Incorporating the Effect of Behavioral States in Multi-Step Ahead Deep Learning Based Multivariate Predictors for Blood Glucose Forecasting in Type 1 Diabetes

Round 1

Reviewer 1 Report

The article is devoted to studying the influence of two behavioral states, physical activity, and stress, on fluctuations in blood glucose in people with type 1 diabetes. The study's relevance is justified by the fact that behavioral factors can affect blood glucose levels in people with type 1 diabetes, so their influence must be considered when controlling blood glucose levels in these people. Therefore, the authors propose two methods to quantify biomarkers associated with physical activity and stress using raw acceleration and electrodermal activity data collected with a wearable device. The authors assess the effects of physical activity and stress on blood glucose fluctuations by adding the resulting behavioral biomarkers to advanced deep learning-based glucose prediction models. Using an ablation study, the paper demonstrates that the inclusion of putative behavioral biomarkers improves the performance of the blood glucose prediction model in terms of relevant metrics such as mean absolute error, standard error, and coefficient of determination across all prediction horizons.

Despite the satisfactory quality of the article, some shortcomings need to be corrected.

1. Expanding the abstract with numerical results obtained within the research is recommended.
2. The aim of the paper should be defined.
3. The methods and approaches proposed by the authors in this study should be separated from known ones and ones already described in the previous research.
4. The neural network architecture selection should be justified.
5. The separation of the data to train and test samples should be described and grounded.
6. It needs to be clarified what the difference is between Figures 2 and 3.
7. The scenarios used for the experimental investigation should be described in more detail.
8. The authors have missed links to Tables 4 and 5.
9. The authors say they have compared obtained results with other research and present them in table 6. However, table 6 is missing.
10. The scientific and practical novelty of the research should be highlighted.
11. The formulas are parts of the sentences, so the correct punctuation should be used.
12. Some references to sources are missed because of technical problems with the references system used by the authors (e.g. lines 170, 171)

In summarizing my comments, I recommend that the manuscript is accepted after major revision. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Behavioral states are important in blood glucose (BG) management since they affect BG levels. In this work, the authors consider two behavioral states, PA and SS, on BG fluctuations. Their corresponding biomarkers were then quantified using ACC and EDA through a wearable device. The impact of PA and SS on BG was evaluated using a deep learning model, consisting of CNN and LSTM modules. The model is proved to be successful through several metrics (MAE, RMSE and R2). The work is well presented and is important in related study.

 

The only thing is that the authors should replace the default texts in Data Availability Statement with their own statement. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Thanks for the authors for considering reviewer's comments and recommendations. In my opinion, now the paper can be accepted.