Evidenced-based medicine views random-assignment clinical trials as the gold standard of evidence. Because patient populations are heterogeneous, large numbers of patients must be studied in order to achieve statistically significant results, but the means or medians of these large samples have weak predictive validity for individual patients. Further, the logic of random-assignment clinical trials allows only the inference that some subset of patients benefits from the treatment. Post-hoc analysis is therefore necessary to identify those patients. Otherwise, many patients may receive treatments that are useless and potentially harmful.