RE-SAMPLE Platform for training and use of COPD exacerbation risk prediction models

<bold>Background:</bold> Many COPD patients experience multiple chronic conditions increasing their burden, healthcare consumption and costs. Due to the interplay of the diseases and overlapping symptoms, disease management is complex. <bold>Objective:</bold> Developing personalised exacerbation risk prediction models for shared decision-making based on combined models of clinical data and real-world data (RWD) to increase patient involvement. <bold>Methods:</bold> Three collaborating European hospitals provided retrospective clinical data enriching holistic data of a current prospective cohort study. The RE-SAMPLE platform manages the clinical data and RWD collected by patients using the Healthentia App at edge nodes on-site in each hospital. This allows for privacy preserving federated training of machine learning (ML) models. The combined models are available in all hospitals to provide personalised predictions and explanations displayed in the clinical dashboard of the Healthentia portal app. <bold>Results:</bold> Edge nodes enable the use of clinician front ends for monitoring and shared decision-making. Cooperative training of models is functional. The analysis of predictive models trained on retrospective data shows that the number of COPD exacerbations in the previous year is the most important predictor for COPD exacerbation risk within following year (single feature model accuracy 75.6% on a balanced dataset, n=1068). This is in line with literature and is evidence for the suitability of the models. <bold>Conclusion:</bold> The RE-SAMPLE platform enables data storage, synchronisation and management for patient monitoring and privacy-preserving training of federated ML models suitable for use in shared-decision making in patients with COPD and comorbidities.