Abstract 4170: An AI-driven multimodal workflow for enhancing late-phase clinical trial outcome prediction

Abstract A major challenge in drug development is the gap between early-phase signals and late-stage success. Despite promising early-stage clinical trial readouts, fewer than 10% of drug candidates ultimately progress to regulatory approval, highlighting substantial gaps in predictive fidelity along the development pathway. This gap complicates decision-making, leading to increased development time, costs and risks. Key contributors to this challenge include the small sample sizes inherent to early-phase cohorts and the limited use of patient stratification, shown to be a contributing factor for higher success rates. Recent advances in artificial intelligence (AI) have introduced powerful tools that have the potential to support clinical trial progress evaluation. By integrating multimodal models with real-world data (RWD) within an AI-driven framework, Imagene AI is developing approaches to address the discrepancies between early and late-phase outcomes, and to support better evaluation of late-phase readouts, such as survival outcomes and biomarkers identification, based on early phase cohorts. In this study, we adopted a multimodal foundation-model strategy, built on a diverse set of foundation models. Among them, our digital pathology foundation model, CanvOI, played a central role in enabling prediction of large-cohort outcomes from small-cohort data. Models were trained on a limited sample set of Trastuzumab (Herceptin)-treated breast cancer patients with outcome data. We then generated Kaplan-Meier survival curves for this small cohort with and without an AI-augmented workflow and compared the results with published outcomes from a Phase III trial. Our findings show that the AI-augmented predictions better align with the Phase III clinical trial outcomes, suggesting that this approach has the potential to support more informed decisions using early-phase data. *ChatGPT was used for editing this abstract Confidentiality Notice: This document is confidential and contains proprietary information and intellectual property of Imagene AI LTD. Neither this document nor any of the information contained herein may be reproduced or disclosed under any circumstances without the express written permission of Imagene AI LTD. Please be aware that disclosure, copying, distribution or use of this document and the information contained therein is strictly prohibited. Citation Format: Inbal Gazy, Assaf Avinoam, Reva Basho, Jonathan Zalach. An AI-driven multimodal workflow for enhancing late-phase clinical trial outcome prediction abstract. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 4170.