91黑料网

Alexander Levis听笔丑顿

Assistant Professor of Biostatistics| University of Pennsylvania

WHEN: Wednesday, February 25, 2026, from 3:30 to 4:30 p.m.
WHERE: Hybrid | 2001 91黑料网 College Avenue, Rm 1140;
NOTE:聽Alexander Levis will be presenting in-person at SPGH聽

Abstract

Biomedical and public health researchers increasingly turn to large observational data, such as those deriving from electronic health records (EHR), to quantify the effects of medical treatments or other exposures. Target trial emulation has emerged as a promising framework for conducting causal inference from these data sources. The approach aims to mitigate bias in part by clarifying a target causal estimand and, critically, a common notion of time zero. Notwithstanding the virtues of this conceptual framework, current practice is often limited by reliance on strong parametric models, implicit and often implausible structural assumptions, and鈥攑erhaps due to a lack of a universally accepted mathematical formalism鈥攁n inability to formally handle commonly faced challenges (e.g., selection due to missing data).In this talk, I will present two recent projects that strive to make progress towards remedying these issues, focusing on estimation of causal effects in the face of (1) missing eligibility status, and (2) possibly calendar time-varying effects. Our development begins by making explicit the structure of the observational (e.g., EHR-based) data used with target trial emulation, the target causal estimands, and assumptions under which the latter are identified from the observed data distribution.The ensuing proposed estimators incorporate machine learning, and can achieve asymptotic normality and semiparametric efficiency under relatively weak assumptions. Finally, we demonstrate practical use of our methods by analyzing an EHR-based study of long-term weight and glycemic outcomes, comparing bariatric surgical intervention to no surgery among patients with severe obesity.

Speaker Bio

Alex Levis is an Assistant Professor of Biostatistics in the Center for Causal Inference and the Department of Biostatistics, Epidemiology and Informatics at the University of Pennsylvania. Before joining Penn, he obtained his PhD in Biostatistics at Harvard University in 2022, and subsequently was a postdoctoral researcher in the Department of Statistics & Data Science at Carnegie Mellon University. Alex's expertise lies in statistical methods development related to causal inference, missing data, non- and semi-parametrics, machine learning, and the use of administrative and electronic health record data for comparative effectiveness research. He is interested in a broad range of application areas spanning the biomedical sciences, focusing recently on problems in surgery, experimental neuroscience, and mental health.聽