Magnitude of effects in clinical trials published in high-impact general medical journals

Methods We scrutinized systematic reviews of the Cochrane Database (Issue 4, 2009) and meta-analyses published in four general journals (2008-09). Eligible articles included epsilon 1 binary outcome meta-analysis(es) pertaining to effectiveness with epsilon 1 clinical trial(s) published in NEJM, JAMA or Lancet. Effect sizes in trials from NEJM, JAMA or Lancet were compared with those from other trials in the same meta-analyses by deriving summary relative odds ratios (sRORs). Additional analyses examined separately early- and late-published trials in HIG journals and journal-specific effects. Results A total of 79 meta-analyses including 1043 clinical trials were analysed. Trials in HIG journals had similar effects to trials in other journals, when there was large-scale evidence, but showed more favourable results for experimental interventions when they were small. When HIG trials had less than 40 events, the sROR was 1.64 [95% confidence interval (95% CI): 1.23-2.18). The difference was most prominent when small early trials published in HIG journals were compared with subsequent trials [sROR 2.68 (95% CI: 1.33-5.38)]. Late-published HIG trials showed no consistent inflation of effects. The patterns did not differ beyond chance between NEJM, JAMA or Lancet. Conclusions Small trials published in the most prestigious journals show more favourable effects for experimental interventions, and this is most prominent for early-published trials in such journals. No effect inflation is seen for large trials.