This project advances personalized oncology by developing a data-driven decision-support framework for metastatic breast cancer patients, grounded in causal inference. Leveraging the large-scale ESME Metastatic Breast Cancer cohort (2008–2024), the project applies modern causal inference techniques to real-world, multivariate time series data characterized by irregular sampling and treatment heterogeneity. By integrating the Potential Outcomes framework with causal graphs and machine learning, the project estimates individual treatment effects (ITE) and uncovers patterns of therapeutic benefit across patient subgroups. A key challenge addressed is the complexity of irregular, longitudinal clinical data and the need to disentangle treatment effects from confounding in a dynamic clinical setting.
Immunotherapy has transformed the therapeutic landscape for metastatic melanoma, offering long-term survival benefits to a subset of patients. However, predicting which individuals will respond remains a major clinical challenge. This project develops explainable and clinically applicable artificial intelligence models that integrate multimodal data—spanning clinical features, routine blood markers, and exploratory biomarkers—to improve the prediction of immunotherapy response and patient survival. A central objective is to design robust methods for selecting predictive features from clinical and routinely collected biological data, both before and after treatment initiation. The project investigates the added value of post-treatment routine biomarkers compared to baseline features, with the hypothesis that dynamic changes may capture early signals of therapeutic efficacy. For more advanced, non-routine markers—such as immune signatures or omics data available on smaller patient subsets—the project evaluates their marginal utility in enhancing prediction performance when combined with more widely available clinical data.