The Usage of Lasso, Ridge, and Linear Regression to Explore the Most Influential Metabolic Variables that Affect Fasting Blood Sugar in Type 2 Diabetes Patients

Abstract

Abstract Background and aims: To explore the most influential variables of fasting blood sugar (FBS) with three regression methods, to identify the existence chance of type 2 diabetes based on influential variables with logistic regression (LR), and to compare the three regression methods according to Mean Squared Error (MSE) value. Material and Methods: In this cross-sectional study, 270 patients suffering from type 2 diabetes for at least 6 months and 380 healthy people were participated. The Linear regression, Ridge regression, and Least Absolute Shrinkage and Selection Operator (Lasso) regression were used to find influential variables for FBS. Results: Among 15 variables (8 metabolic, 7 characteristic), Lasso regression selected HbA1c, Urea, age, BMI, heredity, and gender, Ridge regression selected HbA1c, heredity, gender, smoking status, and drug use, and Linear regression selected HbA1c as the most effective predictors for FBS. Conclusion: HbA1c is the most influential predictor of FBS among 15 variables according to the result of three regression methods. Controlling the variation of HbA1c leads to a more stable FBS. Beside FBS that should be checked before breakfast, maybe HbA1c could be helpful in diagnosis of Type 2 diabetes.