The Gerontologist Advance Access published May 21, 2009 The Gerontologist doi:10.1093/geront/gnp037

© The Author 2009. Published by Oxford University Press on behalf of The Gerontological Society of America. All rights reserved. For permissions, please e-mail: [email protected].

Interventions to Improve Medication Adherence Among Older Adults: Meta-Analysis of Adherence Outcomes Among Randomized Controlled Trials Vicki S. Conn, PhD, RN, FAAN,1,2 Adam R. Hafdahl, PhD,3 2 2 Pamela S. Cooper, PhD, Todd M. Ruppar, PhD, APRN, BC, 4 2 David R. Mehr, MD, MS, and Cynthia L. Russell, PhD, RN Key Words: Medication compliance, Meta-analysis

Purpose: This study investigated the effectiveness of interventions to improve medication adherence (MA) in older adults. Design and Methods: Metaanalysis was used to synthesize results of 33 published and unpublished randomized controlled trials. Random-effects models were used to estimate overall mean effect sizes (ESs) for MA, knowledge, health outcomes, and health services utilization. Results: Data were synthesized across 11,827 participants. Interventions significantly improved MA (ES = 0.33), knowledge (ES = 0.48), and diastolic blood pressure (ES = 0.19). Nonsignificant effects were found for systolic blood pressure (ES = 0.21), other health outcomes (ES = 0.04), and health services utilization (ES = 0.16). Moderator analyses showed larger adherence ESs for interventions employing special medication packaging, dose modification, participant monitoring of medication effects and side effects, succinct written instructions, and standardized (not tailored) interventions. Larger effects were found when a moderate proportion of participants were women, for participants taking 3–5 medications, and when pill count adherence was measured. Implications: The findings document that interventions increase MA in older adults. The considerable heterogeneity in the magnitude of effects across studies and results of the moderator analyses demonstrate the need for additional empirical research to optimize interventions.

adherence,

Patient

Medications to treat chronic conditions typically require a degree of patient adherence to be clinically effective. Unfortunately, older adults’ medication adherence (MA) is often unsatisfactory, with reported rates ranging from 26% to 59% (Botelho & Dudrak, 1992; van Eijken, Tsang, Wensing, de Smet, & Grol, 2003). Limits in achieving therapeutic goals are at least partially attributable to patients failing to follow their providers’ advice (DiMatteo, 2004; DiMatteo, Giordani, Lepper, & Croghan, 2002; Dunbar-Jacob & Mortimer-Stephens, 2001; Vitolins, Rand, Rapp, Ribisl, & Sevick, 2000). The odds of good health outcomes are nearly three times lower for patients who do not adhere to recommended therapies than for patients who follow provider recommendations (DiMatteo et al., 2002). Low MA increases patient and provider frustration and can increase health care costs, including avoidable hospitalizations (DiMatteo et al., 2002; Szeto & Giles, 1997; Vitolins et al., 2000). For older adults, MA difficulties may account for up to 10% of hospital admissions (Col, Fanale, & Kronholm, 1990; Sullivan, Kreling, & Hazlet, 1990). Poor MA can exacerbate disease severity and in some cases is fatal (DiMatteo et al., 2002). The annual cost of managing consequences of inadequate MA may be more than $100 billion (Dunbar-Jacob & Mortimer-Stephens, 2001). Inadequate MA is the single most important modifiable aspect of chronic disease management (Sabate,

1 Address correspondence to Vicki S. Conn, PhD, RN, FAAN, S317 Sinclair School of Nursing, University of Missouri, Columbia, MO 65211. E-mail: [email protected] 2 Sinclair School of Nursing, University of Missouri, Columbia. 3 Department of Mathematics, Washington University, St. Louis, Missouri. 4 Department of Family and Community Medicine, School of Medicine, University of Missouri, Columbia.

1

2003). A meta-analysis investigating the impact of interventions on chronic disease self-management found the largest effect sizes (ESs) for those diseases that responded well to drug therapy (Chodosh et al., 2005). The authors concluded that improved health outcomes were most likely the result of increased MA. Although numerous narrative reviews of MA interventions have been published, no conclusions about the effectiveness of the interventions could be drawn because the preponderance of studies was statistically underpowered (Haynes, Ackloo, Sahota, McDonald, & Yao, 2008). A few meta-analyses have been published, but none has focused exclusively on increasing MA in older adults. Recent syntheses have focused on particular diseases (Iskedjian et al., 2002; Mills et al., 2006; Simoni, Pearson, Pantalone, Marks, & Crapez, 2006) or specific intervention strategies (Bangalore et al., 2007; Heneghan, Glasziou, & Perera, 2006; Iskedjian et al., 2002). Peterson, Takiya, and Finley (2003) synthesized a limited number of MA interventions across participants of all ages and found modest effects, with MA improving 4%–11%. An earlier meta-analysis of studies conducted before 1994 reported larger effects for studies focusing on patients with diabetes and hypertension, particularly when MA was measured by means of pill counts, electronic monitoring, or prescription refills (Roter et al., 1998). The importance of MA for positive health outcomes coupled with low reported rates of adherence by older adults has prompted increased research to identify interventions to increase MA in this population. This article reports a meta-analysis designed to quantitatively synthesize results from primary studies in this rapidly expanding area of investigation. The research synthesis examined the following three questions: What are the overall effects of MA interventions to improve adherence behavior? What are the overall effects of MA interventions on participants’ knowledge about their medications, management of medications, disease symptoms, health outcomes, systolic and diastolic blood pressure, health care services utilization, and quality of life? Do sample demographics, intervention components, or adherence measurement methodologies moderate the effect of interventions on MA behavior?

were conducted in MEDLINE, PsycINFO, HealthStar, Ageline, Dissertation Abstracts International, Cumulative Index of Allied Health Literature, and the Cochrane Library using the following key words: medication compliance, medication adherence, patient compliance, patient adherence, drug counseling, medication education, pharmacist counseling, pharmacist consultation, prescribed regimen, self-medication, and pharmaceutical care. The database searches were performed by an expert health science information specialist, an approach that has been found to retrieve more eligible studies than when research scientists conduct searches (Conn, Isamaralai, et al., 2003; Conn, Valentine, Cooper, & Rantz, 2003; Nony, Cucherat, Haugh, & Boissell, 1995). In addition to database searches, hand searches were conducted in journals where articles on MA are most frequently published. Lastly, ancestry searches were performed on all eligible studies. Using several methods ensured a more thorough search for potentially eligible studies and helped minimize bias from too narrow searches. Broad searches are particularly important in investigational areas such as MA in which database indexing may be imprecise (Conn, Isamaralai, et al., 2003). Inclusion Criteria Inclusion criteria are summarized in Table 1. Unpublished reports were included to prevent the overestimation of ES that can result when only data from published studies are synthesized. The

Table 1. Primary Study Inclusion Criteria 1. Study reported in English between 1970 and 2007 because medication adherence (MA) behavior has changed little since 1970, although prescribed substances have improved dramatically. 2. Mean age of sample at least 60 years. 3. Sample with physical health condition requiring at least one prescription medication. 4. Interventions specifically to increase MA behavior. 5. Interventions to change multiple health behaviors if MA behavior was one of the outcomes measured. 6. Randomized controlled trial design with individual participant randomization. 7. Minimum of 5 participants with outcome data in both treatment and control arms. 8. Measures adherence behavior using actual medications rather than simulated medications. 9. Adequate data to calculate MA effect size: e.g., Ms and SD or SE, t statistics, or an exact p value from an independent t test comparing a treatment group with a control group after intervention.

Methods

Search Strategy Multiple search strategies were employed to identify English-language reports of MA interventions. Comprehensive electronic database searches 2

The Gerontologist

principal difference between published and unpublished studies is the statistical significance of the results (Conn, Valentine, et al., 2003). Studies with small sample sizes, which are common in MA science, were included. Although they often lack statistical power, small studies sometimes contribute novel interventions or target difficult-to-recruit populations. Eligible studies could address MA for any medical condition, other than psychiatric illnesses, for which at least one prescription medication was required. With the exception of psychiatric conditions, there is little evidence for disease-specific MA (Haynes, McKibbon, & Kanani, 1996). Studies in which interventions targeted multiple health behaviors were also included in the meta-analysis if MA was one of the measured outcomes.

wrong time, or wrong quantity) or MA by medication type (Isaac & Tamblyn, 1993; Martin, Bowen, Dunbar-Jacob, & Perri, 2000; Sabate, 2003). Data Management and Analysis To identify reports with potentially overlapping samples, studies were cross-checked by author names to ensure only independent samples were analyzed. Data were extracted by two extensively trained coders with graduate degrees. Discrepancies were resolved by coder consensus, and the senior author adjudicated items on which coders could not reach consensus. A standardized mean difference ES was calculated for each outcome with at least three comparisons, with a Cox approximation used for studies that reported success rates (Sanchez-Meca, MarinMartinez, & Chacon-Moscoso, 2003). Each observed ES was weighted by the inverse of its sampling variance (Hedges & Olkin, 1985). To address dependence due to comparisons with a shared control group, we used multiple-treatment composite ES (Gleser & Olkin, 1994). Outliers detected by externally standardized residuals and leave-one-out statistics (Hedges & Olkin, 1985) were removed from subsequent analyses. Homogeneity (i.e., fixed-effects model specification) was assessed with a Q statistic (Lipsey & Wilson, 2001). Random-effects analyses were used because diverse interventions and study methods were expected (Hedges & Vevea, 1998). Common Language Effect Sizes (CLES) were calculated to facilitate interpretation of findings (McGraw & Wong, 1992). Publication bias estimation was examined with ES plotted against sampling variance (Vevea & Hedges, 1995). Exploratory moderator analyses using mixedeffects analysis of variance and regression analogues were conducted (Hedges, 1994; Raudenbush, 1994). Continuous moderator analyses included both linear and cubic meta-regressions. The latter regression method better detects relationships between ES and moderators that may be more complex than linear analyses might suggest. Provisional multivariate moderator analyses were conducted using meta-regression with selected subsets of moderators. The moderator analyses should be considered suggestive rather than conclusive because small numbers of comparisons were available for several potential moderators and there were moderate to strong associations between some moderators.

Data Coding We developed, pilot tested, and refined a coding scheme to record participant demographic data, study characteristics, intervention components, and outcome data from eligible studies. Study characteristics included publication year, presence of funding, and dissemination vehicle (e.g., journal article, dissertation). Participant demographics included age, sex, race/ethnicity, income, diagnosis, presence of cognitive impairment, literacy level, and prescribed medications. Intervention components were coded (see Table 2), including whether the intervention targeted other health behaviors in addition to MA. Other intervention features coded were the intervention intensity, frequency, and duration; delivery setting; professional credentials of interventionists; and mode of delivery (i.e., either face-to-face or via mediated means such as telephone). In addition to MA outcomes, measures of medication knowledge, health outcomes, health services utilization, participants’ management of their medications, disease symptoms, and quality of life were coded. If outcomes were reported at several time points, the values most distal from the end of the intervention were coded, as these provide the most stringent test of the persistence of MA behavior. When multiple measures were reported for a single dependent variable, the measure coded was chosen based on a prioritized algorithm developed before coding began. This prevented coder and author bias and ensured that the most valid and reliable measures were coded. An insufficient number of studies provided information to permit analysis of specific medication administration errors (underdosing, overdosing, 3

Table 2. Intervention Characteristics Present in at Least Two Reports Characteristic Drug education Written instructions Dose modification Disease education Medication review

Packaging

Succinct written instructions

Side effect management Tailored intervention Medication self-monitoring

Written calendar Disease symptom self-monitoring Integration of provider care

Description

Frequency

Provide information about medications or managing medications, regardless of delivery medium. Education includes counseling. Provision of detailed written information about medications. Does not include “succinct” written instructions. Alter medication regimen to reduce frequency of administration and/or reducing number of different medications. Teaching participants about their specific disease, perhaps in relationship to medications. Pharmacist, physician, or advanced practice nurse (clinical nurse specialist or nurse practitioner) review of prescribed medications to determine potential interactions, problematic medications, and appropriate doses. Use of specific packaging as a component of the intervention. Examples include pillboxes, pill cassettes, blister packages with multiple medications per blister, “unit-of-use” packaging, and special containers indicating time of dose. Examples include drug cards, medication charts, or any written materials in a plastic sheet or laminated. Medication diaries or calendars were not considered written instructions unless they were described as including instructions. Interventions designed to minimize or manage medication side effects. Individualized assessment of participants so intervention content matches specific participant attributes. Participants directed to record medication behavior as a component of the intervention. Includes medication diaries and calendars if participants are instructed to write in these when they take their medications. Medication calendars or schedules provided to participants that specify the time to take medications. Participants instructed to record disease signs and symptoms that might be influenced by medications. Interventions designed to improve integration or coordination of care among a participant’s health care providers.

23

Likewise, the multivariate moderator analyses must be viewed as very preliminary due to substantial missing data when multiple moderators are considered; this is especially the case for continuous moderators. Additional analysis details are available from the senior author.

10 7 6 6

4

4

3 4 2

2 2 2

two were dissertations and one was a conference presentation (s indicates the number of reports, k denotes the number of comparisons). Twenty studies were externally funded. Nine studies were reported in 2000 or later, 16 reports disseminated in the 1990s, 6 in the 1980s, and 2 in the 1970s. Table 3 summarizes studies and Table 4 shows descriptive statistics for primary studies. The number of participants varied considerably from 16 to 6,813 (Mdn = 97). In studies reporting sex composition, women often comprised more than 50% of the sample (Mdn = 59%). The median of participants’ mean age was 67 years. Seven studies included African American participants, two included Hispanic participants, and one study had Native American participants. Some studies focused on participants with specific chronic illnesses (cardiovascular disease, s = 14; diabetes, s = 3; arthritis, s = 2; cancer, s = 2); only three studies reported a mean number of chronic illnesses among participants. Six

Results

Characteristics of Primary Studies Comprehensive searching yielded 43 eligible comparisons from 33 research reports (these studies are denoted with an asterisk in the reference list). Most studies had a single treatment group, but five papers had two treatment groups and three papers had three treatment groups. Eligible studies included a total of 6,235 treatment participants and 5,592 control participants for an overall total of 11,827 participants. Nearly all the primary studies were published journal articles (s = 30); 4

The Gerontologist

5

6,813 participants (50% women, M age 61 years) with hypertension 29 participants (36% women, M age 64 years) with cardiac diseases 16 participants (75% women, M age 69 years) with hypertension consuming M 3.75 drugs, selected because they had adherence problems 252 participants (69% women, M age 63 years) with arthritis

Boissel et al., 1996

114 participants (79% women, M age 77 years) consuming at least 2 drugs

88 participants (24% women, M age 65 years) with cardiac disease, excluded illiterate

Grymonpre, Williamson, & Montgomery, 2001

Halfmann, 2000

Grant, Devita, Singer, & Meigs, 2003

Gabriel, Gagnon, & Bryan, 1977

Faulkner, Wadibia, Lucas, & Hilleman, 2000

Edworthy & Devins, 1999

Burrelle, 1986

30 participants (43% women, M age 63 years) with cardiac diseases, consuming M 5.1 drugs, excluded illiterate 79 participants (M age 65 years) with hypertension consuming M 3.65 drugs (minimum 2 drugs) 115 participants (62% women, M age 67 years) with diabetes, consuming M 5.9 drugs, excluded cognitively impaired

180 participants with hypertension

Blenkinsopp, Phelan, Bourne, & Dakhil, 2000

Brun, 1994

190 participants (59% women) consuming M 5.10 drugs (minimum 3 drugs)

Sample

Begley, Livingstone, Hodges, & Williamson, 1997

Report

Pharmacy refills measured 1 newly prescribed cardiovascular drug 644 days after the intervention

Pill counts measured 1 newly prescribed arthritis drug 56 days after the intervention

Electronic event monitoring measured 1 newly prescribed cardiovascular drug 81 days after the intervention Pill counts measured M of 2.19 cardiovascular drugs

Self-report measured 1 newly prescribed cardiovascular drug 92 days after the intervention

Self-reported measured all prescribed cardiovascular drugs

Pill counts measured all prescribed drugs (M 5.10) 270 days after intervention

Outcome measurement

Drug education, medication review by health care provider to assess prescriptive appropriateness, tailored intervention, written instructions; targeted only drug adherence; delivered face-to-face in multiple settings by pharmacists; 1 session Theory of planned behavior–based intervention, social support; targeted multiple health behaviors including drug adherence; unclear setting and interventions delivered by telephone; 6 sessions for 26 weeks

(Table continues on next page)

Self-report measured all prescribed drugs

Pharmacy refills measured all drugs (M of 6.3 drugs) 365 days after the intervention

Drug adherence self-monitoring, succinct written Pill counts measured M of 3.6 drugs 30 days instructions; targeted only drug adherence; delivered after the intervention face-to-face by pharmacist in ambulatory care facilities; 1 session Drug education, drug side effect self-management training, Self-report measured diabetes drugs 90 days individually tailored; targeted only drug adherence; delivered via after the intervention telephone by pharmacist to participants’ homes; 1 session

Dose modification, drug education, staff-monitored drug adherence, written instructions; targeted drug adherence; delivered face-to-face by pharmacist in participants’ homes; 4 sessions for 12 weeks Drug education, written instructions; targeted drug adherence; delivered face-to-face or phone by pharmacist in pharmacy orparticipants’ homes; 3 sessions averaging 9.39 min/session for 16 weeks Dose modification; targeted only drug adherence; delivered face-to-face by physician in ambulatory care facility; 1 session Dose modification; targeted only drug adherence; delivered face-to-face by physician in ambulatory care facility; 1 session Disease education, drug education, special packaging, medication calendar, written instructions; targeted only drug adherence; delivered face-to-face by pharmacists and other health care providers in participants’ homes; unknown number of sessions for 8 weeks Social cognitive theory–based intervention, drug education, written instructions; targeted only drug adherence; delivered face-to-face in ambulatory care facilities; 1 session Drug education; targeted only drug adherence; delivered by pharmacist in participants’ homes; 12 sessions for 12 weeks

Intervention

Table 3. Studies Included in Medication Adherence Meta-Analysis

6

The Gerontologist

80 participants (M age 70 years) with cancer 81 participants (59% women, M age 67 years) with cardiac diseases 206 participants (M age 75 years) consuming M 5.96 drugs (minimum 3 drugs)

Klein et al., 2006

118 participants (64% women, M age 84 years), consuming M 6 drugs (minimum 4 drugs), excluded cognitively impaired 162 participants (M age 64 years) with diabetes 191 participants (43% women, M age 69 years) consuming M 3.7 drugs (minimum 2), excluded illiterate and vision impaired

Nazareth et al., 2001

Pullar, Birtwell, Wiles, Hay, & Feely, 1988 Raynor, Booth, & Blenkinsopp, 1993

McKenney, Munroe, & Wright, 1992

Lourens & Woodward, 1994

Lipton & Bird, 1994

Laporte et al., 2003

97 participants (20% women, M age 71 years) consuming M 6.25 drugs 67 participants (59% women, M age 73 years)

47 participants (85% women, M age 80 years), excluded cognitively impaired 43 participants (68% women, M age 77 years)

Jennings, Auckland, Franklin, Giles, & Austin, 1992

Kennedy, 1990

137 participants (77% women, M age 61 years)

Sample

Hawkins, Fiedler, Doublas, & Eschbach, 1979

Report

Intervention Increase health care provider time with participants; targeted multiple health care behaviors including drug adherence; delivered face-to-face by pharmacist in ambulatory care facilities; for 52 weeks Drug education; targeted only drug adherence, delivered face-to-face by pharmacist at in-patient facilities; 1 session Drug education, barriers management, packaging, medication calendar, written instructions, individually tailored; targeted only drug adherence; delivered by nurses in in-patient facilities and participants’ homes Dose modification; targeted drug adherence; delivered face-to-face by physician in ambulatory care; 1 session Drug education, written instructions; targeted only drug adherence; delivered face-to-face by physicians and other health care providers; 8 sessions for 1 week Dose modification, drug education, medication review by health care provider to assess prescriptive appropriateness, integrating medical provider care; targeted only drug adherence; delivered by pharmacist via telephone and face-to-face in variety of settings including in-patient care; five 15-min sessions for 12 weeks Drug education, written succinct instructions; targeted only drug adherence; delivered face-to-face by pharmacist in ambulatory care setting; one 15-min session Packaging, self-monitoring disease symptoms, stimulus to take medications for every dose; targeted only drug adherence; delivered face-to-face; interventions unclear; 1 session Drug education, medication review by health care provider to assess prescriptive appropriateness, packaging, integrating medical provider care, tailored, written instructions; targeted multiple behaviors including drug adherence; delivered face-to-face by pharmacist in in-patient setting; 1 session Dose modification; targeted only drug adherence; delivered face-to-face by physician in ambulatory care; 1 session Drug education, succinct written instructions; targeted only drug adherence; delivered by pharmacist in in-patient care; one 8-min session

Table 3 (continued)

(Table continues on next page)

Drug level measured 1 newly prescribed 28 days after the intervention Pill counts measured all prescribed drugs (M 3.7 drugs) 10 days after the intervention

Self-report measured all drugs (M of 6 drugs) 180 days after the intervention

Pill counts measured cardiovascular drugs 84 days after the intervention

Pill count measured all (M of 6.25 drugs) drugs 25 days after the intervention

Electronic event monitoring device measured 1 newly prescribed drug 10 days after the intervention Electronic event monitoring measured 1 newly prescribed cardiovascular drug 83 days after the intervention Self-report measured all (M of 5.96 drugs) drugs 35 days after the intervention

Pill counts measured all prescribed drugs (M of 2.4 drugs) 30 days after the intervention

Pill counts measured all prescribed drugs 7 days after the intervention

Pharmacy refills measured M of 1.5 cardiovascular drugs

Outcome measurement

7

196 participants (44% women) with cancer

84 participants (62% women, M age 80 years) consuming M two drugs 33 participants (3% women, M age 66 years) with cardiac diseases, consuming M of 6.6 drugs (minimum 2), excluded cognitively impaired and illiterate 120 participants (M age 60 years) with hyperlipidemia consuming M of 3.3 drugs 276 participants (M age 65 years) consuming M 2.9 drugs

Rimer et al., 1987

Roden, Harvey, Mayer, & Spence, 1985

132 participants (5% women, M 67 years) with hypertension, excluded illiterate 23 participants (59% women, M age 76 years) with cardiac diseases, excluded cognitively impaired

53 participants (M age 65 years) with hypertension 393 participants (90% women, M age 62 years) with arthritis 46 participants (53% women, M age 60 years) with diabetes

Solomon et al., 1998

Varma, McElnay, Hughes, Passmore, & Varma, 1999

Vivian, 2002

Weinberger, Tierney, Booher, & Katz, 1991

Wood, 1989

Simkins & Wenzloff, 1986

Schectman, Hiatt, & Hartz, 1994

Rozenfeld, Pflomm, Singh, Brazil, & Cheng, 1999

156 participants (67% women, M age 79 years) with cardiac diseases, excluded cognitively impaired, consuming M 5.2 drugs

Sample

Rich, Gray, Beckham, Wittenberg, & Luther, 1996

Report

Intervention

Drug side effect management education; targeted only drug adherence; delivered by telephone from ambulatory care by health care providers; 4.5 twenty-minute sessions Stimulus to refill prescriptions; targeted only drug adherence; pharmacist mailed or telephoned one reminder to participants’ homes Disease education, drug education; targeted multiple health behaviors including drug adherence; delivered face-to-face and by telephone by pharmacist in multiple settings; 4 sessions for 24 weeks Disease education, drug education, medication review by health care provider to assess prescriptive appropriateness, self-monitoring medications, self-monitoring symptoms, written instructions; targeted multiple health behaviors including drug adherence; delivered face-to-face by pharmacist in ambulatory care facility; 4 sessions for 36 weeks Drug education, medication review by health care provider for prescriptive appropriateness; targeted drug adherence; delivered face-to-face by pharmacist in ambulatory care; 6 sessions for 20 weeks Disease education, drug education; targeted multiple health behaviors including drug adherence; delivered by telephone and/or face-to-face in multiple locations; averaged 3–10 sessions for 44 weeks Disease education, drug education; targeted multiple health behaviors including drug adherence; delivered face-to-face while in-patient by health care providers, two 120-min sessions in 1 week

Disease education, dose modification, drug education, medication review by health care provider to assess prescriptive appropriateness; targeted multiple health behaviors including drug adherence; delivered face-to-face by physician and other health care providers in in-patient care; 1 session Drug education, succinct written instructions; targeted multiple health behaviors including drug adherence; delivered face-to-face by health care providers in ambulatory care setting; one 15-min session Drug education, labeling; targeted drug adherence only; delivered face-to-face by pharmacist in in-patient care; 1 session Drug education, drug side effect management education, written instructions; targeted only drug adherence; delivered face-to-face by pharmacist in ambulatory care; 1 session

Table 3 (continued)

Self-report measured diabetes drugs 120 days after the intervention

Self-report measured drugs 30 days after the intervention

Pharmacy refills measured cardiovascular drugs 30 days after the intervention

Pharmacy refills measured cardiovascular drugs

Self-report measured all prescribed cardiovascular drugs 35 days after the intervention

Pharmacy refills measured cardiac drugs

Pharmacy refills measured 1 newly prescribed lipid drug 32 days after the intervention

Electronic event monitoring device measured cardiovascular drugs (M of 2 drugs) 34 days after the intervention

Pill counts measured all drugs (M of 2.1 drugs) 16 days after the intervention

Self-report measured pain drugs 28 days after the intervention

Pill counts measured all prescribed drugs (M of 5.2 drugs) 30 days after the intervention

Outcome measurement

Table 4. Characteristics of Primary Studies in Medication Adherence Meta-Analysis Characteristic Sample size per study Proportion of sample assigned to treatment group Proportion attrition from treatment group Proportion attrition from comparison group Proportion women M age (years) No. of prescribed medications No. of medications measured for adherence Minutes per motivational/educational session No. of motivational/educational sessions No. of weeks intervention was delivered

k

Minimum

Q1

33 33 33 33 26 30 14 23 6 29 33

16 0.45 0.00 0.00 0.00 60 2 1 9 1 1

53 0.49 0.00 0.00 0.43 64 4 1 15 1 1

Mdn 97 0.52 0.04 0.02 0.59 67 5 2 15 1 1

Q3 196 0.57 0.15 0.13 0.67 74 6 4 19 4 12

Maximum 6,813 0.76 0.38 0.45 0.89 84 7 6 120 12 52

Note: Includes all studies that contributed at least one independent-groups effect size to primary analyses. Independent samples within studies aggregated by summing sample sizes before computing proportions and using weighted mean of other characteristics (weighted by sample size). k = number of studies providing data on characteristic; Q1 = first quartile, Q3 = third quartile.

studies excluded cognitively impaired individuals, and one study excluded individuals who had vision problems. Five studies excluded individuals unable to read. None of the studies specifically targeted individuals with cognitive, visual, or reading impairments. Only one study specifically recruited participants with MA difficulties. Most studies employed a no-treatment or usual care comparison group (s = 24), with the remainder using an attention-control group (s = 9). Attrition was modest from both treatment (Mdn = 4%) and comparison (Mdn = 2%) groups. Most studies did not report the number of prescribed medications per participant. In the 14 studies reporting this information, the median medications per participant was 5. The median number of medications used for MA measurement was 2. Thirteen studies measured MA for all prescribed medications. Eight studies measured MA only for newly prescribed medications. MA was measured via pill counts by research staff (s = 11), self-report (s = 10), pharmacy refills (s = 7), electronic event monitoring (s = 4), and monitoring drug levels (s = 1). Although most measures captured the number of doses taken compared with the number prescribed, only five studies assessed when doses were taken relative to the prescribed times. None of the measures addressed drug holidays (days on which medications were stopped). There was some evidence of a ceiling effect on MA, such as less variability in groups with higher means.

many weeks (Table 4). Theoretical frameworks for interventions were rarely mentioned (social cognitive theory, s = 1; theory of planned behavior, s = 1). Intervention strategies employed in at least 2 of the 33 studies in the meta-analysis sample are described in Table 2. Other intervention strategies found only in single reports included labeling changes, monitoring MA by research staff, stimuli or cues for taking medications on time, prompts to refill prescriptions, barriers management, providers spending additional time with participants, and social support. Most interventions targeted only MA, whereas a few interventions focused on multiple health behaviors. Three studies focused on individuals and their family members involved in administering medications, and the rest of the studies focused on medication self-administration. Interventions were most often delivered in ambulatory care settings (s = 13), in-patient facilities (s = 8), participants’ homes (s = 6), or some combination of these settings (s = 4). Interventions were nearly all delivered to participants on an individual basis; only one intervention was delivered in a group setting. Nine reports indicated that participants were in-patients in acute care or rehabilitation facilities when recruited into the study. Pharmacists were the most common interventionists (s = 19), followed by physicians (s = 6) and nurses (s = 1). Most interventions were delivered face-to-face (s = 26); 10 used the telephone and 1 involved mailed materials.

Intervention Features Overall Effects of Interventions

Typical interventions were single sessions of short duration (Mdn = 15 min), although a few studies tested intensive interventions delivered over

Table 5 shows the estimated overall mean ES ( µˆ δ) and associated statistics calculated for MA, 8

The Gerontologist

Table 5. Random-Effects Outcome Variable: Point and Interval Estimates and Tests Variable Adherence Knowledge Health service utilization Systolic blood pressure Diastolic blood pressure Health outcome

k

Q

µˆ δ

SE(µˆ δ)

md 95% CI

σˆ δ

CLES

ab 41 , cd 13 , e 7 8c 8c 5

104.30*** 51.27*** 14.55* 47.55*** 14.11* 6.64

0.33*** 0.48*** 0.16 0.21 0.19* 0.04

0.059 0.143 0.096 0.159 0.091 0.125

0.22 to 0.45 0.20 to 0.76 −0.03 to 0.35 −0.10 to 0.52 0.01 to 0.37 −0.20 to 0.29

0.229 0.355 0.152 0.320 0.147 0.172

0.59 0.63 0.55 0.56 0.55 0.51

Notes: Under homogeneity (H0: di = d), Q is distributed approximately as chi square with df = k – 1, where k is the number of (possibly dependent) observed effect sizes; this test also applies to the between-studies variance component, σˆ 2δ (H0: σˆ 2δ = 0). Weighted method of moments used to estimate σˆ 2δ (tabled value is standard deviation of true effect sizes). Potential outliers excluded based on univariate random-effects standardized residuals. CLES = Common Language Effect Size a Includes four dependent multiple-treatment pairs. b Includes three dependent multiple-treatment triplets. c Includes one dependent multiple-treatment triplet. d Includes two dependent multiple-treatment pairs. e Includes one dependent multiple-treatment pair. *p < .05, **p < .01, ***p < .001 (for d , Q, and µˆ δ).

medication knowledge, health services utilization, diastolic and systolic blood pressure, and other health outcomes. (Quality of life, symptoms, and medication management outcomes were not synthesized because only two comparisons were available for each.) The interventions had their greatest effects on MA ( µˆ δ = 0.33) and medication knowledge ( µˆ δ = 0.48). These values were statistically significant, as was the more modest ES calculated for diastolic blood pressure ( µˆ δ = 0.19). The effect of interventions on systolic blood pressure ( µˆ δ = 0.21) was similar to that of diastolic blood pressure but its confidence interval included 0 (due mainly to more heterogeneity). The ESs were not significantly different from zero for other health outcomes ( µˆ δ = 0.04) or for health services utilization ( µˆ δ = 0.16), such as emergency/urgent care contacts, ambulatory visits, or hospitalizations. The Q statistics documented significant heterogeneity among studies for all variables except health outcome, for which power may be low. One study’s sample was several times larger than those in any other study (Boissel et al., 1996). When analyses were repeated excluding this larger study, results for medication knowledge, health services utilization, and health outcomes were unaffected. ESs were slightly larger for MA ( µˆ δ = 0.34) and diastolic blood pressure ( µˆ δ = 0.27). The only substantial difference in ESs when this study was omitted was for systolic blood pressure ( µˆ δ = 0.56). However, because this large study’s ES did not appear to be an outlier, the study was included in subsequent analyses.

Table 5 also shows the CLES calculated for each outcome’s mean ES. The CLES for a two-group comparison indicates the probability that a random treatment participant will have a better outcome than a random control participant. Thus, the 0.59 CLES for MA indicates that on average (across studies) a random treatment participant will evidence better MA than a random control participant 59% of the time. Although publication bias was not strongly evident in the funnel plot for MA, some sparseness of ESs near zero among smaller studies is consistent with a publication process that favors positive effects. Funnel plots could not be assessed for other outcomes owing to the small number of studies in the samples.

Moderator Analyses for MA Outcomes Results from analyzing dichotomous and continuous moderators are shown in Tables 6 and 7. Multiple–degree of freedom (df) categorical moderator analysis tables are available from the senior author. Neither funding status nor publication status was a significant moderator of MA outcomes (Table 6). Publication year was significantly related to MA in the cubic model but not in the linear model (Table 7); plots of fitted mean ES based on the cubic model suggest that studies reported prior to 1980 and between 1990 and 2000 had larger MA ESs than studies reported between 1980 and 1990 and studies reported after 2000. 9

Table 6. Dichotomous Moderator Analyses for Medication Adherence Moderator

k0

k1

QWa

µˆ δ0

µˆ δ1

QB

σˆ δ

Funded Published Measured all medications Measured only new meds Control group type Low incomeb Specific chronic illness Excluded cognitively impairedb Excluded illiterate Dose modificationb Packaging Professional medication review Side effect management b Self-monitoring medication effects b Stimulus every time Written instructions—succinctb Written instructions—any Disease education b Drug education Tailored interventionb Behavioral targetb Person administering drugs

17 3 20 11 31 37 16 35 35 33 36 35 38 38 38 35 26 35 17 37 10 5

24 38 16 9 10 4 25 6 6 8 5 6 3 3 3 6 15 6 24 4 31 36

102.9 102.4 88.0 57.8 102.8 97.6 100.2 98.2 103.1 97.0 100.8 101.2 101.5 89.9 90.9 97.1 101.5 100.7 98.5 92.2 99.0 100.0

0.44*** 0.27 0.39*** 0.41** 0.32*** 0.36*** 0.29*** 0.36*** 0.33*** 0.32*** 0.30*** 0.36*** 0.34*** 0.30*** 0.30*** 0.29*** 0.28*** 0.36*** 0.42*** 0.37*** 0.24* 0.23

0.28*** 0.34*** 0.30*** 0.45*** 0.39*** 0.13 0.38*** 0.17 0.43** 0.39*** 0.67*** 0.22 0.31 1.18*** 1.06*** 0.61*** 0.45*** 0.24 0.28*** 0.06 0.37*** 0.36***

2.1 0.1 0.6 0.1 0.3 2.1 0.6 1.6 0.3 0.3 4.1* 0.8 0.0 10.6** 9.5** 4.4* 2.3 0.6 1.6 3.6 1.2 0.8

0.248 0.236 0.251 0.311 0.243 0.230 0.245 0.231 0.238 0.253 0.234 0.239 0.235 0.210 0.212 0.227 0.243 0.237 0.247 0.218 0.236 0.238

Notes: kj = number of (possibly dependent) effect size estimates in group coded j. Moderator levels: measured only new meds (0 = some meds measured for adherence not new; 1 = only new meds measured), control group type (0 = true control; 1 = attention control), low income (0 = not focused on low-income participants; 1 = more than 50% of participants of low income), specific chronic illness (0 = participants had diverse/mixed illnesses; 1 = participants had specific illness), behavioral target (0 = multiple behaviors; 1 = meds only), person administering drugs (0 = family involved; 1 = participant only), and all others (0 = absent or no; 1 = present or yes). Heterogeneity statistics: QB = between groups (distributed as chi square on df = 1 under H0: µ δ0 = µ δ1), QW = combined within groups (distributed as chi square on df = k0 + k1 – 2 under H0: σ 2δ0 = σ 2δ1 = 0). Weighted method of moments used to estimate between-studies variance component σ 2δ . Analysis reported if k0 ≥ 3 and k1 ≥ 3. a For all reported moderators, QW yielded p < .001. b Significant fixed-effects QB, p < .05. *p < .05, **p < .01, ***p < .001 (for d0, d1, QB, QW, µˆ δ0, and µˆ δ1).

Sample Demographics.—The number of prescribed medications was related to MA. Studies where participants received three to five medications had larger MA outcomes than studies where participants received fewer or more medications. Mean ES was unrelated to sample age, income, cognitive status, literacy, or chronic illnesses. A quadratic model was significant for proportion of women in the sample (p < .05).

research design features such as number of medications measured, type of control group, attrition, and days to follow-up MA measurement after intervention. Intervention Features.—MA interventions that included medication packaging changes were associated with larger ES ( µˆ δ = 0.67) than interventions that did not include packaging changes ( µˆ δ = 0.30). Examples of packaging interventions include pillboxes, pill cassettes, blister packages with multiple medications in each blister, and special containers that indicate the time of the dose. Interventions with dose modification ( µˆ δ = 0.39) were not more effective than interventions without dose changes ( µˆ δ = 0.32). Interventions that directed participants to selfmonitor symptoms related to medications (including symptom improvement from taking medications

Methodology.—Multiple-df categorical moderator analyses revealed that mean ES varied with the method used to measure adherence (Qbetween = 8.4, p = .038). Measurement methods included electronic event monitoring, pharmacy refills, pill counts, and self-reports. Pairwise comparisons revealed that the dominant significant difference was between pill counts ( µˆ δ = 0.52) and selfreports ( µˆ δ = 0.17). Mean ES was unrelated to other 10

The Gerontologist

Table 7. Linear and Cubic Continuous Mixed-Effects Moderator Analyses for Medication Adherence σˆ δ

Qmodel Moderator

k

M

SD

L

C

L

C

Publication year (compared to 1975) Number of medications measured M age (years) % women No. of prescribed meds Minutes of intervention No. of intervention session (log10) No. of weeks of intervention (log10) Follow-up number of days (log10) Attrition proportion

41 29 37 32 18 8 36 41 33 40

19.63 2.00 64.73 54.41 4.61 50.35 0.14 0.88 1.86 0.08

5.57 2.32 6.96 17.63 1.40 56.78 0.35 0.57 0.39 0.11

3.4 1.3 0.2 0.5 0.0 0.0 0.4 0.0 0.0 1.2

16.1** 4.0 5.2 6.7 11.1* 3.0 0.5 9.1* 1.4 1.4

0.233*** 0.244*** 0.262*** 0.236*** 0.331*** 0.331** 0.269*** 0.244*** 0.256*** 0.247***

0.199*** 0.240*** 0.269*** 0.233*** 0.260** 0.366** 0.283*** 0.229*** 0.289*** 0.258***

Notes: k = number of (possibly dependent) effect size estimates. Each moderator’s weighted mean and standard deviation computed from all available cases, with unconditional weight 1/vi. Polynomial models with degree m = 1 or 3: L = linear; C = cubic. Qmodel = model heterogeneity statistic due to all polynomial terms of moderator (x) in linear (b1x) or cubic (b1x + b2x2 + b3x3) model, distributed as chi square on df = m under H0: b = 0, where b = b1 or [b1 b2 b3]T. Weighted method of moments used to estimate between-studies variance component σ 2δ . Analysis reported if k ≥ m + 5. *p < .05, **p < .01, ***p < .001 (for Qmodel and σ 2δ ).

and medication side effects) were considerably more effective ( µˆ δ = 1.18) than interventions that did not include this component ( µˆ δ = 0.30). Differences in mean ES were not associated with the presence or absence of teaching strategies to manage medication side effects. Likewise, ES differences were not associated with the presence or absence of review of medications by health care professionals to determine prescription appropriateness. Neither medication nor disease education had any significant impact on ES. However, interventions with succinct written instructions achieved better effects on MA ( µˆ δ = 0.61) than studies without succinct written instructions ( µˆ δ = 0.29). The difference between providing any written directions ( µˆ δ = 0.45) and no written directions ( µˆ δ = 0.28) was not significant. Interventions that included a stimulus to take medication, such as an electronic device that makes a sound each time medications should be administered, were more effective ( µˆ δ = 1.06) than interventions without these cues ( µˆ δ = 0.30). Mean ES did not differ between studies where participants administered their own medications and those where informal caregivers administered medications. Interventions that were individually tailored to specific participant characteristics were less effective ( µˆ δ = 0.06) than interventions that were more standardized across participants ( µˆ δ = 0.37). Interventions that focused exclusively on MA were not more effective than those that targeted multiple health behaviors (e.g., MA plus diet). Multiple-df categorical

moderator analyses revealed that interventionists’ professional preparation (e.g., physician, pharmacist), intervention delivery setting (clinic, home, inpatient, or combined), and mediated delivery (face-to-face only, mediated delivered only, or both) had no significant effect on mean ES. Neither the minutes of intervention nor the number of intervention sessions significantly affected mean ES. Cubic polynomial models indicated that MA was related to the number of weeks over which the intervention was delivered (Table 7). Mean predicted ES was highest for interventions delivered for 4 weeks (ES approximately 0.7) but was markedly reduced when interventions were either of brief duration or very prolonged. For example, mean predicted ESs were approximately 0.2 and 0.3 for interventions delivered for 1- or 32-week period, respectively. Multiple-Moderator Models.—Some exploratory mixed-effects multiple-moderator models were examined using subsets of the 20 dichotomous moderators available for all 41 studies (correlations among moderators, some of which are substantial, are available from the senior author). The initial approach was a severely unbalanced and sparse four-way layout based on the four moderators that were individually significant by mixed-effects tests (Table 6); only 6 of the 16 cells had any studies, with 29 studies and most of the weight in the cell with all four components absent. Only main effects were included—the six 2-way marginals were also 11

unbalanced and sparse, and the two estimable interactions were perfectly confounded—and the least significant moderator was successively eliminated until all remaining terms were significant at a = .10. The reduced model contained self-monitoring of medication effects (βˆ = .93, p = .001) and succinct written instructions (βˆ = .35, p = .014) and was statistically significant, Qmodel(2) = 16.8, p < .001, with a significant variance component (σˆ 2δ = 2 0.202 , p < .001). The same approach was used with a 10-way layout based on the 10 moderators individually significant by a fixed- or mixed-effects test (Table 6); only 20 of the 1,024 cells had any studies, and 11 had just one study. This design’s reduced model also contained self-monitoring of medication effects (βˆ = .60, p = .042) and succinct written instructions (βˆ = .40, p = .005) and was statistically significant, Qmodel(5) = 27.5, p < .001, with a sig2 nificant variance component (σˆ 2δ = 0.171 , p = .006). This larger model, however, additionally included dose modification (βˆ = .20, p = .076), packaging (βˆ = .40, p = .051), and tailored intervention (βˆ = −.33, p = .054). Dropping any of these moderators yielded a larger, more significant σˆ 2δ and a less significant model. Third, more exhaustive searches of all 15 and 1,023 nonempty submodels of the 4- and 10-way layouts—by fitting a separate main-effects model with each single moderator, each pair, each triplet, and so forth—yielded the same two- and five-moderator models as the aforementioned backward elimination strategy, although other similar models differing by one or two specific terms were nearly as good. Applying this all-subsets approach to all 20 dichotomous moderators (i.e., 1,048,575 nonempty submodels) yielded a notably different final model that excluded self-monitoring of medication effects but still included tailored intervention (βˆ = −.76, p < .001), packaging (βˆ = .71, p < .001), dose modification (βˆ = .23, p = .026), and succinct written instructions (βˆ = .35, p = .041). This final model also added five new moderators: funded (βˆ = −.33, p = .001), side effect management (βˆ = .60, p = .005), any written instructions (βˆ = .29, p = .014), person administering drugs (βˆ = .23, p = .058), and excluded illiterate (βˆ = −.24, p = .151). The model was significant, Qmodel(9) = 45.1, p < .001, with a nonsignificant variance com2 ponent (σˆ 2δ = 0.132 , p = .074). About 2,000 (0.2%) of the submodels were arguably better than the five-moderator model arrived at from the 10 singly significant moderators. More than 99% of these

better submodels included dose management, side effect management, and tailored intervention, and more than 90% included funded, packaging, and any written instructions. Most of the other 14 moderators occurred in fewer than 50% of the better submodels. Discussion

Overall Effects of Intervention on MA The 0.33 ES on MA in this study is between previously reported values of 0.08 and 0.74 (Devine & Reifschneider, 1995; Peterson et al., 2003), but comparisons with these syntheses are difficult. Whereas the present sample was not restricted to any specific type of medication, Devine and Reifschneider’s synthesis was limited to antihypertensive drugs. When more studies have been completed, comparisons by pharmacological categories may be possible. The meta-analysis by Peterson and colleagues incorporated participants of diverse ages, including children. The authors did not assess the effect of age on MA of ES, so it is unclear if a particular age group contributed to the higher overall ES. The 0.33 overall ES documents that at least some interventions effectively increase MA. The extent of heterogeneity in ES across studies included in the analysis documents that not all interventions are equally effective in improving MA; indeed, a small proportion may have a nil or negative effect. Research participants often receive optimal medical care (Clark, Hartling, Vandermeer, & McAlister, 2005) that may limit further MA improvements from interventions so the amount of improvement in the general population is difficult to predict. Overall Effects on Medication Knowledge, Health Outcomes, and Health Care Services Utilization The ES for knowledge was larger than the ES for any of the other outcomes examined. This finding likely reflects the strong emphasis on patient education activities in the interventions (Table 2). The ES estimates for health outcomes and service utilization were very modest. Reaching conclusions regarding whether the relatively small effect of interventions on these two end points is sufficient or should be of concern is difficult because no clear criteria exist for the minimum MA necessary to affect health (Vitolins et al., 2000). Some authorities suggest that 80% adherence is probably adequate for most drugs (Dunbar-Jacob & Mortimer-Stephens, 2001; Rudd, 12

The Gerontologist

1979), whereas others set more stringent limits— often 90% or above (Szeto & Giles, 1997). The association between MA and health outcomes needs to be more clearly defined either through additional empirical research or through meta-analysis.

confidence in the effectiveness of their medications beyond what is conveyed by health care providers. If patients clinically improve on the medication, self-monitoring may also provide positive feedback that self-administration is successful. This may be a useful strategy for medications where clinical outcomes or measures of mediating constructs can be directly assessed by patients, such as blood pressure or blood glucose. The moderate ES of 0.48 for knowledge outcomes is consistent with the predominant focus on educating patients in interventions intended to increase MA. The univariate moderator analysis hints that interventions that include disease and drug education may actually be less effective than interventions that do not; however, the differences in ES did not achieve statistical significance. Many interventions continue to focus on educating participants despite evidence that knowledge is variably related to MA. Although knowledge may be prerequisite for MA, education alone is generally not adequate to achieve adherence (Mullen et al., 1985; Ryan, 1999; Sabate, 2003). Purely educational interventions may be useful when nonadherence is deliberate, otherwise behavior-based interventions may be more effective (Lehane & McCarthy, 2007). The apparent negative effect of education on MA of ES observed in the moderator analysis might simply be an issue of the fraction of total intervention time spent on education, often at the expense of behavior-based strategies that could have greater impact. In contrast to verbal instruction, written instructions appear to be more effective in increasing MA than other forms of patient education, with succinct written materials being the most effective. Investigation of the moderating effects of study design features documented significant differences in ES depending on the method used to measure MA. Although self-reported adherence measurement may misrepresent true adherence (Szeto & Giles, 1997), this method of assessing MA continues to be widely employed in MA intervention research. The findings from the moderator analyses are intriguing but nonetheless should be interpreted cautiously given the small sample size, absence of predictions from the literature, associations among moderators, and substantial residual betweenstudies heterogeneity. The multivariate moderator analyses on MA showed that this approach could be useful for identifying patterns among studies. Unfortunately, most studies varied in only a small

Moderators of MA Univariate and multivariate moderator analyses were conducted to determine the relative importance of intervention components and study design features on MA. The moderator analysis findings are important because few primary research studies are designed to detect differences related to intervention characteristics (Haynes et al., 2008). Exploratory moderator analysis can therefore suggest future directions for research and practice. The finding that MA interventions may be most effective for participants receiving three to five medications may be explained by differing degrees of difficulty in managing few versus many medications. Taking one or two medications may be easy enough that interventions produce little discernible effect, whereas managing six or more medications may be so difficult that specialized interventions are required. Alternatively, medication number might be a proxy for other attributes such as medication side effects, chronic illnesses features, or even aspects of cognitive function. This is a particularly important issue in chronically ill older individuals, who often take multiple medications. Among the intervention characteristics found to be positive moderators of MA were special packaging, dose modification, and stimuli prompting participants to take medications. That improved MA was associated with three behavior-based strategies is consistent with results from other syntheses in which behavioral interventions outperformed cognitive ones in improving MA. Roter and colleagues (1998) found that behavioral interventions produced better MA than cognitive interventions. Mullen, Green, and Persinger (1985) found behavior-based methods such as practicing self-administration, special pill containers, and memory aids to be more effective than written medication information in reducing patient medication errors. Peterson and colleagues (2003) reported larger ESs for medication prompts than for educational interventions. Interventions that directed participants to selfmonitor medication effects were more effective than interventions without this component. Selfmonitoring drug effects may help increase patients’ 13

number of the moderators found to be linked with ES. Once a more primary research is available and evidence or theory suggests conceptual reasons for examining particular joint moderators, multivariate analysis will be an important strategy to detect which combinations of intervention strategies are most effective in improving MA.

strategies are needed to clarify the relative importance of these two approaches to MA. Continued effort to ensure that studies without statistically significant findings are published is important to ensure that maximum information is available to inform practice and future research. Another area in which additional research is required is the usefulness of tailored interventions in increasing MA. The moderator analysis suggested that individualized interventions are less effective than generalized interventions; however, the analysis was based on a very small sample of tailored interventions and was subject to confounding with other study features. The small sample points to the need for more design and testing of tailored interventions as well as side-by-side comparisons of tailored and nontailored approaches. Carefully designed primary studies or prospective meta-analyses would be effective strategies for disentangling moderating and mediating relationships that are difficult to resolve using retrospective meta-analysis. In the latter case, combining moderators often yields highly unbalanced and sparse data arrangements in which multiple moderators and their joint effects are confounded. Future research should include clinical outcomes so that the consequences of increasing MA can be fully evaluated (Haynes et al., 2008).

Limitations This meta-analysis was limited by the number of studies retrieved despite extensive searching. Limitations on the participant and intervention characteristics that could be coded from the studies were similar to those found by others who have synthesized MA and other health behavior change research (Conn, Hafdahl, Brown, & Brown, 2008; Haynes et al., 2008; Mullen et al., 1985; Peterson et al., 2003). This synthesis was restricted to randomized controlled trials because these designs provide the most rigorous tests of intervention effectiveness. Other research designs may provide information on effectiveness of other kinds of interventions. Sample size placed limitations on the moderator analysis. Some potentially important moderators such as medication type could not be examined because they occurred too infrequently in the sample. The small number of studies in the meta-analysis means that conclusions from the moderator analysis must necessarily be viewed as tentative. Despite these limitations, the synthesis demonstrates that interventions do influence MA and MA-associated outcomes in older adults.

Practice Recommendations and Conclusions These findings, along with other reviews of MA and other health behaviors, suggest that interventions should focus on behavioral strategies to increase MA. For example, interventions that decrease the number of doses and employ special packaging or prompts to take medications may be effective. Prompts could include audible devices that cue medication administration. Some of these devices also dispense the correct medications at the proper time. Interventions should include patients’ selfmonitoring of medication effects when possible. Although teaching patients about their medications and disease is important, less emphasis should be placed on education relative to behavioral strategies to increase MA. Patient education should involve less verbal instruction, and individuals should be provided clear succinct written instructions that they can refer to at home. Interventions can be offered to diverse individuals because most patient characteristics were not found to be significant moderators of MA. However, interventions may be less effective for patients

Future Research The synthesis served to identify several areas in which additional primary research is needed. In particular, the cumulative evidence of this metaanalysis and work of other investigators suggests that interventions emphasizing behavioral rather than cognitive strategies will produce better MA outcomes. Only a very few behavioral strategies were employed in the studies included in this metaanalysis, and some important ones were glaringly absent. For example, although experts recognize the importance of habit in MA (Sabate, 2003), none of the interventions in the meta-analysis targeted habit change. Future research should place greater emphasis on testing behavior-based approaches to MA. Also, rigorous randomized trials that directly compare cognitive versus behavioral 14

The Gerontologist

DiMatteo, M. (2004). Variations in patients’ adherence to medical recommendations: A quantitative review of 50 years of research. Medical Care, 42, 200–209. DiMatteo, M., Giordani, P., Lepper, H., & Croghan, T. (2002). Patient adherence and medical treatment outcomes: A meta-analysis. Medical Care, 40, 794–811. Dunbar-Jacob, J., & Mortimer-Stephens, M. (2001). Treatment adherence in chronic disease. Journal of Clinical Epidemiology, 54(Suppl. 1), S57–S60. *Edworthy, S. M., & Devins, G. M. (1999). Improving medication adherence through patient education distinguishing between appropriate and inappropriate utilization. Patient Education Study Group. Journal of Rheumatology, 26, 1793–1801. *Faulkner, M. A., Wadibia, E. C., Lucas, B. D., & Hilleman, D. E. (2000). Impact of pharmacy counseling on compliance and effectiveness of combination lipid-lowering therapy in patients undergoing coronary artery revascularization: A randomized, controlled trial. Pharmacotherapy, 20, 410–416. *Gabriel, M., Gagnon, J. P., & Bryan, C. K. (1977). Improved patients compliance through use of a daily drug reminder chart. American Journal of Public Health, 67, 968–969. Gleser, L., & Olkin, I. (1994). Stochastically dependent effect sizes. In H. Cooper & L. Hedges (Eds.), The handbook of research synthesis (pp. 339–355). New York: Russell Sage. *Grant, R. W., Devita, N. G., Singer, D. E., & Meigs, J. B. (2003). Improving adherence and reducing medication discrepancies in patients with diabetes. Annals of Pharmacotherapy, 37, 962–969. *Grymonpre, R. E., Williamson, D. A., & Montgomery, P. R. (2001). Impact of a pharmaceutical care model for non-institutionalised elderly: Results of a randomised, controlled trial. International Journal of Pharmacy Practice, 9, 235–241. *Halfmann, S. M. (2000). Peer support with a nurse-managed intervention and compliance after a cardiac event. Unpublished doctoral dissertation, Texas Woman’s University, Denton. *Hawkins, D. W., Fiedler, F. P., Douglas, H. L., & Eschbach, R. C. (1979). Evaluation of a clinical pharmacist in caring for hypertensive and diabetic patients. American Journal of Hospital Pharmacy, 36, 1321–1325. Haynes, R., Ackloo, E., Sahota, N., McDonald, H., & Yao, X. (2008). Interventions for enhancing medication adherence. Cochrane Database of Systematic Reviews (4), CD000011. Haynes, R., McKibbon, K., & Kanani, R. (1996). Systematic review of randomised trials of interventions to assist patients to follow prescriptions for medications. Lancet, 348, 383–386. Hedges, L. (1994). Fixed effects models. In H. Cooper & L. Hedges (Eds.), The handbook of research synthesis (pp. 285–299). New York: Russell Sage. Hedges, L., & Olkin, I. (1985). Statistical methods for meta-analysis. Orlando, FL: Academic Press. Hedges, L., & Vevea, J. (1998). Fixed- and random-effects models in meta-analysis. Psychological Methods, 3, 486–504. Heneghan, C., Glasziou, P., & Perera, R. (2006). Reminder packaging for improving adherence to self-administered long-term medications. Cochrane Database of Systematic Reviews (1), CD005025. Isaac, L., & Tamblyn, R. (1993). Compliance and cognitive function: A methodological approach to measuring unintentional errors in medication compliance in the elderly. McGill-Calgary Drug Research Team. Gerontologist, 33, 772–781. Iskedjian, M., Einarson, T., MacKeigan, L., Shear, N., Addis, A., Mittmann, N., et al. (2002). Relationship between daily dose frequency and adherence to antihypertensive pharmacotherapy: Evidence from a meta-analysis. Clinical Therapeutics, 24, 302–316. *Jennings, M., Auckland, J., Franklin, G., Giles, R., & Austin, C. A. (1992). Counseling does not improve compliance with drug therapy. Age and Ageing, 21(Suppl. 2), 13–14. *Kennedy, L. M. (1990). Effectiveness of a self-care medication education protocol on the home medication behaviors of recently hospitalized elderly. Unpublished doctoral dissertation, University of Texas, Austin. *Klein, C. E., Kastrissios, H., Miller, A. A., Hollis, D., Yu, D., Rosner, G. L., et al. (2006). Pharmacokinetics, pharmacodynamics and adherence to oral topotecan in myelodysplastic syndromes: A Cancer and Leukemia Group B study. Cancer Chemotherapy and Pharmacology, 57, 199–206. *Laporte, S., Quenet, S., Buchmuller-Cordier, A., Reynaud, J., Tardy-Poncet, B., Thirion, C., et al. (2003). Compliance and stability of INR of two

taking only one or two medications compared with those taking three to five. More intense interventions or different approaches altogether may be required for patients taking six or more prescribed drugs. In conclusion, although interventions can increase MA, there is much room for improvement. This research synthesis of interventions and analysis of moderating variables serves as an important basis for developing additional interventions that are particularly effective in older persons. And to ensure that maximum data are available to inform future research and clinical practice, every effort should be made to publish findings regardless of statistical significance. Given the potential major health consequences of medication nonadherence, additional primary research on ways to increase MA is urgently needed.

References *References marked with an asterisk indicate studies included in the meta-analysis. Bangalore, S., Kamalakkannan, G., Parkar, S., & Messerli, F. (2007). Fixed-dose combinations improve medication compliance: A metaanalysis. American Journal of Medicine, 120, 713–719. *Begley, S., Livingstone, C., Hodges, N., & Williamson, V. (1997). Impact of domiciliary pharmacy visits on medication management in an elderly population. International Journal of Pharmacy Practice, 5, 111–121. *Blenkinsopp, A., Phelan, M., Bourne, J., & Dakhil, N. (2000). Extended adherence support by community pharmacists for patients with hypertension: A randomised controlled trial. International Journal of Pharmacy Practice, 8, 165–175. *Boissel, J. P., Meillard, O., Perrin-Fayolle, E., Ducruet, T., Alamercery, Y., Sassano, P., et al. (1996). Comparison between a bid and a tid regimen: Improved compliance with no improved antihypertensive effect. The EOL Research Group. European Journal of Clinical Pharmacology, 50, 63–67. Botelho, R., & Dudrak, R., II (1992). Home assessment of adherence to long-term medication in the elderly. Journal of Family Practice, 35, 61–65. *Brun, J. (1994). Patient compliance with once-daily and twice-daily oral formulations of 5-isosorbide mononitrate: A comparative study. Journal of International Medical Research, 22, 266–272. *Burrelle, T. N. (1986). Evaluation of an interdisciplinary compliance service for elderly hypertensives. Journal of Geriatric Drug Therapy, 1, 23–51. Chodosh, J., Morton, S., Mojica, W., Maglione, M., Suttorp, M., Hilton, L., et al. (2005). Meta-analysis: Chronic disease self-management programs for older adults. Annals of Internal Medicine, 143, 427–438. Clark, A., Hartling, L., Vandermeer, B., & McAlister, F. (2005). Metaanalysis: Secondary prevention programs for patients with coronary artery disease. Annals of Internal Medicine, 143, 659–672. Col, N., Fanale, J., & Kronholm, P. (1990). The role of medication noncompliance and adverse drug reactions in hospitalizations of the elderly. Archives of Internal Medicine, 150, 841–845. Conn, V., Hafdahl, A., Brown, S., & Brown, L. (2008). Meta-analysis of patient education interventions to increase physical activity among chronically ill adults. Patient Education and Counseling, 70, 157–172. Conn, V., Isamaralai, S., Rath, S., Jantarakupt, P., Wadhawan, R., & Dash, Y. (2003). Beyond MEDLINE for literature searches. Journal of Nursing Scholarship, 35, 177–182. Conn, V., Valentine, J., Cooper, H., & Rantz, M. (2003). Grey literature in meta-analyses. Nursing Research, 52, 256–261. Devine, E., & Reifschneider, E. (1995). A meta-analysis of the effects of psychoeducational care in adults with hypertension. Nursing Research, 44, 237–245.

15

oral anticoagulants with different half-lives: A randomised trial. Thrombosis and Haemostasis, 89, 458–467. Lehane, E., & McCarthy, G. (2007). Intentional and unintentional medication non-adherence: A comprehensive framework for clinical research and practice? A discussion paper. International Journal of Nursing Studies, 44, 1468–1477. Lipsey, M., & Wilson, D. (2001). Practical meta-analysis. Thousand Oaks, CA: Sage. *Lipton, H. L., & Bird, J. A. (1994). The impact of clinical pharmacists’ consultations on geriatric patients’ compliance and medical care use: A randomized controlled trial. Gerontologist, 34, 307–315. *Lourens, H., & Woodward, M. C. (1994). Impact of a medication card on compliance in older people. Australian Journal on Ageing, 13, 72–76. Martin, K., Bowen, D., Dunbar-Jacob, J., & Perri, M. (2000). Who will adhere? Key issues in the study and prediction of adherence in randomized controlled trials. Controlled Clinical Trials, 21(5 Suppl.), 195S–199S. McGraw, K., & Wong, S. (1992). A common language effect size statistic. Psychological Bulletin, 111, 361–365. *McKenney, J. M., Munroe, W. P., & Wright, J. T., Jr. (1992). Impact of an electronic medication compliance aid on long-term blood pressure control. Journal of Clinical Pharmacology, 32, 277–283. Mills, E., Nachega, J. K., Buchan, I., Orbinski, J., Attaran, A., Singh, S., et al. (2006). Adherence to antiretroviral therapy in sub-Saharan Africa and North America: A meta-analysis. Journal of the American Medical Association, 296, 679–690. Mullen, P., Green, L. W., & Persinger, G. (1985). Clinical trials of patient education for chronic conditions: A comparative meta-analysis of intervention types. Preventive Medicine, 14, 753–781. *Nazareth, I., Burton, A., Shulman, S., Smith, P., Haines, A., & Timberal, H. (2001). A pharmacy discharge plan for hospitalized elderly patients—A randomized controlled trial. Age & Ageing, 30, 33–40. Nony, P., Cucherat, M., Haugh, M., & Boissel, J. (1995). Critical reading of the meta-analysis of clinical trials. Therapie, 50, 339–351. Peterson, A., Takiya, L., & Finley, R. (2003). Meta-analysis of trials of interventions to improve medication adherence. American Journal of Health-System Pharmacy, 60, 657–665. *Pullar, T., Birtwell, A. J., Wiles, P. G., Hay, A., & Feely, M. P. (1988). Use of a pharmacologic indicator to compare compliance with tablets prescribed to be taken once, twice, or three times daily. Clinical Pharmacology and Therapeutics, 44, 540–545. Raudenbush, S. (1994). Random effects models. In H. Cooper & L. Hedges (Eds.), The handbook of research synthesis (pp. 301–321). New York: Russell Sage. *Raynor, D. K., Booth, T. G., & Blenkinsopp, A. (1993). Effects of computer generated reminder charts on patients’ compliance with drug regimens. British Medical Journal, 306, 1158–1161. *Rich, M. W., Gray, D. B., Beckham, V., Wittenberg, C., & Luther, P. (1996). Effect of a multidisciplinary intervention on medication compliance in elderly patients with congestive heart failure. American Journal of Medicine, 101, 270–276. *Rimer, B., Levy, M. H., Keintz, M. K., Fox, L., Engstrom, P. F., & MacElwee, N. (1987). Enhancing cancer pain control regimens through patient education. Patient Education and Counseling, 10, 267–277. *Roden, S. M., Harvey, P. G., Mayer, P. P., & Spence, L. I. (1985). Evaluation of two techniques to improve drug compliance in the elderly. Journal of Clinical and Experimental Gerontology, 7, 71–82. Roter, D., Hall, J., Merisca, R., Nordstrom, B., Cretin, D., & Svarstad, B. (1998). Effectiveness of interventions to improve patient compliance: A meta-analysis. Medical Care, 36, 1138–1161.

*Rozenfeld, V., Pflomm, J. M., Singh, K. K., Brazil, M. K., & Cheng, J. W. M. (1999). Assessing the impact of medication consultations with a medication event monitoring system. Hospital Pharmacy, 34, 539–549. Rudd, P. (1979). Medication packaging: Simple solutions to nonadherence problems? Clinical Pharmacology & Therapeutics, 25, 257–265. Ryan, A. (1999). Medication compliance and older people: A review of the literature. International Journal of Nursing Studies, 36, 153–162. Sabate, E. (2003). Adherence to long-term therapies: Evidence for action. Geneva, Switzerland: World Health Organization. Sanchez-Meca, J., Marin-Martinez, F., & Chacon-Moscoso, S. (2003). Effect-size indices for dichotomized outcomes in meta-analysis. Psychological Methods, 8, 448–467. *Schectman, G., Hiatt, J., & Hartz, A. (1994). Telephone contacts do not improve adherence to niacin or bile acid sequestrant therapy. Annals of Pharmacotherapy, 28, 29–35. *Simkins, C. V., & Wenzloff, N. J. (1986). Evaluation of a computerized reminder system in the enhancement of patient medication refill compliance. Drug Intelligence & Clinical Pharmacy, 20, 799–802. Simoni, J., Pearson, C., Pantalone, D., Marks, G., & Crepaz, N. (2006). Efficacy of intervention in improving highly active antiretroviral therapy adherence and HIV-1 RNA viral load. A meta-analytic review of randomized controlled trials. Journal of Acquired Immune Deficiency Syndromes, 43(Suppl. 1), S23–S35. *Solomon, D. K., Portner, T. S., Bass, G. E., Gourley, D. R., Gourley, G. A., Holt, J. M., et al. (1998). Clinical and economic outcomes in the hypertension and COPD arms of a multicenter outcomes study. Journal of the American Pharmaceutical Association, 38, 574–585. Sullivan, S., Kreling, D., & Hazlet, T. (1990). Noncompliance with medication regimens and subsequent hospitalizations: A literature analysis and cost of hospitalization estimate. Journal of Research in Pharmaceutical Economics, 2(2), 19–33. Szeto, A., & Giles, J., Jr. (1997). Improving oral medication compliance with an electronic aid. IEEE Engineering in Medicine and Biology Magazine, 16(3), 48–54. van Eijken, M., Tsang, S., Wensing, M., de Smet, P., & Grol, R. (2003). Interventions to improve medication compliance in older patients living in the community: A systematic review of the literature. Drugs and Aging, 20, 229–240. *Varma, S., McElnay, J. C., Hughes, C. M., Passmore, A. P., & Varma, M. (1999). Pharmaceutical care of patients with congestive heart failure: Interventions and outcomes. Pharmacotherapy, 19, 860–869. Vevea, J., & Hedges, L. (1995). A general linear model for estimating effect size in the presence of publication bias. Psychometrika, 60, 419–435. Vitolins, M., Rand, C., Rapp, S., Ribisl, P., & Sevick, M. (2000). Measuring adherence to behavioral and medical interventions. Controlled Clinical Trials, 21(Suppl. 5), 188S–194S. *Vivian, E. M. (2002). Improving blood pressure control in a pharmacistmanaged hypertension clinic. Pharmacotherapy, 22, 1533–1540. *Weinberger, M., Tierney, W. M., Booher, P., & Katz, B. P. (1991). The impact of increased contact on psychosocial outcomes in patients with osteoarthritis: A randomized, controlled trial. Journal of Rheumatology, 18, 849–854. *Wood, E. R. (1989). Evaluation of a hospital-based education program for patients with diabetes. Journal of the American Dietetic Association, 89, 354–358. Received March 20, 2008 Accepted July 21, 2008 Decision Editor: William J. McAuley, PhD

16

The Gerontologist