Prediction of glycosylated hemoglobin level in patients with cardiovascular diseases and type 2 diabetes mellitus...

Area of Science:

• Cardiology
• Endocrinology
• Data Science

Background:

• Glycosylated hemoglobin (HbA1c) monitoring is crucial for managing type 2 diabetes and cardiovascular disease.
• Frequent HbA1c measurements are costly, necessitating reliable estimation methods.
• Predicting future HbA1c levels can enhance treatment strategies and patient outcomes.

Purpose of the Study:

• To develop and validate machine learning models for predicting glycosylated hemoglobin levels in patients with type 2 diabetes and cardiovascular disease.
• To assess the efficacy of various machine learning algorithms in estimating future HbA1c values.

Main Methods:

• Recruited 93 patients with type 2 diabetes and cardiovascular disease, analyzing parameters like age, BMI, blood pressure, and cholesterol.
• Applied eight machine learning methods (KNN, Random Forest, SVM, Extra Trees, XGBoost, Linear Regression, Lasso, ElasticNet) to predict HbA1c levels two years in advance.
• Utilized feature selection and validation techniques, including Pearson's correlation, to refine models and compare performance on external test sets.

Main Results:

• Achieved high prediction accuracy with R² values up to 0.88 and Mean Absolute Error as low as 0.41.
• The best-performing models demonstrated strong predictive capabilities on external validation datasets.
• Specific metrics included R² = 0.88, C Index = 0.857, Accuracy = 0.846, MAE = 0.65 for one group, and R² = 0.86, C Index = 0.80, Accuracy = 0.75, MAE = 0.41 for the other.

Conclusions:

• Developed accurate machine learning algorithms for glycosylated hemoglobin prediction.
• These models can support clinical decision-making for anti-diabetic medication prescription.
• The findings contribute to achieving better glycemic control in patients with type 2 diabetes and cardiovascular disease.