ORIGINAL CONTRIBUTION Empirical Evidence for Selective Reporting of Outcomes in Randomized Trials Comparison of Protocols to Published Articles

SELECTIVE PUBLICATION OF STUD-ies with statistically significantresults has received wide-spread recognition.1 In con-trast, selective reporting of favorable outcomes within published studies has not undergone comparable empirical investigation. The existence of out-come reporting bias has been widely suspected for years,2-12 but direct evi-dence is limited to case reports that have low generalizability13-15 andmay them-selves be subject to publication bias. Our study had 3 goals: (1) to deter-mine the prevalence of incomplete out-come reporting in published reports of randomized trials; (2) to assess the as-sociation between outcome reporting and statistical significance; and (3) to evaluate the consistency between pri-mary outcomes specified in trial pro-tocols and those defined in the pub-lished articles. METHODS In February 2003, we identified proto-cols and protocol amendments for ran-domized trials by reviewing paper files from clinical studies approved by the Scientific-Ethical Committees for Co-penhagen and Frederiksberg, Den