Abstract PO5-21-04: Leveraging Digital Twins for Patient Stratification and Treatment Optimization in geriatric oncology: A Breast Cancer Multivariate Clustering Analysis

Abstract

Abstract Context: Geriatric oncology necessitates an accurate evaluation of a patient's health status and cancer characteristics. Digital twins, virtual replicas of a patient and their cancer, can aid in predicting real-world behaviors and support therapeutic decision-making. Objectives: This study aims to demonstrate how digital twins can optimize treatments by simulating various scenarios based on the individual characteristics of patients and their cancers. Methods: Creation of a digital twin relies on the use of advanced manifold learning technologies. A model is built to represent the variability present in the reference cohort, capturing complex relationships and inherent data structures. A reference digital twin is generated by drawing samples from this model, thus reflecting the characteristics distribution of the reference population. Stratification (MeanShift algorithm) enables cohort structure visualization and classical statistics estimation for each stratum. Multivariate analysis was performed on a French monocentric population of 1345 patients aged over 70 years who underwent surgery for HER2-negative early breast cancer. Results: 8 cluster, created by manifold learning based on biological, demographic, and tumoral variables, exhibits unique five-year survival rates and influential factors. The clusters represent specific patient subgroups, each with its distinct set of characteristics and prognostic indicators. Table 1 provides a summary of these variables' averages within each cluster. we can thus see that the variables of clusters 3 and 6 (including only HR+ cancers) do not show statistically significant differences, unlike their 5-year survival rates (cluster 3 : 80,9% (CI95 :70-88.5) and 63.9% (CI95 :47.6-77.5) for cluster 6, p= 0,04). Provided with model results, the relative importance of these variables varies from one cluster to another, underscoring the heterogeneity among patient subgroups. Interestingly, Hormonal receptor status, nodal involvement and tumoral grade were not the predominant variables in these clusters, which challenges traditional perspectives. Table 1 : Mean value of variables per cluster (SBR grade 0=grade 1 and 2 ; 1= grade 3) Discussion: The findings underscore the vast variability of individual variable importance across clusters. Digital twins enabled the modeling of these complex interactions, providing valuable insights for clinical decision-making. Conclusion: Digital twins offer a valuable tool for personalizing care in oncogeriatrics, facilitating the identification of patient subgroups, yielding more accurate treatment outcomes, and guiding future research directions. The results affirm that incorporating digital twins into oncogeriatric care could potentially lead to the enhanced personalization of therapeutic approaches, ultimately improving patient outcomes. Table. Citation Format: Pierre Heudel, Mashal Ahmed, Felix Renard. Leveraging Digital Twins for Patient Stratification and Treatment Optimization in geriatric oncology: A Breast Cancer Multivariate Clustering Analysis [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO5-21-04.