Swarm Learning And Clinical Data: Evaluating The Benefit of a New Node For The Whole Network

Digital technologies in healthcare have significantly enhanced clinical decision-making with complex data analysis. However, single clinical centers often lack sufficient data volume or diversity for effective modeling. Distributed learning solutions help to overcome these limitations, and in particular, Swarm Learning (SL) offers a privacy-preserving solution by allowing a decentralized and synchronized model training across multiple centers without direct data sharing. Swarm Learning's inherent focus on data privacy makes it well-suited for distributed healthcare analysis. However, no studies have quantified the impact of an external individual center that wants to join the swarm network in terms of performance and fairness. In this paper, we propose an approach for evaluating the benefits from the point of view of both a new center and the existing swarm network, based on the amount of data introduced by the new center. We also present seven different Key Performance Indicators (KPIs) to measure the impact of the new center under different aspects related to the performance variation, model stability, fairness, and quality of the newly introduced data. We apply these indicators using real Intensive Care Unit (ICU) data from the MIMIC-III and MIMIC-IV datasets, and we analyze the impact of the new center based on different data distribution scenarios and different numbers of centers in the network.