Chapter 4 60 Condition-related, and 5. Health system factors [6] using the description of Sabaté [5]. This categorization was chosen, because the WHO’s systems model aims to analyze and provide explanations for non-adherence on a societal and health policy level in a broader sense. In this way, all identified prognostic factors could be placed in an appropriate domain [6]. In case relevant information was missing, we contacted authors via e-mail to obtain the missing information. Risk of bias Assessment Risk of bias was assessed at the study level. Consistent with the Cochrane Collaboration's recommendation, we used the Quality in Prognosis Studies (QUIPS) tool [21]. The tool was used as described in the manual [21]. Statistical Analysis Data preparation To enable pooling of all relevant studies, a single measure of effect size and an indication of the precision of the effect size were required. The most common effect size metric reported were odds ratios. Where another effect size was reported we converted this to the odds ratio metric [22]. Estimates were derived along with their 95% confidence intervals (CI) and p-values. In the absence of confidence intervals or standard errors, we calculated them using the formulas described by Altman and Bland [22]. Where appropriate, direction of effect was converted for consistent reporting (i.e., showing associations between variables and non-adherence rather than adherence). The primary analysis was structured in that the prognostic factors were already categorized into the five WHO-domains as described by Sabaté [5]. A priori considerable clinical heterogeneity was expected in patients, specifically in type of chronic disease. Where relevant, this heterogeneity was addressed by making subgroups based on type of chronic disease (‘cancer’ vs. ‘other diseases’). Methodological heterogeneity was expected based on the variables associated with adherence assessed in the different included studies. After organizing the results, and pooling, we further explored possible sources of heterogeneity. Pooling methods Where relevant, we protocoled a random-effects model to pool the overall effect of each prognostic factor found in one of the five domains. The pooling method used was the Inverse-variance, and the Paule-Mandel procedure to estimate the between-study heterogeneity (t2) [22]. In case an original study did not provide adequate information to extract or calculate an effect size and relevant data could not be obtained from the authors, then this study could not be considered in the meta-analysis.
RkJQdWJsaXNoZXIy MjY0ODMw