2024 Colloquia

= = = = = = = = = = =

November and December 2024 will have closed faculty candidate seminars. Department faculty and students will receive emails with details.

= = = = = = = = = = =

Bayesian Clinical Trial Designs for Utilizing Historical and Real-World Data in Rare and Pediatric Drug Development

Bradley Carlin, Ph.D.
PhaseV Trials

Thanks to the sudden emergence of Markov chain Monte Carlo (MCMC) computational methods in the 1990s, Bayesian methods now have a more than 25-year history of utility in statistical and biostatistical design and analysis. However, their uptake in regulatory science has been much slower, due to the high premium this field places on bias prevention and Type I error control, and its historical reliance on p-values and other traditional frequentist statistical tools. Fortunately, recent actions by regulators at FDA and elsewhere have indicated a new willingness to consider more innovative statistical methods, especially in settings where traditional methods are ill-suited or demonstrably inadequate. 

In this talk, we will begin with very brief review of the Bayesian adaptive approach to clinical trial design and analysis, as well as recent supportive efforts by regulatory agencies in the United States (FDA) and Europe (EMA).  We will then discuss a variety of areas in which Bayesian methods offer a better (and perhaps the only) path to regulatory approval. We begin with an overview of methods for borrowing strength from historical controls and other auxiliary data, including power priors, commensurate priors, robust mixture priors, and more recent techniques that build on these ideas.  We also discuss approaches that incorporate patients’ own natural history data, a generalization of crossover designs in which each patient “acts as their own control.”  We then turn to even more daring designs that use causal inference tools to incorporate real world data (RWD)/real world evidence (RWE), including synthetic controls.  Here it is sometimes useful to combine methods for bias correction (e.g., propensity matching) with methods for cautious borrowing (e.g., commensurate priors).  

Thursday, November 7, 2024

= = = = = = = = = = =

Semi-Parametric Sensitivity Analysis for Trials with Irregular and Informative Assessment Times

Daniel Scharfstein, Sc.D.
 University of Utah School of Medicine

Many trials are designed to collect outcomes at or around pre-specified times after randomization. If there is variability in the times when participants are actually assessed, this can pose a challenge to learning the effect of treatment, since not all participants have outcome assessments at the times of interest. Furthermore, observed outcome values may not be representative of all participants' outcomes at a given time. Methods have been developed that account for some types of irregular and informative assessment times; however, since these methods rely on untestable assumptions, sensitivity analyses are needed.

We develop a methodology that is benchmarked at the explainable assessment assumption, under which assessment and outcomes at each time are related only through data collected prior to that time. Our method uses an exponential tilting assumption, governed by a sensitivity analysis parameter, that posits deviations from this assumption. Our inferential strategy is based on an influence function-based, augmented inverse intensity-weighted estimator. Our approach allows for flexible semiparametric modeling of the observed data, which is separated from specification of the sensitivity parameter. We apply our method to a randomized trial of low-income individuals with uncontrolled asthma and evaluate performance in a realistic simulation study.

This work is joint with Bonnie Smith, Yujing Gao, Shu Yang, Ravi Varadhan and Andrea Apter.

Thursday, September 26, 2024

= = = = = = = = = = =

Quantifying Tumor Geography: Statistical Methods for Spatial Biology

Veera Baladandayuthapani, Ph.D.
 University of Michigan

Visit the Andrei Yakovlev Colloquium for details

Thursday, September 19, 2024

= = = = = = = = = = =

Linking Longitudinal Methods toward Clinical Decision Support

Patrick Heagerty, Ph.D.
University of Washington

Visit the Charles L. Odoroff Memorial Lecture for details

Thursday, May 2, 2024

= = = = = = = = = = =

Estimating Interactions in Chemical Mixtures: Case Studies from Cancer Epidemiology

Paul S. Albert, Ph.D.
 National Institutes of Health (NIH)

Estimating the interactions between mixture components is often of interest in epidemiologic studies. Motivated by studies examining the effects of chemical exposures in the development of cancer, we develop and compare different approaches for estimating interactions among components of these exposures. We develop a latent functions approach where the main and interaction effects are estimated using two separate sets of unobserved functions. We also develop a Bayesian shrinkage approach that incorporates the hierarchical principle which assumes that it is unlikely that there are interactions without the presence of corresponding main effects. These approaches are compared with Bayesian kernel machine regression (BKMR) and LASSO, two approaches currently being used for analyzing interactions in chemical mixture studies. We analyze the data from a series of NCI studies to provide insight into chemical mixture analyses.

This is joint work with Sung Duk Kim and Debamita Kundu.

Thursday, April 18, 2024

= = = = = = = = = = =

Sparse Causal Learning

Linbo Wang, Ph.D.
 University of Toronto

In many observational studies, researchers are interested in studying the effects of multiple exposures on the same outcome. Unmeasured confounding is a key challenge in these studies as it may bias the causal effect estimate. To mitigate the confounding bias, we introduce a novel device, called the synthetic instrument, to leverage the information contained in multiple exposures for causal effect identification and estimation. We show that under linear structural equation models, the problem of causal effect estimation can be formulated as an $\ell_0$-penalization problem, and hence can be solved efficiently using off-the-shelf software. Simulations show that our approach outperforms state-of-art methods in both low-dimensional and high-dimensional settings. We further illustrate our method using a mouse obesity dataset.

Thursday, March 21, 2024

= = = = = = = = = = =