during the previous day or week, he or she will still tend to overestimate the actual rate of compliance (by an average of 17% in 1 study9).
Clinical measures of compliance sometimes can be supplemented or replaced by other methods. Some regimens produce telltale adverse effects, the absence of which suggests low compliance; for example, increased urinary frequency with the initiation of diuretics, dry mouth with anticholinergics, slow heart rate with β-blockers, dark stool with oral iron, and suppression of thyrotropin (thyroid-stimulating hormone) with thyroid hormone replacement.10 Blood level measurements are routinely available for some medications, and these can be used for monitoring compliance, particularly when the serum half-life is relatively prolonged.11 When patients receive all their medications through a single pharmacy, pharmacy records can provide an indirect measure of compliance.12 Medication event monitors, although providing unique information about the pattern of medication taking, are expensive and remain a research tool.13, 14 Tracers, either harmless substances such as riboflavin15 or minute amounts of medications such as phenobarbital that can be easily measured,16 are also research tools.
For pill counts, drug and tracer levels, and surreptitious pill monitors, there are ethical issues to be addressed. Because they invade the patient's privacy and can be used to usurp autonomy, when possible you should inform the patient of their intended use and ask for consent before using them.17 Patients usually agree to monitoring if it is explained that the purpose of the assessments is to help better understand how they are taking their medicine.
Accuracy of Clinical Measures of Compliance
Clinical judgment of compliance has been found wanting in almost every study in which it has been tested. Clinicians who believe that they are exceptions to this finding because they know their patients well should take heed of a study by Gilbert et al.18 Primary care physicians were asked to give compliance estimates only for patients they thought they knew well. The sensitivity of clinical judgment for detecting noncompliance was an embarrassing 10%, and overall performance by clinicians was not better than if they had flipped coins instead of applying their “clinical judgment.” Physicians should not trust their unaided judgment regarding the compliance by individual patients.
19 Richardson et al20 confirmed both that attendance does not ensure compliance with medications and that compliance is even worse among nonattenders: of patients keeping more than 60% of their scheduled clinic appointments, 40% were found to be noncompliant with medication by urine metabolite measurement, whereas 95% of patients with lower appointment compliance demonstrated low compliance.
The patient's response to therapy is also only weakly related to compliance for many treatments,9 but it can be useful when combined with other methods. For example, when Inui et al21 treated patients as noncompliant if they either admitted noncompliance or had uncontrolled pressures, this combined compliance test had a sensitivity of 83% and a specificity of 66%.
Questioning patients about their compliance is the most readily available, valid method of measuring compliance in clinical practice. To review the literature on self-report, we used previously published guidelines for collecting studies and preparing meta-analyses.22 We identified studies comparing self-report with other measures of compliance and uncovered many studies comparing self-report with pill counts. The 4 studies with the strongest research methods9, 13, 18, 23 are summarized in Table 15-3. The results of compliance tests were considered positive if they uncovered noncompliance and negative if they verified compliance. In these studies, self-report yielded a sensitivity of 55%, a specificity averaging 87%, and a likelihood ratio for a positive test result of 4.4 on average. Patients’ reports of compliance with medication were less useful because the patients still may have been noncompliant (likelihood ratio for a negative test result, 0.5). In one study,9 self-report outperformed several direct and indirect measures of compliance. Similarly, when Fletcher et al24 compared the usefulness of interview, pill count, and measurement of serum drug levels of digoxin, they found that interviewing was the most useful method. Unfortunately, to our knowledge there are no studies to date assessing the agreement among clinicians on eliciting compliance information from patients in usual settings or of the effect on self-reports of repeatedly questioning patients about their compliance.
Table 15-3Pooled Data From Methodologically Strong Studies Comparing Pill Count With Self-report9, 13, 18, 23 |Favorite Table|Download (.pdf) Table 15-3 Pooled Data From Methodologically Strong Studies Comparing Pill Count With Self-report9, 13, 18, 23
|Self-report ||Pill Count |
| ||Noncomplianta ||Compliant |
|Missed ≥ 1 ||152 ||34 |
|None missed ||122 ||232 |
Counting the patient's pills is valid for single assessments at the patient's home if the purpose of the visit is not revealed in advance and if care is taken to determine the amount of medication that has been dispensed, the date the most recent prescription refill was begun, how much was left over from the previous prescription when the current prescription was begun, whether there has been any change in the prescription not noted on the pill container, and whether the patient has caches of pills in other locations or has shared them with relatives or friends.9 When all factors are taken into account, the pill count compares favorably with serum drug levels.18 However, pill counts of this rigor are impractical in most clinical settings, and pill counts performed on medications patients bring with them to clinic visits overrepresent compliance when compared with more tamper-proof methods, such as special pill containers that electronically monitor each dose as it is removed.13, 14 The latter devices also show patterns of compliance that cannot be detected by simple pills counts, including increased compliance just before and after appointments and decreasing compliance between appointments.
Although the absence of common adverse effects may be an indication of noncompliance, the link between compliance and adverse effects is either unknown or relatively tenuous. For example, for patients prescribed diuretics, the sensitivity for noncompliance of reductions in serum potassium level was 82% but the specificity was only 48%.9
You can detect most noncompliant patients by watching for nonattenders, watching for nonresponders, and asking nonresponders about their compliance. In addition to clarifying problems of undertreatment and overtreatment, information about patient compliance permits the efficient application of effective methods of increasing compliance.1, 2
Author Affiliations at the Time of the Original Publication
From the Design, Measurement, and Evaluation Program (Drs Rowe, Haynes, Macharia, and Leon and Ms Stephenson) and the Health Information Research Unit (Dr Haynes), McMaster University, Hamilton, Ontario, Canada. Ms Stephenson is now with the Department of Clinical Epidemiology and Biostatistics, McMaster University, Hamilton, Ontario, Canada; Dr Rowe is now with the University of Ottawa, Northeastern Family Medicine Program, Sudbury, Ontario, Canada; Dr Macharia is now with the Department of Pediatrics, Faculty of Medicine, University of Nairobi, Kenya; Dr Leon is now with the Department of Dermatology, Faculty of Medicine, National Autonomous University of Mexico, Mexico City.
This study was supported by the Brian C. Decker Health Informatics Research Fund, Hamilton, Ontario, Canada; a Health Scientist Award from the National Health Research and Development Program, Canada (Dr Haynes); and clinical epidemiology fellowships from the International Clinical Epidemiology Network Program, Philadelphia, Pennsylvania (Drs Macharia and Leon).
P. Helping patients follow the treatments you prescribe. In: Sackett
P, eds. Clinical Epidemiology: A Basic Science for Clinical Medicine. 2nd ed. Boston, MA: Little Brown & Co; 1991:249–281.
JC. The magnitude of compliance
and noncompliance. In: Haynes
DL, eds. Compliance in Health Care.
Baltimore, MD: Johns Hopkins University Press; 1979:11–22.
R. Patient compliance
and the design and interpretation of clinical trials. Control Clin Trials.
D. Process versus outcome in hypertension: a positive result. Circulation.
M. Eradication of rheumatic fever: an unfulfilled hope. Circulation.
L. Behavioral theories and the problem of compliance
. Patient Educ Couns.
RB. Determinants of compliance
: the disease and the mechanics of treatment. In: Haynes
DL, eds. Compliance in Health Care.
Baltimore, MD: Johns Hopkins University Press; 1979:49–62.
J. Can simple clinical measurements detect patient noncompliance? Hypertension.
JM. Serum TSH concentration as an aid to monitoring compliance
with thyroid hormone in hypothyroidism. Am J Med.
JA. Monitoring compliance
through analysis of drug and metabolite levels. Control Clin Trials.
D. Improved compliance
measures: applications in an ambulatory hypertensive drug trial. Clin Pharmacol Ther.
G. Clinical usefulness of riboflavin-tagged isoniazid for self-medication in tuberculosis patients. Am Rev Respir Dis. 1960;82:1–10.
M. Measuring compliance
maintenance patients: use of a pharmacologic indicator to estimate methadone
plasma levels. Clin Pharmacol Ther.
, Lung, and Blood Institute Working Group on Patient Compliance
. Management of patient compliance
in the treatment of hypertension. Hypertension.
P. Predicting compliance
with a regimen of digoxin therapy in family practice. CMAJ.
BP. Measuring intake of a prescribed medication: a bottle count and a tracer technique compared. Clin Pharmacol Ther. 1970;7:49–54.
AM. The effect of compliance
with treatment on survival among patients with haematologic malignancies. J Clin Oncol.
for noncompliance among patients with hypertension: is self-report the best available measure? Med Care.
GH. Guidelines for reading literature reviews. CMAJ.