Unfavourable outcomes of drug therapy--subjective probability versus confidence intervals

To investigate how individual predictions compare with confidence intervals, we asked 50 medical residents and 28 graduate students with biostatistics training to estimate unfavourable outcomes of drug therapy (therapeutic failures and side-effects) in groups of 10 and 20 patients. The predictions made by physicians and graduate students were similar for both sample sizes and types of outcome. The majority (58%) of estimates were greater than the upper limit of the 95% confidence interval, a bias that may hamper the correct interpretation of therapeutic risks in medical decision-making.