Meal estimation from Continuous Glucose Monitor data using Kalman filtering and hypothesis testing

Closed-loop insulin delivery systems, which integrate continuous glucose monitoring (CGM) and algorithms that continuously guide insulin dosing, have been shown to improve glycaemic control. The ability to predict future glucose values can further optimize such devices. In this study, we used machine learning to train models in predicting future glucose levels based on prior CGM and accelerometry data. We used data from The Maastricht Study, an observational population-based cohort that comprises individuals with normal glucose metabolism, prediabetes, or type 2 diabetes. We included individuals who underwent >48h of CGM (n = 851), most of whom (n = 540) simultaneously wore an accelerometer to assess physical activity. A random subset of individuals was used to train models in predicting glucose levels at 15- and 60-minute intervals based on either CGM data or both CGM and accelerometer data. In the remaining individuals, model performance was evaluated with root-mean-square error (RMSE), Spearman's correlation coefficient (rho) and surveillance error grid. For a proof-of-concept translation, CGM-based prediction models were optimized and validated with the use of data from individuals with type 1 diabetes (OhioT1DM Dataset, n = 6). Models trained with CGM data were able to accurately predict glucose values at 15 (RMSE: 0.19mmol/L; rho: 0.96) and 60 minutes (RMSE: 0.59mmol/L, rho: 0.72). Model performance was comparable in individuals with type 2 diabetes. Incorporation of accelerometer data only slightly improved prediction. The error grid results indicated that model predictions were clinically safe (15 min: >99%, 60 min >98%). Our prediction models translated well to individuals with type 1 diabetes, which is reflected by high accuracy (RMSEs for 15 and 60 minutes of 0.43 and 1.73 mmol/L, respectively) and clinical safety (15 min: >99%, 60 min: >91%). Machine learning-based models are able to accurately and safely predict glucose values at 15- and 60-minute intervals based on CGM data only. Future research should further optimize the models for implementation in closed-loop insulin delivery systems.

Dear Editor: We are thankful to Mitre et al.1 for addressing the important issue of how to best measure outcomes during closed-loop studies. Mitre et al.1 contrast four approaches: reference glucose, unmodified continuous glucose monitoring (CGM) data, stochastic transformation of CGM data, and their newly proposed CGM transformation to assess time in, time above, and time below target. We previously proposed the stochastic CGM transformation to correct for the bias introduced by the Navigator® CGM device (Abbott Diabetes Care, Alameda, CA) when using a single CGM device to direct closed-loop insulin delivery and simultaneously to assess outcomes.2 Like us, Mitre et al.1 found that the use of unmodified CGM data, in their case obtained using the Sof-Sensor® probe (Medtronic Diabetes, Northridge, CA), leads to an overestimation of time in target during closed-loop but not during conventional pump therapy. Additionally, they document an underestimation of time above target during closed-loop. Thus Mitre et al.1 concur with us in their generic findings but differ in the transformation needed to eliminate the bias. We are grateful to learn that the 15% measurement error we proposed for the Navigator is not applicable to the Sof-Sensor, which tends to underestimate glucose in hyperglycemia, and a simple mapping correction (4–8 mmol/L mapped to 4–7 mmol/L) is applicable. The corrections proposed by Mitre et al.1 and ourselves2 are relatively simple to implement and are advisable for home or transitional studies of closed-loop systems when reference glucose measurements are not available. We are aware that these relatively simple transformations do not fully reflect the complex relationship between CGM and underlying plasma glucose across the physiological glucose range. Furthermore, it is possible that the CGM–plasma glucose relationship (in mathematical terms the conditional probability of plasma glucose given CGM level) may be altered by closed-loop glucose control. Further research may be needed to clarify these issues and to increase confidence in outcomes derived from CGM levels. Finally, we acknowledge that according to Mitre et al.1 the treatment effect defined as the difference between time in target during closed loop and conventional therapy is unbiased when using unmodified Sof-Sensor data. This is reassuring in case researchers decide to analyze outcomes using unmodified Sof-Sensor data but is not generalizable to other sensors and may prohibit comparison with studies using a different sensor.

Disclosure: K.L. Flint: None. T. Ting: None. K. Rivera: None. P. Tamang: None. C.A. Colling: None. J.H. Li: None. M.S. Putman: Consulting Fee; Self; Synspira Therapeutics. Research Investigator; Self; Dexcom. Other; Self; Vertex Pharmaceuticals Incorporated. Introduction: Continuous glucose monitors (CGM) are FDA-approved for the management of diabetes and have been shown to improve glycemic control. For optimal utilization, CGM can be connected to cloud-based clinic portals for real-time sharing of glycemic data with clinicians to guide clinical care. Although studies have previously identified disparities in access to CGM for patients who receive Medicare, limited data are available examining disparities in CGM utilization and data sharing. Hypothesis: We hypothesized that racial and socioeconomic disparities in CGM access exist both in utilization and in real-time remote sharing of CGM data. Methods: This was a retrospective cohort study examining patients with type 2 diabetes on insulin using Medicare as primary or secondary insurance in a single diabetes clinic affiliated with a tertiary care medical center. Clinical data extracted from the electronic health record (EHR) included age, self-reported race and ethnicity, sex, preferred language, education level, ZIP code-based median household income, and enrollment in the EHR-patient portal. The most recent diabetes clinic note for each patient as of October 2023 was reviewed to assess whether each patient was using CGM at the time of the visit. The clinic’s CGM portal accounts were reviewed to assess whether the patient was connected to and actively sharing CGM data with the clinic. Two sample t-tests and chi-square tests were used to evaluate continuous and categorical predictors of CGM use and real-time remote sharing, respectively. Results: Of the 847 patients who qualified for inclusion, 420 (49.6%) were using CGM and 213 (25.1%) were sharing CGM data in real-time. Compared to patients not using CGM (n=427), patients using CGM were younger (70.8 years vs 73.5 years, p<0.001) and more often enrolled in the EHR-based patient portal (91.9% vs 84.1%, p<0.001); however, there were no differences in racial, ethnic, or socioeconomic factors between the groups (p>0.05 for all). Of the patients using CGM, remote sharing of data was associated with younger age (69.5 years vs 72.3 years, p<0.01), identifying as White (74.2% vs 65.0%, p=0.04), using English as preferred language (92.5% vs 84.2%, p=0.02), higher levels of education (p=0.01), and using the EHR-based patient portal (97.2% vs 86.7%, p<0.001). Conclusions: In this cohort of patients with type 2 diabetes on insulin receiving Medicare, there were no significant socioeconomic disparities in CGM utilization. However, racial and socioeconomic disparities were pronounced in real-time remote sharing of CGM data, suggesting that patients from minoritized racial backgrounds, patients who do not use English as their preferred language, and patients with less education may benefit from additional support and training to connect and share their CGM with their providers. Presentation: 6/2/2024

Glycemic variability contributes to oxidative stress, which has been linked to the pathogenesis of the long-term complications of diabetes. Currently, the best metric for assessing glycemic variability is mean amplitude of glycemic excursion (MAGE); however, MAGE is not in routine clinical use. A glycemic variability metric in routine clinical use could potentially be an important measure of overall glucose control and a predictor of diabetes complication risk not detected by glycosylated hemoglobin (A1C) levels. This study aimed to develop and evaluate new automated metrics of glycemic variability that could be routinely applied to continuous glucose monitoring (CGM) data to assess and enhance glucose control. Individual 24 h CGM tracings from our clinical diabetes research database were scored for MAGE and two additional metrics designed to compensate for aspects of variability not captured by MAGE: (1) number of daily glucose fluctuations >75 mg/dl that leave the normal range (70-175 mg/dl), or excursion frequency, and (2) total daily fluctuation, or distance traveled. These scores were used to train machine learning algorithms to recognize excessive variability based on physician ratings of daily CGM charts, producing a third metric of glycemic variability: perceived variability. Finger stick A1C (average) and serum 1,5-anhydroglucitol (postprandial) levels were used as clinical markers of overall glucose control for comparison. Mean amplitude of glycemic excursion, excursion frequency, and distance traveled did not adequately quantify the glycemic variability visualized by physicians who evaluated the daily CGM plots. A naive Bayes classifier was developed that characterizes CGM tracings based on physician interpretations of tracings. Preliminary results suggest that the number of excessively variable days, as determined by this naive Bayes classifier, may be an effective way to automatically assess glycemic variability of CGM data. This metric more closely reflects 90-day changes in serum 1,5-anhydroglucitol levels than does MAGE. We have developed a new automated metric to assess overall glycemic variability in people with diabetes using CGM, which could easily be incorporated into commercially available CGM software. Additional work to validate and refine this metric is underway. Future studies are planned to correlate the metric with both urinary 8-iso-prostaglandin F2 alpha excretion and serum 1,5-anhydroglucitol levels to see how well it identifies patients with high glycemic variability and increased markers of oxidative stress to assess risk for long-term complications of diabetes.

Analyzing continuous glucose monitoring (CGM) data is a mandatory step for multiple purposes spanning from reporting clinical trial outcomes to developing new algorithms for diabetes management. This task is repetitive, and scientists struggle in computing literature glucose control metrics and waste time in reproducing possibly complex plots and reports. For this reason, to provide the diabetes technology community a unified tool, here we present Automated Glucose dATa Analysis (AGATA), an automated glucose data analysis toolbox developed in MATLAB/Octave. Automated Glucose dATa Analysis is an open-source software program to visualize and preprocess CGM data, compute glucose control metrics, detect adverse events, evaluate the effectiveness of users' prediction algorithms, and compare study arms. Automated Glucose dATa Analysis can be used as a standalone computer application accessible through a dedicated graphical user interface, particularly suitable for clinicians, or by integrating its functionalities in user-defined MATLAB/Octave scripts, which fits the need of researchers and developers. To demonstrate its features, we used AGATA to analyze CGM data of two subjects extracted from a publicly available data set of individuals with type one diabetes. Finally, AGATA's features are compared against those of 12 noncommercial software programs for CGM data analysis. Using AGATA, we easily preprocessed, analyzed, and visualized CGM data in a handy way, in compliance with the requirements and the standards defined in the literature. Compared to the other considered software programs, AGATA offers more functionalities and capabilities. Automated Glucose dATa Analysis is easy to use and reduces the burden of CGM data analysis. It is freely available in GitHub at https://github.com/gcappon/agata.

Chapter 3 Analytical Methods for the Retrieval and Interpretation of Continuous Glucose Monitoring Data in Diabetes

Continuous glucose monitoring (CGM) data stored in data warehouses often include duplicated or time-shifted uploads from the same patient, compromising data quality and accuracy of resulting CGM metrics. We developed a processing algorithm to detect and resolve these errors. We validated the algorithm using two weeks of CGM data from 2038 patients with diabetes. Duplication errors were identified in 528 patients, with 25.7% showing significant differences in at least one metric (Time in Range, Coefficient of Variation, Glycemic Management Indicator, or Glycemic Episode counts) between raw and processed data. Eleven patients crossed clinically meaningful thresholds in one or more metrics after processing. Our results underscore the importance of real-world CGM data processing to maintain accurate and reliable CGM metrics for research and clinical care.

Algorithm-enabled patient prioritization and remote patient monitoring (RPM) have been used to improve clinical workflows at Stanford and have been associated with improved glucose time-in-range in newly diagnosed youth with type 1 diabetes (T1D). This novel algorithm-enabled care model currently integrates continuous glucose monitoring (CGM) data to prioritize patients for weekly reviews by the clinical diabetes team. The use of additional data may help clinical teams make more informed decisions around T1D management. Regular exercise and physical activity are essential to increasing cardiovascular fitness, increasing insulin sensitivity, and improving overall well-being of youth and adults with T1D. However, exercise can lead to fluctuations in glycemia during and after the activity. Future iterations of the care model will integrate physical activity metrics (e.g., heart rate and step count) and physical activity flags to help identify patients whose needs are not fully captured by CGM data. Our aim is to help healthcare professionals improve patient care with a better integration of CGM and physical activity data. We hypothesize that incorporating exercise data into the current CGM-based care model will produce specific, clinically relevant information such as identifying whether patients are meeting exercise guidelines. This work provides an overview of the essential steps of integrating exercise data into an RPM program and the most promising opportunities for the use of these data.

Access to and use of glycemic data are central to optimal management of diabetes. Use of continuous glucose monitoring (CGM) data to guide the management of diabetes has increased dramatically thanks to improved ease of use, accuracy, and availability. Retrospective CGM data collected throughout the day and night allow clinicians to visualize glycemic patterns, and single-page summary views like the Ambulatory Glucose Profile (AGP) Report make rapid interpretation both feasible and intuitive. A systematic approach that integrates retrospective CGM-generated data at clinic visits and other clinical interactions with personal use of CGM data can optimize glycemic management.

Introduction and Objective: Continuous glucose monitors (CGMs) are critical clinical tools whose use in research is rapidly expanding. We aim to understand the current state of knowledge around standard methods and tools for managing and analyzing CGM data. Methods: Literature review: We searched PubMed for articles that: 1) described a clinical trial, 2) were published in the last 5 years, 3) used CGMs to collect data, and 4) included a description of how CGM data was treated in the methods section. Surveys: We built an online survey to understand how researchers work with CGM data. A link to the survey was distributed via email and posted on social media platforms. Interviews: We recruited established diabetes researchers who either completed the survey, attended diabetes research meetings, or published recent articles about CGM data. Results: 89 out of 123 articles screened met inclusion criteria. Fewer than half of those described how their CGM data was extracted and cleaned, how missingness was handled, and which data tools were used. Our survey was completed by 43 researchers with a median of 8 years of experience with CGM research. Only 23 had ever used CGM-specific data tools; others reported using tools like Excel, R, and Python. Twenty-one content area experts from the United States and abroad who use CGM data in their research—averaging nine years of experience—were interviewed about the greatest benefits and challenges of using CGM data and their procedures for treating data. Twenty (over 95%) interviewees were not aware of standardized or universally accepted foundational texts containing information about CGM research methodologies. All 21 agreed that a repository of CGM data would improve diabetes research, increase connectivity, and advance innovation. Conclusion: Few if any standardized methods and tools are designed specifically to help researchers work with CGM data. It is vital to develop tools and resources to improve the quality of CGM-driven research by leveraging standardized practices and simplifying technical processes. Disclosure J.C. Espinoza: Consultant; Sanofi. Other Relationship; Big Health. Speaker's Bureau; Dexcom, Inc. E. Friedman: None. A. DuNova: None. C. Ho: None. A. Ayers: Consultant; Liom Health AG. E. Williams Jr.: None. L. Lett: None. S.N. Shah: None. D.C. Klonoff: Consultant; Synchneuro, Thirdwayv, Tingo, Afon, embecta, Glucotrack, Lifecare, Novo Nordisk, Samsung.

The mySentry system (Medtronic Inc.) is the first to amplify and relay continuous glucose monitoring (CGM) and insulin pump data to a remote site within the house. Its usability and acceptability were evaluated in families having a child with type 1 diabetes. Each enrolled family included a child (age 7-17 years) who used a Paradigm REAL-Time Revel sensor-augmented insulin pump (Medtronic). After a 1-week run-in phase, families set up and used the mySentry system for a 3-week study phase. Opinion surveys were completed by parents, and pump and CGM data were collected and analyzed retrospectively. No formal hypothesis testing was performed, and the study was not powered to detect changes in nocturnal glycemia. Thirty-five families completed the study. Enrolled children (61.1% female) had a mean (± standard deviation) age of 11.9 ± 2.70 years and a mean age at initiation of pump therapy of 7.1 ± 3.19 years. Baseline survey results indicated that most parents were fearful of their unawareness of their children's nocturnal glucose excursions. The mySentry system met the predefined acceptability criteria for general experience, product usability, and training materials. There were no unanticipated device-related adverse effects. Among children who experienced nocturnal hypo- or hyperglycemic episodes in both phases of the study, there was a trend toward less frequent and less prolonged episodes during mySentry use. The mySentry system met all predefined criteria for acceptability and did not demonstrate safety issues. Alerting parents to abnormal glucose values or trends may attenuate nocturnal hypoglycemia and hyperglycemia by prompting appropriate and timely intervention.

Going extended periods of time without eating increases risk for binge eating and is a primary target of leading interventions for binge-spectrum eating disorders (B-EDs). However, existing treatments for B-EDs yield insufficient improvements in regular eating and subsequently, binge eating. These unsatisfactory clinical outcomes may result from limitations in assessment and promotion of regular eating in therapy. Detecting the absence of eating using passive sensing may improve clinical outcomes by facilitating more accurate monitoring of eating behaviours and powering just-in-time adaptive interventions. We developed an algorithm for detecting meal consumption (and extended periods without eating) using continuous glucose monitor (CGM) data and machine learning. Adults with B-EDs (N=22) wore CGMs and reported eating episodes on self-monitoring surveys for 2weeks. Random forest models were run on CGM data to distinguish between eating and non-eating episodes. The optimal model distinguished eating and non-eating episodes with high accuracy (0.82), sensitivity (0.71), and specificity (0.94). These findings suggest that meal consumption and extended periods without eating can be detected from CGM data with high accuracy among individuals with B-EDs, which may improve clinical efforts to target dietary restriction and improve the field's understanding of its antecedents and consequences.

Diabetic peripheral neuropathy (DPN) is a prevalent chronic complication of diabetes which is linked to chronic hyperglycemia and glycemic variability. This study aimed to investigate the association between the glycemia risk index (GRI) and DPN in patients with type 2 diabetes mellitus (T2DM) using continuous glucose monitoring (CGM) data. From 2019 to 2023, 862 adults diagnosed with T2DM were enrolled at a tertiary care diabetes center in Ningbo, China. The medical history and laboratory parameters were recorded. Neurophysiological examinations were performed to evaluate DPN. The CGM data were recorded for 14 days, and the GRI was calculated based on these data. Multivariate logistic regression analyses were conducted to assess the odds ratio (OR) for DPN with an increased GRI. The prevalence of DPN in the ascending GRI quartiles was 41.6%, 47.9%, 49.1%, and 59.5%, respectively (P for trend < 0.001). In the multivariable logistic analysis, the highest GRI quartile exhibited a 63% greater risk of DPN (OR 1.631, 95% CI: 1.071 to 2.484, P = 0.023) than the lowest quartile after adjusted for age, sex, body mass index, diabetes duration, blood pressure, creatinine, urinary albumin-to-creatinine ratio, lipid profile and glycated hemoglobin. High GRI levels, as measured by CGM, were associated with a greater likelihood of DPN in T2DM patients.

CareLink® Therapy Management Software for Diabetes (Medtronic MiniMed, Inc., Northridge, CA) allows information stored in insulin pumps, continuous glucose monitoring (CGM) devices, and blood glucose meters to be uploaded and analyzed retrospectively. Patients typically use CareLink Personal to upload and review CGM and pump data; health care professionals can retrieve this data through CareLink Pro, which allows them to interpret reports and suggest therapy adjustments. Use of the CareLink system has been associated with improved glycemic control in children with type 1 diabetes on insulin pump therapy.1 CareLink Pro 3.0 includes two new reports: (1) Therapy Management Dashboard (Figure 1) and (2) Episode Summary. The Therapy Management Dashboard provides an overview of glycemic control, insulin delivery, and key pump settings. Pattern tables identify repeated episodes of hypoglycemia and hyperglycemia. Sensor use data include average sensor glucose values and the average number of high/low threshold and predictive alarms per day. The Episode Summary (not shown) presents 12 types of events that precede glycemic excursions. Figure 1 Therapy Management Dashboard Display. In a 2010 study funded by Medtronic, a survey was created to evaluate these new reports to verify that they meet acceptability criteria and to assess their value with respect to managing patients on sensor-augmented insulin pumps. Physicians received laptop computers with CareLink Pro 3.0, and were trained on its use. Data from patients’ insulin pumps (Medtronic Paradigm 522 or 722 models) were uploaded into the CareLink database; reports for patient data that included at least 5 days of sensor wear were then generated in CareLink Pro 3.0 and reviewed by the physicians. Physicians completed identical questionnaires after 30 and 120 days. Questions regarding acceptability were rated with a 7-point Likert Scale, and perceived value and benefits were evaluated by whether or not physicians agreed with certain statements. Responses to the Product Acceptability Questionnaire were favorable. The predefined criteria for product acceptability at 30 and 120 days were exceeded, with all investigators agreeing that the CareLink Pro 3.0 reports were acceptable, consistent with earlier versions of CareLink Pro, and consistent with therapy recommendations that they might consider in their own practices. Responses to the Perceived Product Benefits Questionnaire were also favorable. Eight of the 9 physicians agreed that the new reports would help them more easily identify the root causes of hypoglycemic and hyperglycemic events and more easily determine appropriate therapy adjustments. All physicians agreed that the reports would help them gain additional value from CGM data, and would recommend CareLink Pro 3.0 to their peers. Sensor-augmented insulin pumps play a central role in diabetes management by housing data on insulin delivery and glucose readings, and assisting in bolus estimation. Remote data storage and analysis capabilities, such as those provided by CareLink, provide additional opportunities for exploratory data analysis among individuals and large populations.2 Two new reports in CareLink Pro 3.0 were designed to efficiently summarize and present data from sensor-augmented insulin pump systems in order to facilitate patient interactions and inform management decisions. Questionnaire responses described here suggest that the Therapy Management Dashboard and the Episode Summary meet these goals.

Background: Previous InPen decision tree modeling (DTM) in adults indicated that missed meal dosing was a major modifiable factor associated with TIR (70-180mg/dL) . We used DTM to find characteristics/use-behaviors associated with TIR for InPen+CGM users &amp;lt;18 years of age. Methods: InPen+CGM data from youth (N=883) with T1D starting InPen between 7/20 and 6/21 were used to identify characteristics (prior delivery method, diabetes duration) and use-behaviors (frequency of missed/correction doses, priming/bolus count, meal therapy mode, InPen bolus calculator use, app reminders, time settings, and report generation) associated with TIR. A variance inflation factor &amp;gt;10 was used to reduce multicollinearity for feature importance. Those with insufficient CGM and/or bolus data and lack of consistent insulin therapy settings were excluded. Results: DTM of users determined 1) diabetes duration, 2) frequency of missed meal doses, 3) bolus count and 4) correction frequency as the top variables associated with TIR (Figure) . Highest average TIR (68.0%) was observed in newly diagnosed (≤1 year) users. Users with &amp;gt;2 years diabetes duration, &amp;lt;5.5 boluses/day, and &amp;gt;35.1% missed dose frequency demonstrated lowest average TIR (35%) . Conclusions: DTM is a novel application for understanding glycemic outcomes due to modifiable and non-modifiable factors. Disclosure M. Smith: Employee; Medtronic. G. Im: Employee; Medtronic. A. Gaetano: None. S. Thanasekaran: None. J. Macleod: Employee; Medtronic. R. A. Vigersky: Employee; Medtronic.