OBJECTIVE: To study whether sample size of randomized trials has increased over recent years. DESIGN: A systematic review of randomized trials published in 1976, 1981, 1986, or 1991 which were double-blind, had active treatments in both arms, were not of a crossover design, were published in English as full articles, and which had clinical outcomes. RESULTS: We included 386 references. The median total sample size increased from 46 in 1976 to 71 in 1991 (p<0.0001). Sample size was not related to journal impact factor (p = 0.40). The median sample size and impact factor for 106 trials in gastroenterology, cardiology or oncology were larger than for other specialties, 83 vs 60 (p = 0.01) and 1.5 vs 1.2 (p = 0.04), respectively. The use of binary outcomes increased with time (p = 0.00001) as did the proportion of trials with significant results (p = 0.001). CONCLUSION: Although increased, most sample sizes are still too small since several hundred patients are needed to be reasonably sure not to overlook a 25% improvement over standard therapy. A more profound change in sample size could be obtained if the bodies responsible for approving trials rejected small trials, apart from exceptional circumstances, such as very rare diseases.
OBJECTIVE: To study whether sample size of randomized trials has increased over recent years. DESIGN: A systematic review of randomized trials published in 1976, 1981, 1986, or 1991 which were double-blind, had active treatments in both arms, were not of a crossover design, were published in English as full articles, and which had clinical outcomes. RESULTS: We included 386 references. The median total sample size increased from 46 in 1976 to 71 in 1991 (p<0.0001). Sample size was not related to journal impact factor (p = 0.40). The median sample size and impact factor for 106 trials in gastroenterology, cardiology or oncology were larger than for other specialties, 83 vs 60 (p = 0.01) and 1.5 vs 1.2 (p = 0.04), respectively. The use of binary outcomes increased with time (p = 0.00001) as did the proportion of trials with significant results (p = 0.001). CONCLUSION: Although increased, most sample sizes are still too small since several hundred patients are needed to be reasonably sure not to overlook a 25% improvement over standard therapy. A more profound change in sample size could be obtained if the bodies responsible for approving trials rejected small trials, apart from exceptional circumstances, such as very rare diseases.