Statistical resolution of melanin-dependent calibration bias and its correction

Establish, using a substantially larger controlled-desaturation cohort, whether melanin-dependent bias in pulse-oximetry saturation residuals is statistically significant and whether the melanin- and linewidth-corrected calibration removes that bias.

Background

The study compares an empirical linear calibration, a modified Beer–Lambert-law calibration assuming monochromatic sources, and a modified Beer–Lambert-law calibration that accounts for spectral coloring by melanin and source linewidth. The melanin-aware calibration produces a residual slope close to zero, whereas the other approaches produce larger positive slopes with respect to melanin volume fraction.

However, the available calibration cohort contains only 98 patients, and the fold-to-fold uncertainty intervals for all three slopes include zero. Consequently, the observed ordering is physically consistent with the proposed correction but does not statistically establish either the presence of melanin-dependent bias in these data or its removal. The authors estimate that a cohort roughly ten times larger may be required to resolve the effect.