Research

My research develops interpretable, robust, and reproducible statistical methods for data-driven decisions in health and medicine. The work is organized around three connected programs: personalizing decisions, modernizing the evidence generated by clinical studies, and making biomedical AI more trustworthy.

Precision health

Individualized decisions that remain interpretable

I develop interpretable and reproducible methods for precision intervention development and personalized clinical evaluation, with an emphasis on dynamic, data-driven decisions for heterogeneous populations.

This program combines reinforcement learning with robust causal inference to learn adaptive strategies while accounting for measurement cost, clinical utility, and health equity. It also develops transparent tools for individualized variable selection and personalized clinical indices from large-scale EHR data.

Reinforcement learning Robust causal inference Precision intervention development Personalized clinical evaluation

Modern clinical studies

Evidence that respects complex outcomes and imperfect measurement

I build nonparametric and semiparametric methods for clinical studies in which effects vary over time, outcomes are censored, endpoints have different priorities, or no perfect reference standard exists.

The work ranges from global testing and screening for dynamic covariate effects to win statistics for prioritized outcomes. It also develops estimands and inference for duration of response, clinically meaningful ties, and medical-device evaluation with multiple imperfect references.

Quantile regression Win statistics Clinical trials Medical devices

Trustworthy biomedical AI

Reliable uncertainty and reproducible evaluation

I develop statistical and machine-learning methods for trustworthy biomedical AI that quantify uncertainty, withstand rigorous evaluation, and support reliable decisions across clinical and scientific settings.

Current directions include conformal and predictive-distribution inference for censored outcomes; evidence-based evaluation of clinical large language models; generative modeling for synthetic data to support reproducible validation and benchmarking; and reproducible AI methods for drug discovery.

Conformal inference Generative modeling AI for drug discovery Clinical LLM evaluation
View publications Explore software