Patient Clustering Improves Efficiency of Federated Machine Learning to predict mortality and hospital stay time using distributed Electronic Medical Records

TL;DR

This paper introduces a community-based federated learning algorithm that clusters distributed electronic medical records into meaningful groups, improving predictive accuracy and reducing communication costs while preserving data privacy.

Contribution

The novel CBFL algorithm effectively clusters non-IID EMRs into clinically meaningful communities for improved federated learning performance.

Findings

CBFL outperforms baseline FL in ROC AUC and PR AUC.

Communities' performance correlates with their dissimilarity to others.

Data remains local, enhancing privacy and reducing communication costs.

Abstract

Electronic medical records (EMRs) supports the development of machine learning algorithms for predicting disease incidence, patient response to treatment, and other healthcare events. But insofar most algorithms have been centralized, taking little account of the decentralized, non-identically independently distributed (non-IID), and privacy-sensitive characteristics of EMRs that can complicate data collection, sharing and learning. To address this challenge, we introduced a community-based federated machine learning (CBFL) algorithm and evaluated it on non-IID ICU EMRs. Our algorithm clustered the distributed data into clinically meaningful communities that captured similar diagnoses and geological locations, and learnt one model for each community. Throughout the learning process, the data was kept local on hospitals, while locally-computed results were aggregated on a server. Evaluation results show that CBFL outperformed the baseline FL algorithm in terms of Area Under the Receiver Operating Characteristic Curve (ROC AUC), Area Under the Precision-Recall Curve (PR AUC), and communication cost between hospitals and the server. Furthermore, communities' performance difference could be explained by how dissimilar one community was to others.