526562
research-article2014
AOPXXX10.1177/1060028014526562Annals of PharmacotherapyZhang et al.
Review Article
Impact of Health Literacy on Medication Adherence: A Systematic Review and Meta-analysis
Annals of Pharmacotherapy 2014, Vol. 48(6) 741–751 © The Author(s) 2014 Reprints and permissions: sagepub.com/journalsPermissions.nav DOI: 10.1177/1060028014526562 aop.sagepub.com
Ning Jackie Zhang, MD, PhD1, Amanda Terry, MA1, and Colleen A. McHorney, PhD2
Abstract Objective: To systematically review the literature and estimate the effect size of the relationship between health literacy and medication adherence through meta-analysis. Data Sources: Databases searched included Cumulative Index to Nursing and Allied Health Literature (CINAHL; 1982-2013), International Pharmaceutical Abstracts (IPA; 1970-2013), MEDLINE OVID (1966-2013), PubMed (1966-2013), PsycInfo (1966-2013), and Web of Science (1966-2013). Study Selection and Data Extraction: Inclusion criteria were as follows: English language; published through May 1, 2013; medication adherence as the outcome variable; use of validated measures of health literacy and medication adherence; availability of a direct (not mediating) relationship between health literacy and medication adherence; and identifiable effect size and statistical significance of the relationship. Exclusion criteria were as follows: duplicated results, irrelevant results, conference abstracts, proceeding papers, books, dissertations, reviews, editorial letters, continuing education units, or book reviews. Data included author, publication year, disease area, sample size, sampling method, location, study design, effect size of the relationship between health literacy and medication adherence, and measures of health literacy and medication adherence. Data Synthesis: There is a small statistically significant and positive association between health literacy and medication adherence. In the conservative results, the unweighted and weighted correlation coefficients were 0.081 and 0.056 with P values 3 months; not significant MEMS β = 0.62; P = 0.01
Refill records Self-report
NPMG
MMAS-4
Medication Adherence Measures
745
Downloaded from aop.sagepub.com at TEXAS SOUTHERN UNIVERSITY on November 20, 2014
HIV/AIDS
HIV/AIDS
HIV/AIDS
Glaucoma
Disease Area
n = 204 Individuals from specialty center/clinic n = 398 Patients from primary care clinic and specialty center/clinic
Cardiovascular
HIV/AIDS
Noureldin et al (2012)26
Osborn et al (2007)53 Osborn et al (2011)48
Diabetes
n = 314 Patients from primary care clinic
HIV/AIDS
Murphy et al (2010)41
General
Lindquist et al (2012)24 (b) n = 897 Individuals from community n = 310 Patients from primary care clinic n = 197 Individuals from specialty center/clinic n = 186 Individuals from community
n = 254 Patients from hospital discharge
General
Lindquist et al (2012)24 (a)
Cardiovascular, Marcum et al diabetes (2013)55 Mosher et al General (2012)58 Muir et al (2006)25 Glaucoma
n = 254 Patients from hospital discharge
Cardiovascular
n = 204 Individuals from specialty center/clinic n = 182 Individuals from community n = 138 Individuals from community n = 145 Patients from primary care clinic, individuals from specialty center/clinic, and from community n = 92 Individuals from community n = 434 Patients from primary care clinic
Sample Size and Sample Location
Kripalani et al (2010)52
Kim et al (2004)40 Diabetes
Juzych et al (2008)22 Kalichman et al (1999)49 Kalichman et al (1999)51 Kalichman et al (2008)23
Author (Year)
Table 1. (continued)
Cross-sectional study
REALM
REALM
S-TOFHLA
Adapted TOFHLA
REALM
REALM
Observational study Cross-sectional study Questionnaire
S-TOFHLA
S-TOFHLA
REALM
TOFHLA-S
TOFHLA
SDSCA
PMAQ
Module 1 of Pediatric Adherence Questionnaire MEMS
Self-report, refill records Refill records
MMAS-4
Self-report
Self-report
CMG, MMAS-4
SDSCA
Pill counting
Adapted TOFHLA Self-report
REALM-SF
Secondary data analysis after RCT Interview
Medication Adherence Measures Relationship Between Literacy and Adherence
AOR (95% CI) = 2.12 (1.93, 2.32); significant P < 0.02
P = 0.001
OR (95% CI) for 0% adherence = 1.00; P = 0.98
P = 0.003
χ2 = 0.91
(continued)
CMG OR (95% CI) = 1.7 (1.0, 3.1), P < 0.001 for inadequate HL; CMG OR (95% CI) = 1.8 (1.0, 3.2) for marginal HL; significant Unintentional nonadherence, adequate HL as reference: marginal HL AOR (95% CI) = 4.22 (1.63, 10.88); inadequate HL AOR (95% CI) = 5.42 (2.05, 14.37); P = 0.003 Intentional nonadherence, adequate HL as reference: marginal HL AOR (95% CI) = 0.170 (0.057, 0.509); inadequate HL AOR (95% CI) = 0.203 (0.070, 0.590); P = 0.003 AOR(95% CI) = 1.12 (0.59, 2.10); P = 0.74
P = 0.751
AOR (95% CI) = 3.77 (1.46, 9.93); P < 0.01
χ2 = 4.99; P < 0.05
Self-report, P < 0.001 medical records Adapted TOFHLA Self-report OR (95% CI) = 3.9 (1.1, 13.4); P < 0.05
TOFHLA
Questionnaire
Interview
Interview
Observational study Cross-sectional study
Interview
Interview
Observational study Interview
Design
Health Literacy Measures
746
HIV/AIDS
General
Abbreviations: REALM, Rapid Estimate of Adult Literacy in Medicine; TOFHLA, Test of Functional Health Literacy in Adults; AOR, adjusted odds ratio; MEMS, Medication Event Monitoring System; MARS, Medication Adherence Report Scale; RCT, randomized controlled trial; MMAS, Morisky Medication Adherence Scale; NPMG, New Prescription Medication Gap; S-TOFHLA, short form of Test of Functional Health Literacy in Adults; CMG, Cumulative Medication Gap; HBCS, Hill-Bone Compliance Scale; SDSCA, Summary of Diabetes Self-Care Activities Measure; HL, health literacy; REALM-SF, short form of Rapid Estimate of Adult Literacy in Medicine; PMAQ, Patient Medication Adherence Questionnaire; and CT, controlled trials.
PMAQ
REALM; S-TOFHLA Prospective cohort REALM-SF study Interview REALM Questionnaire
n = 57 Individuals from community n = 111 Caregivers of children and hospital discharges n = 204 Individuals from specialty center/clinic Cardiovascular
Raehl et al (2006)27 Rosman et al (2012)59 Wolf et al (2007)28
HIV/AIDS
Self-report
Adjusted β for REALM = 0.666; R2 = 0.271; P = 0.001 AOR (95% CI) = 0.78 (0.14, 4.92); not significant AOR (95% CI) = 2.1 (0.8, 5.5) for marginal HL; AOR (95% CI) = 3.3 (1.3, 8.7) for low HL; P < 0.001
AOR (95% CI) = 1.93 (0.86, 4.31); not significant
AIDS Clinical Trials Group Adherence Instrument MedTake Test
Doser CT, MEMS P = 0.45
n = 46 Patients from hospital Prospective cohort TOFHLA-S discharge study n = 235 Individuals from primary Longitudinal study REALM care clinic, specialty center/ clinic, and community Asthma
Paasche-Orlow et al (2005)56 Paasche-Orlow et al (2006)42
Disease Area Author (Year)
Table 1. (continued)
Sample Size and Sample Location
Design
Health Literacy Measures
Medication Adherence Measures
Relationship Between Literacy and Adherence
Annals of Pharmacotherapy 48(6) coefficient (r)—using the Meta-Analysis program44 and Comprehensive Meta-Analysis45 for cross-validation. The correlation coefficient is one of the most commonly used effect size measures in meta-analysis because it represents both magnitude and direction of a statistical association.46 A positive r suggests that a higher level of health literacy is associated with better medication adherence. Although multiple methods to synthesize effect sizes from independent studies exist, random-effect models were used for the meta-analyses in this study because a random-effect model allows generalization of results from included studies to more studies by assuming that the observed variability in sample estimates of effect sizes come from both the variations in the population parameters and sampling effects of the estimator.43,47 Whereas the fixed-effect model underscores weighted mean analyses and the homogeneity of the effect size parameters, the random-effect model assumes unweighted mean analyses and heterogeneity of the parameters. The random-effect model uses a variance component on betweenstudy variance to calculate the weights, whereas the fixedeffect model uses the within-study variance.43 To compare the results of both approaches, both weighted and unweighted means of effect sizes were presented in results. In addition, the 95% confidence interval of the means of effect sizes, their P values, and fail-safe n for r of 0.05 were presented. Fail-safe n is a measure of the filedrawer problem of sampling bias in meta-analysis. It represents the number of file-drawer studies required to bring the mean effect size down to a statistically insignificant level as an unknown number of studies with zero effect sizes that have remained in file-drawers, which means that they have not been submitted or accepted for publication.44,45 One potential problem with the above analysis is that the distributions of the effect size parameters are largely unknown and may not be normal. Monte-Carlo simulation can construct bootstrap sampling–based confidence intervals for the effect size comparison. The Monte Carlo simulation was implemented through bootstrap resampling, with replacement to the sample of effect sizes drawn from included studies.43 Bootstrapping was undertaken using 1000 iterations from which the distribution of correlation coefficients was generated. Sample sizes of individual studies were used to assign weights to effect sizes because larger samples should carry more weight than smaller samples, the latter of which have larger variances and might be less precise. Sample-size-weighted averages and corresponding confidence intervals are expected to be more accurate than unadjusted statistics. The probability density function of the relationship between health literacy and adherence was plotted to demonstrate the distribution of the estimated effect size. If the confidence interval of the effect size included zero, the effect size estimation was considered to be statistically significant. R software and self-developed coding were used to conduct the analysis.43
Downloaded from aop.sagepub.com at TEXAS SOUTHERN UNIVERSITY on November 20, 2014
747
Zhang et al. Table 2. Meta-analysis Results.a
Approach 1 Unweighted analysis Weighted analysis Approach 2 Unweighted analysis Weighted analysis
Population Effect Size (Mean r)
Significance (P Value)
95% CI of Population Effect Size
0.08127 0.05558
research-article2014
AOPXXX10.1177/1060028014526562Annals of PharmacotherapyZhang et al.
Review Article
Impact of Health Literacy on Medication Adherence: A Systematic Review and Meta-analysis
Annals of Pharmacotherapy 2014, Vol. 48(6) 741–751 © The Author(s) 2014 Reprints and permissions: sagepub.com/journalsPermissions.nav DOI: 10.1177/1060028014526562 aop.sagepub.com
Ning Jackie Zhang, MD, PhD1, Amanda Terry, MA1, and Colleen A. McHorney, PhD2
Abstract Objective: To systematically review the literature and estimate the effect size of the relationship between health literacy and medication adherence through meta-analysis. Data Sources: Databases searched included Cumulative Index to Nursing and Allied Health Literature (CINAHL; 1982-2013), International Pharmaceutical Abstracts (IPA; 1970-2013), MEDLINE OVID (1966-2013), PubMed (1966-2013), PsycInfo (1966-2013), and Web of Science (1966-2013). Study Selection and Data Extraction: Inclusion criteria were as follows: English language; published through May 1, 2013; medication adherence as the outcome variable; use of validated measures of health literacy and medication adherence; availability of a direct (not mediating) relationship between health literacy and medication adherence; and identifiable effect size and statistical significance of the relationship. Exclusion criteria were as follows: duplicated results, irrelevant results, conference abstracts, proceeding papers, books, dissertations, reviews, editorial letters, continuing education units, or book reviews. Data included author, publication year, disease area, sample size, sampling method, location, study design, effect size of the relationship between health literacy and medication adherence, and measures of health literacy and medication adherence. Data Synthesis: There is a small statistically significant and positive association between health literacy and medication adherence. In the conservative results, the unweighted and weighted correlation coefficients were 0.081 and 0.056 with P values 3 months; not significant MEMS β = 0.62; P = 0.01
Refill records Self-report
NPMG
MMAS-4
Medication Adherence Measures
745
Downloaded from aop.sagepub.com at TEXAS SOUTHERN UNIVERSITY on November 20, 2014
HIV/AIDS
HIV/AIDS
HIV/AIDS
Glaucoma
Disease Area
n = 204 Individuals from specialty center/clinic n = 398 Patients from primary care clinic and specialty center/clinic
Cardiovascular
HIV/AIDS
Noureldin et al (2012)26
Osborn et al (2007)53 Osborn et al (2011)48
Diabetes
n = 314 Patients from primary care clinic
HIV/AIDS
Murphy et al (2010)41
General
Lindquist et al (2012)24 (b) n = 897 Individuals from community n = 310 Patients from primary care clinic n = 197 Individuals from specialty center/clinic n = 186 Individuals from community
n = 254 Patients from hospital discharge
General
Lindquist et al (2012)24 (a)
Cardiovascular, Marcum et al diabetes (2013)55 Mosher et al General (2012)58 Muir et al (2006)25 Glaucoma
n = 254 Patients from hospital discharge
Cardiovascular
n = 204 Individuals from specialty center/clinic n = 182 Individuals from community n = 138 Individuals from community n = 145 Patients from primary care clinic, individuals from specialty center/clinic, and from community n = 92 Individuals from community n = 434 Patients from primary care clinic
Sample Size and Sample Location
Kripalani et al (2010)52
Kim et al (2004)40 Diabetes
Juzych et al (2008)22 Kalichman et al (1999)49 Kalichman et al (1999)51 Kalichman et al (2008)23
Author (Year)
Table 1. (continued)
Cross-sectional study
REALM
REALM
S-TOFHLA
Adapted TOFHLA
REALM
REALM
Observational study Cross-sectional study Questionnaire
S-TOFHLA
S-TOFHLA
REALM
TOFHLA-S
TOFHLA
SDSCA
PMAQ
Module 1 of Pediatric Adherence Questionnaire MEMS
Self-report, refill records Refill records
MMAS-4
Self-report
Self-report
CMG, MMAS-4
SDSCA
Pill counting
Adapted TOFHLA Self-report
REALM-SF
Secondary data analysis after RCT Interview
Medication Adherence Measures Relationship Between Literacy and Adherence
AOR (95% CI) = 2.12 (1.93, 2.32); significant P < 0.02
P = 0.001
OR (95% CI) for 0% adherence = 1.00; P = 0.98
P = 0.003
χ2 = 0.91
(continued)
CMG OR (95% CI) = 1.7 (1.0, 3.1), P < 0.001 for inadequate HL; CMG OR (95% CI) = 1.8 (1.0, 3.2) for marginal HL; significant Unintentional nonadherence, adequate HL as reference: marginal HL AOR (95% CI) = 4.22 (1.63, 10.88); inadequate HL AOR (95% CI) = 5.42 (2.05, 14.37); P = 0.003 Intentional nonadherence, adequate HL as reference: marginal HL AOR (95% CI) = 0.170 (0.057, 0.509); inadequate HL AOR (95% CI) = 0.203 (0.070, 0.590); P = 0.003 AOR(95% CI) = 1.12 (0.59, 2.10); P = 0.74
P = 0.751
AOR (95% CI) = 3.77 (1.46, 9.93); P < 0.01
χ2 = 4.99; P < 0.05
Self-report, P < 0.001 medical records Adapted TOFHLA Self-report OR (95% CI) = 3.9 (1.1, 13.4); P < 0.05
TOFHLA
Questionnaire
Interview
Interview
Observational study Cross-sectional study
Interview
Interview
Observational study Interview
Design
Health Literacy Measures
746
HIV/AIDS
General
Abbreviations: REALM, Rapid Estimate of Adult Literacy in Medicine; TOFHLA, Test of Functional Health Literacy in Adults; AOR, adjusted odds ratio; MEMS, Medication Event Monitoring System; MARS, Medication Adherence Report Scale; RCT, randomized controlled trial; MMAS, Morisky Medication Adherence Scale; NPMG, New Prescription Medication Gap; S-TOFHLA, short form of Test of Functional Health Literacy in Adults; CMG, Cumulative Medication Gap; HBCS, Hill-Bone Compliance Scale; SDSCA, Summary of Diabetes Self-Care Activities Measure; HL, health literacy; REALM-SF, short form of Rapid Estimate of Adult Literacy in Medicine; PMAQ, Patient Medication Adherence Questionnaire; and CT, controlled trials.
PMAQ
REALM; S-TOFHLA Prospective cohort REALM-SF study Interview REALM Questionnaire
n = 57 Individuals from community n = 111 Caregivers of children and hospital discharges n = 204 Individuals from specialty center/clinic Cardiovascular
Raehl et al (2006)27 Rosman et al (2012)59 Wolf et al (2007)28
HIV/AIDS
Self-report
Adjusted β for REALM = 0.666; R2 = 0.271; P = 0.001 AOR (95% CI) = 0.78 (0.14, 4.92); not significant AOR (95% CI) = 2.1 (0.8, 5.5) for marginal HL; AOR (95% CI) = 3.3 (1.3, 8.7) for low HL; P < 0.001
AOR (95% CI) = 1.93 (0.86, 4.31); not significant
AIDS Clinical Trials Group Adherence Instrument MedTake Test
Doser CT, MEMS P = 0.45
n = 46 Patients from hospital Prospective cohort TOFHLA-S discharge study n = 235 Individuals from primary Longitudinal study REALM care clinic, specialty center/ clinic, and community Asthma
Paasche-Orlow et al (2005)56 Paasche-Orlow et al (2006)42
Disease Area Author (Year)
Table 1. (continued)
Sample Size and Sample Location
Design
Health Literacy Measures
Medication Adherence Measures
Relationship Between Literacy and Adherence
Annals of Pharmacotherapy 48(6) coefficient (r)—using the Meta-Analysis program44 and Comprehensive Meta-Analysis45 for cross-validation. The correlation coefficient is one of the most commonly used effect size measures in meta-analysis because it represents both magnitude and direction of a statistical association.46 A positive r suggests that a higher level of health literacy is associated with better medication adherence. Although multiple methods to synthesize effect sizes from independent studies exist, random-effect models were used for the meta-analyses in this study because a random-effect model allows generalization of results from included studies to more studies by assuming that the observed variability in sample estimates of effect sizes come from both the variations in the population parameters and sampling effects of the estimator.43,47 Whereas the fixed-effect model underscores weighted mean analyses and the homogeneity of the effect size parameters, the random-effect model assumes unweighted mean analyses and heterogeneity of the parameters. The random-effect model uses a variance component on betweenstudy variance to calculate the weights, whereas the fixedeffect model uses the within-study variance.43 To compare the results of both approaches, both weighted and unweighted means of effect sizes were presented in results. In addition, the 95% confidence interval of the means of effect sizes, their P values, and fail-safe n for r of 0.05 were presented. Fail-safe n is a measure of the filedrawer problem of sampling bias in meta-analysis. It represents the number of file-drawer studies required to bring the mean effect size down to a statistically insignificant level as an unknown number of studies with zero effect sizes that have remained in file-drawers, which means that they have not been submitted or accepted for publication.44,45 One potential problem with the above analysis is that the distributions of the effect size parameters are largely unknown and may not be normal. Monte-Carlo simulation can construct bootstrap sampling–based confidence intervals for the effect size comparison. The Monte Carlo simulation was implemented through bootstrap resampling, with replacement to the sample of effect sizes drawn from included studies.43 Bootstrapping was undertaken using 1000 iterations from which the distribution of correlation coefficients was generated. Sample sizes of individual studies were used to assign weights to effect sizes because larger samples should carry more weight than smaller samples, the latter of which have larger variances and might be less precise. Sample-size-weighted averages and corresponding confidence intervals are expected to be more accurate than unadjusted statistics. The probability density function of the relationship between health literacy and adherence was plotted to demonstrate the distribution of the estimated effect size. If the confidence interval of the effect size included zero, the effect size estimation was considered to be statistically significant. R software and self-developed coding were used to conduct the analysis.43
Downloaded from aop.sagepub.com at TEXAS SOUTHERN UNIVERSITY on November 20, 2014
747
Zhang et al. Table 2. Meta-analysis Results.a
Approach 1 Unweighted analysis Weighted analysis Approach 2 Unweighted analysis Weighted analysis
Population Effect Size (Mean r)
Significance (P Value)
95% CI of Population Effect Size
0.08127 0.05558
