Psychosocial predictors of non-adherence to chronic medication: systematic review of longitudinal studies

Abstract

Objectives

Several cross-sectional studies suggest that psychosocial factors are associated with non-adherence to chronic preventive maintenance medication (CPMM); however, results from longitudinal associations have not yet been systematically summarized. Therefore, the objective of this study was to systematically synthesize evidence of longitudinal associations between psychosocial predictors and CPMM non-adherence.

Materials and methods

PUBMED, EMBASE, CINAHL, and PsychINFO databases were searched for studies meeting our inclusion criteria. The reference lists and the ISI Web of Knowledge of the included studies were checked. Studies were included if they had an English abstract, involved adult populations using CPMM living in Western countries, and if they investigated associations between psychosocial predictors and medication non-adherence using longitudinal designs. Data were extracted according to a literature-based extraction form. Study quality was independently judged by two researchers using a framework comprising six bias domains. Studies were considered to be of high quality if ≥four domains were free of bias. Psychosocial predictors for non-adherence were categorized into five pre-defined categories: beliefs/cognitions; coping styles; social influences and social support; personality traits; and psychosocial well-being. A qualitative best evidence synthesis was performed to synthesize evidence of longitudinal associations between psychosocial predictors and CPMM non-adherence.

Results

Of 4,732 initially-identified studies, 30 (low-quality) studies were included in the systematic review. The qualitative best evidence synthesis demonstrated limited evidence for absence of a longitudinal association between CPMM non-adherence and the psychosocial categories. The strength of evidence for the review’s findings is limited by the low quality of included studies.

Conclusion

The results do not provide psychosocial targets for the development of new interventions in clinical practice. This review clearly demonstrates the need for high-quality, longitudinal research to identify psychosocial predictors of medication non-adherence.

Introduction

In conditions such as rheumatoid arthritis, diabetes, and hypertension, long-term therapy with chronic preventive maintenance medication (CPMM) is essential for reducing risks of disease progression, comorbidity, and mortality. However, sufficient medication adherence to CPMM is a prerequisite for reducing these risks.¹

Medication non-adherence, or the extent to which patients do not take their medications as agreed with their health care provider, averages 50% among patients suffering from chronic diseases in developed countries.² Non-adherence can result in poorer health outcomes and a lower quality of life in patients.³ For example, patients who did not adhere to beta-blocker therapy were four and a half times more likely to have complications from coronary heart disease than those who adhered to therapy.⁴ Non-adherence also affects health care utilization. For instance, poorer adherence among elderly patients with moderate-to-severe asthma was associated with a 5% increase in annual physician visits, whereas better adherence was associated with a 20% decrease in annual hospitalization.⁵

Considering the undesired consequences of non-adherence to CPMM, interventions are needed to improve medication non-adherence. According to the World Health Organization (WHO), possible targets for these interventions can be divided into five domains:² socio-economic factors, health care system factors, condition-related factors, therapy-related factors, and patient-related factors. Although none of the factors within these domains are consistently associated with non-adherence across conditions, some tend to be better predictors of non-adherence than others (like poverty, the nature of the disease, and side-effects).^1,2 Also, psychosocial factors like beliefs about medication, self-efficacy, and social support can be promising intervention targets. These are mostly modifiable (in contrast to factors like poverty or side-effects), and according to reviews of cross-sectional studies, they appear to be associated with non-adherence in various somatic, chronic conditions.^6–13 Beliefs about medication were the most powerful predictors of adherence (among demographic and medical factors) in one cross-sectional study,⁹ while another cross-sectional study identified low-self-efficacy as a significant predictor of non-adherence across different countries, for example.¹¹ However, there is no insight into psychosocial factors predicting non-adherence in longitudinal studies with a longer follow-up period (≥3 months). Such knowledge would be helpful in designing effective adherence interventions in clinical practice.

This is the first review which aims to systematically synthesize evidence of longitudinal associations between psychosocial predictors and CPMM non-adherence across adult patients living in Western countries. Since non- adherence literature is scattered across diseases,¹⁴ we combined studies from various somatic, chronic conditions to increase the robustness of our findings.

Methods

PRISMA-guidelines were followed in performing this systematic review.¹⁵ The steps taken regarding data searches, study selection, data extraction, study quality assessment, data synthesis, and data analyses are elaborated below.

Data sources and searches

In March 2011, according to a pre-defined search strategy, four electronic databases (PUBMED, EMBASE, CINAHL, and PsychINFO) were searched for studies up to February 2011. With this search, a first set of studies was included, the reference lists of these studies were hand searched to find additional studies. The studies were also entered into the ISI Web of Knowledge citation index (August 2011). The resulting list of studies, citing one of the initial included studies in our review, was also searched.

The search strategy (see Supplementary Materials) contains key words on medication adherence, chronic, somatic diseases, adults, longitudinal designs, and Western countries. Countries in Africa, Latin-America, South-America, Asia (excluding Indonesia and Japan), and Turkey were considered as non-Western according to Statistics Netherlands.¹⁶ Non-Western countries were excluded because underlying mechanisms of medication non-adherence could differ from those in Western countries due to socio-economic and cultural differences.¹⁷

In this review, we focused on two of the three components of adherence (ie, on initiation and implementation adherence, thus the extent to which a patient’s actual medication dosing regimen corresponds with the prescribed dosing regimen from initiation to last dose). We did not include discontinuation of medication.¹

As using CPMM terms in the search strategy was unfeasible, we used the corresponding diseases for which the CPMMs were prescribed as search terms instead. The disease terms were selected as follows:

1. Chronic preventive maintenance medications were defined. CPMMs were regarded as drugs that 1) are intended to be used chronically to prevent the occurrence or worsening of a disease or its complications; and 2) may have an immediate effect, but must also have a long-term effect (>3 months).

2. From the full November 2010 Anatomical Therapeutic Chemical Classification System (ATC)-7 medication list of drugs available in the Netherlands, 246 CPMMs (Supplementary Materials) were independently selected by two pharmacists (BvdB and VH). There was an initial agreement of 96% on medications being CPMM. Disagreements were resolved by discussion between the pharmacists.

3. Disease indications for the 246 CPMMs were subsequently clustered by BvdB according to the International Classification of Diseases (WHO). Finally, 20 disease terms were used in the search strategy.

Study selection

Studies were selected based on the criteria in Table 1.

Table 1.

Inclusion and exclusion criteria

Domain	Inclusion criteria	Exclusion criteria
Study population	Study population ≥18 years, living in a Western country and using chronic preventive maintenance medication for one or more somatic, chronic condition as specified in the search strategy	Studies exclusively recruiting subpopulations in special conditions, like alcohol addicts, prisoners, pregnant women*
Study types	Longitudinal retrospective or prospective study design, at least examining associations between predictor ‘X’ measured at baseline and outcome ‘Y’ measured ≥3 months after baseline†	Study is cross-sectional, controlled trial, case report, (systematic) review, meta-analysis, editorial, letter, comment, interview, newspaper article, case-control study, intervention study, thesis* or validation study*
Outcome measure	Medication non-adherence is (one of) the primary outcomes of the study. All adherence instruments (eg, different questionnaires, refill data, MEMS) are eligible for inclusion	Outcome is discontinuation of medication
Psychosocial predictors	Psychosocial predictors are defined as predictors, pertaining to the influence of social factors on an individual’s mind or behavior, and to the interrelation of behavioral and social factors18 The term psychosocial also covers internal, psychological predictors (eg, anxiety) in this systematic review. All predictor instruments (eg, different questionnaires/scales) are eligible for inclusion	Predictors measuring addiction to stimulating agents, psychosocial co-morbidity (eg, diagnosed depression according to DSM-IV criteria, and cognitive impairment. Illness symptoms, however, like depressive mood states and anxiety, are included in the review), socio-demographics, knowledge, cognitive status, behavior, satisfaction about treatment and health care, overall outcome measures (eg, social functioning of patient, general health status, perceived quality of life, behavioral intentions), predictors outside perception of individual patient (eg, beliefs of physicians)* In addition, predictors for which it was unclear what they measured (eg, ‘HIV-mastery’19 or ‘coping’ without specifying the type of coping), predictors for which results had not been reported in studies*
Other	English abstract available‡	No English abstract available, unpublished studies which could not be retrieved after substantial efforts

Notes:

^*These criteria were formulated during the selection process. We did not exclude subpopulations based on socio-demographic features. Veterans or government employees, for example, are not in a special condition per se; ^†when the outcome is measured multiple times after baseline, and one summary measure over the total, observational time after baseline is calculated, than the observational time should be at least 6 months. For example, studies measuring daily adherence for 3 months after baseline and calculating one summary adherence measure for a patient over these 3 months are excluded, because the mean time point of the summary adherence measure is 1.5 months after baseline; ^‡please note that studies of all languages are eligible, but at least an English abstract should be available.

Abbreviations: DSM-IV, Diagnostic and Statistical Manual of Mental Disorders, 4th Edition, Text Revision; HIV, human immunodeficiency virus; MEMS, medication event monitoring System.

Studies exclusively recruiting subpopulations in special conditions (like prisoners, pregnant women) were excluded. Their results only pertain to a specific group of patients, therefore, including them might have introduced bias into this systematic review.

Two reviewers (BvdB and HZ) independently assessed studies for eligibility in two phases: 1) screening based on title and abstract; and 2) screening based on full text. Disagreements between BvdB and HZ were resolved by discussion; a third reviewer (CvdE) made decisions in case disagreements could not be resolved. Studies in Spanish or Portuguese were judged by LvdA. During the study selection process, three authors were contacted about statistics, outcome measure, or study design to determine eligibility for this review.^20–22

Data extraction and quality assessment

For data extraction, a literature-based, standard form was developed.^23,24 Information regarding study setting, design, descriptive statistics, measures, and analysis were extracted by HZ; BvdB arbitrarily selected 15% of the included studies to check appropriateness of all extracted data of these studies, and also checked all doubts indicated on the form by HZ.

If multiple adherence measures were presented in one study (eg, about dosing, timing, or taking medication)²⁵, we only extracted data about taking medication. Two authors were contacted during the extraction process to check the duration of a follow-up period of ≥3 months²⁶ or to explain ambiguities.¹⁹

We adapted the framework developed by Hayden et al²⁷ to judge methodologic study quality. Our framework contained 23 items divided into six bias domains: study participation, study attrition, prognostic, outcome and confounding measurement, and analyses. Each item was scored as ‘yes’ (no unacceptable amount of bias introduced), ‘partly’ (/unsure), and ‘no’ (unacceptable amount of bias introduced). For every bias domain, a transparent method was used to reach overall judgment about the presence or absence of bias (see Table S1). Studies with ≥four domains judged as ‘yes’ were considered high-quality studies; studies with <four domains were considered low-quality studies.

Using three randomly selected studies not included in the review, the framework was piloted by BvdB and HZ, who also performed the actual quality assessment. Disagreements were resolved by discussion and, when necessary, a third reviewer (CvdE) made final decisions. On the domain level, a weighted extent of agreement between BvdB and HZ (quadratic weighting scheme) was calculated due to the ordinal nature of the scores.^28,29

Data synthesis and analysis

Because over 70 non-identical psychosocial predictors (non-identical by name and/or measurement instrument) were studied in this review, and because of the variety of instruments used to measure non-adherence, a qualitative instead of a quantitative analysis was considered to be appropriate.³⁰ Therefore, the results regarding associations between psychosocial predictors and medication non-adherence were qualitatively synthesized in four steps.

In step 1, psychosocial categories were formulated. Initially, all psychosocial elements as mentioned in general health behavior models and theories^31,32 were listed (HZ). Subsequently, based on consensus, the elements were clustered by HZ and three psychologists (SvD, JV, and LK) resulting in the categories of Figure 1.

Figure 1.

Psychosocial categories.

Next, the psychosocial predictors within the studies of the review were assigned to one of the categories in Figure 1 (HZ and the psychologists). In this way, the considerable number of single, non-identical predictors was dealt with.

In step 2, for each psychosocial predictor within a category and within a study, the presence of a significant univariate and multivariate association with medication non-adherence was determined (see Table S2). Statistical significance was set at P<0.05.

In step 3, results within studies were synthesized per psychosocial category. When ≥75% of variables within a single psychosocial category were significantly and consistently (ie, same predictors in same direction) associated with non-adherence, a ‘yes’ was assigned (ie, association present). When ≥75% of variables were significantly, but inconsistently, associated (eg, four of five predictors in category about depressive symptoms, of which two are positively related to non-adherence and two are negatively related), the term ‘conflicting’ was assigned. When <75% of variables were significantly and consistently associated, a ‘no’ was assigned. Multivariate results were preferably used to synthesize results in this step. When multivariate results were not reported, univariate results were used.

In the fourth and final step, a best evidence synthesis (BES) per psychosocial category between studies was performed to summarize evidence of longitudinal associations between the predictors in the psychosocial categories and medication non-adherence. We defined four levels of evidence as used in previous reviews of longitudinal studies:^{60 –62}

1. Strong evidence: consistent findings (≥75% of studies within psychosocial category report same conclusion about association; ie, ‘yes, present’ or ‘no, not present’) in at least two high-quality studies.

2. Moderate evidence: consistent findings in one high-quality study AND at least two low-quality studies.

3. Limited evidence: findings in one high-quality study OR consistent findings in at least two low-quality studies.

4. Conflicting evidence: inconsistent findings in at least two studies irrespective of study quality (ie, <75% of studies report same conclusion about association). Note that this level of evidence was checked first before assigning strong, moderate or limited evidence level to a category.

The level of evidence was undeterminable when ≤one study of low quality was available for a psychosocial category.

Sensitivity analyses were performed to examine the robustness of findings, regarding the cut-off point for methodological quality, diseases, adherence measurement, and statistical analyses (ie, focusing on univariate analyses only). Also, an additional analysis on single predictors was carried out, since associations between single predictors like ‘avoidance coping’ and non-adherence could be overshadowed by combining them into a single category with generally non-significant psychosocial predictors, such as hopelessness and confusion. Three steps were taken: 1) all significant predictors (P≤0.05) were listed; 2) each of these predictors was grouped with identically named, significant and non-significant predictors; and 3) when at least two studies were available for those predictors, the BES rules were applied.

Results

Study inclusion

Of 4,732 non-duplicate references, 30 met our inclusion criteria (Figure 2).^{19,25,26,33–59} In all, 1,255 records were identified by screening the reference lists and the ISI Web of Knowledge citation index of the initial included studies.

Figure 2.

Flowchart of study inclusion process.

Abbreviation: CPMM, chronic preventive maintenance medication.

Initially, the percentage of agreement regarding the eligibility of studies was 86% (of the 272 studies selected on title and abstract, agreement was obtained in about 235 studies after reading the full-text). Disagreements were mainly due to misconceptions about psychosocial predictors (eg, clinically diagnosed depression versus symptoms of depression), study design, and adherence measure (ie, discontinuation or execution adherence). For one study,⁵² disagreement could not be resolved by discussion and thus a final decision was made by CvdE.

Study characteristics and quality assessment

Table 2 displays study characteristics, measures, and results. A comprehensive table of measures and results is presented in Table S2.

Table 2.

Study characteristics and results^*

First author	Setting	Sample characteristics			Measures and results
		Sample size, % loss to follow-up	Age†, % female	Disease duration†	Adherence‡, follow-up period§	Psychosocial category, number of predictors||	Association present between category and adherence/non-adherence?¶	Number domains bias free**
Asthma (inhaled corticosteroids)
Ponieman33	USA; patients from general internal medicine clinic	261, 23%	48 (13), 82%	Age of onset ≤20 years: 50% of sample	Self-report (MARS), 3 months	AI, n=5 AIII, n=3	No (U: yes, M: no) No (U: no, M: no)	0 of 6
Diabetes (oral and/or parenteral antidiabetics)
Venturini34,††	USA; patients from HMO-providing health services	786, 0%	59 (mean), 24–92 (range), 49%	NR	Record review, last time point flexible, but within 24 months	EI, n=1	No (U: NR, M: no)	2 of 6
Heart disease and hypertension (cardiovascular medication)
Gazmararian35	USA; community-dwelling patients‡‡	1,549, UD	Age: 65–69: 35%, 70–74: 28%, 75–79: 20%, 80–84: 12%, >84: 6%, (female): 58%	UD	Record review, 12 months	CIII, n=1	No (U: no, M: NT)	3 of 6
Nabi26	Finland; local government employees	1,021, UD	26–63 (range), 32%	0–2 years: n=311 2–5 years: n=222 5–10 years: n=292 >10 years: n=196	Record review, 12 months	D, n=4 EI, n=2	Yes (U: no, M: yes) No (U: no, M: NT)	1 of 6
Grégoire36	Canada; hypertensive adults with prescriptions from network of pharmacies	692, 26%	59 (13), 56%	47 months (adherent group), 44 months (non-adherent group)	Self-report (Morisky scale), 3 months	AI, n=1 AII, n=5 CIII, n=1	No (U: no, M: no) No (U: no, M: no) No (U: no, M: no)	0 of 6
Miller37	Site not reported: patients from institutions providing cardiac rehabilitation programs¶¶	141, 21%	56 (mean), 32–70 (range), 22%	NR	Self-report (HBS), 6–9 months	AI, n=1 CII, n=1	No (U: NR, M: no) No (U: NR, M: no)	0 of 6
Molloy38	UK; patients admitted to hospitals with acute coronary syndrome¶¶	295, 11%	61 (mean), 32–87 (range), 23%	0 years (acute)	Self-report, 12 months	CIII, n=2	No (U: no, M: no)	1 of 6
HIV (antiretroviral medication)
Deschamps25	Belgium; outpatients at university hospital	60, 28%	43 (9) adherent group, 41 (8) non-adherent group, 16%	NR	MEMS, 5–6 months after measuring psychosocial constructs	AI, n=3 AIII, n=1 BI, n=3 BII, n=4 CIII, n=2 D, n=1 EI, n=2	No (U: no, M: NR) No (U: no, M: NR) No (U: no, M: NR) No (U: no, M: NR) No (U: no, M: NR) No (U: no, M: NR) No (U: no, M: NR)	1 of 6
Holmes19	USA; HIV-clinic patients	116, 0%§§	44 (median), 25–69 (range), 19%	5 years (median)	MEMS, 12 months (or when viral load of ≥1,000 copies/mL was reached)	AI, n=1 AII, n=2 CI, n=1 CIII, n=1 EI, n=1 EII, n=1	No (U: no, M: no) No (U: no, M: no) No (U: no, M: NT) No (U: no, M: NT) No (U: no, M: no) No (U: no, M: no)	2 of 6
Delgado39	Canada; patients enrolled in community drug treatment program	316, 0%	NR, NR	NR	Record review, 12 months	EI, n=1	No (U: yes, M: no)	1 of 6
Singh40	USA; new, veteran patients seen at medical center	52, 12%	40 (median), 23–68 (range), 0%	NR	Record review, 6 months	BII, n=1 CIII, n=2 EI, n=4	No (U: no, M: no) No (U: no, M: no) No (U: no, M: no)	1 of 6
Singh41	Site not reported: patients in HIV-medical centers	138, 11%	41 (median), 24–71 (range), 7%	NR (but 7% therapy-naive)	Record review, 6 months	BI, n=3 BII, n=6 CIII, n=4 EI, n=1	No (U: no, M: NR) No (U: no, M: NR) No (U: no, M: NR) No (U: no, M: NR)	1 of 6
Bottonari42	USA; patients treated in immunodeficiency clinic	78, 69%	36 (7), 4%	NR	Self-report (straightforward), 6–9 months	D, n=2 EI, n=1 EII, n=3	No (U: no, M: NR) No: (U: no, M: NR) No: (U: no, M: NR)	0 of 6
Godin43	Canada; patients from medicalHIV-clinics	400, 6%	43 (8), 4%	>5 years HIV- infected: 73%	Self-report (straightforward), 12 months	AI, n=1 AIII, n=2 CI, n=1 CIII, n=1 D, n=1	Yes (U: NR, M: yes) No (U: NR, M: no) No (U: NR, M: no) No (U: NR, M: no) No (U: NR, M: no)	1 of 6
Kacanek44	USA; patients recruited by media and physician networks	225, 0%	45 (7), 23%	NR	Self-report (straightforward); maximum 30 months	EI, n=1	Yes (U: yes, M: NT)	2 of 6
Martini45	Italy; outpatients using combination therapy¶¶	214, 71%	<30: 13%, 30–39: 56%, >39: 31%, (female): 36%	NR	Self-report (straightforward); 12 months	AI, n=2 CI, n=1	No (U: no, M: NR) Yes (U: yes, M: NR)	0 of 6
Mellins46	USA; HIV-infected mothers recruited in waiting room of adult clinic	128, 25%	38 (mean), 22–66 (range), 100%	5 years	Self-report (AACTG, straightforward), T1 after 4–5 months, T2 8–18 months after T1	AIII, n=1 EI, n=1 EII, n=2	No (U: no, M: NR) No (U: no, M: NR) Yes (U: yes, M: NR)	0 of 6
Nilsson Schönnesson47	Sweden; patients recruited by clinic nurses	203, 29%	45 (9), 22%	Mean year of diagnosis =1990	Self-report (straightforward), 24 months	AI, n=3 AII, n=1 AIII, n=2 BII, n=2 CI, n=1 CIII, n=2 D, n=1 EI, n=3 EII, n=1	No (U: NR, M: no) No (U: NR, M: no) No (U: NR, M: no) No (U: NR, M: no) No (U: NR, M: no) No (U: NR, M: no) No (U: NR, M: no) No (U: NR, M: no) No (U: NR, M: no)	1 of 6
Thrasher48	USA; patients in public use of HCSUS data set	1,911, 33%§§	Minority versus non-minority: <35: 35% minority group, 30% non-minority group. % female: 33% versus 12%, respectively	Mean year first diagnosed with HIV: 1992, minority group; 1990, non-minority group	Self-report (straightforward), 12 months	CIII, n=1 EI, n=2	No (U: no, M: NR) Yes (U: yes, M: NR)	1 of 6
Horne49	UK; outpatients, eligible to receive HAART	136, 14%	38 (9), NR	5 years	Self-report (straightforward), 12 months	AI, n=2 EI, n=1	Yes (U: yes, M: yes) No (U: no, M: NT)	3 of 6
Mugavero50	USA; patients receiving care at infectious disease clinics	474, 39%	40 (median), 35–46 (IQR), 29%	NR	Self-report (AACTG, straightforward), 27 months	EII, n=4	No (U: yes, M: no)	3 of 6
Carrieri51	France; patients starting HAART-regimen	1,110, 13%	37 (median), 22%	First time since first positive HIV-test in years: 3.8 (median), 0.5–8.2 (IQR)	Self-report (AACTG, straightforward), 60 months	CII, n=1 EI, n=1	Yes (U: yes, M: yes) Yes (U: yes, M: yes)	2 of 6
Transplant-related (immunosuppressant medication)
Stilley52	USA; transplant patients, recruited before hospital discharge or at early clinic visit	152, 29%	55 (10), 33%	NR	MEMS, 6 months	BI, n=1 CII, n=1 D, n=2 EI, n=1	No (U: no, M: NR) No (U: no, M: NR) No (U: no, M: NR) No (U: no, M: NR)	1 of 6
De Geest53	Belgium; convenience sample of outpatients	101, 0%	56 (median), 20–69 (range), 13%	3 (median), 1–6 (range) years since transplantation	MEMS, 6 months	AIII, n=1 CIII, n=1 EI, n=1 EII, n=1	Yes (U: NR, M: yes) No (U: NR, M: no) No (U: NR, M: no) No (U: NR, M: no)	2 of 6
Russell54	USA; convenience sample of renal transplant patients	50, 26%	60 (5), 38%	NR	MEMS, 12 months	AIII, n=1 CIII, n=1 EI, n=2	No (U: no, M: NR) No (U: no, M: NR) No (U: no, M: NR)	0 of 6
Weng55	USA; patients recruited at time of renal transplantation	829, 66%	48 (median), 39–57 (IQR), 39%	NR	MEMS, 12 months post-transplantation	AIII, n=1 CII, n=1 EI, n=1 EII, n=1	No (U: yes, M: no) No (U: no, M: NT) No (U: no, M: NT) No (U: no, M: NT)	2 of 6
Dew56	USA; heart transplant patients from academic hospital|| ||	108, 22%	<50 years: 49%, (female): 16%	NR	Self-report (straightforward), 12 months post-transplantation	AIII, n=1 BI, n=2 BII, n=1 CII, n=2 EI, n=3	No (U: no, M: NT) No (U: no, M: NT) Yes (U: yes, M: yes) No (U: no, M: no) No (U: no, M: no)	2 of 6
Dew57	USA; patients receiving first lung transplantation in academic hospital	178, 29%	37% <50 years, (female): 48%	NR	Self-report (straightforward), 24 months	AIII, n=3 CII, n=2 D, n=1 EI, n=3	No (U: yes***, M: no) No (U: yes, M: no) No (U: yes, M: no) No (U: yes, M: no)	1 of 6
Dobbels58	Belgium; heart, liver and lung transplant patients listed at university hospitals	186, 24%	52 (12), 33%	NR	Self-report (straightforward), 12 months post-transplantation	CII, n=2 D, n=5 EI, n=2	No (U: NR, M: no) No (U: NR, M: no) No (U: NR, M: no)	1 of 6
Other (diabetes and/or hypertension and/or heart disease)
DiMatteo59	USA; patients from five medical specialties in HMOs, large multispecialty groups or solo practices‡‡,¶¶	Max 1,828, UD	60 (8), 54%	NR	Self-report (straightforward), 24 months	BII, n=1 CI, n=2 CII, n=1 EI, n=1	No (U: NR, M: no) No (U: NR, M: no) No (U: NR, M: no) Yes (U: NR, M: yes)	0 of 6

^*NS (non significant): as reported in the concerning study. UD (undetermined): because of inadequate description in the concerning study;

^†mean (and for age: standard deviation) in years reported unless indicated otherwise;

^‡with straightforward, we mean that participants were directly asked to indicate how many medication doses they missed. For example: “How many pills did you take this week?”;

^§follow-up period = number of months between baseline (unless indicated otherwise) and last adherence measurement;

^||this column shows the number of psychosocial predictors measured in the concerning study, and the assigned psychosocial category. Details about the single predictors are presented in Table S2. A = Beliefs and cognitions about I) medication and treatment; II) illness; III) self-efficacy and locus of control. B = coping styles I) task oriented, II) emotion oriented. C = Social influences and social support I) regarding medical caregiver; II) regarding friends and family; III) in general. D = personality traits. E = psychological well-being: I) mood state; II) perceived stress/stressors;

^¶no = no significant association between psychosocial category and medication adherence/non-adherence within study when P≤0.05; Yes = significant association when P≤0.05; U: univariate; M: multivariate;

^**to determine methodological quality, six bias domains per study were judged. Here, the total amount of bias free domains is reported (for further details, see Table S3);

^††retrospective design;

^‡‡diagnosis for coronary heart disease, hypertension, diabetes mellitus, and/or hyperlipidemia;

^§§% loss to follow-up assumed by HZ/BvdB;

^{|| ||}type of medication is immunosuppressants, antihypertensives, and/or antivirals;

^¶¶use of chronic preventive medication assumed;

^***significance of P≤0.05 assumed by HZ/BvdB.

Abbreviations: AACTG, adult AIDS clinical trials group; HAART, highly active antiretroviral therapy; HBS, health behavior scale; HCSUS, HIV cost and services utilization study; HIV, human immunodeficiency virus; HMO, health maintenance organization; IQR, interquartile range; MARS, medication adherence report scale; MEMS, medication event monitoring system; NR, not reported; NS, non significant; NT, not tested; UD, undetermined.

The included studies (all based on different data sets) covered CPMMs for asthma, diabetes, heart diseases/hypertension, human immunodeficiency virus (HIV), and organ transplants. Medication type was not explicitly mentioned in four studies,^37,38,45,59 but we assumed CPMM was used since CPMM is the standard medical treatment for the 20 selected diseases in this review. In most studies, patients were recruited from medical clinics or hospitals and the sample size ranged from 50–1,911. Attrition rates varied from 0%–71%. Participants were predominantly men and often ≥37 years of age and a disease duration of >2 years. The observation period between baseline and last adherence measurement was ≥3 and <12 months in ten studies and ≥12 months in 20 studies, with a maximum of 60 months. Medication adherence was mostly measured by self-report (18 studies, predominantly questionnaires); seven studies used a validated adherence questionnaire.^{33,36,43,46,49–51} Other adherence measurements were carried out by reviewing medical records or the medication event monitoring system (MEMS). In 15 studies, both univariate and multivariate analyses were reported.

All 30 included studies were judged to be ‘low-quality’ (Table S3). This was mainly due to poor descriptions and/or bias regarding the study sample, the use of non-validated questionnaires, the lack of accounting for confounding variables, and a poor description of the data analyses. Most studies, moreover, did not appropriately describe actions taken in case of missing data.

A total of 180 bias domains were judged (30 studies by six domains). Initially, BvdB and HZ fully agreed on 78 domains, partially agreed (ie, ‘partly’ versus ‘no’ or ‘partly’ versus ‘yes’) on 79 domains and fully disagreed (eg, ‘yes’ versus ‘no’) on 23 domains, resulting in a weighted agreement of 76%. Disagreements were caused by poor description of methods, different interpretations of missing data, differences in calculating study attrition rates, and different interpretations regarding the appropriateness of study sample descriptions. On this latter point, disagreements about three studies^35,48,52 could not be resolved by discussion between BvdB and HZ and, thus, CvdE made the final decision.

Best evidence synthesis

Table 3 shows there is limited evidence for the absence of a longitudinal association with medication non-adherence in all of the eleven psychosocial subcategories.

Table 3.

Level of evidence for longitudinal associations between psychosocial categories and medication non-adherence

Psychosocial category	N of studies	Quality	Longitudinal association	Level of evidence
A. Beliefs and cognitions
I. About medication and treatment	9	All low	2 × yes43,49 7 × no19,25,33,36,37,45,47	No association (limited evidence)
II. About illness	3	All low	3 × no19,36,47	No association (limited evidence)
III. Self-efficacy and locus of control	10	All low	1 × yes53 9 × no25,33,43,46,47,54–56,57	No association (limited evidence)
B. Coping styles
I. Task-oriented	4	All low	4 × no25,41,52,56	No association (limited evidence)
II. Emotion-oriented	6	All low	1 × yes56 5 × no25,40,41,47,59	No association (limited evidence)
C. Social influences and social support
I. Regarding medical caregiver	5	All low	1 × yes45 4 × no19,43,47,59	No association (limited evidence)
II. Regarding friends and family	6	All low	1 × yes51 5 × no37,52,55–57	No association (limited evidence)
III. In general	14	All low	14 × no19,25,35,36,38,40,41,43,47,48,53,54,58,59	No association (limited evidence)
D. Personality traits	8	All low	1 × yes26 7 × no25,42,43,47,52,57,58	No association (limited evidence)
E. Psychosocial well-being
I. Mood state	21	All low	3 × yes44,48,51 18 × no19,25,26,34,39–42,46,47,49,52–58	No association (limited evidence)
II. Perceived stress/stressors	8	All low	2 × yes46,59 6 × no19,42,47,50,53,55	No association (limited evidence)

Beliefs and cognitions

Regarding category AI (beliefs and cognitions about medication and treatment), two of nine studies found a longitudinal, multivariate association between having a positive attitude towards taking medication and adherence (odds ratio [OR] =1.56, 95% confidence interval [CI] 1.18, 2.06),⁴³ and between necessity beliefs and concern beliefs about medication and adherence (OR =2.19, 95% CI 1.02, 4.71 and OR =0.45, 95% CI 0.22, 0.96, respectively).⁴⁹ One other study³³ found univariate associations between necessity and concern beliefs about medication and adherence, but these associations did not hold in the multivariate analysis.

One study demonstrated a longitudinal, multivariate association between low self-efficacy and medication non-adherence;⁵³ however, the effect size was small. Univariate, but not multivariate associations between self-efficacy and adherence were demonstrated in two studies.^55,57

Coping styles

No univariate and multivariate associations were found between the task-oriented coping style category and medication adherence.

Regarding emotion-oriented coping styles, one of six studies revealed a multivariate association with non-adherence (eg, OR of 9.71 for avoidance coping).⁵⁶ Furthermore, avoidance coping as a single predictor was associated with non-adherence in three of four studies measuring this construct.^25,40,56

Social influences and social support

Two of the 25 studies demonstrated significant associations between predictors within the category social influences and social support and (non-)adherence, but only one of these studies reported on a multivariate association between having support from a partner and non-adherence (regression coefficient =−0.15, 95% CI −0.25, −0.05).⁵¹ Receiving practical social support was associated with better adherence as a single predictor.^38,41

Personality traits

One of eight studies showed a multivariate, longitudinal association between the category of personality traits and medication non-adherence:²⁶ a lower sense of coherence (a global life orientation in which life is perceived as comprehensible, manageable and meaningful)⁶³ was associated with greater non-adherence (OR=0.55, CI 0.31–0.96). Associations between other predictors within the personality traits category and non-adherence were lacking.

Psychological well-being

Regarding categories EI (mood state) and EII (perceived stress/stressors), no associations between predictors in those categories and medication non-adherence could be established for the vast majority of studies (24 out of 29). Two of the five studies which did show significant associations reported on multivariate analyses: the regression coefficient for depressive symptoms was 0.18 (95% CI 0.07, 0.29) in predicting non-adherence;⁵¹ the standardized beta for health distress was −0.22 (CI not reported) for predicting adherence.⁵⁹

Table S2 can be consulted for detailed information about associations between single psychosocial predictors and medication adherence/non-adherence.

Sensitivity analyses

The sensitivity analyses confirmed that, generally, no association was found between the psychosocial categories and medication non-adherence (Table S4).

The additional analysis on single predictors showed no association between most single, psychosocial predictors and medication non-adherence. However, conflicting evidence was found for having a positive attitude towards taking medication,^37,43 necessity beliefs and concern beliefs about medication,^33,49 self-efficacy in medication-taking,^{25,33,43,47,53,54} the coping style “planful problem solving”,^25,41 and (the number of) stressful (life) events.^38,42,46 Limited evidence was found for an association between escape-avoidance coping and medication non-adherence,^25,41,56,59 and for an association between receiving practical, social support and medication adherence.^38,41

Discussion

To the best of our knowledge, this is the first systematic review summarizing evidence of longitudinal associations between psychosocial factors and non-adherence to CPMM, irrespective of somatic disease. Due to the low quality of the included studies, limited evidence was found for absence of longitudinal associations between categories of psychosocial predictors and medication non-adherence. In general, findings were robust according to sensitivity analyses.

Our findings of longitudinal associations between psychosocial factors and medication non-adherence are in line with the few conducted cross-sectional studies about associations between medication adherence, coping styles, personality traits, and psychosocial well-being (except depressive symptoms) in somatic conditions. The findings in these cross-sectional studies are ambiguous at best.^8,64–68 For example, an active coping style was associated with medication adherence in some studies^8,68 but not in others,^64,66 and stress was associated with lesser adherence in a study of Holt et al,⁶⁷ but was unrelated to non-adherence in a study of Ediger et al.⁶⁵

In contrast to coping styles and personality traits, depression is often studied as possible predictor of medication non-adherence. Here, our results are not in line with results from other reviews, reporting depression to be a predictor of medication non-adherence.^6,69–74 Initially, this discrepancy might be explained by the fact that clinical depression is within the scope of most other studies, but beyond the scope of our systematic review since we did not study morbidity as a predictor of non-adherence; instead, we studied depressive symptoms. Second, an explanation might be that those other reviews included studies with mainly cross-sectional designs. Feelings of depression might increase and decrease over the course of a disease. A high degree of depressive feelings might correlate well with non-adherent behavior at that same time, but just might not be predictive of non-adherent behavior in the future due to this changeability. Thus, longitudinal associations between depressive feelings and non-adherence might not be applicable.

This thought might also apply to discrepancies in findings between our review and other reviews on associations between beliefs about medication/treatment, poor social support, and non-adherence. These other reviews underline the importance of beliefs about medication/treatment and poor social support in predicting medication non-adherence^6,10,69–76 in contrast to our review findings, but again, those other reviews are mainly based on studies with cross-sectional designs.

In terms of internal validity, a strength of this review is that we, in contrast to others, systematically defined and categorized psychosocial factors. By doing so, we were able to 1) draw a concise number of conclusions about associations between psychosocial predictors and medication non-adherence in a reproducible manner; 2) address the heterogeneity between single, psychosocial predictors; and 3) address an important goal of a systematic review: converging information. The pitfall of categorization (eg, the possibility of overlooking significant associations between certain, single predictors and non-adherence, by pooling them with other types of [non-significant] predictors), was avoided by performing an extended sensitivity analysis on single predictors. This analysis revealed our conclusions to be robust for almost all single, psychosocial predictors included in this review.

Another strength of this review is that we systematically synthesized results using a best evidence synthesis in contrast to most other reviews, which tend to be characterized by narrative designs.^{6,10,69,70,73,74,76} Narrative designs often do not rely on systematic methods to assign weight of evidence; eg, by incorporating methodological quality of included studies.⁷⁷ Although no review procedure eliminates the chance that reviewers’ biases will affect the conclusions drawn,⁷⁷ the application of a best evidence synthesis makes a review procedure transparent and reproducible.

A limitation of this systematic review is that we used chronic disease terms instead of medication terms in the search strategy and, consequently, we may have missed relevant studies about chronic preventive maintenance medication. However, we assume that the number of missed studies is minimal, since diseases are usually mentioned in medication adherence studies.

Another limitation could be the use of results of univariate analyses to draw conclusions about associations in the absence of multivariate analysis data, as univariate analyses could lead to an overestimating of the strength of associations. However, our sensitivity analyses on data from univariate analyses confirmed the robustness of our findings.

Concerning external validity, a strong feature of this review is that it focused exclusively on longitudinal associations between psychosocial predictors and medication non-adherence, thereby providing insight into the temporality and robustness of associations. However, only 5 of the 30 studies included in our review corrected for baseline non-adherence.^{34,50,53,58,59} Failure to account for baseline non-adherence when suggesting predictive longitudinal associations is considered a liberal approach,⁷⁸ since baseline non-adherence is likely to explain a substantial part of the variance in non-adherence over time. Because we did not find any associations using a liberal approach, however, we believe it is unlikely that handling a strict longitudinal approach in this review would have altered our findings.

Another limitation concerning external validity is that the poor quality of the included studies prevented us from drawing firm conclusions about the lack of associations between psychosocial predictors and medication adherence The lack of a gold standard for adherence measurement⁷³ also restricts the validity of our findings. The adherence measures used in the included studies of this review (self-report, refill data, and electronic monitoring) do not measure actual ingestion, and the use of self-report and electronic monitoring might have introduced response bias because of participants’ awareness of the measurements. However, all medication adherence related research has to deal with the limitations of adherence measurements. For now, our review provides the best evidence currently available, and clearly demonstrates the need for more high-quality, longitudinal research into associations between psychosocial predictors and medication non-adherence.

Two recommendations for future research can be made. First, future longitudinal research into psychosocial predictors of medication non-adherence should be of high quality. Researchers should, for example, use valid measures of psychosocial predictors and medication non-adherence and should thoroughly describe which steps were performed in the study, especially those relating to handling missing data and avoiding bias.

Second, the research gap in longitudinal studies into associations between psychosocial predictors and medication non- adherence in patients with conditions such as rheumatic diseases, migraine disorders, gout, glaucoma, and stomach ulcers (see Supplementary Materials) should be complemented. Although we assume that review findings will also apply to these diseases, this assumption needs to be confirmed.

The conclusion of this systematic review is that there is limited evidence for absence of longitudinal associations between psychosocial predictors and medication non-adherence. Consequently, our results do not provide psychosocial targets for the development of new interventions in clinical practice. However, the usefulness of psychosocial predictors in improving medication adherence should not be ruled out, as more high-quality research is needed to confirm or refute the conclusion of this review. Such future research could also further explore the associations found in this review between escape-avoidance coping and medication non-adherence, and between receiving practical, social support and medication adherence.

Supplementary materials

Pubmed search strategy

(((adult[MeSH Terms] OR mature[tw] OR adult[tw])

AND

((Ischaemic heart diseases[TW] OR angina pectoris[TW] OR Myocardial Ischemia[TW] OR asthma[TW] OR Diabetes mellitus[TW] OR diabetes mellitus[TW] OR hypercholesterolaemia[TW] OR hyperlipidaemia[TW] OR Dyslipidemias[TW] OR Gastric ulcer[TW] OR Duodenal ulcer[TW] OR Stomach Ulcer[TW] OR glaucoma[TW] OR glaucoma[TW] OR heart failure[TW] OR Heart failure[TW] OR arrhythmias[TW] OR Arrhythmias, Cardiac[TW] OR “Human immunodeficiency virus” OR HIV disease[TW] OR HIV-disease[TW] OR HIV infections[TW] OR HIV-infections[TW] OR Hypertensive diseases[TW] OR Hypertension[TW] OR Ulcerative colitis[TW] OR Crohn’s disease[TW] OR Inflammatory Bowel Diseases[TW] OR Arthropathies[TW] OR gout[TW] OR Malignant neoplasm of breast[TW] OR Breast Neoplasms[TW] OR Hereditary angioedema[TW] OR Angioedemas, Hereditary[TW] OR transplantation[TW] OR Organ Transplantation[TW] OR migraine[TW] OR Migraine Disorders[TW] OR osteoporosis[TW] OR arthropathy[TW] OR Systemic connective tissue disorders[TW] OR psoriatic arthropathy[TW] OR rheumatoid arthritis[TW] OR Systemic lupus erythematosus[TW] OR Systemic sclerosis[TW] OR Arthritis, Psoriatic[TW] OR Arthritis, Rheumatoid[TW] OR Lupus Erythematosus, Systemic[TW] OR Scleroderma, Systemic[TW] OR Arterial embolism[TW] OR thrombosis[TW] OR venous embolism[TW] OR Embolism and Thrombosis[TW] OR Paget Disease[TW] OR Osteitis Deformans[TW]) OR (Myocardial Ischemia[MH] OR asthma[MH] OR diabetes mellitus[MH] OR Dyslipidemias[MH] OR Stomach Ulcer[MH] OR glaucoma[MH] OR Heart failure[MH] OR Arrhythmias, Cardiac[MH] OR HIV infections[MH] OR Hypertension[MH] OR Inflammatory Bowel Diseases[MH] OR gout[MH] OR Breast Neoplasms[MH] OR Angioedemas, Hereditary[MH] OR Organ Transplantation[MH] OR Migraine Disorders[MH] OR osteoporosis[MH] OR Arthritis, Psoriatic[MH] OR Arthritis, Rheumatoid[MH] OR Lupus Erythematosus, Systemic[MH] OR Scleroderma, Systemic[MH] OR Embolism and Thrombosis[MH] OR Osteitis Deformans[MH]))

AND

((medication adherence[MH] OR patient compliance[MH]) OR (medication compliance[TW] OR medication noncompliance[TW] OR medication non compliance[TW] OR medication noncompliance[TW] OR medication adherence[TW] OR medication non-adherence[TW] OR medication non adherence[TW] OR medication nonadherence[TW] OR medication adherance[TW] OR medication non-adherance[TW] OR medication non adherance[TW] OR medication nonadherance[TW] OR medication persistence[TW] OR medication non-persistence[TW] OR medication non persistence[TW] OR medication nonpersistence[TW] OR medication persistance[TW] OR medication non-persistance[TW] OR medication non persistance[TW] OR medication nonpersistance[TW] OR medicine compliance[TW] OR medicine non-compliance[TW] OR medicine non compliance[TW] OR medicine noncompliance[TW] OR medicine adherence[TW] OR medicine non-adherence[TW] OR medicine non adherence[TW] OR medicine nonadherence[TW] OR medicine adherance[TW] OR medicine non-adherance[TW] OR medicine non adherance[TW] OR medicine nonadherance[TW] OR medicine persistence[TW] OR medicine non-persistence[TW] OR medicine non persistence[TW] OR medicine nonpersistence[TW] OR medicine persistance[TW] OR medicine non-persistance[TW] OR medicine non persistance[TW] OR medicine nonpersistance[TW] OR medical compliance[TW] OR medical non-compliance[TW] OR medical non compliance[TW] OR medical noncompliance[TW] OR medical adherence[TW] OR medical non-adherence[TW] OR medical non adherence[TW] OR medical nonadherence[TW] OR medical adherance[TW] OR medical non-adherance[TW] OR medical non adherance[TW] OR medical nonadherance[TW] OR medical persistence[TW] OR medical non-persistence[TW] OR medical non persistence[TW] OR medical nonpersistence[TW] OR medical persistance[TW] OR medical non-persistance[TW] OR medical non persistance[TW] OR medical nonpersistance[TW] OR drug compliance[TW] OR drug non-compliance[TW] OR drug non compliance[TW] OR drug noncompliance[TW] OR drug adherence[TW] OR drug non-adherence[TW] OR drug non adherence[TW] OR drug nonadherence[TW] OR drug adherance[TW] OR drug non-adherance[TW] OR drug non adherance[TW] OR drug nonadherance[TW] OR drug persistence[TW] OR drug non-persistence[TW] OR drug non persistence[TW] OR drug nonpersistence[TW] OR drug persistance[TW] OR drug non-persistance[TW] OR drug non persistance[TW] OR drug nonpersistance[TW] OR drugs compliance[TW] OR drugs non-compliance[TW] OR drugs non compliance[TW] OR drugs noncompliance[TW] OR drugs adherence[TW] OR drugs non-adherence[TW] OR drugs non adherence[TW] OR drugs nonadherence[TW] OR drugs adherance[TW] OR drugs non-adherance[TW] OR drugs non adherance[TW] OR drugs nonadherance[TW] OR drugs persistence[TW] OR drugs non-persistence[TW] OR drugs non persistence[TW] OR drugs nonpersistence[TW] OR drugs persistance[TW] OR drugs non-persistance[TW] OR drugs non persistance[TW] OR drugs nonpersistance[TW]))

AND

(Prospective Studies[MH] OR Longitudinal Studies[MH] OR Cohort Studies[MH] OR Follow-up Studies[MH] OR Retrospective Studies[MH] OR Prospective Studies[TIAB] OR Longitudinal Studies[TIAB] OR Cohort Studies[TIAB] OR Follow-up Studies[TIAB] OR Retrospective Studies[TIAB] OR observational stud*[TIAB] OR predict*[TW] OR prognos*[TW] OR prognostic factor*[TW] OR course[TW] OR determinant*[TW]))

NOT

“Africa”[Mesh] OR “Latin America”[Mesh] OR “Asia, Central”[Mesh] OR “Borneo”[Mesh] OR “Brunei”[Mesh] OR “Cambodia”[Mesh] OR “East Timor”[Mesh] OR “Laos”[Mesh] OR “Malaysia”[Mesh] OR “Mekong Valley”[Mesh] OR “Myanmar”[Mesh] OR “Philippines”[Mesh] OR “Singapore”[Mesh] OR “Thailand”[Mesh] OR “Vietnam”[Mesh] OR “Bangladesh”[Mesh] OR “Bhutan”[Mesh] OR “India”[Mesh] OR “Afghanistan”[Mesh] OR “Bahrain”[Mesh] OR “Iran”[Mesh] OR “Egypt”[Mesh] OR “Iraq”[Mesh] OR “Israel”[Mesh] OR “Jordan”[Mesh] OR “Kuwait”[Mesh] OR “Lebanon”[Mesh] OR “Oman”[Mesh] OR “Qatar”[Mesh] OR “Saudi Arabia”[Mesh] OR “Syria”[Mesh] OR “United Arab Emirates”[Mesh] OR “Yemen”[Mesh] OR “Nepal”[Mesh] OR “Pakistan”[Mesh] OR “Sri Lanka”[Mesh] OR “China”[Mesh] OR “Korea”[Mesh] OR “Mongolia”[Mesh] OR “Taiwan”[Mesh]

NOT

(youth[TIAB] OR child*[TIAB])

NOT

(Clinical Trial[MH] OR case reports[PT] OR review[PT] OR meta-analysis[MH] OR Cross-sectional Studies[MH] OR Case-control Studies[MesH:NoExp] OR Clinical Trial*[PT] OR case report*[PT] OR review*[PT] OR meta-analys*[PT] OR case report*[TIAB] OR case-report*[TIAB] OR review*[TIAB] OR systematic review*[TIAB] OR meta-analys*[TIAB] OR randomized controlled trial*[TIAB] OR randomised controlled trial*[TIAB] OR clinical trial*[TIAB] OR controlled clinical trial*[TIAB] OR cross-sectional*[TIAB] OR cross sectional*[TIAB] OR Case-control Studies[TIAB] OR case-control[TIAB] OR case control[TIAB] OR Editorial[ptyp] OR Letter[ptyp] OR Comment[ptyp] OR Interview[ptyp] OR Newspaper Article[ptyp])

Chronic preventive maintenance medication

Table S1.

Framework for judging methodological quality

Bias domain	Criterion	Score	Judgment	Final score
1. Study participation	1.1. The setting of the source population is adequately described by key characteristics (setting/geographical location)	○ Yes ○ Partly ○ No	5 × yes = yes 1 × no = no Else = partly	○ Yes ○ Partly ○ No
	1.2. The (baseline) study sample is adequately described by key characteristics (descriptive data about age, sex, diagnosis, disease duration and medication type/group), and no unacceptable level of bias is present	○ Yes ○ Partly ○ No
	1.3. The method of recruitment or sampling is adequately described. If method of recruitment is not ‘consecutive’, then, for example, descriptions are given about the sampling frame, numbers, methods to identify the sample (such as a description of referral patterns in health care) and period of recruitment, and no unacceptable level of bias is present	○ Yes ○ Partly ○ No
	1.4. Inclusion and exclusion criteria are adequately described, and no unacceptable level of bias is present	○ Yes ○ Partly ○ No
	1.5. There is adequate participation in the study by eligible individuals (power analysis is described or the sample size (n) is adequate in relation to the number of prognostic variables (K) in the statistical analyses (ratio n:K exceeds 10:1)	○ Yes ○ Partly ○ No
2. Study attrition	2.1. Response rate (ie, proportion of study sample completing the study and providing outcome data) is adequate If study sample size ≤50 participants: ‘yes’ when total number of participants lost to follow-up was <10% at follow-up ≥three months. ‘Partly’: if this percentage was between 10% and 20%. ‘No’: if this percentage was ≥20% If study sample size >50 participants: ‘yes’, when total number of participants lost to follow-up was <20% at follow-up $three months. ‘Partly’: if this percentage was between 20% and 33%. ‘No’: if this percentage was ≥33%	○ Yes ○ Partly ○ No	2.1 yes = yes (you can leave 2.2 open) 2.1 no = no OR 2.1 partly, 2.2 no = no Else = partly	○ Yes ○ Partly ○ No
	2.2. Attempts to collect information about participants who dropped out of the study are described: 1) reasons for loss to follow-up are provided OR 2) participants lost to follow-up are adequately described by key characteristics and outcomes. No unacceptable level of bias is present	○ Yes ○ Partly ○ No
3. Prognostic factor measurement	3.1. A clear description of the main prognostic factors is provided (not covariates) AND/OR measures/methods regarding the main prognostic factors, at baseline and follow-up are adequately described to allow assessment of their validity and reliability. No unacceptable level of bias is present  ○ Objective measures (such as number of life-changing events) and clear description is ‘yes’. Poor/no description = ‘partly’  ○ Validated, subjective measures (eg, opinions) and clear description = ‘yes’. Poor/no description = ‘partly’  ○ Non-validated, subjective measures and clear description = ‘partly’. Poor/no description = ‘no’	○ Yes ○ Partly ○ No	4 × yes = yes 3.1 or 3.2 no = no OR 3.1 or 3.2 partly (no no’s), 3.3 or 3.4 no = no Else = partly	○ Yes ○ Partly ○ No
	3.2. The method and setting of measurement are the same for all study participants at baseline and follow-up	○ Yes ○ Partly ○ No
	3.3. Continuous variables are reported or appropriate cut-off points are used	○ Yes ○ Partly ○ No
	3.4. Authors appropriately described and dealt with missing data on prognostic factors	○ Yes ○ Partly ○ No
4. Outcome measurement	4.1. A clear description of medication adherence is provided AND/OR measures/methods of medication adherence (at baseline and follow-up) are adequately described, to allow assessment of their validity and reliability. No unacceptable level of bias is present  ○ Objective measures (such as pill count, refill rates, MEMS) and clear description = ‘yes’. Poor/no description is ‘partly’  ○ Validated, subjective measures (eg, questionnaires) and clear description = ‘yes’. Poor/no description = ‘partly’  ○ Non-validated, subjective measures and clear description = ‘partly’. Poor/no description = ‘no’	○ Yes ○ Partly ○ No	3 × yes = yes 4.1 or 4.2 no = no OR 4.1 or 4.2 partly (no no’s), 4.3 no = no Else = partly	○ Yes ○ Partly ○ No
	4.2. The method and setting of measurement are the same for all study participants at baseline (if measured) and follow-up	○ Yes ○ Partly ○ No
	4.3. Authors appropriately described and dealt with missing outcome data	○ Yes ○ Partly ○ No
5. Confounding measurement and account	5.1. The most important confounders are measured Examples of possible confounders: age; socioeconomic status/educational level/financial situation/illiteracy; social support/networks; depression/anxiety/emotional distress/lack of acceptance of disease; fatigue/pain/physical disability; self-efficacy/coping; regimen complexity/route of administration/number of medications; satisfaction with patient-provider relationship/autonomy	○ Yes ○ Partly ○ No
	5.2. A clear description of the most important confounders measured is provided AND/OR measures/methods of the most important confounders (at baseline) are adequately described to allow assessment of their validity and reliability. No unacceptable level of bias is present  ○ Objective measures (such as age, sex) and clear description = ‘yes’. Poor/no description is ‘partly’  ○ Validated, subjective measures (eg, opinions) and clear description = ‘yes’. Poor/no description = ‘partly’  ○ Non-validated, subjective measures and clear description = ‘partly’. Poor/no description = ‘no’	○ Yes ○ Partly ○ No	One of 5.1 to 5.4 = no (if 5.1 no, you can leave 5.2 to 5.5 open) OR 5.1 to 5.4 partly, 5.5 no = no All partly = partly OR 5.1 to 5.4 partly, 5.5 yes = partly OR none of 5.1 to 5.4 no, 5.5 no = partly OR 5.1 to 5.4 yes, 5.5 not yes = partly Else = yes	○ Yes ○ Partly ○ No
	5.3. The method and setting of confounding measurement are the same for all study participants at baseline	○ Yes ○ Partly ○ No
	5.4. Important potential confounders are accounted for in the study design (eg, matching for key variables/restriction) OR in analysis (stratification/multivariate techniques)	○ Yes ○ Partly ○ No
	5.5. Authors appropriately described and dealt with missing confounding data	○ Yes ○ Partly ○ No
6. Analysis	6.1. There is sufficient presentation of data to assess the adequacy of the analysis ‘Yes’, if main findings of the study and statistical methods used are clearly described: simple outcome data, crude data and estimates of random variability should be reported, so that the reader can check the major analyses and conclusions	○ Yes ○ Partly ○ No	4 × yes = yes At least 1 × no = no Else = partly	○ Yes ○ Partly ○ No
	6.2. The statistical tests used to assess the main outcome are appropriate For example, non-parametric methods should be used for small sample sizes	○ Yes ○ Partly ○ No
	6.3. The strategy for model building (ie, inclusion of variables) is appropriate, and is based on conceptual thoughts, a framework or a model For example: variables that do not correlate with the main outcome of interest are not used in multivariate analysis. Proper variables are entered in logical steps into the multivariate model	○ Yes ○ Partly ○ No
	6.4. The selected model is adequate for the design of the study For example: in repeated measures, a repeated-measure model should be used. If outcome is binominal, logistic regression should be used, etcetera. If delta outcome is being investigated, models should to be adjusted for baseline outcome values	○ Yes ○ Partly ○ No

Table S2.

explanation of measures and results^*

First author	Setting, n patients	Measures		Psychcat||	Results¶		Direction of association (regarding adherence)**	N domains bias free††
		Adherence†, follow-up period‡	Psychosocial predictors§		Univariate	Multivariate
Asthma (inhaled corticosteroids)
Ponieman1	USA; patients from general internal medicine clinic, n=261	Adherence by self-report (MARS), 3 months	(Items derived from BMQ and Self-Regulation Theory): concerns beliefs: worried about side effects of ICS? Concerns beliefs: worried about getting addicted to ICS?	AI	OR =0.3 (0.2, 0.7), P<0.05	OR =0.52 (0.36, 0.74), P<0.001	U: − M: −	0 of 6
			Concerns beliefs: if I use ICS all the time they will stop working	AI	OR =0.4 (0.2, 0.8), P<0.05	NS‡‡	U: − M: 0
			Necessity beliefs: important to use ICS when symptomatic?	AI	OR =0.4 (0.2, 0.9), P<0.05	NS‡‡	U: − M: 0
			Necessity beliefs: important to use ICS when asymptomatic?	AI	NS	NS‡‡	U: 0 M: 0
			Self-efficacy: confident in ability to use ICS as prescribed	AI	OR =5.8 (2.3, 14.6), P<0.05	OR =4.15 (2.54, 6.77), P<0.001	U: + M: +
			Self-efficacy: confident in ability to control asthma	AIII	OR =3.5 (1.6, 7.6), P<0.05	OR =2.23 (1.42, 3.52), P<0.001	U: + M: +
			Self-efficacy: confident in controlling future health	AIII	NS	NS‡‡	U: 0 M: 0
				AIII	NS	NS‡‡	U: 0 M: 0
Diabetes (oral and/or parenteral anti-diabetics)
Venturini2,§§	USA; patients from HMO-providing health services, n=786	Adherence by record review (continuous measure corrected for self-reported baseline adherence), last time point flexible, but within 24 months	Perception of mental health (mood state, SF-36)	EI	NR	NS	U: NR M: 0	2 of 6
Heart disease and hypertension (cardiovascular medication)
Gazmararian3,|| ||	USA; community-dwelling patients, n=1,549	Non-adherence by record review, 12 months	Social support (instrument NR)	CIII	NS	NT	U: 0 M: NT	3 of 6
Nabi4	Finland: local government employees, n=1,021	Non-adherence by record review (ordinal measure), 12 months	Anxiety (ATS)	EI	NS	NT	U: 0 M: NT	1 of 6
			Hostility (FTSSH)	D	NS	NT	U: 0 M: NT
			Optimism (LOT-R)	D	NS	NT	U: 0 M: NT
			Pessimism (LOT-R)	D	NS	NT	U: 0 M: NT
			Psychological distress (GHQ)	EI	NS	NT	U: 0 M: NT
			Sense of coherence (SOC)	D	OR =0.62 (0.36, 1.05), P<0.10	OR =0.55 (0.31, 0.96), P<0.05	U: 0 M: +
Grégoire5	Canada: hypertensive adults with prescription from network of pharmacies, n=692	Non-adherence by self-report (Morisky Scale), 3 months	(Interview, self-developed items): beliefs concerning efficacy of antihypertensive medication	AI	NS	NS	U: 0 M: 0	0 of 6
			Beliefs concerning hypertension as risk factor for other diseases	AII	“No effect” versus “a lot of effect” (ref cat): OR =1.74 (1.08, 2.81), P=0.02	“No effect” versus “a lot of effect”: OR =2.00 (1.21, 3.33), P≤0.05	U: − M: −
			How much are you at risk of a heart attack because of your hypertension if you follow your doctor’s advice?	AII	NS	NS	U: 0 M: 0
			How much are you at risk of a stroke because of your hypertension if you follow your doctor’s advice?	AII	NS	NS	U: 0 M: 0
			How much are you at risk of heart attack because of your hypertension if you do not do anything about it?	AII	“Do not know” versus “no to moderate risk” (ref cat): OR =0.46 (0.19, 1.12), P=0.09	NS	U: 0 M: 0
			How much are you at risk of stroke because of your hypertension if you do not do anything about it?	AII	“Do not know” versus “no to moderate risk” (ref cat): OR =0.44 (0.17, 1.16), P=0.10	“Do not know” versus “no to moderate risk”: OR =0.40 (0.15, 1.09), P=0.07	U: 0 M: 0
			Social support (Pearlin et al31)	CIII	NS	NS	U: 0 M: 0
Miller6,§§§	Site not reported: patients from institutions providing cardiac rehabilitation programs, n=141	Adherence by self-report (continuous measure, HBS), 6–9 months	Attitude towards medication taking (MAS)	AI	NR	NS	U: NR M: 0	0 of 6
			Beliefs about which steps of the medical regimen people most important to them think they should perform (HIS)	CII		NS	U: NR M: 0
Molloy7,§§§	UK; patients admitted to one of four London hospitals with Acute Coronary Syndrome, n=295	Adherence by self-report, 12 months	Emotional support (derived from Berkman et al32 and Seeman et al33)	CIII	NS	NS	U: 0 M: 0	1 of 6
			Practical support	CIII	Number of patients providing practical support: 0: 39.7% adherent. 1: 40.5% adherent. Two or more: 59.2% adherent, P=0.004	OR =2.12 (1.06, 4.26), P=0.03	U: + M: +
HIV (antiretroviral medication)
Deschamps8	Belgium; outpatients of university hospital, n=60	Non-adherence by MEMS, 5–6 months after measuring psychosocial constructs	Anxiety (AMHI)	EI	NS	NR	U: 0 M: NR	1 of 6
			Coping style: confrontational (AWC)	BI	NS		U: 0 M: NR
			Coping style: distancing	BII	NS		U: 0 M: NR
			Coping style: self-controlling	BII	NS		U: 0 M: NR
			Coping style: seek social support	CIII	NS		U: 0 M: NR
			Coping style: accept responsibility	BII	NS		U: 0 M: NR
			Coping style: escape-avoidance (higher score = more escape-avoidance)	BII	Adherent patients 7.2, (2.2) versus non-adherent patients 10.1 (2.8), P=0.003		U: − M: NR
			Coping style: planful problem solving (higher score = more planful problem solving)|| || ||	BI	Adherent patients 7.5 (median), 3 (IQR) versus non-adherent patients 9 (median), 2 (IQR), P=0.049		U: − M: NR
			Coping style: positive reappraisal	BI	NS		U: 0 M: NR
			Depression (AMHI)	EI	NS		U: 0 M: NR
			Perceived benefits of treatment (APIAQ)	AI	Adherent patients 21 (3.5) versus non-adherent patients 18.7 (3.9), P=0.07		U: 0 M: NR
			Perceived severity of seriousness of implications when not taking medications adequately	AI	NS		U: 0 M: NR
			Perceived susceptibility of developing AIDS when not taking medications as prescribed	AI	NS		U: 0 M: NR
			Positive affect (eg, happiness person)	D	NS		U: 0 M: NR
			Received social support (AGSRP)	CIII	NS		U: 0 M: NR
			Self-efficacy in taking HAART medication (ALTMBSES)	AIII	NS		U: 0 M: NR
Holmes9	USA; HIV-clinic patients, n=116	Adherence by MEMS, 12 months (or when viral load of ≥1,000 copies/mL was reached)	Depressive symptoms (CES-D)	EI	High adherence 12.6 (11.3), low adherence 16.5 (11.7), P=0.06	NS	U: 0 M: 0	2 of 6
			HIV-disclosure worries (HAT-QOL)	AII	NS	NT	U: 0 M: NT
			Health worries (higher score = fewer worries)	AII	High adherence 79.2 (23.9), low adherence 70.4 (28.9), P=0.06	NS	U: 0 M: 0
			Medication worries (higher score = fewer worries)	AI	High adherence 86.1 (20.4), low adherence 83.3 (18.3), P=0.06	NS	U: 0 M: 0
			Provider trust	CI	NS	NT	U: 0 M: NT
			Social support (ISEL)	CIII	NS	NT	U: 0 M: NT
			Stress (PSS)	EII	High adherence 12.4 (7.8), low adherence 15.3 (8.2), P=0.07	NS	U: 0 M: 0
Delgado10	Canada; patients enrolled in community drug treatment program, n=316	Adherence by record review, 12 months	Depressive symptoms (CES-D)	EI	Not reporting depression: 79.8% adherent, reporting depression: 68.1% adherent, P=0.02	NS	U: − M: 0	1 of 6
Singh11	USA; new veteran patients seen at medical center, n=52	Non-adherence by record review, 6 months	Confusion and bewilderment (POMS)	BII	NS	NT	U: 0 M: NT	1 of 6
			Depression and dejection	EI	Adherent 14.2 (SEM 1.9), non-adherent 22.1 (SEM 3.4), P=0.04	NS	U: − M: 0
			Mood disturbance	EI	39% in adherent patients, 76% in non-adherent patients, P=0.03	OR =1.4 (1.1, 1.8), P=0.01	U: − M: −
			Religious support (instrument NR)	CIII	NS	NT	U: 0 M: NT
			Social support (instrument NR)	CIII	NS	NT	U: 0 M: NT
			Symptoms of depression (BDI)	EI	NS	NT	U: 0 M: NT
			Tension and anxiety (POMS)	EI	NS	NT	U: 0 M: NT
Singh12	Site not reported: patients in HIV-medical centers, n=138	Non-adherence by record review, 6 months	Coping style: active-behavioral focused (higher score = greater applicability of coping style to patient, BMICIS)	BI	(Mean score, SEM): non-adherent 5.2 (0.5) versus adherent 6.6 (0.2), P=0.01	NR	U: + M: NR	1 of 6
			Coping style: active-cognitive focused	BI	NS		U: 0 M: NR
			Coping style: avoidant coping	BII	Non-adherent 3.3 (0.3) versus adherent 2.6 (0.2), P=0.02		U: − M: NR
			Coping style: emotion-focused	BII	NS		U: 0 M: NR
			Coping style: problem-focused	BI	Non-adherent 6.0 (0.5) versus adherent 7.1 (0.2), P=0.02		U: + M: NR
			Hopelessness: future expectations	BII	NS		U: 0 M: NR
			Hopelessness: loss of motivation (higher score = more hopelessness, BHS)	BII	Non-adherent 1.75 (0.5), adherent 0.6 (0.1), P=0.006		U: − M: NR
			Hopelessness: negative feelings about future	BII	NS		U: 0 M: NR
			Hopelessness: total score	BII	NS		U: 0 M: NR
			Quality of life: psychological functioning (MOS SF-36)	EI	NS		U: 0 M: NR
			Satisfaction with social support: emotional (SSQ)	CIII	NS		U: 0 M: NR
			Satisfaction with social support: informational (higher scores = less satisfaction)	CIII	Non-adherent 7.9 (1.1), adherent 6.1 (0.3), P=0.04		U: + M: NR
			Satisfaction with social support: tangible	CIII	Non-adherent 7.7 (1.1), adherent 5.5 (0.3), P=0.07		U: 0 M: NR
			Satisfaction with social support: total score	CIII	Non-adherent 22.9 (3.3), adherent 16.8 (0.75), P=0.03		U: + M: NR
Bottonari13	USA; patients treated in immunodeficiency clinic, n=78	Adherence by self-report (straightforward), 6–9 months	Depressive symptoms (IDD)	EI	NS	NR	U: 0 M: NR	0 of 6
			Experience of general (stressful) life events (LES)	EII	NS		U: 0 M: NR
			HIV-specific (stressful) life events (BHLES)	EII	NS		U: 0 M: NR
			Neuroticism: personality style indicative of affective instability (NSEPQSS)	D	NS		U: 0 M: NR
			Perceived stress (PSS)	EII	OR =0.88 (0.77, 0.98), P=0.04		U: − M: NR
			Self-esteem (RSEQR)	D	NS		U: 0 M: NR
Godin14	Canada; patients from medical HIV-clinics, n=400	Adherence over time by self-report (straightforward), 12 months	Change in predictors related to adherence over time: attitude towards medication-taking (more positive attitude = greater adherence, self-developed scale)	AI	NR	OR =1.56 (1.18, 2.06), P≤0.05	U: NR M: +	1 of 6
			Optimism (DOS)	D		NS	U: NR M: 0
			Outcome expectations (eg, believe that specific course of action will lead to desired outcome, self-developed scale)	AIII		NS	U: NR M: 0
			Patient-doctor satisfaction (Pat SS)	CI		NS	U: NR M: 0
			Self-efficacy regarding medication taking (self-developed scale)	AIII		OR =1.68 (1.27, 2.22), P≤0.05	U: NR M: +
			Social support (SPS)	CIII		NS	U: NR M: 0
Kacanek15	USA; patients recruited by media and physician networks, n=225	Suboptimal adherence by self-report (straightforward): max 30 months	Development of depressive symptoms (BST)	EI	Suboptimal adherence in those who developed depressive symptoms =45.1% versus 25.9% in those with no depressive symptoms, P=0.01	NT	U: − M: NT	2 of 6
Martini16,§§§	Italy; outpatients using combination therapy, n=214	Adherence by self-report (ordinal measure, straightforward questionnaire), 12 months	(Interview, instrument NR): perception of therapy: reliable?	AI	In “high adherence” category, therapy perceived as “reliable” by 15.6%, and “not reliable” by 84.4%. In “variable adherence” cat 4.8% versus 95.2%. In “low adherence” cat 0% versus 100%, P=0.02	NR	U: + M: NR	0 of 6
			Perception of therapy: enslaving?	AI	NS		U: 0 M: NR
			Satisfied about doctor/patient discussion regarding clinical and therapeutic aspects of treatment?	CI	In “high adherence” category: “sufficient/highly satisfied” = 73.9%, “little/not satisfied” =26.1%. In “variable adherence” cat 80% versus 20%. in “low adherence” cat 50% versus 50%, P=0.05		U: ? M: NR
Mellins17	USA; HIV-infected mothers recruited in waiting room of adult clinic, n=128	Non-adherence by self-report (AACTG, straightforward), T1 after 4–5 months, T2 8–18 months after T1	Negative stressful events (PEI)	EII	OR =1.27 (1.09, 1.49), P<0.01 at T1, OR =1.28 (1.05, 1.57), P=0.02 at T2	NR	U: − M: NR	0 of 6
			Parenting stress (low scores = more stress, PPCS)	EII	OR =0.86 (0.76, 0.98), P=0.02 at T2		U: − M: NR
			Psychological distress (aggregated demoralization score, DSPERI)	EI	NS		U: 0 M: NR
			Self-efficacy in carrying out health-related behaviors (Chesney et al34)	AIII	NS		U: 0 M: NR
Nilsson Schönnesson18	Sweden; patients recruited by clinic nurses, n=203	Adherence by self-report (straightforward), 24 months	Anxiety symptoms (ASBSI)	EI	NR	NS	U: NR M: 0	1 of 6
			Belief in adherence necessity (one item)	AI		NS	U: NR M: 0
			Belief that ART prolongs one’s life (one item)	AI		NS	U: NR M: 0
			Belief in future HIV-related health problems (self-developed scale)	AII		NS	U: NR M: 0
			Belief in influencing HIV disease (MAH)	AII		NS	U: NR M: 0
			Beliefs in ART health concerns (eg, believe that medication makes sicker, one item)	AI		NS	U: NR M: 0
			Coping mode: helplessness (MAH)	BII		NS	U: NR M: 0
			Coping mode: resilience (MAH)	D		NS	U: NR M: 0
			Depressive symptoms (DSBSI)	EI		NS	U: NR M: 0
			Global social support satisfaction (one item)	CIII		NS	U: NR M: 0
			Hopelessness (BHS)	BII		NS	U: NR M: 0
			Life stress (LSS)	EII		NS	U: NR M: 0
			Patient-provider relationship (self-developed scale)	CI		NS	U: NR M: 0
			Perceived pressure to take HIV medication (self-developed scale)	CIII		NS	U: NR M: 0
			Posttraumatic stress disorder symptoms related to HIV diagnosis (HIE)	EI		NS	U: NR M: 0
			Self-efficacy in taking medication (self-developed scale)	AIII		NS	U: NR M: 0
Thrasher19	USA; patients in public use of HCSUS data-set, n=1,911	Adherence by self-report (straightforward),12 months	(Instruments NR): depressive symptoms	EI	OR =0.98 (0.96, 0.99), P=0.007	NR	U: − M: NR	1 of 6
			Dysthymia symptoms	EI	OR =0.92 (0.87, 0.96), P=0.001		U: − M: NR
			Social support	CIII	NS		U: 0 M: NR
Horne20	UK; outpatients, eligible to receive HAART, n=136	Adherence by self-report (VAS-scale from MASRI, straightforward), 12 months	Depressive symptoms (HADS)	EI	NS	NT	U: 0 M: NT	3 of 6
			HAART concern beliefs about medication (BMQ)	AI	High adherence 2.9 (0.6) versus low adherence 3.3 (0.6), P=0.005	OR =0.45 (0.22, 0.96), P=0.038	U: − M: −
			HAART necessity beliefs about medication	AI	High adherence 4.0 (0.6) versus low adherence 3.7 (0.6), P=0.006	OR =2.19 (1.02, 4.71), P=0.045	U: + M: +
Mugavero21	USA; patients receiving care at one of eight infectious disease clinics, n=474	Non-adherence by self-report (AACTG, straightforward, corrected for baseline non-adherence), 27 months	Number of severe stressful events (LES, modified version)	EII	OR (per event) =1.14 (1.03, 1.26)	NS	U: − M: 0	3 of 6
			Number of stressful events (moderate + severe stressful events)	EII	OR (per event) =1.09 (1.04, 1.13)	OR (per event) =1.10 (1.04, 1.16)	U: − M: −
			Number of traumatic events	EII	OR (per event) =1.73 (1.24, 2.39)	NS	U: − M: 0
			Number of types of lifetime traumatic experiences (composite measure of diverse questionnaires)	EII	NS	NS	U: 0 M: 0
Carrieri22	France; patients starting HAART-regimen including at least oneprotease inhibitor, n=1,110	Non-adherence by self-report (AACTG,straightforward), 60 months	Depressive symptoms (CES-D)	EI	b =0.22 (95% CI =0.12, 0.32), P<0.001	b =0.18 (0.07, 0.29)	U: − M: −	2 of 6
			Support from partner (whether principal or not, instrument NR)	CII	b =−0.16 (−0.26, −0.07), P=0.001	b =−0.15 (−0.25, −0.05)	U: + M: +
Transplant-related (immunosuppressant medication)
Stilley23	USA; transplant patients, recruited before hospital discharge or atearly clinic visit, n=152	Adherence by MEMS (continuous measure), 6 months	Affective dysregulation (degree of negative affectivity and irritability, DI)	EI	Correlation coefficient: NS¶¶	NR	U: 0 M: NR	1 of 6
			Behavioral dysregulation (impulsivity, sensation seeking, aggression)	D	r=0.26, P≤0.05***		U: −††† M: NR
			Cognitive dysregulation (less strategic thinking, problem solving, self-monitoring)	BI	NS¶¶		U: 0 M: NR
			Family environment (family support, FRI)	CII	NS¶¶		U: 0 M: NR
			Hostility (CMHS)	D	NS¶¶		U: 0 M: NR
De Geest24	Belgium; convenience sample of outpatients, n=101	Non-adherence by MEMS (ordinal measure, correction for past adherence), 6 months	Depressive symptoms (BDI)	EI	NR	NS	U: NR M: 0	2 of 6
			Self-efficacy in taking medication (LTMSES)	AIII		Median =4.85 (Q1 =4.70, Q3 =5.00) for excellent adherers, 4.81 (Q1 =4.70, Q3 =4.89) for moderate non-adherers, 4.41 (Q1 =4.30, Q3 =4.81) for minor adherers, P=0.04	U: NR M: +
			Social support (PRQ)	CIII		NS	U: NR M: 0
			Symptom distress (ATSFDS)	EII		NS	U: NR M: 0
Russell25	USA; convenience sample of renal transplant patients, n=50	Adherence by MEMS (ordinal measure), 12 months	Depressive symptoms (BDI)	EI	NS	NR	U: 0 M: NR	0 of 6
			Emotional burden (MS)	EI	NS		U: 0 M: NR
			Self-efficacy in taking medication (LTMSES)	AIII	NS		U: 0 M: NR
			Social support (SSAI)	CIII	NS		U: 0 M: NR
Weng26	USA; patients recruited at time of renal transplantation, n=829	Adherence by MEMS (ordinal measure), 12 months posttransplantation	Beliefs regarding who or what controls and influences one’s health (MHLCS)	AIII	OR =1.05 (1.00, 1.11), P=0.05 (powerful others subscale)	NS	U: + M: 0	2 of 6
			Depressive symptoms (CES-D)	EI	NS	NT	U: 0 M: NT
			Perceived stressfulness of transplant-related issues (TSQ)	EII	NS	NT	U: 0 M: NT
			Perceptions that social needs are being met (friends and family sub-score, SSAS)	CII	NS	NT	U: 0 M: NT
Dew27	USA; heart transplant patients at academic hospital‡‡‡, n=108	Non-adherence by self-report (straight-forward), 12 months post-transplantation	Coping strategies: use of active-behavioral coping (Coping checklist)	BI	NS	NT	U: 0 M: NT	2 of 6
			Coping strategies: use of active-cognitive coping	BI	NS	NT	U: 0 M: NT
			Coping strategies: use of avoidance coping (% high)	BII	Non-adherent 58.8%, adherent 29.9%, P<0.05	OR =9.71, P<0.05	U: − M: −
			Emotional status: anger-hostility symptoms (SCL-90)	EI	Non-adherent 47.1%, adherent 12.1%, P<0.001	OR =13.40, P<0.05	U: − M: −
			Emotional status: anxiety symptoms	EI	Non-adherent 82.4%, adherent 53%, P<0.05	NS	U: − M: 0
			Emotional status: depressive symptoms	EI	NS	NT	U: 0 M: NT
			Sense of mastery (ie, control over life, SMS)	AIII	NS	NT	U: 0 M: NT
			Social support: caregiver support (% poor) (Spanier,35 Pearlin and Schooler)36	CII	Non-adherent 52.9%, adherent 27.0%, P<0.05	NS	U: − M: 0
			Social support: friend support (Moos)37	CII	NS	NT	U: 0 M: NT
Dew28	USA; patients receiving first lung transplantation in academic hospital, n=178	Non-adherence by self-report (straightforward), 24 months	Anger-hostility symptoms (SC)	EI	(Correlation coefficient, significant if r≥0.15***): r≥0.15	NS	U: ? M: 0	1 of 6
			Anxiety symptoms (SC)	EI	r≥0.15	NS	U: ? M: 0
			Care provider locus of control (health outcomes due to professional? MHLCS)	AIII	r≥0.15	NS	U: ? M: 0
			Chance locus of control (health outcomes occur by chance?)	AIII	r≥0.15	NS	U: ? M: 0
			Degree to which one can rely on friends for emotional/practical support/friend support (Moos)37	CII	r≥0.15	NS	U: ? M: 0
			Depressive symptoms (SC)	EI	r≥0.15	NS	U: ? M: 0
			Expectations about the future/optimism (LOT)	D	r≥0.15	NS	U: ? M: 0
			Internal locus of control (can I influence my health outcome? MHLCS)	AIII	r≥0.15	NS	U: ? M: 0
			Supportiveness (both emotionally and practically) of recipient’s relationship with their primary family caregiver (when low = higher odds) (DAS)	CII	r≥0.15	OR =2.59 (1.20, 5.58), P<0.05	U: ? M: −
Dobbels29	Belgium: heart, liver and lung transplant patients listed at university hospitals, n=186	Non-adherence by self-report (straightforward, corrected for pre-transplant adherence), 12 months posttransplantation	Agreeableness (one’s orientation along continuum from compassion to antagonism, NEO-FFI)	D	NR	NT or NS	U: NR/NS M: 0	1 of 6
			Anxiety symptoms (HADS)	EI		NT or NS	U: NR/NS M: 0
			Conscientiousness (ie, degree of organization, NEO-FFI)	D		OR =0.80 (0.67, 0.95), P=0.01	U: NR/NS M: +
			Depressive symptoms (HADS)	EI		NT or NS	U: NR/NS M: 0
			Extraversion (capacity for joy, need for stimulation, NEO-FFI)	D		NT or NS	U: NR/NS M: 0
			General received practical and informational support (SSQ)	CIII		NT or NS	U: NR/NS M: 0
			Neuroticism (NEO-FFI)	D		NT or NS	U: NR/NS M: 0
			Openness to experience (toleration for and exploration of the unfamiliar, NEO-FFI)	D		NT or NS	U: NR/NS M: 0
			Received specific support with medication taking (SSQ)	CIII		OR =0.94 (0.89, 0.99), P=0.03	U: NR/NS M: +
DiMatteo30,|| ||	USA: patients from five medical specialties in HMOs, large multispecialty groups or solo practices, n=max 1,828§§§	Adherence by self-report (straightforward, continuous measure, correction for baseline adherence), 24 months	Health distress (instrument NR)	EII	NR	ß =−0.22, P=0.05	U: NR M: −	0 of 6
			Perceptions of physician’s authoritativeness (self-developed scale)	CI		NT or NS	U: NR M: 0
			Satisfaction with interpersonal medical care (Sherbourne)38	CI		NT or NS	U: NR M: 0
			Social support (composite measure, Sherbourne and Stewart)39	CIII		NT or NS	U: NR M: 0
			Tendency to use avoidance coping (instrument NR)	BII		NT or NS	U: NR M: 0

^*NS (non significant): as reported in the concerning study. UD (undetermined): because of inadequate description in the concerning study.

^†Binary outcome measure, unless indicated otherwise. With a straightforward question, we mean that participants were directly asked to indicate how many medication doses they missed. For example: “How many pills did you take this week?”;

^‡follow-up period = number of months between baseline (unless indicated otherwise) and last adherence measurement;

^§if no instrument is mentioned for predictor, then previous mentioned instrument is applicable;

^||psychosocial category, to which a predictor was assigned. A = Beliefs and cognitions about: I) medication and treatment; II) illness; III) self-efficacy and locus of control. B = coping styles: I) task oriented, II) emotion oriented. C = Social influences and social support: I) regarding medical caregiver; II) regarding friends and family; III) in general. D = personality traits. E = psychological well-being: I) mood state; II) perceived stress/stressors;

^¶OR: Odds Ratio (95% confidence interval). OR <1 = lower chance of being adherent or non-adherent (for direction in relevant study, see column “Adherence, follow-up period”) when predictor increases or when predictor ≠ reference category. OR > 1 = greater change of being adherent or non-adherent when predictor increases (or when predictor ≠ reference category). Scores other than OR are the mean predictor scores with standard deviation, unless indicated otherwise;

^**+ = higher level of predictor implies higher adherence at level P≤0.05; − = higher level of predictor implies less adherence at P≤0.05; 0 = no significant association between predictor and adherence at P≤0.05; ? = association present, but direction unclear;

^††to determine methodological quality, six bias domains per study were judged. Here, the total amount of bias free domains is reported (for further details, see table S3);

^‡‡assumed that all variables, tested by univariate analysis, were also tested by multivariate analysis;

^§§retrospective design;

^{|| ||}Diagnosis for coronary heart disease, hypertension, diabetes mellitus and/or hyperlipidaemia;

^¶¶not reported in study is interpreted by HZ/BvdB as not significant;

^***significance of P≤0.05 assumed by HZ/BvdB;

^†††negative association assumed;

^‡‡‡type of medication is immunosuppressants, antihypertensives, and/or antivirals;

^§§§use of chronic preventive medication assumed;

^{|| || ||}unexpected direction.

Abbreviations: AACTG, adult AIDS clinical trials group; ALTMBSES, adapted long term medication behavior self efficacy scale; AGSRP, adapted gay service research project; AIDS, acquired immunodeficiency syndrome; AMHI, adapted mental health inventory; APIAQ, adapted protease inhibitor attitude questionnaire; ART, antiretroviral therapy; ASBSI, anxiety subscale of brief symptom inventory; ATS, anxiety trait scale; ATSFDS, adapted version of transplant symptom frequency and distress scale; AWC, adapted ways of coping BDI, beck depression inventory; BHLES, buffalo HIV life events survey; BHS, beck hopelessness scale; BMICIS, Billings and Moos inventory of coping with illness styles; BMQ, beliefs about medication questionnaire; BST, Burnam interviewer-administered 8-item screening tool; CES-D, center for epidemiologic studies depression scale; CMHS, Cook-Medley hostility scale; DAS, dyadic adjustment scale; DI, dysregulation inventory; DOS, dispositional optimism scale; DSBSI, depression subscale of brief symptom inventory; DSPERI, demoralization scale of psychiatric epidemiology research interview; FRI, family relations index (from family environment scale); FTSSH, Finnish twin study scale of hostility; GHQ, general health questionnaire; HAART, highly active antiretroviral therapy; HADS, hospital anxiety and depression scale; HAT-QOL, HIV/AIDS-targeted quality of life instrument; HBS, health behaviour scale; HCSUS, HIV cost and services utilization study; HIE, Horowitz impact of events scale; HIS, health intention scale; HIV, human immunodeficiency virus; HMO, health maintenance organization; ICS, inhaled corticosteroids; IDD, inventory to diagnose depression; IQR, interquartile range; ISEL, interpersonal support evaluation list; LES, life experience survey; LOT-R, life orientation test; LSS, life stressors scale; LTMSES, long term medication self-efficacy scale; MAH, mental adjustment to HIV; MARS, medication adherence report scale; MAS, Miller attitude scale; MASRI, medication adherence self-report inventory; MEMS, medication even monitoring system; MHLCS, multidimensional health locus of control scale; MOS, medical outcome study health survey; MS, Memphis survey; NEO-FFI, NEO five factor inventory; NR, not reported; NS, non-significant; NSEPQSS, neuroticism scale of the Eysenck personality questionnaire-revised short scale; NT, not tested; OR, odds ratio; Pat SS, patient satisfaction scale; PEI, psychiatric epidemiology interview; POMS, profiles of mood states; PPCS, perceived parenting competence scale; PRQ, personal resource questionnaire; PSS, perceived stress scale; RSEQR, Rosenberg self-esteem questionnaire; SC, symptom checklist; SCL-90, Symptom Checklist-90-R; SEM, standard error of the mean; SF-36, short form-36 health survey; SMS, sense of mastery scale; SOC, sense of coherence; SPS, social provision scale; SSAI, social support appraisals index; SSAS, social support appraisal scale; SSQ, social support questionnaire; TSQ, transplant stress questionnaire; VAS, visual analog scale.

Table S3.

Results of judging methodologic quality

First author	Overall quality	Domain free of bias?
		Study participation	Study attrition	Prognostic factor measurement	Outcome measurement	Confounding measurement and account	Analysis
Bottonari13	Low	No	No	Partly	Partly	No	No
De Geest24	Low	No	Yes	Partly	Yes	No	Partly
Delgado10	Low	Partly	Yes	Partly	Partly	Partly	Partly
Deschamps8	Low	No	Partly	No	Yes	No	No
Dew27	Low	No	Yes	Yes	Partly	Partly	Partly
Dew28	Low	Yes	Partly	Partly	Partly	Partly	Partly
DiMatteo30	Low	Partly	No	Partly	No	No	No
Dobbels29	Low	Yes	Partly	Partly	Partly	Partly	No
Gazmararian3	Low	Yes	Partly	Partly	Yes	Partly	Yes
Godin14	Low	Partly	Yes	No	Partly	Partly	Partly
Grégoire5	Low	Partly	No	No	Partly	No	Partly
Holmes9	Low	Partly	Yes	Partly	Partly	Partly	Yes
Kacanek15	Low	No	Yes	Partly	Partly	No	Yes
Martini16	Low	Partly	No	No	Partly	No	No
Mellins17	Low	Partly	Partly	Partly	No	No	No
Miller6	Low	No	Partly	Partly	No	Partly	Partly
Nabi4	Low	Partly	Partly	Partly	Partly	Partly	Yes
Nilsson Schönnesson18	Low	Partly	Yes	Partly	No	Partly	No
Ponieman1	Low	No	No	Partly	Partly	Partly	Partly
Russell25	Low	No	No	Partly	Partly	No	Partly
Singh11	Low	No	Yes	Partly	Partly	Partly	No
Singh12	Low	Partly	Yes	Partly	Partly	No	No
Stilley23	Low	Yes	Partly	Partly	No	No	No
Thrasher19	Low	Yes	Partly	Partly	Partly	Partly	Partly
Venturini2	Low	Yes	Partly	Partly	Yes	Partly	Partly
Weng26	Low	Partly	No	Yes	Partly	Partly	Yes
Molloy7	Low	No	Yes	Partly	No	Partly	Partly
Horne20	Low	Yes	Yes	Partly	Partly	No	Yes
Mugavero21	Low	Yes	No	Yes	Partly	Partly	Yes
Carrieri22	Low	No	Yes	No	Partly	Partly	Yes

Table S4.

Sensitivity analyses: methodological quality, disease, adherence measures, and statistical analyses

Alteration	Relevant studies	Categories affected	Change in level of evidence
Alterations in methodological quality cut-offs
High-quality study when all six bias domains judged at least as partly (and no no-judgment) instead of ≥four domains judged as yes	19,26,34,35,42,52,64 now high-quality, all other studies low-quality	AI, II, III and CI, II and EII	No association: moderate instead of limited evidence
		CIII and EI	No association: strong instead of limited evidence
Low-quality study when ≥four domains judged as no instead of <four domains judged as yes	19,26,33–36,38,39,42–44,46,47,49,51–60,64,65 now high-quality, all other studies still low-quality	All categories	No association: strong instead of limited evidence
Alterations in disease
Only focus on HIV disease	19,25,42–49,51–55 (studies in analysis)	AI	No association: conflicting instead of limited evidence
		CII and D	Level undetermined (≤one study available)
Only focus on transplant-related diseases	56–60,64,65 (studies in analysis)	AI, II, BII and CI	Level undetermined (≤one study available)
Focus on asthma, diabetes, heart disease/hypertension	26,33–36,38,39,66 (studies in analysis)	AII, III, BI, II, CI, II, D, EII	Level undetermined (≤one study available)
Alterations in adherence measures
Focus on objective adherence measures (MEMS, record review)	19,25,26,34,35,42–44,56–59 (studies in analysis)	D	No association: conflicting instead of limited evidence
		AII and CI	Level undetermined (≤one study available)
Focus on subjective adherence measures (self-report)	33,36,38,39,45–49,51–55,60,64–66 (studies in analysis)	AI	No association: conflicting instead of limited evidence
		BI	Level of evidence undetermined (≤one study available)
		BII and EI	No association: conflicting instead of limited evidence
Alterations in statistical analysis
Only focus on univariate analysis instead of multivariate analysis	34,38,46,51,57,65,66 (studies omitted due to lack of univariate analysis)	AI, AIII, CI, CIII, EI and EII	No association: conflicting instead of limited evidence

Notes: A = Beliefs and cognitions about: I) medication and treatment; II) illness; III) self-efficacy and locus of control. B = coping styles: I) task oriented, II) emotion oriented. C = Social influences and social support: I) regarding medical caregiver; II) regarding friends and family; III) in general. D = personality traits. E = psychological well-being: I) mood state; II) perceived stress/stressors.

Abbreviations: HIV, human immunodeficiency virus; MEMS, medication event monitoring system.