Predicting and Preventing Nocturnal Hypoglycemia in Type 1 Diabetes Using Big Data Analytics and Decision Theoretic...

Area of Science:

• Biomedical Engineering
• Endocrinology
• Data Science

Background:

• Nocturnal hypoglycemia remains a significant challenge for type 1 diabetes (T1D) management, as individuals may not detect symptoms or alarms during sleep.
• Predictive models for nighttime hypoglycemia are crucial for preventing adverse events and improving patient safety.

Purpose of the Study:

• To develop and validate a machine learning model for predicting nocturnal hypoglycemia in people with T1D before bedtime.
• To assess the accuracy and potential impact of the predictive algorithm on reducing the incidence of nighttime hypoglycemia.

Main Methods:

• A support vector regression (SVR) model was trained using continuous glucose monitoring and insulin data from 124 individuals with T1D (22,804 nights).
• A decision theoretic criterion was applied to optimize the minimum glucose threshold for hypoglycemia risk prediction.
• Model accuracy was evaluated on a validation set using sensitivity, specificity, correlation, and root-mean-square error.

Main Results:

• The SVR model achieved 94.1% sensitivity in predicting nocturnal hypoglycemia events (<3.9 mmol/L).
• High correlation (R=0.71, P<0.001) was observed between predicted and actual minimum nocturnal glucose levels.
• In-silico simulations indicated a potential 77.0% reduction in nocturnal hypoglycemia without affecting time in target glucose range.

Conclusions:

• A data-driven SVR model, optimized with decision theory, can accurately predict nocturnal hypoglycemia risk at bedtime for individuals with T1D.
• This predictive capability holds promise for minimizing dangerous nighttime glucose lows and enhancing the safety of diabetes management.