Interventions to Enhance Medication Adherence in Chronic Medical Conditions: A Systematic Review

Background  Approximately 20% to 50% of patients are not adherent to medical therapy. This review was performed to summarize, categorize, and estimate the effect size (ES) of interventions to improve medication adherence in chronic medical conditions.

Methods  Randomized controlled trials published from January 1967 to September 2004 were eligible if they described 1 or more unconfounded interventions intended to enhance adherence with self-administered medications in the treatment of chronic medical conditions. Trials that reported at least 1 measure of medication adherence and 1 clinical outcome, with at least 80% follow-up during 6 months, were included. Study characteristics and results for adherence and clinical outcomes were extracted. In addition, ES was calculated for each outcome.

Results  Among 37 eligible trials (including 12 informational, 10 behavioral, and 15 combined informational, behavioral, and/or social investigations), 20 studies reported a significant improvement in at least 1 adherence measure. Adherence increased most consistently with behavioral interventions that reduced dosing demands (3 of 3 studies, large ES [0.89-1.20]) and those involving monitoring and feedback (3 of 4 studies, small to large ES [0.27-0.81]). Adherence also improved in 6 multisession informational trials (small to large ES [0.35-1.13]) and 8 combined interventions (small to large ES [absolute value, 0.43-1.20]). Eleven studies (4 informational, 3 behavioral, and 4 combined) demonstrated improvement in at least 1 clinical outcome, but effects were variable (very small to large ES [0.17-3.41]) and not consistently related to changes in adherence.

Conclusion  Several types of interventions are effective in improving medication adherence in chronic medical conditions, but few significantly affected clinical outcomes.

An estimated 20% to 50% of patients do not take their medications as prescribed and are said to be nonadherent or noncompliant with therapy.^1-3 In the setting of chronic medical conditions such as hypertension and hypercholesterolemia, medication nonadherence leads to worse medical treatment outcomes, higher hospitalization rates, and increased health care costs.^4,5 Because of this, adherence has been called “the key mediator between medical practice and patient outcomes.”⁶

Researchers have tested a variety of interventions to improve patient adherence, ranging from simple adjustments in the medication regimen to complex multidisciplinary interventions that address health system barriers and communication between patients and health care professionals.^7-10 However, the overall quality of the literature is poor, with wide variability in study design, including patient population, outcome measure, and duration of follow-up.¹¹ It is desirable to draw recommendations from a smaller set of high-quality studies.

In a recent review for the Cochrane Database,¹² several simple interventions appeared to improve adherence with short-term regimens, such as a course of antibiotics. However, interventions to improve medication use for chronic conditions appeared less effective overall and were often multifaceted, making it more difficult to synthesize published evaluations.¹² In this article, we extend the findings of the Cochrane review by providing a more detailed examination of interventions to improve adherence in chronic medical conditions. We offer a framework for categorizing these interventions and report their relative impact using a standardized measure of effect size (ES). The goal of this review is to help physicians better understand the strengths and limitations of tested interventions for improving long-term medication use and identify those that are most successful.

Electronic searches of the published literature from January 1967 to September 2004 were conducted through MEDLINE, CINAHL, PsycINFO, SOCIOFILE, International Pharmaceutical Abstracts, EMBASE, and The Cochrane Library, without language restriction. The exact search strategy varied across databases and generally included terms for adherence (eg, compliance, noncompliance, adherence, dropouts, treatment refusal), medication use (eg, medication, medicine, drug, treatment, regimen, pharmacotherapy), and clinical trial design (eg, clinical trial, intervention, outcome, randomized, control). A full description of the search strategy is provided elsewhere.¹²

Reviewers screened citation titles, index terms, and abstracts (if available) to identify potentially relevant articles, which were retrieved for full-text review. Two reviewers independently assessed the articles for possible inclusion. Differences in assessment were resolved by discussion or with assistance from a third reviewer.

Articles were selected if they reported a randomized controlled trial that described 1 or more unconfounded interventions intended to enhance adherence with self-administered medications used in the treatment of chronic medical conditions. Confounding was judged to be present if study groups were treated unequally except for the intervention intended to enhance adherence (eg, patients in the intervention group received not only more encouragement to comply but also a different medication or dose than those in the control group). Included studies were required to report at least 1 measure of medication adherence and 1 clinical outcome, with at least 80% follow-up of participants during 6 months if the study results were initially positive. Studies with negative results and less than 6 months of follow-up were included because initial failure was unlikely to be followed by success. Studies of short-term regimens were included in the Cochrane review¹² but excluded from the present analysis to permit a focus on chronic medical conditions, for which adherence is a greater concern. Trials that pertained to psychiatric disorders were excluded because adherence is generally lower in patients with psychiatric disorders and unique challenges may also be present, potentially limiting the generalizability of interventions tested in that context.¹³

For each eligible study, 1 reviewer extracted features of the patient population, study design, interventions, and control, as well as results for adherence and clinical outcomes. When multiple time points were provided, data were extracted for the longest period of follow-up, provided follow-up remained at least 80%. Each extraction was confirmed by at least 1 other reviewer. Other articles on the same trial were retrieved, and authors were contacted as needed for additional details or verification of reported analyses.

The eligible studies differed substantially in patient population, intervention, adherence measure, and clinical outcome measure, making a pooling of results inappropriate. Instead, studies were grouped by intervention type, using a taxonomy developed from other sources.^8,9,14,15Informational interventions described cognitive strategies designed primarily to educate and motivate patients by instructional means, based on the concept that patients who understand their condition and its treatment will be more informed, empowered, and likely to comply. Informational sessions conducted individually or in a group setting, as well as didactic and interactive approaches, were included. Examples of informational interventions are face-to-face oral, telephone, written, or audiovisual education; didactic group class; and mailed instructional material (not including reminders or prompts to comply). Behavioral interventions were strategies designed to influence behavior through shaping, reminding (cues), or rewarding desired behavior (reinforcement). Examples include skill building by a health care professional; pillboxes, calendars, a change in packaging, or other steps intended to remind the patient; changes in dosage schedule to simplify the regimen or tailor the regimen to the patient's daily routine (ie, reduce its behavioral demands); and rewards and reinforcement (eg, assessment of adherence with feedback to the patient). Family and social interventions involved social support strategies, whether provided by family or another group. Examples are support groups and family counseling. Group sessions that were primarily didactic or informational, rather than supportive, were categorized as informational. Combined interventions included features of 2 or 3 of the preceding categories. Each included study was classified by 1 author using these categories, descriptors, and examples. This classification was reviewed by at least 1 other author, and disagreements were resolved by consensus.

For each outcome, an ES (Cohen d) with 95% confidence interval was calculated from information provided in the article or obtained from the authors. The ESs compare the difference in effect between study groups, divided by the standard deviation of this difference, resulting in standard deviation units.¹⁶ This measure is independent of the method of measurement used, thus permitting comparison of different interventions across studies. In this way, ES provides more information than a simple test of significance comparing outcomes in intervention and control groups. The ES can be positive or negative, depending on the effect of the treatment on the particular outcome measure. If the study results are statistically significant, then the ES is generally significant as well.

We calculated ESs according to established methods.^16,17 For studies that provided range instead of a standard deviation, we converted range to standard deviation.¹⁸ The absolute value of each statistically significant ES was categorized as very small (<0.20), small (0.20 to <0.50), medium (0.50 to <0.80), or large (≥0.80).¹⁶ We calculated ESs between each intervention and control group. Some studies had more than 2 arms and provided only an overall test of significance comparing results across experimental groups. In these cases, the significance of the ES for intervention-control pairs may have differed from the overall significance of results. Also, if study authors did not specify the number of patients involved in the final analysis, we used the baseline sample size of each group to calculate the ES. Calculating the ES in this way, analogous to an intention-to-treat analysis, may underestimate the true effect of the intervention.

The electronic searches returned 13 061 citations (including 458 review articles), 955 of which were judged to merit full-text review. A total of 38 articles describing 37 randomized controlled trials met all inclusion criteria, including 2 articles from the same trial.^19,20 Among the included studies, 13 reported on 12 informational interventions,^19-31 10 on behavioral interventions,^32-41 and 15 on combined interventions.^42-56 None described purely family or social interventions (Figure).

Most studies reported a single measure of adherence. However, the choice of measure varied widely, from self-report scales to more objective measures obtained by auditing refill rates, performing pill counts, and assessing the timing of prescription bottle opening by Medication Event Monitoring System caps. Approximately half of the studies reported a single clinical outcome, which also varied greatly, depending on the patient population.

The ES could be calculated in 30 studies for adherence outcomes and in 30 studies for clinical outcomes. Most studies (71% for adherence and 100% for clinical outcomes) that did not provide enough information to calculate the ES were not statistically significant, which illustrated a form of reporting bias.

The studies of informational interventions each compared intensive education to limited education or usual care (Table 1). Sample size varied from 46 to 350. The interventions ranged in duration from a single session of less than 1 hour to several hours over many sessions. Counseling was provided by a physician, nurse, health educator, or pharmacist. Both individual and group sessions were described, sometimes in combination, and they were often accompanied by written materials.

Half (6 of 12) of these studies reported a significant increase in at least 1 measure of adherence.^{19-21,23,25,28,30} However, 3 of these 6 trials had mixed results. Gallefoss et al^19,20 reported an improvement among only patients with asthma not those with chronic obstructive pulmonary disease. Levy et al²⁵ demonstrated improvement in adherence only for severe, not mild, asthma attacks. In the study by Schaffer and Tian³⁰ of 3 educational strategies, 2 were effective.

Among the 10 studies for which an ES for adherence outcomes could be calculated,^19-25,28-31 a large ES was observed only in a trial conducted by Pradier et al²⁸ of intensive, multisession counseling for human immunodeficiency virus (HIV) treatment adherence (ES, 1.13 for self-reported adherence). Four other studies^19-21,23,25 that provided counseling over multiple sessions had a small to medium ES (range, 0.35-0.68), and the remainder were not significant.

Most informational interventions (8 of 12) did not improve clinical outcomes. Of the 4 studies with at least 1 positive clinical result, 2 had a small ES (Pradier et al²⁸: ES, −0.39 for HIV RNA level; Levy et al²⁵: ES, 0.34 for asthma symptom scores). The other 2 trials had medium to large ESs (Gallefoss et al^19,20: ES, 0.52 for forced expiratory volume in 1 second in patients with asthma; Cote et al²²: ES, −0.64 for urgent asthma-related visits and 3.41 for peak expiratory flow). Each of these 4 studies provided fairly intensive counseling with reinforcement over time. Only the interventions used by Cote et al²² and Pradier et al²⁸ consistently led to better clinical outcomes, despite the study by Cote et al having no effect on adherence owing to a high rate of adherence in the control group (97%). Gallefoss et al^19,20 again saw improvement only among the asthma subgroup, and Levy et al²⁵ had mixed results based on the clinical outcome measure.

The sample size of the behavioral studies ranged from 27 to 497 (Table 2). The most common and effective forms of behavioral intervention were dosage simplification^32,34,36 and repeated assessment of medication use with feedback.^35,37-39 Other trials tested specialized packaging,³³ directly observed therapy,⁴⁰ and cognitive behavior therapy,⁴¹ but none of these techniques significantly affected adherence or clinical outcomes.

All 3 trials of dosage simplification demonstrated an improvement in adherence, whether switching from 2 doses to 1 dose per day or from 4 to 2.^32,34,36 The ESs that could be calculated were large (range, 0.89-1.20).^34,36 The impact on clinical outcomes was mixed. Brown et al³⁴ found a significantly higher percentage of patients reached their low-density lipoprotein cholesterol goal when prescribed extended-release niacin compared with the short-acting form (medium ES of 0.66). However, the study by Baird et al³² of once-daily metoprolol demonstrated no significant improvement in blood pressure, whereas Girvin et al³⁶ actually found a trend toward increased blood pressure in the once-daily therapy group.

The 4 trials that involved assessment and feedback asked patients to monitor and report their medication use and/or blood pressure.^35,37-39 The 3 studies that then provided patients with tailored feedback, reinforcement, or rewards demonstrated a significant improvement in adherence (ESs small to large; range, 0.27-0.81).^35,37,39 The fourth study, which simply made home blood pressure values available to the patient and physician, did not improve adherence.³⁸ Only 1 study in this group produced a clear improvement in clinical outcomes (Marquez Contreras et al³⁹: ES, 0.89 for total cholesterol and 0.78 for low-density lipoprotein cholesterol). Friedman and colleagues³⁵ reported a decrease in the diastolic blood pressure only after controlling for baseline blood pressure in adjusted analyses (small ES of 0.29); they found no significant change for systolic blood pressure.

Most (13 of 15) combined interventions included informational and behavioral components,^42-45,47-55 and 2 joined social support strategies with either informational or behavioral elements (Table 3).^46,56 Sample size again varied widely, from 37 to 1113. Of the 13 studies with informational and behavioral elements, 5 demonstrated improvement in all adherence measures,^{42,47,48,50,51} and 3 others reported mixed or nearly significant results.^43,45,53 The significant effects ranged in size from small to large (absolute ES values of 0.43-1.20). No studies in this group were able to demonstrate a significant change in all measured clinical parameters, although several trials noted a significant improvement in at least 1 clinical outcome^42,48,51,55 or a strong trend.⁵³ The ESs ranged widely across studies (very small to large ES [0.17-1.58]). No consistent relationships were observed between intervention characteristics or disease state and the success of the interventions. For example, 3 trials implemented relatively similar interventions among patients with asthma, but their reported effects varied substantially.^42,45,47 Regarding the 2 studies with strong social support elements, both successfully improved adherence (unable to calculate ES), but neither led to better clinical outcomes.^46,56

In this systematic review of randomized controlled trials designed to enhance medication adherence in chronic medical conditions, 16 of 37 trials reported a consistent improvement in adherence.^{21,23,28,32,34-37,39,42,46-48,50,51,56} Four other studies noted a significant change in adherence among a particular patient subgroup, with some but not all of the adherence measures used, or in a certain arm of a study testing multiple interventions.^{19,20,25,30,43} Nine of the 20 trials that had some impact on adherence were able to demonstrate an improvement in some^{19,20,25,35,42,48,51} or all^28,34,39 of the measured clinical outcomes. Of the 17 studies that did not show an improvement in adherence, 2 improved clinical outcomes; the apparent lack of effect on adherence in these studies was probably due to insensitive adherence measures or a ceiling effect.^22,55

Previous reviews have found few if any consistent relationships between the characteristics and effectiveness of interventions to improve adherence.⁷ In the present study, categorization of the trials by intervention type (informational, behavioral, or combined informational, behavioral, and/or social support) and calculation of ESs permit a few general observations. Behavioral interventions that reduced the dosing demands of individual therapies consistently improved adherence with a large ES (0.89-1.20). Apart from this, other successful interventions usually contained multiple elements delivered over time. For example, 6 informational interventions that provided educational counseling over a few sessions and addressed self-care issues improved adherence with small to large ESs (0.35-1.13). Successful behavioral interventions based in monitoring and feedback and the most successful combined interventions also included various elements, such as self-management plans, reinforcement, and occasionally rewards.

Despite the ability of these interventions to improve adherence, however, a positive effect on clinical outcomes was demonstrated infrequently, and ESs pertaining to clinical outcomes were highly variable (very small to large ES [0.17-3.41]). Moreover, results for clinical outcomes were not consistently related to the magnitude of adherence results or the characteristics of the intervention. Unfortunately, many of the studies were relatively small and not sufficiently powered to detect differences in clinical outcomes. Thus, the results of many of the studies are indeterminate for clinical outcome, rather than clearly negative.

The adherence literature and present study had several additional limitations. First, because we included only published trials, the reported findings may overestimate the true effect of such interventions due to publication bias. Second, our inclusion criteria were somewhat stringent. Many studies with shorter follow-up or no measurement of clinical outcomes were excluded, and they are listed elsewhere.¹² Third, our ability to categorize the trials into distinct modes of intervention was limited by the frequently complex nature of the interventions, requiring us to make judgments about the predominant components. Fourth, although we calculated ES and can make general comments about the magnitude of adherence and clinical outcome differences, the included studies were too heterogeneous to warrant pooling in a formal meta-analysis. Fifth, the literature review was completed in September 2004 and trials published since then may differ in their results, although they would be unlikely to substantially change the conclusions drawn from the 37 trials included in this review.

Because most of the available literature does not separate out the effects of the individual components of multifaceted interventions, it is not possible to draw definitive conclusions about which features of combined interventions are most beneficial. Additional research is needed to clarify which features are most responsible for changes in adherence and clinical outcomes, with the caveat that individual components may not prove powerful enough to show important effects. Future studies should also examine the effect of varying the intensity of interventions to determine dose-response relationships. Such findings would have important implications for health systems considering the implementation of patient adherence programs on a large scale. Investigations should be conducted with clinically meaningful outcomes as the primary end points and be sufficiently powered to detect a difference in these measures. Most important, future research should seek to understand the determinants of adherence behavior and to develop and test innovative ways to help people adhere to prescribed medication regimens, rather than persisting with existing approaches.

In summary, we found that several types of interventions are effective in improving medication adherence, but few were able to demonstrate an impact on clinical outcomes. Based on this review of the literature, the most effective interventions appear to be those that simplify dosing demands. Interventions that involve monitoring and feedback, as well as informational interventions delivered over multiple sessions, are probably also effective.

Correspondence: Sunil Kripalani, MD, MSc, Division of General Medicine, Department of Internal Medicine, Emory University School of Medicine, 49 Jesse Hill Jr Dr SE, Atlanta, GA 30303 (skripal@emory.edu).

Accepted for Publication: August 23, 2006.

Author Contributions: Dr Kripalani had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Kripalani and Haynes. Acquisition of data: Kripalani, Yao, and Haynes. Analysis and interpretation of data: Kripalani, Yao, and Haynes. Drafting of the manuscript: Kripalani, Yao, and Haynes. Critical revision of the manuscript for important intellectual content: Kripalani and Yao. Statistical analysis: Kripalani and Yao. Obtained funding: Haynes. Administrative, technical, and material support: Yao and Haynes. Study supervision: Kripalani and Haynes.

Financial Disclosure: None reported.

Funding/Support: Dr Kripalani was supported by a Mentored Patient-Oriented Research Career Development Award (grant K23 HL077597) and formerly by the Emory Mentored Clinical Research Scholars Program (National Institutes of Health/National Center for Research Resources grant K12 RR017643). Dr Haynes is supported by the Michael Gent Professorship.

Role of the Sponsors: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; or the preparation, review, or approval of the manuscript.

Acknowledgment: We thank Stephen Walter, PhD, McMaster University, for his assistance with ES calculations and comments on an early draft of the manuscript. We also thank Heather McDonald, MSc, Aqeel Degani, MSc, and Amit Garg, MD, PhD, for their contributions to previous Cochrane reviews.