Decision Theoretic Procedure for Multiple Testing through Deep Neural Networks

Meg Gamalo, Pfizer

Margaret (Meg) Gamalo, PhD is currently VP, Statistics Head for Inflammation and Immunology in Pfizer Global Product Development. She combines expertise in biostatistics, regulatory science and adult and pediatric clinical development in inflammation and immunology, endocrine, neurodegenerative and infectious diseases. Prior to joining Pfizer, she was Research Advisor, Global Statistical Sciences at Eli Lilly and Company and Mathematical Statistician at the Food and Drug Administration. Meg led the Pediatric Innovation Task Force at the Biotechnology Innovation Organization and a member of the European Forum for Good Clinical Practice – Children’s Medicine Working Party that provided guidance on inclusion of adolescents in adult research. She co-leads the scientific workgroups on Statistics in Pediatric Drug Development and the Statistical Perspective on AI/ML in Pharmaceutical Development within the Biopharmaceutical Section of the ASA. Meg is a Fellow of the American Statistical Association.

Abstract:

Analytics, Biostatistics, AI/ML, Gen-AI, Oh My!

Greg Szwartz, Managing Director, Life Sciences Data Science Practice Lead, Deloitte

Greg has over 25 years of consulting experience in Life Sciences, with the majority focused on commercial analytics, market access, and patient safety. He is focused on applying quantitative analysis to strategic and operational decisions. As the leader of Deloitte’s precision engagement practice, Greg is on the Advisory Board of Upenn’s Master’s in Behavioral and Decision Sciences program, as well as the anti-microbial resistance program at Virginia Tech. Greg has a Bachelor’s and Master’s degree in Engineering from Columbia University, and a MS from the Harvard School of Public Health. Greg is a member of the Society of Decision Professionals and the System Dynamics Society. His publications include a marquis study on behavior change published in JAMA and a broad piece on data + behavioral science for health.

Abstract:

As life science companies continue to invest in analytics and AI, the new emphasis on generative AI adds to a growing pressure to show benefits of these investments - at scale - and in line with expectations set (often by non-Analytics professionals) around “democratization of data science” and “insight to action decisioning”. To work at scale and deliver on business cases, we need to address challenges around the consistent use of AI (and Gen-AI) and ML models, the measurement of their incremental value over traditional methods, and their large-scale application. Additionally, health equity is a key part of the challenge for us in 2024. Increasingly patient-centric analytics call for data science applications that can combat systematic bias and enable equitable access to effective healthcare.

There's also a need for a more rounded approach to scaled analytics that not only focuses on technology but also considers aspects like business strategy, talent, and operations. A self-funding approach, where the financial benefits derived from analytical models are reinvested into their further development, can also help ensure the full potential of these models is realized. And the growing demand for data science applications also implies a changing landscape for data science and analytics skills and a new operation models in organizations looking for scaled applications.

Estimating Individual Treatment Effects in Randomized Controlled Trials Using Machine Learning

Kenneth Verstrate, Researcher, KU Leuven

Kenneth Verstraete is a doctoral researcher at the KU Leuven in Belgium since 2019. He obtained a Master in Computer Science in 2017 and a Master in Artificial Intelligence in 2019 at the KU Leuven. He is currently doing research on machine learning in pulmonology with a focus on chronic obstructive pulmonary disease (COPD) and pulmonary hypertension. His primary research is on the estimation of individual treatment effects using causal machine learning, focusing on methodological aspects as well as applications. He also focuses on the use of machine learning to support multidisciplinary board discussions to aid in complex diagnoses and treatment decisions.

Abstract: Our study aimed to develop machine learning models for estimating individual treatment effects (ITE) in COPD interventions, utilizing data from randomized controlled trials (RCTs). Using Causal Forest as the causal inference model, we optimized and tested the models on COPD patients from the SUMMIT and IMPACT trials. The novel Q-score metric was introduced to assess the power of causal inference models. Results indicated that patients with the strongest ITE consistently experienced the most significant reductions in exacerbation rates in both trials. Poor lung function and blood eosinophils emerged as strong predictors of ITE. The study suggests that ML models can effectively identify individual responses to COPD treatments, potentially serving as valuable tools for personalized treatment decisions.