Abstract
Deep learning models on medical images are now widely used to capture signals regarding patients’ outcomes. In this talk, we discuss how AI models can improve the clinical development pipeline by reducing the time to market and enhancing prioritization of drugs using external datasets. In oncology, both the sample size requirements and the duration are keys for phase III trials. Recent randomized trials in newly diagnosed patients with DLBCL have taken on average more than 5 years between first enrollment and publication [Batlevi2018]. We will present how this can be reduced by adjusting the primary analysis on deep learning predictions of the outcome. Another challenge in clinical development is the decision whether to launch a phase III trial based on the results of a single-arm short-term phase II trial. This decision process can be enriched by an estimation of treatment effect based on the single-arm data and historical controls. We will discuss how recently developed synthetic control arm methods, which rely on machine learning, allow more precise estimation of treatment effect. Additionally, to overcome the short-term follow-up in early phases, we will discuss how digital surrogate endpoints, trained on previous phase III trials, can help to extract information on efficacy. Keywords: Bioinformatics; Computational and Systems Biology Conflicts of interests pertinent to the abstract P. Trichelair Employment or leadership position in a company: Owkin France SESSION 1: NEW THERAPEUTICS
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call