Dataset Information

Machine learning-based glucose prediction with use of continuous glucose and physical activity monitoring data: The Maastricht Study.

ABSTRACT:

Background

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.

Methods

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).

Results

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%).

Conclusions

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.

SUBMITTER: van Doorn WPTM

PROVIDER: S-EPMC8224858 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:Type 2 diabetes (T2D) and prediabetes are classically defined by the level of fasting glucose or surrogates such as hemoglobin HbA1c. This classification does not take into account the heterogeneity in the pathophysiology of glucose dysregulation, the identification of which could inform targeted approaches to diabetes treatment and prevention and/or predict clinical outcomes. We performed gold-standard metabolic tests in a cohort of individuals with early glucose dysregulation and quantified four distinct metabolic subphenotypes known to contribute to glucose dysregulation and T2D: muscle insulin resistance, β-cell dysfunction, impaired incretin action, and hepatic insulin resistance. We revealed substantial inter-individual heterogeneity, with 34% of individuals exhibiting dominance or co-dominance in muscle and/or liver IR, and 40% exhibiting dominance or co-dominance in β-cell and/or incretin deficiency. Further, with a frequently-sampled oral glucose tolerance test (OGTT), we developed a novel machine learning framework to predict metabolic subphenotypes using features from the dynamic patterns of the glucose time-series ("shape of the glucose curve"). The glucose time-series features identified insulin resistance, β-cell deficiency, and incretin defect with auROCs of 95%, 89%, and 88%, respectively. These figures are superior to currently-used estimates. The prediction of muscle insulin resistance and β-cell deficiency were validated using an independent cohort. We then tested the ability of glucose curves generated by a continuous glucose monitor (CGM) worn during at-home OGTTs to predict insulin resistance and β-cell deficiency, yielding auROC of 88% and 84%, respectively. We thus demonstrate that the prediabetic state is characterized by metabolic heterogeneity, which can be defined by the shape of the glucose curve during standardized OGTT, performed in a clinical research unit or at-home setting using CGM. The use of at-home CGM to identify muscle insulin resistance and β-cell deficiency constitutes a practical and scalable method by which to risk stratify individuals with early glucose dysregulation and inform targeted treatment to prevent T2D.

Project description:BACKGROUND:The purpose of this study was to examine the relationships between glucose parameters obtained by continuous glucose monitoring and clinical outcomes in acute stroke patients. METHODS AND RESULTS:Consecutive patients with acute ischemic stroke or intracerebral hemorrhage within 24 hours after onset were included. A continuous glucose monitoring device (iPro2) was attached for the initial 72 hours after emergent admission. Eight glucose parameters were obtained from continuous glucose monitoring: maximum, minimum, mean, and SD of blood glucose levels, as well as area under the curve more than 8 mmol/L of blood glucose, distribution time more than 8 mmol/L of blood glucose, coefficient of variation (%CV), and presence of time less than 4 mmol/L over 72 hours. The primary outcome measure was death or dependency at 3 months (modified Rankin Scale score ?3). One hundred patients with acute ischemic stroke (n=58) or intracerebral hemorrhage (n=42) were included. Blood glucose levels varied between 5.2±1.4 and 11.4±3.2 mmol/L over 72 hours, with area under the curve more than 8 mmol/L of blood glucose of 0.7±1.4 min×mmol/L, distribution time more than 8 mmol/L of blood glucose of 31.7±32.7%, coefficient of variation of 15.5±5.4%, and presence of hypoglycemia in 20% of overall patients. Mean glucose level (adjusted odds ratio, 1.60, 95% confidence interval, 1.12-2.28/1 mmol/L), area under the curve more than 8 mmol/L of blood glucose (2.13, 1.12-4.02/1 min×mmol/L), and distribution time more than 8 mmol/L of blood glucose (1.25, 1.05-1.50/10%) were related to death or dependency for overall patients, as well as for acute ischemic stroke patients (2.05, 1.15-3.65; 2.38, 1.04-5.44; 1.85, 1.10-3.10, respectively). CONCLUSIONS:High mean glucose levels, distribution time more than 8 mmol/L of blood glucose, and areas under the curve more than 8 mmol/L of blood glucose during the initial 72 hours of acute stroke were associated with death or dependency at 3 months.