Adherence influencing factors – a systematic review of systematic reviews

A systematic literature search was performed in MEDLINE (via Pubmed) and Embase (via Embase). The search strategy combined various terms and medical subject headings related to adherence, and oral medical medication (the full search strategies for each database are available in Additional file 1). The search was limited to a publication date after 1990. The search was performed on July the 15th 2013.

To be eligible for this review the studies had to meet the following inclusion criteria:

It was decided to focus on factors unrelated to the health care system (e.g. type of insurances), condition (e.g. symptoms) and medication (e.g. side effects) i.e. factors that are strongly related to health care system, condition or medications were excluded. For example side effects or symptoms are strongly associated with the respective medication disease respectively and can consequently have different influence on adherence. The purpose of focusing on disease and medication unrelated factors was on the one hand to insure comparability, consistency and clarity of results and on the other hand to identify factors that are widely applicable in clinical practice. A preliminary unsystematic search was performed to identify articles on adherence influencing factors. Based on identified articles a list of potential adherence influencing factors was prepared and chosen which factors are unrelated to the disease and therefore should be included in the analysis. The decision to include a factor was made independently by two reviewers and discussed until consensus in case of discrepancies. The following factors were eligible: age, gender, ethnic status, education, employment, financial status/income, marital status/not living alone, social support, measure of intake complexity (e.g. number of tablets, number of medications, frequency of intake), duration of therapy, duration of disease, comorbidity, co-payments, medication costs, insurance status (insured/not ensured).

Furthermore it was decided to include systematic reviews only if the risk of bias of included primary studies was assessed, documented and relatable to the respective factor because an examination of the evidence for an effect of a potential adherence influencing factor is only possible considering the validity of evidence of the primary studies.

Two reviewers performed the study selection independently according to the inclusion criteria above. Titles and abstracts of all articles were screened. The full-texts of all potentially relevant articles were than obtained and screened in detail. Any differences between the reviewers were discussed until consensus. In addition to identify grey literature the reference lists of all full-texts were cross-checked and an additional search in Google scholar performed.

The methodological quality of included studies was assessed using the AMSTAR instrument that was found to be valid and reliable to assess SRs [8, 9]. Each assessment question was rated with “yes”, “no”, “unclear” or “not applicable”. The instrument was applied independently by two reviewers. Disagreements were resolved in a discussion. The question “was the scientific quality of the included studies assessed and documented” was skipped because we formulated quality assessment of included studies as inclusion criterion (see inclusion criterion 5).

Data were extracted by one reviewer in standardized summary tables and were verified by a second reviewer. Any disagreements were discussed till consensus. For each systematic review, characteristics were extracted on the condition/medication (marked bold), inclusion and exclusion criteria for primary studies (only other than our applied inclusion criteria) and search period limits. Results were extracted according to the type of evidence synthesis. In the case of a narrative synthesis, results were abstracted by modified vote counting [10]. This contained data to the effect direction, all comparisons showing this effect direction, statistically significant comparisons showing this effect direction and total number of comparisons for this factor. This method has been suggested for presenting results of qualitative synthesis, overcoming problems arising when simple vote counting is used by relying either on the number of comparisons with a positive direction of effect or the number of comparisons reaching statistical significance. For all meta-analyses, pooled effect sizes and measures with confidence intervals and test and measure for statistical heterogeneity were extracted. All data in the tables were commutated so that the influence on adherence refers throughout to an increase of the respective factor independently whether the factor is positive (e.g. educational level) or negative (adverse events). A p-level of less than 0.05 was considered statistical significant. Overlaps (multiple included primary studies) were assessed in case of multiple systematic reviews on the same indication.

For all factors a summary evaluation of the influence was made. Two reviewers rated the evidence for an effect, considering the number of included systematic reviews for a comparison, the effect direction, the significance and consistency (within and between systematic reviews) of results, the methodological quality of systematic reviews and the methodological quality/risk of bias of included primary studies. If only one high quality study was included in a comparison that showed a statistical significant positive effect for a certain factor, this was rated as tendency for a positive effect. Discrepant judgments were discussed until consensus.

The electronic literature search resulted in 1604 hits after duplet removal (EndNote ×5). Of these 72 titles and abstracts seemed potential relevant and full-text versions of the publications were screened in detail. Seven studies met all inclusion criteria and were finally included in this overview [11‐17]. The process of study selection is illustrated in the flowchart (see Figure 1). Most systematic reviews were excluded because a risk of bias assessment was not performed or the quality of evidence was not attributable to the influencing factors (inclusion criteria 5).

The included systematic reviews assessed adherence influencing factors in the following conditions/medications; chronic malignant pain, Parkinson disease, heart failure, inflammatory arthritis, prophylactic treatment of hemophilia A or B, intake of oral anti-cancer agents [11‐15, 17]. Sinoott et. al did not restrict the condition or medication but included all studies that analysed public insured patients which were exposed to co-payments for medications [16]. Overlaps were not assessed, because all included systematic reviews were on different indications. The characteristics of the included systematic reviews are presented in Table 1.

Overall the quality of included systematic reviews was moderate to high One systematic review satisfied all quality criteria [16] and two did not fulfil only one quality criterion [12, 17]. At most three items were not answered with “no” or “unclear” (three studies [11, 14, 15]). In three systematic reviews the study selection and/or data extraction was not performed in duplicate or the item was rated as unclear because of missing information [11, 14, 15]. The most frequent flaw was that a list of excluded studies was not provided. The assessment questions relating to combing findings and publication bias were rated as not applicable in 6 studies because a quantitative data-synthesis was not performed. The quality of each systematic review is presented in Table 2.

The following factors were included in the analysis: age, comorbidity, co-payments, duration of disease, duration of therapy, education, ethnic status, gender, regime complexity (e.g. number of daily intake, number of tablets, frequency of intake, daytime of intake), marital status (including domestic partnerships), medication costs, different medications, social support, financial status/income, employment.

The results of individual systematic reviews are presented in Additional file 2 and the evidence synthesis of individual systematic reviews is presented in Table 3.

The evidence for effect of education was mostly judged as unclear. Only in chronic malignant pain there is a tendency that education has effect on adherence [11]. However the effect direction was not reported in this systematic review. In the systematic reviews that reported effect direction, education showed a positive influence on adherence [13, 17].

Employment was only analysed in the systematic review on inflammatory arthritis [14]. There is a tendency of effect but the effect direction was not reported.

In Parkinson diseases there is the tendency that ethnic minorities are less adherent [12]. In inflammatory arthritis whites seem to be more adherent [14].

Financial status seems to have no effect on adherence. The three systematic reviews that analysed this outcome showed no effect, the tendency of no effect, or unclear effect respectively [12, 13, 17].

The effect of social support was unclear in both systematic reviews that analysed this exposure [13, 14].

Duration of therapy was only analysed in the systematic review on oral anticancer agents [17]. The effect direction was negative. But the evidence for effect was unclear.

The effect direction of frequency of intake was negative in one primary study on patients with Parkinson diseases [12]. However, the study was not statistical significant. There is no effect of frequency of intake on adherence in patients with heart failure or inflammatory arthritis patients [14].

The effect of tablets taken per day in heart failure patients is unclear, because one of the two studies that analysed this factor was statistical significant and one was not. Moreover effect direction was not reported [13].

Most of the systematic reviews that analysed intake of different medications showed a negative effect on adherence (Parkinson, oral anticancer therapy, inflammatory arthritis) [12, 14, 17]. In patients with chronic malignant pain the effect direction was heterogenic for this factor [11]. For taking different medications there was some evidence for effect in heart failure patients. But effect direction was unclear [13].

Taking medications at meals showed a positive effect direction on adherence in patients taking oral anticancer agents. The evidence for effect was unclear [17].

Duration of diseases seems to have no influence on adherence in patients with chronic malignant pain and tendentially not in patients with inflammatory arthritis [11, 14]. For patients taking oral anticancer agents the effect direction was negative, but evidence for effect was unclear [17].

With one exception [17], all studies showed a positive effect of higher age or middle age on adherence. There was a tendency for effect in patients with Parkinson and heart failure [12, 13]. In patients with inflammatory arthritis the effect of higher age was unclear. Patients aged between 35–56 showed higher adherence than other age groups in one study [14]. For the other indications (chronic malignant pain, hemophilia, oral anticancer therapy) the evidence of effect was unclear [11, 15, 17].

The evidence for effect of comorbidity was mostly unclear [12‐14]. Only in heart failure patients there was the tendency of negative effect of mental comorbidities [13].

The effect direction of gender was heterogeneous in patients with heart failure and patients on oral anticancer therapy [13, 17]. In patients with chronic malignant pain there was the tendency of higher adherence in women [11].

Co-payments and medication costs showed a negative effect direction in all systematic reviews. For both factors there was a tendency of effect in inflammatory arthritis patients and unclear effect in patients taking oral anticancer agents [14, 17]. The meta-analysis on co-payments in public insurance systems showed a clear negative effect [16].