Bayesian Modeling and Computation in Causal Inference – Applications in Sequential Decision-Making
Causal inference provides a suite of tools for defining and identifying causal effects of non-randomized treatments in complex settings. Though agnostic to inferential paradigms, Bayesian estimation of causal effects has several advantages including full posterior inference for functionals, uncertainty quantification about partially-identifiable aspects of the causal structure via priors, and principled shrinkage. Lastly, an array of nonparametric Bayesian models combines flexible point estimation with the uncertainty estimation required for inference. Crucially, these methods require tailored Markov Chain Monte Carlo sampling and integration methods to obtain posterior draws of the causal effects. We discuss some recent advances and applications of these methods for fitting nonparametric causal models to estimate effects of sequential treatment decisions.
Arman Oganisian is an Assistant Professor of Biostatistics at Brown University. His methodological research centers around developing Bayesian methods for flexible causal effect estimation with observational data. These methods blend principled causal reasoning, nonparametric Bayesian modeling, and efficient computation to build tools for data-driven decision making. His substantive area of interest is primarily oncology, where he has developed Bayesian nonparametric methods cost estimation and cost-effectiveness estimation for endometrial cancer therapies and novel Bayesian bootstrap procedures for assessing treatment effect heterogeneity of proton therapies. He is currently a PI of a PCORI subaward developing Bayesian semiparametric methods for estimating and optimizing effects of sequential treatment strategies with motivating applications in acute myeloid leukemia.
He received his PhD in Biostatistics from the University of Pennsylvania where he was an active member of the Center for Causal Inference. More information on his work can be found at stablemarkets.netlify.app.