Application of Machine Learning Algorithms to Predict Uncontrolled Diabetes Using the All of Us Research Program Data

Area of Science:

• * Computational biology and bioinformatics
• * Health informatics and predictive modeling

Background:

• * A significant need exists for robust predictive models for uncontrolled diabetes mellitus.
• * Existing models often lack the predictive power to identify individuals at high risk.

Purpose of the Study:

• * To evaluate the efficacy of various machine learning algorithms in predicting uncontrolled diabetes mellitus.
• * To identify key patient characteristics that serve as important predictors for uncontrolled diabetes.

Main Methods:

• * Application of machine learning algorithms including Random Forest, Extreme Gradient Boost, Logistic Regression, and Weighted Ensemble.
• * Utilized a dataset from the All of Us Research Program, including adult patients with diabetes.
• * Included demographic data, biomarkers, and hematological indices as predictive features.

Main Results:

• * The Random Forest model achieved the highest prediction accuracy (0.80) and Area Under the Curve (0.77).
• * Key predictors identified include potassium levels, body weight, aspartate aminotransferase, height, and heart rate.
• * Serum electrolytes and physical measurements emerged as crucial features for prediction.

Conclusions:

• * Machine learning, particularly the Random Forest model, shows significant promise for predicting uncontrolled diabetes.
• * Incorporating clinical characteristics like serum electrolytes and physical measurements enhances predictive capabilities.
• * These findings support the development of data-driven tools for proactive diabetes management.