Sample Size and Effect Estimates in Randomized Trials in Pediatric Critical Care: a Methodologic Review
Duffett M, Choong K, Hartling L, Menon K, Thabane L, Cook DJ.
Aims & objectives
Randomized controlled trials (RCTs) are ideally adequately powered to detect an important difference in the primary outcome. To describe the methods and reporting of sample size estimates of RCTs of children in pediatric critical care.
We included published English-language trials from the Evidence in Pediatric Intensive Care database (epicc.mcmaster.ca) of RCTs administering any intervention to children. We excluded trials conducted in pre-term infants, and cross-over trials.
101 (40%) of 251 RCTs published between 1986 and 2013 reported the intended sample size and some detail of their sample size estimation. 18 (18%) reported or referenced the formula used for sample size calculations and 7 (7%) reported the software used. 42 (42%) cited published data upon which their assumptions were based. The effect size actually observed was smaller than that expected in the sample size calculation in 30 (73%) of the 41 RCTs with binary primary outcomes that reported the expected effect size. The median (IQR) expected relative risk was 2.00 (1.67, 2.67) and the observed relative risk was 1.32 (0.97, 1.74) p=0.002. 42 (17%) of trials reported conducting interim analyses, 76% of which were planned a priori. Of the 25 RCTs (25% of the trials reporting a sample size calculation) that were stopped early, 8 (32%) described a stopping rule.
Reporting of sample size estimation in pediatric critical care RCTs remains sub-optimal. Researchers frequently over-estimate the expected treatment effect when planning RCTs. Reporting transparency needs to be improved.