A scalable application of Artificial Intelligence-Driven Insulin Titration Program to transform Type 2 Diabetes Management.

BACKGROUND: Despite new pharmacotherapy, most patients with long-term Type 2 Diabetes are still hyperglycemic. This could have been solved by insulin with its unlimited potential efficacy, but its dynamic physiology demands frequent titrations which are overdemanding. This report provides a real-life account for a scalable transformation of diabetes care in a community-based endocrinology center by harnessing AI-based autonomous insulin titration. METHODS: The center embedded the d-Nav® technology and its dedicated clinical support. Reported outcomes include treatment efficacy/safety in the first 600 patients and use of cardiorenal-risk reduction pharmacotherapy. FINDINGS: Patients used d-Nav for 8.2±3.0 months with 82% retention. Age was 67.1±11.5 years and duration of diabetes was 19.8±11.0 years. During the last 3 years before d-Nav, HbA1c had been overall higher than 8% and at the beginning of the program it was as high as 8.6%±2.1% with 29.3% of the patients with HbA1c>9%. With d-Nav, HbA1c decreased to 7.3%±1.2% with 5.7% of patients with HbA1c>9%. During the first 3 months, d-Nav reduced total daily dose of insulin in 1 of every 5 patients due to relatively low glucose levels to minimize the risk of hypoglycemia. GLP-1 or dual GLP-1 and GIP receptor agonists were prescribed in about a half of the patients and SGLT2 inhibitor in a third. The frequency of hypoglycemia (<54mg/dl) was 0.4±0.6/month and severe hypoglycemia 1.7/100-patient-years. INTERPRETATION: The use of d-Nav allowed for improvement in overall diabetes management with appropriate use of both insulin and non-insulin pharmacologic agents in a scalable way.

Challenging the 50-50 rule for the basal-bolus insulin ratio in patients with type 2 diabetes who maintain stable glycaemic control.

BACKGROUND: For patients using basal-bolus insulin therapy, it is widespread clinical practice to aim for a 50-50 ratio between basal and total daily bolus. However, this practice was based on a small study of individuals without diabetes. To assess the rule in real-world practice, we retrospectively analyzed patients on basal-bolus therapy that was adjusted at least weekly by an artificial intelligence-driven titration within the d-Nav® Insulin Management Technology. MATERIALS AND METHODS: We obtained de-identified data from the Diabetes Centre of Ulster Hospital for patients with four inclusion criteria: type 2 Diabetes (T2D), on d-Nav >6 months, on basal-bolus insulin therapy >80% of the time (based on insulin analogs), and no gap in data >3 months. RESULTS: We assembled a cohort of 306 patients, followed by the d-Nav service for 3.4 ± 1.8 years (mean ± SD), corresponding to about 180 autonomous insulin dose titrations and about 5000 autonomous individual dose recommendations per patient. After an initial run-in period, mean glycated hemoglobin (HbA1c) values in the cohort were maintained close to 7%. Surprisingly, in just over three-quarters of the cohort, the average basal insulin fraction was <50%; in half of the cohort average basal insulin fraction <41.2%; and in one-quarter the basal insulin fraction was <33.6%. Further, the basal insulin fraction did not remain static over time. In half of the patients, the basal insulin fraction varied by ≥1.9×; and, in 25% of the patients, ≥2.5×. CONCLUSION: Our data show that a 50-50 ratio of basal-to-bolus insulin does not generally apply to patients with T2D who successfully maintain stable glycemia. Therefore, the 50-50 ratio should not serve as an ongoing treatment guide. Moreover, our results emphasize the importance of at least weekly insulin titrations.

Automated insulin dosing guidance to optimise insulin management in patients with type 2 diabetes: a multicentre, randomised controlled trial.

BACKGROUND: Insulin therapy is most effective if dosage titrations are done regularly and frequently, which is seldom practical for most clinicians, resulting in an insulin titration gap. The d-Nav Insulin Guidance System (Hygieia, Livonia, MI, USA) is a handheld device that is used to measure glucose, determine glucose patterns, and automatically determine the appropriate next insulin dose. We aimed to determine whether the combination of the d-Nav device and health-care professional support is superior to health-care professional support alone. METHODS: In this multicentre, randomised, controlled study, we recruited patients from three diabetes centres in the USA (in Detroit MI; Minneapolis, MN; and Des Moines IA). Patients were eligible if they were aged 21-70 years, diagnosed with type 2 diabetes with a glycated haemoglobin (HbA1c) concentration of 7·5% or higher (≥58 mmol/mol) and 11% or lower (≤97 mmol/mol), and had been using the same insulin regimen for the previous 3 months. Exclusion criteria included body-mass index of 45 kg/m2 or higher; severe cardiac, hepatic, or renal impairment; and more than two severe hypoglycaemic events in the past year. Eligible participants were randomly assigned (1:1), with randomisation blocked within each site, to either d-Nav and health-care professional support (intervention group) or health-care professional support alone (control group). Both groups were contacted seven times (three face-to-face and four phone visits) during 6 months of follow-up. The primary objective was to compare average change in HbA1c from baseline to 6 months. Safety was assessed by the frequency of hypoglycaemic events. The primary objective and safety were assessed in the intention-to-treat population. We used Student's t test to assess the primary outcome for statistical significance. This study was registered with ClinicalTrials.gov, number NCT02424500. FINDINGS: Between Feb 2, 2015, and March 17, 2017, 236 patients were screened for eligibility, of whom 181 (77%) were enrolled and randomly assigned to the intervention (n=93) and control (n=88) groups. At baseline, mean HbA1c was 8·7% (SD 0·8; 72 mmol/mol [SD 8·8]) in the intervention group and 8·5% (SD 0·8; 69 mmol/mol [SD 8·8]) in the control group. The mean decrease in HbA1c from baseline to 6 months was 1·0% (SD 1·0; 11 mmol/mol [SD 11]) in the intervention group, and 0·3% (SD 0·9; 3·3 mmol/mol [9·9]) in the control group (p<0·0001). The frequency of hypoglycaemic events per month was similar between the groups (0·29 events per month [SD 0·48] in the intervention group vs 0·29 [SD 1·12] in the control group; p=0·96). INTERPRETATION: The combination of automated insulin titration guidance with support from health-care professionals offers superior glycaemic control compared with support from health-care professionals alone. Such a solution facilitated safe and effective insulin titration in a large group of patients with type 2 diabetes, and now needs to be evaluated across large health-care systems to confirm these findings and study cost-effectiveness. FUNDING: US National Institutes of Health, National Institute of Digestive and Kidney Diseases.

Temporary Reductions in Insulin Requirements Are Associated with Hypoglycemia in Type 2 Diabetes.

Background: In patients with type 2 diabetes, insulin therapy necessitates regular and frequent dosage titration to overcome variations in insulin requirements. The goal of this study was to evaluate changes in insulin requirements, using data from a technology-based insulin-titration service. Methods: To keep glycemia stable, the service adjusts and records insulin dosage at least weekly. Therefore, insulin dosage closely tracks insulin requirement. Events of considerable and persistent decrease in insulin requirements were identified by reductions in total daily dose (TDD) of insulin ≥25%. Periods ended when a persistent increase in TDD of insulin has started. The average frequency of hypoglycemia was expressed as any glucose reading <54 mg/dL (both inside or outside periods of decrease in insulin dosage) divided by the total number of months for each patient. Results: Patients (n = 246) were followed for 2.8 ± 0.9 years. Reductions of TDD of insulin were experienced by 70.3% of the patients, occurred 0.8 ± 0.5 times per year, lasted 10.0 ± 7.7 weeks, and insulin requirements declined by 39.9% ± 12.6%. The frequency of hypoglycemia (<54 mg/dL) was low, at 0.5 ± 0.6 per month, and the difference in frequencies in biphasic/premixed and basal-bolus insulin regimens was not statistically significant. Hypoglycemia was 6.5 times more prevalent during reductions in TDD of insulin. Conclusions: Sizeable changes in insulin requirements occur over time, which demand persistent and frequent titration to preserve treatment safety.

A novel approach to optimise glycaemic control in insulin users.

Insulin therapy has been available for almost a century. However, its success rate is still disappointing where the majority of users sustain harmfully elevated glycated haemoglobin (HbA1c) levels. The key element essential for effective and safe insulin therapy is frequent dosage titration to overcome constant variations in insulin requirements. In reality, dosage titration is done sporadically during clinic visits. A scalable solution to this problem is being reviewed. A diabetes nurses service improves glycaemic control without overburdening the health system. The service relies on a handheld device, which provides patients with an insulin dose recommendation for each injection while using the device to monitor glucose. Similar to the approach providers use during clinical encounters, the device analyses stored glucose trends and constantly titrates insulin dosage without care providers' supervision. In this report, we describe the logic behind the technology by providing examples from users.

Can a tool that automates insulin titration be a key to diabetes management?

BACKGROUND: Most patients who use insulin do not achieve optimal glycemic control and become susceptible to complications. Numerous clinical trials have shown that frequent insulin dosage titration is imperative to achieve glycemic control. Unfortunately, implementation of such a paradigm is often impractical. We hypothesized that the Diabetes Insulin Guidance System (DIGS™) (Hygieia, Inc., Ann Arbor, MI) software, which automatically advises patients on adjustment of insulin dosage, would provide safe and effective weekly insulin dosage adjustments. SUBJECTS AND METHODS: In a feasibility study we enrolled patients with type 1 and type 2 diabetes, treated with a variety of insulin regimens and having suboptimal glycemic control. The 12-week intervention period followed a 4-week baseline run-in period. During the intervention, DIGS processed patients' glucose readings and provided insulin dosage adjustments on a weekly basis. If approved by the study team, the adjusted insulin dosage was communicated to the patients. Insulin formulations were not changed during the study. The primary outcome was the fraction of DIGS dosage adjustments approved by the study team, and the secondary outcome was improved glycemic control. RESULTS: Forty-six patients were recruited, and eight withdrew. The DIGS software recommended 1,734 insulin dosage adjustments, of which 1,731 (99.83%) were approved. During the run-in period the weekly average glucose was stable at 174.2±36.7 mg/dL (9.7±2.0 mmol/L). During the following 12 weeks, DIGS dosage adjustments resulted in progressive improvement in average glucose to 163.3±35.1 mg/dL (9.1±1.9 mmol/L) (P<0.03). Mean glycosylated hemoglobin decreased from 8.4±0.8% to 7.9±0.9% (P<0.05). Concomitantly, the frequency of hypoglycemia decreased by 25.2%. CONCLUSIONS: The DIGS software provided patients with safe and effective weekly insulin dosage adjustments. Widespread implementation of DIGS may improve the outcome and reduce the cost of implementing effective insulin therapy.

Frequent insulin dosage adjustments based on glucose readings alone are sufficient for a safe and effective therapy.

PROBLEM: Frequent dosage adjustments are necessary to achieve effective insulin therapy. However, a controversy surrounds the pertinent clinical parameters required to make effective and safe insulin titrations. We hypothesize that glucose readings are sufficient to adjust insulin dosage provided that it is done on a weekly basis. METHODS: In a prospective pilot study, we recruited 14 subjects with suboptimally controlled insulin-treated Type-2 and Type-1 diabetes. Subjects were treated with basal-bolus insulin therapy that was titrated weekly for 12 weeks. Dosage adjustments were made by the study Endocrinologist by reviewing subjects' glucose readings, exclusively based on logsheets and contingent upon the approval of the on-site study team. To corroborate that the glucose readings were sufficient for making dosage adjustments, we used software to process only glucose readings and recommend insulin dosage adjustments. The recommendations made by the software were retrospectively compared to the ones made by the study Endocrinologist. RESULTS: All N=568 recommendations were approved by the study team and in 99.3% of the cases the recommendations were clinically similar to the ones made by the software. No hazardous disagreements were found. The mean A1C improved from 9.8% (± 2.0) to 7.9% (± 1.3) (p=0.001) in 12 weeks and the weekly mean glucose progressively improved from 220.3 mg/dl (± 51.9) to 151.5 mg/dl (± 19.2) (p<0.0001). The frequency of minor hypoglycemia was 22.7 per patient-year in subjects with Type-2 diabetes and 42.7 in the subjects with Type-1 diabetes. No severe hypoglycemic events occurred. CONCLUSIONS: Glucose readings are sufficient to adjust insulin therapy in a safe and effective manner, when adjustments are made on a weekly basis. Thus, dedicated software may help adjust insulin dosage between clinic visits.

The effort required to achieve and maintain optimal glycemic control.

BACKGROUND: Insulin therapy is most effective when dosage is frequently adjusted. We sought to evaluate the effort required to maintain A1C below 7% once attained in older patients with type 2 diabetes. METHODS: A total of 2380 insulin dosage adjustment episodes were analyzed for their intensity and frequency. The data were divided into an "induction period" (n=608), defined as the time before subjects' A1C dropped below 7% for the first time, and a "maintenance period" (n=1772), defined as the remaining study period. The data originated from a published study and included 26 older subjects with suboptimally controlled type 2 diabetes treated for a year with intensive insulin therapy. To achieve therapy goals, the study team contacted the subjects once every few weeks, reviewed records and optimized the insulin dosage. RESULTS: During both the induction and maintenance periods, insulin dosage (both long-acting and fast-acting) was adjusted by more than 20%. Maintaining A1C below 7% required dosage adjustments every 2.7(±1.0) weeks, averaging 11.4% (±4.0) in 2.0 (±0.3) different components of insulin dosage (i.e., two of either long-acting or short-acting for breakfast, lunch or dinner) per contact. CONCLUSIONS: Considerable effort was required to maintain optimal A1C levels in older patients with type 2 diabetes. Since the full benefit of insulin therapy is attained only when multiple components of insulin dosage are frequently adjusted and given the growing shortage of care providers' availability, innovative approaches are needed to empower patients to safely make their own insulin adjustments.

Are glucose readings sufficient to adjust insulin dosage?

AIMS/HYPOTHESIS: insulin therapy is effective predominantly when dosage is frequently adjusted. However, a controversy surrounds the pertinent clinical parameters required to make effective and safe frequent dosage adjustments. We hypothesize that glucose readings are sufficient to adjust insulin dosage provided that dosage is adjusted every 1-4 weeks. METHODS: to test the hypothesis, we generated several algorithms implemented in software to process glucose readings and recommend insulin dosage adjustments. A post hoc analysis was made on 630 log sheets (2,520 insulin dosage adjustments) from 26 older adults with suboptimally controlled type 2 diabetes. The subjects were followed for a year and treated with intensive insulin therapy that was titrated every 1-4 weeks by a trained study team. More than 88% of subjects attained the treatment goal (hemoglobin A1c <7%) without excessive hypoglycemia. Glucose readings from each log sheet were used as an input to the software, and its recommendations for insulin dosage adjustments were compared to the original ones made by the study team. While the study team could have been exposed to multiple clinical parameters, the software relied solely on glucose readings. RESULTS: the software recommendations for dosage adjustments were clinically equivalent to the original study team's recommendations in more than 95% of the cases, unrelated to patients' insulin sensitivity. The remaining 4.4% (n = 111) were thoroughly examined, yet we did not find any recommendations suggested by the software to be unsafe or unreasonable. CONCLUSIONS/INTERPRETATION: glucose readings are sufficient to effectively adjust insulin dosage provided that adjustments are made every 1-4 weeks. Therefore, dedicated software can help adjusting insulin dosage between clinic visits.