Exploring COVID-19 Daily Records of Diagnosed Cases and Fatalities Based on Simple Nonparametric Methods

Round 1

Reviewer 1 Report

The authors provide an important study of comparative epidemiological data analysis elucidating the the dynamics of the current COVID-19 pandemics. They study fatality to case ratios as a key indicator. It is an important indicator for which they find consistent cross-country estimates, however, the authors could also elaborate on their finding that/how fatality to case ratios seem to correlate with disease prevalence (and and potentially overburdened health systems?).

There are a few remarks that I would like to add

• try to use consistent notation (in the abstract 0.02 vs. 2.4%)
• page 3, lines 114-120. I cannot support scepticism with respect to modelling approaches as they make assumptions explicit. In this as in any study, assumptions are still there, however only made implicitly, e.g. the assumption of a constant delay between infection/diagnosis and death. In the same line, instead of refering to a "well known approximation for R0" on page 7, line 262 it would be more instructive to mention on which model assumptions the approximation is built. In this light I cannot fully agree with the overall attitude summarized in the conclusions (page 25, lines 717-723), but nonetheless appreciate the findings.
• page 4, 180ff - avoid complaining, clear defintions that follow are appreciated

 

 

Author Response

Reviewer 1

The authors provide an important study of comparative epidemiological data analysis elucidating the the dynamics of the current COVID-19 pandemics. They study fatality to case ratios as a key indicator. It is an important indicator for which they find consistent cross-country estimates, however, the authors could also elaborate on their finding that/how fatality to case ratios seem to correlate with disease prevalence (and and potentially overburdened health systems?).

Answer: This is a very interesting point. We checked the correlation between fatality to case ratio and the number of new cases. This yields a negative correlation for DE and IT. We added subsection 3.4 to discuss this remarkable aspect. Thanks for pointing us to this aspect.

 

There are a few remarks that I would like to add

•  

try to use consistent notation (in the abstract 0.02 vs. 2.4%)

Agree: We improved consistency by switching to proportions only instead of percentages

•  

page 3, lines 114-120. I cannot support scepticism with respect to modelling approaches as they make assumptions explicit. In this as in any study, assumptions are still there, however only made implicitly, e.g. the assumption of a constant delay between infection/diagnosis and death. In the same line, instead of refering to a "well known approximation for R0" on page 7, line 262 it would be more instructive to mention on which model assumptions the approximation is built. In this light I cannot fully agree with the overall attitude summarized in the conclusions (page 25, lines 717-723), but nonetheless appreciate the findings.

Answer: We agree that our scepticism has been formulated in a rather hostile way. In fact, we previously started with SIR-modelling, thence encountering with the problematic issue of estimating highly time-dependent parameters directly from fitting the model to data. In the long run we aim at an incorporation of the parameter estimates in form of a lookup table into SIR models hoping for a synergistic gain. We reformulated our apparently too sceptical attitude.

We added an explanation of how to derive the “well known approximation for R_0”. Indeed, one way to derive this approximation is given by a conditional analysis of an SIR model.

•  

page 4, 180ff - avoid complaining, clear defintions that follow are appreciated
Agree: We shortened and reformulated this part

Please confer the attached manuscript with tracked changes

Author Response File: Author Response.pdf

Reviewer 2 Report

This is a valuable paper as it uses nonparametric methods to estimate needed parameters that their modeling involve more complex methods otherwise. While I like the paper and the proposed methods, I have some concerns if addressed could substantially improve this work:

(1) The abstract sounds as a full paper. It needs restructuring and shortening highlighting the main findings. The same comment holds for the rest of the paper.

(2) I would like to see the syntax provided for the readers to use (minor).

(3) More information on the time series normalization is needed (minor). 

(4) It's essential to see some sub-analysis using standard methods to estimate the SEIR and SIR mechanistic models to obtain R0 and 
compare the findings with that in this paper. Otherwise, it will be very hard to judge the proposed calculations (major).

(5) While assuming the counts of cases to be Poisson variates, were there any examination about the relationship between the mean and variance?
An overdispersion should have triggered the use of Negative Binomial instead.

(6) Other studies have shown that COVID-19 behaves in a self-similar fashion geospatially in its spread trajectory or growth, could this provide 
any explanation for the convergence of the case-fatality ratios of the 8 countries? 

Author Response

This is a valuable paper as it uses nonparametric methods to estimate needed parameters that their modeling involve more complex methods otherwise. While I like the paper and the proposed methods, I have some concerns if addressed could substantially improve this work:

(1) The abstract sounds as a full paper. It needs restructuring and shortening highlighting the main findings. The same comment holds for the rest of the paper.

Answer: We shortened and restructured the abstract. We also removed some redundant parts in the main text as far as consistent with our style of occasionally referring to the broader context.

(2) I would like to see the syntax provided for the readers to use (minor).

Answer: Following the MDPI template, we used the section “abbreviations” at the end of the article. We added some more notions that may be unknown to the readers.

(3) More information on the time series normalization is needed (minor).

Answer: Population sizes which we used for per capita calculations are part of the data. We added corresponding explanations.

(4) It's essential to see some sub-analysis using standard methods to estimate the SEIR and SIR mechanistic models to obtain R0 and 
compare the findings with that in this paper. Otherwise, it will be very hard to judge the proposed calculations (major).

Answer: Thanks for pointing us to this important issue. The used approximation for R0 stems from an analysis of a simple SIR model and our estimates are, therefore, consistent with results obtained from fitting an SIR model to data at the outset of the epidemic where the reproduction number is assumed to be constant. We elaborated on R0 as well as on the effective (time-dependent) R(t) during the epidemic in the main text.

(5) While assuming the counts of cases to be Poisson variates, were there any examination about the relationship between the mean and variance?
An overdispersion should have triggered the use of Negative Binomial instead.

Answer: Thanks for pointing us to this interesting aspect. We added a discussion on this aspect in combination with the Reviewers concern (6). In fact, Negative Binomial has been used in a recent publication who picked up our delay correlation method and applied it to a geospatial analysis. As far as the approximation of the mean is addressed, we do not expect differences.

(6) Other studies have shown that COVID-19 behaves in a self-similar fashion geospatially in its spread trajectory or growth, could this provide 
any explanation for the convergence of the case-fatality ratios of the 8 countries? 

Answer: Extremely interesting point! Self-similarity is a remarkable aspect which we stressed by adding an extensive paragraph in our main text. We are very grateful for pointing us to this important characteristic of the Corona epidemic. Along these lines, we added some proper references. We agree, it appears to be evident that the similarity between the countries results from the characteristic of geospatial self-similarity.

 

Please confer the attached manuscript with tracked changes

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Thanks for addressing the comments. I still have some concerns about the edits as follows:

(1) Syntax is different than Abbreviations and hence it should be added into a supplemental file. I am having concerns that the authors are not willing to disclose the syntax for usability and reproducibility purposes.

(2) The references used in the newly added self-similarity paragraph are not related to COVID-19. The ones I have found online that are related to COVID-19 are as follows but more could be found:

https://arxiv.org/pdf/2003.14284.pdf
https://iopscience.iop.org/article/10.1088/1742-6596/1797/1/012010/pdf
https://www.mdpi.com/1660-4601/17/8/2750/pdf

 

Author Response

Answers to Reviewer 2:

Thanks for addressing the comments. I still have some concerns about the edits as follows:

(1) Syntax is different than Abbreviations and hence it should be added into a supplemental file. I am having concerns that the authors are not willing to disclose the syntax for usability and reproducibility purposes.

Answer: It was not clear to us what the reviewer means by “syntax”. We apologize for this mis-interpretation. We provided a link to the github repository where the reader can find the R source code used for our analysis. The link has been added to the data availability statement.

(2) The references used in the newly added self-similarity paragraph are not related to COVID-19. The ones I have found online that are related to COVID-19 are as follows but more could be found:

https://arxiv.org/pdf/2003.14284.pdf
https://iopscience.iop.org/article/10.1088/1742-6596/1797/1/012010/pdf
https://www.mdpi.com/1660-4601/17/8/2750/pdf

Answer: 3 references are directly related to COVID-19, 2 references provide evidence for this aspect in a more general way. However, we agree that there a publications which are highly relevant in this context. We apologize that we have not been aware of these publications. We extended the bibliography accordingly.

Please find attached the pdf of the revised manuscript.

Author Response File: Author Response.pdf