Objective: To compare primary outcomes reported in publications, protocols and registries and to evaluate the contribution of available protocols to assess selective outcome reporting (SOR) as compared with registration alone.
Study design and setting: We included all randomized controlled trials (RCTs) published in 2015 and 2016 in the five leading general medical journals. For each RCT, we evaluated whether the protocol was available and searched for registration. We extracted all primary outcomes reported in publications, registries, and protocols. We evaluated whether SOR was suspected (i.e., at least one discrepancy in primary outcomes), unclear, or not suspected based on comparisons of publications and (1) trial registration alone or (2) protocols in addition to registration.
Results: Selective outcome reporting was suspected for 77/274 (28.1%), unclear for 30 (10.9%), and not suspected for 167 (60.9%) when comparing publications and trial registration alone. With protocols available, the classification changed for 38 RCTs (13.9%): 11 not suspected of SOR based on registration became suspected of SOR with protocols available, and 27 with unclear assessment based on registration became suspected of SOR (n = 7) and not suspected of SOR (n = 20) with protocols available.
Conclusions: Compared to registration alone, making protocols available allows for a more precise evaluation of SOR.
Keywords: Bias; Outcome; Protocols; Publishing/standards; Randomized controlled trial; Registries; Selective outcome reporting.
Copyright © 2018 Elsevier Inc. All rights reserved.