17 Prevalence of secondary prevention medication use in South America 2 A tool for the quality assessment of studies reporting prevalence estimates was adapted from the previous work by Zhao et al (13), and Li et al (18) (Supplementary File 3A). For overall risk of bias assessment, we summarized risk of bias as follows: for risk of bias in each domain (study design, study population, participation rate, participants’ characteristics and outcome) 2 points were given for low risk, 1 point for moderate risk and 0 points for high or unclear risk. Publications with a score lower than 6 out of 10 were excluded. The remaining publications were classified as following: moderately low risk of bias (6-7 points), low risk of bias (8-9 points) and very low risk of bias (10 points). Reviewers ED and AMF assessed the quality of articles for which they extracted the data, and additionally they assessed the quality of 10% of the articles examined by the other reviewer. Discrepancies were discussed until consensus was achieved. Data analysis Data analysis was conducted using R Studio (19). Data on medication prevalence is expressed in percentages by class of medication. We reported the prevalence of each kind of medication separately. We presented separate categories for those articles reporting general classes of medications instead of specific drugs. In the case of antiplatelet drugs, we additionally showed the estimates of articles reporting the prevalence of aspirin, clopidogrel, and not-specified antiplatelet drugs combined because of the shared indication for these medications, and to be able to explore the use of this medication class in general. Meta-analysis was performed using a mixed model from the R package “metafor” (20) for each class of medication. The results were expressed as pooled prevalence with 95% confidence intervals (CI) and random effects, and displayed in forest plots by care setting. Heterogeneity was quantified with the I2 test. The same statistical package was used in a sensitivity analysis to analyse potential differences in prevalence between studies conducted in Brazil and in other countries. In order to explore time trends in medication use, mixed meta-regression models were fitted with the starting year of the study as covariate for each medication class. The reported prevalence of medication and the model prediction were plotted against the year the studies commenced using bubble plots to illustrate time-trends in medication use. Meta-regression models were performed to discern potential factors contributing to medication use. A mixed meta-regression model was run for each class of medication, including the following covariates: the proportion of women included, time of outcome measurement since the start of the study, diagnosis of the patients included in the study, urban region, and care setting. The full models were reduced and simpler models were compared against the full models and among them with the AIC fitting statistic. Models with the lowest AIC were selected. Results were expressed as odds ratios (OR) and 95% CI. RESULTS Study selection The search strategy resulted in 7388 publications: 2660 in LILACS, 1810 in Embase, 1538
RkJQdWJsaXNoZXIy MTk4NDMw