(C-123) Application of Artificial Intelligence (AI) and Machine learning (ML) in oncology drug development using real-world data: A Targeted Review

Background: Remarkable progress in AI/ML and the availability of RWD in healthcare have sparked interest in integrating both fields within oncology. We aimed to examine literature evaluating AI/ML applications in RWD across oncology.

Objectives: To summarize and evaluate current AI/ML applications using RWD, specifically highlighting clinical drug development (CDD) within oncology.

Methods: A targeted literature review was conducted focused on studies published from Jan 2020-May 2024. Embase®, PubMed, and Cochrane library were searched using cancer-related, AI/ML techniques/algorithm, and RWE (real world evidence)/RWD terms. Research-related themes were prioritized.

Results: Of the 5156 citations initially screened, 4988 were excluded for reasons such as reviews/editorial, non-cancer studies, non-RWE/AI studies or non-prioritized themes. The remaining 168 studies were included and categorized into prioritized themes of CDD (n=19 studies), real-world data management and analysis (33), disease burden (27), and treatment outcomes & mortality prediction (89). Majority of CDD studies were conducted in North America (10), followed by Europe (3), China (3), and Australia (3). Lung cancer was the most explored (7), followed by breast cancer (5) and colorectal cancer (2). Most CDD studies focused on clinical trial enrollment screening and matching (11), followed by adverse event detection (3). Other areas included real-world response rates (1), survival prediction using real-world data (1), identifying clinical inflection points (1), drug repurposing (1), and generating synthetic patient data for clinical trials (1). Of the 11 clinical trial enrollment/ matching studies, 5 reported clinical trial phases, featuring Phase 3 and Phase 2 trials. Among all CDD studies, 9 studies used pre-developed AI algorithms (e.g., CTMS (clinical trial matching system), Mendel.AI, and Normalizing flows (NFlow)), 8 created their own (e.g., AI-powered Case Matching Assistant (ACMA) system, Digital Research Assistant (DRA) platform and Novel trial search tool) and 2 included both. AI-based CTMS (4 studies) was the most commonly used pre-developed AI system. Common evaluation metrics included accuracy, sensitivity, specificity, precision, area under the curve (AUC) and F1 score, e.g., 4 studies reported that AI-based CTMS improved patient matching accuracy (ranges: 64.3%-100%). In addition, 5 studies reported AI/ML saved screening time significantly, in which one study showed a reduction in screening time from 150 hours to 2 hours.

Conclusions: In CDD, the majority of AI/ML applications were for clinical trial enrollment and matching. AI/ML applications show promise to enhance oncology clinical trials by improving efficiency and reducing time and resources.