A Dual Mode Adaptive Basal-Bolus Advisor Based on Reinforcement Learning.

Area of Science:

• Artificial Intelligence in Medicine
• Diabetes Technology
• Computational Endocrinology

Background:

• Type 1 diabetes (T1D) management relies on accurate glucose monitoring.
• Self-monitoring of blood glucose (SMBG) and continuous glucose monitoring (CGM) are primary methods for glucose measurement.
• Personalized insulin dosing remains a challenge in T1D care.

Purpose of the Study:

• To develop and validate an adaptive basal-bolus algorithm (ABBA) for personalized insulin dose suggestions in T1D.
• To assess the ABBA's performance using both SMBG and CGM data inputs.
• To evaluate the algorithm's ability to optimize glucose control independent of monitoring technology.

Main Methods:

• The adaptive basal-bolus algorithm (ABBA) was developed using reinforcement learning, a type of artificial intelligence.
• The ABBA was validated in silico using a simulated population of 100 adults with T1D.
• Simulations incorporated realistic daily meal patterns and variability in insulin sensitivity, meal timing, and glucose measurements over three months.

Main Results:

• The ABBA demonstrated comparable performance when using either CGM or SMBG data as input.
• The algorithm provided personalized suggestions for basal rates and prandial insulin doses.
• The total daily insulin dose was not significantly influenced by the ABBA's recommendations.

Conclusions:

• AI-driven algorithmic approaches, like the ABBA, show promise for personalized adaptive insulin optimization in T1D.
• Glucose control can be achieved effectively regardless of the specific glucose monitoring technology used (CGM or SMBG).
• The ABBA represents a significant step towards intelligent, technology-independent diabetes management systems.