Causal effect on the number of life years lost due to a specific event: Average treatment effect and variable importance

TL;DR

This paper introduces a causal framework for estimating the impact of specific events on life years lost, utilizing machine learning for efficient, model-agnostic estimators, with applications in biostatistics and antidepressant response analysis.

Contribution

It develops new estimators for causal effects on life years lost, integrating machine learning and providing a variable importance measure for biostatistical time-to-event data.

Findings

Estimators are asymptotically normal and efficient.

Simulation studies demonstrate good performance.

Application to antidepressant data illustrates practical utility.

Abstract

Competing risk is a common phenomenon when dealing with time-to-event outcomes in biostatistical applications. An attractive estimand in this setting is the "number of life-years lost due to a specific cause of death", Andersen et al. (2013). It provides a direct interpretation on the time-scale on which the data is observed. In this paper, we introduce the causal effect on the number of life years lost due to a specific event, and we give assumptions under which the average treatment effect (ATE) and the conditional average treatment effect (CATE) are identified from the observed data. Semiparametric estimators for ATE and a partially linear projection of CATE, serving as a variable importance measure, are proposed. These estimators leverage machine learning for nuisance parameters and are model-agnostic, asymptotically normal, and efficient. We give conditions under which the estimators are asymptotically normal, and their performance are investigated in a simulation study. Lastly, the methods are implemented in a study concerning the response to different antidepressants using data from the Danish national registers.