Sample size calculations in acute stroke trials: A systematic review of their reporting, characteristics, and relationship with outcome

Background and Purpose-Only a few randomized controlled trials in acute stroke have shown a treatment-related benefit. Inadequate trial design, especially low sample size, may partly explain this failure. We investigated sample size calculations (SSCs) in a systematic review of acute stroke trials. Methods-Full reports of nonconfounded randomized controlled trials that recruited patients within 1 week of stroke onset and were published before the end of 2001 were identified from the Cochrane Library and other bibliographic databases. Information on the SSC and outcome event rates was collected for each trial. Results-Of 189 identified trial reports, 57 (30%) reported greater than or equal to1 components of the SSC, phase II 14/129 (11%) versus phase III 43/60 (72%) (P<0.001), with 32 (56%) giving all the required parameters. Significance (alpha) was mentioned in 54 (96%) reports; 53 used a significance level of alpha=0.05. And 55 (98%) reports gave the power (1-beta) of the study (median [25th and 75th percentile] 0.80 [0.80, 0.90]). The anticipated percentage of control subjects having a primary outcome event was given in 24 (42%) articles: case fatality 21.8% (11.8%, 23.5%, n=4) and combined death or disability/dependency 55.5% (44.5%, 66.3%, n=20); 25 studies used other outcomes and 8 studies gave insufficient information. Four of the 22 trials achieved a control rate within 5% of their prediction. 49 (86%) reports gave the anticipated treatment effect; case fatality: anticipated 9.5% (1.1%, 12.5%, n=6), achieved -0.3% (-4.1%, +2.4%); combined death or disability/dependency: anticipated 13.0% (10.0%, 16.0%, n=25), achieved 1.8% (-0.5%, +5.4%). The median calculated sample size was 600 (198, 995, n=54). Conclusions-Too few trial publications report the assumptions underlying their SSC. Most trials were underpowered, ie, power <0.90, used inappropriate assumptions for event rates, and were grossly overoptimistic in their expectation of treatment effect. These deficiencies will together have resulted in trials being far too small and reduced their chance of being able to detect real treatment effects.