Poor agreement in significant findings between meta-analyses and subsequent large randomized trials in perioperative medicine

Background: The reliability of meta-analysis (MA) in predicting the findings of subsequent large randomized controlled trials (RCTs) has not been assessed in perioperative medicine and anaesthesia. Methods: Using Medline and PubMed, large RCTs (n >= 1000) published since 2000 in the anaesthesia and perioperative medicine/critical care literature were identified. All previous MAs of RCTs investigating the same intervention and population were sourced. For all reported major morbid endpoints common to each, results (significant/non-significant P<0.05) were compared. Results: 18 large RCTs and 44 prior MAs investigating the effects of 16 interventions were identified. Where endpoint results in the large RCTs were each compared with the single largest recent preceding MA, 35 of a total of 57 outcomes were predicted correctly by the MAs (61.4%). The odds ratio for a significant result from MA compared with the subsequent large RCT was 3.6, P=0.033 Bonferroni corrected. The positive predictive value of MAs was 22.7%; the negative predictive value was 85.7%, Kappa was 0.094 indicating slight agreement. The estimated power for each endpoint for large RCTs and MAs were similar, but the median study size for large RCTs was larger than that of the MAs, n=4,482 vs 1,389, P<0.0001. Conclusions: There was a strong tendency towards positive findings in MA not substantiated by subsequent large RCTs, which was not attributable to differences in study power. This finding suggests caution in clinical decision-making in anaesthesia and perioperative medicine based on findings of meta-analysis.