Machine Learning Approach for Care Improvement of Children and Youth with Type 1 Diabetes Treated with Hybrid Closed-Loop System

Round 1

Reviewer 1 Report

The article presents an extremely interesting topic and raises a broad problem affecting society. However, it needs some corrections and additions. 

·       In the introduction, the authors focus quite a lot on medical issues and closed-loop systems, while they write little about the use of machine learning methods in diabetology. Please elaborate on the paragraph on ML.

·       Please provide the total size of the input data set for ML, as it may be misleading to mention only 16 patients.

·       I suggest adding a flow chart in the preprocessing section to visualise the process better.

·       Precision and recall are not always indicators that speak directly to the accuracy of the classifiers. I suggest adding sensitivity and specificity for each classifier.

·       The authors mention that normalisation did not improve the accuracy of the classifier performance. What type of normalisation was used?

·       Please elaborate on 'prec' and 'rec' as these are skins.

·       Please explain in detail the coefficients b0 - b4 in Table 2.

·       Please include in the methodology what was the null hypothesis for MLR (this is only in the discussion).

·       In the discussion, the authors mention features extracted from the data. Which features are referred to? Please list them in the preprocessing section. Please also add information on how the feature extraction was changed.

Author Response

Please see the attachment. 

Author Response File: Author Response.pdf

Reviewer 2 Report

This is of great significance for the researchers to present their attempt to create a fully performative algorithm that could be embedded into a single sensor capable of automatically modifying and adjusting pump settings, based on the glycaemic values, and finally delivering the right insulin dose. Although this is a preliminary study, it can contribute experience for the future work and include more related features. However, there are still some details that need to be improved.

 

Details and Suggestions:

1.In the Abstract, the authors said that “The dataset is 7 composed of 90 days of recordings taken from 16 children and adolescents. ”, while in the Materials and Methods, why it become “91days”——(Data covering 3 112 months (91 days) for each patient was downloaded and used for the analyses.)？

 

2.How to decide and compute the sample size？

 

3.How to measure the intake of carbohydrate？

 

4.Whether the term “Insulin Sensitivity Factor” belongs to proper noun or not? If not, please don’t capitalize it.

 

5.In Results, when the evaluation indicators like “rec”, “prec” appeared for the first time, the full titles are preferred.

 

6.In Results, explanations for some indicators, such as “R²”, “F-stat”, can be put in the Materials and Methods section.

 

7.In Results, apart from the pictures and tables, some brief descriptions along with key data are advised to present.

 

8.Authors said that “LR didn’t provide good results”. In this case, why authors only choose RF instead of other algorithms? Did authors try other algorithms?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Overall, I am satisfied with the authors responses to my review.

In Results, the authors have added some descriptions, however, with few data presented.

The whole article still need to be checked carefully. Some details such as “The usage 98 of ML in diabetes care is huge and it varies from the predictions and analysis of glucose 99 levels to to the optimization of hypo and hyperglycemia situations [27], and for insulin 100 pumps improvements（line 98-101）”, double “to” appeared, need to be corrected.

 

Author Response

We thank the reviewer for the provided comment that have allowed us to improve the work.

All the mentioned errors have been corrected.