Cellular Immunity of SARS-CoV-2 in the Borriana COVID-19 Cohort: A Nested Case–Control Study

Round 1

Reviewer 1 Report

I reviewed the epidemiologia-2899273 manuscript. The most important disadvantages are the following:

1) The retrospective nature of the study and the small number of participants.

2) The selection of a relative large number of selected variables tested as confounders in the multiple logistic regression models, which lead to model overfitting. Thus, the findings may not easily be generalized.

3) The lack of novelty.

4) The fact that the researchers evaluated the presence of post-COVID sequelae by a single question made by a phone call, which is a potential source of bias  

Minor English editing is required.

Author Response

1. First reviewer:

I reviewed the epidemiologia-2899273 manuscript. The most important disadvantages are the following:

Thank you very much for your revision and comments on our manuscript.

1) The retrospective nature of the study and the small number of participants.

We valued your comment. However, we can argue about the prospective design of our study. This design is a nested control study in a prospective cohort (1). A measure of cellular immune response (CIR) of SARS-CoV-2 (outcome) was obtained in December 2022 after a follow-up of two years from 2020 to 2022, and the different explanatory variables (exposure) were measured during the follow-up. On the other hand, the complexity of CIR determinations made difficult the study of large samples, and the relative small size of our sample was indicated as a limitation in the discussion. Many studies of CIR have a small size, and the samples may be no representative of the general population. Another option is the study of a large sample size for SARS-CoV-2 humoral immunity with a small sample size for SARS-CoV-2 cellular immunity. We have reviewed the sample size of the cohort or cross-sectional studies with CIR determinations included in the references of our manuscript. In total, there are 24 references with a median sample size of 175 (percentiles 1%-99%: 43-2720). The sample size of our study is 225, higher than the median of reference’s studies. In the context of CIR research, our sample size could be considered sufficient to obtain valid results.     

  1.Kelsey JL, Whittemore AS, Evans AS, Douglas-Thompson, W. Methods in observational epidemiology. Second edition. Oxford University Press, Inc. New York 1996: pp 122-125

2) The selection of a relative large number of selected variables tested as confounders in the multiple logistic regression models, which lead to model overfitting. Thus, the findings may not easily be generalized.

We appreciate this indication. Application of the Directed Acyclic Graphics method permits that each explanatory variable has adjusted only for their potential confounders in the model. In this situation, the possibility of overfitting may be reduced for some the explanatory variables. In addition, the confounders in our models such as age, sex, blood groups, body mass index, chronic disease, smoker, alcohol consumption, physical exercise, SARS-CoV-2 vaccine doses, and time elapsed since the last vaccine dose / infection are well known, and comparisons with other studies could be more or less performed.    

3) The lack of novelty

We agree with the reviewer on the lack of novelty of our manuscript. However, some points would be addressed about their usefulness. The nested case-control design of the manuscript is an epidemiological approach to CIR with a random sample of a population-based cohort, including naïve participants and SARS-CoV-2 patients, in contrast with frequent studies, which use convenient samples of hospitalized SARS-CoV-2 patients, healthcare workers, and nursing home residents. The employed technique for cellular immunity determination presents a higher sensitivity compared with other techniques used in several studies. Our results are adjusted for potential confounding factors, a method which is not very usual in SARS-CoV-2 clinical microbiology studies, and 15 explanatory variables were employed. 

4) The fact that the researchers evaluated the presence of post-COVID sequelae by a single question made by a phone call, which is a potential source of bias.

Thank you very much for your indication. This limitation of our study, considering that information about the sequelae was obtained from participants themselves, was included in the discussion. However, the Borriana COVID-19 cohort comprised a follow-up from 2020 to 2022, and three surveys were carried out with this cohort. The single question about the presence of post-COVID-19 sequelae represents a summing up of several questions of post-COVID-19 sequelae in order to obtain a simplified measure. The cumulative incidence of of sequelae in our sample was 40.7% (95% Confidence Interval 33.7%-48.1%) (77/189), and it is in the intervals of the incidence of long COVID-19 symptomatology in no-hospitalized patients reported by Di Gennaro and co-authors in a systematic review and meta-analysis was 53.0% (95% Confidence Interval 38.5%-67.4%) (2).          

2. Di Gennaro F, Belati A, Tulone O, Diella L, Fiore Bavaro D, Bonica R, Genna V, Smith L, Trott M, Bruyere O, et al. Incidence of long COVID-19 in people with previous SARS-Cov2 infection: a systematic review and meta-analysis of 120,970 patients. Intern Emerg Med. 2023,18;1573-1581. doi: 10.1007/s11739-022-03164-w.

Reviewer 2 Report

I thank the authors for providing me with the opportunity to review their research. The study design focused on determining the cellular immune response (CIR) in a random sample of the Borriana COVID-19 cohort. The article is well done, however it has some shortcomings that can easily be overcome.

1.      In statistical section, the author reported that they use a multivariate logistic regression (line 168), but reading the results I suppose they use a multivariable logistic regression model to calculate the adjusted Odds Ratio (Hidalgo B, Goodman M. Multivariate or multivariable regression?. Am J Public Health. 2013;103(1):39-40. doi:10.2105/AJPH.2012.300897). The variable selection of this multivariable model was done with causal inference.

2.      In lines 162-165 the authors describe the detection of potential confounders and the use of DAGs but do not report it.

3.      The authors write in line 190-192:"The naïve group was older that the SARS CoV-2 infection patients (p=0.010), and the patients with sequelae group older than patient without sequela group (p=0.01)". The median test to compare this difference is reported in the statistical section, but table 1 shows the age of the patients with mean and standard deviation. Why don't use the T Student test?

4.      The asterisks marking meaning in Table 1 are not very clear. Specifically, they are not clear which comparison they are referring to. Furthermore, I consider it appropriate to report the p-values for each comparison.

5.      Table 2 has some formatting issues that make it unclear.

6.     Table 3 is not clear. You can divide with a line the comparison of each groups.

Author Response

2. Second Reviewer.

Thank the authors for providing me with the opportunity to review their research. The study design focused on determining the cellular immune response (CIR) in a random sample of the Borriana COVID-19 cohort. The article is well done, however it has some shortcomings that can easily be overcome.

We are grateful for your revision and comments on our manuscript.

1. In statistical section, the author reported that they use a multivariate logistic regression (line 168), but reading the results I suppose they use a multivariable logistic regression model to calculate the adjusted Odds Ratio (Hidalgo B, Goodman M. Multivariate or multivariable regression?. Am J Public Health. 2013;103(1):39-40. doi:10.2105/AJPH.2012.300897). The variable selection of this multivariable model was done with causal inference.

Thank you very much for this indication. There is a mistake in our text, and we are agreeing with the reviewer. This is multivariable and not multivariate. We change the word. The reference of Hidalgo and Goodman. Am J Public Health 2013;103:39-40 is very clarified.

2. In lines 162-165 the authors describe the detection of potential confounders and the use of DAGs but do not report it.

Thank you for indication. We have not drawn the graphic of DAGs to avoid a higher complexity of the manuscript, and we reported the potential confounding factors included in the DAGs in the Material and Methods section. In addition, each explanatory variable is adjusted for different confounding factors, which are reported at the bottom of several Tables.

3. The authors write in line 190-192:"The naïve group was older that the SARS CoV-2 infection patients (p=0.010), and the patients with sequelae group older than patient without sequela group (p=0.01)". The median test to compare this difference is reported in the statistical section, but table 1 shows the age of the patients with mean and standard deviation. Why don't use the T Student test?

We appreciate your question. We have preferred using no-parametric tests like Kruskal-Wallis and median tests. T Student test can be used with variables, which a normal distribution and it may be not the case.

4. The asterisks marking meaning in Table 1 are not very clear. Specifically, they are not clear which comparison they are referring to. Furthermore, I consider it appropriate to report the p-values for each comparison.

Thank you very much for your suggestion. We explain in the title of Table 1 the comparison between groups and the significant differences. Including all p-values in Table 1 appears very complicate and we prefer to maintain the same format.

5. Table 2 has some formatting issues that make it unclear.

Thank you very much for your indication. We change the format as you recommend.

6. Table 3 is not clear. You can divide with a line the comparison of each groups.

We appreciate your indication and we have made some change to get Table 3 more understand.

Reviewer 3 Report

The discussion session is too verbose. It's not necessary to list and compare all different COVID-19 studies in discussion. Not too many insights are provided by these discussions. 

There are too many grammar mistakes and spelling errors, which make it hard to understand this paper. The authors should pay more attention to the writing and proofread the manuscript before submission. 

Author Response

3. Third Reviewer.

Thank you very much for your revision and comments on our manuscript.

The discussion session is too verbose. It's not necessary to list and compare all different COVID-19 studies in discussion. Not too many insights are provided by these discussions. 

 We appreciate your indications and we are decreasing the extension of the discussion with a reduction of the information of different studies. On the other hand, we have tried to simplify the discussion to make more precise. With respect to the list of COVID-19 studies, we prefer to maintain all our references because are relevant to better understand the complexity and dynamic of SARS-CoV-2 cellular immune response. No limit of references’ number of manuscripts is a wise rule of the Epidemiologia journal.     

There are too many grammar mistakes and spelling errors, which make it hard to understand this paper. The authors should pay more attention to the writing and proofread the manuscript before submission.

Thank you very much for the comment. We have consulted with an expert in English language to improve the English level of the manuscript, and a completed revision of the manuscript has been made. 

Round 2

Reviewer 2 Report

Dear authors, the revised work appears to be clearer and more organic after the revision process. However, I would like to bring to your attention the answer you provided to question #3 in the previous review:

We appreciate your question. We have preferred using no-parametric tests like Kruskal-Wallis and median tests. T Student test can be used with variables, which a normal distribution and it may be not the case.

I agree with the authors that if the variable does not follow the Gaussian distribution it is necessary to carry out non-parametric tests, but in this case I believe it is more correct to report the median value and the IQR in the tables (for example variable Vit D9 in table 1) than the mean and the standard deviation.

Author Response

Thank you very much for your indication. In Table 1, we have changed the mean and deviation standard by the median and ranges of the quantitative variables.