Title: A Framework for Interpretable Causal Inference with PK/PD Models with Applications to Seizure-like Activity 

Abstract: 

Evaluating treatment for critically-ill patients is an extremely challenging task: treatment decisions are high-stakes, clinical trials may be difficult to arrange, observational datasets can be limited in scope or quality, drug interactions can be complex, and there may be potentially important unobserved variables, such as drug absorption rates, which are typically not recorded in the EHR. Ignoring these variables can induce a bias in our analysis, which can make it difficult to estimate the desired causal effects. Ideally, we need an interpretable and auditable framework for these types of high-stakes causal analyses. To this end, we will be introducing a new framework for analyzing the causal effect of seizure-like behavior in critically ill patients while taking into consideration the valuable information one can obtain from PK/PD modeling. Moreover, we will seek to demonstrate the additional insights one can gain by taking an interpretable approach to matching and the additional reliability we gain from the ability for physicians to externally validate the results via a chart review. Time permitting, we will also discuss some new developments in Demspter-Shafer Inference for Multinomial Data.

Share this Event