All Times EDT
Keywords: causal inference; subgroup; odds ratio; hazard ratio; marginal; conditional
Targeted therapies tend to have biomarker defined subgroups that derive differential efficacy from treatments. This presentation shows that using efficacy measure such as odds ratio and hazard ratio can potentially deprive patients from efficacious therapies, even from a randomized clinical trial. We prove this negative phenomenon analytically and demonstrate it with the OAK bTMB study. On the other hand, we prove positively that using efficacy measures such as relative response and ratio of time will provide logic-respecting causal inference, so long as inference is in accordance with the (newly proposed) Subgroup Mixable Estimation (SME) principle.
SME proceeds theoretically and is coded by mixing within each treatment arm first and then calculating efficacy for their mixtures. This guarantees marginal and conditional efficacy agree, for logic-respecting efficacy measures (be it a ratio or a difference), no matter the outcome is continuous, binary, or time-to-event. One does not have to choose between marginal and conditional.