The association between biochemical factors and type ۲ diabetes: a machine learning approach

Background: Several blood biomarkers have been related to the risk of type ۲ diabetes mellitus(T۲D); however, their predictive value has seldom been assessed using machine learning algorithms.Method: This cohort study was conducted on ۹۶۵۰ participants recruited from the MAshhadStroke and Heart Atherosclerotic disorders (MASHAD) study from ۲۰۱۰ to ۲۰۲۰. Individualswith previous T۲D (Free Blood Sugar >۱۲۶) were excluded. Serum levels of biochemical factorssuch as creatinine (Cr), high sensitivity C reactive protein (hs-CRP), Uric acid, alanine aminotransferase(ALT), aspartate aminotransferase (AST), direct and total bilirubin (BIL.D, BIL.T),lipid profile, besides body mass index (BMI), blood pressure, and age were evaluated throughLogistic Regression (LR) and Decision Tree (DT) methods to develop a predicting model forT۲D.Results: The comparison between diabetic and non-diabetic participants represented higher levelsof triglyceride (TG), LDL, cholesterol, ALT, BIL.D, and Uric acid in diabetic cases (p-value<۰.۰۵). The LR model indicated a significant association between TG, Uric acid, and hs-CRP, besidesage, sex, BMI, and blood pressure, with T۲D development. DT algorithm demonstrated Uricacid as the most determining factor in T۲D prediction, followed by age and TG. Furthermore, itobtained a ۳.۱ mg/dl cut-off for Uric acid, so that Uric acid <۳.۱, Age >=۴۷ and, TG >۲۰۰ resultedin an ۸۰% probability of developing T۲D.Conclusion: There was a significant association between triglyceride, Uric acid, and hs-CRP withT۲D development, along with age, BMI, and blood pressure through the LR and DT methods.

کلیدواژه ها:

Biochemical factors ، Type ۲ diabetes ، Machine learning ، Decision Tree ، Uric acid