Inference for Heterogeneous Treatment Effects with Deep Survival Learners

Conditional average treatment effects (CATEs) provide a principled framework for assessing treatment-effect heterogeneity and identifying patients who may benefit from an intervention. For survival outcomes, CATEs may also evolve over time and pose challenges for estimation and inference. Existing methods can accommodate complex covariate heterogeneity but often fail to exploit shared temporal structure or provide valid inference for time-varying treatment effects and their contrasts. We propose a deep survival learning framework that combines neural networks for nonparametric covariate effects with a shared spline representation of temporal variation, enabling joint estimation of CATE trajectories. Using doubly robust pseudo-outcomes and sample splitting, we establish nonasymptotic oracle bounds accounting for network and temporal approximation, nuisance estimation, and optimization errors. Our theory shows that exploiting shared temporal structure yields sharper error bounds for temporal contrasts than those derived from separate time-specific estimators. For inference, we develop a subsampled ensemble procedure and establish asymptotic normality for pointwise CATEs, temporal contrasts, and covariate contrasts, including temporal contrasts over shrinking time intervals. Under suitable regularity and bias-control conditions, consistent infinitesimal jackknife variance estimators yield asymptotically valid confidence intervals. Simulations evaluate estimation accuracy, interval coverage, and sensitivity to temporal dimension and basis mismatch. An application to lung cancer survival data illustrates the framework's ability to characterize heterogeneous treatment benefits over time.

Authors

Publication Details

Journal
arXiv (Cornell University)
Published
2026-11-17
DOI
https://doi.org/10.48550/arxiv.2611.23090
Primary Topic
Advanced Causal Inference Techniques
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Inference for Heterogeneous Treatment Effects with Deep Survival Learners

Sattwik Ghosal, Yuming Sun, Yi Li, Xuran Meng
arXiv (Cornell University)
Advanced Causal Inference Techniques
preprint

Inference for Heterogeneous Treatment Effects with Deep Survival Learners

Sattwik Ghosal, Yuming Sun, Yi Li, Xuran Meng
preprint en

Abstract

Conditional average treatment effects (CATEs) provide a principled framework for assessing treatment-effect heterogeneity and identifying patients who may benefit from an intervention. For survival outcomes, CATEs may also evolve over time and pose challenges for estimation and inference. Existing methods can accommodate complex covariate heterogeneity but often fail to exploit shared temporal structure or provide valid inference for time-varying treatment effects and their contrasts. We propose a deep survival learning framework that combines neural networks for nonparametric covariate effects with a shared spline representation of temporal variation, enabling joint estimation of CATE trajectories. Using doubly robust pseudo-outcomes and sample splitting, we establish nonasymptotic oracle bounds accounting for network and temporal approximation, nuisance estimation, and optimization errors. Our theory shows that exploiting shared temporal structure yields sharper error bounds for temporal contrasts than those derived from separate time-specific estimators. For inference, we develop a subsampled ensemble procedure and establish asymptotic normality for pointwise CATEs, temporal contrasts, and covariate contrasts, including temporal contrasts over shrinking time intervals. Under suitable regularity and bias-control conditions, consistent infinitesimal jackknife variance estimators yield asymptotically valid confidence intervals. Simulations evaluate estimation accuracy, interval coverage, and sensitivity to temporal dimension and basis mismatch. An application to lung cancer survival data illustrates the framework's ability to characterize heterogeneous treatment benefits over time.

arXiv (Cornell University)
Advanced Causal Inference Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon