Fall 2026 Departmental Seminars & Lectures

During the Fall and Spring semesters, the Department of Biostatistics holds regular seminars on Thursdays, called the Levin Lecture Series, on a wide variety of topics which are of interest to both students and faculty. Over each semester, there are also often guest lectures outside the regular Thursday Levin Lecture Series, to provide a robust schedule the covers the wide range of topics in Biostatistics. The speakers are invited guests who spend the day of their seminar discussing their research with Biostatistics faculty and students. All Levin Lectures will be taking place in-person on Thursdays beginning at 11:45am.

Fall 2026 Schedule

Thursday, September 10th, 532A/532B - SMS

Alexander Levis, PhD

Assistant Professor of Biostatistics, Department of Biostatistics, Epidemiology and Informatics

Perelman School of Medicine, University of Pennsylvania

Sharp mediation analysis with recanting witnesses

 

Abstract:

Causal mediation analysis invariably relies on strong untestable identifying assumptions. The standard approach for point identifying natural direct and indirect effects assumes that there are no "recanting witnesses", that is, post-treatment confounders of the mediator-outcome relationship. Alternatives include replacing estimands with less interpretable but identifiable randomized analogues, or assuming the treatment decomposes into components that do not simultaneously affect the same post-treatment confounders. Here, we pursue partial identification of true mediation quantities: natural path-specific effects under a nonparametric equation model with independent errors. We show that sharp bounds arise as solutions to a challenging optimal transport problem. We provide a general solution by regularizing the bound functionals using entropic optimal transport, and derive flexible, nonparametric efficient estimators for the smoothed bounds. Further, we present preliminary results for separable effect estimands under a weaker structural model. In this case, we assume that the treatment can be decomposed into two distinct components, but allow both components to arbitrarily impact the same set of post-treatment confounders. Finally, we demonstrate the proposed methods in simulations and empirical examples.

Thursday, September 17th, 532A/532B - SMS

Yu Cheng, PhD

Professor, Department of Statistics

Kenneth P. Dietrich School of Arts and Sciences, University of Pittsburgh

Win Odds Comparison in Sequential Multiple Assignment Randomized Trial

 

Abstract:

Sequential multiple-assignment randomized trials (SMARTs) are designed to construct and evaluate dynamic treatment regimes (DTRs), sequences of treatment decisions tailored to a patient's evolving status across multiple stages of care. In a two-stage SMART, patients are randomized twice: first among initial treatments, and subsequently among second-stage options conditional on response to the initial treatment, resulting in multiple embedded DTRs. In biomedical settings, patients often experience multiple, possibly competing events (e.g., an adverse event occurring before a therapeutic response), complicating personalized decision-making. We adopt win odds (WO) to assess the comparative effectiveness of one DTR versus another. WO and related win statistics have gained traction as intuitive metrics that prioritize clinically important outcomes and accommodate multiple endpoints. However, existing methods do not directly extend to SMARTs because embedded DTRs can share patient paths, inducing dependence. To address this challenge, we use inverse probability weighting to account for missingness induced by subsequent randomizations and inverse probability of censoring weighting to address comparisons that are indeterminate due to censoring. We consider two inference strategies: (i) deriving the WO variance using V-statistic theory and related asymptotic tools and (ii) a hull-based confidence interval approach. Both methods perform well in numerical studies.

Thursday, September 24th, ARB 8th Floor Auditorium

Joseph W. Hogan, ScD

Chair of Biostatistics; Carole and Lawrence Sirovich Professor of Public Health; Professor of Biostatistics

Brown University School of Public Health

Clinical decision support for HIV care: development, implementation, evaluation

 

Abstract:

Key milestones in HIV care are represented in a cascade framework that prioritizes case identification, linkage and retention in care, compliance with antiviral medication, and suppression of viral load. Many prediction models have been developed to target these endpoints. In this talk, we describe a comprehensive approach to development, implementation, and formal, trial-based evaluation of a machine-learning-based approach to clinical decision support that is focused on retention in care. A point of emphasis is that, even in a randomized trial, the impact of prediction models or AI interventions on patient outcomes typically needs to be represented as a multi-stage process. We highlight key early findings about effectiveness from pilot data, and describe methodologic innovations motivated by the goals of the project. This work is based at AMPATH, a large HIV care program in western Kenya, and is a collaboration between Brown University (US) and Moi University (Kenya).

Thursday, October 1st, Hess Commons

Amanda Mejia, PhD

Associate Professor, Department of Statistics

College of Sciences, North Carolina State University

Statistical methods for individual precision in functional brain imaging

 

Abstract:

Functional magnetic resonance imaging (fMRI) enables noninvasive mapping of human brain function and is a valuable resource for neuroscience research and in clinical care. However, extracting accurate individual-level information from this high-dimensional, noisy, and biologically complex data remains a major challenge, which limits clinical translation. In this talk, I will present statistical methods developed to enhance individual precision in functional brain mapping by leveraging information from the population and across the brain. These methods are designed to respect key features of the data, including its spatiotemporal structure, neuroanatomy, and individual variability in both spatial and temporal brain organization. Finally, I will describe data science tools that facilitate adoption of these statistical techniques, including open source software, reproducible preprocessing pipelines, and resource sharing, with the ultimate goal of advancing the utility of functional neuroimaging for both research and clinical care.

Thursday, October 8th, ARB 8th Floor Auditorium

Yuchen Hu, PhD

Assistant Professor, Department of Industrial Engineering and Operations Research

Columbia Engineering, Columbia University

Spectral Design of Random-Duration Switchbacks

 

Abstract:

A switchback experiment alternates an entire system between treatment and control over time. It is especially useful when interactions between units can undermine standard unit-level experiments. Switchback experiments are commonly implemented on fixed temporal grids, which impose highly structured restrictions on when treatment can switch. We study a broader class of random-duration switchbacks, in which treatment and control alternate across runs whose durations are drawn from a common distribution. Under finite carryover, we show that the mean squared error of the Horvitz-Thompson estimator has a simple frequency-domain representation governed by how temporal outcome patterns align with design-induced imbalance and contamination patterns. This representation yields a tractable worst-case design criterion that can be computed directly from the duration distribution. Optimizing over even a simple two-rate family produces a run-age-dependent switching rule that reduces the asymptotic worst-case mean squared error by at least 32% relative to the standard independently randomized fixed-block switchback.

Thursday, October 15th, Hess Commons

Benjamin Risk, PhD

Associate Professor, Department of Biostatistics and Bioinformatics

Rollins School of Public Health, Emory University

Motion artifacts, selection bias, and solutions to improve pediatric brain connectivity studies

 

Abstract:

Resting-state functional magnetic resonance imaging (rs-fMRI) is used to examine correlations between brain regions, known as functional connectivity. Head motion during a scan creates artifacts that distort functional connectivity. Motion quality control removes high-motion participants. We examine how this introduces selection bias in autism and other studies of pediatric populations. We develop methods to address this bias and more efficiently use all data. First, we introduce a method that works on the processed connectivity measures. The method uses ensemble learning and a stochastic intervention to estimate what connectivity would look like under low motion. Second, we introduce a method that works on the time series of the rs-fMRI data, where we use ensemble learning to remove motion artifacts during preprocessing. We integrate this approach into a causal inference framework to examine the impact of a treatment on functional connectivity. We discuss the merits of different approaches to addressing selection bias and the implications for understanding autism and neurodevelopmental disorders.

Thursday, October 22nd, Hess Commons

Chong Wang, PhD

Professor, Department of Statistics and Department of Veterinary Diagnostic and Production Animal Medicine

College of Liberal Arts and Sciences and College of Veterinary Medicine, Iowa State University

Where Does the Evidence Come From? Exact Study-Level Attribution of Direct and Indirect Evidence in Network Meta-Analysis

 

Abstract:

Network meta-analysis (NMA) borrows strength across a connected network of trials, combining direct and indirect comparisons into a single estimate of relative treatment effect. A persistent difficulty in reporting such analyses is that widely used contribution measures do not reproduce the estimate they are meant to explain: applied as ordinary weights to the component effects, published contribution proportions generally fail to return the reported NMA estimate, and multi-arm trials, which induce within-study correlation, make the discrepancy worse.

In this talk I describe a contrast-space projection formulation of NMA that closes this reproducibility gap. Writing the estimator as an orthogonal projection of the observed pairwise contrasts onto the consistency-constrained contrast space gives an explicit linear map from the data to the fitted estimate. A canonical within-study reduction then removes the algebraic redundancy created by multi-arm trials and yields a study-based definition of direct and indirect evidence that is unique and invariant to how contrasts are parameterized. The result is an exact, covariance-aware decomposition of the NMA estimator into study-level direct and indirect contributions, with the indirect component resolved further into path-level pieces. The resulting weights are directly analogous to inverse-variance weights in pairwise meta-analysis and support, to our knowledge, the first forest plot that exactly reconstructs an NMA estimate, together with tension plots and path-based displays that show where evidence and disagreement reside. I illustrate the framework on empirical treatment networks and close with ongoing work on when a path-level allocation is itself identified by the fitted network, and what should be reported when it is not.

Thursday, October 29th, Hammer 301

Roderick J. A. Little, PhD

Richard D. Remington Distinguished University Professor of Biostatistics; Professor, Department of Statistics; Research Professor, Institute for Social Research

University of Michigan

Seminal Ideas and Controversies in Statistics

 

Abstract:

The evolution of statistics is marked by breakthrough developments but also by lively controversies that have shaped the discipline. Appreciating these seminal ideas and controversies deepens understanding and hones critical thinking skills. This seminar discusses a number of these topics, which are discussed in more detail in my recent book, Seminal Ideas and Controversies in Statistics. Chapman Hall/CRC. I'll touch on the following major themes:

1. Estimating equations vs likelihood-based methods
2. Conditionality principle, strengths and weaknesses of frequentist and Bayesian inference approaches, the Calibrated Bayes compromise
3. Fisher and Neyman's approaches to hypothesis testing, the recent ASA statement on P-Values and reactions
4. Empirical Bayes and random-effects models
5. Data science and exploratory data analysis
6. The role of randomization in surveys and experiments
7. Causal inference

Thursday, November 5th, Hess Commons

Yi Zhao, PhD

Associate Professor of Biostatistics & Health Data Sciences

Indiana University School of Medicine

Regression Modeling of Correlation and Covariance Matrices

 

Abstract:

Correlation and covariance matrices are fundamental objects in multivariate analysis, arising naturally in neuroimaging, genomics, finance, and many other fields as representations of pairwise linear associations among variables. In resting-state functional magnetic resonance imaging (fMRI) studies, for example, the correlation or covariance matrix of regional brain activity time courses provides a compact representation of functional connectivity. A central scientific question is how such connectivity patterns vary with subject-level characteristics, including age, cognitive performance, and clinical status. Modeling these matrix-valued outcomes presents important methodological challenges. Full-rank covariance matrices are constrained to be symmetric and positive definite, while correlation matrices must additionally have unit diagonal entries. Consequently, they lie on non-Euclidean spaces, making standard regression methods designed for unconstrained Euclidean outcomes generally inappropriate. In this presentation, I will introduce a geodesic regression framework for full-rank correlation matrix outcomes on the correlation manifold. The proposed model relates Euclidean covariates to manifold-valued responses through the Riemannian exponential map. We equip the correlation manifold with a geodesically complete and permutation-invariant family of metrics. For model estimation, we develop an iterative algorithm that alternates between tangent-space linearization and Riemannian gradient descent. We further extend the framework to high-dimensional covariate settings through penalized estimation. I will conclude by briefly discussing a complementary set of parsimonious modeling frameworks that we have developed for covariance matrix outcomes. These approaches transform covariance matrices into suitable linear representations, allowing statistical models formulated in Euclidean spaces to be applied while preserving essential features of the original matrix-valued data.

Thursday, December 3rd, ARB 8th Floor Auditorium

Fei Zou, PhD

Professor of Biostatistics

Gillings School of Global Public Health, University of North Carolina at Chapel Hill

Semiparametric Neural Network Models for Estimating Heterogeneous Treatment Effects

 

Abstract:

In this talk, we introduce deepHTL (deep Heterogeneous Treatment Learning), a semiparametric framework for testing and estimating heterogeneous treatment effects (HTE) using observational data, such as hospital registries. In deepHTL, we incorporate three key innovations to deliver robust causal inference in the presence of complex confounding.

First, we employ bagged deep neural networks to estimate high-dimensional nuisance functions, improving accuracy and addressing the instability commonly observed in standard DNNs with finite samples. Second, we estimate heterogeneous treatment effects through a bias-reduced semiparametric regression. Third, to rigorously assess treatment effect heterogeneity, we implement both a global kernel score test and a cross-fitted permutation test, ensuring valid finite-sample inference.

Across extensive simulations designed to mimic confounding structures typical of clinical registry data, deepHTL achieves proper type I error control and yields more accurate individualized treatment effect estimates than existing methods. In an application to lung transplant cohort data, we establish strong heterogeneous treatment effects between bilateral and single lung transplantation, providing statistically grounded insights to guide patient selection and donor organ allocation.

Thursday, December 10th, Hess Commons

Walter Dempsey, PhD

Associate Professor of Biostatistics; Research Associate Professor, Survey Research Center, Institute for Social Research

University of Michigan School of Public Health

Future-Path Incremental Excursion Effects in Sequentially Randomized Trial

Abstract:

In sequentially randomized trials, researchers are often interested in assessing time-varying effects of the interventions on both proximal and distal outcomes. When considering impact on distal outcomes, it is common to either estimate the value function under strong assumptions such as time-homogeneous Markov dynamics or to assess a localized contrast of an intervention versus no intervention at a single decision point on the distal outcome which requires minimal assumptions but averages over the remaining treatment path. In this talk, we introduce a future-path incremental excursion effect defined through the local derivative of expected downstream utility under an infinitesimal log-odds tilt of future treatment probabilities from a decision time onward which bypasses the curse of the horizon problem without strong assumptions on the underlying dynamics. The parameter is anchored to the randomized trial policy while allowing analysis of both cumulative long-term proximal outcomes and distal. We will introduce likelihood-ratio, score, outcome-regression, and recursive representations, yielding direct and cross-fitted augmented estimators. The same local gradient supports KL-calibrated pessimistic policy updates that recommend changes only when a lower confidence bound is positive. Simulations show that the estimand captures future-path heterogeneity and that recursive augmentation can substantially improve precision. An application to LowSalt4Life 2, a micro-randomized trial, illustrates long-run engagement and a secondary clinical endpoint. Time permitting, we will discuss extensions of this work to AI evaluation.