Chair: Matt Austin, Amgen

Matt’s been in involved in analytics in the pharmaceutical industry for 25 years having spent more than 20 years at Amgen. He was the global statistical lead for products in oncology and bone health. Matt currently leads a data science team focusing on addressing clinical development issues.

Applications of Deep Learning Based Medical Image Segmentation in Drug Development

David Paulucci, Director, Data Science, BMS

David Paulucci is Director of Data Science in the Department of Global Biometrics and Data Sciences at Bristol Myers Squibb. In his role, David leads a team responsible for exploratory analyses in the domains of biomarker, imaging, real world data, and machine learning, to support global drug development in oncology. Prior to working at BMS, he spent 3 years at Mount Sinai Hospital, carrying out analyses leading to more than 30 publications on surgical outcomes for localized kidney and prostate cancer. David received his MS in Biostatistics from the Icahn School of Medicine at Mount Sinai. His key areas of interest include supervised machine learning, and predictive modeling of heterogenous treatment effects.

Abstract: The advent of 3D medical image segmentations models in recent years has been driven by methodological advances in deep learning based architectures, accessibility of high performance compute, and the proliferation of datasets available for training. Two of the more prominent approaches include 3D fully convolutional networks and UNet Transformers. Furthermore, open source frameworks for deep learning in health care such as MONAI, have made medical image segmentation tools readily accessible for drug development organizations. There are many potential applications of deep learning based medical image segmentation in drug development. Segmentation of tumors and extraction of sizebased features at screening and on treatment may allow for fully automated assessments of tumor progression and response. Features beyond size including shape and density may also be extracted from the segmented tumor to identify predictive and prognostic features for stratification, patient selection and prediction of long term outcome. Segmentation can also be used for downstream assessments of genetic mutation status or gene expression through imaging phenotypes. Deep learning based medical image segmentation offers a variety of unique opportunities to facilitate more rapid development of safe and effective therapies. However, significant investment is required to maximize the development and potential utility of these approaches in drug development, such as the time and cost to acquire the large amount annotated data required for training, and collaboration from multiple disciplines providing appropriate subject maPer expertise to limit model bias.

AI Guided Clinical Trial Design

Rishu Batra, Associate Principal, ZS Associates

Rishu is an Associate Principal at ZS Associates in New Delhi (India), bringing over 12 years of experience in management consulting and analytics within the healthcare domain. His career is marked by a deep commitment to advancing healthcare through innovative solutions. Currently, he leads the Clinical AI team in India, focusing on AI-driven solutions to accelerate drug development and enhance clinical trial efficiency. A key area of his work involves exploring the possibilities of Classical and Generative AI to solve complex challenges in clinical trials and drug development.

Ronnie Du, Associate Principal, ZS Associates

Ronnie is an Associate Principal at ZS Associates in Los Angeles, bringing over 18 years of experience in clinical development, marketing, technology, and data science applied to accelerating drug development and commercialization. Ronnie’s focus is on biopharmaceutical data strategy and data-driven initiatives where he has helped clients maximize the value of their internal and external data assets to drive meaningful business outcomes. Most recently Ronnie has focused on enabling data-driven clinical development design, planning, and operations.

Ronnie has a Bachelors degree in electrical engineering and computer science and a Masters in Business Administration, both from the University of California Los Angeles.

Abstract:

This session will illuminate the transformative role of AI in reshaping clinical trial design. We will explore how AI-driven strategies are essential in optimizing critical elements of trial design, including optimizing inclusion/exclusion criteria, determining right endpoints, reduction of amendment risks, and alleviation of patient burden. These advancements are not only making trials more patient-centric but also ensuring more successful and efficient operational outcomes.

Strategies for Tackling Data Abstraction in Large Scale Enterprises

David Edwards, Director, Data Science Engineering, Amgen

David has worked in the Pharmaceutical / Biotechnology industry for 25+ years. He has held several different data and analytics roles working primarily with Clinical Trial Patient and Operational data. He currently works at Amgen where he leads a team of data science engineers within the Center for Design and Analysis. He has held several different data and analytics roles working primarily with Clinical Trial Patient and Operational data. He currently works at Amgen where he leads a team of data science engineers within the Center for Design and Analysis.

Abstract: In today's era of data-driven decision-making, grappling with the vast and ever-growing volumes of data has become a significant challenge for companies. This presentation will delve into the intricacies surrounding the wrangling of data within organizational ecosystems. As businesses accumulate diverse datasets from both internal and external sources, the process of accessing, integrating, and transforming this information into meaningful insights emerges as a critical bottleneck. The discussion will explore strategies for mitigating the complexities associated with accessing and wrangling structured data within the enterprise. This involves leveraging a set of simple abstractions, supported by open-source solutions like Vault by HashiCorp & DuckDB. At Amgen, the implementation of these abstractions aims to simplify data analysis for data scientists and statisticians while enabling engineers to deliver production-ready data pipelines.