Background: Evidence for bias in the results of studies using nonrandomised allocation has previously been examined by comparing the results of non-randomised with randomised studies of the same intervention, and accumulating such comparisons across many interventions. Previous studies have been criticised for being subject to selection bias (by not systematically sampling clinical topics), and as potentially confounded due to other differences in study design, interventions, participants and outcomes between the randomised and non-randomised studies. Objectives: To quantify the bias in the results of non-randomised studies (NRS) compared with randomised controlled trials (RCTs), while avoiding biased selection of topics and minimising the potential for confounding. Methods: We did a meta-epidemiological study, pooling comparisons of RCTs with NRS across clinical topics. We identified topics by randomly selecting RCTs from the Cochrane Central Register of Controlled Trials (CENTRAL), then locating an NRS addressing the same topic. RCTs reporting all cause mortality as an outcome were selected, as this outcome is the most objective and least likely to be affected by other aspects of study design. All RCTs and NRS addressing the topic were then located and meta-analysed. Ratios of odds ratios (ROR) were calculated for each topic and then combined using random-effects meta-analysis. Results: Preliminary findings are based on the first six topics where both RCTs and NRS were found (54 RCTs and 27 NRS). Overall, intervention effect estimates tended to be more beneficial in NRS (ROR 0.90, 95% CI 0.77 to 1.06), with some evidence of between-meta-analysis heterogeneity in bias (I² = 0.2) (Figure). Results based on 20 comparisons will be presented. Conclusions: Compared with previous studies, our approach minimised selection bias via random sampling of clinical topics. These preliminary results suggest that NRS overestimated treatment effects by 10% on average, compared with RCTs. The precision of our findings will be improved by the time of the Colloquium, when complete results will be available.