|Home | About | Journals | Submit | Contact Us | Français|
The definition and assessment of adherence vary considerably across studies. Increasing consensus regarding these issues is necessary to improve our understanding of adherence and the development of more effective treatments. We review the adherence literature over the past 3 decades to explore the definitions and assessment of adherence to oral antipsychotics in schizophrenia patients. A total of 161 articles were identified through MEDLINE and PsycINFO searches. The most common method used to assess adherence was the report of the patient. Subjective and indirect methods including self-report, provider report, significant other report, and chart review were the only methods used to assess adherence in over 77% (124/161) of studies reviewed. Direct or objective measures including pill count, blood or urine analysis, electronic monitoring, and electronic refill records were used in less than 23% (37/161) of studies. Even in studies utilizing the same methodology to assess adherence, definitions of an adherent subject varied broadly from agreeing to take any medication to taking at least 90% of medication as prescribed. We make suggestions for consensus development, including the use of recommended terminology for different subject samples, the increased use of objective or direct measures, and the inclusion in all studies of an estimate of the percentage of medication taken as prescribed in an effort to increase comparability among studies. The suggestions are designed to advance the field with respect to both understanding predictors of adherence and developing interventions to improve adherence to oral antipsychotic medications.
There is no question that adherence to medication is essential to maximizing outcomes for individuals with schizophrenia.1,2 While adherence is poor across a wide variety of physical and psychiatric conditions,1,3–5 the consequences of poor medication adherence can be devastating in schizophrenia, where the personal and societal costs of relapse are very high.1,2 Although we continue to develop new antipsychotic and adjunctive treatments with broader efficacy and improved side-effect profiles, levels of adherence remain alarmingly low.1,3
For decades, researchers have worked to explain the causes of poor adherence and to develop interventions.6–8 Unfortunately, there has been remarkably little agreement regarding the definition of adherence or how it is best measured. Medication adherence is often defined as “the extent to which a person's behavior coincides with medical … advice.”5(p2) However, different operational definitions and different assessment methods identify different subgroups of patients. If the same patient can be identified as adherent in one study and nonadherent in the next, how are we to combine information across studies? If we continue to proceed as if we are speaking about the same thing when we use the term “adherence,” our efforts to understand and improve it are likely to remain largely unsuccessful. Changing our terms from “noncompliance” to “adherence” to “concordance” has served to promote the idea that medication treatment should be a collaborative effort between doctor and patient, but it has done little to address the fundamental methodological problems in adherence research. An agreed-upon set of definitions and a better understanding of the measurement problems and how to address them are necessary if we are going to unravel the complex nature of adherence and intervene effectively in schizophrenia. With these goals in mind, we review the literature on adherence to oral antipsychotics in schizophrenia patients over the past 3 decades.
We searched the literature published between 1970 and February 2006 using both MEDLINE and PsycINFO for articles containing the terms “adherence,” “concordance.” or “compliance” in combination with “schizophrenia.” In addition, we utilized the reference sections of review articles to identify articles we may have missed using these search terms. We eliminated case studies, articles that did not include at least 1 measure of medication adherence, articles examining only or primarily depot medications, studies involving mainly inpatients who were administered medication during the time adherence was assessed, studies using dropout from a clinical drug trial as the primary measure of adherence, and review articles. We identified a total of 161 articles. A minimum of 2 investigators read all studies and agreed that the assessment of adherence for each study fit into 1 or more of the following discrete categories: self-report, significant other report, treatment provider report, chart review, pill count, electronic refill records, electronic monitoring, blood or urine levels, urine level of tracer substances, and ability to take medication. A priori we defined self-report, provider report (physician, nurse, or caseworker), and chart review as more subjective/indirect assessments of adherence, and pill counts, electronic monitoring, blood or urine sampling, and electronic refill records as either more objective or more direct measures. Chart review was not included as a direct/objective measure due to the heavy reliance on chart information in the report of the patient and opinion of the treatment provider. In addition, each author made a determination as to whether the article represented a specific adherence study. An investigation was classified as an adherence study if the goal were to identify predictors of adherence to oral medication, to examine relationships between adherence and other variables, to investigate the assessment of medication adherence, or to examine the effects of a treatment specifically designed to improve medication adherence. Using these criteria, studies examining the effects of family therapy on outcomes and those examining reasons for relapse were not classified specifically as adherence studies.
Table 1 lists the identified studies, the number of subjects, the methods of assessing adherence, and the criteria used to define adherence. Of 161 studies that were identified, 93 were classified as specifically adherence studies, and 68 were classified as general studies that included a measure of adherence to oral antipsychotic medication. The most common method used to assess adherence in both general and adherence studies was the report of the patient. Self-report was utilized alone or in combination with other methods a total of 107 times in 161 studies. Moreover, 25% (17/68) of general studies and over 36% (34/93) of all studies that specifically examined adherence to medication used self-report as the sole assessment of medication adherence. Self-report methodologies themselves differed greatly among studies and included ad hoc measures, interview (unspecified), semistructured interviews (unspecified), and semistructured interviews using the Rating of Medication Influences (ROMI),9 Treatment Compliance Interview (TCI),9 Drug Attitude Inventory (DAI),10 the medication compliance item from the Multnomah Community Ability Scale (MCAS),11 Medication Adherence Rating Scale (MARS),12 knowledge level, medication checklist, attitudes and insight, which asked subjects if there was a week or 2 weeks in the past variable period of time during which they stopped medication, and medication refusal.
Other methods were used less frequently. In order of frequency, methods were treatment provider report, family or significant other report, chart review, pill count, electronic refill information, electronic monitoring, blood level, urine analysis of medication, and urine analysis of a tracer substance. Figure 1 illustrates the specific frequencies of use of each method in the 161 studies. The subjective/indirect measures of assessing adherence were used 218 times (some studies used more than 1 of these methodologies) and were the only measures used in 124 of the 161 studies. Out of 161 studies, pill counts, blood levels, urine analysis, electronic monitoring, electronic refill records, or tracers that could provide objective or direct data regarding medication adherence were used only 43 times, representing a total of 37 of 161 studies. Of these 37 studies, 27 were specifically adherence studies and 10 were general studies. In one-fourth of these studies, the objective methodology was used only in part of the sample, only when reports by the patient were questioned (no criteria provided), only when the patient brought the urine or pills in to be examined, or only when available (usually with no information on how many subjects had such data available). One study used ability to take medication in a performance-based assessment.
Even in cases where studies used the same methodology to assess adherence, the definition of adherence varied greatly. Figure 2 describes the various levels of discrimination between adherence and nonadherence and the numerous ways to define nonadherence to oral medication. Any inception cohort of individuals for whom it is recommended that medication be taken continuously can be divided into those who refuse and those who accept medication. In a study using this first level of discrimination depicted in Figure 2, those who agree to take medication (accept) would be adherent, and all others would be considered nonadherent. For those who accept the medication, the next level divides subjects into those who continue to use it for a period of time and those who discontinue use for some time period, which varies across studies. At this level, those who discontinue use are nonadherent, all those who continue using the medication, even if dosages vary considerably from what is prescribed, are considered adherent. The next level of inquiry examines only those who continue to take medication. These individuals may take medication as prescribed or not. How much an individual can vary from the prescribed dosage and still be considered compliant differs by study. Dosage deviations can be due to a decision that less medication is better, due to unintentional factors such as forgetting, or due to environmental barriers such as poverty and lack of transportation.
With respect to dosage deviation, in the 161 articles we found dosage cutoffs that ranged from 50% to 90% and categorical classifications ranging from taking any of the prescribed medication to taking nearly every dose. Likert-type scales that were not divided into percentage of medication taken varied from 3 points to 7 points, with a variety of different terms for each point, including “overreliance on medication” at the high end of 1 scale.
For electronic refill data, a common adherence measure was the mean gap or the length of time for which no medication was available to the patient. An alternative measure examined patients who had gone a quarter without a claim for medication.
Several studies reported using Medication Event Monitoring System, or MEMS—pill bottle caps capable of recording the time and date each time the bottle is opened. Openings for other reasons (eg, filling), if known to the researchers, must be deleted to provide an accurate estimate of the number of doses taken. While MEMS is sometimes described as a “gold standard” of adherence assessment, in the studies using MEMS for schizophrenia patients, the fact of missing data was identified as a problem. One study reported close to 45% of data missing due to failures on the part of the patient to bring in the MEMS caps in order to download the information.13
The review illustrates the heavy reliance in the field on self-report and other subjective/indirect measures of adherence, which are known to be significantly flawed. Unfortunately, each method used to assess adherence to oral medications in this population has its own drawbacks. Self-report often exaggerates the degree of adherence. A commonly cited quotation from subjects regarding self-report of adherence behavior goes as follows: “How do you expect me to remember when I forget to take my medication?” Provider report may be based on the report of the patient or on a worsening clinical condition, which may be related either to poor adherence or to a failure of the chosen medication to control symptoms. The report of significant others is affected by how much time the respondent spends with the identified subject and how directly involved the significant other may be in the subject's care. This method cannot be used when patients do not have sufficiently involved informants. Studies using chart review often did not specify the information available in the chart to make a determination about adherence. References to medication adherence in the chart may turn out to be based largely on self-report.
More direct or objective measures of assessment also have problems. Pill counts and refill records can be affected by the use of samples and old medications that are still available to the patient. Pill counts are often complicated by medication from earlier time periods that are added to current prescription bottles. This leaves the researcher with more pills to count than the number indicated on the bottle as dispensed by the pharmacy. Most studies reviewed expected patients to bring in bottles for the pill count; some required patients to bring in urine samples or electronic monitors (MEMS caps). This methodology is likely to bias results toward finding higher levels of compliance. Electronic monitoring can also be problematic when subjects do not replace the caps or take medication out of alternative bottles. While electronic monitoring is considered a gold standard in adherence research with other populations, the cognitive impairments and unstable living environments often found in patients with schizophrenia may make it necessary to use alternative methodologies such as home visits to retrieve MEMS data or using a system that automatically downloads adherence information.57,174
Electronic prescribing records provide an objective assessment of the medication obtained by the patient. Unfortunately, just because medication is available does not mean that it is taken. While this method is likely to underreport problems with adherence, utilizing electronic refill data in large samples can provide substantial power to examine relationships between adherence and outcomes from a cost perspective. Certain assumptions must be made in utilizing such data (eg, that prescriptions are not filled outside of the system). In addition, decisions must be made regarding how to deal with subjects taking more than 1 antipsychotic medication (eg, delete all subjects with multiple antipsychotic prescriptions during the specified period). Accuracy problems may occur when subjects are given unrecorded samples, when medications are filled outside the system, and when old medications from prior episodes of poor adherence are available to the patient.
Even blood levels cannot be considered a gold standard for adherence as they can be highly affected by behaviors in the days immediately prior to the blood draw. Thus, the sample obtained may not represent the level of adherence over a more extended time period (1 month, 3 months). Moreover, there is a great deal of individual variability in blood levels across patients. With the atypical antipsychotic medications there is very little data about appropriate or therapeutic blood levels to use as a criterion.100 In fact, in a recent study of olanzapine, blood levels were taken but not used as the primary measure of adherence for this reason.100 Obviously, the more intrusive or elaborate the method of adherence assessment, the more it will deviate from clinical practice and be less appropriate for use by practicing clinicians. Moreover, some adherence assessments can influence adherence behavior.
Few studies appropriately critique their own methods for assessing adherence or point out the problems encountered in obtaining the data using the chosen method(s). This can make it somewhat unclear whether the conclusions of the studies are fully supported by the data presented.
With respect to the definition of adherence, the field may not have as many definitions as studies, but the number of definitions is clearly problematic. More often than not, the definitions and rationales for the choices are not clearly explained. Because definitions differ, the same subject could be categorized differently depending on the study. For example, a patient who takes about 55% of his or her antipsychotic medication would be classified as adherent in a study examining medication refusal as the criterion for nonadherence. This same patient would be classified as nonadherent if he or she discontinued medication for at least a 1-week period in a study using this as the method to define the nonadherent group. In a study using a cutoff of taking at least 50% of medication, this patient would fall into the adherent group, even if he or she did not take any medication for a 1-week period because the measure used averages over a period of 1 month. The same patient would be nonadherent in a study using a higher cutoff percentage during the specified time period.
What is an appropriate percentage of medication for an individual to take before he or she is considered poorly adherent? Based on the review, the answer to this question is far from clear. Percentage of medication taken as prescribed could be used as a continuous variable. However, the percentage may have high variability, necessitating very large sample sizes before statistical significance could be found in treatment studies in which a treatment would have a moderate effect. In addition, the difference between taking 0% versus 25% or 80% versus 100% of prescribed medication may not be clinically significant. Therefore, grouping subjects into clusters of 0 to 29%, 30–69%, and 70–75%+ may make conceptual sense. Part of the difficulty may be that each investigator examines their data and chooses cut points based on natural breaks in their data. Arbitrarily choosing 80% as a cutoff for adherence may yield no adherent patients in some studies.
The issue of percentage of medication taken as prescribed to classify adherence is further complicated by the tendency over time for physicians to increase the dosage of medication when symptoms are not well controlled. Over time, dosages can creep up because patients are not fully adherent, making the full dose clinically inappropriate when the patient is taking medication as prescribed.
It is common to validate adherence measures against clinical outcome. This is best seen in studies of electronic pharmacy records in which gaps in having medication available predict hospitalization. However, there may be problems in using some clinical outcome data to validate adherence measures. It is not uncommon for treating physicians to base their impression of adherence on clinical state (symptomatology, clinical global impression). Therefore, it is possible that physician-rated adherence would have a stronger relationship with clinical state than adherence assessed through more objective means. It is also true that we do not know the clinical consequences of many types of nonadherence.
Given the confusion in the field, we would propose defining those who do not accept medication as “medication refusers” to distinguish them from individuals who continue to take medication but may have adherence problems. This latter group of medication acceptors can then be further divided by degrees of adherence. This is important because complete refusal may begin as a function of missed or skipped doses, either intentional or accidental. It is likely that what predicts medication refusal and what predicts irregular compliance, or what we call “dosage deviation,” may be very different. It is also likely that treatment approaches to these 2 groups of individuals may need to target very different variables, insight in the former case and cognitive deficits or environmental problems in the latter. It is unclear how patients who discontinue the use of medication for 1 week or longer compare with those who somewhat consistently take half of their prescribed dosage. Moreover, it is likely to be very important to distinguish those who deviate in dosage by choice versus those who inadvertently miss medication due to forgetting, misunderstanding, poverty, or other environmental barriers. The focus of intervention for these groups could vary considerably.
Prospective studies that follow patients over time and examine adherence are necessary to determine predictors of problem adherence. The difficulties with retrospective data, particularly in the schizophrenia population, are numerous and include problems with inaccurate recall and backward reasoning (“I got sick, so I must have forgotten to take my medication”). Longer follow-up periods are likely to minimize the impact of assessing adherence on adherence behavior.
To increase comparability among studies, it would be helpful for each investigation to report an estimate of the mean percentage of medication taken during the follow-up period, even if the primary measure of adherence is operationalized otherwise. This would allow studies to be compared on a common variable. At the same time, this would allow investigators to group patients according to natural breaks in their data and the overall adherence level in their samples.
We would suggest that all studies include at least 2 measures of adherence and that at least 1 of these be a direct or objective measure such as pill count, urine analysis, blood analysis (if problems discussed below have been addressed), electronic monitoring, pharmacy refill records, or the examination of tracer substances in blood or urine. Pharmacy refill records have been found to be useful for large samples, but it is unclear whether this method will be sensitive in smaller samples investigating the effects of clinical treatment. While it may be economically prohibitive in some studies and inconvenient in all studies, doing pill counts, downloading electronic monitoring devices, and collecting blood or urine are likely to be best accomplished during home visits. We have been able to decrease loss of data to less than 5% by downloading MEMS information from the caps on home visits using laptop computers. Alternatively, using more sophisticated electronic devices such as the Med-eMonitor174 is recommended. This device records the same type of data as MEMS but stores up to 5 different medications. Using an LCD readout, the monitor can query the subject as to whether a specific compartment has been opened for the purpose of taking medication or for some other reason. Most important, the monitor downloads adherence data automatically to a secure Web site, decreasing problems with data retrieval.174
An in-home setup at the beginning of studies that use electronic monitoring and pill counts can cut down on problems. For example, extra medications can be bagged and stapled and only recounted if the seal is broken. A box can be provided for patients to store empty bottles, making determinations of pills dispensed more accurate over time. Pill counts should be cross-referenced with prescribing records to deal with some of the problems described above, such combining bottles together. Training on the use of electronic monitoring devices, providing belt bags to carry the larger electronic pill containers, and using the more sophisticated devices may advance the field.
Because there is so much variability in blood-level data from patient to patient for the atypical antipsychotic medications, if blood samples are collected, obtaining individualized baseline levels during a period in which all medication taking is monitored may be ideal. Randomly drawn blood levels during a follow-up period can then be compared with the baseline levels for consistency between consecutive levels and consistency in plasma level/dose. However, unless better procedures for interpreting blood-level data become available in the near future for the atypical antipsychotic medications, it is unclear how blood-level data will assist the field in better understanding adherence.
The use of multivariable algorithms that combine measures to make a determination of adherence level needs further investigation. One approach would involve assessing larger samples of patients with multiple measures of adherence. With samples approaching 200, factor analytic techniques could be used to determine whether adherence could be conceptualized as a latent variable in which the various sources of error could be explicitly modeled.
While the assessment of the ability to take medication is important for this population, the ability to perform a task does not always translate into performing the task in the natural environment. Because adherence problems are likely to be multidetermined in the natural environment, the usefulness of ability as a proxy for actual adherence may be limited.
Irrespective of method, we recommend briefly describing the reasons for selection of the method, the pros and cons of the method selected, and the actual problems encountered in obtaining the data. For example, a recent study by Grymonpre et al.4 in an elderly population describes the problems encountered in their sample while obtaining pill counts. Including this type of information in a report would assist readers in evaluating the conclusions of the study.
In summary, to be successful in identifying predictors of adherence and developing interventions we need to do better in defining and assessing adherence. By putting into place some standardization of terms and procedures, the field is more likely to accomplish these goals. These suggestions for consensus development most closely apply to adherence to oral antipsychotic medications in psychosis patients. However, given that the vast majority of prescriptions for atypical antipsychotic medications are “off-label,” the problems in assessment and the recommendations made may be relevant for a wider range of populations.
Dr. Velligan, Dr. Lam, Dr. Ereshefsky, and Dr. Miller were supported in part during the writing of this paper by National Institutes of Mental Health grant R01 MH62850-05. Dr. Velligan, Dr. Miller, and Dr. Glahn were supported in part during the writing of this paper by National Institutes of Mental Health grant R01 MH61775-04.