The program times are listed in Central European Time (CEST)

Displaying One Session

ORAL PRESENTATION SESSION
Date
Wed, 02.06.2021
Session Type
ORAL PRESENTATION SESSION
Session Time
18:35 - 19:55
Room
Hall C
ORAL PRESENTATION SESSION

A DATA-DRIVEN CLASSIFIER OF DAILY CONTINUOUS GLUCOSE MONITORING (CGM) PROFILES

Abstract

Background and Aims

With the proliferation of CGM, massive databases of CGM traces (daily profiles) are constantly growing. A question therefore arises: are all of these profiles substantially different or is there a finite set of distinct daily profiles which sufficiently approximate all possible profiles? We propose a data-driven approach to determine a finite set of “motifs” – representative daily profiles – such that almost any daily profile can be matched to one of the motifs.

Methods

Data: 595 individuals with type 1 or type 2 diabetes (T1D, T2D) participating for 3-6 months in either the International Diabetes Closed-loop (iDCL) Trial or in Dexcom’s DIaMonD study. A set of 226 motifs was constructed by clustering 4,802 (training) profiles from the iDCL Protocol 1 study (T1D) and identifying the motif for each cluster. The representative set of motifs was then tested using profiles from the iDCL Protocol 3 and DIaMonD studies (T1D and T2D), which included a variety of treatment modalities, e.g. daily insulin injections, insulin pumps, and artificial pancreas.

Results

Over 98.8% of the 39,916 testing profiles were successfully classified using the motifs. Each cluster of profiles from the testing data had similar clinical characteristics (e.g., time within or above range) to the corresponding cluster of profiles from the training data.

Conclusions

The finite set of motifs can sufficiently describe almost any daily profile, and the clinical characteristics of each motif are representative of the CGM profiles clustered around it. The motifs can be used for predictive modeling, decision support, or automated closed-loop control.

Hide
ORAL PRESENTATION SESSION

VALIDATION OF A PREDICTIVE PHYSIOLOGICAL MODEL TO SUPPORT PERSONALIZED MEDICINE IN TYPE 2 DIABETES

Abstract

Background and Aims

We evaluated the accuracy of a predictive physiological model for simulating the predicted glucose response to metformin therapy in individuals with type 2 diabetes (T2D). The model is a computer-based, interactive decision support system that enables clinicians to perform individualized simulations that test various treatment options for the prediction of glucose profiles in response to various metabolic interventions.

Methods

This two-step validation study used an independent dataset collected in a clinical trial that assessed glucose response in a cohort of individuals with type 2 diabetes who transitioned from diet/exercise treatment to metformin therapy. The primary objective was to evaluate the quality of the model both in terms of fitting the dataset as well as the accuracy of prediction.

Results

Data from 16 T2D patients were included in the analysis. The overall fit between observed and modeled glucose profile values showed concordance pre- and post-metformin treatment (Figure) with notable accuracy as measured by mean absolute relative difference (MARD): 8.1% and 12.6%, respectively. Parkes Error Grid analysis also showed strong correlation pre- and post-therapy: r2 = 0.89 & SSE = 18.3 and r2 = 0.55 & SSE = 27.9, respectively, which was confirmed by two-sample Kolmogorov-Smirnov test.

Conclusions

The integration of predictive modeling approaches into clinical decision support tools has the potential to optimize clinician time and accuracy in determining the most effective treatment regimen for each T2D patient.

Hide
ORAL PRESENTATION SESSION

AN INTERPRETABLE LSTM-BASED PREDICTION MODEL FOR ASSESSING THE RISK OF HOSPITALIZATION AND RE-HOSPITALIZATION IN YOUTH WITH TYPE 1 DIABETES MELLITUS 

Abstract

Background and Aims

Diabetic Ketoacidosis (DKA) and hyperglycemia with ketosis in the absence of acidosis constitute major causes of hospital admission and morbidity in children and adolescents with Type 1 Diabetes Mellitus (T1DM). This study aims at the development of an interpretable prediction model for the risk assessment of hospitalization and re-hospitalization in children and adolescents with T1DM.

Methods

Data collected from a two-year follow-up of 127 T1DM patients at the “Agia Sofia” Children’s Hospital, within the framework of the “SWEET” Initiative, were used for development and evaluation purposes. Frequently identified risk factors for recurrent DKA admissions were considered to compose the input space.

The model was based on Long Short-Term Memory Neural Networks (LSTM) in order to leverage LSTM’s efficiency in handling sequential data. The unbalanced nature of the dataset was addressed by applying an ensemble learning method, based on a sub-sampling approach. Interpretation of the model’s decisions was achieved by deploying the Local Interpretable Model-agnostic Explanations (LIME) technique.

data_table.png

Results

The 3-fold cross-validation criterion was applied to assess the model’s generalization ability. The obtained results in terms of C-statistic (71.34 ± 0.05%) indicate acceptable discrimination ability. Meaningful insights about the influence of the considered risk factors on the model’s decisions were provided by applying the LIME technique.

lime_image_abstract.png

Conclusions

The obtained results demonstrate the potential of the proposed method to achieve high accuracy while ensuring user’s trust by providing appropriate explanations on the produced decisions. Acknowledgment: Supported within the framework of the ENDORSE project, which is funded by the NSRF (Grant agreement: Τ1ΕΔΚ-03695).

Hide
ORAL PRESENTATION SESSION

THE ROLE OF EGDR AS A PREDICTOR OF INSULIN RESISTANCE AND CARDIOVASCULAR DISEASE

Abstract

Background and Aims

People with type 1 diabetes (T1D) have a high risk of cardiovascular disease (CVD) which may be accelerated by insulin resistance. Estimated glucose disposal rate (eGDR) correlates well with the euglycaemic clamp. We aimed to assess the association between eGDR, liver steatosis and CVD.

Methods

Adult T1D subjects were consecutively screened for liver steatosis using ultrasound (US), Fatty Liver Index (FLI) and controlled attenuation parameter (CAP). The eGDR was calculated based on hypertension, HbA1c and waist circumference. CVD was assessed based on patient files.

Results

CVD was present in 34 out of 355 subjects. Divided into tertiles (<5.39,5.39-7.79,>7.79), 36.6% expressed low eGDR; 32.7% intermediate eGDR and 30.7% high eGDR. There was moderate correlation between eGDR and FLI (r=0.68,p<0.001) and weak correlation with US (r=0.33,p<0.001) and CAP (r=0.50,p<0.001). In the low eGDR group (=insulin resistant group) not only steatosis (38.5% vs. 11.2% (intermediate eGDR) and 12.8% (high eGDR)) but also composite CVD (18.5% vs. 6.0% and 2.8%) were significantly more present (p<0.001 for both). Low eGDR (OR:4.2[2.2-8.2],p<0.001), but not BMI or dyslipidaemia was independently associated with US-defined liver steatosis. Low eGDR was also independently associated with FLI-determined steatosis (OR:5.5[1.7-17.6],p=0.004) together with BMI (OR:1.6[1.4-1.9],p<0.001). Low eGDR (OR:8.0[2.3-27.4],p=0.001) and liver steatosis (OR:2.7[1.2-6.1],p=0.022 (US-defined), OR:2.9[1.4-6.0],p=0.005 (FLI-defined)) were independently associated with composite CVD, but presence of metabolic syndrome, dyslipidemia and BMI were not.

Conclusions

Insulin resistance is prevalent in T1D. eGDR correlates with the presence of liver steatosis. Both eGDR and liver steatosis correlate with prevalent CVD.

Hide
ORAL PRESENTATION SESSION

ENABLING PATIENT-DATA INTERACTIONS IN TYPE 1 DIABETES THROUGH A CLOUD-BASED SIMULATION TOOL

Abstract

Background and Aims

Advanced insulin therapy in type 1 diabetes (T1D) relies on key individual treatment profiles such as basal rate, carbohydrate ratio, and correction factor. Periodic adjustments of these profiles are needed based on review of data that usually require manual downloads from multiple devices. The aim of this project is to design and test a novel, cloud-based, centralized platform – the Web-Based Simulation Tool (WST) – that allows users to quickly and safely explore changes to their treatment practices.

Methods

WST automatically collects data from the patients’ insulin pumps via Tandem t:connect technology and generates personalized models of their glucose metabolism on a daily basis. It is equipped with a simple user-interface where users can visualize their data, run simulations with modified meals and insulin parameters, and generate reports. An outpatient pilot study with fifteen adult participants with T1D is currently being conducted to evaluate WST usability and performance.

Results

WST has already processed 233 days of data from which it was able to generate 195 models (83.69% success rate) with an average RMSE and MARD of 15.98±6.45 mg/dl and 7.95±3.15%, respectively, and with 99.59% of reconstructed glucose values in the A- and B-zones of the Clarke Error Grid. Analysis of responses to technology expectation/acceptance and psychobehavioral questionnaires will be completed at the end of the trial.

Conclusions

Simulation technologies helps leverage the vast amount of diabetes data currently available, enabling novel patient-data interactions that could facilitate decision-making processes related to the optimization of T1D treatment strategies.

Hide
ORAL PRESENTATION SESSION

EVALUATION OF THE HYPOGLYCAEMIA PREDICTIVE ALGORITHM IN THE INSULCLOCK® INSULIN PEN CAP DIGITAL PLATFORM IN TYPE 1 DIABETES TREATED WITH INSULIN MULTIDOSE.

Abstract

Background and Aims

Insulclock® is a small electronic device that functions as a cap fitted to the available insulin pens and monitors the date, time and dose of insulin, and information from glucometers and continuous glucose monitors (CGM). Store the information in an app designed for this purpose. Our goal was to evaluate the accuracy of an algorithm for hypoglycemia (HG) prediction by the Insulclock app using the device's information exclusively.

Methods

An original HG (glucose <70mg/dL) predictive algorithm was developed that uses data from Insulclock® and the Freestyle Libre® (Abbott) CGM. It alerts the risk of HG and the expected time up to it. Intakes are automatically detected using the GRID method. Subsequently, it has been evaluated for 180 days in a patient 47 years old with DM1 30 years ago. We consider correct alarms real HG avoided with an intake; false positive if after the expected time for HG, it does not arrive without having eaten; false negative, HG without previous alarm. Additionally, the error in the calculated time and the total number of HG events were evaluated.

Results

132 alarms issued. Correct alarms 90 (84.9%); false positives 42 (31.8%); false negatives 16 (15.1%); 116 actual HG events (77.5% detected). The average advance time detection was 87 minutes. The average absolute error value in the time prediction for hypoglycemia is 35 minutes.

Conclusions

The predictive algorithm tested allows to detect and alert in advance to potentially avoid a high number of hypoglycemia events using only information obtained automatically by the Insulclock®device. New studies should expand and confirm this experience.

Hide
ORAL PRESENTATION SESSION

TIME IN RANGE AND GLUCOSE CONTROL IN PATIENTS WITH TYPE 1 DIABETES USING A MOBILE APP-ASSISTED CARBOHYDRATE COUNTING

Abstract

Background and Aims

Background: carbohydrate (CHO) counting is often performed inaccurately by patients with type 1 diabetes (T1D). We hypothesized that mobile App “Dietrometro”, that estimates CHO content of food pictures and weights, would ameliorate glucose control in T1D subjects.

Aim: to study the effect of Dietometro on glucose control.

Methods

Methods: 60 T1D subjects (aged 18-60 years, 31 males), on multiple daily injections (n=26) or continuous subcutaneous insulin infusion (n=34), were randomly assigned to three groups: no CHO counting (group 1; n=21), “self-managed” counting (group 2; n=19) and App-assisted counting (group 3; n=20). Main outcomes were TIR at one-month and HbA1c at three-months follow-up. Time above (TAR) and below range (TBR) were estimated by flash (Freestyle Libre-1) or continuous glucose monitoring (Guardian, Dexcom G6).

Results

Results: Age, gender, type of insulin, glucose monitoring system and baseline TIR were similar between groups, while baseline HbA1c was lower in group 3 compared to group 1 (6.9±1.06 vs. 7.8±0.85%; p<0.05). At one-month follow-up, TIR was higher in group 3 compared to group 1 (66.53±14.71 vs. 54.38±14.23; p=0.02), but similar between other groups (p>0.11), although this difference disappeared after adjustment for baseline HbA1c. At three-months follow-up, groups 2 and 3 had a lower HbA1c than group 1 (7.17±0.86 vs. 6.88±1.05 vs. 8.18±1.06%, respectively; p=0.001). TAR and TBR were similar between groups at baseline and after one-month follow-up.

Conclusions

Conclusions: app-assisted CHO counting might improve TIR in T1D. Larger sample size and longer follow-up are needed to define the long-term effect of this system and its advantage over self-managed counting.

Hide
ORAL PRESENTATION SESSION

AN INNOVATIVE ONLINE SELF-DETERMINATION PROGRAM TO IMPROVE SELF-MANAGEMENT IN YOUNG ADULTS WITH TYPE 1 DIABETES

Abstract

Background and Aims

Background: For young adults with type 1 diabetes (YAWD), it can be difficult to manage diabetes due to competing lifestyles and work commitments. The Guided Self-determination (GSD) method is an evidence-based self-management program that aims to improve motivation to make positive lifestyle changes.

The study aim was to evaluate the feasibility and efficacy of the online GSD program in improving diabetes self-management skills among YAWD.

Methods

Methods Nine Diabetes Educators (DEs) attended a 1.5 day face-to-face training course. YAWD aged 18 to 30 were recruited from a Young Adults Diabetes Service clinic, from Consumer organisation and from University student forums. DEs completed the online GSD program flexibly with YAWD over 3 to 6 months. Online validated surveys measures of autonomy, competency and communication with healthcare providers were completed by YAWD before and after participation. Follow-up surveys invited comments on YAWDs’ and DEs’ experience of the program and DEs participated in a focus group.

Results

Results 15 young adults have completed the program. DEs indicated that the program has changed the way they communicate with their clients, and that the online GSD approach was most successful when applied flexibly, to suit YAWDS’ preferred time of day, learning style and mode of conversation (eg telephone versus videoconferencing).

Conclusions

Conclusions. The online GSD program is feasible; efficacy in improving self-management skills has yet to be assessed in a larger sample of YAWD, however, has the potential to empower YAWD to improve diabetes care and facilitate access to healthcare 24/7 and regardless of location.

Hide