Multimodal machine learning integrates clinical and comorbidity data to predict breast cancer prognosis and treatment
Yongsheng Luo1, Hai Huang2, Tet Khuan Chen3
1Zhuhai College of Science and Technology, Zhu Hai, Guang Dong, China. vincent@zcst.edu.cn.
Abstract:
Breast cancer prognosis and treatment outcomes are shaped by intricate interplay between tumor biology, systemic comorbidities, and patient-reported functional status. This paper introduces a framework to address critical gaps in precision oncology by systematically integrating multimodal data-including clinical parameters, comorbidity profiles, and patient-reported outcomes (PROs)-to refine prognostic stratification and therapeutic decision-making. Section 1 contextualizes the clinical relevance of comorbidity burden and quality-of-life metrics in breast cancer care, emphasizing their underutilization in conventional models. Section 2 synthesizes prior efforts in machine learning and oncology, identifying limitations in single-modality approaches and underscoring the necessity for holistic data integration. Section 3 describes the harmonization of structured clinical variables, binary comorbidity flags, and EORTC QLQ-C30/BR23 scores from a prospective cohort of 1727 patients, alongside methodological innovations for handling missingness and feature engineering. Section 4 outlines experimental protocols for model training, cross-validation, and ablation studies to quantify the incremental value of multimodal inputs. Section 5 highlights key insights into prognostic heterogeneity, including the differential impacts of specific comorbidities (e.g., urinary tract infections, depression) and PRO domains (e.g., fatigue, global health status) on survival trajectories. Section 6 interprets these findings through the lens of clinical utility, addressing scalability challenges and ethical considerations for real-world deployment. Finally, Sect. 7 proposes translational pathways for embedding multimodal analytics into risk-adapted treatment paradigms. This work bridges computational innovation with patient-centered care, offering a roadmap for advancing precision oncology through data-driven, multidimensional modeling.
