Personalized Insulin Adjustment With Reinforcement Learning: An In-Silico Validation for People With Diabetes on Intensive Insulin Treatment

Despite recent advances in insulin preparations and technology, adjusting insulin remains an ongoing challenge for the majority of people with type 1 diabetes (T1D) and longstanding type 2 diabetes (T2D). In this study, we propose an enhanced version of the Adaptive Basal-Bolus Advisor (ABBA), a personalized insulin treatment recommendation system based on an actor-critic, model-free reinforcement learning approach. ABBA is designed for individuals with T1D and T2D, performing self-monitoring blood glucose measurements and multiple daily insulin injection therapy. We developed and evaluated the effectiveness of the enhanced version of ABBA to achieve better time-in-range (TIR) for individuals with T1D and T2D, compared to the use of a standard basal-bolus advisor (BBA). The in-silico test was performed using an FDA-accepted population, including 101 simulated adults with T1D and 101 with T2D. The in-silico evaluation shows that the updated version of ABBA significantly improved TIR by $9.54 \pm 7.76\%$ and $11.80 \pm 10.76\%$ in individuals with T1D and T2D, respectively, and significantly reduced both times below- and above-range, compared to BBA. After two months, TIR increased by $11.94 \pm 8.39\%$ and $7.74 \pm 5.53\%$ in T1D and T2D, respectively, on ABBA, while BBA showed only modest changes over time with variations of $1.32 \pm 1.41\%$ and $1.45 \pm 1.47\%$ , respectively. On a subgroup of people with T1D, the old version of ABBA was outperformed by $6.4 \pm 4.9\%$ , $5.8\pm 2.1\%$ , and $0.6 \pm 5.1\%$ in TIR, TBR, and TAR, accordingly. This personalized method for adjusting insulin has the potential to further optimize glycemic control and support people with T1D and T2D in their daily self-management. Our results warrant ABBA to be trialed for the first time in humans.