The detection of adverse events in randomized clinical trials: can we really say new medicines are safe?

Background: While it is well known that randomized controlled trials (RCTs) are usually designed with sufficient sample size and power to detect the efficacy but not safety of a medicine, the extent to which RCTs quantify safety has not been well ascertained.

Purpose: The aim of this study was to assess the safety data available for five commonly prescribed medicines at the time of marketing.

Methods: Published RCTs for five medicines risperidone, sertraline, donepezil, strontium ranelate and tramadol extended release were identified. All adverse events (AEs) in the trials were independently extracted by two clinical researchers. Using the sample size in the trials, the power to detect the observed difference in AEs rates between the treatment and placebo groups was calculated. A power of 80% or more was deemed adequate to detect AEs; studies with power of < 80% were deemed insufficiently powered to detect AEs.

Results: 12 RCTs were identified. Six trials were insufficiently powered to detect any of the potential AEs reported. Of the 150 evaluated AEs, the trials were insufficiently powered to detect 81% (122/150) of the AEs reported. For the adverse events that were detected with adequate powered clinical trials, only 53% (10/19) of potentially very common AEs (≥10%) and 17% (18/106) of potentially common AEs (1%-<10%) were identified.

Conclusion: Trials are insufficiently powered to detect the majority of adverse events that are reported in clinical trials, even for common adverse events. Observations other than primary efficacy endpoints such as AEs that are not prespecified with adequate power should be treated as hypothesis generating only and not justification of evidence. Claims of safety based on trial evidence not designed for the safety endpoint are often premature.