Accepted for/Published in: JMIR Medical Informatics
Date Submitted: Apr 3, 2023
Date Accepted: Oct 12, 2023
Warning: This is an author submission that is not peer-reviewed or edited.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.
Machine learning models for blood glucose prediction in patients with diabetes mellitus: a systematic review and network Meta-analysis
ABSTRACT
Background:
Background:
Machine learning (ML) algorithms have provided more choices for diabetic patients to manage blood glucose (BG) more properly. However, because of abundant types of algorithms, choosing appropriate models is vital important.
Objective:
Objective:
Therefore, this study aimed to comprehensively assess the performance of ML algorithms to predict BG values with a systematic review and network Meta-analysis, and those to detect and predict adverse BG events were assessed by calculating pooled estimates of sensitivity and specificity.
Methods:
Methods:
PubMed, EMbase, Web of Science, and IEEE explore were systematically searched for studies on predicting BG values and predicting or detecting adverse BG events using ML algorithm from their inception to November 2022. Primary outcomes were relative ranks of ML algorithms for predicting BG values in different predict horizons (PH), and pooled estimates of sensitivity and specificity of algorithms for detecting or predicting adverse BG events. This study is registered with PROSPERO, registration ID: CRD42022375250.
Results:
Results:
In total, 46 eligible studies were included for Meta-analysis. For ML algorithms predicting BG values, the means of root mean absolute error (RMSE(SD)) in PH of 15 min, 30min, 45min, and 60min were 18.88 (19.71), 21.40 (12.56), 21.27 (5.17), and 30.01 (7.23) mg/dl, respectively. Neural network related model (NNM) showed relatively highest performance in different PH. Furthermore, the pooled estimates (95%CI) of positive likelihood ratio (PLR), and negative likelihood ratio (NLR) of ML algorithms were 8.3 (5.7-12.0), and 0.31 (0.22-0.44) for predicting hypoglycemia, and 2.4 (1.6-3.7) and 0.37 (0.29-0.46) for detecting hypoglycemia.
Conclusions:
Conclusions:
In summary, for predicting precise BG values, the RMSE increases with the rise of PH, and NNM showed relatively highest performance among all the algorithms. Meanwhile, current machine learning algorithms have sufficient ability to predict hypoglycemia events, while their ability for detecting hypoglycemia events needs to be enhanced. Clinical Trial: Trial Registration: The study protocol has been registered in the international prospective register of systematic reviews (PROSPERO; Registration ID: CRD42022375250).
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.