Systematic review of the predictors of statin adherence for the primary prevention of cardiovascular disease

Introduction Previous research has shown that statin adherence for the primary prevention of CVD is lower compared to secondary prevention populations. Therefore the aim of this systematic review was to review predictors of statin adherence for the primary prevention of CVD. Methods A systematic search of papers published between Jan 1984 and May 2017 was conducted in PubMed, PsycINFO, EMbase and CINAHL databases. A study was eligible for inclusion if; 1) it was a study of the general population or of patients with familial hypercholesterolemia, hypertension, diabetes or arthritis; 2) statins were prescribed; 3) adherence was defined and measured as the extent to which patients followed their statin regimen during the period of prescription, and 4) it was an original trial or observational study (excluding case reports). A study was subsequently excluded if 1) results were not presented separately for primary prevention; 2) it was a trial of an intervention (for example patient education). Papers were reviewed by two researchers and consensus agreed with a third. A quality assessment (QA) tool was used to formally assess each included article. To evaluate the effect of predictors, data were quantitatively and qualitatively synthesised. Results In total 19 studies met the inclusion criteria and nine were evaluated as high quality using the QA tool. The proportion of patients classed as “adherent” ranged from 17.8% to 79.2%. Potential predictors of statin adherence included traditional risk factors for CVD such as age, being male, diabetes and hypertension. Income associated with adherence more strongly in men than women, and highly educated men were more likely and highly educated women less likely to be adherent. Alcohol misuse and high BMI associated with non-adherence. There was no association between polypharmacy and statin adherence. The evidence base for the effect of other lifestyle factors and health beliefs on statin adherence was limited. Conclusion Current evidence suggests that patients with more traditional risk factors for CVD are more likely to be adherent to statins. The implications for future research are discussed.


Introduction
Previous research has shown that statin adherence for the primary prevention of CVD is lower compared to secondary prevention populations. Therefore the aim of this systematic review was to review predictors of statin adherence for the primary prevention of CVD.

Methods
A systematic search of papers published between Jan 1984 and May 2017 was conducted in PubMed, PsycINFO, EMbase and CINAHL databases. A study was eligible for inclusion if; 1) it was a study of the general population or of patients with familial hypercholesterolemia, hypertension, diabetes or arthritis; 2) statins were prescribed; 3) adherence was defined and measured as the extent to which patients followed their statin regimen during the period of prescription, and 4) it was an original trial or observational study (excluding case reports). A study was subsequently excluded if 1) results were not presented separately for primary prevention; 2) it was a trial of an intervention (for example patient education). Papers were reviewed by two researchers and consensus agreed with a third. A quality assessment (QA) tool was used to formally assess each included article. To evaluate the effect of predictors, data were quantitatively and qualitatively synthesised.

Results
In total 19 studies met the inclusion criteria and nine were evaluated as high quality using the QA tool. The proportion of patients classed as "adherent" ranged from 17.8% to 79.2%. Potential predictors of statin adherence included traditional risk factors for CVD such as PLOS  Introduction HMG-CoA (3-hydroxy-3-methyl-glutaryl-coenzyme) reductase is the rate-controlling enzyme of the mevalonate pathway, the metabolic pathway that produces low density lipoprotein cholesterol (LDLc). HMG-CoA reductase inhibitors (statins) reduce the level of LDL-C and other isoprenoids (lipids) and thereby lower the risk of atherosclerosis and cardiovascular disease (CVD) [1]. Statins also have multiple other (so called pleiotropic) effects which may contribute to the reduction of CVD risk including effects that stabilise atherosclerotic plaques, support endothelial function and reduce inflammation of the vasculature [2] Whilst statins are used for the primary and secondary prevention of CVD the risk of death is lower for the primary prevention population, therefore it is important to evaluate the benefit of statins in this setting separately. Data from a prior meta-analysis of eleven randomised controlled trials (RCTs) indicated there was no association between statin use and all-cause mortality (RR 0.91, 95% CI 0.81-1.01) [3]. More recently, Taylor et al. conducted a Cochrane review of 18 RCTs and 19 observational studies and reported compared to placebo, statins reduced lipid levels and also the risk of experiencing a fatal or non-fatal cardiovascular event (CVE) by a quarter (RR 0.73, 95% CI 0.67-0.80), the equivalent number needed to treat (NNT) for five years was 56 (95% CI 46-75) [4]. They also analyzed the incidence of adverse events, including cancer, myalgia and rhabdomyolysis, arthritis, and increased liver enzyme, and found no evidence of increased risk for statin users compared to placebo or control participants, except in one trial in which there was an increased risk of type II diabetes [4]. The authors postulated there may be a risk of stroke, but there was no data to investigate this, and the authors recognized that not all trials in their review investigated side effects. Thus, there is evidence for cardiovascular but not wider benefits from statins use in the primary prevention setting.
Current UK clinical guidelines recommend that a person with at least a one in ten risk of experiencing a fatal or non-fatal CVE in the next ten years should be offered atorvastatin at 20mg daily or an equivalent dose of another licensed statin for its primary prevention [1]. The risk of primary CVD is calculated using a cardiovascular risk calculator. The QRISK3 calculator is most commonly used in the UK [5], the Framingham Risk Score (FRS) in the US [6], and the Systematic Coronary Risk Evaluation (SCORE) in Europe [7]. All of these calculators include the following components to calculate the risk of CVD; age, gender, smoking status, systolic blood pressure, the level of LDL-C and presence of co-morbid diabetes and hypertension, and some contain additional factors [8].
adherence specific to the primary prevention setting is merited. Lemstra et al. lamented the lack of studies investigating the effect of psychological and lifestyle factors on adherence [20]. Factors such as depression, medication and illness beliefs have been shown to predict nonadherence to other long-term medications [25]. A contemporary meta-analysis of interventions to improve statin adherence found strategies such as patient education, counselling, simplifying regimens, issuing reminders and even interventions classed as multi-faceted that included a combination of the above strategies only achieved small positive effects upon adherence (Hedges g<0.5) [26]. Clearly there is a need to investigate the reasons for and predictors of adherence in more depth as this will allow for better targeted and tailored strategies to optimise statin adherence for the primary prevention of heart disease.
Therefore the aim of this systematic review is to identify predictors of statin adherence for the primary prevention of CVD.

Search strategy
EMbase, Medline, CINAHL (Cumulative Index to Nursing and Allied Health Literature) and PsycInfo databases were searched from January 1984 (when the first trials of statins were published) until May 2017 [27], using Patient Intervention Comparison Outcome (PICO) search methodology to build the following strategy [28]: P)Primary prevention as the patient population; I) one or more statins as an intervention; C) predictors of adherence as comparators and O) adherence as a measured study outcome. The PICO comparison category would be identified at subsequent stages of the selection process. Synonyms for each PICO category were defined and the databases searched to identify abstracts that included a synonym from each category in the title, original title, abstract, subject heading, name of substance, or registry word fields (S1 Table). The apriori review protocol is available upon request.

Study inclusion
Studies obtained from the systematic search were eligible for inclusion if: 1) People were receiving treatment for the primary prevention of CVD or the results were given separately for primary prevention; 2) a statin was prescribed; 3) adherence was defined as the extent to which patients followed their statins regimen during the period of prescription, rather than the length of time till statin discontinuation; 4) predictors of adherence were defined and measured and 5) the study was a piece of original research (including abstract, thesis or conference proceedings). Titles and abstracts obtained from the search were independently evaluated by two researchers HH, SF and, where there was a disagreement, adjudicated by a third reviewer (DS). If the cohort was not defined as primary or secondary prevention cohort, for example if registry or pharmacy refill datasets were used to create 'incident statin user' cohorts, the reviewers assumed these analyses would include at least some incident statin use after a CV event or diagnosis and therefore these studies were excluded from the review. If primary and secondary prevention populations were jointly investigated, studies were only included if results specific to the primary prevention sample were evident. If the number or proportion of patients in adherent and non-adherent groups were presented these data were extracted and odds ratios for adherence calculated. Trials were included if there were secondary analyses of both arms of the study that investigated predictors of adherence.
If original research met the inclusion criteria but only existed as an abstract, thesis or conference proceedings, and the effect of factors on adherence was available it was included. Relevant reviews and opinion articles were retrieved in order to cross reference to ensure all relevant articles were included.

Quality assessment
The quality of the included studies was formally assessed using the quality assessment tool measure used in a previous published systematic review by Hope et al [25]. The quality assessment consisted of sixteen items, adapted from the recommendations of Sanderson et al. which state that observational studies should be evaluated on the use of appropriate methods to: 1) select participants, 2) measure exposure and outcome variables, 3) control for confounding, 4) reduce bias and 5) analyse data [29] (S2 Table). The authors judged papers that scored fourteen or more as high quality (range (0-17)). Trials were judged using the same criteria, since the data were analysed as if they were prospective cohort studies.

Evidence synthesis
Quantitative synthesis. Rates of non-adherence were inverted to calculate a rate of adherence for each study. A predictor was selected for quantitative pooling if there were at least three studies with a combined sample �1000 that investigated the same or similar predictor using the same or similar analysis (binary versus continuous data). Where cohorts were stratified by age, the effect size and the sample size of said strata were entered as separate effects into the meta-analysis. Where studies had stratified their cohort and obtained separate estimates for each cohort these were treated as separate cohort studies. Where studies had used the same data source the study with the larger sample size was included in the meta-analysis unless the effect was only investigated in the smaller study. These estimates were pooled using a fixed effects meta-regression analysis that adjusted for the study sample size. The I 2 statistic was used to evaluate the proportion of variance across the studies attributable to study heterogeneity. Sensitivity analyses of adjusted and unadjusted effect sizes, period of follow-up (� 1 year, 1 year, > 1 year), region, gender distribution, age range, measure of adherence and % adherent (< 50% versus � 50%) were conducted to identify possible sources of heterogeneity.
Qualitative synthesis. Predictor data that did not meet criteria for quantitative synthesis, or data where the pooled estimate possessed high heterogeneity (I 2 >50%), were qualitatively compared across studies and evaluated based on the definitions of strong, moderate, limited and conflicting evidence of van Tulder and colleagues [30]. Strength of evidence for an association was graded 1-5, where 5 meant there were multiple high quality studies, where high quality meant the study scored �14 on the QA score and the specific analysis adjusted for potential confounding, with a total sample size �1000. To score four there had to be a total sample size �1000 from several studies including one high quality study. To score 3 there had to be evidence from one high quality study, or several low quality studies with a total sample size �1000. To score 2 the evidence was taken from several low quality studies or one high quality study with a total sample <1000, and 1 was scored where there was only one low quality study with a sample less than 1000. Where there were inconsistent findings with the same level of evidence these were classed as '0' to indicate conflicting evidence. The evidence could be conflicting in relation to the presence of or direction of an effect (Table 1). All unique predictors were included in the qualitative synthesis. Where studies utilised the same cohorts and duplicate effects existed then the effect size from the higher quality study was included in the synthesis.

Results
The systematic search generated 2049 abstracts, a further 12 were included after snowballing and after duplicates (n = 284) were removed. 1777 abstracts were screened. After screening 257 abstracts fulfilled the inclusion criteria and a full paper review was performed. Some papers had to be excluded because the patient cohort was not clearly defined (n = 23) or the primary and secondary prevention cohorts were analysed together (n = 118). Other papers were excluded because variables were not compared across adherence levels (n = 34), or there was insufficient data to include (n = 28), discontinuation or persistence were measured rather than adherence (n = 23), or they were fixed dose combination (FDC) therapies (n = 11). A total of 19 papers fulfilled all inclusion criteria (Fig 1).
The proportion of patients included in the review defined as adherent ranged from 17.8% to 79.2%, which indicates that overall adherence to statins for the primary prevention of CVD, however it was measured, appears to be sub-optimal. There was some evidence that the wide variation of the number of adherent patients reflects the heterogeneity across studies with respect to the characteristics of the sample and study design ( Table 2). The length of follow-up and type of adherence measure appeared to account for some of the variability observed across studies (Fig 2). Quantitative synthesis of the data that met the conditions for a meta-analysis revealed pooled estimates with high heterogeneity (I 2 >90%), therefore only the qualitative synthesis is reported.

Traditional cardiovascular risk factors
The standard components of the CVD risk scores (i.e. age, male gender, LDL-C levels, the presence of comorbid diabetes and the presence of co-morbid hypertension) are used by physicians to decide who should be prescribed a statin for the primary prevention of a CV event.  We considered these patient factors as potential predictors of adherence because the physician may make the patient aware that possessing these characteristics increases their ten year risk for a CV event.
Age-positively associates with statin adherence (Strength of evidence = 5). There is strong evidence that older age predicts statin adherence. Four studies (three high quality & one low quality), including a total of 662638 participants, that adjusted for confounders [31,34,35,44] and six studies (Total N = 496921, two high quality & four low quality) with unadjusted effects found adherence increased with older age [37,40,42,43,47]. One small (N = 169) high quality study and one small low quality study (N = 158) found age did not associate with adherence [32,45] (Fig 3). Braamskamp et al may have found a different effect because they investigated adherence in a cohort of young adults with familial hypercholesterolemia; this population has an average age of 24 years, this is much younger than the typical population who commence statins [32]. Wallach-Kildemoes et al. found that the adjusted odds of adherence increased by up to a factor of 2 per five year increase in age in their male and female cohorts aged 40 to 65 years [44]. From the same cohort study the odds of being adherent decreased by up to 60% per five year increase in age in the male and female cohorts aged 65-80 years.
Men are more adherent than women-Strength of evidence = 4. One large high quality study that adjusted for other factors and four low quality studies with unadjusted effect sizes (Total N = 301106), reported men were more adherent than women [34,35,37,42,43]. In one high quality and one low quality study women were more adherent than men (Total N = 318952), but these effects were not adjusted for other factors [40,44]. Finally there was limited evidence that gender has no effect upon statin adherence from three studies (Total N = 527, n high quality = 1) [31,32,45] (Fig 4).
High cholesterol / Dyslipidemia does not associate with statin adherence-Strength of evidence = 3. In one large high quality general population study the presence of dyslipidemia was not associated with the odds of being adherent in the female or male cohorts [32]. Braamskamp et al. found, in a population of young adults with familial hypercholesterolemia, baseline LDL-C levels were not associated with self-reported adherence over the past month (OR 0.90, 95%CI;0.70-1.19).
Diabetes associates with statin adherence-Strength of evidence = 4. Eight studies examined the relationship between diabetes and being adherent [31,34,35,37,40,42,43,45]. There was strong evidence that that people with diabetes are more likely to adhere to statins, four large studies including Aarnio et al. that adjusted for other confounders found that people with diabetes or who used of antidiabetic medications had an increased odds of being adherent compared to non-diabetics (Total N = 376694). There was moderate evidence that diabetes does not associate with adherence from three low quality studies (Total N = 14576, mean QA score = 11) and limited evidence that diabetes reduced the odds of being adherent (Total N 79010, QA score = 13) (Fig 5).
Hypertension/blood pressure associates with adherence-Strength of evidence = 5. Qualitative synthesis of the evidence indicated that there is very strong evidence that hypertension positively associates with statin adherence even after adjusting for other factors (Total N = 551094, n High quality = 2) [31,34,37,40,42], there was moderate evidence of no effect of hypertension on statin adherence in women (Total N = 65256, QA score = 16) [34], and no evidence of a negative effect (Fig 6). Aarnio et al. (2016) also noted that the odds of adherence to statins increased for every additional class of CV medications prescribed, and this effect was observed in both men (OR 1.03, 95%CI; 1.0-1.06) and women (OR 1.04, 95%CI;1.01-1.08).
Being an ex-smoker associates with adherence-Strength of evidence = 3. Smoking status was investigated in two studies. In one large high quality study being a current smoker (vs being a non-smoker) did not predict being adherent (OR 1.01 (0.86-1.18)) (n = 6458, QA score = 16), whilst in this same study being an ex-smoker predicted good adherence (OR 1.20, 95%CI;1.0-1.3) [46]. One small low quality study compared current smokers versus nonsmokers and former smokers grouped together and found a non-significant negative effect of smoking (yes v no) (OR 0.69 95%CI: 0.23-2.07) (n = 200, QA score = 13) [31].

Socioeconomic factors
Low socioeconomic status indicated by lower levels of income, education and work status are known to associate with CVD. One of the mechanisms through which this association may occur is lower levels of adherence to medications such as stains.
Higher income associates with adherence and interacts with gender-Strength of evidence = 5. Two high quality studies that adjusted for other confounders including socioeconomic factors compared adherence across income quintiles [34,44]. Wallach-Kildemoes et al.
(2014) stratified their cohort by age and gender and then split their samples into quintiles of income that took into account family composition; the exact income thresholds for each quintile were not presented. Compared to participants in the lowest income quintile, participants in the higher income quintiles were more likely to be adherent, after adjustment for age, income, education and hypertension. This effect was observed in men and women of middle and post-retirement age; the strongest effects were observed for men of middle age (OR 1.56, Systematic review of the predictors of statin adherence 95%CI; 1.54-1.56) [43]. Aarnio et al (2016) used the taxable income per year to calculate income quintiles and used the wealthiest quintile as their reference category [33], in men there was a strong positive effect of income on the odds of adherence; compared to men in the wealthiest quintile men in succeeding lower income quintiles were less likely to adhere and the strongest effect was observed with men in the poorest quintile (OR 0.74, 95%CI; 0.68-0.79). The strength of these associations was attenuated in the cohort of women; only women in the poorest quintile were less likely to be adherent compared to women in the highest income quintile (OR 0.93, 95%CI; 0.86-1.00). These analyses adjusted for other socioeconomic and clinical factors but not smoking status.
Higher level of education associates with statin adherence and interacts with gender-Strength of evidence = 5. Four studies provided data on the level of education and statin adherence [34,44,45,47]. In studies where more than 50% of the sample were male a higher level of education increased the likelihood of adhering (OR 1.07, 95%CI;1.04-1.10), whereas in studies where 50% or more of the sample were women a good education reduced the likelihood of adhering to statins (OR 0.92, 95%CI; 0.89-0.95). These estimates included two studies that were of high quality and adjusted for other confounders including socioeconomic factors [34,44]. Aarnio et al. found that the likelihood of being adherent was lower for men if they had a basic level or secondary level education compared to those with a degree [33]. Wallach-Kildemoes et al reported a similar positive effect in men who had 12 or more years of education compared to those with 7-10 years or 10-12 years of education [43]. This effect attenuated once income and age were controlled for in the analyses but remained for the most versus the least educated men. The opposite effect was observed in women; increasing levels of education were associated with a lower odds of being adherent, and these effects remained even after Systematic review of the predictors of statin adherence controlling for other covariates, the strongest effect was observed in Finnish women aged 40-64 years (OR 0.85, 95%CI; 0.85-0.85) (Fig 7).
The relationship between work status and statin adherence-Strength of evidence = 0. Two large high quality studies and one small low quality study investigated the effect of work on adherence [31,34,38]. Being in work reduced the likelihood of adhering to statins. In particular compared to employed people, retired people were 11% more likely to be adherent, and this appears to be the case for both men and women after adjustment for other factors including age and other comorbidities [34]. However, Halava et al. followed a Swedish cohort over the transition from employment to retirement and captured the prevalence of non-adherence [38]. Using a repeated measures design they found that adherence to statins was lower after retirement (PR 0.85, 95%CI; 0.80-0.88). Halava et al. adjusted for the calendar year, time in study, and age at retirement, but not for other factors such as the number or type of co-morbidities that may have confounded this relationship. Given there were only two studies with conflicting findings further research is needed. Simple comparisons of retired and employed groups should be avoided since they may be confounded by age, which strongly associates with both statin adherence and retirement.
The effect of region on adherence-Strength of evidence = 4. Four studies provided data on adherence across different regions including one large high quality study that adjusted for confounders and several studies with unadjusted effects Several studies looked at adherence in regions within countries and observed significant differences across regions of the USA, Finland and Kuwait [34,42,43,45]. The effect sizes observed suggest the effect of living in a particular place could reduce the odds of adhering by between 10 to 50% ( Table 3). The reason for these differences could be variations in health, health services or socioeconomic factors.

Other demographic factors
Three studies provided data on other demographic factors [34,35,47].
Being married negatively associates with statin adherence-Strength of evidence = 3. Aarnio et al. reported that, compared to single men, married men after adjustment for other covariates were less likely to adhere to statins (OR 0.85, 955CI;0.80-0.91), as were divorced men (OR 0.61 95%CI; 0.56-0.67) and widowed men (OR 0.79, 95%CI; 0.69-0.79). The effects were very similar in the cohort of women [34]. The finding that married men and women are less adherent than single men and women is surprising, since married individuals generally benefit from spousal support that should ease not hinder adherent behaviour, therefore further research is required.
Racial background associates with adherence-Strength of evidence = 3. The effect of race on statin adherence was investigated in two US studies. In adjusted analyses Mann [35]. These findings can be interpreted within the context of previous findings that in the US Black and Hispanic Americans face more barriers to adhering to statins such as lower levels of insurance and access to care.   Co-payment dispensation (euro cents/ tablet) <20 Ref.
(Continued )           Given the increased prevalence of CVD risk factors such as hypertension and diabetes in Black and Hispanic US populations, this finding is particularly of note.

Comorbid conditions
It was possible to investigate the association between co-morbidity and adherence on ten studies [34,35,37,[39][40][41][42]45,47,49]. Given the increasing likelihood of comorbid conditions as people age and the positive association between age and adherence the unadjusted analyses should be interpreted cautiously.
The effect of increasing comorbidity on statin adherence: Strength of evidence = 0. The number of co-morbid conditions as a measure of disease burden was counted in six studies [34,35,39,40,42,47]. One large high quality study reported the odds of being adherent increased per additional comorbidity, after adjustment for age, gender, ethnicity and co-payment status [35]. One small high quality study reported that any co-morbidity increased the odds of self-reported adherence by a factor of ten, but this was unadjusted for other factors [47]. Three of these seven studies [34,39,42] calculated the Charlson Comorbidity index (CCI) [50], a validated measure of disease burden. Unadjusted effects from Lavikainen et al. reveal that participants with a CCI�1 were more likely to be adherent than those with no comorbidity (OR 1.08 95%CI;1.00-1.16)) [39]. Similar sized unadjusted effects of having a score of one or two on the CCI compared to zero were reported by Rublee et al. [42]; however adherence was lower in the group with a CCI �3 compared to zero comorbidities (OR 0.75, 95%CI; 0.70-0.80). In contrast, Aarnio et al. who used the same registry data as Lavikainen et al. found no association between a one point increase in the CCI and the odds of being adherent in either men or women after adjusting for the other covariates [34]. Perrault et al. used a different measure of comorbidity, the chronic disease score [51], and people classed as having a high chronic disease score (�4) were more adherent (OR 1.43, 95%CI;1.14-1.20) [39].
Depression inversely associates with statin adherence-Strength of evidence = 3. The existing evidence from five studies suggests that a diagnosis of depression does impact statin adherence [31,34,40,42,49]. In one study (n = 116846, QA score = 16) depression (identified from ICD-10 codes in Finnish registers) inversely associated with good adherence after adjustment for other covariates including age and socioeconomic factors in men (OR 0.85, 95%CI; 0.79-0.93) and in women (OR 0.91, 95%CI; 0.85-0.95) [34]. One low quality study (n = 158, QA score = 10) found an inverse association between depression and statin adherence using the Hamilton Depression Rating Scale (HDRS), where higher scores indicate increasingly depressive symptomatology [52], and adherence was measured using a Medication Event Monitoring System (MEMS) [49]. Unadjusted analyses from one large low quality study that identified depression from ICD-9 classification codes found no association between depression and adherence, nor did one small low quality study that used the Depression Anxiety and Stress Scale to measure depression [31,42]. Finally, data from one low quality study included use of antidepressants and this associated positively with adherence (OR 1.21, 95%CI; 1.17-1.24).
Association of anxiety with statin adherence-strength of evidence = 0. Four studies captured anxiety data and unadjusted effects were calculated [31,39,40,49]. Two studies found (n = 42046, n high quality = 0) that anxiety did not associate with adherence [30,38]. Two studies found anxiety did associate with adherence (n = 243072, n high quality = 0) [40,49]. Currently, the extent to which anxiety associates with adherence is poorly understood and evidenced.
The association of other mental health diagnoses and statin adherence-strength of evidence = 3. Aarnio et al found the presence of a "mental disorder" increased the odds of being adherent to statins by approximately 40% in men and women after adjustment for other comorbid conditions, socioeconomic, demographic and clinical factors [34]. Mental disorder here refers to the ICD10 codes for schizophrenia, psychotic, bipolar and manic disorders; there is evidence that cardiovascular mortality is higher in these groups and people may be informed of their increased risk and therefore adhere accordingly.
Obesity inversely associated with statin adherence-Strength of evidence = 3. Aarnio et al. found that obesity was associated with a lower odds of being adherent in women by about 25%, the same size of effect was observed in men but the confidence intervals crossed one (OR 0.81, 95%CI; 0.64-1.02) [33]. Rublee et al reported a very similar size of negative effect (OR 0.83, 95%CI; 0.78-0.88) [42]. Three studies collected data on BMI; Halava et al (n = 6458, QA score = 16) found, after adjustment for other clinical (depression, cancer and self-rated health), demographic and lifestyle factors, people classified as being obese (BMI>29.9) or overweight (25<BM<29.9 kg/m2) were approximately 15% less likely to be adherent than people with a BMI�25 kg/m 2 [46]. Importantly, Halava et al. considered cardiovascular comorbidities and risks for CVD including diabetes and hypertension to moderate the relationship between lifestyle and adherence and these people were excluded from this particular analysis. In contrast, two studies (n = 327, n high quality = 0) found that BMI did not differ across adherent and non-adherent groups [31,45].
Other co-morbid conditions: Strength of evidence = 3. Aarnio et al. and Rublee et al. included data on other comorbid conditions, these were identified from ICD10 and ICD9 codes collected in their respective Finnish and US registers (n = 195856, n high quality = 1) [34,42]. Aarnio et al. (QA score = 16) calculated adjusted odds ratios for men and women separately [33], whereas data provided in the Rublee study (QA score = 13) were used to calculate univariate odds ratios without stratification by gender [41].
Data were collected on cancer, respiratory disease, renal disease and rheumatoid arthritis.
Alcohol misuse inversely associated with statin adherence-Strength of evidence = 4. Four studies evaluated this association. Two studies (n = 151140, n high quality = 2) reported that severe alcohol misuse, or alcoholism nearly doubled the risk of non-adherence after adjustment for other factors, in particular Bryson et al. were able to control for potential confounding from level of education and smoking status [34,36]. Two studies (n = 6616, n high quality = 1) found no effect of alcohol use or extreme drinking on statin adherence [45,46]. However the level of drinking captured in these studies may not be comparable to a diagnosis of alcoholism or self-reported severe alcohol misuse.
Physical activity does not associate with adherence-strength of evidence = 3. Halava et al. (QA = 16, n = 6458) found no relationship between the level of self-reported activity measured using the metabolic equivalent of task (MET) index and adherence to statins either unadjusted or adjusted analyses [46]. Accurately capturing physical activity via self-report is challenging and further research using objective methods would better determine if there is a link between adhering to statins and exercise.
Health seeking behaviours associated with adherence-Strength of evidence = 3. Three studies [42,47,48] contained information on other types of health seeking behaviours. In two studies seeing a general practitioner increased the odds of being adherent by 20 to 25% [42,48]. Mann et al. reported that visiting a healthcare practitioner for a cholesterol check increased the odds of being adherent by a factor of four [46]. Rublee et al. (N = 79010, QA score = 13) captured data on the number of people who attended clinic for preventive services [41]. These included vaccinations, screening tests for bowel cancer, screening for osteoporosis, cervical and breast cancer in women and prostate screening in men. The size of the association varied depending upon the preventive service used but unadjusted analyses revealed people who undertook these health seeking behaviours were 10 to 30% more likely to be adherent during the one year adherence assessment period.
Other health behaviours associate with statins adherence-Strength of evidence = 1-2. Guthrie et al investigated the association between other self-reported health behaviours and adherence to statins (N = 4548, QA score = 5) [48]. Self-reported use of blood pressure control treatments, trying to quit smoking and increased physical activity increase the odds of being adherent by 40 to 50%. These were all unadjusted analyses and the study was deemed low quality, therefore extrapolation of these results to the primary prevention population in general is limited.
Health beliefs associate with adherence-Strength of evidence = 2. Mann et al. (N = 71, QA score = 14) investigated the association between health beliefs and adherence using a bespoke health belief questionnaire. Within the study there were conflicting findings (Table 3). Participants who endorsed the beliefs "plan to use statin: < 6 months", "personal risk of a heart attack: less than average", and "statins may be harmful to me" were two to three times less likely to adhere than veterans who did not endorse these views [47].

Treatment-related predictors
Polypharmacy is not associated with statin adherence-Strength of evidence = 3. Four studies investigated the association between the total number of medications a person received and adherence to statins [32,34,39,45]. Aarnio et al (QA score = 16, N = 116846) found that after adjustment for all other variables there was no increased likelihood of adherence per additional medicine for men and women [34]. Lavikainen et al. (QA score = 13, N = 42807) used a subsample from the same female cohort as Aarnio et al. but categorised the total number of medications into groups of 1-2, 3-5 and 6-31 medications. In unadjusted analyses patients who took more medications were 10 to 20% more likely to be adherent [39]. Farsaei et al reported adherent patients took 1.4 more medications than nonadherent patients (QA score = 7, n = 158) [45], and Braamskamp et al. found no association after adjusting for other factors between using other medications (yes v no) and adherence in their young familial hypercholesterolemia cohort (QA score = 13, N = 169) [32]. It is likely people on statins are prescribed other preventive CVD treatments such as antihypertensive therapies.
Type of statin associates with adherence-Strength of evidence = 3. Four studies examined the type of statin and the direction and size of effect for particular statins varied greatly across studies [31,34,37,40]. Given the high likelihood of bias from indication for particular statins only Aarnio et al. is reported here because these analyses adjusted for potential confounders (Table 3). Compared to simvastatin, people were more likely to adhere to fluvastatin and rosuvastatin, and less likely to adhere to lovastatin. After adjustment for other factors, Finnish people using pravastatin were no more adherent than people who used simvastatin [34].
Intensity of statin dose inversely associated with adherence = strength of evidence = 3. Aarnio et al. set cut-offs of intensity for each type of statin and then classed people as having a low, moderate or high dose of statins. Men on a moderate daily dose (OR 0.89, 95%CI;0.84-0.94) and men on a high daily dose were less likely to adhere compared to men on a low daily dose of statins (OR 0.70, 95%CI; 0.54-0.92) [34]. Similar and larger effects were observed in the cohort of women, women on a high daily dose of statins were 60% less likely to adhere compared to women on a low daily dose.
Timing  [39]. In a cohort of patients (QA score = 13, N = 169) with familial hypercholesterolemia, those who initiated statins before puberty were no more likely to be currently adherent than those who were first prescribed statins post puberty [32].
Longer pharmacy prescription associates with statin adherence = strength of evidence = 3. Batal et al. (QA score = 15, N = 3386) demonstrated that a longer supply of statins, 60 versus 30 days, was associated with an increased likelihood of being adherent after adjustment for clinical and demographic factors (RR 1.40, 95%CI;1.27-1.55) [35].
Other treatment related factors = strength of evidence = 1-2. Al-Foraih reported an unadjusted positive effect of longer statin duration on adherence; however, the authors did not describe over what period of time this was measured, and this study may be susceptible to left censorship [31]. Finally one large high quality study captured data on which professional had prescribed the statin [43]. People were nearly four times more likely to be adherent if the initial prescriber was a cardiologist than if the prescriber was a general practitioner, but this was an unadjusted analysis, without adjusting for the number of CVD risk factors. This effect is likely to be confounded by factors that influence whether the patient has a consultation with a cardiologist rather than a general practitioner. Braamskamp et al. also reported that self-reported side effects did not associate with adherence but noted that a minority of the cohort reported side effects.
Medication costs association with adherence-Strength of evidence = 0. Two studies using three different measures evaluated the impact of medication cost [34,35]. Aarnio et al. 2016 calculated the total out of pocket prescription costs for all medications and secondly by calculating the specific co-payment patients made with respect to their first statin prescription [34]. For every 50 euro increase in total costs there was approximately 10% increase in the likelihood of adhering. However, men and women who paid the highest tariff were over two times more likely to non-adhere than those paying the lowest tariff. The analyses adjusted for other socioeconomic factors such as income and education that may have confounded the association between cost and adherence. One other study investigated co-payments and found making co-payments (yes v no) did not affect adherence after adjustment for other factors [35], however given that 80% of this cohort made some kind of co-payment a comparison akin to Aarnio which compared different levels of co-payment may have demonstrated a difference. Given the heterogeneity of how cost was considered in the studies drawing a firm conclusion on the impact of cost is difficult.
No association between type of healthcare organisation and statin adherence-Strength of evidence = 3. Three studies investigated the type of healthcare delivery system and adherence [35,42,43]. Slejko et al and Batal et al. extracted data from US drug registries and investigated if the type of health plan associated with being adherent (Total N = 793396, N high quality = 1). Batal et al. compared those with or without insurance cover and found those with insurance cover were no more likely to be adherent than those without, after adjustment for other factors [35]. Data from Slejko et al. in unadjusted analyses indicate that members of commercial health insurance plans were no more likely to be adherent than those covered by Medicaid or Medicare insurance plans [43]. One study (N = 11126, QA score = 13) did report different rates of adherence for different healthcare organisations, based on the type of cover [42]. Compared to people with point of service (POS) plans, which allow people to access any healthcare professional at the point of service without paying into a plan, people with indemnity cover which does require regular payments but pays out when sickness occurs were twice as likely to be adherent. People with POS plans that limited who they could see in the healthcare service were twice as likely to be non-adherent. However these were unadjusted analyses that did not account for the different characteristics of people who use comprehensive indemnity insurance and those using basic insurance provided by the state, such as income and education level.

Discussion
This is the first systematic review specifically focused on predictors of adherence to statins in the primary prevention setting. In total nineteen studies were included, and many more could have been included if results had been stratified by primary and secondary prevention. This was not a review of adherence rates overall, but the level of adherence in these reviewed studies was sub-optimal, and further emphasises the importance of focusing on improving adherence in the primary prevention population. There was moderate to strong evidence that individuals with traditional CV risk factors have better adherence. In particular older age, male gender, a diagnosis of diabetes, and a diagnosis of hypertension predicted better adherence. In contrast, the evidence that adherent patients adopt other healthy behaviours to protect their heart was limited; only the evidence for an inverse relationship between alcohol misuse and obesity (in women) and adherence was convincing. These findings challenge the concept of the "healthy adherer effect", since those who are more ill appear to adhere to statins better. These findings are predicted by the need-concern framework of health beliefs which postulates those who perceive the greatest need despite medication concerns are more likely to adhere [53].
There was moderate to strong evidence that socioeconomic predictors such as wealth, employment status and level of education associate with statin adherence, and these effects have been observed in systematic reviews that have included primary and secondary prevention cohorts combined [20,21]. The association between higher socioeconomic status and adherence may be related to fewer practical barriers to adhering. However in this review factors such as longer prescriptions fills, medication costs and type of healthcare provision, in contrast to previous reviews that included secondary prevention populations, were inconsistently associated with statin adherence. The type and intensity of statin dose independently associated with statin adherence [34]. Whether these associations are underpinned by medication concern, such as those reported by Mann et al. or increased levels of adverse events is unknown. Only one study examined the relationship between side effects and non-adherence and returned a null finding [32]. However side effects were given as a reason for both noncommencement [33] and non-adherence [45] in two small studies, therefore this merits further investigation.
These findings align with results from previous reviews that increased health risks and male gender associate with better adherence [20,21]. In this review most studies reported a positive linear relationship between age and adherence. However, previous reviews indicate that age is nonlinearly related to adherence, suggesting age is an important modifier of the relationship between perceived risk of CVD and adherence.
In this review the association between higher income and better adherence was much clearer for men than women, and higher levels of education associated with lower statin adherence in women and higher statin adherence in men. The apparent sex dependent effect of socioeconomic status upon adherence was discussed by Aarnio et al. and they cite the unmeasured association between unhealthy lifestyles and low health literacy as an explanation [34]. There was strong evidence from this review that excess alcohol consumption in men and women was associated with lower adherence to statins, and strong evidence for women but not men with obesity to be less adherent. These associations may be partly underpinned by low health literacy. Educated men and women are likely to be more health literate. Men and women may perceive their susceptibility to primary CVD differently because sex informs the calculation of CVD risk. Alternatively, men and women may balance the overall need for a preventive medication with medication concerns differently.
These data suggest that individuals who are younger, female, or do not have diabetes or hypertension may have a lower perceived risk of developing CVD or experiencing a CVE [54], and reminders alone may not be sufficient to change behaviour [55]. The very limited evidence from this review supports the hypothesis that low perceived susceptibility to CVD independently predicted poor adherence [47]. Interventions that aim to improve patient and/or physician understanding of personal risk of incident CVD have demonstrated improved statin adherence in the short-term [56,57]. Lower risk individuals may also have less contact with medical professionals and there was limited evidence that attending clinic, or having a cholesterol check associated with better statin adherence.

Methodological limitations
The findings of this review must be interpreted in the context of its limitations. What constitutes a primary prevention population was fairly consistent across studies. One study excluded patients with diabetes and hypertension as Halava et al. considered these CVD risks may modify the relationship between lifestyle factors and adherence [46]. Data pertaining to potential predictors were extracted from ten studies to calculate univariable associations between these factors and statin adherence. The interrelatedness of clinical, socioeconomic and lifestyle factors cannot be addressed in such analyses. Where studies conducted multivariable analyses with large numbers of variables collinearity may not have been sufficiently addressed. Further research is therefore required to understand the relationship between the presence of cardiovascular risks, adoption of healthy lifestyles and statin adherence.
A limitation of measuring adherence is the risk of measurement reactivity; this was recently demonstrated in two RCTs designed to improve adherence. Patients were screened for sub optimal adherence based on their pharmacy refill records before entry in to the trial. These trials objectively measured adherence using a MEMS. After 3 months the level of adherence was very high irrespective of treatment arm with no significant differences. Given that only 1% of the sampling frame agreed to participate in these trials and the median patient adherence before entry into these trials was~60% it appears that the effective intervention to optimise adherence involved the patient's decision to participate in these studies, and potentially the patient's response to the electronic monitoring of their medication [58]. However, most of the studies included in this review used objective indirect methods to assess adherence, meaning the participant was unaware that they were being monitored. Therefore, there is limited bias in these data from measurement reactivity. Prescription refill data still have limitations both because there is an underlying assumption that all prescribed pills were taken between fills and because it is not possible to identify periods of time when a prescription is not filled because of medical advice to temporarily stop a statin.
Fixed dose combination (FDC) therapy or "polypill" therapies that combine blood pressure, cholesterol lowering and antiplatelet treatments into a single pill, are hypothesised as one strategy to improve adherence. The authors did not include FDC trials in this review because adherence to a polypill necessarily cannot be specific to a statin. Adherence to FDC compared to multiple CVD medications was investigated as part of a Cochrane review and levels of adherence were higher in the FDC arm but this was investigated in just one study [59]. In our review, four studies investigated polypharmacy and there was no evidence that polypharmacy lowered adherence to statins; one study reported that adherence increased per additional medication. Similar inconsistent effects have been observed in previous reviews. Given the current evidence base, FDC therapy is unlikely to increase adherence to statins. Other features of medicine use (time taken to complete, regimen complexity) may have a greater impact on longterm adherence than simply the number of medications.

Future research
This review draws attention to the limited number of well-designed observational studies examining multivariable predictors of adherence. Synthesis of the existing data allows one to infer possible mechanisms but there are a number of areas where more research is required. Firstly, is high perceived risk of primary CVD the mechanism that explains the observed associations between traditional CVD risk factors and statin adherence? Secondly, given there is evidence of sex dependent effects of socioeconomic factors on adherence, what are the drivers of these sex differences? Potentially, there is an interaction between gender and level of health literacy which results in gendered beliefs about the need for and concerns about statins, but this still needs demonstrating empirically. Thirdly, given that side effects and fear of side effects were given as reasons for discontinuation, it is remarkable that only one study investigated the relationship between side effects and adherence [32]. Patient tolerance of side effects could explain the observed links between dosing and type of statin and adherence [34], alternatively, people with prior negative expectations may misattribute symptoms such as muscular pain (myalgia) to statins use [60]. The low level of reporting of adverse events prevents the authors from drawing any conclusion about the effect of side effects on statin adherence. Therefore, future research should address the relationship between prior concerns about side effects, reported side effects and statins adherence. Fourthly, the relationship between lifestyle factors and statins adherence is poorly understood, nevertheless there is some evidence these associations may also be modified by patient factors. Future analyses using high quality prospective data could investigate if gender and age modify the association between lifestyle factors and adherence. Such analyses would allow one to infer the possible drivers of these differences. For example, high physical activity in a middle aged woman may predict poor statin adherence, because this woman perceives herself to be at low risk of disease, whereas high physical activity in an older man may predict high adherence because both his behaviours are underpinned by a high perceived risk of personal morbidity/mortality. Similarly, data from studies that investigate if the presence or absence of diabetes and/or hypertension alters the relationship between lifestyle factors and statins adherence could be used to infer the latent effects of health beliefs.

Conclusion
There is an ongoing debate about how widespread the use of statins should be in the field of primary prevention. Hence, improving adherence should not be at the expense of supporting people to make healthy lifestyle changes. This review makes clear the predictors of adherence common to primary and secondary prevention settings. There also appear to be important sex and age dependent differences that are specific to adherence to statins prescribed for the primary prevention of CVD. Further research is needed to understand better the underlying mechanisms of statin adherence.