# Continuous Glucose Monitoring Prediction

**Authors:** Julia Ann Jose, Trae Waggoner, Sudarsan Manikandan  
**arXiv:** [2101.02557](https://arxiv.org/abs/2101.02557)  
**Date:** 2021-01-04  
**Summary page:** https://www.summarizepaper.com/en/paper/2101.02557v1/

## Key points

- Diabetes is a highly prevalent and dangerous disease, affecting nearly 10 percent of the global adult population.
- Advances in technology have provided new opportunities for effective treatment plans, such as continuous blood glucose monitoring (CGM).
- Accurate detection of meals is crucial for managing diabetes effectively, as meal intake can significantly impact glucose levels in the body and may lead to hyperglycemia.
- The study explores three different algorithmic approaches to detect meals using CGM data: Auto-regression based model, Kalman-Filter based approach, and LSTM-RNN based approach.
- The researchers synchronize CGM time series data with bolus ground truth information before developing and implementing the three algorithms.
- Train and test accuracies are reported, along with analysis of execution time.
- An initial algorithm for predicting meals from CGM data is proposed, providing valuable insights into future meal patterns.
- This research contributes to improving the understanding of continuous meal detection algorithms using CGM data for people with diabetes.
- By refining these algorithms and developing predictive models, healthcare professionals can enhance treatment plans tailored to individual needs, leading to better management of diabetes and improved patient outcomes.

## Summary

Diabetes is a highly prevalent and dangerous disease, affecting nearly 10 percent of the global adult population. Advances in technology have provided new opportunities for effective treatment plans, such as continuous blood glucose monitoring (CGM). This review focuses on the development and evaluation of three different meal detection algorithms using CGM data from patients with diabetes. Meal intake can significantly impact glucose levels in the body and may lead to hyperglycemia; therefore, accurate detection of meals is crucial for managing diabetes effectively. The study explores various algorithmic approaches to detect meals online: an Auto-regression based model, a Kalman-Filter based approach, and an LSTM-RNN based approach. The researchers synchronize the CGM time series data with bolus ground truth information before developing and implementing the three algorithms. They report their train and test accuracies as well as analyze execution time. Additionally, they propose an initial algorithm for predicting meals from CGM data that provides valuable insights into future meal patterns. This prediction algorithm allows individuals with diabetes to plan their treatment strategies more effectively. Overall, this research contributes to improving the understanding of continuous meal detection algorithms using CGM data for people with diabetes. By refining these algorithms and developing predictive models, healthcare professionals can enhance treatment plans tailored to individual needs which leads to better management of diabetes and improved patient outcomes.