Accounting for Differences Among Patients in the FDA Approval Process

The FDA employs an average-patient standard when reviewing drugs: it approves a drug only if the average patient (in clinical trials) does better on the drug than on control. It is common, however, for different patients to respond differently to a drug. Therefore, the average-patient standard can result in approval of a drug with significant negative effects for certain patient subgroups (false positives) and disapproval of drugs with significant positive effects for other patient subgroups (false negatives). Drug companies have a financial incentive to avoid false negatives. After their clinical trials reveal that their drug does not benefit the average patient, they conduct what is called post hoc subgroup analysis to highlight patients that benefit from the drug. The FDA rejects such analysis due to the risk of spurious results. With enough data dredging, a drug company can always find some patients that benefit from their drug. This paper asks whether there workable compromise between the FDA and drug companies. Specifically, we seek a drug approval process that can use post hoc subgroup analysis to eliminate false negatives but does not risk opportunistic behavior and spurious correlation. We recommend that the FDA or some other independent agent conduct subgroup analysis to identify patient subgroups that may benefit from a drug. Moreover, we suggest a number of statistical algorithms that operate as veil of ignorance rules to ensure that the independent agent is not indirectly captured by drug companies. We illustrate our proposal by applying it to the results of a recent clinical trial of a cancer drug (motexafin gadolinium) that was recently rejected by the FDA.