Background and Aims

Telehealth can bring care into the homes of patients. However, there is a risk that telehealth may worsen health care disparities.

Methods

Prior to the COVID-19 pandemic, patients were only eligible for telehealth if referred by their diabetes provider. Providers assisted patients with device downloads and technical issues. There was no access to interpreters, social workers, or nutritionists.

The COVID-19 pandemic necessitated transition to telehealth in March 2020. Certified diabetes educators helped obtain device downloads, medical assistants provided connection support, and workflows were developed to incorporate interpreters, social workers, and nutritionists into telehealth visits (Fig1).

Chart review was performed for telehealth visits between July 1, 2017 and April 30, 2020. Visits for children with public insurance, a marker of lower socioeconomic status, and those who were non-English speaking were determined.

Results

In the 31 months prior to COVID-19, 195 telehealth visits were performed and in the first 6 weeks of the pandemic, another 436 telehealth visits were completed. In our practice, 38.4% of children have public insurance and 17.4% of the population is non-English speaking. The percentage of children with public insurance who accessed telehealth increased from 24.1% to 39.9% with the new workflow (p=0.004). The percentage of non-English speakers accessing telehealth increased from 3.1% to 13.5% (p<0.01) with the new workflow.

Conclusions

Prior to the COVID-19 pandemic, our workflows were sub-optimal and this increased disparities for children who were non-English speaking or from lower socioeconomic status. The creation of inclusive workflows and support for patients and providers helped close the disparities gap.

Background and Aims

Hypoglycemia is one of the most relevant acute complications of diabetes mellitus and is associated with an increased risk of driving accidents. Today's cars gather a broad spectrum of real-time driving parameters. Based on changes in driving behaviour during hypoglycemia we aim at establishing algorithms capable of discriminating eu- and hypoglycemic driving patterns using artificial intelligence.

Methods

We included active drivers with type 1 diabetes mellitus (T1DM). Using an adapted hypoglycemic clamp protocol and a professional driving simulator, driving data was recorded in 2 glycemic states: euglycemia (deu, 5-8 mmol/l) and hypoglycemia (dhypo, 2.0-2.5 mmol/l). In each glycaemic state the participants drove for 15 min through a random sequence of 3 environments: highway, rural and town. Car-based sensor data were sliced into overlapping 45-second windows. Finally, predictive performance in hypoglycemia detection was evaluated using gradient boosted decision trees.

Results

The study encompassed 15 participants with T1DM (11 male, HbA1c 7.2±0.6 %). Mean blood glucose in deu and dhypo was 5.9±0.6mmol/l and 2.4±0.25 mmol/l (p<0.001), respectively. Car-based data provided 466,303 measurements in deu and 481,497 samples in dhypo. 1-fold cross-validation on subject level resulted in a ROC-AUC in hypoglycemia prediction of 0.85. ROC-AUC for the highway, rural and town environment was 0.80, 0.98, and 0.78, repectively.

Conclusions

Our study applying machine learning models on driving simulator-based data shows robust between-subject predictability of hypogylcemia. This confirms the effectiveness of artificial intelligence in hypoglycemia detection while driving and may represent a promising novel approach to increase traffic safety in people with diabetes.

Background and Aims

Connected insulin pens have the potential to objectively assess the impact of insulin dosing behaviors on postprandial glucose control.

Methods

This observational study enrolled 68 people with T1D or insulin-using T2D with A1C ≥8% and ≥3 reported insulin boluses/day at baseline. Connected pen and CGM data were used to identify meal-related glucose excursions defined as glucose increase ≥50 mg/dL within 2 hours with no glucose <70 mg/dL in the preceding 30 min. A pre-meal bolus (PB) was taken within one h prior to the nadir of excursion or during excursion with maximum of rate-of-change of BG within preceding 20 minutes less than 1.0 mg/dL/min. Doses taken during the excursions that were not labeled as PB were defined to be delayed bolus (DB), while doses taken within 6 h of nadir and after the peak but before the next nadir of excursions were defined as correction bolus (CB). We defined an excursion to have a missed bolus (MB) if neither PB nor DB was taken.

Results

Participants were (mean±SD) 48 years (± 12), 44% female, 59% T1D, 51% BMI ≥30, and 65% with A1C ≥9.0%. DB and MB significantly impaired postprandial glucose control compared to PB (table). Use of a CB reduced the impact of DB and MB, but slightly increased time below range.

Conclusions

These findings highlight the clinical utility of connected pens to identify insulin dosing practices that contribute to suboptimal postprandial control. Individuals who do not inject consistently premeal could potentially benefit from faster-acting bolus insulins.

Background and Aims

Many people with T2D will eventually need insulin treatment to achieve glycaemic control and to reduce complications from chronic hyperglycaemia. However, insulin treatment initiation is often delayed, and HbA_1c targets are frequently not achieved in insulin-treated patients; the latter is due, in part, to insufficient titration, inadequate dosing, or missed doses. We aim to highlight current challenges in managing insulin treatment and how new technologies may address these barriers.

Methods

We searched PubMed and materials from four national and international diabetes conferences in this focussed literature review. The term ‘diabetes’ was combined with the following: insulin titration, insulin AND digital, smart phone, digital health technology, smart pen, connected pen, connected device. More than 300 resulting publications were manually filtered.

Results

A number of publications reported technology-based interventions including software tools or devices to overcome barriers to effective treatment. Of the software-based technologies selected for discussion (Table), most were developed for smartphones and demonstrated equivalent or improved glycaemic outcomes and required less contact time with healthcare practitioners (HCPs) versus controls or previous care setting. The new devices chosen generally aimed to track doses and dose timing.

Conclusions

Key features that new technology should offer include efficacy at improving glycaemic control, ease of use, accurate data capture, accessibility of data to the HCP and insulin user, and data security. A solution that connects continuous glucose monitoring, dose recording, help with titration, and recording of lifestyle factors might reduce treatment complexity and burden and result in improved titration and higher treatment adherence.

Study funded by Sanofi

Background and Aims

Type 2 diabetes (T2D) can be diagnosed with the oral glucose tolerance test or with hemoglobin A1C (HbA1c); however, the reproducibility and concurrence between these tests is suboptimal. Continuous glucose monitoring (CGM) may allow for convenient and accurate T2D diagnosis. We assessed whether a factory-calibrated CGM system (Dexcom G6), worn in blinded mode for a single wear period, can be used to diagnose T2D.

Methods

We developed a binary classification diagnostic CGM ("dCGM") algorithm based on CGM and HbA1c data using a dataset of 716 individual CGM sensor sessions with associated HbA1c measurements from seven clinical trials. Data from 623 sensor sessions were used for training and 93 subjects for testing (49 normals [HbA1c <5.7%], 27 prediabetes, and 17 T2D [HbA1c ≥6.5%] not using pharmacotherapy). dCGM performance was evaluated against the accompanying HbA1c measurement which was assumed to provide the correct diagnosis.

Results

The dCGM algorithm's overall sensitivity, specificity, positive predictive value, and negative predictive value were 71%, 93%, 71%, and 93%, respectively. At other clinically relevant HbA1c thresholds, dCGM specificity among normals was 98% (48/49 correctly classified) and for subjects with suboptimally-controlled diabetes (HbA1c ≥7%, above the ADA recommended HbA1c goal) the sensitivity was 100% (8/8 subjects correctly diagnosed with T2D).

Conclusions

We have shown in a small dataset that dCGM has good performance for the diagnosis of T2D. Thus a factory-calibrated CGM system with a dCGM algorithm is a feasible alternative for the diagnosis of T2D and warrants further investigation.

Background and Aims

One in five youth with T1D experience worsening HbA1c values between quarterly visits. We evaluated the effectiveness of KidCare Anywhere (KCA), a direct-to-consumer telehealth intervention offering problem-solving and education to identify and manage glucose patterns for youth predicted to experience a rise in A1c 70-110 days following routine clinical visits.

Methods

Patients received care at a tertiary diabetes clinic in the U.S. Midwest. Supervised machine learning was used to develop a random forest-based model to predict 90-day change in A1c. Clinic staff reviewed weekly lists of patients with a predicted 90-day rise in A1c of ≥3 mmol/mol. From these lists, 61 patients under 20yrs old with baseline A1c ≥55 mmol/mol were enrolled in KCA. Youth received 1-6 brief telehealth sessions with a trained interventionist over 90 days before their next routine clinic visit. During each session, families reviewed device data with the interventionist and received personalized insulin regimen adjustments and problem-solving support.

Results

Study cohort was 73% white, 3% Hispanic, 62% female, 53% on CGM, 64% on insulin pump, median age 13.97yrs (IQR=10.39,16.13), baseline A1c 64 mmol/mol (60,73), and follow-up A1c 66 mmol/mol (58,79). Actual 90-day A1c change was significantly lower than the predicted 90-day A1c change (p=0.0088). Of 61 KCA patients predicted to have A1c rise ≥3 mmol/mol, only 21 (34%) did.

Conclusions

Findings suggest KCA may lead to improved glycemic levels by preventing clinically significant 90-day rise in A1c. Future research should evaluate the efficiency of this intervention in a randomized controlled setting to better define factors associated with intervention efficacy.

Background and Aims

Clinical trials are an essential instrument to test newly developed solutions for diabetes management and care. Multivariable and multisensor data that are usually collected in such studies should be structured to enhance their fruition, synchronized to a telemonitoring interface, and finally analyzed to assess the methodology under examination. The process of collecting and organizing data is complex and requires ad-hoc multidomain infrastructures. This work presents a newly developed telemedicine platform that allows real-time data gathering and monitoring during clinical studies.

Methods

The platform is composed by a cloud database, a mobile application and a web interface (see Figure, upper panel). The mobile application allows to log daily-life events and automatically collects data from Dexcom’s CGM devices and health vitals from both Apple Watch and Fitbit smartwatches. Data are streamed to the cloud database through secure RESTful APIs and ultimately exposed in real-time to clinicians through an easy-to-use web interface. The platform complies with the General Data Protection Regulation and ensure modularity to allow fast implementation of new algorithms for their assessment.

Results

Currently undergoing tests on a diabetic individual show that the platform is robust and performant. An example of data collected in such pilot study is reported (Figure, lower panel). No errors or disconnections have been experienced so far.

Conclusions

The developed platform is proving to be an efficient tool to gather and visualize in real-time multivariable and multisensor data. Next steps include an ad-hoc study to extensively test and validate it on a large population.

Background and Aims

During the COVID-19 pandemic, appointments for diabetes clinic visits rapidly switched to a telemedicine format, impacting all aspects of the routine clinical care, including: sharing data, clinic flow, technological readiness, and billing. To understand this shift, we surveyed member clinics of the T1D Exchange QI Collaborative.

Methods

A total of 21 clinics across the USA completed surveys. Telemedicine was defined as any video visit or telephone visit that took place in lieu of a face to face in response to social distancing measures during COVID-19 pandemic. Outcome metrics included survey responses and monthly metrics regarding telemedicine visits (January - August, 2020). The survey covered topics related to access to technology tools, the telehealth visit process, and insurance coverage.

Results

Of 21 clinics (16 pediatric and 5 adult), 62% used both video software and phone calls. Clinics reported that insurance covered 95% of telemedicine visits during the pandemic (see Table 1). All clinics had access to Carelink, T-Connect, Glooko, and Clarity platforms to support remote monitoring of patients.

Over half (62%) of clinics instituted workflows to obtain patient lab results, less (38%) had a system for conducting depression screening. Only 3 clinics had psychologists available to participate in telemedicine.

Clinics described similar rates of prescribing for CGM and pumps (62%). . Clinics continued to provide support for pumps (100%) and CGM (70%).

Conclusions

Physicians and insurers have adopted telemedicine with remarkable speed. Future studies will assess the effectiveness of telemedicine visits during this pandemic in different patient populations.

Background and Aims

Successful management of patients with T2D relies on successful treatment initiation and adherence. A mobile health (mHealth) app that includes planning, body weight and dose tracking features has been developed to support patients during the initiation (first 12 weeks) of once-weekly (OW) semaglutide, a glucagon-like peptide-1 receptor agonist approved for type 2 diabetes. Here, we provide an overview of users’ interaction with the app since its launch in August 2018.

Methods

Data were analysed both overall and separately for active (≤3 consecutive weeks without interacting with [opening] the app) and less active (>3 consecutive weeks without interacting with the app) users.

Results

Of 7,789 registered users, 2,820 were active and 4,969 were less active. ‘How to’ resources were frequently viewed during the first week by both user groups. Resources most viewed were ‘when to take’ and ‘side effects’. Overall, 73.1% registered users set up injection reminders during the first week and 39.8% set up anchoring plans (where the user anchors injection behaviour to an existing habit in their routine). Setting up anchoring plans resulted in more consistent self-recorded injection rates throughout the 12-week dose escalation period in active (+7%) and less active (+20%) users vs users not setting up plans; both p<0.05 (Figure)

Conclusions

This analysis suggests that use of the mHealth app and setting up anchoring plans is associated with more consistent use of treatment as assessed by self-recorded injection rates, regardless of overall engagement levels. This highlights the benefits of medication-specific mHealth apps as patient support tools when initiating new medications.