571213

research-article2015

CNU0010.1177/1474515115571213European Journal of Cardiovascular NursingRuppar et al.

EUROPEAN SOCIETY OF CARDIOLOGY ®

Original Article

Medication adherence interventions for heart failure patients: A meta-analysis

European Journal of Cardiovascular Nursing 2015, Vol. 14(5) 395­–404 © The European Society of Cardiology 2015 Reprints and permissions: sagepub.co.uk/journalsPermissions.nav DOI: 10.1177/1474515115571213 cnu.sagepub.com

Todd M Ruppar1, Janet M Delgado1 and Jonathon Temple2

Abstract Background: Adherence to medications is an essential part of heart failure self-care. Poor medication adherence leads to increased rates of exacerbation causing hospitalizations and increased morbidity and mortality. Aims: This meta-analysis aimed to quantify the effect of interventions to improve adherence to heart failure medications on adherence outcomes. Methods: Comprehensive search methods identified studies testing interventions designed to improve medication adherence among patients with heart failure. Data from eligible studies were independently coded by two coders and analyzed using random-effects meta-analysis methods. Moderator analyses to explain heterogeneity among the studies were conducted using meta-regression and ANOVA for moderators with sufficient numbers of comparisons. Results: Searching yielded 6665 potential study reports. From these studies, we identified 29 eligible treatment versus control comparisons of heart failure medication adherence interventions (total n=4285). The mean effect size (d-index) for two-group comparisons was 0.29 (SE=0.09, p=0.004). Moderator analyses found effect sizes were larger as samples were older. Medication adherence effect sizes were larger for studies conducted in Europe or Asia versus North America, and for interventions focused on changing only medication adherence. Smaller effect sizes were seen for interventions with components directed at health care providers, and those including social support as an intervention component. Conclusion: While the medication adherence effect size across all studies was significant, the effect was modest. Approaches to improving heart failure medication adherence may be most effective when focused on medication adherence alone, and when seeking to change behavior of patients, rather than health care provider behavior. Keywords Heart failure, medication adherence, meta-analysis, intervention, systematic review, self-care Date received: 1 October 2014; accepted: 15 January 2015

Introduction Heart failure is a significant cause of morbidity and mortality. Heart failure prevalence is estimated at over 5.8 million persons in the USA and 23 million persons worldwide.1,2 By 2030, the prevalence in the USA is expected to increase 46% to about 8.5 million persons.2,3 The prognosis for heart failure patients remains poor. Heart failure leads to considerable functional impairment, symptom burden, and health care costs. Heart failure is associated with high mortality; in 2009, one in nine death certificates listed heart failure as a cause or contributing factor.3 Heart failure has a high rate of hospitalizations and negatively impacts patients’ physical function and quality of life.2,3 Management of heart failure requires multiple

health behavior changes, including close attention to diet, physical activity, and cardiovascular medications.4,5 Medication is essential to control heart failure symptoms and prevent exacerbations. Poor adherence to heart failure medication is related to increased mortality,

1Sinclair 2School

School of Nursing, University of Missouri, Columbia, USA of Pharmacy, University of Missouri–Kansas City, USA

Corresponding author: Todd Ruppar, Sinclair School of Nursing, University of Missouri, S423 School of Nursing, Columbia, MO 65211, USA. Email: [email protected]

396 morbidity, increased hospitalization rates, and health care costs.3,6–10 Adequate adherence to heart failure medication, however, is a challenge for heart failure patients, their families, and health care providers. About 50% of patients are not adherent to their chronic medication regimens.10–12 One approach to mitigate poor heart failure outcomes is through interventions designed to improve heart failure patients’ adherence to medication regimens. Although many studies have tested interventions to improve heart failure medication adherence, very little has yet been done to synthesize the results of such studies, evaluate the state of the science, or compare the effectiveness of interventions. While several systematic reviews and narrative reviews have addressed aspects of the issue,7,13,14 the findings of the reviews have been mixed. No meta-analyses have synthesized the impact of medication adherence interventions on heart failure medication adherence outcomes. One meta-analysis by Wakefield and colleagues15 reported an effect size on medication adherence outcomes from three studies, but the search did not focus on interventions to improve medication adherence. This paper reports the adherence outcome results of a meta-analysis of interventions to improve adherence to medications among adults with heart failure. Our approach not only synthesizes the outcome effects across the eligible studies, but also conducts moderator analyses to determine the intervention, design, and patient population characteristics that are associated with the greatest improvement in medication adherence. Such analyses will lead to refining of existing and future interventions.

Methods We used systematic review and meta-analysis methods, reported here according to the PRISMA guidelines,16 to synthesize results of reports of studies testing interventions to improve adherence to medications for heart failure. The review protocol is available from the lead author.

Eligibility criteria Studies were eligible if they reported testing an intervention designed to improve adherence to heart failure medications in a sample of adults (age ⩾ 18 years) with heart failure. Interventions were not required to be solely for the purpose of improving adherence to medications, but some component of the intervention must have been designed to improve medication adherence. This review conceptualized medication adherence according the recent taxonomy of Vrijens and colleagues, which views adherence as encompassing the entire process of taking medication as prescribed.17 Differences based on specific adherence component or by adherence measure could then be assessed empirically.

European Journal of Cardiovascular Nursing 14(5) We did not limit eligibility by study design; however, two-group trials were analyzed separately from singlegroup, pre-post designs. Since very few studies used single-group approaches, this paper will focus on the results of the two-group comparisons. Studies must have included a medication adherence outcome measure and reported sufficient data for calculating a medication adherence effect size. Studies also were not excluded a priori based on length of follow-up; rather, the duration of follow-up was addressed empirically as a moderator variable in the meta-analysis.

Search strategies Our team used comprehensive search strategies developed for this project by an expert health sciences research librarian on our team to identify all potentially eligible studies. Search terms were tailored to each database, and included both indexed search terms and free text keywords. Databases searched included MEDLINE, CINAHL, PsycINFO, Cochrane Central Register of Controlled Trials, Scopus, ProQuest, International Pharmaceutical Abstracts, DARE, and Highwire. Searches were conducted on each database from inception through 2013. Hand searching of selected journals was done for the last 10 years to identify any recent studies that may not have been identified in our database searches. Reference lists of review articles were also searched to identify any further studies. Studies were separately evaluated for eligibility by two research staff. Similarly, data extraction and coding of eligible studies was done by two trained research staff, and data were compared to ensure accuracy of data extraction. If eligible study reports did not report adequate data for calculating an adherence outcome effect size, the primary study authors were contacted with a request for additional data. If a study reported results for more than one intervention group, each intervention was coded as a separate comparison. Our initial searches did not exclude based on language, as some non-English publications include English-language abstracts with sufficient data for meta-analysis coding.

Data management and analysis Study citation data were managed using bibliographic management software and tracked using a Microsoft Access database. Coded data were duplicate-entered into Excel, corrected for errors, and imported into Comprehensive Meta-Analysis Software for analysis. A standardized mean difference effect size (Cohen’s d) was calculated for each comparison. The overall mean effect size was calculated using random-effects meta-analysis methods. Random-effects models account for heterogeneity due to both within- and between-study variance. The random-effects approach was chosen due to the expected heterogeneity between samples with respect to

397

Ruppar et al.

Figure 1.  Study flow diagram.

s: number of studies; k: number of comparisons; HF: heart failure.

both sample and intervention characteristics. Such heterogeneity is expected in health behavior research. In calculating the mean effect size, each comparison’s effect size was weighted by the inverse of its variance plus the between-study variance estimate. This gave more weight to studies with greater measurement precision. Heterogeneity was assessed by calculating the Q statistic and I2, which is the proportion of variance explained by the meta-analytic model. This proportion (I2) serves as a “signal-to-noise” ratio that indicates that additional unexplained variance exists that may be explained by moderator analyses. We assessed for individual study outliers by checking for large standardized residuals. Publication bias was assessed visually using funnel plots. Other sources of potential bias were coded and tested empirically for their effects on study effect sizes across the sample of studies. We conducted moderator analyses using meta-regression (for continuous variables) and ANOVA (for categorical variables). In the moderator analyses, we analyzed study characteristics, such as year of publication and presence of funding; sample characteristics, such as mean age, gender, race and ethnicity; and intervention characteristics, such as use of medication education, feedback, intervention delivery type, and intervention dose. Dichotomous moderator analyses were performed for those moderators with at least five comparisons in each group. This investigation conformed to the principles outlined in the Declaration of Helsinki. As this study was a metaanalysis and did not enroll human subjects, it did not require institutional review board approval.

Results Comprehensive literature searching yielded 6665 citations for screening to identify studies testing medication adherence interventions in samples of heart failure patients. From these studies, we identified 29 eligible treatment versus control comparisons with sufficient data for coding medication adherence outcomes (see Figure 1 for flow diagram in accordance with PRISMA recommendations16). The studies, published between 1995 and 2012, report data on a total of 4285 study participants. The median of mean sample age was 69.5 years. Eighteen comparisons were from studies conducted in North America, seven in Europe, and four in Asia. Co-morbidities were common among the subjects in the included samples. The most common reported co-morbidities were hypertension, coronary artery disease, diabetes mellitus, and chronic respiratory disease. The percentage of patients containing different co-morbidities in each sample was not consistently reported across all eligible studies. The most common intervention components were medication education (k=26) and disease education (k=20). This education was often delivered as part of a medication counseling intervention approach (k=13) and/ or through written instructions about medication-taking (k=14). Eleven interventions had patients self-monitor heart failure signs and symptoms, but only four interventions asked patients to self-monitor their medicationtaking behavior. Study attributes including design, sample size, interventions, and adherence measure used are listed in Table 1.18–40

398

European Journal of Cardiovascular Nursing 14(5)

Table 1.  Characteristics of studies included in the meta-analysis. Primary author (year) Region / Study duration

Study design MA outcome sample sizes

Intervention

Adherence measure

Antonicelli (2010)18  Europe   12 months Bisharat (2012)19  Asia   12 months Bouvy (2003)20  Europe   Six months Dawson (1998)21   North America   One month

RCT Tx: 29 Co: 28 Non-randomized trial Tx=33 Co=41 RCT Tx: 48 Co: 43 Quasi-experimental Tx: 8 Co: 2

Weekly telemonitoring to assess adherence and symptoms and adjust treatment as needed

Self-report by telephone interview Pharmacy refills

Falces (2008)22  Europe   12 months Goodyer (1995)23  Europe   Three months Gwadry-Sridhar (2005)24   North America   12 months Jerant (2003)25   North America   Two months

RCT Tx: 29 Co: 20 RCT Tx: 42 Co: 40 RCT Tx: 48 Co: 47 RCT Tx-A: 13 Tx-B: 11 Co: 12 RCT Tx: 121 Co: 109

Laramee (2003)26   North America   Three months Lopez Cabezas (2006)27  Europe   12 months Murray (2007)28   North America   12 months Nimpitakpong (2002)29  Asia   11 days

RCT Tx: 40 Co: 23 RCT Tx: 122 Co: 192 Quasi-experimental Tx-A: 29 Tx-B: 33 Co: 30

Nucifora (2006)30  Europe   Six months Powell (2010)31   North America   24 months Rich (1996)32   North America   One month

RCT Tx: 85 Co: 93 RCT Tx: 451 Co: 451 RCT, individual randomization Tx: 80 Co: 76 Controlled trial, allocated by pharmacy Tx: 195 Co: 181

Ringer (2001)33   North America   15 months

Patient counseling by nurse (pre-discharge) and pharmacist (post-discharge) Structured interviews with pharmacists covering medication use, reasons for non-adherence, and reinforcing adherence; monthly follow-up contacts Patient education and goal setting with an advanced practice nurse on HF pathophysiology, medication effects and side effects, diet, physical activity, symptom monitoring, and follow-up appointments Education about heart failure, medication, diet with telephone follow-up

Electronic monitoring

Counseling on correct medication use by a pharmacist; also provided medication calendars

Pill counts by research staff

HF information booklets, HF video, and patient education on MA, diet, and lifestyle recommendations from nurse and pharmacist A: Home video-based telecare visits B: Telephone contact by nurse

Pharmacy refills

Inpatient care coordination by a nurse case manager; patient and family education; 12 weeks of telephone follow-up; adjustment of HF medications to optimal regimen according to guidelines Education about HF, diet, medications and MA; telephone follow-up

Self-report by telephone interview

Pharmacist-delivered medication education, health literacy tools, medication calendar; MA and weight monitoring A: Discharge nursing consultation and written materials with education, behavioral, and support strategies; pharmacist consultation B: All of elements of (A) plus follow-up home visit for monitoring, problem solving, and reinforcing MA Pre-discharge HF and treatment education by a cardiovascular nurse; telephone follow-up; assessment of MA and symptoms Self-management counseling and skills training in a group format

Electronic monitoring

Pre-discharge HF education by a nurse, dietary and social service consultations, and medication review by a geriatric cardiologist; post-discharge followup by a nurse Training pharmacists to provide medication counseling to patients

Pill counts by research staff

Self-report by telephone interview Pill counts by research staff

Self-report

Self-report

Self-report by face-to-face interview Unclear Electronic monitoring

Pharmacy refills

399

Ruppar et al. Table 1. (Continued) Primary author (year) Region / Study duration

Study design MA outcome sample sizes

Intervention

Adherence measure

Sadik (2005)34  Asia   12 months

RCT Tx: 104 Co: 104

Self-report

Tierney (2003)35   North America   12 months

2 × 2 Factorial design, block randomization Tx-A: 71 Tx-B: 53 Tx-C: 62 Co: 73 RCT Tx: 140 Co: 136

Regimen simplification (where possible); HF, medication, and symptom management education by a pharmacist; self-monitoring of HF symptoms and MA A: Automated guideline-based care suggestions provided to physicians B: Prompts to pharmacist to review guidelinebased cardiac care suggestions C: Includes the content of both treatments A and B HF and medication education, adherence aids, written materials, and event diary provided predischarge; follow-up telephone contact for six months post-discharge Once-daily dosing regimen (versus twice-daily regimen) HF, medication, and symptom management education by a pharmacist; self-monitoring diaries

Self-report by written questionnaire Self-report

Tsuyuki (2004)36   North America   Six months Udelson (2009)37   North America   Five months Varma (1999)38  Europe   12 months Wakefield (2009)39   North America   Six months

RCT Tx: 135 Co: 131 RCT Tx: 26 Co: 23 RCT Tx-A: 28 Tx-B: 26 Co: 35

Wu (2012)40   North America   Nine months

RCT Tx-A: 27 Tx-B: 27 Co: 28

A: Post-discharge telephone follow-up by a nurse to monitor symptoms and reinforce treatment plan over three months B: Same intervention as A, but using a videophone instead of telephone A: Education by a nurse about HF symptoms, MA, beliefs/perceptions about medication; support to improve perceived behavioral control; feedback from electronic monitoring medication caps B: Same as A, but without adherence feedback

Combination of measures

Pharmacy refills

Electronic monitoring

Electronic monitoring

RCT: randomized controlled trial; Tx: treatment group; Co: control group; HF: heart failure; MA: medication adherence.

Main effects and moderator analyses The overall mean adherence effect size (Cohen’s d) was 0.29 (SE=0.09, p=0.004) for treatment versus control outcome comparisons (see Figure 2). Adherence outcome effect sizes were larger for comparisons where samples were older (see Figure 3). Effect sizes did not significantly differ based on year of publication, percentage of women in the samples, racial composition of samples, duration of the intervention, the number of intervention sessions, or the amount of time between the end of the intervention and the outcome measurement. Looking at dichotomous moderators, we found that studies conducted in Europe or Asia reported higher medication adherence effect sizes than did studies conducted in North America. We also found that funded studies had smaller effect sizes than studies that did not report research funding. Studies where patients were individually randomized to study groups had significantly larger effect sizes than did studies without individual randomization. Blinding of data collectors to subjects’ group assignment

was associated with a larger effect size when compared with studies where data collectors were not blinded or incompletely blinded. Significant dichotomous moderator results are shown in Table 2. There were no significant differences in effect size based on whether studies reported using a theoretical framework to guide the intervention (0.28 vs. 0.29, p = 0.988). Provider-focused interventions were associated with lower effect sizes. Interventions specifically incorporating methods to improve health care providers’ skills for improving patients’ medication adherence were less effective than interventions that did not attempt to improve providers’ skills. In general, interventions which targeted providers in addition to patients and/or caregivers were less effective than studies which targeted only patients or patient/caregiver groups. Studies that attempted to enhance social support as part of the intervention were less effective than studies that did not include a social support component. Not all studies focused solely on changing medication adherence, however interventions focusing on changing medication adherence behavior alone had a larger mean ES than did studies

400

European Journal of Cardiovascular Nursing 14(5)

Figure 2.  Forest plot of included two-group studies.

Summary effect heterogeneity statistics: Q=145.14, p