A0850
Title: Multilayer perceptron for the economic analysis of diagnostic inappropriateness in cancer networks
Authors: Anna Pia Di Iorio - University of Naples Parthenope (Italy) [presenting]
Giorgia Rivieccio - Parthenope University (Italy)
Sandro Pignata - Istituto Nazionale Tumori IRCCS Fondazione Pascale (Italy)
Francesco Schiavone - University of Naples Parthenope (Italy)
Abstract: Innovation is pivotal for optimizing patient care pathways in healthcare, yet its integration into clinical practice demands models that effectively incorporate new technologies. Regional cancer networks serve as strategic structures to coordinate care and promote diagnostic appropriateness, key to cost control and sustainability of health systems. In this context, machine learning-based predictive models are increasingly utilized. The aim is to present a model designed to estimate costs related to diagnostic inappropriateness in oncology, defined as tests not aligned with clinical guidelines or unnecessarily repeated. A multi-layer perceptron (MLP) artificial neural network was used to capture nonlinear relationships among clinical, organizational, and sociodemographic variables. The model was trained on a structured dataset including patients' sociodemographic profiles and oncological care journeys. The target variable was the proportion of inappropriate diagnostic costs per patient. To ensure explainability, SHAP (SHapley Additive exPlanations) analysis was applied to identify key predictors. The most significant factor was time to first multidisciplinary team evaluation, followed by diagnostic modality, distance from the treatment center, and patient age. The results support the development of targeted interventions to reduce inefficiencies and improve diagnostic appropriateness in oncology, contributing to more sustainable and patient-centered care.
