C5 – AI-Assisted Bayesian Causal ML Methods and Software for Modern Biopharmaceutical Development

Chair:  Brad Carlin, PhD (PhaseV Trials, Inc)

Instructors:
Brad Carlin, PhD (PhaseV Trials, Inc.)
Raviv Pryluk, PhD (PhaseV Trials, Inc.)

Course Description:

Artificial intelligence (AI) tools for large language modeling (LLM) such as ChatGPT, Claude, and others have had a rapid and accelerating impact on scientific practice.  In biopharmaceutical development, LLMs are already used to develop standard forms and reports, summarize literature, write computer code, draft technical reports, and many other functions. Now, FDA has announced that they too will scale use of a generative AI pilot to assist scientific reviews.  Despite these developments, AI and machine learning (ML) tools have yet to become standard in other areas of drug development, such as in the design, analysis, and execution of clinical trials. This is due at least in part to the fact that AI tools require training with both data and domain expertise, and they remain prone to mistakes that must be spotted and corrected by humans. 

This short course offers an introduction to AI-assisted methods for biopharmaceutical development, with particular focus on their use in adaptive clinical trials.  We will begin with a brief overview of the Bayesian adaptive approach, and describe trial optimization approaches that seek to control Type I error, maximize power, minimize sample size, and ensure other favorable operating characteristics while still greatly reducing the simulation burden. We may then further improve the adaptive trial design via a stochastic optimization algorithm that uses local information to limit the search space and avoid a time-consuming grid search. Next, we discuss the use of causal ML to evaluate heterogeneity of treatment effects and identify promising patient subgroups. We obtain improved estimates by combining multiple algorithms, all while accounting for uncertainty in the causal ML estimates. Finally, we describe AI-enhanced tools for improved clinical operations, such as incorporation of external data, identifying recruiting centers, and monitoring drug supply, with key results displayed in a convenient computer “dashboard”. 

Additional topics to be covered may include:

•       AI-powered causal inference tools to incorporate real-world data (RWD)/real-world evidence (RWE), including propensity score matching
•       AI-assisted tools for literature search and meta-analysis (for setting placebo rates) and clinical site identification
•       AI-guided tools for clinical site selection and prioritization
•       Causal modeling powered by AI to inform indication selection and sequencing

Time permitting, we will also offer a brief demonstration of the PhaseV computing platform, which incorporates many of these features.