A Digital Twins Machine Learning Model for Forecasting Disease Progression in Stroke Patients

Round 1

Reviewer 1 Report

Nice preliminary work on the topic. It needs further investigation but it shows the potential of digital twins in this field.

I suggest you better describe your data, and the preprocessing performed. For example, what do you mean precisely by canonical form at page 4?

I am curious whether you investigated the use of GANs to show that the resulting twin is indistinguishable from the real data.

All of your references are wrong, there is no correspondence between the numbers in the body of the article and the list o references at the end. Please double-check the references section.

The correct reference for \beta-VAE is missing and is:

Higgins, I., Matthey, L., Pal, A., Burgess, C., Glorot, X., Botvinick, M., Mohamed, S., Lerchner, A. 57202655646;57202449516;57202450281;57202449548;49861305800;6602977616;57209146118;57208442315; Β-VAE: Learning basic visual concepts with a constrained variational framework (2017) 5th International Conference on Learning Representations, ICLR 2017 - Conference Track Proceedings, .

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

I understand the importance of the suggested paper, but it doesn't seem to have a high level of completion on the whole. I consider some parts require revisions. 

 

1. First, related works were not fully researched. There have been a lot of related studies, but it is hard to find them in this paper. There is a need to survey related studies deeper to see how they have been conducted.
2. Due to a lack of related works, it is hard to understand the contribution of the suggestion. For improving this, you need to add related works and reorganize the contribution of the paper. 
3. About the method, many details are omitted. As an example, it is hard to see how beta-VAE was applied. 
4. It is needed to find how effective the suggested method is. The current paper makes it hard to figure out the strengths of the suggestion. 
5. In relation to the above, deep learning-based approaches are often applied recently. For instance, the below paper can be a good reference to improve this paper. 

Choi, Edward, et al. "Generating multi-label discrete patient records using generative adversarial networks." Machine learning for healthcare conference. PMLR, 2017.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The abstract fail to clarify what features are used for the machine learning process. Moreover, is not clear what the methods are and what data will be used to access the accuracy of the used algorithms. At a certain point in the abstract is mentioned simulated data and hinted it to be some sort of the assessing data set. Digital twins are mentioned later but for the sake of clarity the authors should do it earlier. So, the abstract needs to be clearer and “to the point”.

The selection of the covariates should be further explained using along with an explanation of the MIMIC database demographics relatively to the covariates. Do all patients have the same covariates?

The log transformation of the inputs should be justified qualitatively. Also, the authors should enunciate the drawbacks of such an operation. The statement “All values were reverted to canonical form before analysis” should be further explained.

The used 90-day data windowing should be justified besides the vague declaration “was selected to balance data availability with feature missingness”. Also, the authors should explain if the edge effects caused by data segmenting was addressed in any way.  Table 1 should also refer the covariates.

What is the real value of the several plots in Figure 5? The same for Figure 4. The authors only offer a description of the figure layout and parameters. There is a general lack of and analytic description of what is going on relatively to the results in the figures. The discussion chapter should address this competently.

Has the  MIMIC IV non-compliance data that needed to be addressed?

Author Response

Please see the attachment. 

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Thanks for your thorough revision of the manuscript.