Artificial Intelligence Benchmarking of Federated Sepsis Early-Warning Models Under Heterogeneous District Hospital Label Noise: A Healthcare Fairness-Audit Evaluation

Federated learning promises to train sepsis early-warning models across district hospitals without centralising patient data, but the label noise generated by heterogeneous local coding practices threatens both the validity and the fairness of the resulting models. The framework specifies a before-after-control-impact design with a declared variance structure (between-cluster SD = 6.0, residual SD = 4.5, intraclass correlation = 0.640) and characterises the operating characteristics of the design across 2000 replications per cell for cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points.54 clusters are required to detect a 5-percentage-point effect at 80% power, and it derives the bias and coverage consequences of site-heterogeneous label noise for federated averaging. The contribution is a reproducible audit protocol that separates the statistical power of the evaluation design from the fairness properties of the model being evaluated, so that benchmarking claims about federated sepsis models can be stated with explicit boundary conditions.