High Seroprevalence of SARS-CoV-2 among Healthcare Workers in a North Italy Hospital

Round 1

Reviewer 1 Report

This paper describes a cross-sectional study performed in Health Care Workers in Northern Italy to assess SARS-CoV-2 seroprevalence and identify associated determinants.

The researchers collected an interesting set of samples, the number of included HCW was high and the participation rate considerable amongst both permanent and temporary workers. Only a limited number of studies alike have been performed, so this study can be considered relevant and has a scientific merit. However, in my opinion data analyses performed are insufficient and interpretation of results and discussion can be improved. Given the considerable sample size combined with information on job titles and activities, there is much more information to gain from the data collected and the way it is currently analyzed/presented might be flawed and is certainly prone to misinterpretation. 

Data analyses and presentation

The way data has been analyzed and presented enhances the risk of misinterpretation. Many of the determinants taken into account, can be considered to be potentially related to each other (no data/description is provided on this so I can only make guesses). For example age, sex and job title; nurses probably are more often female and are typically younger compared to physicians. So I would have refrained from focusing on statistically testing of differences per factor and presenting it as such. I miss an actual descriptives table (typically Table 1) listing the most relevant variables and showing these per group (seropositives and seronegatives) without any testing. I would move table 3 to the supplements as the results are unadjusted. Table 4 presents the main results as determinants are mutually adjusted (multivariable logistic model), in this table I miss presentation of the OR before adjustment for age and sex. This would make it transparent if these are these indeed confounding the associations. In the methods section it is missing how confounders were selected. Given a multivariable model is presented, it would have been good to include in the supplements a table showing the associations observed in univariable logistic regression modelling. In addition, it would have been good to provide insight into the relation between all variables, eg how is the COVID-19 exposure risk distributed amongst the different job titles? This would have facilitated interpreting the results of the multivariable model because results now might be distorted as there might be collinearity issues. Interestingly, the one hospital ‘Vercelli’ strikes out as a risk factor. Considering this, it might be worthwhile to repeat the modelling stratified per hospital as I can imagine this can affect the outcomes and yield interesting insights, eg direct contact with patients might be a risk factor in the one hospital and not in the other if for example COVID-19 protective measures are better in the one hospital than the other. I am lacking sufficient insight in the data, but I can imagine it might be worthwhile to consider implementing an interaction effect, eg the effect of having direct contact with patients might be different when COVID-19 exposure risk is assessed low versus high.

So by showing univariable analyses, relations between variables, stratified analyses and potentially implementing interaction terms, better insight can be gained into what are the actual risk factors. I think that is highly interesting and highly relevant as this would lead to better understanding of what is underlying this increased SARS-CoV-2 risk observed for HCW. This would enable discussions on potential implications and consequences.

Besides the just mentioned matters, the discussion section would also strengthen from more in-depth considerations of findings from other studies. In comparisons to other studies on seroprevalence in HCW, what were characteristics of serology testing performed (sensitivity, specificity), study period (SARS-CoV-2 incidence, policy) and timing of the study (duration of detectable IgG levels?). And very importantly COVID-19 measures in place in the hospital, this should be properly described as well for this study as it is really important for putting results in context and providing suggestions on effectiveness of interventions. There have been studies performed to assess SARS-CoV-2 transmission in hospitals by means of sampling of surfaces, air etc; what has been identified and how does this related to the findings of this study? Currently the conclusions statement is quite generic which I think is a missed opportunity. Given the considerable sample size combined with information on job titles and activities, there is much more information to gain from the data collected. Then with this high seroprevalence observed amongst these Italian HCW and to be attained understanding of risk factors, evidence-based conclusions can be made.

Minor comments and questions:

The risk of exposure per HCW is determined by means of job administration registries, to what extent are these accurate and up-to-date? Has there been a validation by means of questionnaires distributed amongst HCW to verify their actual job task and assigned location/ward?

Table 2 can be presented more clearly. I suggest it merely presents the results of the subset of HCW that underwent NP swab testing. Presenting it in regards to the 2250 HCW is in that respect confusing as not all of them were tested.

What is meant with veterinarians mentioned? What is their role in a human hospital?

Informed consent statement: waiving of patient consent because data was used anonymously is a questionable argument I would say.  

Overall, a language check would be beneficial especially considering terminology and structure of sentences.

Author Response

R1: This paper describes a cross-sectional study performed in Health Care Workers in Northern Italy to assess SARS-CoV-2 seroprevalence and identify associated determinants.The researchers collected an interesting set of samples, the number of included HCW was high and the participation rate considerable amongst both permanent and temporary workers. Only a limited number of studies alike have been performed, so this study can be considered relevant and has a scientific merit. However, in my opinion data analyses performed are insufficient and interpretation of results and discussion can be improved. Given the considerable sample size combined with information on job titles and activities, there is much more information to gain from the data collected and the way it is currently analyzed/presented might be flawed and is certainly prone to misinterpretation. 

A: We thank revisor for his consideration, for highlighting the strengths of our study, including the large sample size of our study. We thank the reviewer for this criticism and some concerns have been raised and discussed below.

Data analyses and presentation.

R1: The way data has been analyzed and presented enhances the risk of misinterpretation. Many of the determinants taken into account, can be considered to be potentially related to each other (no data/description is provided on this so I can only make guesses). For example age, sex and job title; nurses probably are more often female and are typically younger compared to physicians. So I would have refrained from focusing on statistically testing of differences per factor and presenting it as such. I miss an actual descriptives table (typically Table 1) listing the most relevant variables and showing these per group (seropositives and seronegatives) without any testing. I would move table 3 to the supplements as the results are unadjusted. Table 4 presents the main results as determinants are mutually adjusted (multivariable logistic model), in this table I miss presentation of the OR before adjustment for age and sex. This would make it transparent if these are these indeed confounding the associations. In the methods section it is missing how confounders were selected. Given a multivariable model is presented, it would have been good to include in the supplements a table showing the associations observed in univariable logistic regression modelling.

A: We thank the reviewer for this criticism previous version of our manuscript showed a high risk of misinterpretation; for this reason, we decided to perform new statistical analyses

In response to Reviewer’s request, in table 1 we reported data on job title of the participants of, distinguishing them in seropositive and seronegative. However, we decided also to include information on seroprelavence (including 95 % IC) because they are useful to comment and interpret the data highlining the categories with higher seroprevalence. In the main text, we also decided to delete the p-value information and we kept only the description of difference in seroprevalence among groups with a detailed focus on specific unit wards.

As suggested by the reviewer, many of the determinants taken into account can be potentially related to each other. For this reason, firstly, we aggregated some categories with few subjects to avoid problem of sparse data. Secondly, to evaluate presence of multicollinearity, we used the LASSO model and we removed the regression coefficients that may be co-dependent. Then, we used the remaining covariates for the univariate and multivariate models. To be more specific, in the multivariate model we included covariates associated with seroprevalence (p-value <0.10) and then we used a backward procedure in order to select and identify terms that have to be included in the final model. Age and sex were included independently from statistical significance. In table 3 to avoid incorrect interpretation of data, we prepared a column with the univariate ORs and a column with the results of multivariable model.

In addition, it would have been good to provide insight into the relation between all variables, eg how is the COVID-19 exposure risk distributed amongst the different job titles? This would have facilitated interpreting the results of the multivariable model because results now might be distorted as there might be collinearity issues. Interestingly, the one hospital ‘Vercelli’ strikes out as a risk factor. Considering this, it might be worthwhile to repeat the modelling stratified per hospital as I can imagine this can affect the outcomes and yield interesting insights, eg direct contact with patients might be a risk factor in the one hospital and not in the other if for example COVID-19 protective measures are better in the one hospital than the other. I am lacking sufficient insight in the data, but I can imagine it might be worthwhile to consider implementing an interaction effect, eg the effect of having direct contact with patients might be different when COVID-19 exposure risk is assessed low versus high. So by showing univariable analyses, relations between variables, stratified analyses and potentially implementing interaction terms, better insight can be gained into what are the actual risk factors. I think that is highly interesting and highly relevant as this would lead to better understanding of what is underlying this increased SARS-CoV-2 risk observed for HCW. This would enable discussions on potential implications and consequences.

A: According to the reviewer’s comment, we implemented two multivariate models separately for the two hospitals involved in the analysis. However, we did not perform analysis using interaction terms as convergence problem were reported.

R1: Besides the just mentioned matters, the discussion section would also strengthen from more in-depth considerations of findings from other studies. In comparisons to other studies on seroprevalence in HCW, what were characteristics of serology testing performed (sensitivity, specificity), study period (SARS-CoV-2 incidence, policy) and timing of the study (duration of detectable IgG levels?). And very importantly COVID-19 measures in place in the hospital, this should be properly described as well for this study as it is really important for putting results in context and providing suggestions on effectiveness of interventions. There have been studies performed to assess SARS-CoV-2 transmission in hospitals by means of sampling of surfaces, air etc; what has been identified and how does this related to the findings of this study? Currently the conclusions statement is quite generic which I think is a missed opportunity. Given the considerable sample size combined with information on job titles and activities, there is much more information to gain from the data collected. Then with this high seroprevalence observed amongst these Italian HCW and to be attained understanding of risk factors, evidence-based conclusions can be made.

A: We improved the discussion as suggested by the reviewer with more in depth considerations. We compared our study with others present in literature.  

Minor comments and questions:

R1: The risk of exposure per HCW is determined by means of job administration registries, to what extent are these accurate and up-to-date? Has there been a validation by means of questionnaires distributed amongst HCW to verify their actual job task and assigned location/ward?

A: The job administration registries are accurate and up-to-date as they are used to payment of health workers. Moreover, the categorization was manually done by one physician and one hospital director, who was aware of the organization of the hospitals. No questionnaire was distributed among HCWs to verify their actual job task.  

R1: Table 2 can be presented more clearly. I suggest it merely presents the results of the subset of HCW that underwent NP swab testing. Presenting it in regards to the 2250 HCW is in that respect confusing as not all of them were tested.

A: we agree with the Reviewer 1.the table is confused, so we reported only e data about the subjects with previous NP swab. 

R1: What is meant with veterinarians mentioned? What is their role in a human hospital?

A: Veterinarians are professionals who treat diseases and disorders in non-human animals. In prevention department, a section of Local Health Unit,  they deal exclusively or primarily with animals raised for food. They have to advise on and monitor the handling, preparation, and storage of food of animal origin. We decided to include these 4 subjects in the physician category because we can consider them they medical executives. To avoid confusion, we removed the term in the table. 

R1: Informed consent statement: waiving of patient consent because data was used anonymously is a questionable argument I would say.  

A: Data was anonymous. Because of it, ethical approval code was not warranted; the research work does not use personal data. Our study was promoted by Italian Regional Council. Furthermore, on April 21, 2020 IRC decide to organize a seroprevalence screening among healthcare professionalsthe HWCs participation was on voluntary basis.

Reviewer 2 Report

Thank you for a great read. The paper is important and timely. It contains a large sample size and draws meaningful conclusions.

I am unsure of what the last sentence of the introduction mean "Finally, we decided also stratify seroprevalence by different risk of exposure classes and intensity of care services." It seems that there are a lot of assumptions that the readers are familiar with the survey and additional explanation is needed in the introduction.

Methods, were there any exclusion criteria for enrollment?

for stratified categories, I recommend included references to support the grouping.

Author Response

R2: Thank you for a great read. The paper is important and timely. It contains a large sample size and draws meaningful conclusions.

A: We thank the Reviewer for comments and considerations.

R2: I am unsure of what the last sentence of the introduction mean "Finally, we decided also stratify seroprevalence by different risk of exposure classes and intensity of care services." It seems that there are a lot of assumptions that the readers are familiar with the survey and additional explanation is needed in the introduction.

A: We completely agree with the review, so we changed the sentence that was very confusing. “Moreover, we assessed if seroprevalence was influenced by job title, Covid-19 exposure risk, contact with patients, unit wards and intensity of care.”

 Methods, were there any exclusion criteria for enrollment? for stratified categories, I recommend included references to support the grouping.

A: Italian Region Council decided to organize a seroprevalence screening among all the healthcare workers. The only exclusion criteria was the absence of patient’s consent. in the analysis but we included all subjects who worked in the two Vercelli LHS that decided to participated on voluntary basis. We add a sentence in materials and methods

Reviewer 3 Report

The topic, the seroprevalence of SARS-CoV-2 IgG anti-bodies among the personnel (HCWs and administrative) of a Local Health Service involved in the fight against the COVID-19 in a high prevalence zone of North of Italy, is very relevant in the current pandemic. While the paper is well written and structured, I have a few potential points for improvement:

Major:

1. The repetition of the word "seroprevalence" in the title seems somewhat redundant. You may consider to change it to "High Seroprevalence of SARS-CoV-2 among Health Care Workers in a Local Health Service of North Italy" or "A Seroprevalence Study of SARS-CoV-2 IgG anti-bodies among personnel working in a Local Health Service of North Italy" for a better definition of the topic of the study.

2. At the end of the introduction authors stated that the aim of the study was "to evaluate the seroprevalence of SARS-CoV-2 IgG anti-bodies among the personnel (HCWs and administrative) of Azienda Sanitaria Locale - Local Health Service (ASL) of Vercelli...". Moreover, as clearly pointed out in the materials and methods section, "the tested population included HCWs as well as technical and administrative staff" employed in "two main hospitals (Vercelli “Sant’Andrea” Hospital and Borgosesia “Santi Pietro e Paolo” Hospital), as well as Territorial Medical and Administrative services". Please better define your study population and settings also when you explain the aims of the study and the tested subjects in the abstract.

3. If possible, it would be interesting to know how many of the enrolled subjects were totally asymptomatic and how many presented, even if minimal, symptoms suggestive of SARS-Cov2 infection before the test.

Despite this, the discussion section the provides fairly detailed explanations for the high seroconversion recorded among HCWs an potential risk factors associated with SARS-CoV-2 diffusion and may suggest future strategies for infection control in occupational health practices.

Minor

1. I suggest to add an English term as explanation for Azienda Sanitaria Locale (ASL) (e.g. Local Health Service or Local Health-care Unit

2. A the end of the discussion "Finally, our results may be influenced by: inclusion criteria, e type of serological screening test performed, time elapsed from the first case in the specific region, infection
containment measures and prevalence in the general population" should be corrected in "Finally, our results may be influenced by: inclusion criteria, type of serological screening test performed, time elapsed from the first case in the specific region, infection containment measures and prevalence in the general population".

Author Response

R3: The topic, the seroprevalence of SARS-CoV-2 IgG anti-bodies among the personnel (HCWs and administrative) of a Local Health Service involved in the fight against the COVID-19 in a high prevalence zone of North of Italy, is very relevant in the current pandemic. While the paper is well written and structured, I have a few potential points for improvement:

A: We thank the Reviewer for the comments and considerations

Major:

R3-1. The repetition of the word "seroprevalence" in the title seems somewhat redundant. You may consider to change it to "High Seroprevalence of SARS-CoV-2 among Health Care Workers in a Local Health Service of North Italy" or "A Seroprevalence Study of SARS-CoV-2 IgG anti-bodies among personnel working in a Local Health Service of North Italy" for a better definition of the topic of the study.

A: As suggested by the Reviewer, we omitted the final part of the title to avoid repetition

R3-2. At the end of the introduction authors stated that the aim of the study was "to evaluate the seroprevalence of SARS-CoV-2 IgG anti-bodies among the personnel (HCWs and administrative) of Azienda Sanitaria Locale - Local Health Service (ASL) of Vercelli...". Moreover, as clearly pointed out in the materials and methods section, "the tested population included HCWs as well as technical and administrative staff" employed in "two main hospitals (Vercelli “Sant’Andrea” Hospital and Borgosesia “Santi Pietro e Paolo” Hospital), as well as Territorial Medical and Administrative services". Please better define your study population and settings also when you explain the aims of the study and the tested subjects in the abstract.

A: We better defined the study population and settings also in the abstract in order to better explain our study population. 

R3-3. If possible, it would be interesting to know how many of the enrolled subjects were totally asymptomatic and how many presented, even if minimal, symptoms suggestive of SARS-Cov2 infection before the test.

A: We did not record symptoms of subjects who tested positive for SARS COV 2.  For this reason, it seems not possible to know the prevalence of asymptomatic among subjects with COVID-19. We describe this limitation of our study in the discussion.  

Despite this, the discussion section the provides fairly detailed explanations for the high seroconversion recorded among HCWs an potential risk factors associated with SARS-CoV-2 diffusion and may suggest future strategies for infection control in occupational health practices.

A: We thank the Reviewers for highlighting the strengths of our proposal

Minor

R3-1. I suggest to add an English term as explanation for Azienda Sanitaria Locale (ASL) (e.g. Local Health Service or Local Health-care Unit

A: we translated the term in English

R3-2. A the end of the discussion "Finally, our results may be influenced by: inclusion criteria, e type of serological screening test performed, time elapsed from the first case in the specific region, infection
containment measures and prevalence in the general population" should be corrected in "Finally, our results may be influenced by: inclusion criteria, type of serological screening test performed, time elapsed from the first case in the specific region, infection containment measures and prevalence in the general population".

A: We decided to remove the phrase as considered a generally and not specific sentence.

Round 2

Reviewer 1 Report

My compliments to the authors for this considerably improved manuscript. Most of my concerns/comments have been covered.

Thank you for the revised table 3 showing the results of the univariable and multivariable modelling transparently. What I miss in this table is a mark (eg *) for pointing out the significant associations. As actually in the multivariable model, it can be seen that for the different variables, most of the categories have non-significant associations (OR of 1 in 95% CI). For job title it is solely healthcare assistant, for COVID-19 exposure risk solely medium category, for intensity of cares solely low intensity. This is now not that clear from the text in the results section as well. Eg it is described ‘In terms of job title, the estimates in multivariable models were weaker (closer to 1) than 195 the univariate ones;’ but associations are not only weaker but also non-significant.

A comment related to terminology: univariate – multivariate refers to the number of outcomes of the modelling, so 1 versus many. Univariable – multivariable refers to the number of variables (aka determinants) considered in the model, so 1 versus many. So in this paper there is solely one outcome, SARS-CoV-2 serological status (so by definition univariate modelling) and there have been univariable modelling and multivariable modelling performed. So eg table 3 heading should be ‘univariable and multivariable logistic models’, and outcome should be 'SARS-CoV-2 serological status (0/1)'. Please also check carefully, throughout the text, whether correct terms are used otherwise correct accordingly. 

I was wondering, why aren’t there estimates mentioned for the Borgosesia hospital multivariable model?

Some textual comments, lines 276-278 are duplicated at 283-285. Please carefully check the language and style, especially in the added/revised sentences throughout the manuscript.

Author Response

R1:My compliments to the authors for this considerably improved manuscript. Most of my concerns/comments have been covered.

A: We wanted to thank the reviewer for his suggestions; his observations were very useful in order to better describe our study. 

R1:Thank you for the revised table 3 showing the results of the univariable and multivariable modelling transparently. What I miss in this table is a mark (eg *) for pointing out the significant associations. As actually in the multivariable model, it can be seen that for the different variables, most of the categories have non-significant associations (OR of 1 in 95% CI). For job title it is solely healthcare assistant, for COVID-19 exposure risk solely medium category, for intensity of cares solely low intensity. This is now not that clear from the text in the results section as well. Eg it is described ‘In terms of job title, the estimates in multivariable models were weaker (closer to 1) than 195 the univariate ones;’ but associations are not only weaker but also non-significant.

A: As suggested by the reviewer, we add an * in table 3. in order to point out the association that were statistically significant (p<0.05). In the text we write that “In terms of job title, the estimates in multivariable models were weaker (closer to 1) than 195 the univariate ones”. We refered to p-value that we obtained using type three test. We found  statistically significant associations. To be more specific, healthcare assistants showed an iC of 1.03-3.62; for this reason, we considered the association weaker but statically significant.

R1: A comment related to terminology: univariate – multivariate refers to the number of outcomes of the modelling, so 1 versus many. Univariable – multivariable refers to the number of variables (aka determinants) considered in the model, so 1 versus many. So in this paper there is solely one outcome, SARS-CoV-2 serological status (so by definition univariate modelling) and there have been univariable modelling and multivariable modelling performed. So eg table 3 heading should be ‘univariable and multivariable logistic models’, and outcome should be 'SARS-CoV-2 serological status (0/1)'. Please also check carefully, throughout the text, whether correct terms are used otherwise correct accordingly.

A:We thanks the reviewer for his suggestions. We agree with him and so we decided to change the term univariate into univariable.

I was wondering, why aren’t there estimates mentioned for the Borgosesia hospital multivariable model?

A:We explained this aspect in “materials and method”; for this reason, we decided to “conduct different univariable models and each variable with p-value less than 0.10 was considered for the multivariable ones.” As no variable was statistically significant, multivariable model was not performed. We added a sentence in the section results.

R1:Some textual comments, lines 276-278 are duplicated at 283-285. Please carefully check the language and style, especially in the added/revised sentences throughout the manuscript.

A: We checked the language and style

Reviewer 3 Report

Thank you for doing the correction. Changes made clearly improve the text.

Author Response

Thank you!