Regional analysis of socioeconomic indicators and federated learning performance gain.

The table presents a systematic comparison across Brazil’s five macro-regions, linking key socioeconomic and health indicators with the study’s sample characteristics (number of hospitals and average patient cohort size per hospital). The primary outcome, the mean performance gain from federated learning (ΔAUC = AUC - AUC), is presented for each of the three models: Logistic Regression (LR), Multilayer Perceptron (MLP), and Random Forest (RF). The mean ΔAUC values for each region were calculated by averaging the hospital-specific mean ΔAUCs reported in S1 Table (for Logistic Regression), S2 Table (for MLP), and S3 Table (for Random Forest) across all hospitals within that geographical region. This allows for a higher-level investigation of the relationship between regional characteristics and the benefits of the federated approach.
(XLSX)