Differentially Private Learners for Heterogeneous Treatment Effects

2025-03-05

AI Summary

This paper addresses the problem of estimating conditional average treatment effects (CATE) from observational healthcare data under rigorous privacy constraints. We propose DP-CATE, the first unified framework that simultaneously satisfies Neyman orthogonality and ε-differential privacy. Methodologically, we design a Neyman-orthogonal loss function, integrate two-stage meta-learners (e.g., T-learner or R-learner) with arbitrary base models, and perform private CATE function release via regression in a reproducing kernel Hilbert space (RKHS). Strict ε-differential privacy is ensured through sensitivity analysis and calibrated noise injection. Experiments on synthetic and real-world healthcare datasets demonstrate that DP-CATE achieves asymptotically unbiased, robust, and statistically efficient CATE estimation. To our knowledge, this is the first work to jointly guarantee theoretical Neyman orthogonality and strict differential privacy for CATE estimation—bridging causal inference and privacy-preserving machine learning in sensitive healthcare applications.

Technology Category

• Machine Learning: Privacy
• Reasoning under Uncertainty: Causality
• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy

Application Category

• Security and Privacy: Data transparency and provenance
• Economics, Online Markets and Human Computation: Fairness, privacy, and diversity in economic environments
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

Patient data is widely used to estimate heterogeneous treatment effects and thus understand the effectiveness and safety of drugs. Yet, patient data includes highly sensitive information that must be kept private. In this work, we aim to estimate the conditional average treatment effect (CATE) from observational data under differential privacy. Specifically, we present DP-CATE, a novel framework for CATE estimation that is Neyman-orthogonal and further ensures differential privacy of the estimates. Our framework is highly general: it applies to any two-stage CATE meta-learner with a Neyman-orthogonal loss function, and any machine learning model can be used for nuisance estimation. We further provide an extension of our DP-CATE, where we employ RKHS regression to release the complete CATE function while ensuring differential privacy. We demonstrate our DP-CATE across various experiments using synthetic and real-world datasets. To the best of our knowledge, we are the first to provide a framework for CATE estimation that is Neyman-orthogonal and differentially private.

Problem

• Estimating heterogeneous treatment effects from sensitive patient data.
• Ensuring differential privacy in CATE estimation from observational data.
• Developing a general framework for private CATE estimation using machine learning.

Innovation

• DP-CATE framework for private CATE estimation
• Neyman-orthogonal loss ensures robust estimation
• RKHS regression for private CATE function release