Accepted for/Published in: Journal of Medical Internet Research
Date Submitted: Jun 13, 2022
Open Peer Review Period: Jun 13, 2022 - Aug 8, 2022
Date Accepted: Jan 21, 2023
(closed for review but you can still tweet)
Warning: This is an author submission that is not peer-reviewed or edited. Preprints - unless they show as "accepted" - should not be relied on to guide clinical practice or health-related behavior and should not be reported in news media as established information.
The effectiveness of wearable devices utilizing AI for blood glucose level forecasting/prediction: Systematic Review
ABSTRACT
Background:
In 2019 alone Diabetes Mellitus, a metabolic disorder primarily characterized by abnormally high Blood Glucose (BG) levels affected 463 million people globally, and over four million deaths were reported. The use of non-invasive technologies, such as Wearable Devices (WDs), to regulate and monitor BG in diabetics is a relatively new concept and yet in its infancy. Non-invasive WDs coupled with Machine Learning (ML) techniques have the potential to understand and conclude meaningful information from the gathered data and provide clinically meaningful advanced analytics for the purpose of forecasting and/or prediction.
Objective:
To provide a systematic review complete with quality assessment looking at diabetes effectiveness of utilizing Artificial Intelligence (AI) in WDs for forecasting/predicting BG levels
Methods:
We searched seven of the most popular bibliographic databases. Two reviewers' performed study selection and data extraction independently before cross checking of the extracted data was performed. A narrative approach was used to synthesize the data. Quality assessment was performed using an adapted version of QUADAS-2 tool.
Results:
From an initial 3,872 studies, we report the features from 12 studies after filtering according to our pre-defined inclusion criteria. The reference standard in all studies overall (n=11, 92%) was classified as low, as all ground truths were easily replicable. Since the data input to AI technology was highly standardized and there was no effect of flow and time frame on the final output, both the factors were categorized in a low-risk group (n=10,83%). We observed Classical ML approaches were deployed by half of the studies, the most popular being ensemble boosted trees (RF). The most common evaluation metrics used was CGE (n=7, 58%) followed by RMSE (n=5, 42%). We observed the wide usage of PPG and NIR sensors on wrist worn devices.
Conclusions:
This review has provided the most extensive work to date summarizing WDs that utilize ML for diabetics related BG level forecasting/prediction. Although current studies are few, our study suggests quality and measure of performance of ground truth techniques against invasive devices such as CGM was high. Further validation is needed for commercially available devices, we envisage that WDs in general have the potential to remove the need for invasive devices completely for glucose monitoring in the not-too-distant future.
Citation
Request queued. Please wait while the file is being generated. It may take some time.
Copyright
© The authors. All rights reserved. This is a privileged document currently under peer-review/community review (or an accepted/rejected manuscript). Authors have provided JMIR Publications with an exclusive license to publish this preprint on it's website for review and ahead-of-print citation purposes only. While the final peer-reviewed paper may be licensed under a cc-by license on publication, at this stage authors and publisher expressively prohibit redistribution of this draft paper other than for review purposes.