Global Burden, Regional Differences, Trends, and Health Consequences of Medication Nonadherence for Hypertension During 2010 to 2020: A Meta‐Analysis Involving 27 Million Patients

Abstract

Background

Nonadherence to antihypertensive medications is the leading cause of poor blood pressure control and thereby cardiovascular diseases and mortality worldwide.

Methods and Results

We investigated the global epidemiology, regional differences, and trend of antihypertensive medication nonadherence via a systematic review and meta‐analyses of data from 2010 to 2020. Multiple medical databases and clinicaltrials.gov were searched for articles. Observational studies reporting the proportion of patients with anti‐hypertensive medication nonadherence were included. The proportion of nonadherence, publication year, year of first recruitment, country, and health outcomes attributable to antihypertensive medication nonadherence were extracted. Two reviewers screened abstracts and full texts, classified countries according to levels of income and locations, and extracted data. The Joanna Briggs Institute prevalence critical appraisal tool was used to rate the included studies. Prevalence meta‐analyses were conducted using a fixed‐effects model, and trends in prevalence were analyzed using meta‐regression. The certainty of evidence concerning the effect of health consequences of nonadherence was rated according to Grading of Recommendations, Assessment, Development and Evaluations. A total of 161 studies were included. Subject to different detection methods, the global prevalence of anti‐hypertensive medication nonadherence was 27% to 40%. Nonadherence was more prevalent in low‐ to middle‐income countries than in high‐income countries, and in non‐Western countries than in Western countries. No significant trend in prevalence was detected between 2010 and 2020. Patients with antihypertensive medication nonadherence had suboptimal blood pressure control, complications from hypertension, all‐cause hospitalization, and all‐cause mortality.

Conclusions

While high prevalence of anti‐hypertensive medication nonadherence was detected worldwide, higher prevalence was detected in low‐ to middle‐income and non‐Western countries. Interventions are urgently required, especially in these regions. Current evidence is limited by high heterogeneity.

Registration

URL: www.crd.york.ac.uk/prospero/; Unique identifier: CRD42021259860.

Nonstandard Abbreviations and Acronyms

MMAS

Morisky Medication Adherence Scale

MPR

medication possession ratio

WHO

World Health Organization

Clinical Perspective.

What Is New?

• Anti‐hypertensive medication nonadherence was common globally (27%–40%), was more prevalent in low‐ to middle income and non‐Western countries, and did not improve between 2010 and 2020.

What Are the Clinical Implications?

• Policymakers and clinicians should incorporate validated methods (eg, validated questionnaires, medication procession ratio, pill counting, electronic pills or pillbox, and biochemical detection by drug assays) into health care systems to routinely detect anti‐hypertensive medication nonadherence.

• Once detected, clinicians could conceptualize the reasons for nonadherence, using the World Health Organization model, and manage them accordingly.

Medication adherence is defined as the “extent to which patients take their medication as prescribed.” ^{1 , 2} Although hypertension is one of the most common chronic conditions and a leading cause of death globally, ³ medication nonadherence among patients with hypertension is highly prevalent. Up to 50% of patients stop taking their prescribed antihypertensive medications within 1 year of initiation. ⁴ The high prevalence of antihypertensive medication nonadherence has contributed to poor blood pressure (BP) control worldwide. Accordingly, optimal control of BP is attained in less than one‐third and one‐tenth of patients with hypertension in high‐income and low‐ to middle‐income countries, respectively. ^{1 , 3} This poor control has consequently led to a high global burden of cardiovascular diseases, chronic kidney disease, dementia, and mortality.

The World Health Organization (WHO) has provided a conceptual framework to explain the multifactorial reasons underlying antihypertensive medication nonadherence, including socioeconomic factors (eg, age, sex, and educational status), patient‐related factors (eg, readiness to change and self‐efficacy), therapy‐related factors (eg, complexity of treatment and out‐of‐pocket costs), comorbidities (eg, comorbid cardiovascular diseases and mental illnesses), and health care system factors (eg, doctor‐patient relationships and doctors' burnout). ⁵ Clinically, antihypertensive medication nonadherence is detected by various methods, including validated self‐reported questionnaires, pill counting (by counting the pills left over since the last prescription), prescription refills (eg, medication possession ratio [MPR] and proportion of days covered by prescriptions by reviewing medication databases), electronic pill boxes (typically detect the opening of the pill box), blood/urine biomarkers or drug assays (detect the presence of drug metabolites in biological samples), and, recently, electronic medication monitors that directly detect gastric juice. ²

Despite the importance of antihypertensive medication nonadherence, a comprehensive meta‐analysis investigating its global epidemiology is yet to be conducted. Previous meta‐analyses included only certain countries or populations, for example, low‐ to middle‐income countries and only patients with resistant hypertension. ^{6 , 7 , 8 , 9} Furthermore, previous meta‐analyses only included self‐reported questionnaires or used both validated and nonvalidated methods to define medication nonadherence. ^{6 , 7 , 8 , 9 , 10} Moreover, the high heterogeneity of results from previous meta‐analyses has not been adequately investigated using subgroup analyses or meta‐regressions, despite the presence of multiple and complex factors associated with medication adherence. ^{6 , 7 , 10} Finally, although trends and regional prevalence of uncontrolled hypertension have been well studied, there is a lack of similar research on anti‐hypertensive medication nonadherence. ³

Therefore, the primary objective of this meta‐analysis was to estimate the global prevalence of antihypertensive medication nonadherence. Additionally, the prevalence was compared among different regions and countries. We hypothesized that antihypertensive medication nonadherence would be more prevalent in low‐ to middle‐income countries, attributable to lower availability and affordability of medication, and in non‐Western countries, attributable to different beliefs/cultures. ¹¹ Trends in antihypertensive medication nonadherence from 2010 to 2020 were also examined. We hypothesized that because of the considerable research efforts and development of interventions for antihypertensive medication nonadherence over time, its prevalence would have decreased in the previous decade. ^{12 , 13} Additionally, the health consequences of antihypertensive medication nonadherence (eg, poor BP control) were investigated. The results of this study can inform patients, physicians, researchers, and policymakers regarding managing antihypertensive medication nonadherence.

METHODS

This meta‐analysis was registered in the International Prospective Register of Systematic Reviews (CRD42021259860) and reported according to the Meta‐Analyses of Observational Studies in Epidemiology standard of reporting and Preferred Reporting Items for Systematic Reviews and Meta‐Analyses guidelines. ¹⁴ Two of the authors (E.K.P.L. and P.P.) had full access to all the data and take full responsibility for its integrity and analysis. The data that support the findings of this study are available from the corresponding author upon reasonable request.

Study Eligibility

Observational studies were included if they (1) included patients with hypertension; (2) reported prevalence of antihypertensive medication nonadherence; (3) included ≥100 participants; (4) measured antihypertensive medication adherence using at least 1 of the following methods: validated questionnaire (eg, 4‐item or 8‐item Morisky Medication Adherence Scale [MMAS]), pill counting, prescription refills, electronic pill boxes, biochemical assays, or electronic medication monitoring ^{1 , 2} ; (5) used the validated or conventional cutoff of these methods (eg, scores of MMAS‐8 <6); and (6) were published in Chinese or English. The eligibility criteria were determined before the assessment of study eligibility (Table S1). ^{2 , 15} Studies were excluded if they included patients who (1) were aged <18 years, (2) had no hypertension, (3) received no antihypertensive medications, and (4) were pregnant. ⁷ Furthermore, studies that included only patients with resistant hypertension were excluded because these patients may have a higher prevalence of nonadherence and represent a different spectrum of nonadherence behaviors. Interventional trials, qualitative studies, animal studies, commentaries, and reviews were also excluded.

Information Sources

Chinese and English databases, such as the Cumulated Index to Nursing and Allied Health Literature Complete, Cochrane Library, Embase, Ovid Medline, PubMed, Scopus, Web of Science, and China Academic Journals Full‐text Database were searched for articles published up to December 2020.

Search Strategy

Keywords such as medication adherence, compliance, hypertension, antihypertensive medications, and medication adherence scale, were used as search terms (Table S2). The search was limited to studies of adults. In addition to English, studies published in Chinese were also included. Additionally, reference lists of relevant published systematic reviews were searched. ^{6 , 7 , 10} Clinicaltrials.gov was searched for unpublished trials, and the authors were contacted whenever possible.

Selection Process

All studies from the search were entered into the Covidence program (Covidence Systematic Review Software, Veritas Health Innovation, Melbourne, Australia; available at www.covidence.org). Two reviewers (from among E.K.P.L., P.P., Y.B., M.T.Z., and A.C.H.N.) independently assessed the eligibility of studies by screening the title/abstracts followed by the full texts in Covidence.

Data Collection Process

Data were dual extracted by reviewers (2 from among E.K.P.L., P.P., Y.B., M.T.Z., and A.C.H.N.) independently into Covidence. Discrepancies were compared and resolved by 2 reviewers (E.K.P.L. and P.P.).

Data Items

Extracted data included (1) details of the studies (eg, sample size, country, settings [ie, specialist center/hospital settings versus other settings], study design, inclusion/exclusion criteria). Countries were classified independently by 2 reviewers (from among E.K.P.L., P.P., M.T.Z., and A.C.H.N.) as Western or non‐Western (Western countries included Australia, New Zealand, Canada, all member countries of the European Union, the European microstates, the United Kingdom, and the United States) and high‐ or low‐ to middle‐income (as defined by the World Bank); (2) details of anti‐hypertensive medication nonadherence (methods used, cutoff, prevalence); (3) details for trend analyses (year of first recruitment and publication year); (4) socioeconomic and demographic variables of the participants that may affect adherence as defined by the WHO (age/sex, proportion with tertiary education or above, presence of cardiovascular diseases/renal diseases/diabetes/hyperlipidemia, number of years since hypertension diagnosis, the use of single‐pill combination and once‐daily medications, number of antihypertension classes, and proportion of current smokers); and (5) health consequences of nonadherence (systolic BP and diastolic BP differences between adherent and nonadherent participants and odds ratios [ORs] of suboptimal BP).

For cohort or case–control studies, health consequences, including ORs of suboptimal BP control, cardiovascular diseases, renal diseases, hospitalization, and death were also extracted. For cohort studies that reported adherence at multiple time points, the baseline value was used for analysis of comparability with cross‐sectional studies.

When only abstracts were found, the authors of the papers were contacted for published reports or articles. Abstracts were included only if they provided adequate information (ie, clear inclusion criteria, definition of anti‐hypertensive medication nonadherence, number of participants, and proportion of participants with antihypertensive medication nonadherence). For duplicated studies and cohort studies using potentially overlapping databases with overlapping dates, the latest study with the most extractable data was selected by 2 reviewers (E.K.P.L and P.P.).

Furthermore, the study by Saleem and colleagues was excluded post hoc because it reported a 100% nonadherence rate at a predetermined cutoff and could not be analyzed in Stata. ¹⁶

Study Risk‐of‐Bias Assessment

The Joanna Briggs Institute prevalence critical appraisal tool, a validated instrument, was used to rate the included studies. ¹⁷ Included studies were rated as having a low risk of bias only when no concern was raised regarding all questions in the instrument. All other included studies were rated as having unknown risk or high risk of bias. Quality assessments were conducted by 2 independent reviewers (from among E.K.P.L., P.P., Y.B., M.T.Z., and A.C.H.N.), and all discrepancies were resolved through discussion with E.K.P.L. and P.P. The certainty of evidence concerning the effect of health consequences of nonadherence was rated according to Grading of Recommendations, Assessment, Development and Evaluations.

Data Analysis

All meta‐analyses were conducted using Stata software (Stata Statistical Software: Release 15, StataCorp LLC, College Station, TX).

Global prevalence was estimated through the “metaprop” function, using a fixed‐effects model, which is the recommended and valid method to estimate prevalence from given populations. ¹⁸ Subgroup analyses were conducted on the basis of (1) the methods used to define nonadherence (eg, questionnaires, biochemical assays), (2) the countries where the studies were performed (Western versus non‐Western), and (3) the income level of these countries (high‐ versus low‐ to middle‐income). The nonadherence trend was analyzed using publication year and year of first recruitment. Heterogeneity, differences, and trends were further investigated by meta‐regression analyses using the “metareg” function. Heterogeneity across studies was assessed using I ² statistics and P values. Furthermore, the effect of nonadherence on BP level and OR was investigated by comparing between adherent and nonadherent patients using the “metan” function and a random‐effects model because of a difference in population characteristics in the included studies. P values were 2‐tailed, considering those <0.05 to be statistically significant. Examples of the Stata commands can be found in Data S1.

Sensitivity analyses were conducted to include only studies with a low risk of bias and larger studies (n>500 and n>3000 [when an adequate number of studies were available]). Within the subgroup of studies that used questionnaires, sensitivity analyses were conducted by (1) replacing studies in which the MMAS‐8 cutoff was <6 with studies that used cutoffs of ≤6; (2) including only studies that used MMAS‐4; and (3) including only studies that used MMAS‐8 because MMAS‐4 and MMAS‐8 were the most commonly used questionnaires. For cohort studies that reported adherence data after 1 year, the prevalence of nonadherence at the last follow‐up was used for the sensitivity analysis. For health consequences attributable to anti‐hypertensive medication nonadherence, sensitivity analysis was conducted using results from cohort studies only.

Publication bias was assessed by visual examination of a funnel plot, plotting the log of prevalence against the standard error of prevalence, and Egger's test.

RESULTS

Characteristics of Included Studies and Population

Of the 7004 studies identified, a total of 161 studies from 68 countries were included, with a sample size ranging from 100 to 23 833 000 (Figure 1). Over half of the included studies were conducted in low‐ to middle‐income countries (n=88). Only a few studies used biochemical assays (n=5), pill counting (n=4), and electronic pill boxes (n=3) to detect nonadherence. Therefore, meaningful corresponding subgroup and meta‐regression analyses in these subgroups was not possible. Furthermore, studies in low‐ to middle‐income and non‐Western countries predominantly used questionnaires to measure adherence during the study period, with no studies using biochemical assays or electronic pill boxes. Moreover, the sample size of studies conducted in low‐ to middle‐income countries was small, and only 1 had a sample size of >3000. Among the studies that used questionnaires, the MMAS‐8 (n=73) and MMAS‐4 (n=45) questionnaires were most commonly used (Table S5). Only 23 studies were rated as having a low risk of bias (Tables S6 through S8). Our study population consisted of 27 785 595 patients with hypertension, with a mean age of 57 (42.9% men). Other demographic data and the list of included studies are presented in Tables S3 and S4.

Figure 1. Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) flowchart.

HT indicates hypertension.

Global Prevalence, Regional Differences, and Trends in Antihypertensive Medication Nonadherence

The prevalence varied with methods used to define nonadherence: 40% by questionnaires (95% CI, 40%–40%), 28% by prescription refill (95% CI, 28%–28%), 28% by pill counting (95% CI, 26%–29%), 28% by electronic pill boxes (95% CI, 25%–31%), and 27% by biochemical assays (95% CI, 26%–29%) (Figure 2, Table S9).

Figure 2. Prevalence of nonadherence presented with 95% CIs (subgroup: nonadherence definitions, West vs non‐West, income levels).

Nonadherence was more prevalent in low‐ to middle‐income countries than in high‐income countries, when defined by questionnaires (43% versus 38%; P=0.145), prescription refill (50% versus 28%; P=0.37), and pill counting (66% versus 25%; P=0.382). Similarly, nonadherence was more prevalent in non‐Western countries than in Western countries, when defined by questionnaires (43% versus 38%; P=0.108), and prescription refill (49% versus 26%; P=0.086; Figure 2, Table S9). Although nonadherence was less prevalent in non‐Western countries than in Western countries when pill counting was used, this included only 4 unclear to high risk‐of‐bias studies (22% versus 49%; P=0.974; Figure S1). Depending on the method used to define nonadherence, the prevalence of nonadherence ranged from 20% to 49% among continents (Tables S9 through S13, Figure S1).

No significant trend in antihypertensive medication nonadherence was detected over the past decade in all meta‐regression analyses, including subgroup analyses, using publication year or year of first recruitment (Figure 3, Tables S14 and S15).

Figure 3. Trend of medication nonadherence according to nonadherence prevalence and included studies' publication year.

A, By any definition: regression coefficient: 0.004, P=0.434; (B) by questionnaires: regression coefficient: −0.0002, P=0.977; (C) by prescription refills: meta‐regression coefficient: 0.010, P=0.416. DBP indicates diastolic blood pressure; HT, hypertension; and SBP, systolic blood pressure.

When using meta‐regression to explore heterogeneity, in the subgroup analysis of studies using the prescription refill method of adherence, nonadherence was less common in older patients (P=0.001), patients receiving free medical service or insurance (P=0.044), and patients receiving more classes of antihypertensive medications (P=0.014; Table S16). Other factors, such as the presence of cardiovascular diseases and medication frequency, were not significantly associated with the prevalence of nonadherence (Table S16). These meta‐regression analyses did not explain the heterogeneity, and all residual I ² remained >95%.

Consequences of Antihypertensive Medication Nonadherence

Compared with adherent patients, patients with antihypertensive medication nonadherence had higher systolic BP (mean difference, 3.76 mm Hg [95% CI, 2.23–5.28 mm Hg]; I ², 87.1%; P<0.001), and diastolic BP (mean difference, 3.11 mm Hg [95% CI, 2.24–3.99 mm Hg]; I ², 76%; P<0.001; Figure 4). ^{19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 , 38 , 39} Furthermore, patients with antihypertensive medication nonadherence had increased odds of having suboptimal BP control (OR, 2.15 [95% CI, 1.84–2.5]; I ², 97.4%; P<0.001), complications from hypertension (OR, 2.08 [95% CI, 0.99–4.35]; I ², 94.2%; P<0.001), all‐cause hospitalization (OR, 1.38 [95% CI, 1.35–1.41]; I ², 0; P=0.64), and all‐cause mortality (OR, 1.38 [95% CI, 1.35–1.41]; I ², 0; P=0.509; Figure 5). ^{19 , 21 , 33 , 34 , 35 , 40 , 41 , 42 , 43 , 44 , 45 , 46 , 47 , 48 , 49 , 50 , 51 , 52 , 53 , 54 , 55 , 56 , 57 , 58 , 59 , 60 , 61 , 62 , 63 , 64 , 65 , 66 , 67 , 68 , 69 , 70} Sensitivity and subgroup analyses revealed similar results but did not resolve high heterogeneity (Figures S2 and S3, Tables S17 and S18). According to Grading of Recommendations, Assessment, Development and Evaluations, the certainty of evidence was low for all health outcomes, owing to inclusion of observational studies only.

Figure 4. Blood pressure difference attributable to medication nonadherence.

A, Systolic blood pressure difference attributable to medication nonadherence; (B) diastolic blood pressure difference attributable to medication nonadherence.

Figure 5. Health consequence attributable to medication nonadherence.

DBP indicates diastolic blood pressure; DL, xxx; and SBP, systolic blood pressure.

Sensitivity Analyses

Sensitivity analyses generally showed a decrease in nonadherence prevalence when only larger studies were included. This result is congruent with our findings on regional differences because larger studies were predominantly from high‐income countries. Moreover, almost all sensitivity analyses consistently found lower nonadherence prevalence in Western and high‐income countries. For instance, this was observed when only low‐risk‐of‐bias and questionnaire studies (prevalence, 38% [95% CI, 37%–39%]; Figure S3), and only studies using MMAS‐4 (prevalence, 41% [95% CI, 41%–42%]; Figure S3) were included. The differences in systolic BP/diastolic BP and health outcomes between adherent and nonadherent participants remained similar in the sensitivity analyses. Moreover, no significant trend in prevalence of nonadherence was detected in various sensitivity analyses (Tables S17 and S18, Figure S3). However, no sensitivity analysis adequately explained the results' high heterogeneity (Tables S17 and S18, Figure S3).

Publication Bias

The funnel plots and Egger's test did not show a significant small study bias (Egger's test, P=0.332; Figure S4).

DISCUSSION

Main Findings and Comparison With Previous Literature

Subject to different detection methods, the global prevalence of antihypertensive medication nonadherence ranged from 27% to 40%. Furthermore, antihypertensive medication nonadherence was more prevalent in low‐ to middle‐income countries and non‐Western countries. For instance, using our results from prescription refill and the latest WHO data, this translates to ≈426 million people from low‐ to middle‐income countries, and 119 million people from high‐income countries. ⁷¹ Our results are similar to those of another meta‐analysis that reported a global prevalence of 45%, but that meta‐analysis included only studies that used MMAS. ¹⁰ Our results are also similar to those of previous large observational studies revealing that antihypertensive medication nonadherence led to poor BP control, higher health care resource use, cardiovascular complications, and death. ^{72 , 73} However, this is the first study to suggest that, in addition to the known factors of underdiagnosis and undertreatment, nonadherence plays an important role in the differential poor hypertension control in low‐ to middle‐income countries. ³ The exact reasons underlying these regional differences cannot be determined from our data, but they could be attributed to differences in cultures, beliefs, the use of alternative medicine, health care systems, and drug affordability and availability. ³² To date, there has been a lack of primary studies that directly investigate regional differences (eg, Western versus non‐Western or high‐income versus low‐ to middle‐income countries) in antihypertensive medication nonadherence.

Although a decreasing trend in nonadherence has been described in a few US studies, this trend has not been observed globally. ^{74 , 75} This suggests that, although evidence‐based interventions, such as reduction in daily number of pills and single‐pill combinations, can reduce medication nonadherence, they were not adequately implemented in clinical practice. ^{13 , 76 , 77} A Cochrane review also suggested that significant improvements in adherence and clinical outcomes were uncommon in well‐conducted randomized controlled trials, and these called for advances and more interventional studies in the field. ⁷⁸

Our results also suggested that the prevalence of nonadherence was generally lower when more objective detection methods were used (ie, electronic pill boxes and biochemical assays). However, these studies were conducted only in Western and high‐income countries. This difference could, therefore, be attributable to the regional differences described. These differences could also result from the Hawthorne effect, that is, an improved nonadherence rate when patients know that they are being monitored. ⁷⁹ In the current study, nonadherence was detected by these objective methods in only 8 studies, and no study used electronic medications.

Clinical and Research Implications

Our results are consistent with international guidelines that state that antihypertensive medication nonadherence is highly prevalent and clinicians treating hypertension should screen for nonadherence during every clinician visit. ⁸⁰ However, clinicians' predictions of drug nonadherence are known to be no better than “a coin toss.” Therefore, policymakers and clinicians should incorporate validated methods into health care systems to routinely detect anti‐hypertensive medication nonadherence. ⁸¹ However, all existing methods, including the use of questionnaires, calculation of MPR, or telemonitoring by electronic pill boxes, would require extra time and resources, which could be difficult to implement. Newer methods, including the use of dried blood samples and oral fluid assays, are being developed and investigated to provide reliable and quick methods for clinicians to routinely detect nonadherence. ^{82 , 83} Once detected, clinicians could conceptualize the reasons for nonadherence, using the WHO model, and manage them accordingly. ⁵

Our results also call for implementation research to examine how the latest evidence can be translated into clinical practice and trials to investigate interventions that can effectively improve medication adherence and clinical outcomes. ⁷⁸ While most existing research investigated single interventions, clinical practice guidelines suggest that complex interventions combining several interventions to target the factors listed by the WHO are most likely needed. ⁸⁰ However, real‐life data concerning such complex interventions are scarce. Furthermore, there is still no reference standard for the detection of medication nonadherence. Even biochemical assays, which are one of the most objective measures, suffer from the white‐coat adherence effect, in which patients have improved adherence only before doctor visits. ⁸⁴ A feasible, affordable, and reliable reference standard to define nonadherence would facilitate research and its clinical detection. Additionally, existing validated antihypertensive medication nonadherence detection methods could not provide a comprehensive assessment of patients' adherence behaviors, which include the processes of “initiation,” “implementation,” and “discontinuation.” ⁸⁵ For instance, although 90% MPR signified good medication adherence using a conventional cutoff, the missing 10% can represent both occasional drug holidays or complete discontinuation. Moreover, the reasons for the higher nonadherence prevalence in low‐ to middle‐income and non‐Western countries could be explored and examined further. Finally, large population‐based studies on antihypertensive medication nonadherence from low‐ to middle‐income countries are lacking.

Strengths and Limitations

This study has many strengths. This is the first meta‐analysis that describes not only global prevalence but also regional differences and trends in antihypertensive medication nonadherence in the previous decade. This study represented the best available evidence in view of the lack of similar primary research across continents with different income levels. Our meta‐analysis also involved a comprehensive search, including Chinese databases, with the largest number of studies among similar meta‐analyses. Meta‐regressions were conducted to investigate the relationship between prevalence of nonadherence and patients' determining factors (eg, presence of cardiovascular diseases), and treatment factors (eg, once‐daily or combined‐pill treatments, number of medications; Table S16). There was no significant publication bias, and the sensitivity analyses showed congruent results.

However, all results were highly heterogeneous because studies included different populations, used different definitions of nonadherence, and included diverse factors that this study could not encompass (eg, characteristics of health care and insurance systems, quality of doctor‐patient relationships, and level of doctors' burnout). Furthermore, questionnaires had different sensitivities and specificities to detect medication nonadherence and measured different aspects of nonadherence (beliefs, barriers, and actual use of medications), which could partially explain the statistical heterogeneity. ⁸⁶ To minimize heterogeneity, we included only studies that used validated or conventional definitions and cutoffs for antihypertensive medication nonadherence. Relevant subgroups, meta‐regression, and sensitivity analyses were also used to investigate heterogeneity; however, these did not adequately explain the heterogeneity. Although the use of only population‐based samples may further reduce heterogeneity (a methodology commonly used in other meta‐analyses that investigated hypertension epidemiology), this was not possible because large studies from low‐ to middle‐income countries were not available. For instance, only 1 study from a low‐ to middle‐income country had a sample size >3000. ³ Moreover, our sensitivity analyses, which included only large studies, did not resolve heterogeneity (Tables S17 and S18, Figure S3).

Second, methods including prescription refills, pill counting, electronic pill boxes, and biochemical assays were rarely used in studies from non‐Western or low‐ to middle‐income countries. These precluded comparative analyses or statistical significance in several subgroups. Therefore, prevalence estimates from these countries were derived primarily using questionnaire methods, which are prone to self‐reporting bias and have poor agreement with objective methods. ⁸⁷ Furthermore, many questionnaires, such as MMAS‐8, cannot provide the exact timing and number of doses missed. However, since questionnaires tended to underestimate nonadherence as compared with objective methods (eg, biochemical assays), this strengthens our conclusion that nonadherence was more prevalent in non‐Western or low‐ to middle‐income countries. ² Third, we included only studies published in English or Chinese. Nevertheless, of the 677 full‐text studies screened, only 34 were excluded because of language issues. Fourth, interventional trials were excluded because patients who volunteered and consented to these trials (especially trials to improve drug adherence) could be systematically different from other patients with hypertension. Strict inclusion and exclusion criteria of randomized controlled trials often results in the selection of patients with similar characteristics, which may bias our results. Nevertheless, including baseline data from these intervention trials could further enhance our comprehensiveness and sample size. Fifth, high heterogeneity of the results could hinder the detection of trends of antihypertensive medication nonadherence in the meta‐regression analysis.

Sixth, although we used the most validated and conventional cutoffs for questionnaires and MPR, these cutoffs can still be questioned. For example, at a cutoff of 6, MMAS‐8 has only a sensitivity and specificity to detect nonadherence of ≈0.43 and 0.74, respectively. ¹⁵ Similarly, the MPR cutoff of 0.82, instead of 0.80, may be more appropriate to detect antihypertensive medication nonadherence. ⁸⁸ However, alternate cutoffs (eg, MPR <0.82), were not used by the current studies and therefore could not be used in the current meta‐analyses. We have presented questionnaire data using MMAS‐8 <6 and ≤6 (sensitivity analysis in Tables S17 and S18 and Figure S3). Finally, although the results of the health consequences of antihypertensive medication nonadherence were rated low according to Grading of Recommendations, Assessment, Development and Evaluations because of the inclusion of only observational studies, this matter is difficult and unethical to investigate using clinical trials.

CONCLUSIONS

Globally, ≈27% to 40% of patients with hypertension are nonadherent to their medications. A higher prevalence of antihypertensive medication nonadherence was detected in low‐ to middle‐income and non‐Western countries. Interventions are urgently required to detect antihypertensive medication nonadherence and improve medication adherence, especially in countries where antihypertensive medication adherence is suboptimal.

Sources of Funding

None.

Disclosures

None.

Supporting information

Data S1

Tables S1–S18

Figures S1–S4

References 16, 65, 75, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 125, 126, 127, 128, 129, 130, 131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 145, 146, 147, 148, 149, 150, 151, 152, 153, 154, 155, 156, 157, 158, 159, 160, 161, 162, 163, 164, 165, 166, 167, 168, 169, 170, 171, 172, 173, 174, 175, 176, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 191, 192, 193, 194, 195, 196, 197, 198, 199, 200, 201, 202, 203