Agreement of treatment effects from observational studies using causal modelling and randomised trials: Meta-epidemiological study

Ewald H1, Ladanie A1, Ioannidis JP2, Mc Cord K1, Bucher HC3, Hemkens LG3
1Basel Institute for Clinical Epidemiology and Biostatistics (ceb), University Hospital Basel, University of Basel; Swiss Tropical and Public Health Institute, Basel, 2Stanford Prevention Research Center; Meta-Research Innovation Center at Stanford (METRICS); Department of Health Research and Policy; Department of Biomedical Data Science; Department of Statistics; Stanford University, CA, 3Basel Institute for Clinical Epidemiology and Biostatistics (ceb), University Hospital Basel, University of Basel

Background:Randomised controlled trials (RCTs) are not available for many important healthcare questions. Observational studies using causal modelling such as marginal structural models (MSM) are increasingly proposed as useful alternative. A recent Cochrane review (1), comparing observational studies with randomised trials, found no empirical analysis of observational studies using causal modelling.

Objectives:To evaluate the agreement of treatment effects estimated by observational studies with causal modelling using MSM with effects of RCTs on the same clinical question.

Methods: In a comprehensive meta-epidemiological study, we included any observational study comparing any defined treatment with any comparator providing an MSM-based effect estimate on any binary outcome. We identified 100 eligible studies via PubMed (last search October 2014), supplemented by screening of citations of key references of causal inference literature. In each eligible MSM-study, we identified any clinical question with a reported MSM-based treatment effect estimate and conducted systematic, peer-reviewed searches on PubMed (last search April 2016), supplemented by citation screenings, for RCT evidence on the same clinical question. Multiple RCTs were combined with random-effects meta-analyses to obtain one summary odds ratio for each clinical question. We then compared the direction of treatment effects, effect sizes and confidence intervals between MSM-studies and RCTs and used the ratio of odds ratios approach to evaluate the overall relationship of causal modelling effects and RCT results. We conducted several sensitivity analyses to explore effect modifications, in particular by risk of bias, mortality/non-mortality outcomes, active/passive comparators, missing data, and research topic.

Results:Results will be available at the time of the Summit.

Conclusions:Results will indicate whether observational studies using causal modelling give different answers than randomised-controlled trials that evaluate the same clinical question.

1. Anglemyer A, et al. Healthcare outcomes assessed with observational study designs compared with those assessed in randomized trials.