Bayesian Matrix Learning by Principle Eigenvector for Completing Missing Medical Data

Round 1

Reviewer 1 Report (New Reviewer)

This is an interesting paper which presents a way to fill in missing data in a matrix which links types of symptoms to treatments for people with Diabetes.

There is no "accuracy" measure in the paper, so the word should be omitted in the abstract.

Line 9: missing "is" after the word "learning".

line 127 there is an "in" to be deleted.

line 132 please delete "the only", duplicated.

line 136 please add a space after the full stop.

line 173: English to be corrected

lines 174 to 188: please put the numbers of the experiment in TABLE form (two tables or part A and part B)

Please complete the Captions of Figures 5 and 6. Add a good explanation of these two figures in the text as well. All captions must be improved, actually, explaining acronyms and giving basic information to understand the figure.

line 294: surprising !!!! "microelectronics" is not the concept nor the word to use here. Please rephrase the sentence to say what you mean, and please do not refer to wearable devices or diabetes monitors ...

The Discussion section is missing and the Conclusion does not fulfil the need of the reader to re-order ideas and get the paper's whole dimension. The paper is VERY interesting and deserves a good Discussion where the authors argue that with little sparse information, links between cause and effect can be reconstructed. A mention to Information Theory would be very interesting stating that Shannon´s theorem is respected but at the same time specific information is dug out of overall links. This is due to the eigenvectors! Please also state how the clinical members of the interdisciplinary authoring team plan to implement the article´s suggestion for medical care and what verification parametres will be used. Please invite the reader to a pleasant walk recalling your findings and putting them in plain English, boasting about your work, in a way.

 

Author Response

The replies are recorded in the uploaded file.

Author Response File: Author Response.docx

Reviewer 2 Report (New Reviewer)

The authors should consider the following comments to improve the quality of their research before it can be considered for acceptance. 

1. The novelty of the research is weak. The introduction can be extensively revised to show the contributions and novelty of the research.

2. There are a couple of gramartical errors which makes the readability weak. E.g. in the introduction, the authors stated thus, "Chen et al. purposed a Matrix Completion for Planning 33 Diabetes Treatment by using nonlinear convex optimization to generate information [9]", i think 'purposed' is meant to be 'proposed'. This example is just one out of many. The authors should thoroughly revise and improve the language of the entire manuscript. 

3. A comparison against other existing techniques to show the outperformance of the approach should be given in the result and analysis.

4. More recent researches can be reviewed and included in the research. 

Author Response

The replies are recorded in the uploaded file.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report (New Reviewer)

minimum points for publication of the article applsci-2093102 for the reasons given in my first review

Author Response

Modify the manuscript according to the Review Report Form. 

Author Response File: Author Response.docx

Reviewer 2 Report (New Reviewer)

The authors have address most of the concerns raised in the earlier review session. However, the language of the research still still be improved for better readability.

Author Response

The language of the research has been modified to improve the readability.

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

The contribution the paper gives to the literature is not clear to me. The algorithm the authors can be applied to the problem of matrix completion.
Medical applications are introduced in the title, and it can be ok. In the final concluding section the authors narrow the scope of the paper as  "This article establishes a DTBM to describe the relationship between diabetes symptoms and medication regimens".

What is the principle eigenvector? The notion of principal eigenvector is familiar to me but I doubt it can be a typo (the notion is mentioned in the title itself). To avoid confusion a definition or reference should be added.

The English is not easy to read. Senteces like "The principle eigenvector with the eigenvalue of the DTBM is 1, that is, the probability of different symptoms in diabetic patients and the probability of each medication regimen being used as the matrix stable state probability distributions" are obscure.

The literature on Bayesian matrix factorization is not adequately referenced in the introduction.

The content of section 6 is not clear. Is it a real data analysis? Details about the data source are not given. Nonetheless, why should the section named "Experiments"? If the aim is that of proving the accuracy and applicability of the completion algorithm, I think that a comparative discussion of alternative algorithms already known in the literature should be considered.

Reviewer 2 Report

Novelty missing

Reviewer 3 Report

GENERIC COMMENTS

 

Overall, the paper requires sizeable reworking to reach the required quality. While the motivation is evident, the practical value and significance is not clear, neither do the authors propose to position their contribution in relation to the state of the art. There is a need for a clearer theoretical framework and methodology, while experimental design needs to be linked to and informed by the latest literature. Similarly, outcomes of the experiments need to be clearly unpacked for their theoretical and practical significance. A discussion of results is missing and the conclusions appear to be generic. 

 

 

SPECIFIC COMMENTS

 

This paper discusses the use of Bayesian matrix eigenvector component for complementing missing data on diabetes treatment data. The work could clearly benefit from far higher clarity and depth, as there are a number of improvements to be sought. The abstract provides the problem domain and hints to some contribution, but a clear deliverable and objective should have been set out to begin with. This may even take the form of an enumerated list, to be revisited in a Discussion section in the end to explain how each objective is met. 

 

The background review is in significant need of more depth; there is literature (not excluding most recent work published in the last year) that needs to be discussed on both DTBM and methodological approaches to probabilistic definitions and prediction of missing data. 

There is also a need for a clearly outlined methodology; while a set of experiments is set out, there is little in explaining the experimental design, or discussing alternatives and ultimately inform the decision was to why the specific approach is suitable.

There is further a crucial need of outlining the results and their significance, as this is left open following each experiment. As such, the practical (and theoretical) importance of the work is not presented and a lot more needs to be articulated to achieve that.  

 

Some specific comments: in section 6, you need to clarify where the probabilities are derived from (I.e. do they come from data analysis, or are they part of the dataset definition?)

 

P187 reads: “In the statistics of medications for diabetes symptoms, it is assumed that there is no 190 mutual influence between medications” - can you please provide a reference to this?

 

Also figures and graphs are presented out of context and are not linked-to from the text. 

In general, there needs to be much more structure and context to the design and execution of the experiments. In addition more contextual relevance is required and the contribution to practical and theoretical sides of the problem need to be clearly provided.