Overall bias methods and their use in sensitivity analysis of Cochrane reviews were not consistent.

OBJECTIVE: The objective of the study was to analyze methods of assessing "overall bias" in Cochrane reviews of interventions published in the Cochrane Database of Systematic Reviews and sensitivity analyses related to overall risk of bias (RoB). STUDY DESIGN AND SETTING: From Cochrane reviews published within 3 years, from July 2015 to June 2018, we extracted data regarding methods of judging overall bias for a single trial, as well as details regarding methods used in frequency of RoB in sensitivity analyses. RESULTS: Of the 1,452 analyzed Cochrane reviews, 409 mentioned assessment of overall RoB on a study level. In 107 reviews, authors clearly specified key domains that determined the overall RoB, whereas in the remaining reviews, assessment of overall bias was not in line with the Cochrane Handbook. Among 268 Cochrane reviews that had any RoB-related sensitivity analysis, in 56 (21%) reviews, the authors reported a significant change for at least one outcome compared with the initial analysis. CONCLUSION: Highly heterogeneous approaches to summarizing overall RoB on a study level and using RoB for sensitivity analyses may yield inconsistent and incomparable results across Cochrane reviews.

Selective reporting bias due to discrepancies between registered and published outcomes in osteoarthritis trials.

Aim: Outcome reporting bias (ORB) occurs when outcomes planned in a study protocol are subsequently not reported or are partially reported. Our aim was to analyze ORB in randomized controlled trials (RCTs) about conservative interventions for osteoarthritis (OA) by comparing registered protocols and published manuscripts, as well as association between study funding type and intervention type, and ORB in those RCTs. Materials & methods: We analyzed RCTs that were published in a peer-review journal and analyzed any type of conservative intervention for treatment of OA in humans that reported in the manuscript registration in a public clinical trial registry and provided unique registration identifier. We extracted data indicating ORB by comparing outcomes in protocol and published article, and characteristics of trials. Results: In 190 (57%) of 334 included RCTs, it was indicated in the manuscript that a trial was registered. In 48% of trials we found discrepancies in number, type or time point of primary efficacy outcome between protocol and manuscript. Significantly less discrepancies in primary efficacy outcomes between protocols and published articles were found in trials funded by a commercial sponsor (p = 0.0062) and trials of pharmacological interventions (p = 0.0016). Conclusion: Trials about conservative therapies for OA have high prevalence of discrepancies between protocol and publication, and frequent ORB. This may mislead readers of published results because it has been shown that ORB can lead to both overestimation and underestimation of effects of interventions, depending on the intervention and outcome. Efforts to prevent nonregistration of protocols and selective reporting are needed.