Journal of Diabetes Science and Technology

Personal insulin pumps have shown to be effective in improving the quality of therapy for people with type 1 diabetes (T1D). However, the safety of this technology is limited by the possible infusion site failures, which are linked with hyperglycemia and ketoacidosis. Thanks to the large availability of collected data provided by modern therapeutic technologies, machine learning algorithms have the potential to provide new way to identify failures early and avert adverse events.

A clinical dataset (N = 20) is used to evaluate a novel method for detecting real-time infusion site failures using unsupervised anomaly detection algorithms, previously proposed and developed on in-silico data. An adapted feature engineering procedure is introduced to make the method able to operate in the absence of a closed-loop (CL) system and meal announcements.

In the optimal configuration, we obtained a performance of 0.75 Sensitivity (15 out of 20 total failures detected) and 0.08 FP/day, outperforming previously proposed literature algorithms. The algorithm was able to anticipate the replacement of the malfunctioning infusion sets by ~2 h on average.

On the considered dataset, the proposed algorithm showed the potential to improve the safety of patients treated with sensor-augmented pump systems.

There is an increasing use of continuous glucose monitoring (CGM) by people with diabetes. Measurement performance is often characterized by the mean absolute relative difference (MARD). However, MARD is influenced by a number of factors and little is known about whether MARD is stable throughout the day.

A total of 24 participants with type 1 diabetes were enrolled in the study. The study was performed for seven in-patient days. Participants wore two CGM systems in parallel and performed additional frequent blood glucose (BG) measurements. On two days, glucose excursions were induced.

MARD was calculated between pairs of CGM and BG values, with BG values serving as reference values. ARD values calculated from CGM-BG pairs were grouped by hour of the day. Results were analyzed separately for glucose excursion days and for regular days.

Total MARDs for the complete study duration were 12.5% ± 3.6% and 13.2% ± 2.4% (n = 24). Throughout the day marked variability of MARD was observed (8.0% ± 1.3%-16.3% ± 2.9% (G5); 9.1% ± 1.4%-16.3% ± 5.3% (FL), up to n = 157 each). Low(est) MARD values were observed before breakfast and dinner, when subjects were in or near a fasting state. Especially after breakfast and lunch, MARD values were higher than average.

Analytical performance of the two CGM systems, assessed by MARD, was found to vary markedly throughout the day. Activities of daily life likely triggered these variations. An increasing number of CGM users base therapeutic decisions on CGM values, and they should be aware of these variations of performance throughout the day.

Hybrid closed-loop (HCL) insulin pump therapy (Medtronic 670G) is an emerging technology that is growing in use worldwide. Initial clinical trials demonstrated the effectiveness of HCL in reducing hypoglycemia and improving glucose control; however, these subjects were intensely monitored and supervised. There has been concern regarding the ability of patients to remain in auto mode. We aimed to assess HCL when used in a typical outpatient endocrine clinic.

We initially analyzed data from 80 individuals with type 1 diabetes managed in an endocrine clinic by a single certified diabetes educator (CDE). We then included our other providers and had 230 subjects by the end of the study. Patients were either transitioned from traditional insulin pump or multiple daily insulin injection therapy (MDI) to HCL. Patients initiated to HCL pump therapy from July 2017 through February 2020 were studied. Endpoints of change in time in hypoglycemic/hyperglycemic range and time in target range were analyzed. The primary outcome was a change in percent time in the target range during manual mode compared with auto mode.

There was an 18.2% increase in average time in target range when comparing manual mode to auto mode (59.3% vs 70.1%, P < .0001). Average time in hyperglycemic range was significantly reduced by 26.7% (39.0% vs 28.6%, P < .0001) but without increasing average time in hypoglycemic range (1.7% vs 1.3%, P = 0.95).

HCL was effective in reducing hyperglycemia and increasing time in the target range but did not increase hypoglycemia. These data suggest HCL will improve the metrics of glucose control.

Older adults with type 1 diabetes (≥65 years) are often under-represented in clinical trials of automated insulin delivery (AID) systems. We sought to test the efficacy of a recently FDA-approved AID system in this population.

Participants with type 1 diabetes used sensor-augmented pump (SAP) therapy for four weeks and then used an AID system (Control-IQ) for four weeks. In addition to glucose control variables, patient-reported outcomes (PRO) were assessed with questionnaires and sleep parameters were assessed by actigraphy.

Fifteen older adults (mean age 68.7 ± 3.3, HbA1c of 7.0 ± 0.8) completed the pilot trial. Glycemic outcomes improved during AID compared to SAP. During AID use, mean glucose was 146.0 mg/dL; mean percent time in range (TIR, 70-180 mg/dL) was 79.6%; median time below 70 mg/dL was 1.1%. The AID system was in use 92.6% ± 7.0% of the time. Compared to SAP, while participants were on AID the TIR increased significantly (+10%, P = .002) accompanied by a reduction in both time above 180 mg/dL (−6.9%, P = .005) and below 70 mg/dl (−0.4%, P = .053). Diabetes-related distress decreased significantly while using AID (P = .028), but sleep parameters remained unchanged.

Use of this AID system in older adults improved glycemic control with high scores in ease of use, trust, and usability. Participants reported an improvement in diabetes distress with AID use. There were no significant changes in sleep.

Physical activity can cause glucose fluctuations both during and after it is performed, leading to hurdles in optimal insulin dosing in people with type 1 diabetes (T1D). We conducted a pilot clinical trial assessing the safety and feasibility of a physical activity-informed mealtime insulin bolus advisor that adjusts the meal bolus according to previous physical activity, based on step count data collected through an off-the-shelf physical activity tracker.

Fifteen adults with T1D, each using a continuous glucose monitor (CGM) and an insulin pump with carbohydrate counting, completed two randomized crossover daily visits. Participants performed a 30 to 45-minute brisk walk before lunch and lunchtime insulin boluses were calculated based on either their standard therapy (ST) or the physical activity-informed bolus method. Post-lunch glycemic excursions were assessed using CGM readings.

There was no significant difference between visits in the time spent in hypoglycemia in the post-lunch period (median [IQR] standard: 0 [0]% vs physical activity-informed: 0 [0]%, P = NS). Standard therapy bolus yielded a higher time spent in 70 to 180 mg/dL target range (mean ± standard: 77% ± 27% vs physical activity-informed: 59% ± 31%, P = .03) yet, it was associated with a steeper negative slope in the early postprandial phase (P = .032).

Use of step count to adjust mealtime insulin following a walking bout has proved to be safe and feasible in a cohort of 15 T1D subjects. Physical activity-informed insulin dosing of meals eaten soon after a walking bout has a potential of mitigating physical activity related glucose reduction in the early postprandial phase.

Excess carbohydrate intake during hypoglycemia can lead to rebound hyperglycemia (RH). We investigated associations between RH and use of real-time continuous glucose monitoring (rtCGM) and an rtCGM system’s predictive alert.

RH events were series of sensor glucose values (SGVs) >180 mg/dL starting within two hours of an antecedent SGV <70 mg/dL. Events were characterized by their frequency, duration (consecutive SGVs >180 mg/dL × five minutes), and severity (area under the glucose concentration-time curve). To assess the impact of rtCGM, data gathered during the four-week baseline phase (without rtCGM) and four-week follow-up phase (with rtCGM) from 75 participants in the HypoDE clinical trial (NCT02671968) of hypoglycemia-unaware individuals were compared. To assess the impact of predictive alerts, we identified a convenience sample of 24 518 users of an rtCGM system without predictive alerts who transitioned to a system whose predictive alert signals an SGV ≤55 mg/dL within 20 minutes (Dexcom G5 and G6, respectively). RH events from periods of blinded versus unblinded rtCGM wear and from periods of G5 and G6 wear were compared with paired t tests.

Compared to RH events in the HypoDE baseline phase, the mean frequency, duration, and severity of events fell by 14%, 12%, and 23%, respectively, in the follow-up phase (all P < .05). Compared to RH events during G5 use, the mean frequency, duration, and severity of events fell by 7%, 8%, and 13%, respectively, during G6 use (all P < .001).

Rebound hypreglycemia can be objectively quantified and mitigated with rtCGM and rtCGM-based predictive alerts.

Continuous subcutaneous insulin infusion (CSII) is a common diabetes treatment modality. Glycemic outcomes of patients using CSII in the first 24 hours of hospitalization have not been well studied. This timeframe is of particular importance because insulin pump settings are programmed to achieve tight outpatient glycemic targets which could result in hypoglycemia when patients are hospitalized.

This retrospective cohort study evaluated 216 hospitalized adult patients using CSII and 216 age-matched controls treated with multiple daily injections (MDI) of insulin. Patients using CSII did not make changes to pump settings in the first 24 hours of admission. Blood glucose (BG) values within the first 24 hours of admission were collected. The primary outcome was frequency of hypoglycemia (BG < 70 mg/dL). Secondary outcomes were frequency of severe hypoglycemia (BG < 40 mg/dL) and hyperglycemia (BG ≥ 180 mg/dL).

There were significantly fewer events of hypoglycemia [incident rate ratio (IRR) 0.61, 95% confidence interval (CI) 0.42–0.88, p = 0.007] and hyperglycemia (IRR 0.79, 95% CI 0.65–0.96, p = 0.02) in the CSII group compared to the MDI group. There was a trend toward fewer events of severe hypoglycemia in the CSII group (IRR 0.15, 95% CI 0.02–0.93, p = 0.06).

Patients using CSII experienced fewer events of both hypoglycemia and hyperglycemia in the first 24 hours of hospital admission than those treated with MDI. Our study demonstrates that CSII use is safe and effective for the treatment of diabetes within the first 24 hours of hospital admission.

Portable retinal cameras and deep learning (DL) algorithms are novel tools adopted by diabetic retinopathy (DR) screening programs. Our objective is to evaluate the diagnostic accuracy of a DL algorithm and the performance of portable handheld retinal cameras in the detection of DR in a large and heterogenous type 2 diabetes population in a real-world, high burden setting.

Participants underwent fundus photographs of both eyes with a portable retinal camera (Phelcom Eyer). Classification of DR was performed by human reading and a DL algorithm (PhelcomNet), consisting of a convolutional neural network trained on a dataset of fundus images captured exclusively with the portable device; both methods were compared. We calculated the area under the curve (AUC), sensitivity, and specificity for more than mild DR.

A total of 824 individuals with type 2 diabetes were enrolled at Itabuna Diabetes Campaign, a subset of 679 (82.4%) of whom could be fully assessed. The algorithm sensitivity/specificity was 97.8 % (95% CI 96.7-98.9)/61.4 % (95% CI 57.7-65.1); AUC was 0·89. All false negative cases were classified as moderate non-proliferative diabetic retinopathy (NPDR) by human grading.

The DL algorithm reached a good diagnostic accuracy for more than mild DR in a real-world, high burden setting. The performance of the handheld portable retinal camera was adequate, with over 80% of individuals presenting with images of sufficient quality. Portable devices and artificial intelligence tools may increase coverage of DR screening programs.

HbA1c result provide information on metabolic control in diabetes mellitus (DM) and could also be used for its diagnosis. For its determination, the laboratory must be certified by the National Glycohemoglobin Standardization Program (NGSP) or the International Federation of Clinical Chemistry (IFCC) and comply with a strict quality control program.

To determine the correlation and agreement between HbA1c results measured by three analytical methods (enzymatic, turbidimetric, and capillary electrophoresis) versus HPLC.

Method comparison—1245 samples from equal number of subjects at 45 Association of High Complexity Laboratories (Asociación de Laboratorios de Alta Complejidad—ALAC) centers, centralizing sample processing and operator. Statistical analysis—analysis of variance (ANOVA) and nonparametric Friedman ANOVA test for related samples, means, and medians. Correlation and concordance—Pearson’s correlation and linear regression, intraclass correlation coefficient (Passing and Bablock and Bland and Altman).

The comparison of mean values obtained by the four methods showed statistically significant, but clinically irrelevant, differences: HbA1c by HPLC versus Electrophoresis 0.06% (0.42 mmol/mol) P = .000 (± 1.96 DS -0.070 -0.047), Enzymatic 0.087% (1 mmol/mol) P = .000 (± 1.96 DS 0.077 0.098), Turbidimetric 0.056% (0.38 mmol/mol) P = 0.000 (± 1.96 DS -0.067 -0.044). Their concordance showed intraclass correlation of single measures of 0.982 P < .001 (95% CI 0.987 - 0.9838).

The three methods present low variability and high correlation versus the HPLC.

The ability to measure insulin secretion from pancreatic beta cells and monitor glucose-insulin physiology is vital to current health needs. C-peptide has been used successfully as a surrogate for plasma insulin concentration. Quantifying the expected variability of modelled insulin secretion will improve confidence in model estimates.

Forty-three healthy adult males of Māori or Pacific peoples ancestry living in New Zealand participated in an frequently sampled, intravenous glucose tolerance test (FS-IVGTT) with an average age of 29 years and a BMI of 33 kg/m². A 2-compartment model framework and standardized kinetic parameters were used to estimate endogenous pancreatic insulin secretion from plasma C-peptide measurements. Monte Carlo analysis (N = 10 000) was then used to independently vary parameters within ±2 standard deviations of the mean of each variable and the 5th and 95th percentiles determined the bounds of the expected range of insulin secretion. Cumulative distribution functions (CDFs) were calculated for each subject for area under the curve (AUC) total, AUC Phase 1, and AUC Phase 2. Normalizing each AUC by the participant’s median value over all N = 10 000 iterations quantifies the expected model-based variability in AUC.

Larger variation is found in subjects with a BMI > 30 kg/m², where the interquartile range is 34.3% compared to subjects with a BMI ≤ 30 kg/m² where the interquartile range is 24.7%.

Use of C-peptide measurements using a 2-compartment model and standardized kinetic parameters, one can expect ~±15% variation in modelled insulin secretion estimates. The variation should be considered when applying this insulin secretion estimation method to clinical diagnostic thresholds and interpretation of model-based analyses such as insulin sensitivity.

School-aged children often participate in type 1 diabetes (T1D) self-care tasks. Despite widespread discussion about the importance of developing self-care skills in childhood, few explain how the health care team should assess the skills of children with T1D when performing insulin injections.

We sought to assess content validity evidence in two checklists regarding injection technique performed by children.

Two checklists were designed based on a systematic review of the insulin injection technique. Experts in pediatric diabetes, health literacy, and diabetes education assessed the checklists regarding their clarity, objectivity, and relevance. Content validity was assessed using the content validity ratio (CVR).

Eleven providers (72% nurses or physicians, professional experience 19.4 ± 10.1 years, 45% of specialists in endocrinology, and 18% in pediatrics) participated in the assessment. Experts considered items containing the word homogeneity inappropriate. Items related to the needle insertion angle and the skin fold did not reach the CVR critical value. The final version of the checklist for syringe injection comprised 22 items with CVR = 0.91, and the checklist for pen injection comprised 18 items with CVR = 0.87.

The checklists presented clear, objective, and relevant content that assesses the skills of children with T1D for insulin injection. The checklists formally present the order of the technique and all the steps for insulin injection and allow a quantitative assessment of the operational skills of children. The developed instruments offer providers the possibility of continuous assessment of the progress of the pediatric clientele until they reach independence in diabetes self-care.