A new way to address missing data in late-stage clinical trials.

According to ICH E9(R1), defining the estimand comes before defining the analysis approach, and the strategies for addressing intercurrent events are key components of the estimand. With the composite strategy, the problem of missing data disappears, because it becomes part of the endpoint definition. It is this perspective that we adopt in addressing the problem of missing data. We propose comparing patients in a pairwise manner to determine which patient fared better, one patient from each group, taking into account the reason for and timing of missingness. For purposes of illustration, reasons for missingness are placed into three categories: (1) death or adverse events, (2) administration of rescue medication (treated as missing even if patient continues in the study, or a poor score is assigned), and (3) other reasons such as loss to follow-up or withdrawal of consent. Each pair of patients is compared on the endpoint of interest. The comparison outcomes are determined based on the presence of missing data, its category, and timing. For instance, if both patients in a pair have received rescue medication, the patient with the later time of rescue medication is considered to have fared better. The overall treatment effect is estimated from combining results across all pairwise comparisons. This method allows the reason and timing of missing data to contribute to the assessment of treatment effects, providing a solution to some limitations of existing methods. Thus, the pairwise comparison approach addresses the missing data problem transparently via the composite strategy.