New DOOR causal framework: CVTMLE-SL wins for clinical trial benefit-risk

A new arXiv paper from a multi-institutional team (Yuan Feng, Shiyu Shu, Yixin Fang, Ionut Bebu, Toshimitsu Hamasaki, Scott Evans, and Guoqing Diao) introduces a unified covariate-adjusted causal inference framework for estimating the desirability of outcome ranking (DOOR) probability — a metric used to balance benefits and risks in randomized trials and observational studies. The approach expresses the DOOR probability as a bilinear functional of marginal ordinal outcome distributions under two treatment strategies, then estimates conditional ordinal distributions through sequential risk-set hazards. The authors derive the efficient influence function (EIF) for the DOOR probability, enabling robust inference.

In extensive simulations, the team compared G-computation, normalized inverse probability weighting (IPW), augmented IPW (AIPW), and targeted maximum likelihood estimation (TMLE), using generalized linear models or Super Learner (SL) for nuisance functions. TMLE-SL showed the strongest and most consistent point-estimation performance, with AIPW-SL ranking second. For inference, cross-fitted TMLE-SL (CVTMLE-SL) outperformed all alternatives across DOOR-scale bias, recovery of ordinal distributions, standard-error accuracy, and confidence-interval coverage. The methodology was illustrated using data from the multidrug-resistant organism network of the Antibacterial Resistance Leadership Group, demonstrating its practical value in real clinical settings.