Implementation of five machine learning methods to predict the 52-week blood glucose level in patients with type 2 diabetes

Abstract

ObjectiveFor the patients who are suffering from type 2 diabetes, blood glucose level could be affected by multiple factors. An accurate estimation of the trajectory of blood glucose is crucial in clinical decision making. Frequent glucose measurement serves as a good source of data to train machine learning models for prediction purposes. This study aimed at using machine learning methods to predict blood glucose for type 2 diabetic patients. We investigated various parameters influencing blood glucose, as well as determined the most effective machine learning algorithm in predicting blood glucose.Patients and methods273 patients were recruited in this research. Several parameters such as age, diet, family history, BMI, alcohol intake, smoking status et al were analyzed. Patients who had glycosylated hemoglobin less than 6.5% after 52 weeks were considered as having achieved glycemic control and the rest as not achieving it. Five machine learning methods (KNN algorithm, logistic regression algorithm, random forest algorithm, support vector machine, and XGBoost algorithm) were compared to evaluate their performances in prediction accuracy. R 3.6.3 and Python 3.12 were used in data analysis.ResultsThe statistical variables for which p< 0.05 was obtained were BMI, pulse, Na, Cl, AKP. Compared with the other four algorithms, XGBoost algorithm has the highest accuracy (Accuracy=99.54% in training set and 78.18% in testing set) and AUC values (1.0 in training set and 0.68 in testing set), thus it is recommended to be used for prediction in clinical practice.ConclusionWhen it comes to future blood glucose level prediction using machine learning methods, XGBoost algorithm scores the highest in effectiveness. This algorithm could be applied to assist clinical decision making, as well as guide the lifestyle of diabetic patients, in pursuit of minimizing risks of hyperglycemic or hypoglycemic events.

Prediction of glycosylated hemoglobin level in patients with cardiovascular diseases and type 2 diabetes mellitus with respect to anti-diabetic medication

“…Fu et al. ( 7 ) used XGBoost that is prone to overfitting. The contrast between performance of their model on training set (99%) and test set (68%) demonstrates overfitting of the training process.…”

“…To develop the predictive model, various machine learning algorithms were utilized, including Random Forest, Support Vector Machine (SVM), XGBoost, Linear Regression (Least Squares, Lasso, ElasticNet), Extra Trees, and k nearest neighbors (k-NN). These algorithms were chosen based on their suitability for regression task, as well as their ability to handle complex and high-dimensional data in similar environment ( 7 ).…”

“…Five different machine learning algorithms were used by Fu et al. ( 7 ). They found statistically significant variables such as body mass index (BMI), pulse, Na, Cl, Alkaline Phosphatase (AKP).…”

“…Our models outperformed Tao et al's best model ( 5) that got ROC AUC of 0.8 while our C Index achieved 0.86 in the first group. Fu et al (7) used XGBoost that is prone to overfitting. The contrast between performance of their model on training set (99%) and test set (68%) demonstrates overfitting of the training process.…”

A Supervised Machine Learning Approach with Feature Selection for Sex-Specific Biomarker Prediction

“…They found extreme Gradient Boost to be the preferred model for predicting blood glucose concentrations. The features they considered for training included age, sex, experimental grouping, family history, education level, dietary assessment, complications (retinopathy, kidney disease, peripheral neuropathy, peripheral atherosclerosis, intermittent claudication), hypertension, drinking status, smoking status, BMI, pulse rate, and key biochemical indicators (HDL, Hb, K, Na, Cl, CO2, Ca, P, AKP, GPT, GOT, rGT), however, they did not stipulate the final features selected for training [27]. Van Doorn et al’s study considered predicting blood glucose at 15 minutes and 60 minutes intervals.…”