216 Chapter 10 Study selection, data extraction, and quality assessment We included studies that met the following conditions: (1) randomized or observational study evaluating outcomes in adults with SAB, (2) included 200 or more patients, (3) reported mortality at or before 90 days following SAB, and (4) reported mortality stratified by sex. Exclusion criteria were studies on specific subpopulations (eg, dialysis, intensive care unit, hematological or oncological patients), studies that included SAB patients as a subgroup (eg, patients with bacteremia by any microorganism) that did not report SAB-specific data, and studies using (partially) the same cohort as another study included in this review. In this latter scenario, the study with the largest cohort was included. Titles and abstracts of articles (with authors and institutions visible) identified through our primary search were screened independently by two reviewers (A.W. reviewed all; R.K., M.W., J.K., F.R., J.P., S.M., S.K., M.L., V.F., and J.T. were second reviewers). Conflicts at this stage were resolved by a third person. Articles marked for full-text review underwent full-text screening by two independent reviewers. Conflicts at this stage were resolved by consensus or by obtaining a third reviewer’s opinion when consensus could not be reached. Data extraction and quality assessment was done by one reviewer and verified by a second reviewer. Extracted variables included lead author, journal, year of publication, start and end year of inclusion, country, aim of study, study design, number of hospitals, number of patients, population description, and whether methicillin-resistant S aureus (MRSA), methicillin-susceptible S aureus (MSSA), or both were addressed. Unadjusted mortality stratified by sex was extracted, as well as adjusted mortality when reported, the statistical model and the covariates for which mortality was adjusted. If a study described mortality for two subgroups (eg, for MSSA and MRSA bacteremia separately), both were included. Risk of bias and quality were assessed with the Newcastle-Ottawa Quality Assessment Scale [11] (eAppendix 2 in Supplement 1) because only observational studies were identified. Statistical analysis Mortality data were combined as odds ratios (ORs). If ORs were not reported in a study, we calculated ORs from raw mortality by sex if such data was available. If raw data was not available either, then ORs were calculated from the provided risk ratio (RR) or hazard ratio (HR) values based on previously published methods [12,13]. In the single study that reported a rate ratio [14], this rate ratio was used to estimate the OR [15]. Sensitivity analyses involving only studies that directly reported an OR (as opposed to estimating OR based on HR or RR) were conducted. ORs were combined using inverse variance with random effects models. We used the Knapp and Hartung method to adjust the standard errors of the estimated coefficients [16,17]. Robustness of findings were assessed through influence and sensitivity analyses as detailed in the
RkJQdWJsaXNoZXIy MTk4NDMw