Welcome to the ATTD 2023 Interactive Program

Background and Aims

The frequency and cause of hypoglycemia in various categories of septic patients have not been adequately explored. In this study, we focused on sepsis-associated hypoglycemia in the early phase of critically ill patents and evaluated the impact of hypoglycemia on mortality.

Methods

We performed a retrospective cohort study using the Medical Information Mart for Intensive Care IV, anonymised database (MIMIC-IV), based on the data of Intensive Care Unit (ICU) admissions between 2008 and 2012 at Beth Israel Deaconess Medical Center, USA. The study protocol was approved by the respective Institutional Review Boards.
We selected 31461 patients with Sepsis-3 criteria from the MIMIC-IV database for the analysis. Figure 1 depicts the stages of patient inclusion in the study.

Figure 1.

Results

We performed survival analysis with ICU mortality as the target, stratified per glucose level. Figure 2 shows the Kaplan-Meier curves.

Figure 2. Kaplan-Meier survival curves for patients with sepsis in the five categories of blood glucose levels.

The following table shows the results of a Cox proportional hazards model, per glucose category, after adjusting for age, sex, OASIS (Oxford Acute Severity of Illness Score) and SOFA (Sequential Organ Failure Assessment) scores.

Conclusions

The survival curves for severe and mild hypoglycemia seem to be quite below the curve for euglycemia, indicating lower survival rates, as is also the case for severe hyperglycemia, although less prominently.

The proportional hazards model suggests that severe and mild hypoglycemia are significantly (p < 0.05) associated with increased mortality rates for septic patients, as is also severe hyperglycemia.

Background and Aims

The abundance and complexity of CGM-data constitute a challenge. Therefore, we created a composite score combining all aspects of glucose control in order to maximize its use. By using real-world data, we evaluated how our score correlates with traditional CGM-metrics.

Methods

CGM-data was collected from 194 pediatric type 1 diabetes subjects (age 13.8±4.2, disease duration 5.2±3.6), corresponding to approximately 5200 patient-weeks. Correlations between our score and CGM-metrics for a 14-day period were computed. The latest 14-day period of each patient was categorized into three groups based on the median score (low<60, medium 60-75 and high >75; maximum score = 100).

Results

We observed a positive correlation between our score and time in range (TiR) (r=0.84, p<0.001). A negative correlation was observed for eHbA1c (r=-0.77, p <0.001), time above range (r=-0.77, p<0.001), coefficient of variation (r=-0.39, p<0.001). Also, the number of severe hyperglycemic- (r=-0.49, p<0.001) and severe hypoglycemic- (r=-0.11, p<0.001) episodes correlated negatively with our score. The three different groups differed both for eHbA1c (72±15 vs. 53±7 vs. 45±7 mmol/mol, p<0.001) and TiR (41±12 vs. 63±10 vs. 77±11 %, p<0.001). However, it should be noted that also in the group with the lowest score there was a wide range of eHbA1c (49-124 mmol/mol) and TiR (10-65%).

Conclusions

Our composite score can be used to rapidly assess several aspects of CGM-data. By grouping patient-weeks we can stratify patients while still accounting for several aspects of glycemic control.

Background and Aims

Background and Aims

In people with type 1 diabetes (T1D), physical activity (PA) affects blood glucose (BG) concentration during exercise and for up to 12 to 24 hours in recovery [1]; therefore, it is essential to design appropriate treatment decisions to prevent PA-induced hypoglycemia in T1Ds.

Methods

Methods

We utilized a publicly available dataset [2] to examine the effect of various strategies on a group of T1Ds who participated in four 45-minute fasted aerobic exercise sessions as described in Table 1. We examined the effect of each treatment on post-exercise glycemic stability by analyzing the total number of hyperglycemia and hypoglycemia events across all participants.

Results

Results

Each exercise session contained fourteen participants. Figure 1 depicts the glucose profiles measure by continuous glucose monitoring (CGM) sensor for all patients receiving various treatments. Fig.2 depicts the total number of hyperglycemia and hypoglycemia events for 0-3 hours (during exercise and early recovery) and 3-12 hours after exercise.

Figure.1

Figure.2

Conclusions

Conclusion

In conclusion, we observe that during the exercise session and early recovery, all treatments function adequately; however, for the time interval of 3 to 12 hours after the experiment, mini dose glucagon treatment can reduce the number of both hypoglycemia and hyperglycemia events in T1Ds.

Background and Aims

As of today,the optimal basal to total insulin(BD/TD) has not yet been determined,and also there is no consensus how to determine basal insulin dose with using diabetes technology.The aim of this study is to determine the variability of insulin doses and basal/bolus insulin ratios according to insulin treatment modality and diabetes technologies from the Better Control in Pediatric and Adolescent Diabetes:Working to Create Centers of Reference(SWEET)registry.

Methods

The study cohort included the patients in SWEET database with Type 1 diabetes onset <18 years with at least one clinic visit between June 2010 and June 2021 and Type 1 diabetes for at least 2 years.

Results

In this study; 38,889 patients included.48.6% were female, median age was15.2(11.9;17.2)years, and median diabetes duration was5.9(3.7; 9.0)years.The distribution of treatment modality was as follows: multiple daily injection(MDI)without continuous glucose monitoring(CGM), 32.8%;MDI with CGM,18.0%; subcutaneous insulin infusion(CSII) without CGM,11.7%; and CSII with CGM37.3%.Data of the participants were analyzed for each treatment modality separately.In unadjested data, a significant association with BD/TDwas shown in all analyses,regardless of treatment modality:male gender,younger age group, and lower HbA1c were all related to decreased BD/TD(all p< 0.05).After adjustment for age, sex, and diabetes duration, there is no association remained between BD/TD and using diabetes technologies.

Conclusions

This study shows that similar basal to total insulin proportion is associated with using diabetes technologies,after adjustment for sex,age,and diabetes duration.On the other hand it remains to be investigated in a large prospective long-term study if reducing BD/TD insulin will improve metabolic control in children with type 1 diabetes.

Background and Aims

To assess the association of the key parameters of type 1 diabetes (T1D) control with treatment and monitoring modalities including the newly introduced hybrid closed-loop (HCL) algorithms in children with T1D (CwD) using the data from the national pediatric diabetes registry ČENDA.

Methods

CwD younger than 19 years with T1D duration >1 year were divided according to the treatment modality and type of CGM used: multiple daily injections (MDI), insulin pump without (CSII) and with HCL function, intermittently scanned continuous glucose monitoring (isCGM), real-time CGM (rtCGM) and intermittent or no CGM (CGM-). HbA1c, times in glycemic ranges and glycemia risk index (GRI) were compared between the groups.

Results

A total of 3251 CwD data (mean age 13.4 years) were analyzed. 2187 (67.3%) were treated with MDI, 1064 (32.7%) with insulin pump, 585/1064 (55%) with HCL. The HCL users achieved the best CGM-derived parameters results with median TIR 75.4% and GRI 29.1 (p<0.001 compared to the other groups), followed by MDI rtCGM and CSII groups with TIRs 68.8 and 69.0%, GRIs 38.8 and 40.1, respectively (non-significant p-values). These three groups did not differ in the HbA1c medians (51.8, 50.7, and 52.7 mmol/mol, respectively). The poorest T1D control was observed in CGM non-users (defined as CGM use <70% of the time) regardless of the treatment modality.

Conclusions

This population-based study shows that HCL technology is superior to other treatment modalities in a pediatric population.

Background and Aims

The prospective study was designed to determine the effect of Twin Precision Treatment Technology (TPT) on change in A1C and T2DM remission and regression of microalbuminuria. The TPT intervention uses the Whole-Body Digital Twin Platform, with AI and Internet of Things, to integrate multidimensional data to give precision nutrition and health recommendations via the TPT app and by coaches

Methods

We analysed the data for 6 months (intervention n= 206, control n= 71).Microalbuminuria was defined as urinary albumin excretion of 30-300 mg/day

Results

No. of patients with microalbuminuria (MIC) and macroalbuminuria reduced from 42 to 6 (-85.7%) and 4 to 1 (-44.4%), p<0.001, respectively. 151 patients who did not have microalbuminuria at baseline increased to 191 (26.5% increase). Mean duration of diabetes was 3.6 years (±2.6, 95% CI 3.3 to 4.04). Mean age was 43.3 years (±8.8, 95% CI 42.1 to 44.4). Based on ADA criteria, 83.4% (n=172/206) achieved diabetes remission vs 71.4% (n=10/14) in MIC group and 84.3% (n=161/191) without MIC; p=0.257 (ns). One patient had macroalbuminuria. There was significant change in HbA1c and cardiometabolic parameters (Table). The changes noted were: mean HbA1c % (9.2 to 6), HOMA2B % (54.04 to 81.26), HOMA2IR % (1.9 to 1.13), body weight kg (79.06 to 67.09), hsCRP mg/dL (3.21 to 2.53), SBP mmHg (130 to 118.5), DBP mmHg (87 to 79.3), ASCVD risk score % (10.46 to 4.29)

Conclusions

A significant number of patients achieved regression of microalbuminuria and improvement in metabolic markers. TPT is useful for mitigating the incipient renal complications attributed to T2DM.

Background and Aims

Composite metrics are potential screening tools for quality of glucose profiles. Q-Score has been constructed using main factors of the glucose profile, which are central glycemic tendency, hyperglycemia, hypoglycemia, intra- and inter-daily variability. Here, Q-Score was further developed for assessment of short-term glycemic control and identification of glucose profiles requiring therapeutic action.

Methods

CGM-profiles were from non-interventional, retrospective cross-sectional studies. The Q-Score-parameter‚ time above target range’ (TAR) was adjusted from 8.9 to 10 mmol/l using 3-day-sensor profiles (n=1,562). Profiles with 21 days of recording, obtained from 251 people with diabetes using the flash-glucose monitoring (FM) system, were applied to investigate time to Q-Score stability as well as correlation of Q-Score with fructosamin, Glucose management indicator % (GMI) and time in range % (TIR) as parameters of short-term metabolic control.

Results

The linear relation between the Q-Scores applying both TARs was Q-Score₁₀= -0.03 + 1.00 Q-Score_8.9 (r=0.997, p<0.001). Q-Score was stable after 11 days prior to TIR with 12 and variability given as % CV within 14 days. The Q-Score components reached stability between 12 and 13 days. Hypoglycemia was the slowest with 15 days. Q-Score had a correlation to fructosamin, GMI and TIR of r=0.715, 0.884 and -0.869, respectively. Q-Score indicated insufficient glycemic control in 218/251 profiles, mostly belonging to insulin-treated people with diabetes. Q-Score components responsible were variability plus hypoglycemia in type 1 and hyperglycemia in type 2 diabetes, respectively.

Conclusions

Q-Score is a potential metric for short-term glycemic control and can be used for personalized evaluation of glucose profiles by quantifying Q-Sore components.

Background and Aims

Rapidly improving diabetes device technology offers great promise for improving diabetes care. However, the patient data necessary for evidence based medical management exist outside the EMR and creates a massive burden for providers to access and evaluate. Meaningful care plans depend on assessing glycemic control over time and pulling this data involves multiple steps and inefficiencies. Various technologies have been deployed to address this problem. To date, however, these have been deployed mostly outside the EMR workflow, and only address the challenge acquiring data, with limited functionality to automatically import the data into the EMR and process clinical data into a meaningful treatment decision support.

Methods

Data was collected during routine clinical care at three sites of Yale Health and North East Medical Group (NEMG) endocrinology centers. Time it takes clinical staff members in treating patients with diabetes on an intensive insulin regimen was captured before and after integration of the technology into the Yale Health EMR system. Number of “clicks” in the process was also analyzed.

Results

Medical assistants and providers from three clinical sites were included in this study, before and after integration of the technology into the EMR. Data collected compared time it takes for the complete patient flow and time it takes to perform each step.

Results to be presented.

Conclusions

Results to be presented.

Background and Aims

The OneGlance® EMR and Smart Phone App is a unique platform customised for diabetes care. The app is synchronized with EMR and enables direct connectivity between patients and doctors. The patient can record and share their SMBG readings, receive timely treatment using the available chat option.

The study aims to evaluate the effectiveness of the app based connectivity in helping the patients to achieve a better individualized glycaemic targets, minimize hypoglycaemic episodes, with a better adherence to treatment regime.

Methods

A retrospective data analysis over five years, with selection criterion of 18 months treatment. The selected 731 patients were classified as App users (288) and non-app users (443) and then app users subclassified as insulin users and insulin non-users. The primary outcome was the glycaemic control measured by HbA1c, FBS, SMBG before and after using the app, while secondary outcomes included reported hypoglycemia events.

Results

In the study population of 288 app users (137 males, 151 females), they have mean age 53.1±13.1 years, baseline HbA1c 7.2±1.9% and FBS 139.6±57.7 mg/dl. Before and after using the app mean difference in their HbA1c was 0.44 (p<0.0001), FBS was 18.53mg/dl (p<0.0001) respectively, amongst them the insulin takers had a mean difference of 0.76 in HbA1c(p<0.0001). Average monthly incidence of reported hypoglycaemia reduced from 0.82 to 0.47 in the app users. Patients using the app had 3.3% overall better glycaemic profile on a like for like comparison to the non-users.

Conclusions

OneGlance® app is effective tool helping patients achieve better individualised glycaemic control with minimum hypoglycaemic episodes.

Background and Aims

We aim to develop a system enabling personalised weight prediction and providing data-based personalised nutrition advice for weight management.

Methods

We developed an interpretable weight prediction model based on the energy balance equation (energy intake is a function of recorded calorie consumption and a latent part dependent on observed weight changes, whereas energy expenditure is a function of weight and physical activity). We modelled the change in the caloric intake rate as random walk. Weight measurement errors and uncertainty of caloric intake were assumed to be independent and normally distributed. Parameters were estimated using Bayesian inference across weight, meal and activity data of 9500 users undergoing App-based nutrition and lifestyle coaching.

Results

The model proved useful in revealing preceding cause, i.e. caloric deficit or excess, from the observed body weight. This facilitates early detection of unwanted dietary habit shifts. The model also informs about main drivers of weight change by attributing it to quality and frequency of meals and activities. The causal design of the model allows estimating the effect of candidate dietary interventions on body weight. The predicted likelihood of achieving predefined weight targets (e.g., 73% accuracy predicting ≥5% weight loss at week 4 of 12) further supports tailored counselling.

Conclusions

The developed interpretable model empowering dietitians to individually tailored nutritional advice may facilitate the application in public health care, a sector otherwise cautious when adopting AI systems. Moving forward, we will further tailor the model output to fit in with clinical workflows, end users and therapeutic decisions.

Background and Aims

Determining user profile settings is an important part of insulin pump initiation and long-term use. We designed and evaluated an automated system to initialize and adjust insulin pump settings.

Methods

Adults with type 1 diabetes (N=29, mean age 35.4 (10.1) years, 62.1% female, diabetes duration 20.7 (11.2) years), completed 2 weeks of CGM run-in with multiple daily injections, then 13 weeks of Control-IQ technology use. Initial basal rate, carbohydrate ratio, and correction factor were determined by an algorithm based on prior insulin use, with follow-up settings adjusted weekly by the automated system. Providers could override the automated settings changes for safety concerns.

Results

Time 70-180 mg/dL (3.9-10 mmol/L) improved from 45.7% during run-in to 69.1% during the last 30 days of Control-IQ use. This improvement was evident after just one week (median improvement 18.8%, p<0.001, 95% CI 13.6 to 23.9). However time <70 mg/dL (<3.9 mmol/L) gradually decreased from 1.8% during run-in to 1.0% over 6 weeks and then stayed stable (p=0.03) (Table 1). Percentage of participants achieving HbA1c <7% (<53 mmol/mol) went from zero at baseline to 55% at study end (p<0.001). Only six of the 318 automated settings adaptations were manually overridden.

Conclusions

Automated therapy settings initialization and adaptation, implemented with Control-IQ technology, reduced hypoglycemia over time while showing immediate and sustained improvement in time in range. Use of this simplified technology may allow primary care and other providers, less comfortable with pump technology, to improve outcomes and reduce burden of care by increasing uptake of insulin pump use.