Targeted Learning of Causal Effects with Continuous Time-to-Event Outcomes

• Activity Number: 537 - Recent Developments in Causal Inference with Real World Evidence in Drug Development
• Type: Invited
• Date/Time: Thursday, August 11, 2022 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Mark Van der Laan*
• Companies: University of California Berkeley

Abstract

Targeted maximum likelihood estimation (TMLE) provides a general methodology for estimation of causal parameters in presence of high-dimensional nuisance parameters. Generally, TMLE consists of a two-step procedure that combines data-adaptive nuisance parameter estimation with semiparametric efficiency and rigorous statistical inference obtained via a targeted update step. In this talk we demonstrate the practical applicability of TMLE for standard survival and competing risks settings where event times are not confined to take place on a discrete and finite grid. We demonstrate TMLE updates that simultaneously target point-treatment specific survival curves and or treatment-cause-specific subdistributions in the competing risk setting, across treatment and time-points. We consider the case that we only observe baseline covariates as well as the case that we also track time-dependent covariates that potentially inform censoring/drop-out. This results in estimates that are not only fully efficient, but also respect the natural monotonicity of survival functions and cause specific subdistributions and makes sure that the sum of subdistributions and survival equals 1. We propose a su