Abstract A034: Model-free scoring system for risk prediction and early detection

There is a growing need to develop risk-prediction scoring systems for survival data that can facilitate early detection and guide clinical decision-making for early-onset cancers. Identifying important risk factors (e.g., genetic or molecular biomarkers) is particularly critical for patient screening and timely treatment recommendations in this population. However, most existing methods either rely on strong modeling assumptions (such as proportional hazards) or are restricted to binary outcomes, limiting their utility in this context. In this article, we propose a flexible approach that simultaneously selects important risk factors and constructs an optimal linear risk score by maximizing a pseudo-likelihood function based on the time-dependent area under the receiver operating characteristic curve. This framework is especially relevant for evaluating risk profiles and recommending personalized surveillance or early interventions for patients at elevated risk of early-onset cancers. We establish desirable theoretical properties of the method, including asymptotic normality and the oracle property after variable selection. Its performance is further assessed through simulation studies and illustrated with an application to hepatocellular carcinoma data, where the method is used to identify biomarkers predictive of early-stage disease and to support strategies for early detection and intervention. Citation Format: Ying Yuan. Model-free scoring system for risk prediction and early detection [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: The Rise in Early-Onset Cancers—Knowledge Gaps and Research Opportunities; 2025 Dec 10-13; Montreal, QC, Canada. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(23_Suppl):Abstract nr A034.