Advanced decision support system for individuals with diabetes on multiple daily injections therapy using reinforcement learning and nearest-neighbors: In-silico and clinical results.

Many individuals with diabetes on multiple daily insulin injections therapy use carbohydrate ratios (CRs) and correction factors (CFs) to determine mealtime and correction insulin boluses. The CRs and CFs vary over time due to physiological changes in individuals' response to insulin. Errors in insulin dosing can lead to life-threatening abnormal glucose levels, increasing the risk of retinopathy, neuropathy, and nephropathy. Here, we present a novel learning algorithm that uses Q-learning to track optimal CRs and uses nearest-neighbors based Q-learning to track optimal CFs. The learning algorithm was compared with the run-to-run algorithm A and the run-to-run algorithm B, both proposed in the literature, over an 8-week period using a validated simulator with a realistic scenario created with suboptimal CRs and CFs values, carbohydrate counting errors, and random meals sizes at random ingestion times. From Week 1 to Week 8, the learning algorithm increased the percentage of time spent in target glucose range (4.0 to 10.0 mmol/L) from 51 % to 64 % compared to 61 % and 58 % with the run-to-run algorithm A and the run-to-run algorithm B, respectively. The learning algorithm decreased the percentage of time spent below 4.0 mmol/L from 9 % to 1.9 % compared to 3.4 % and 2.3 % with the run-to-run algorithm A and the run-to-run algorithm B, respectively. The algorithm was also assessed by comparing its recommendations with (i) the endocrinologist's recommendations on two type 1 diabetes individuals over a 16-week period and (ii) real-world individuals' therapy settings changes of 23 individuals (19 type 2 and 4 type 1) over an 8-week period using the commercial Bigfoot Unity Diabetes Management System. The full agreements (i) were 89 % and 76 % for CRs and CFs for the type 1 diabetes individuals and (ii) was 62 % for mealtime doses for the individuals on the commercial Bigfoot system. Therefore, the proposed algorithm has the potential to improve glucose control in individuals with type 1 and type 2 diabetes.

Postprandial glucose-management strategies in type 1 diabetes: Current approaches and prospects with precision medicine and artificial intelligence.

Postprandial glucose control can be challenging for individuals with type 1 diabetes, and this can be attributed to many factors, including suboptimal therapy parameters (carbohydrate ratios, correction factors, basal doses) because of physiological changes, meal macronutrients and engagement in postprandial physical activity. This narrative review aims to examine the current postprandial glucose-management strategies tested in clinical trials, including adjusting therapy settings, bolusing for meal macronutrients, adjusting pre-exercise and postexercise meal boluses for postprandial physical activity, and other therapeutic options, for individuals on open-loop and closed-loop therapies. Then we discuss their challenges and future avenues. Despite advancements in insulin delivery devices such as closed-loop systems and decision-support systems, many individuals with type 1 diabetes still struggle to manage their glucose levels. The main challenge is the lack of personalized recommendations, causing suboptimal postprandial glucose control. We suggest that postprandial glucose control can be improved by (i) providing personalized recommendations for meal macronutrients and postprandial activity; (ii) including behavioural recommendations; (iii) using other personalized therapeutic approaches (e.g. glucagon-like peptide-1 receptor agonists, sodium-glucose co-transporter inhibitors, amylin analogues, inhaled insulin) in addition to insulin therapy; and (iv) integrating an interpretability report to explain to individuals about changes in treatment therapy and behavioural recommendations. In addition, we suggest a future avenue to implement precision recommendations for individuals with type 1 diabetes utilizing the potential of deep reinforcement learning and foundation models (such as GPT and BERT), employing different modalities of data including diabetes-related and external background factors (i.e. behavioural, environmental, biological and abnormal events).

A Randomized Crossover Trial to Compare Automated Insulin Delivery (the Artificial Pancreas) With Carbohydrate Counting or Simplified Qualitative Meal-Size Estimation in Type 1 Diabetes.

OBJECTIVE: Qualitative meal-size estimation has been proposed instead of quantitative carbohydrate (CHO) counting with automated insulin delivery. We aimed to assess the noninferiority of qualitative meal-size estimation strategy. RESEARCH DESIGN AND METHODS: We conducted a two-center, randomized, crossover, noninferiority trial to compare 3 weeks of automated insulin delivery with 1) CHO counting and 2) qualitative meal-size estimation in adults with type 1 diabetes. Qualitative meal-size estimation categories were low, medium, high, or very high CHO and were defined as <30 g, 30-60 g, 60-90 g, and >90 g CHO, respectively. Prandial insulin boluses were calculated as the individualized insulin to CHO ratios multiplied by 15, 35, 65, and 95, respectively. Closed-loop algorithms were otherwise identical in the two arms. The primary outcome was time in range 3.9-10.0 mmol/L, with a predefined noninferiority margin of 4%. RESULTS: A total of 30 participants completed the study (n = 20 women; age 44 (SD 17) years; A1C 7.4% [0.7%]). The mean time in the 3.9-10.0 mmol/L range was 74.1% (10.0%) with CHO counting and 70.5% (11.2%) with qualitative meal-size estimation; mean difference was -3.6% (8.3%; noninferiority P = 0.78). Frequencies of times at <3.9 mmol/L and <3.0 mmol/L were low (<1.6% and <0.2%) in both arms. Automated basal insulin delivery was higher in the qualitative meal-size estimation arm (34.6 vs. 32.6 units/day; P = 0.003). CONCLUSIONS: Though the qualitative meal-size estimation method achieved a high time in range and low time in hypoglycemia, noninferiority was not confirmed.

Low-Dose Empagliflozin as Adjunct to Hybrid Closed-Loop Insulin Therapy in Adults With Suboptimally Controlled Type 1 Diabetes: A Randomized Crossover Controlled Trial.

OBJECTIVE: To assess whether low doses of empagliflozin as adjunct to hybrid closed-loop therapy improve glycemia compared with placebo in adults with type 1 diabetes (T1D) who are not able to achieve targets with the system alone. RESEARCH DESIGN AND METHODS: A double-blind crossover randomized controlled trial was performed in adults with suboptimally controlled T1D (HbA1c 7.0-10.5%) who were not able to achieve a target time in range (3.9-10.0 mmol/L) ≥70% after 14 days of hybrid closed-loop therapy. Three 14-day interventions were performed with placebo, 2.5 mg empagliflozin, or 5 mg empagliflozin as adjunct to the McGill artificial pancreas. Participants were assigned at a 1:1:1:1:1:1 ratio with blocked randomization. The primary outcome was time in range (3.9-10.0 mmol/L). Analysis was by intention to treat, and a P value <0.05 was regarded as significant. RESULTS: A total of 24 participants completed the study (50% male; age 33 ± 14 years; HbA1c 8.1 ± 0.5%). The time in range was 59.0 ± 9.0% for placebo, 71.6 ± 9.7% for 2.5 mg empagliflozin, and 70.2 ± 8.0% for 5 mg empagliflozin (P < 0.0001 between 2.5 mg empagliflozin and placebo and between 5 mg empagliflozin and placebo). Mean daily capillary ketone levels were not different between arms. There were no serious adverse events or cases of diabetic ketoacidosis or severe hypoglycemia in any intervention. CONCLUSIONS: Empagliflozin at 2.5 and 5 mg increased time in range during hybrid closed-loop therapy by 11-13 percentage points compared with placebo in those who otherwise were unable to attain glycemic targets. Future studies are required to assess long-term efficacy and safety.

Empagliflozin add-on therapy to closed-loop insulin delivery in type 1 diabetes: a 2 × 2 factorial randomized crossover trial.

There is a need to optimize closed-loop automated insulin delivery in type 1 diabetes. We assessed the glycemic efficacy and safety of empagliflozin 25 mg d-1 as add-on therapy to insulin delivery with a closed-loop system. We performed a 2 × 2 factorial randomized, placebo-controlled, crossover two-center trial in adults, assessing 4 weeks of closed-loop delivery versus sensor-augmented pump (SAP) therapy and empagliflozin versus placebo. The primary outcome was time spent in the glucose target range (3.9-10.0 mmol l-1). Primary comparisons were empagliflozin versus placebo in each of closed-loop or SAP therapy; the remaining comparisons were conditional on its significance. Twenty-four of 27 randomized participants were included in the final analysis. Compared to placebo, empagliflozin improved time in target range with closed-loop therapy by 7.2% and in SAP therapy by 11.4%. Closed-loop therapy plus empagliflozin improved time in target range compared to SAP therapy plus empagliflozin by 6.1% but by 17.5% for the combination of closed-loop therapy and empagliflozin compared to SAP therapy plus placebo. While no diabetic ketoacidosis or severe hypoglycemia occurred during any intervention, uncomplicated ketosis events were more common on empagliflozin. Empagliflozin 25 mg d-1 added to automated insulin delivery improves glycemic control but increases ketone concentration and ketosis compared to placebo.

Alleviating carbohydrate counting with a FiASP-plus-pramlintide closed-loop delivery system (artificial pancreas): Feasibility and pilot studies.

AIM: To assess whether a FiASP-and-pramlintide closed-loop system has the potential to replace carbohydrate counting with a simple meal announcement (SMA) strategy (meal priming bolus without carbohydrate counting) without degrading glycaemic control compared with a FiASP closed-loop system. MATERIALS AND METHODS: We conducted a 24-hour feasibility study comparing a FiASP system with full carbohydrate counting (FCC) with a FiASP-and-pramlintide system with SMA. We conducted a subsequent 12-day outpatient pilot study comparing a FiASP-and-placebo system with FCC, a FiASP-and-pramlintide system with SMA, and a FiASP-and-placebo system with SMA. Basal-bolus FiASP-and-pramlintide were delivered at a fixed ratio (1 U:10 µg). Glycaemic outcomes were measured, surveys evaluated gastrointestinal symptoms and diabetes distress, and participant interviews helped establish a preliminary coding framework to assess user experience. RESULTS: Seven participants were included in the feasibility analysis. Time spent in 3.9-10 mmol/L was similar between both interventions (81%-84%). Four participants were included in the pilot analysis. Time spent in 3.9-10 mmol/L was similar between the FiASP-and-placebo with FCC and FiASP-and-pramlintide with SMA interventions (70%), but was lower in the FiASP-and-placebo with SMA intervention (60%). Time less than 3.9 mmol/L and gastrointestinal symptoms were similar across all interventions. Emotional distress was moderate at baseline, after the FiASP-and-placebo with FCC and SMA interventions, and fell after the FiASP-and-pramlintide with SMA intervention. SMA reportedly afforded participants flexibility and reduced mealtime concerns. CONCLUSIONS: The FiASP-and-pramlintide system has the potential to substitute carbohydrate counting with SMA without degrading glucose control.

Long-term use of the hybrid artificial pancreas by adjusting carbohydrate ratios and programmed basal rate: A reinforcement learning approach.

BACKGROUND AND OBJECTIVES: The hybrid artificial pancreas regulates glucose levels in people with type 1 diabetes. It delivers (i) insulin boluses at meal times based on the meals' carbohydrate content and the carbohydrate ratios (CRs) and (ii) insulin basal, between meals and at night, continuously modulated around individual-specific programmed basal rate. The CRs and programmed basal rate significantly vary between individuals and within the same individual with type 1 diabetes, and using suboptimal values in the hybrid artificial pancreas may degrade glucose control. We propose a reinforcement learning algorithm to adaptively optimize CRs and programmed basal rate to improve the performance of the hybrid artificial pancreas. METHODS: The proposed reinforcement learning algorithm was designed using the Q-learning approach. The algorithm learns the optimal actions (CRs and programmed basal rate) by applying them to the individual's state (previous day's glucose levels and insulin delivery) based on an exploration and exploitation trade-off. First, outcomes from our simulator were compared to those of a clinical study in 23 individuals with type 1 diabetes and have yielded similar results. Second, the learning algorithm was tested using the simulator with two scenarios. Scenario 1 has fixed meal sizes and ingestion times and scenario 2 has a more realistic eating behavior with random meal sizes, ingestion times, and carbohydrate counting errors. RESULTS: After about five weeks, the reinforcement learning algorithm improved the percentage of time spent in target range from 67% to 86.7% in scenario 1 and 65.5% to 86% in scenario 2. The percentage of time spent below 4.0 mmol/L decreased from 9% to 0.9% in scenario 1 and 9.5% to 1.1% in scenario 2. CONCLUSIONS: Results indicate that the proposed algorithm has the potential to improve glucose control in people with type 1 diabetes using the hybrid artificial pancreas. The proposed algorithm is a key in making the hybrid artificial pancreas adaptive for the long-term real life outpatient studies.