Tiansheng Wang

Latest News

• Oct 2025: Honored to receive the Alzheimer’s Association & NACC New Investigators Award .
• Oct 2025: Start adjunct assistant professor position at Dept. of Epidemiology, University of North Carolina at Chapel Hill Gillings School of Global Public Health
• Sep 2025: Present at 11th Annual UNC Center for Pharmacoepidemiology Methods Symposium on Using of Proximal Causal Inference to Address Unmeasured Confounding in Pharmacoepidemiology Studies.
• Sep 2025: Start tenure-track assistant professor position at Dept. of Pharmaceutical Health Outcomes and Policy, University of Houston College of Pharmacy.
• Aug 2025: Present at ISPE 2025 on developing novel methods for confounding control using longitudinal pattern at baseline period. Talk 1: preventative drug trajectories for confounding control; Talk 2: temporal high-dimensional propensity score.

Active Research Areas

Precision Medicine

• Iterative Causal Forest (iCF) — Am J Epidemiol, 2024
We've developed a machine learning subgrouping algorithm based on causal forest to better identify subpopulation who will benefit (or be harmed) more from a specific treatment. The iCF algorithm iteratively develops shallow causal forests at different depths to obtain a family of subgroup decisions, then picks the cross-validated subgroup decision that best predicts causal treatment effect as the final output (subgroup decision).

• High-dimensional Iterative Causal Forest (hdiCF) — Am J Epidemiol, 2024
We developed hdiCF, the high-dimensional extension of our iterative Causal Forest (iCF) algorithm. hdiCF uses diagnosis/procedure/prescription codes (i.e., does not require predefined covariates) to identify subgroups with heterogeneous treatment effects in claims data. We demonstrate its utility using a case study of SGLT2i and hospitalized heart failure.

• High-dimensional method for assessing heterogeneous treatment effect in Claims — Am J Epidemiol, 2025
We introduced a new high-dimensional setting (one ordinal variable per diagnosis/procedure/prescription code) to better identify and evaluate treatment effect heterogeneity in large real-world datasets.

Confounding Control

• Caution in Handling Switchers in Pharmacoepidemiologic Studies Estimating Treatment Effects — Am J Epidemiol, 2025
We highlighted the study design flaw in a specific type of time-varying exposure design, which only allows individuals from the comparator group to switch to the treatment group but not the other way around (asymmetrical follow-up), inducing time-related bias.

• Temporal pattern for confounding control
Work in progress, more coming soon.

Using multimodal data (genetics, imaging, real-world data) for assessing drug effects and drug repurposing

• Using integrated genetic and real-world data to assess heterogeneous treatment effects — Alzheimers Dement, 2026
We assessed the heterogeneous treatment effects of statins on dementia in a target trial emulation with iterative causal forest (iCF) causal machine learning subgrouping using integrated genetic and real-world data in UK Biobank data.

• Claims-based Dementia Identification Algorithm — Am J Epidemiol, 2025
Using Medicare claims from participants of the ARIC Study and ARIC’s rigorous syndromic dementia classification, we developed and validated algorithms to identify incident and prevalent dementia, addressing a key need in pharmacoepidemiology and aging research.

• Deep learning-powered neuroimaging analysis to assess drug effects on dementia
Work in progress, more coming soon.