Dataset Information

Optimized glycemic control of type 2 diabetes with reinforcement learning: a proof-of-concept trial.

ABSTRACT: The personalized titration and optimization of insulin regimens for treatment of type 2 diabetes (T2D) are resource-demanding healthcare tasks. Here we propose a model-based reinforcement learning (RL) framework (called RL-DITR), which learns the optimal insulin regimen by analyzing glycemic state rewards through patient model interactions. When evaluated during the development phase for managing hospitalized patients with T2D, RL-DITR achieved superior insulin titration optimization (mean absolute error (MAE) of 1.10 ± 0.03 U) compared to other deep learning models and standard clinical methods. We performed a stepwise clinical validation of the artificial intelligence system from simulation to deployment, demonstrating better performance in glycemic control in inpatients compared to junior and intermediate-level physicians through quantitative (MAE of 1.18 ± 0.09 U) and qualitative metrics from a blinded review. Additionally, we conducted a single-arm, patient-blinded, proof-of-concept feasibility trial in 16 patients with T2D. The primary outcome was difference in mean daily capillary blood glucose during the trial, which decreased from 11.1 (±3.6) to 8.6 (±2.4) mmol L^-1 (P < 0.01), meeting the pre-specified endpoint. No episodes of severe hypoglycemia or hyperglycemia with ketosis occurred. These preliminary results warrant further investigation in larger, more diverse clinical studies. ClinicalTrials.gov registration: NCT05409391 .

SUBMITTER: Wang G

PROVIDER: S-EPMC10579102 | biostudies-literature | 2023 Oct

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Optimized glycemic control of type 2 diabetes with reinforcement learning: a proof-of-concept trial.

Wang Guangyu G Liu Xiaohong X Ying Zhen Z Yang Guoxing G Chen Zhiwei Z Liu Zhiwen Z Zhang Min M Yan Hongmei H Lu Yuxing Y Gao Yuanxu Y Xue Kanmin K Li Xiaoying X Chen Ying Y

Nature medicine 20230914 10

The personalized titration and optimization of insulin regimens for treatment of type 2 diabetes (T2D) are resource-demanding healthcare tasks. Here we propose a model-based reinforcement learning (RL) framework (called RL-DITR), which learns the optimal insulin regimen by analyzing glycemic state rewards through patient model interactions. When evaluated during the development phase for managing hospitalized patients with T2D, RL-DITR achieved superior insulin titration optimization (mean absol ...[more]

PMID: 37710000

Similar Datasets

Project description:BackgroundWe recently reported improved glycemic control with reduced insulin dose in subjects with type 1 diabetes treated with the sodium glucose co-transporter-2 inhibitor empagliflozin. To further characterize the effects, we analyzed diurnal glycemic patterns by continuous glucose monitoring (CGM).MethodsIn an 8-week single-arm open-label pilot study of empagliflozin, we compared ambulatory glucose profiles produced from CGM data during 2-week intervals in a placebo run-in baseline period, end-of-treatment, and post-treatment. Change in glycemic exposure was evaluated by area under the median curve according to time of day (AUCTOTAL 12:00am-11:55pm; AUCDAY 7:05am-10:55pm, AUCNIGHT 11:00pm-7:00am), as well as glycemic variability, glycemic stability and time-in-target (≥70 to ≤140mg/dL).ResultsThe 40 patients (26 on insulin pump) were aged 24±5 years and BMI 24.5±3.2 kg/m2. Consistent with the observed HbA1c decrease (8.0±0.9% to 7.6±0.9%, p<0.0001), normalized AUCTOTAL CGM decreased from 153.7±25.4 to 149.0±30.2mg/dL∙h at end-of-treatment (p = 0.31), and significantly increased post-treatment (164.1±29.5mg/dL∙h, p = 0.02). The numerical decrease in normalized AUCNIGHT (152.0±36.6 to 141.9±34.4mg/dL∙h, p = 0.13) exceeded AUCDAY (154.5±24.5 to 152.6±30.4mg/dL∙h, p = 0.65). Trends toward lower glycemic variability (83.1±18.9 to 75.6±28.6mg/dL, p = 0.06) and little change in glycemic stability (10.8±3.6 to 10.3±4.5mg/dL/h, p = 0.51) were observed. When empagliflozin was discontinued, these worsened relative to baseline (89.3±19.3mg/dL, p = 0.04 and 11.8±3.7mg/dL/hr, p = 0.08). Time-in-target numerically increased (40.2±11.9 to 43.1±13.5%, p = 0.69) at end-of-treatment but reversed post-treatment. Findings were similar on stratification of pump and MDI subjects.ConclusionsWe observed that empagliflozin was associated with patterns of improved nighttime glycemia more prominent than daytime.Trial registrationClinicaltrials.gov NCT01392560.

Project description:IntroductionTight glycemic control (TGC) has shown benefits but has been difficult to achieve consistently. STAR (Stochastic TARgeted) is a flexible, model-based TGC approach directly accounting for intra- and inter- patient variability with a stochastically derived maximum 5% risk of blood glucose (BG) < 4.0 mmol/L. This research assesses the safety, efficacy, and clinical burden of a STAR TGC controller modulating both insulin and nutrition inputs in pilot trials.MethodsSeven patients covering 660 hours. Insulin and nutrition interventions are given 1-3 hourly as chosen by the nurse to allow them to manage workload. Interventions are calculated by using clinically validated computer models of human metabolism and its variability in critical illness to maximize the overlap of the model-predicted (5-95th percentile) range of BG outcomes with the 4.0-6.5 mmol/L band while ensuring a maximum 5% risk of BG < 4.0 mmol/L. Carbohydrate intake (all sources) was selected to maximize intake up to 100% of SCCM/ACCP goal (25 kg/kcal/h). Maximum insulin doses and dose changes were limited for safety. Measurements were made with glucometers. Results are compared to those for the SPRINT study, which reduced mortality 25-40% for length of stay ≥3 days. Written informed consent was obtained for all patients, and approval was granted by the NZ Upper South A Regional Ethics Committee.ResultsA total of 402 measurements were taken over 660 hours (~14/day), because nurses showed a preference for 2-hourly measurements. Median [interquartile range, (IQR)] cohort BG was 5.9 mmol/L [5.2-6.8]. Overall, 63.2%, 75.9%, and 89.8% of measurements were in the 4.0-6.5, 4.0-7.0, and 4.0-8.0 mmol/L bands. There were no hypoglycemic events (BG < 2.2 mmol/L), and the minimum BG was 3.5 mmol/L with 4.5% < 4.4 mmol/L. Per patient, the median [IQR] hours of TGC was 92 h [29-113] using 53 [19-62] measurements (median, ~13/day). Median [IQR] results: BG, 5.9 mmol/L [5.8-6.3]; carbohydrate nutrition, 6.8 g/h [5.5-8.7] (~70% goal feed median); insulin, 2.5 U/h [0.1-5.1]. All patients achieved BG < 6.1 mmol/L. These results match or exceed SPRINT and clinical workload is reduced more than 20%.ConclusionsSTAR TGC modulating insulin and nutrition inputs provided very tight control with minimal variability by managing intra- and inter- patient variability. Performance and safety exceed that of SPRINT, which reduced mortality and cost in the Christchurch ICU. The use of glucometers did not appear to impact the quality of TGC. Finally, clinical workload was self-managed and reduced 20% compared with SPRINT.

Project description:BackgroundDespite the increasing prevalence of type 2 diabetes in youth, there are few data to guide treatment. We compared the efficacy of three treatment regimens to achieve durable glycemic control in children and adolescents with recent-onset type 2 diabetes.MethodsEligible patients 10 to 17 years of age were treated with metformin (at a dose of 1000 mg twice daily) to attain a glycated hemoglobin level of less than 8% and were randomly assigned to continued treatment with metformin alone or to metformin combined with rosiglitazone (4 mg twice a day) or a lifestyle-intervention program focusing on weight loss through eating and activity behaviors. The primary outcome was loss of glycemic control, defined as a glycated hemoglobin level of at least 8% for 6 months or sustained metabolic decompensation requiring insulin.ResultsOf the 699 randomly assigned participants (mean duration of diagnosed type 2 diabetes, 7.8 months), 319 (45.6%) reached the primary outcome over an average follow-up of 3.86 years. Rates of failure were 51.7% (120 of 232 participants), 38.6% (90 of 233), and 46.6% (109 of 234) for metformin alone, metformin plus rosiglitazone, and metformin plus lifestyle intervention, respectively. Metformin plus rosiglitazone was superior to metformin alone (P=0.006); metformin plus lifestyle intervention was intermediate but not significantly different from metformin alone or metformin plus rosiglitazone. Prespecified analyses according to sex and race or ethnic group showed differences in sustained effectiveness, with metformin alone least effective in non-Hispanic black participants and metformin plus rosiglitazone most effective in girls. Serious adverse events were reported in 19.2% of participants.ConclusionsMonotherapy with metformin was associated with durable glycemic control in approximately half of children and adolescents with type 2 diabetes. The addition of rosiglitazone, but not an intensive lifestyle intervention, was superior to metformin alone. (Funded by the National Institute of Diabetes and Digestive and Kidney Diseases and others; TODAY ClinicalTrials.gov number, NCT00081328.).

Project description:BackgroundNo targeted immunotherapies reverse type 1 diabetes in humans. However, in a rodent model of type 1 diabetes, Bacillus Calmette-Guerin (BCG) reverses disease by restoring insulin secretion. Specifically, it stimulates innate immunity by inducing the host to produce tumor necrosis factor (TNF), which, in turn, kills disease-causing autoimmune cells and restores pancreatic beta-cell function through regeneration.Methodology/principal findingsTranslating these findings to humans, we administered BCG, a generic vaccine, in a proof-of-principle, double-blind, placebo-controlled trial of adults with long-term type 1 diabetes (mean: 15.3 years) at one clinical center in North America. Six subjects were randomly assigned to BCG or placebo and compared to self, healthy paired controls (n = 6) or reference subjects with (n = 57) or without (n = 16) type 1 diabetes, depending upon the outcome measure. We monitored weekly blood samples for 20 weeks for insulin-autoreactive T cells, regulatory T cells (Tregs), glutamic acid decarboxylase (GAD) and other autoantibodies, and C-peptide, a marker of insulin secretion. BCG-treated patients and one placebo-treated patient who, after enrollment, unexpectedly developed acute Epstein-Barr virus infection, a known TNF inducer, exclusively showed increases in dead insulin-autoreactive T cells and induction of Tregs. C-peptide levels (pmol/L) significantly rose transiently in two BCG-treated subjects (means: 3.49 pmol/L [95% CI 2.95-3.8], 2.57 [95% CI 1.65-3.49]) and the EBV-infected subject (3.16 [95% CI 2.54-3.69]) vs.1.65 [95% CI 1.55-3.2] in reference diabetic subjects. BCG-treated subjects each had more than 50% of their C-peptide values above the 95(th) percentile of the reference subjects. The EBV-infected subject had 18% of C-peptide values above this level.Conclusions/significanceWe conclude that BCG treatment or EBV infection transiently modified the autoimmunity that underlies type 1 diabetes by stimulating the host innate immune response. This suggests that BCG or other stimulators of host innate immunity may have value in the treatment of long-term diabetes.Trial registrationClinicalTrials.gov NCT00607230.

Dataset Information

Optimized glycemic control of type 2 diabetes with reinforcement learning: a proof-of-concept trial.

Publications

Optimized glycemic control of type 2 diabetes with reinforcement learning: a proof-of-concept trial.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets