Artificial Intelligence Neonatal Jaundice Risk Estimation Under Variable Transcutaneous Bilirubinometer Probe Wavelength Drift: A Spectral Calibration Transfer Evaluation of Healthcare Phototherapy Referral Equity

Transcutaneous bilirubinometry is the standard non-invasive screen for neonatal hyperbilirubinaemia, yet probe-to-probe wavelength drift silently displaces the spectral basis of every risk estimate built on it. the present analysis treats that displacement as a calibration-transfer problem and evaluates, by formal analysis rather than by field measurement, how much phototherapy-referral equity an artificial-intelligence risk model can lose before its decisions become unreliable. A linear measurement-error model is developed in which the indicated bilirubin is the true value plus a probe-specific spectral bias, and the bias is propagated through a fixed-threshold referral rule to give closed-form expressions for referral-probability distortion and for the equity gap between two probes of differing drift. The analysis is a design study: no observational data are used or claimed, and the operating characteristics are characterised under a declared variance structure with between-cluster SD 6.0, residual SD 4.5 and intraclass correlation 0.64. Under that structure, 52 clusters are required for 80% power to detect a 5-percentage-point effect in a before-after-control-impact design, and power was estimated across 2000 replications per design cell for cluster counts of 8, 16, 24 and 32 and effects of 3, 5 and 8 percentage points. The contribution is a transfer-aware referral rule and an explicit equity criterion that a calibration programme must meet before an AI jaundice model can be deployed across probes.