Reporting of systematic reviews and meta-analysis of observational studies

Abstract

Although randomized controlled trials (RCTs) are generally seen as providing more certainty about the evidence of an effect, many interventions in infectious diseases are examined in observational studies. These include rare cases of treatment that cannot be examined in a RCT (e.g. appropriate vs. inappropriate empirical antibiotic treatment) [1,2], complex interventions that are difficult to examine in RCTs (e.g. infection control or antibiotic stewardship) [3,4], comparative effective research and quality improvement programmes (comparisons between commonly used, approved, antibiotics, monotherapy vs. combinations or treatment durations) for which resources for RCTs are lacking [5e10], or studies rapidly launched during outbreaks, such as the coronavirus disease 2019 (COVID-19) pandemic [11e13]. Summarizing these in systematic reviews is appealing; however, systematic reviews of observational studies suffer from the same shortcomings as primary observational studies when causal inferences need to be made. Consistent errors in observational studies may even be magnified by narrowing the confidence intervals through meta-analysis.In this editorial note, we provide our recommendations and expectations for systematic reviews and meta-analyses of observational studies examining an exposure or intervention. Guidance on performance and reporting of such studies has been published, and we recommend following these [14,15]. We explain the specific pitfalls of such studies and what we require from authors of systematic reviews of observational studies.

Effect of calcium-channel blockers on the risk of active tuberculosis and mortality: systematic review and meta-analysis

“…Therefore, our results are robust and consistent with the available primary studies. One issue with pooling raw effect sizes in a meta-analysis of non-randomized observational studies is that it provides no more information than a univariate analysis of the original observational studies ( Liu et al, 2017 ; Paul and Leeflang, 2021 ). The Cochrane Handbook advises adopting the adjusted model estimate with the greatest number of confounding variables ( Higgins et al, 2023 ), as combining uncorrected data may result in the observation of a significant impact that may be diminished or even eliminated when controlled for these covariates ( Higgins et al, 2003 ).…”

Adverse Effects of Dipyrone (Metamizole) Use During Pregnancy on Offspring Health: A Systematic Review and Meta-Analysis

“…The outcomes are reported as dichotomous data and were compared using odds ratios (ORs) with 95% con dence intervals (CIs). The adjusted ORs (aORs) were extracted from studies to minimize confounding factors 13 . Heterogeneity was evaluated with tau and I 2 statistics; p values less than 0.10 and I 2 values > 25% were considered signi cant for heterogeneity.…”

Predisposing factor, impactful adjunctive treatment or unfavorable prognostic marker: a meta-analysis on the elusive role of dexamethasone for Listeria monocytogenes meningitis

“…At once, it represents the first evidence synthesis on the specific topic of the use of AD for neurolisteriosis, presenting a pooling of raw and adjusted data, as recommended when meta-analyzing observational studies [17].…”