Primary Submission Category: Machine Learning and Causal Inference
Incremental Causal Effects for Time to Treatment
Authors: Zhichen Zhao, Zhichen Zhao, Andrew Ying, Ronghui (Lily) Xu,
Presenting Author: Zhichen Zhao*
We consider time to treatment initiation, which commonly arises in preventive medicine, such as disease screening and vaccination, and in non-fatal health conditions, such as HIV infection prior to AIDS onset. While traditional causal inference has focused on deterministic interventions that assign treatment according to fixed rules, including whether or when treatment is assigned and allowing dependence on subject characteristics, we study the incremental causal effect of intervening on the intensity of treatment initiation.
We establish identification and derive the efficient influence function for this effect without requiring the commonly assumed positivity condition. Building on this characterization, we propose efficient nonparametric estimators based on augmented inverse probability weighting that can attain fast convergence rates while accommodating flexible machine learning estimation of nuisance functions. We also develop general efficiency theory for the proposed estimators.
We illustrate the finite-sample performance of our methods through simulation studies and apply them to a rheumatoid arthritis study to evaluate the incremental effect of time to methotrexate initiation on joint pain, as well as to a Norwegian women’s study to evaluate the incremental effect of time to subsequent HPV testing on detection of cervical intraepithelial neoplasia grade 2 or 3 (CIN2+).