Beta, or type II error in psychiatric controlled clinical trials

Abstract

Fifty-two reportedly 'negative' psychiatric controlled clinical trials from 1980-1984 were re-examined to note if large enough samples had been studied to give a high probability (greater than 0.90) of detecting a 20% or 50% improvement in therapeutic response, defined as a reduction in the non-response rate. Forty-four of the trials had a larger than 10% risk of missing a true 20% difference in non-response rates, and with the same risk, 25 of the trials could have missed a 50% difference between treatments. Based on estimates of 90% confidence intervals, a potential 20% difference was possible for 47, and 30 had a potential true difference of 50% in non-response rates between experimental and control conditions. Far more reports describing 'non-significant' studies could not be re-examined, due to lack of reporting part or all of the relevant results. The possibility of missing considerable therapeutic improvement, or important clinical differences, because of small sample size demands more concern in psychiatry