Missed and Mistimed Insulin Doses in People with Diabetes: A Systematic Literature Review

Nocturnal Glucose Profile According to Timing of Dinner Rapid Insulin and Basal and Rapid Insulin Type: An Insulclock® Connected Insulin Cap-Based Real-World Study

BACKGROUND

A study to assess the glucose levels of people with type 1 diabetes (T1D) overnight, based on the insulin type and timing.

METHODS

A real-world, retrospective study of T1D, using multiple daily insulin injections. Continuous glucose monitoring and insulin injection data were collected for ten hours after dinner using the Insulclock® connected cap. Meal events were identified using the ROC detection methodology. The timing of the rapid insulin, second injections, and the type of insulin analogs used, were evaluated.

RESULTS

The nocturnal profiles (n = 775, 49 subjects) were analyzed. A higher glucose AUC of over 180 mg/dL was observed in subjects with delayed injections (number; %; mg/dL × h): -45-15 min (n = 136; 17.5%, 175.9 ± 271.0); -15-0 min (n = 231; 29.8%, 164.0 ± 2 37.1); 0 + 45 min (n = 408; 52.6%, 203.6 ± 260.9), (p = 0.049). The use of ultrarapid insulin (FiAsp®) (URI) vs. rapid insulin (RI) analogs was associated with less hypoglycemia events (7.1 vs. 13.6%; p = 0.005) and TBR70 (1.7 ± 6.9 vs. 4.6 ± 13.9%; p = 0.003). Users of glargine U300 vs. degludec had a higher TIR (70.7 vs. 58.5%) (adjusted R-squared: 0.22, p < 0.001). The use of a correction injection (n = 144, 18.6%) was associated with a higher number of hypoglycemia events (18.1 vs. 9.5%; p = 0.003), TBR70 (5.5 ± 14.2 vs. 3.0 ± 11.1%; p = 0.003), a glucose AUC of over 180 mg/dL (226.1 ± 257.8 vs. 178.0 ± 255.3 mg/dL × h; p = 0.001), and a lower TIR (56.0 ± 27.4 vs. 62.7 ± 29.6 mg/dL × h; p = 0.004).

CONCLUSION

The dinner rapid insulin timing, insulin type, and the use of correction injections affect the nocturnal glucose profile in T1D.

Impact of Missed and Late Meal Boluses on Glycemic Outcomes in Automated Insulin Delivery-Treated Children and Adolescents with Type 1 Diabetes: A Two-Center, Population-Based Cohort Study

Objective: To evaluate the impact of missed or late meal boluses (MLBs) on glycemic outcomes in children and adolescents with type 1 diabetes using automated insulin delivery (AID) systems. Research Design and Methods: AID-treated (Tandem Control-IQ or Medtronic MiniMed 780G) children and adolescents (aged 6-21 years) from Stanford Medical Center and Steno Diabetes Center Copenhagen with ≥10 days of data were included in this two-center, binational, population-based, retrospective, 1-month cohort study. The primary outcome was the association between the number of algorithm-detected MLBs and time in target glucose range (TIR; 70-180 mg/dL). Results: The study included 189 children and adolescents (48% females with a mean ± standard deviation age of 13 ± 4 years). Overall, the mean number of MLBs per day in the cohort was 2.2 ± 0.9. For each additional MLB per day, TIR decreased by 9.7% points (95% confidence interval [CI] 11.3; 8.1), and compared with the quartile with fewest MLBs (Q1), the quartile with most (Q4) had 22.9% less TIR (95% CI: 27.2; 18.6). The age-, sex-, and treatment modality-adjusted probability of achieving a TIR of >70% in Q4 was 1.4% compared with 74.8% in Q1 (P < 0.001). Conclusions: MLBs significantly impacted glycemic outcomes in AID-treated children and adolescents. The results emphasize the importance of maintaining a focus on bolus behavior to achieve a higher TIR and support the need for further research in technological or behavioral support tools to handle MLBs.

Expert Panel Recommendations for a Standardized Ambulatory Glucose Profile Report for Connected Insulin Pens

Background: Connected insulin pens capture data on insulin dosing/timing and can integrate with continuous glucose monitoring (CGM) devices with essential insulin and glucose metrics combined into a single platform. Standardization of connected insulin pen reports is desirable to enhance clinical utility with a single report. Methods: An international expert panel was convened to develop a standardized connected insulin pen report incorporating insulin and glucose metrics into a single report containing clinically useful information. An extensive literature review and identification of examples of current connected insulin pen reports were performed serving as the basis for creation of a draft of a standardized connected insulin pen report. The expert panel participated in three virtual standardization meetings and online surveys. Results: The Ambulatory Glucose Profile (AGP) Report: Connected Insulin Pen brings all clinically relevant CGM-derived glucose and connected insulin pen metrics into a single simplified two-page report. The first page contains the time in ranges bar, summary of key insulin and glucose metrics, the AGP curve, and detailed basal (long-acting) insulin assessment. The second page contains the bolus (mealtime and correction) insulin assessment periods with information on meal timing, insulin-to-carbohydrate ratio, average bolus insulin dose, and number of days with bolus doses recorded. The report's second page contains daily glucose profiles with an overlay of the timing and amount of basal and bolus insulin administered. Conclusion: The AGP Report: Connected Insulin Pen is a standardized clinically useful report that should be considered by companies developing connected pen technology as part of their system reporting/output.

Retrospective observational study on real world use of the Minimed™ 780G automated insulin delivery system: Impact of the settings on autocorrection and omitted meal boluses

INTRODUCTION

The Medtronic MiniMed™ 780G (MM780G) system uses an algorithm that includes autocorrection bolus (AB) delivery. This study evaluates the impact of omitted meal boluses and the system settings, glucose target and active insulin time (AIT), on the AB.

METHOD

Retrospective observational study on data uploaded by all MiniMed 780G users in our healthcare area, obtained through the remote monitoring platform Care Connect, from April to August 2023. Downloads with a sensor usage time <95% were excluded.

RESULTS

235 downloads belonging to 235 users were analysed. AB delivery was significantly higher at 2 h AIT (36.08 ± 13.17%) compared to the rest of settings (2.25-4 h) (26.43 ± 13.2%) (p < 0.001). AB differences based on the glucose target were not found. Patients with <3 meal boluses per day had higher AB delivery (46.91 ± 19.00% vs 27.53 ± 11.54%) (p < 0.001) and had more unfavourable glucometric parameters (GMI 7.12 ± 0.45%, TIR 67.46 ± 12.89% vs GMI 6.78 ± 0.3%, TIR 76.51 ± 8.37%) (p < 0.001). However, the 2-h AIT group presented similar TAR, TIR and GMI regardless of the number of meal boluses.

CONCLUSION

The fewer user-initiated boluses, the greater the autocorrection received. The active insulin time of 2 h entails a more active autocorrection pattern that makes it possible to more effectively compensate for the omission of meal boluses without increasing hypoglycaemias.

Carbohydrate counting knowledge and ambulatory glucose profile in persons living with type 1 diabetes

CONTEXT

The amount of consumed carbohydrates is the strongest factor influencing glucose levels during the four hours following a meal. Our aim was to evaluate the association between carbohydrate counting knowledge and continuous glucose monitoring (CGM) parameters in patients with type 1 diabetes (T1D) using different insulin regimens.

METHOD

In this multicenter prospective study, the GluciQuizz questionnaire was used to evaluate carbohydrate knowledge. CGM data for the 14 days preceding completion of the questionnaire were analyzed. The primary endpoint was evaluation of the correlation between the GluciQuizz total score and time in range (TIR) in the study population.

RESULTS

The mean age of the 170 participants was 40.7 ± 14.8 years and duration of T1D 18.8 ± 12.1 years. The mean GluciQuizz total score for all participants was 66 ± 13 %. Mean TIR was 58.6 ± 18.7 %. GluciQuizz total score positively correlated with TIR (r = 0.3001; p < 0.0001). This correlation was observed in CSII users (r = 0.2526; p < 0.05) but not in MDI (r = 0.2510; p = 0.1134) and HCL users (r = -0.1065; p = 0.4914). TIR was also negatively correlated with the mean carb count error in all study participants (r = -0.2317; p < 0.01).

CONCLUSION

In conclusion, as the Gluciquizz score was associated with metabolic control, this easy-to-use self-administered questionnaire could be used widely on a routine basis to assess the carbohydrate knowledge of T1D patients and to offer them targeted education tailored to their needs.

An ensemble machine learning approach for the detection of unannounced meals to enhance postprandial glucose control

BACKGROUND

Hybrid automated insulin delivery systems enhance postprandial glucose control in type 1 diabetes, however, meal announcements are burdensome. To overcome this, we propose a machine learning-based automated meal detection approach; METHODS:: A heterogeneous ensemble method combining an artificial neural network, random forest, and logistic regression was employed. Trained and tested on data from two in-silico cohorts comprising 20 and 47 patients. It accounted for various meal sizes (moderate to high) and glucose appearance rates (slow and rapid absorbing). To produce an optimal prediction model, three ensemble configurations were used: logical AND, majority voting, and logical OR. In addition to the in-silico data, the proposed meal detector was also trained and tested using the OhioT1DM dataset. Finally, the meal detector is combined with a bolus insulin compensation scheme; RESULTS:: The ensemble majority voting obtained the best meal detector results for both the in-silico and OhioT1DM cohorts with a sensitivity of 77%, 94%, 61%, precision of 96%, 89%, 72%, F1-score of 85%, 91%, 66%, and with false positives per day values of 0.05, 0.19, 0.17, respectively. Automatic meal detection with insulin compensation has been performed in open-loop insulin therapy using the AND ensemble, chosen for its lower false positive rate. Time-in-range has significantly increased 10.48% and 16.03%, time above range was reduced by 5.16% and 11.85%, with a minimal time below range increase of 0.35% and 2.69% for both in-silico cohorts, respectively, compared to the results without a meal detector; CONCLUSION:: To increase the overall accuracy and robustness of the predictions, this ensemble methodology aims to take advantage of each base model's strengths. All of the results point to the potential application of the proposed meal detector as a separate module for the detection of meals in automated insulin delivery systems to achieve improved glycemic control.

Evaluating Patient and Provider Preferences for a Once-Weekly Basal Insulin in Adults with Type 2 Diabetes

Purpose

The global burden of disease of type 2 diabetes (T2D) is significant, and insulin currently plays a central role in T2D management. This study sought to assess the preferences of patients with T2D and healthcare providers (HCPs) involved in T2D care regarding a hypothetical once-weekly basal insulin in comparison to current basal insulin options.

Patients and Methods

In a survey-based study in the United States that included a discrete choice experiment (DCE), patients with T2D (insulin naïve and current insulin users) and providers who treat individuals with T2D were asked to evaluate current basal insulins and identify attributes of importance regarding a hypothetical once-weekly basal insulin. A regression analysis was conducted to identify drivers of preference by relevant demographics, attitudes, and behaviors.

Results

Most respondents (91% of patients with T2D and 89% of HCPs in the base case scenario) would choose a once-weekly basal insulin product over another type of basal insulin. Both patients with T2D and HCPs rated insulin type and delivery method to be attributes of highest importance in the discrete choice exercise. Current basal insulin users ("insulin experienced") reported higher levels of confidence that a once-weekly insulin would help them to achieve their desired blood sugar levels compared to their current basal insulin (5.7 vs 5.2 on a 7-point Likert scale). Most insulin-experienced respondents (88%) were likely to inquire about once-weekly basal insulin, and most HCPs (85%) indicated willingness to educate patients on management of their T2D using a once-weekly basal insulin.

Conclusion

Discussing preferences for T2D medication management is important for patients and HCPs to ensure treatments are offered for patients based on their preferences. This study showed that patient and provider preferences are similar towards a once-weekly basal insulin over current basal insulin preparations.

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.

Delayed prandial insulin boluses are an important determinant of blood glucose control and relate to fear of hypoglycemia in people with type 1 diabetes on advanced technologies

AIMS

Automated insulin delivery systems improve blood glucose control in patients with type 1 diabetes (T1D). However, optimizing their performance requires patient's proper compliance to meal insulin bolus administration. We explored real-life prevalence of delayed prandial boluses (DBs) in adults with T1D on advanced technologies, and their association with glycemic control and fear of hypoglycemia (FH).

METHODS

In the last two-week web-based reports of 152 adults with T1D on Hybrid Closed Loop Systems (HCLS) or Sensor Augmented Pump (SAP), DBs were identified when a steep increase in blood glucose occurred at CGM before the prandial bolus, and CGM metrics were evaluated. All participants completed an online questionnaire on FH.

RESULTS

Mean DBs over two weeks were 10.2 ± 4.7 (M ± SD, range 1-23) and more frequent in women than men (11.0 ± 4.6 vs. 9.4 ± 4.7, p = 0.036). Participants with more DBs (>12) showed significantly lower Time-In-Range (62.4 ± 13.8 vs. 76.6 ± 9.0 %) than those with less DBs (<7.7), along with higher Time-Above-Range, GMI, and Coefficient-of-Variation (ANOVA, p < 0.001 for all). Participants with higher FH score showed more DBs (11.6 ± 5.0) than those in lower tertiles (9.57 ± 4.59 and 9.47 ± 4.45, ANOVA p = 0.045).

CONCLUSIONS

In patients on advanced technologies, delayed boluses are extremely common, and associate with significantly worse glycemic control. Utmost attention is needed to bolus timing, mainly tackling fear of hypoglycemia.

The Burden of Suboptimal Insulin Dosing in People with Diabetes in Spain: Barriers and Solutions from the Physician Perspective

Background

This study aimed to determine physicians' perceptions of the extent of suboptimal insulin dosing and the barriers and solutions to optimal dosing in people with diabetes (PwD) treated with insulin.

Methods

A cross-sectional online survey was conducted in four countries with primary care physicians and endocrinologists treating PwD using insulin pens, which included 53 questions on physicians' characteristics and their perceptions of the behaviors of PwD in relation to insulin dosing routines, unmet needs and potential solutions. Analyses were descriptive.

Results

Of the 160 physicians (80 primary care physicians, 80 specialists) surveyed in Spain, 58.1% were male and 88.8% had been qualified to practice for more than five years. Most physicians (>65%) indicated that 0-30% of PwD missed or skipped, mistimed, or miscalculated an insulin dose in the last 30 days. Common reasons for these actions were that PwD forgot, were out of their normal routine, were too busy or distracted, or were unsure of how much insulin to take. To optimize insulin dosing, over 75% of physicians considered it very helpful for PwD to have real-time insulin dosing calculation guidance, mobile app reminders, a device automatically recording glucose measurements and/or insulin, having insulin and glucose data in one place, and having the time for more meaningful conversations about insulin dosing routines.

Conclusion

According to physicians' perspectives, suboptimal insulin dosing remains common among PwD. This survey highlights the need for integrated and automated insulin dosing support to manage the complexity of insulin treatment, improve communications between PwD and physicians, and ultimately improve outcomes for PwD.

Shining the Spotlight on Multiple Daily Insulin Therapy: Real-World Evidence of the InPen Smart Insulin Pen

Objective: Connected insulin pens are creating opportunities for the millions of individuals with diabetes using multiple daily injections (MDI) therapy across the globe. Continuous glucose monitoring (CGM) data from connected insulin pens are revealing gaps and opportunities to significantly improve care for this population. In this article, we report real-world findings of the InPen™ smart insulin pen paired with CGM (InPen system), used by persons with type 1 diabetes (T1D) and type 2 diabetes (T2D). Methods: A retrospective cohort analysis was conducted with the real-world data collected from the InPen system of individuals (N = 3793 with T1D, N = 552 with T2D, and N = 808 unidentified) who used the system from January 01, 2020, to December 31, 2021. Diabetes management (e.g., missed and mistimed insulin dosing, mismatched food intake, and correction dose delivery) and glycemic outcomes were assessed. Results: In the overall and T1D populations, a dosing frequency of ≥3 doses per day and a missed dose frequency of <20% was associated with improved glycemia. In adults with T2D, missing <20% of doses was the significant factor determining improved glycemia. Conclusion: This analysis, integrating data from a smart insulin pen and CGM, provides insights into the impact of dosing behavior on glycemic outcomes and informs counseling strategies for the diabetes care team, through technologically advanced insulin management for those using MDI therapy.

Ultrarapid Insulin Use Can Reduce Postprandial Hyperglycemia and Late Hypoglycemia, Even in Delayed Insulin Injections: A Connected Insulin Cap-Based Real-World Study

Objectives: Reaching optimal postprandial glucose dynamics is a daily challenge for people with type 1 diabetes (T1D). This study aimed to analyze the postprandial hyperglycemic excursion (PHEs) and late postprandial hypoglycemia (LPH) risk according to prandial insulin time and type. Research Design and Methods: Real-world, retrospective study in T1D using multiple daily injections (MDI) analyzing 5 h of paired continuous glucose monitoring and insulin injections data collected from the connected cap Insulclock®. Meal events were identified using the rate of change detection methodology. Postprandial glucometrics and LPH (glucose <70 mg/dL 2-5 h after a meal) were evaluated according to insulin injection time and rapid (RI) or ultrarapid analog, Fiasp® (URI), use. Results: Meal glycemic excursions (n = 2488), RI: 1211, 48.7%; UR: 1277, 51.3%, in 82 people were analyzed according to injection time around the PHE: -45 to -15 min; -15 to 0 min; and 0 to +45 min. In 63% of the meals, insulin was injected after the PHE started. Lower PHE was observed with URI versus RI (glucose peak-baseline; mg/dL; mean ± standard deviation): 106.7 ± 35.2 versus 111.2 ± 40.3 (P = 0.003), particularly in 0/+45 injections: 111.6 ± 40.2 versus 118.1 ± 43.3; (P = 0.002). One third (29.1%) of participants added a second (correction) injection. The use of URI and avoiding a second injection were independently associated with less LPH risk, even in delayed injections (0/+45), (-36%, odds ratio [OR] 0.641; confidence interval [CI]: 0.462-0.909; P = 0.012) and -56% (OR 0.641; CI: 0.462-0.909 P = 0.038), respectively. Conclusions: URI analog use as prandial insulin reduces postprandial hyper- and hypoglycemia, even in delayed injections.

Beneficial Effect of Remote Glucose Monitoring and Computerized Insulin Dose Adjustment Algorithms Independent of Insulin Dose Increases in Sizeable Minorities of Patients

This article describes a program through which interactions every 2-3 weeks between patients and primary care clinicians (PCCs), with recommendations based on analysis of remote glucose monitoring by computerized insulin dose adjustment algorithms, significantly improved diabetes control. Insulin doses increased by 30% in the majority of patients. A sizeable minority (36%) had a decrease or no increase in insulin doses, but still showed an improvement in diabetes control. Frequent interactions allowed PCCs the opportunity to recognize and address medication nonadherence.

Insulin Therapy for the Management of Diabetes Mellitus: A Narrative Review of Innovative Treatment Strategies

The discovery of insulin was presented to the international medical community on May 3, 1922. Since then, insulin has become one of the most effective pharmacological agents used to treat type 1 and type 2 diabetes mellitus. However, the initiation and intensification of insulin therapy is often delayed in people living with type 2 diabetes due to numerous challenges associated with daily subcutaneous administration. Reducing the frequency of injections, using insulin pens instead of syringes and vials, simplifying treatment regimens, or administering insulin through alternative routes may help improve adherence to and persistence with insulin therapy among people living with diabetes. As the world commemorates the centennial of the commercialization of insulin, the aims of this article are to provide an overview of insulin therapy and to summarize clinically significant findings from phase 3 clinical trials evaluating less frequent dosing of insulin and the non-injectable administration of insulin.

A Probabilistic Approach to Blood Glucose Prediction in Type 1 Diabetes Under Meal Uncertainties

Currently, most reliable and commercialized artificial pancreas systems for type 1 diabetes are hybrid closed-loop systems, which require the user to announce every meal and its size. However, estimating the amount of carbohydrates in a meal and announcing each and every meal is an error-prone process that introduces important uncertainties to the problem, which when not considered, lead to sub-optimal outcomes of the controller. To address this problem, we propose a novel deep-learning-based model for probabilistic glucose prediction, called the Input and State Recurrent Kalman Network (ISRKN), which consists in the incorporation of an input and state Kalman filter in the latent space of a deep neural network so that the posterior distributions can be computed in closed form and the uncertainty can be propagated using the Kalman equations. In addition, the proposed architecture allows explicit estimation of the meal uncertainty distribution, whose parameters are encoded in the filter parameters. Results using the UVA/Padova simulator and data from a clinical trial show that the proposed model outperforms other probabilistic models using several probabilistic metrics across different degrees of distributional shifts.

Assessment of Adherence to Insulin Injections among Diabetic Patients on Basal-Bolus Regimen in Primary and Secondary Healthcare Centers in Al-Jouf Region of Saudi Arabia; A Descriptive Analysis

BACKGROUND

Patient adherence to insulin therapy is one of the major challenges during the treatment of diabetes mellitus. Considering the dearth of investigations, this study aimed to determine the adherence pattern and factors linked with nonadherence among diabetic patients using insulin in Al-Jouf region of Saudi Arabia.

METHODS

This cross-sectional study included diabetic patients using basal-bolus regimens, whether they had type 1 or type 2 diabetes. This study's objective was determined using a validated data collection form that included sections on demographics, reasons for missed insulin doses, list of barriers to therapy, difficulties during insulin administration, and factors that may improve insulin inaction adherence.

RESULTS

Of 415 diabetic patients, 169 (40.7%) were reported to forget doses of insulin every week. The majority of these patients (38.5%) forget one or two doses. Away from home (36,1%), inability to adhere to the diet (24.3%) and embarrassment to administer injections in public (23.7%) were frequently cited as reasons for missing insulin doses. The occurrence of hypoglycemia (31%), weight gain (26%), and needle phobia (22%) were frequently cited as obstacles to insulin injection use. Preparing injections (18.3%), using insulin at bedtime (18.3%), and storing insulin at a cold temperature (18.1%) were the most challenging aspects of insulin use for patients. Reduction in the number of injections (30.8%) and convenient timing for insulin administration (29.6%) were frequently cited as factors that may improve participant adherence.

CONCLUSIONS

This study revealed that the majority of diabetic patients forget to inject insulin, primarily as a result of travel. By identifying potential obstacles faced by patients, these findings direct health authorities to design and implement initiatives to increase insulin adherence among patients.

Classification of Hypoglycemic Events in Type 1 Diabetes Using Machine Learning Algorithms

INTRODUCTION

To improve the utilization of continuous- and flash glucose monitoring (CGM/FGM) data we have tested the hypothesis that a machine learning (ML) model can be trained to identify the most likely root causes for hypoglycemic events.

METHODS

CGM/FGM data were collected from 449 patients with type 1 diabetes. Of the 42,120 identified hypoglycemic events, 5041 were randomly selected for classification by two clinicians. Three causes of hypoglycemia were deemed possible to interpret and later validate by insulin and carbohydrate recordings: (1) overestimated bolus (27%), (2) overcorrection of hyperglycemia (29%) and (3) excessive basal insulin presure (44%). The dataset was split into a training (n = 4026 events, 304 patients) and an internal validation dataset (n = 1015 events, 145 patients). A number of ML model architectures were applied and evaluated. A separate dataset was generated from 22 patients (13 'known' and 9 'unknown') with insulin and carbohydrate recordings. Hypoglycemic events from this dataset were also interpreted by five clinicians independently.

RESULTS

Of the evaluated ML models, a purpose-built convolutional neural network (HypoCNN) performed best. Masking the time series, adding time features and using class weights improved the performance of this model, resulting in an average area under the curve (AUC) of 0.921 in the original train/test split. In the dataset validated by insulin and carbohydrate recordings (n = 435 events), i.e. 'ground truth,' our HypoCNN model achieved an AUC of 0.917.

CONCLUSIONS

The findings support the notion that ML models can be trained to interpret CGM/FGM data. Our HypoCNN model provides a robust and accurate method to identify root causes of hypoglycemic events.

A Review of Precision Insulin Management With Smart Insulin Pens: Opening Up the Digital Door to People on Insulin Injection Therapy

Although advances in insulin therapy and delivery have been made, global evidence indicates sub-optimal glycemic management in people on insulin therapy with either type 1 diabetes (T1D) or type 2 diabetes (T2D). In this review, we discuss connected insulin pens that include tracking insulin pens (TIPs) and smart insulin pens (SIPs) and caps, as approaches to improving mean glucose or time in range while minimizing exposure to hypoglycemia or time below range (TBR) in people with diabetes (PwD) on multiple daily injection (MDI) therapy. We discuss various factors offered by SIPs that can facilitate precision insulin management, that is, delivering the right dose at the right time. These factors include the automatic recording of insulin dose size and delivery time; differentiating prime from therapy doses; active insulin tracking; dose calculators that provide individualized dosing recommendations; alerts for missed doses (ie, rapid-acting or long-acting insulin), insulin temperature, and insulin age monitoring; and integrated data reports for the clinical care team. A data-driven approach to care is critical to precision insulin management and includes helping PwD make informed choices regarding their preferred method of insulin delivery and ensuring insulin delivery technology tools are configured for their personal therapy plan. The data-driven approach involves developing a plan for ongoing collaborative use of the resulting data with their care team that may include adjusting insulin regimen and optimizing the care plan on a timely basis. We conclude with a list of practice protocols that are needed to support data-driven precision insulin management. This review includes a summary of research including various stages of connected insulin pens and caps.

Vasodilatory effects of glucagon: A possible new approach to enhanced subcutaneous insulin absorption in artificial pancreas devices

Patients with diabetes mellitus type 1 depend on exogenous insulin to keep their blood glucose concentrations within the desired range. Subcutaneous bihormonal artificial pancreas devices that can measure glucose concentrations continuously and autonomously calculate and deliver insulin and glucagon infusions is a promising new treatment option for these patients. The slow absorption rate of insulin from subcutaneous tissue is perhaps the most important factor preventing the development of a fully automated artificial pancreas using subcutaneous insulin delivery. Subcutaneous insulin absorption is influenced by several factors, among which local subcutaneous blood flow is one of the most prominent. We have discovered that micro-doses of glucagon may cause a substantial increase in local subcutaneous blood flow. This paper discusses how the local vasodilative effects of micro-doses of glucagon might be utilised to improve the performance of subcutaneous bihormonal artificial pancreas devices. We map out the early stages of our hypothesis as a disruptive novel approach, where we propose to use glucagon as a vasodilator to accelerate the absorption of meal boluses of insulin, besides using it conventionally to treat hypoglycaemia.

Socio-cognitive determinants affecting insulin adherence/non-adherence in late adolescents and young adults with type 1 diabetes: a systematic review protocol

Objective

This systematic review aims to investigate the key socio-cognitive determinants associated with adherence/non-adherence to insulin treatment in late adolescents and young adults in the age range of 17–24 years with T1D.

Methods

A pre-specified search strategy will be used to search for studies in the electronic databases and citation indexes: PubMed, EMBASE, Web of Science, and PsycINFO. Two researchers will screen the title and the abstract independently, then will read and critically appraise the full text of each included study. A third independent reviewer will resolve disagreements in data extraction until consensus. Data will be extracted using the Population, Exposure, Outcomes, Study characteristics framework. Study selection will follow the updated guideline for reporting systematic reviews (PRISMA 2020) and will take place from 15 October 2021 to 1 January 2022. The methodological quality and risk of bias of the observational studies will be assessed by the JBI Critical Appraisal Checklist for Cohort and JBI Critical Appraisal Checklist for Analytical Cross Sectional Studies.

Results

A qualitative narrative synthesis will present the characteristics and the quality of studies and the outcomes of concern.

Conclusion

Based on the contemporary literature, this review will synthesize the evidence on the socio-cognitive determinants associated with adherence/non-adherence to insulin treatment in late adolescents and young adults in the age range of 17–24 years with T1D. The findings will help design patient-centered interventions to promote adherence to insulin in this age group, guide patients’ consultations and diabetes self-management education (DSME) programs.

Protocol registration: PROSPERO ID: CRD42021233074.

A Deep Learning Framework for Automatic Meal Detection and Estimation in Artificial Pancreas Systems

Current artificial pancreas (AP) systems are hybrid closed-loop systems that require manual meal announcements to manage postprandial glucose control effectively. This poses a cognitive burden and challenge to users with T1D since this relies on frequent user engagement to maintain tight glucose control. In order to move towards fully automated closed-loop glucose control, we propose an algorithm based on a deep learning framework that performs multitask quantile regression, for both meal detection and carbohydrate estimation. Our proposed method is evaluated in silico on 10 adult subjects from the UVa/Padova simulator with a Bio-inspired Artificial Pancreas (BiAP) control algorithm over a 2 month period. Three different configurations of the AP are evaluated -BiAP without meal announcement (BiAP-NMA), BiAP with meal announcement (BiAP-MA), and BiAP with meal detection (BiAP-MD). We present results showing an improvement of BiAP-MD over BiAP-NMA, demonstrating 144.5 ± 6.8 mg/dL mean blood glucose level (-4.4 mg/dL, p< 0.01) and 77.8 ± 6.3% mean time between 70 and 180 mg/dL (+3.9%, p< 0.001). This improvement in control is realised without a significant increase in mean in hypoglycaemia (+0.1%, p= 0.4). In terms of detection of meals and snacks, the proposed method on average achieves 93% precision and 76% recall with a detection delay time of 38 ± 15 min (92% precision, 92% recall, and 37 min detection time for meals only). Furthermore, BiAP-MD handles hypoglycaemia better than BiAP-MA based on CVGA assessment with fewer control errors (10% vs. 20%). This study suggests that multitask quantile regression can improve the capability of AP systems for postprandial glucose control without increasing hypoglycaemia.

The automated pancreas: A review of technologies and clinical practice

Insulin pumps and glucose sensors are effective in improving diabetes therapy and reducing acute complications. The combination of both devices using an algorithm-driven interoperable controller makes automated insulin delivery (AID) systems possible. Many AID systems have been tested in clinical trials and have proven safety and effectiveness. However, currently, none of these systems are available for routine use in children younger than 6 years in Europe. For continued use, both users and prescribers must have sound knowledge of the features of the individual AID systems. Presently, all systems require various user interactions (e.g. meal announcements) because fully automated systems are not yet developed. Open-source systems are non-regulated variants to circumvent existing regulatory conditions. There are risks here for both users and prescribers. To evaluate AID therapy, the metric data of the glucose sensors, 'time in target range' and 'glucose management index', are novel recognized and suitable parameters allowing a consultation based on real glucose and insulin pump download data from the daily life of people with diabetes. Read out via cloud-based software or automatic download of such individual treatment data provides the ideal technical basis for shared decision-making through telemedicine, which must be further evaluated for general use.