AI-PRO and AI-DEA: Artificial Intelligence for Patient-Reported Outcomes and Deployment of EPC Access.

The present project aims to advance the clinical management of patients with multiple myeloma (MM) through the development and validation of two innovative artificial intelligence (AI)-based models: AI-PRO (AI for Patient-Reported Outcomes) and AI-DEA (AI for the Deployment of EPC Access). These models are specifically designed to enhance the integration of Patient-Reported Outcomes (PROs) and Early Palliative Care (EPC) into routine clinical practice, addressing two critical unmet needs: the limited predictive utility of current PROs and the underutilization of EPC services. AI-PRO is an advanced model based on Transformer Reinforcement Learning, a cutting-edge AI architecture that combines the interpretative capacity of Transformers with the iterative optimization of Reinforcement Learning. AI-PRO will optimize the selection, interpretation, and prognostic value of PROs by integrating longitudinal patient-reported data with other unstructured clinical data (e.g., laboratory and diagnostic results, clinical letters, interview transcripts). It will be trained on a retrospective dataset of 500 MM patients and validated both retrospectively and prospectively on a cohort of 650 MM patients. AI-PRO will refine PRO instruments by eliminating redundant items, identifying clinically informative symptoms, and linking them to disease progression. It will also detect inconsistencies between patient-reported symptoms and clinical observations, monitor symptom dynamics over time, and uncover symptom clusters based on disease stage and treatment response. Explainability mechanisms inherent in the Transformer model will allow clinicians to visualize the most informative PRO items and the interactions driving the model’s predictions, facilitating trust and clinical usability. AI-DEA, building on results from AI-PRO, will predict the optimal timing for EPC initiation on a per-patient basis. It will be trained on a large retrospective cohort of 1,100 MM patients and prospectively validated on 150 MM patients. By integrating PRO dynamics with molecular and genetic disease markers, toxicity profiles, anticipated treatment transitions, and real-world constraints (e.g., EPC workforce and patient load), AI-DEA will provide clinically actionable recommendations that support hematologists in a decision domain historically marked by uncertainty and inconsistency. Expected outcomes include the development of next-generation PROs with validated prognostic utility and a clinical tool for EPC referral, directly usable by clinicians. This project is original in both methodology and clinical application. It transforms PROs into predictive, AI-driven tools for clinical decision-making and introduces a novel, data-informed approach to EPC timing in hematology. The proposal builds on solid preliminary work conducted by the PI and the proposing group, relies on collaboration with the GIMEMA network, and is fully aligned with the EU Horizon Europe “Cancer” Mission.