Abstract / Summary
Abstract Objective: Chagas disease is a parasitic infection that is endemic to Latin America and primarily transmitted by insects. When untreated, Chagas disease can cause cardiomyopathy. Serological testing capacities for Chagas disease are limited, but Chagas cardiomyopathy often manifests in electrocardiograms (ECGs), providing an opportunity to use ECG-based screening to prioritize patients for testing and treatment. Approach: The George B. Moody PhysioNet Challenge 2025 invited teams to develop algorithmic approaches for identifying potential cases of Chagas disease from ECGs. The Challenge metric, denoted by TPR@5%, measured the true positive rate of the top 5% of patients determined by an algorithm to be the most likely to have Chagas disease, reflecting the Brazilian serological testing capacity. The data included more than 360,000 ECGs from five Brazilian databases. We also developed an ensemble algorithm for the best-performing entries. Main results: This Challenge provided multiple innovations. First, we leveraged several datasets with labels from either patient reports or serological testing, providing a large dataset with weak labels and smaller datasets with strong labels. Second, we augmented the data to support model robustness and generalizability to unseen data sources. Third, we applied an evaluation metric, TPR@5%, that captured the local serological testing capacity for Chagas disease to frame the Challenge as a triage task. Fourth, we developed an ensemble model that combined the individual Challenge models to improve performance. These innovations provided reliable sensitivity at clinically relevant operating points. The performance of the top-performing Challenge model (Biomed-Cardio) varied performed well across datasets (TPR@5% of 0.468 on REDS-II and TPR@5% of 0.125 on ELSA-Brasil), and our ensemble-based approach demonstrated consistent performance across datasets (TPR@5% of 0.389 on REDS-II and TPR@5% of 0.146 on ELSA-Brasil). Significance: Over 630 participants from 111 teams submitted over 1300 entries during the Challenge, representing diverse approaches from academia and industry worldwide.