Artificial Intelligence Sparse-Data Transfer Learning for Neonatal Jaundice Risk Inference Under Variable Transcutaneous Bilirubin Device Calibration: A computed Healthcare Safety Evaluation

Transcutaneous bilirubin (TcB) devices enable non-invasive neonatal jaundice screening, but their calibration drifts across device generations, skin phototypes and clinical sites, and sparse local data prevent site-specific recalibration. The framework combines a shared spectral representation with device-specific affine correction and a risk-calibrated decision threshold, and it is evaluated through a replication study that characterises the operating characteristics of the before-after-control-impact design under a declared variance structure (between-cluster SD = 6.0, residual SD = 4.5, intraclass correlation = 0.64). Power was estimated across 2000 design cells with 8, 16, 24 and 32 clusters and effect sizes of 3, 5 and 8 percentage points.50 clusters are required to detect a 5-percentage-point effect at 80% power, and that the minimum detectable effect at 80% power falls steeply as cluster count rises from 8 to 32. The framework proposes that calibration-aware transfer learning can reduce false-negative risk under device heterogeneity, but the design analysis indicates that sparse-data safety claims require cluster counts well above those typical of single-site evaluations. The contribution is a formalisation of calibration shift as a domain-adaptation problem and a design-level account of the evidence needed before such inference can be deployed safely.