Personalized models improve disease diagnosis across hospitals nearby
Geographically Regularized AUC-Maximizing Personalized Federated Learning
Machine Learning
Summary
During disease outbreaks, hospitals need good computer models to help predict who might be sick. But hospitals usually cannot share detailed patient data with each other because of privacy rules. The authors created a way for hospitals close to each other to build similar but customized models without sharing private data, and these models focus on improving the AUC, a way to measure accuracy. Their method helps predict diseases better, especially when nearby hospitals have similar types of patients.
What this means in practice
- •For hospital data teams: Build customized diagnostic models that respect patient privacy and improve predictions by linking nearby hospitals with similar patient patterns.
- •For epidemiology modelers: Develop better risk prediction tools that account for regional differences in disease data without pooling sensitive patient information.
Authors
Mayu Hiraishi, Kensuke Tanioka, Toshio Shimokawa
Abstract
Accurate diagnostic and risk-prediction models are important for supporting clinical decision-making during infectious disease outbreaks. However, privacy and governance requirements may restrict patient-level data sharing across healthcare institutions, and data distributions often vary. Moreover, AUC is widely used to evaluate discriminative performance, motivating its direct optimization in model development. We propose geographically regularized AUC-maximizing personalized federated learning (GrAUC-PFL), which directly optimizes a smooth pairwise AUC surrogate to learn personalized models while keeping patient-level data local and accounting for institutional heterogeneity. Graph-based regularization encourages geographically neighboring institutions to have similar coefficient vectors while retaining a personalized models. Simulations and a real-data application suggest improved discriminative performance, particularly when geographically neighboring institutions have similar data-generating characteristics.