Development of an equation to predict delta bilirubin levels using machine learning

Area of Science:

• Clinical Chemistry
• Biochemical Analysis
• Medical Diagnostics

Background:

• Delta bilirubin, a form of albumin-bound bilirubin, indicates impaired hepatic excretion of conjugated bilirubin.
• Current methods for measuring delta bilirubin are not suitable for real-time clinical laboratory diagnostics.

Purpose of the Study:

• To develop a practical, real-time method for estimating delta bilirubin concentrations.
• To identify key clinical variables that can predict high delta bilirubin levels.

Main Methods:

• Collected 210 samples and measured delta bilirubin using high-performance liquid chromatography.
• Utilized linear regression machine learning (1,000 iterations) for feature selection among 20+ clinical variables.
• Identified an optimal combination of variables for predicting delta bilirubin.

Main Results:

• Selected variables included total bilirubin, direct bilirubin, total protein, albumin, hemoglobin, icterus value (Iv), icterus index (Ii), and lipemia value (Lv).
• The combination of albumin, direct bilirubin, hemoglobin, Iv, and Lv provided the best predictive performance for high delta bilirubin.
• A predictive equation was derived: delta bilirubin = 0.35*Iv + 0.05*Lv - 0.23*direct bilirubin - 0.05*hemoglobin - 0.04*albumin + 0.10.

Conclusions:

• The developed equation offers a practical and easily implementable method for real-time delta bilirubin estimation.
• This approach can enhance the clinical utility of assessing hepatic excretion in routine laboratory settings.