The results suggested that very accurate short-term prediction can be achieved by only monitoring interstitial glucose data over a very short time period and by using a low sampling frequency.
A study investigating the minimum data variety, volume, and velocity required to create accurate person-centric short-term prediction models with the use of wearable electronics for type 1 diabetes patients found that accurate short-term prediction can be achieved by monitoring interstitial glucose data over a very short time period and sampling at a low frequency.
The study, published by Sensors, built an appropriate data set through passive patient monitoring in real-world conditions; a subset of type 1 diabetic subjects wore the flash glucose monitoring system. Also, the researchers comparatively and quantitatively evaluated the performance of the developed data-driven prediction models and the corresponding machine learning techniques, according to the study.
“Wearable devices broadly available on the market deliver multiple measurements per minute of the physiological condition and the daily activity of an individual,” explained the authors. “These data could potentially help to better predict the magnitude, probability, frequency, and duration of fluctuations in glucose levels. These predictions, when combined with reliable insulin pumps, can realize the vision of closed-loop control of blood glucose, also known as the artificial pancreas.”
The results suggested that very accurate short-term prediction can be achieved by only monitoring interstitial glucose data over a very short time period and by using a low sampling frequency. The models, that the researchers developed, predicted glucose levels within a 15-min horizon with an average error of 15.43 mg/dL by using only 24 historic values collected within a period of 6 hours, the results explained. Additionally, by increasing the sampling frequency to include 72 values, the average error reduced to 10.15 mg/dL.
“In this sense, we provided concrete evidence for wearable device manufacturers: (a) they do not need to use large memory units to increase the volume of the stored data and (b) they do not need to use high frequency for collecting data from the sensors. As such, the hardware requirements and the power consumption of the next-generation wearable devices can be reduced while delivering products of (a) smaller size, thus less weight, and (b) a longer lifespan, thus requiring less maintenance from the end users,” concluded the authors.
The study suggested that future research should monitor beyond 10 days and the data should be confirmed by examining the possibility of missing data. Also, the researchers hope that their results can eventually be used to involve patients with type 2 diabetes.
Rodriguez-Rodriguez, I, Chatzigiannakis I, et al. Utility of big data in predicting short-term blood glucose levels in type 1 diabetes mellitus through machine learning techniques [published online October 16, 2019]. Sensors. doi:10.3390/s19204482.