Hierarchical decision model for in vitro bilirubin content prediction from absorption spectrum of whole blood.

Significance: Bilirubin forms by the breakdown of heme proteins in the liver, but a newborn's sluggish liver can lead to elevated serum bilirubin levels that cross the blood-brain barrier and result in kernicterus. Earlier studies have used the 400 to 500 nm optical wavelength range to characterize the bilirubin content. There is not a universally established correlation among other wavelengths and the amount of bilirubin in clinical whole blood samples. Aim: We demonstrated that the amount of bilirubin could be quantified with ∼82% accuracy in a label-free, self-referenced manner using only a few wavelengths, viz. 468, 492, 500, 560, 605, 645, 660, and 675 nm, wherein band-averaged absorption measurements are used. Approach: We addressed the above problem by conducting a preliminary study containing 50 neonates through an absorption spectrum measurement of whole blood in 3 to 5 µl samples from the neonates. We constructed a hierarchical decision method that first grossly divides the 30 neonates of the training set into <10 mg/dl and ≥10 mg/dl bilirubin level cohorts. A subsequent boundary condition further divides the ≥10 mg/dl group into two >15 mg/dl and ≤15 mg/dl bilirubin level cohorts. A finer measure later predicted the bilirubin content of each of these groups as low (<10 mg/dl), medium (10 to 15 mg/dl), and high (>15 mg/dl). Results: Using this hierarchical decision model statistical approach, we quantified the amount of bilirubin in the 20 testing set samples with 82% accuracy. Conclusions: We formulated a biostatistical model in which we automated the spectrometric determination of total bilirubin in the whole blood for patients of neonatal hyperbilirubinemia.