Positive affect and medication adherence in chronic conditions: A systematic review

Abstract

Objective:

This review aims to inform research and clinical care on the current state of knowledge on the relationship between positive affect and medication adherence.

Methods:

Searches were carried out in PsycINFO, PubMed MEDLINE, Cochrane Central Register of Controlled Trials (CENTRAL), CINAHL, and Embase. There were no limits on study type, publication date, language, or participant demographics. Studies reporting a relationship between positive affect and medication adherence were eligible for inclusion if positive affect was measured prior to or concurrently with medication adherence.

Results:

Nine studies met inclusion criteria. All studies were prospective cohort or cross-sectional, and examined positive affect and medication adherence in people living with HIV or cardiovascular conditions. The majority of results indicated positive associations between positive affect and medication adherence, with Cohen’s d effect sizes ranging from −0.40 to 1.27.

Conclusions:

Consistent with previous theoretical work, this systematic review provides evidence of a link between positive affect and improved medication adherence. Better measurement of both affect and medication adherence across chronic conditions is an important focus for future research and will inform targeted interventions to improve adherence and, ultimately, decrease the morbidity, mortality, and cost associated with suboptimal adherence in chronic physical conditions.

Up to 80% of deaths in the United States can be attributed to chronic diseases, leading the World Health Organization to emphasize prevention of chronic diseases and to increase affordability and accessibility of medications that help extend the lives of people with chronic conditions (World Health Organization, 2017). Approximately half of patients with chronic illness do not take their medication as prescribed (Brown & Bussell, 2011; Osterberg & Blaschke, 2005). Across a range of approaches to medication adherence (MA) measurement, research has shown that poor MA can lead to increased morbidity and mortality, and has been shown to significantly increase the cost of healthcare in the United States (Brown & Bussell, 2011; Osterberg & Blaschke, 2005). For example, a lack of adherence to blood pressure medication, as measured by prescription claims records, is the single most important cause of inadequate blood pressure control (Sokol, McGuigan, Verbrugge, & Epstein, 2005). This, in turn, significantly increases the risk of myocardial infarction, stroke, and hospitalization (Brown & Bussell, 2011). Indeed, between 33 and 69% of hospital visits in the United States for medication-related issues (e.g., adverse drug events, drug labeling issues, a lack of adherence) are due to non-adherence to medication regimens, resulting in an increased cost of $100 billion per year (Brown & Bussell, 2011; Osterberg & Blaschke, 2005). In contrast, research has indicated that better MA lowers disease-related medical costs as well as total healthcare costs for the chronic conditions of diabetes and hypercholesterolemia, even when factoring in the cost of medications for these conditions (Sokol et al., 2005). Similarly, patients with diabetes, hypertension, hypercholesterolemia, and congestive heart failure who maintained at least 80% MA were significantly less likely to be hospitalized when compared to their less adherent counterparts (Sokol et al., 2005), thereby reducing the likelihood of the physical and financial health ramifications associated with poor adherence.

Predictors of Adherence

Determinants of MA can be classified into five factors (Sabaté, 2003): social and economic factors (e.g., income, level of education), healthcare team- and system-related factors (e.g., reimbursement by health insurance plans), condition-related factors (e.g., severity of symptoms), therapy-related factors (e.g., treatment duration), and patient-related factors (e.g., forgetfulness). There have been two overviews that have examined how these broad factors influence MA (Kardas, Lewek, & Matyjaszczyk, 2013; Mathes, Jaschinski, & Pieper, 2014). Across both reviews, authors found that poverty, high cost of medications, and minority status were associated with decreased MA, whereas older age, employment, and higher education were positively associated with MA (Kardas et al., 2013; Mathes et al., 2014). In the review by Kardas et al. (2013), the authors noted that many potential determinants of MA are not modifiable. As such, it is imperative to understand potential determinants of MA that are modifiable, such as positive affect (PA) (Moskowitz, 2010), so that they can be targeted for intervention.

Affect and Health

The majority of work on affect and health behaviors has focused on negative affective (NA) states (Kardas et al., 2013; Mathes et al., 2014). Meta-analyses examining the relationship between MA in chronic conditions and NA demonstrate that increased depression is linked to significantly worse MA in chronic conditions such as HIV, diabetes, hypertension, and asthma, (Grenard et al., 2011; Sin & DiMatteo, 2014). Specifically, results of one meta-analysis indicated that over a number of chronic conditions, depression was associated with a 1.76 higher odds of medication non-adherence compared to those without depression (Grenard et al., 2011), and another meta-analysis indicated that when people living with HIV are treated for depression, their odds of adhering to medication were 1.83 times higher than those who do not receive treatment for depression (Sin & DiMatteo, 2014).

However, NA is only one piece of the puzzle. Positive affect (PA), defined as feelings reflecting pleasurable engagement with the environment, includes emotional states such as interest, joy, and enthusiasm (Clark, Watson, & Leeka, 1989). PA and NA are not always two ends of an emotional continuum (Clark et al., 1989) and can co-occur, even during times of severe life stress (Folkman & Moskowitz, 2000). Studies have shown unique positive associations between PA and health-related outcomes, including viral load suppression in those living with HIV (Wilson et al., 2017), as well as increased functional ability during recovery from stroke, heart attack, or hip fracture (Ostir et al., 2002). Other research indicates that PA increases physical activity (Emerson, Dunsiger, & Williams, 2018; Ryff, Singer, & Love, 2004; Watson, 1988a), greater fruit and vegetable intake in young adults (Conner, Brookie, Carr, Mainvil, & Vissers, 2017), and healthier eating behaviors in those living with diabetes (Patel, 2013).

Thus, given a significant body of work showing that NA is related to MA, and growing evidence that PA is associated with a host of other health behaviors, we conducted a systematic review to synthesize information on the association between PA and MA, paying special attention to chronic medical conditions due to the challenge of suboptimal adherence in patients suffering from these ailments. The objective of this systematic review was to characterize the relationship between PA and MA in adults with chronic medical conditions.

Methods

Eligibility Criteria

Original studies published in peer-reviewed journals that quantitatively examined the relationship between PA and MA were eligible for inclusion if MA was measured either concurrently with, or subsequent to, PA and participants had chronic medical conditions. We excluded grey literature such as conference proceedings, meeting abstracts, and dissertations, due to the lack of peer-review. There were no restrictions on language, study design, year of publication, or participant demographics including age, gender, or race/ethnicity.

Chronic Conditions

The chronic conditions eligible for inclusion were selected based upon Goodman and colleagues’ 2013 synthesis of chronic conditions (Goodman, Posner, Huang, Parekh, & Koh, 2013) and included asthma, hypertension/high blood pressure, congestive heart failure, coronary artery disease/coronary heart disease/ischemic heart disease, cardiac arrhythmias, hyperlipidemia, cerebrovascular disease/stroke/transient ischemic attack, arthritis, cancer, chronic kidney disease, chronic obstructive pulmonary disease, diabetes, hepatitis, HIV, and osteoporosis. Psychological conditions were excluded due to potential confounding with PA.

PA/MA Terms

PA terms eligible for inclusion in the database search were derived from Pressman and Cohen’s (2005) seminal paper exploring the effect of PA on health. Additional terms (e.g., “positive states of mind”) were included after a preliminary search. Terms highly correlated with physiological states (e.g., vigor) were excluded due to potential confounding with physiological arousal and the side effects of some medications. PA terms used for this review were: positive affect, positive emotion, happy, cheerful, joy, excited, elated, enthusiastic, content, amused, humor, calm, relaxed, happiness, and positive states of mind.

Medication use was operationalized as: medication prescribed by a doctor for the chronic condition examined in the study, and self-administered by the patient on a daily basis. MA search terms included: medication adherence, adherence to medication, medication compliance, adherence, compliance, treatments, therapies, non-adherence, and non-compliance. Full lists of search strategies and terms are included in the supplementary materials.

Information sources

[Authors 1 and 3] developed the search strategies based on the specific inclusion criteria. Searches were carried out by [Author 3] in September 2018 in PubMed MEDLINE, PsycINFO (Ebsco), Cochrane Central Register of Controlled Trials (CENTRAL), CINAHL (Ebsco), and Embase (embase.com). Searches of the other databases were adapted from the PubMed strategy

Data collection process

[Authors 1 and 3] developed list of search terms (Appendix A; Online Supplement), and [Author 3] carried out the searches and eliminated duplicate results. [Authors 1, 2, and 4] independently screened titles, abstracts, and full-text of inclusion, and then [Authors 1 & 2] independently extracted data from the nine included studies. The authors used Covidence (Veritas Health Innovation, 2018) to manage the process of screening studies and extracting information. Discrepancies were resolved through discussion between [Authors 1 and 2]. See the protocol we adhered to, accessible in Appendix B (Online Supplement), for additional information.

Methodological quality of studies

We evaluated quality of the studies using the National Heart, Lung, and Blood Institute’s (NHLBI) Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies (2014) which is appropriate for cross-sectional and cohort studies (National Heart Lung and Blood Institute, 2014). When utilizing this checklist, reviewers must decide whether studies meet a criterion in question such as risk of bias (including selection bias, information bias, measurement bias and confounding bias), whether the answer to the question cannot be determined or is not reported, or whether a question is not applicable to the study (e.g., due to study design; (National Heart Lung and Blood Institute, 2014).

Synthesis of Findings

We used a narrative synthesis approach for the current study rather than a meta-analysis due to the heterogeneity of both measures and populations in relatively low number of studies eligible for inclusion in the review. The said heterogeneity between measures and populations between the included studies would have rendered a quantitative synthesis uninterpretable.

Results

Study selection

See Figure 1 for a PRISMA diagram (Moher, Liberati, Tetzlaff, Altman, & Group, 2009). Seven hundred ninety seven articles were retrieved from relevant databases, and 275 were removed due to duplication. Nine studies were deemed appropriate for data extraction. Extracted information appears in Table 1.

Figure 1.

2009 PRISMA Flow Diagram

Table 1.

Study type and characteristics

Study	Study Type and Characteristics	Inclusion/Exclusion Criteria	Participant Characteristics	Self-report PA/MA Measures	Outcomes/Covariates
Bardeguez et al. (2008)	Prospective cohort study: Antepartum follow-up = variable Postpartum follow-up = 48 weeks ARV MA in HIV-positive pregnant women was observed during (follow-up = variable) and after pregnancy (postpartum follow-up = 48 weeks)	Inclusion: Enrolled during first pregnancy, 1+ completed MA form in study database, on tablet-based ARV regimen Exclusion: NR	Chronic condition: HIV N: 519 Age: ≥13; ≤21 = 13.5%, >21–30 = 49.7%, >30 = 36.8% Race: White = 14.1%, Black non-Hispanic = 55.5%, Hispanic = 30.4% Sex: Female = 100% SES: Education = ≤11th grade = 42.1%, HS = 40.2%, >HS = 17.7%	PA: “How much of the time in the past month have you been a happy person?” responses dichotomized - “all/most of the time,” “never/some of the time.” MA: Current ARV regimen, number of doses missed for each ARV in four days prior to study visit, last time ARV was missed. Adherence dichotomized - perfect/imperfect	Outcome: Participants who felt happy all/most of the time were more likely to report perfect MA postpartum (OR = 1.69; 95% CI = 1.14 – 2.51, d = 0.42). Covariates: Use of ARVs prior to pregnancy, CDC disease category, ever missing prenatal vitamins, use of marijuana ante/postpartum Outcome: Participants who felt happy all/most of the time were more likely to report perfect MA antepartum (OR = 1.78, 95% CI = 1.24 – 2.54. d = 0.46). Covariates: Use of ARVs prior to pregnancy, CDC disease category, ever missing prenatal vitamins, use of marijuana ante/postpartum Outcome: Participants who felt happy all/most of the time postpartum were more likely to have had a longer time since last missed medication dose than those who felt happy never/some of the time (OR = 1.56, 95% CI = 1.04 – 2.34, d = 0.35) Covariates: Use of ARVs prior to pregnancy, ever missing prenatal vitamins, use of marijuana ante/postpartum, use of alcohol ante/postpartum Outcome: Participants who felt happy all/most of the time postpartum were more likely to have had a longer time since last missed medication dose antepartum (OR - 1.50, 95% CI = 1.06 – 2.12, d = 0.32). Covariates: Use of ARVs prior to pregnancy, ever missing prenatal vitamins, use of marijuana ante/postpartum, use of alcohol ante/postpartum
Carrico et al. (2010)	Cross-sectional study: HIV-positive MSM/transgender individuals who reported past-month methamphetamine use self-reported HIV-specific traumatic stress, drug use, PA, and ART adherence.	Inclusion: ≥18 years of age, English-speaking, MSM or transgendered, evidence of current ART prescription, taking ART for ≥1 month, report using methamphetamine during past 30 days Exclusion: NR	Chronic condition: HIV N: 122 Age: M = 44, SD = 6 Race: African American = 32%, Hispanic = 14%, Asian/Pacific Islander = 2%, Native American/Alaskan Native = 2%, Caucasian = 41%, Multi-cultural = 9% Sex: Male = 94%, MTF = 4%, FTM = 2% SES: Education = <HS = 14%, HS = 26%, Some college = 34%, ≥College = 26%. Income mode = $5,000 - $11,999 (44%)	PA: PANAS MA: ART adherence (visual analogue scale): Compared 100% past-month adherence to <100% past-month adherence	Outcome: Participants with higher PA were more likely to report perfect ART adherence (AOR = 1.79, 95% CI = 1.06 – 2.12, d = 0.46). Covariates: Age, ethnicity, time since HIV diagnosis, time since starting ART, NA, HIV-specific traumatic stress
Carrico and Moskowitz (2014)	Prospective cohort study: Follow-up = 18 months Observational study of people who had recently been diagnosed with HIV. PA, ART persistence, and viral load were measured over time. (follow-up = 18 months)	Inclusion: Informed of HIV-positive status within past 8 weeks, English or Spanish-speaking, ≥18 years of age; provide informed consent Exclusion: NR	Chronic condition: HIV N: 153 Age: M = 38, SD = 9 Race: Caucasian = 47%, African American = 27%, Hispanic/Latino = 12%, Asian/Pacific Islander = 5% Sex: Male = 89%; 76% of whom were gay or bisexual SES: 52% = not college graduates, 57% = <$30,000/year	PA: Modified Differential Emotions Scale - 9 PA items. Participants rated frequency of certain emotions over the past week on a 0–4 (never-most of the time) scale. MA: Participants reporting starting and remaining on ART during all assessments over 18-month study period classified ART persistent and compared with those who did not start ART/used ART inconsistently	Outcome: Participants with PA one SD above mean at baseline were more likely to report medication persistence over 18-month follow-up (OR = 1.08, 95% CI = 1.01 – 1.16, d = 0.04). Covariates: Education, CD4+ count, HIV viral load, NA
Cuffee et al. (2012)	Cross-sectional study: Hypertensive patients in safety net hospital selfreported PA and MA.	Inclusion: African American, receive primary care from specific hospital, hypertensive, ≥19 years old, provide informed consent, not pregnant Exclusion: Missing MA data, missing happiness data	Chronic condition: Hypertension N: 536 Age: M = 53.6, SD = 9.7 Race: African American = 100% Sex: Female = 71.6% SES: Income: Low-income; Household income of >$16,000 = 16.7% Education: >HS education = 68.2%	PA: Subjective Happiness Scale. Low, medium & high happiness defined by tertiles. MA: 4-item Morisky Medication Adherence Scale: combined into tertiles for analyses 0–1, 2–3, 4.	Outcome: Participants reporting high happiness were more likely to be in a better MA category compared to those with low happiness (OR = 2.26, 95% CI = 1.52 – 3.37, d = 0.65), as were participants reporting medium compared to low happiness (OR = 1.53, 95% CI = 1.02 – 2.27, d = 0.34). Covariates: Income, education, difficulty paying for medical care
Fipro-Perretti et al. (2018)	Cross-sectional study: HIV+ women completed oral autobiographical narratives about 3 selfdefining memories they considered turning points in their lives at time and self-reported MA. M number of words per narrative = 2072.10, SD = 1134.72. Prospective cohort study: Follow-up = 3–9 months (M = 5.72, SD = 1.51) HIV+ women completed oral autobiographical narratives about 3 selfdefining memories they considered turning points in their lives at time 1. MA data was collected at time 2.	Inclusion: Participation in the Women’s Interagency HIV Study, HIV+ Exclusion: Not English-speaking, not taking part in bi-annual inperson clinic visits	Chronic condition: HIV N: 98 at time 1; N: 89 at time 2 Age: M = 45.32, SD = 8.87 Race: African American = 90.8%, Non-Hispanic White = 4.1%, Hispanic = 4.1% Other = 1% Sex: Female = 100% SES: Education: range = Less than 7th grade - attended/completed graduate school. Grades 7–11 = 43.9, HS graduate = 30.6%, Some college = 19.4%. Income: <=$12,000/year = 68.3%	PA: Frequency of positive affect words (e.g., love, nice, sweet) during oral autobiographical narratives, as measured by LIWC 2007’s internal positive emotion word dictionary. MA: Percentage of ART pills taken in past 6 month using one of five category responses: 1 = 100% of the time, 2 = 95–99% of the time, 3 = 75–94% of the time, 4 = 75% of the time, 5 = I haven’t taken any of my prescribed medications.	Outcome: Higher frequency of PA word use was not associated with greater odds of better MA cross-sectionally, OR = 1.48, 95% CI = 0.78 – 2.78, d = 0.22 or prospectively, OR = 0.83, 95% CI = 0.42, 1.63, d = −0.10. Covariates: Age, WIHS enrollment cohort, past tense, present tense, future tense word use, negative affect word use, MA at time 1 (prospective analysis). Outcome: Higher frequency of PA word use was not associated with greater odds of better MA cross-sectionally, OR =1.55, 95% CI = 0.72, 3.33, d = 0.24 or prospectively, OR = 0.89, 95% CI = 0.39, 2.03, d = −0.06. Outcome: Higher frequency of PA word * past tense word use was not associated with greater odds of better MA cross-sectionally, OR =0.60, 95% CI = 0.21, 1.72, d = −0.28 or prospectively, OR 1.51, 95% CI = 0.64, 3.56, d = 0.23. Outcome: Higher frequency of PA word * present tense word use was not associated with greater odds of better MA cross-sectionally, OR = 1.24, 95% CI = 0.66, 2.31, d = 0.12 or prospectively, OR = 1.50, 95% CI = 0.51, 4.47, d = 0.22. Outcome: Higher frequency of PA word * future tense word use not associated with greater odds of better MA cross-sectionally, OR = 0.48, 95% CI = 0.20, 1.14, d = −0.40 or prospectively, OR = 1.41, 95% CI = 0.70, 2.84, d = 0.19. Covariates: Age, WIHS enrollment cohort, past, present, & future tense words, negative affect words, tense * positive and negative affect word interactions, MA at time 1 (prospective analysis)
Reis et al. (2013)	Cross-sectional study: HIV+ outpatients at two Portuguese hospitals self-reported adherence, positive and NA, satisfaction with life. HIV information obtained via patient records. Disease stages included = asymptomatic, symptomatic, & AIDS	Inclusion: ≥18 years of age, informed consent, diagnosis of HIV, ART for ≥3 months Exclusion: NR	Chronic condition: HIV N: 197 Age: M = 40.3, SD = 9.1 Race: (Not reported) Sex: Male = 69.5% SES: Education: range = <4 years - university. 53.9% = only elementary school, 44.2% unemployed	PA: PANAS (Portuguese) MA: Cuestionario para la Evaluation de la Adhesion al Tratamiento Antiretroviral (Portuguese)	Outcome: PA was positively correlated with MA (r = 0.535, p < 0.01, d = 1.27). Covariates: NR
Sin et al. (2015)	Cross-sectional study: Outpatients with CHD reported PA and health behaviors including physical activity, sleep quality, MA, cigarette smoking, and alcohol use. Prospective cohort study: Follow-up = 5 years Participants reported PA and health behaviors including physical activity, sleep quality, MA, cigarette smoking, and alcohol use 5 years postbaseline.	Inclusion: At least one of: history of myocardial infarction/coronar y revascularization, ≥50% stenosis in 1+ coronary vessels, exercise-induced ischemia Exclusion: NR	Chronic condition: Coronary heart disease N: Cross sectional = 1022, Cohort = 662 Age: M = 66.8, SD = 10.9 Race: White = 60.14% Sex: Male = 81.68% SES: Education = HS grad = 87.41% Income <$20,000/year = 48.93%	PA: PANAS MA: “In the past month, how often did you take your medications as the doctor prescribed?” Adherent = took medications 90–100% of the time, Non-adherent = took medications ≤75% of time. MA also evaluated as continuous variable ranging from 0 to 4; higher scores indicating more adherence.	Outcome: At baseline, higher PA was associated with greater odds of better MA in unadjusted, OR = 1.61, 95% CI = 1.28 – 2.02, d = 0.26, and fully adjusted models, OR = 1.46, 95% CI = 1.12 – 1.90, d = 0.30. Covariates: Age, education, income, BMI, HDL cholesterol, revascularization history, baseline depressive symptoms. Outcome: Five-year increases in PA were correlated with 5-year increases in MA in unadjusted, B = 0.015, SE = 0.004, p < 0.001, d = 0.03, and fully adjusted models, B = 0.014, SE = 0.004, p < 0.001, d = 0.03. Covariates: Age, education, income, BMI, HDL cholesterol, history of revascularization, five year change in depressive symptoms. Outcome: Five-year increases in MA were predicated by baseline PA in unadjusted analyses, p = 0.003. Baseline PA did not predict five year MA when controlling for baseline health behaviors, B = 0.005, SE = 0.003, p = 0.16, d = 0.01. Covariates: Baseline health behaviors
Williams et al. (2000)	Cross-sectional study: Participants interviewed and asked about demographics, drug use, sexual behaviors, health status, HIV/drug treatment history, health beliefs (including PA).	Inclusion: African American, used illicit substances within past year or currently injecting drugs/smoking crack cocaine, HIV-positive and under medical care, informed consent Exclusion: NR	Chronic condition: HIV N: 47 Age: ≥18; 18–29 = 6%, 30–39 = 43%, ≥40 = 51% Race: African American = 100% Sex: Male=68% SES: Education = <HS = 50%, ≥HS = 50%. Currently homeless = 55%	PA: Participants were asked how much they were currently experiencing different emotions, including joy. Intensity measured using a 10-point scale (no feeling at all - very intense feeling). MA: Perception of compliance measured by asking how often participants believed that they missed taking a specified medication using a 5-point, Likert scale (never - always).	Outcome: Joy was positively associated with perceptions of compliance, r = 0.311, p = 0.035, d = 0.65. Covariates: NR
Wilson et al. (2017)	Prospective cohort study: Follow-up = 18 months Three waves of six-month data of women on ART analyzed to assess PA in relation to viral load over time. MA assessed during waves one/two.	Inclusion: Female Western-blot confirmed HIV, ≥13 years of age, ability to answer questions in English/Spanish Exclusion: NR	, Chronic condition: HIV N: 995 Age: M = 48, SD = 8. Race: Black = 58%, Hispanic = 27%, White = 10.94%, Other = 3.75% Sex: Female = 100% SES: NR	PA: Four-item PA subscale of CES-D. MA: Assessed every 6 months Participants selfreported % of ART regimen taken over previous 6 months. Responses dichotomized as adherence <95%; adherence ≥95%	Outcome: Significant differences in MA between those with low PA (% adherent = 80.9) and high PA (% adherence = 87.8) over time, x2(1) = 7.9, p = 0.005, d = 0.09. Covariates: Age, race, recruitment site, enrollment cohort, born in U.S., interviewed in Spanish, presence of partner/spouse, viral suppression, substance use, heavy drinking, interpersonal problems, NA, somatic burden.

Note. AOR = adjusted odds ratio; ART = antiretroviral therapy; ARV = antiretroviral; FTM = female to male transgender; HS = high school; LIWC = Linguistic Inquiry and Word Count; M = mean; MSM = men who have sex with men; MTF = male to female transgender; NR = not reported; OR = odds ratio; SES = socioeconomic status; SD = standard deviation

Study and participant characteristics

As seen in Table 1, sample size for included studies ranged from 22 – 995, and mean participant ages, when reported, ranged from 38 – 66.8 years of age. Three studies included only female participants (Bardeguez et al., 2008; Firpo-Perretti, Cohen, Weber, & Brody, 2018; Wilson et al., 2017), one study included only male or transgendered individuals (Carrico, Johnson, Colfax, & Moskowitz, 2010), and all other studies included both male and female participants. All studies except one were conducted in the United States. Two studies had only African American participants (Cuffee et al., 2012; Williams, Bowen, Ross, Freeman, & Elwood, 2000), one did not provide information on participant race or ethnicity (Reis et al., 2013), and all other studies included participants of multiple races. Seven studies examined the relationship between PA and MA in patients with HIV (Bardeguez et al., 2008; Carrico et al., 2010; Carrico & Moskowitz, 2014; Firpo-Perretti et al., 2018; Reis, Guerra, & Lencastre, 2013; Williams et al., 2000; Wilson et al., 2017), and two examined this relationship in patients with cardiovascular conditions (Cuffee et al., 2012; Sin, Moskowitz, & Whooley, 2015).

Study design

Three studies were prospective cohort studies (Bardeguez et al., 2008; Carrico & Moskowitz, 2014; Wilson et al., 2017), four were cross-sectional (Carrico et al., 2010; Cuffee et al., 2012; Reis et al., 2013; Williams et al., 2000), and two employed both prospective and cross-sectional analyses in longitudinal cohorts (Firpo-Perretti et al., 2018; Sin et al., 2015). Cohort studies differed in length of follow-up ranging from 18 months to 5 years. Bardeguez and colleagues (2008) measured antiretroviral (ARV) adherence both antepartum and postpartum, with a follow-up of 48 weeks plus an antepartum assessment period. Two studies had follow-up periods of 18 months (Carrico & Moskowitz, 2014; Wilson et al., 2017). The length of follow-up in the study by Sin and colleagues (2015) was five years, and the length of the study by Firpo-Perretti et al. (2018) ranged from 3–9 months.

PA/MA Terms

While all studies except Firpo-Perretti et al. (2018) utilized self-report PA, studies employing self-report differed in choice of measurement instrument. Specifically, three studies employed the Positive and Negative Affect Schedule (PANAS; (Watson, Clark, & Tellegen, 1988), which measures PA and NA over the past week (Carrico et al., 2010; Reis et al., 2013; Sin et al., 2015). Carrico and Moskowitz (2014) used a modified version of the Differential Emotions Scale (Fredrickson, 2013) which includes a broader set of emotion terms. Cuffee and colleagues (2012) employed the Subjective Happiness Scale (Lyubomirsky & Lepper, 1999), and Wilson and colleagues (2017) used the four-item PA subscale of the Centers for Epidemiologic Studies – Depression (CES-D) scale (Radloff, 1977). One study used a dichotomous scale of all/most of the time versus never/some of the time to answer the question “How much of the time in the past month have you been a happy person?” (Bardeguez et al., 2008), and one asked participants how intensely they were currently experiencing various emotions, including joy on a 1 −10 (not at all - very intense) scale (Williams et al., 2000). The study by Firpo-Perretti et al. (2018) measured the percentage of positive emotion words used in oral autobiographical narratives that women living with HIV told to supportive listeners (e.g., nurses). Participants told these listeners about three events that were turning points in their lives, and were asked to describe the events including when they happened, who was involved, and how participants felt. Researchers subsequently transcribed the interviews, and then assessed positive emotion words using the Linguistic Inquiry and Word Count (LIWC) 2007 internal positive emotion dictionary (Firpo-Perretti et al., 2018; Pennebaker, Booth, & Francis, 2007).

Assessment of MA

All studies utilized self-report to assess MA. Self-report differed by measure type (e.g. visual analogue scale, questionnaire) and timespan. Cuffee (2012) modified the Morisky Medication Adherence scoring such that potential answers were put into tertiles of 0–1, 2–3, and 4 (low – high). Reis and colleagues (2013) used a Portuguese version of the Cuestionario para la Evaluación de la Adhesión al Tratamiento Antiretroviral, which measures adherence continuously (Remor, 2013). Bardeguez and colleagues (2008) measured adherence by asking participants to report the numbers of doses missed for each ARV in the four days preceding the study visit, and the most recent time an ARV dose was missed. Adherence was dichotomized as perfect/imperfect (Bardeguez et al., 2008). Carrico and colleagues (2010) asked participants to self-report antiretroviral therapy (ART) adherence using a visual analogue scale, and compared participants who had 100% adherence to those with less than 100% adherence over the past month. Carrico and Moskowitz (2014) compared participants who initiated and remained persistently on ART at follow-up assessments over 18 months to those who did not initiate or remain persistently on ART, a notable departure from asking about percent adherence. Sin and colleagues (2015) asked participants, “In the past month, how often did you take your medication as the doctor prescribed?” Participants were categorized as adherent if they took their medication as prescribed 90–100% of the time, and non-adherent if they took their medications as prescribed 75% or less of the time. Information regarding those who took their medication 75–90% of the time was not included in this manuscript. MA was also examined on a continuous scale ranging from 0 – 4 (least adherent - most adherent) (Sin et al., 2015). Williams and colleagues (2000) asked participants how often they had missed taking a specific medication using a Likert-type scale, with answers ranging from never to always, and Wilson and colleagues (2017) and Firpo-Perretti and colleagues (2018) asked participants to report the percentage of their medication regimen taken over the six months preceding the study.

Measurement of covariates

As seen in Table 1, while the majority of studies employed demographic and disease-related covariates, the use of covariates differed by study. Six studies controlled for demographic variables (e.g., age, education, income, race), but no two studies employed similar demographic covariates (Carrico et al., 2010; Carrico & Moskowitz, 2014; Cuffee et al., 2012; Firpo-Perretti et al., 2018; Sin et al., 2015; Wilson et al., 2017). The same six studies controlled for health variables (e.g., body mass index, time since HIV diagnosis, viral suppression, history of revascularization) using differing covariate combinations (Carrico et al., 2010; Carrico & Moskowitz, 2014; Cuffee et al., 2012; Firpo-Perretti et al., 2018; Sin et al., 2015; Wilson et al., 2017), and two controlled for health behaviors (e.g., substance use, heavy drinking, missing prenatal vitamins) (Bardeguez et al., 2008; Wilson et al., 2017). Half controlled for depression or negative affect (Carrico et al., 2010; Carrico & Moskowitz, 2014; Cuffee et al., 2012; Sin et al., 2015; Wilson et al., 2017). Two studies included only bivariate analyses of the relationship between PA and MA (Reis et al., 2013; Williams et al., 2000), and one study (Sin et al., 2015) included both bivariate and multivariate.

Study Quality

See Table 2 for an assessment of individual study quality by NHLBI-specified domains (National Heart Lung and Blood Institute, 2014). Study quality was mediocre overall. Importantly, the NHLBI notes that its questions regarding quality are not intended to provide a definitive cut-off for high or low quality, but rather to help reviewers critically appraise study quality and focus on important concepts that might induce study bias (National Heart Lung and Blood Institute, 2014). All studies included in the current review had some potential bias, driven by factors including 1) no sample size justification, power description, or variance and effect estimates, 2) correlational studies lacking information to support a direction of effect. However, the majority of prospective cohort studies assessed PA more than once over time, which is beneficial for study quality (National Heart Lung and Blood Institute, 2014).

Table 2.

Study Quality

	Was the research question or objective in this paper clearly stated?	Was the study population clearly specified and defined?	Was the participation rate of eligible persons at least 50%?	Were all the subjects selected or recruited from the same or similar populations (including the same time period)? Were inclusion and exclusion criteria for being in the study prespecified and applied uniformly to all participants?	Was a sample size justification, power description, or variance and effect estimates provided?	For the analyses in this paper, were the exposure(s) of interest measured prior to the outcome(s) being measured?	Was the timeframe sufficient so that one could reasonably expect to see an association between exposure and outcome if it existed?	For exposures that can vary in amount or level, did the study examine different levels of the exposure as related to the outcome (e.g., categories of exposure, or exposure measured as continuous variable)?	Were the exposure measures (independent variables) clearly defined, valid, reliable, and implemented consistently across all study participants?	Was the exposure(s) assessed more than once over time?	Were the outcome measures (dependent variables) clearly defined, valid, reliable, and implemented consistently across all study participants?	Were the outcome assessors blinded to the exposure status of participants?	Was loss to follow-up after baseline 20% or less?	Were key potential confounding variables measured and adjusted statistically for their impact on the relationship between exposure(s) and outcome(s)?
Bardeguez 2008	Yes	Yes	NR	No	No	Yes	Yes	Yes	No	Yes	No	CD	No	Yes
Carrico 2010	Yes	Yes	Yes	No	No	NA	NA	Yes	Yes	NA	CD	CD	NA	Yes
Carrico 2014	Yes	Yes	Yes	Yes	No	Yes	Yes	Yes	Yes	No	No	CD	No	Yes
Cuffee 2012	Yes	Yes	Yes	Yes	No	NA	NA	Yes	Yes	NA	Yes	CD	NA	Yes
Fipro-Perritti 2018 - XS	Yes	Yes	NA	Yes	No	NA	NA	Yes	Yes	NA	Yes	NR	NA	Yes
Fipro-Perritti 2018 - CH	Yes	Yes	NA	Yes	No	Yes	Yes	Yes	Yes	No	Yes	NR	Yes	Yes
Reis 2013	Yes	Yes	NR	CD	No	NA	NA	Yes	CD	NA	Yes	CD	NA	No
Sin 2015 - XS	Yes	Yes	NR	No	No	NA	NA	Yes	Yes	NA	CD	CD	NA	Yes
Sin 2015 - CH	Yes	Yes	NR	No	No	Yes	Yes	Yes	Yes	Yes	CD	CD	No	Yes
Williams 2000	Yes	Yes	NR	CD	No	NA	NA	Yes	CD	NA	CD	CD	NA	No
Wilson 2017	Yes	Yes	Yes	Yes	No	Yes	Yes	Yes	Yes	Yes	No	CD	No	Yes

Note. CD = cannot determine; CH = cohort; NA = not applicable; NR = Not reported; XS = cross-sectional

Relationship between PA and MA

Seven of the nine studies demonstrated a positive association between PA and better MA. This positive relationship is seen in HIV in both cross-sectional (Carrico et al., 2010; Reis et al., 2013; Williams et al., 2000) and prospective cohort studies (Bardeguez et al., 2008; Carrico & Moskowitz, 2014; Wilson et al., 2017) in which PA was self-reported. However, when PA was measured as percentage of PA word use in oral autobiographical narratives, no association between PA and MA was seen (Firpo-Perretti et al., 2018). In cardiovascular conditions, positive associations between PA and MA were found over both cross-sectional (Cuffee et al., 2012; Sin et al., 2015) and prospective cohort study designs (Sin et al., 2015), with the exception of one null result indicating that PA was not predictive of MA five years later when controlling for baseline MA in coronary heart disease (Sin et al., 2015). When standardized effect sizes were not given, most studies gave sufficient information with which to calculate Cohen’s d. Standardized effect sizes (range of Cohen’s d = −0.40 – 1.27) are included in Table 1.

Studies that include self-report NA and other covariates also show an association over and above the effects of self-reported NA in both HIV (Carrico & Moskowitz, 2014; Wilson et al., 2017) and cardiovascular conditions (Sin et al., 2015). However, the study that employed frequency of PA word use in narratives also controlled for frequency of NA word use using LIWC 2007 internal dictionaries (Pennebaker et al., 2007), and found no association between PA and MA in these, or other, analyses (Firpo-Perretti et al., 2018).

The studies that reported uncontrolled associations using self-report measures for both PA and MA (Reis et al., 2013; Sin et al., 2015; Williams et al., 2000) indicated positive correlations between these two variables. Specifically, Sin et al. (2015) found that baseline PA and baseline MA, as well as five-year increases in PA and MA, were correlated in both unadjusted and adjusted models, while baseline PA only predicted five-year increases in MA before adjusting for baseline health behaviors, including MA in patients with cardiovascular conditions. Reis et al., 2013 found that PA was positively correlated with MA cross-sectionally in a mostly male sample of people living with HIV in Portugal, and Williams et al. (2000) found that PA was positively associated with perceptions of MA cross-sectionally in mostly male African American recent or current drug users.

Discussion

While the literature on the relationship between PA and MA in chronic conditions is nascent, the current review provides some indication that PA is positively associated with MA. Despite a number of measurement limitations including varying time periods and covariates, a positive relationship between PA on MA was found in the majority of studies. The fact that the relationship between PA and MA can be seen over differing study lengths and an array of self-report scales indicates that this relationship is worth studying further. This review thus serves as a foundation for further research into the unique beneficial role that PA might play in supporting health.

There are multiple potential pathways linking PA and MA. One potential pathway is through PA acting as a buffer to the stress of chronic illness or life stress more generally. MA is significantly influenced by psychological stress (Morisky, Ang, Krousel‐Wood, & Ward, 2008) and the occurrence of PA may help individuals to cope better with stress, thus reducing the impact of stress on adherence. According to the seminal hypothesis by Lazarus, Kanner, and Folkman (1980) when negative affect is dominant in the midst of stressful situations as is common in people coping with chronic illness, PA can act as a psychological respite, replenish resources diminished by stress, and sustain coping efforts. Coupled with evidence that positive affect increases motivation (Aarts, Custers, & Marien, 2008), it is plausible that when people feel more PA, their motivation to adhere to their medication is replenished, and it becomes easier to adhere to medication in a sustained manner. Another potential pathway linking PA and MA is suggested by the Broaden and Build Theory of positive emotion, which posits that PA broadens the scope of thoughts, urges to action, and perceptions that one has as well as builds cognitive and social resources (Fredrickson, 1998). For example, it is possible that PA incites creative solutions to remember medication, or helps people build supportive relationships that encourage adherence.

There are a number of factors to consider in drawing conclusions from this review. First, we are unable to analyze the associations of PA and MA in any conditions besides HIV and cardiovascular conditions because none of the studies in other chronic conditions (e.g., asthma, arthritis, cancer, chronic kidney disease) met all the inclusion criteria. Future high-quality work that examines PA and MA is needed, particularly in these chronic conditions where adherence is paramount (e.g., diabetes, asthma).

There was a lack of bivariate analyses (e.g., analysis of the uncontrolled correlation) between PA and MA reported in the examined literature, with only three studies including unadjusted analyses of this relationship. Importantly, a greater number of uncontrolled analyses would allow researchers to compare the relationship between PA and MA across very different samples. While the majority of studies employed covariates, use of specific covariates varied between studies. For example, a small number of studies controlled for NA, and of the longitudinal studies, only one controlled for baseline MA (Sin et al., 2015).

The range of effect sizes was larger among studies examining the relationship between PA and MA in HIV (d = −0.40 – 1.27) than in cardiovascular disorders (d = 0.01 – 0.65), though there was a smaller sample size for studies examining this relationship in cardiovascular conditions than in HIV. The single study for which there were negative effect sizes was the study by Fipro-Perretti et al. (2018), which examined the relationship between frequency of PA word use in oral autobiographical narratives and self-reported MA 3–9 months later. There were no significant relationships between PA and MA in this study (Firpo-Perretti et al., 2018). However, the directionality of the findings and effect sizes suggest that the methods by which PA is measured are impactful.

Indeed, measurement of both PA and MA varied widely and impacts the strength of the conclusions we are able to draw. For example, some researchers note 80% adherence as adequate, while for HIV in particular, at least 95% adherence to medication is critical (Osterberg & Blaschke, 2005; Paterson et al., 2000). Classifying MA as adequate by percentages can be further complicated by the fact that patients can take more medication than prescribed, leading to rates of greater than 100% adherence (Osterberg & Blaschke, 2005), although this theoretical complication did not influence present results.

Eight of nine included studies employed both self-report PA and MA and, as with most self-report measures, responses may be skewed by social desirability (Lucas, Diener, & Larsen, 2003; Wagner & Miller, 2004). Misreporting of both self-report PA and MA, may also be an issue, although reasons for misreporting may differ: There is evidence suggesting that people self-enhance PA when measuring this construct (Wojcik & Ditto, 2014), and have difficulty recalling accurate MA when reporting adherence (Lam & Fresco, 2015). This being said, evidence suggests self-report measures of adherence are consistent with electronic data and viral load among people living with HIV (Walsh, Mandalia, & Gazzard, 2002), and are consistent with pill counts among hypertensive adults (Haynes et al., 1980).

In addition, people may inaccurately report retrospective PA or MA due to normatively biased retrospective recall (Lam & Fresco, 2015). Correlations between self-report PA and self-report MA may also be inflated by common-method variance (Podsakoff, MacKenzie, Lee, & Podsakoff, 2003). It is unclear whether the relationship between PA and objective measures of MA (e.g., levels of antiretroviral therapy medications in hair samples (Hickey et al., 2014)) would yield similar results to the results seen in the included studies. Although there are a number of weaknesses to self-report measures of MA, non-self-report approaches also have drawbacks. For example, counting doses of medication remaining in a bottle during doctor visits or employing electronic monitors of medication bottle opening may yield inaccurate results due to pocket dosing (which could under-report adherence) or patients opening their medication bottles but discarding medication (which could over-report adherence) (Osterberg & Blaschke, 2005).

For the measurement of PA, there are aspects of the emotional experience (Lucas et al., 2003), such as physiological activation and neurological activity, that are important for understanding emotional experiences in ways that self-reports are unable to capture. It is as yet unknown if or how these more objective measures of emotion are associated with MA. Although non-self-report measures of emotion are often only weakly correlated with self-reported emotion experiences (Lucas et al., 2003), the complexity of accurate emotion assessment is important to consider when drawing conclusions about relationships between PA and MA or other health behaviors. Especially important for the measurement of PA in this review is the qualitative difference seen in measuring PA via linguistic coding as compared to self-report. In particular, as noted by Pressman and Cohen (2012), as reporting positive emotions is socially desirable (Diener & Suh, 2003) and differing reporting styles lead to different self-report emotion results (Lucas et al., 2003), there has been a call for using alternative methods of assessing emotion (Diener, Suh, Lucas, & Smith, 1999), which Firpo-Perretti et al. (2018) answered by employing LIWC to analyze emotion word use. It is notable that this study did not find a positive relationship between PA and MA, and points to the importance of measuring emotion in a multifaceted manner.

Irrespective of measurement of PA and MA, no studies can rule out the “third variable problem,” which states that two variables may be correlated due to a common correlation with a third variable (American Psychological Association, 2018). For example, it is possible that the number of close others with whom one interacts may influence both PA and MA by raising their PA and reminding them about their MA. Similarly, engaging in health behaviors such as regular exercise may increase both PA and MA by both directly boosting PA and increasing attentiveness to other health behaviors.

Limitations

In additions to the limitations in the studies reviewed, discussed above, there are three important limitations to note in the present review. Although we followed published guidelines for systematic reviews (Moher et al., 2009), we did not register our search protocol prior to the start of the review as is required for registration with PROSPERO (University of York Centre for Reviews and Dissemination, n.d.). The protocol is available in Appendix B. A second weakness that plagues virtually all systematic reviews is publication bias or the “file drawer” problem (Rosenthal, 1979). While it is possible to calculate Rosenthal’s fail-safe N (Rosenthal, 1979) in meta-analyses, thus accounting for the file drawer problem statistically, it was not possible to calculate Rosenthal’s fail-safe N for this qualitative synthesis. In addition, we only searched peer-reviewed journals. As it is more likely for significant findings than non-significant findings to be published, it is possible additional studies showing no relationship between PA and MA have been conducted but not published in peer-reviewed journals (Rosenthal, 1979). Finally, we restricted our review to one specific health behavior – MA. Although MA is a good starting point, additional work examining associations of PA with other critical health behaviors would add significantly to the literature and serve as important guidance in developing health behavior interventions.

Future Directions

Although both cross-sectional and prospective cohort studies have examined the relationship between PA and MA, no randomized controlled trials in which PA is manipulated have examined this association. Ogedegbe et al. (2012) conducted a randomized controlled trial demonstrating that an intervention that targeted PA led to higher adherence to hypertensive medication, but this study did not explicitly test whether increased PA led to the intervention effects. Future research should explicitly examine whether randomized controlled trials that increase PA result in improved adherence in chronic conditions.

Future studies should include bivariate analyses as well as relevant covariates when multivariate analyses are warranted based upon previous literature and statistical analyses. Inclusion of the uncontrolled association between PA and MA will facilitate comparison across diverse studies. Similarly, researchers must take advantage of the fact that while PA and NA are relatively independent, they do overlap to some extent (Watson, 1988b). As such, researchers would be advised to control NA when conducting analyses, to allow more certain conclusions about the unique effects of PA. Moreover, the quality and quantity of social relationships, health behaviors that could influence both PA and MA, as well as other potential unmeasured but correlated variables should be controlled for.

As the wide variability of PA and MA self-report scales renders comparison across studies and synthesis of the current studies challenging, standardized and validated PA and MA scales should be used in future work. Ideally, multiple methods of PA and MA measurement would be employed within studies so as to capture different facets of emotion and to overcome biases occurring with reliance on a single measurement approach. Using multiple methods of measurement may also decrease the third variable problem; especially that of potential misreporting when self-reporting PA and MA. This would also increase confidence that the effects seen in the studies that use both self-report PA and MA are real and not the result of misreporting and human error.

Implications

While nascent, the literature on the relationship between PA and MA in chronic conditions indicates that PA is likely correlated with and may potentially influence MA. It is crucial for more work to be done in this area so that researchers and clinicians can have a better understanding of whether this is a relationship that they can capitalize upon. Future researchers can take advantage of the emerging state of the literature and conduct randomized controlled trials that have the potential to not only inform science, but also to improve health of people living with chronic conditions. Moreover, researchers have the opportunity to use multiple measurement methods and add to certainty that findings in this nascent literature are real, and not the result of measurement error. Better measurement and more longitudinal studies would move this area of research forward in important directions, allowing more concrete conclusions to be drawn. Indeed, future robust research could very well improve our ability to increase MA, and perhaps, to ultimately decrease the morbidity, mortality, and cost associated with suboptimal MA in chronic physical conditions.