P-L02 Machine learning to evaluate liver reserve function based on venous blood biochemistry. (16th December 2021)

• Record Type: Journal Article
• Authors: Wang, Zhengfeng; long shi, zhi; zheng, jnxing; rui, shaozhen; zhang, hui; li, xun; zhou, wence
• Is Part Of: British journal of surgery. Volume 108:Supplement 9(2021)
• Publication Date: 2021-12-16
• Subjects: Surgery -- Periodicals; 617.005
• DOI: 10.1093/bjs/znab430.095
• Languages: English
• ISSNs: 0007-1323

Abstract

Abstract: Background: The accurate and comprehensive evaluation of liver reserve function is crucial for daily follow-up and medical treatment of patients with liver disease. However, existing techniques are too complex and costly for universal implementation.To develop a convenient, reliable method to evaluate liver reserve function based on eight biochemical indicators measured from venous blood. Methods: Blood test results (albumin (Alb), total bilirubin (TBIL), prothrombin time (PT), international normalized ratio (INR), total cholesterol (TC), cholinesterase (ChE), aspartate amino transferase (AST), and alanine transaminase (ALT)) were collected retrospectively from 660 patients treated at the first hospital of Ianzhou University from 2016 to 2018. As the reference standard for liver reserve function, indocyanine green (ICG) clearance test results were also collected from the same patients at the same times. The patient data were processed and analyzed to construct a machine learning model, eXtreme Gradient Boosting (XGBoost), and a generalized linear model (GLM) to predict liver reserve function based on the eight biochemical indicators. Results: Results showed that the predicted XGBoost values were closely correlated with the actual ICG 15-minute retention rates (R = 0.969, R2 = 0.939), while the GLM values had a relatively low correlation (R = 0.566, R2 = 0.320). These findings indicate that the developed model can be used to evaluate liver reserve function with comparable performance to the ICG clearance test. Furthermore, the XGBoost model exhibited superior prediction compared with the GLM. Hence, the XGBoost model developed using machine learning can be utilized to evaluate liver reserve function from eight biochemical indicators that are closely related to liver function, commonly used clinically, and easier to obtain than ICG clearance measures. Conclusions: The results predicted by the XGBoost model were highly accurate when compared with the results of the actual ICG test, demonstrating the strong practical clinical value of the model.