Reinforcement Learning for Multiple Daily Injection (MDI) Therapy in Type 1 Diabetes (T1D)
Round 1
Reviewer 1 Report
Dear authors, in this research, you are dealing with ML, which is a new approach important for modern pharmacotherapy management. Besides, you are still far from a real clinical setting and your results need to be verified for greater interest of clinicians. Your writing is clear and scholarly, the references are well chosen.
In the methodological part, you must add an explanation of why your choice was precisely those three scenarios.
I suggest you expand your research to include more parameters that could affect glucose levels, such as physical condition, activities, stress, etc.
Kind regards,
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The paper deals with important topics in biomedinformatics. The authors have reported the case of utilization of the Reinforcement Learning approach in MDI therapy for diabetes (T1D).
However, I have a number of suggestions:
1. Please provide optimize image with summarized Reinforcement Learning model data flow, including methods you describe in section 2.
2. Please provide a link to an open-access repository with the dataset you attached at the end of the paper, and add them into references, it'll be a far better way, to organize it.
Everything else, looks so far so good. Wish you good luck with your research.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Dear Authors,
Manuscipt is intesresting. However, the manuscript can be improved a lot. Here are some of my suggestions.
Minor comments:
1. Need a schematic diagram for the methodology used to build the RL model.
2. Parameters and features implementation could be summarized in the table.
3. Demographics such as age and ethnicity of the test population sample should be mentioned.
5. Block diagram of the closed-loop control for AP.
Major comments:
6. The model must be validated using more samples and different test cohorts.
7. The rationale that this is a better model must be provided by comparing the specificity and sensitivity etc.
Minor editing of the English language required
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
The study proposes a closed-loop insulin administration model with multiple daily injections (MDI) treatment using a reinforcement learning (RL) agent. The proposed model is proved to significantly reduce glucose variability and increases the percentage of time the glucose levels remain within the target range using simulation.
In general, the paper was clear in terms of the purpose of the study, the methodologies used in the study and the conclusions of the study.
A few minor questions:
- In the simulation setting, the meal was taken at 7am, 1pm and 7pm. Does the time taken the meal or the time interval between breakfast, lunch and dinner also impact the result? For example, what if the we shorten the time interval between breakfast and lunch to 4 hours (8am to 12pm) instead of the proposed 6 hours. Will it affect the simulation result?
- In the simulation result, the mean and standard deviation of bolus has been increased significant compared to open loop. Will this cause some safety concerns when applied to clinical trial?
Minor editing of English language required
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
Dear Authors,
All of my concerns are addressed. Wish you good luck.
Minor edits required.
