Greatest Risk Factor for Death from COVID-19: Older Age, Chronic Disease Burden, or Place of Residence? Descriptive Analysis of Population-Level Canadian Data
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsI thank the authors for putting together the manuscript titled “At greatest risk from COVID-19: older age, chronic disease burden, place of residence? Descriptive analysis of population-level Canadian data”. While the paper tends to pass an important message, there are some concerns with the results and the writing style.
1. There are some concerns with the use of terms and symbols. COVID or Covid should be replaced with “COVID-19” throughout the manuscript.
2. There are places “+” sign was used to mean “and above” (P1. L17), while in other places, E.g. Table 5 it meant “positive”.
3. There are conjectures and imprecise results/data throughout the manuscript. For instance, in P1. L12, state the exact population of Ontario using the most recent sources. P5. L 145, & 159 and P6. L 182 and throughout the manuscript, stated the exact OR results.
4. P2. L45, “populace province” check grammatical error and the use of terms throughout the paper, especially when the test, virus, or disease is to be described. Check P2. L59, “SARS-C0V-2” is not figure “0”, it should be the letter “O”. And so many other places with grammatical errors.
5. In the text, use “80 years and older” instead of “>80”. Be consistent with “%” or “per cent” or “percent”.
6. The statistical methods section needs to be developed further. How were missing variables, univariate and multivariate outliers treated and other relevant assumptions of logistics regression, were they tested and met? Why were comorbidity categories 3 and 4 collapsing together for inferential analysis – any citation for the rationale?
7. Please mention all the statistical tools used, it seems Peason’s chi-square was also used.
8. Result section: check the tables for proper rounding. The total percentages for the residence distribution (Table 1) were 100.1%.
9. Similarly in Table 1, it is better to calculate the percentages of chronic disease burden across the residence as a proportion within each type of residence, than the current format which is a ratio of the entire population. This is because the sample of each residence type differs greatly.
10. I am not sure of the best choice of statistics reported from the SPSS binary logistics table. The output often contains the standardized regression coefficient (beta) which is interpreted as the odd ratio (OR). It appears the authors are interpreting something else based on the magnitude of the figures, perhaps “Wald”. If so, this is not a correct use of Wald estimates. This observation may impact the reported results significantly. Where 95% CI was used in a table, please create an extra column for p-value.
11. P5. L143-145, the authors reported that the chi-square of large sample sizes often turns significant. This is true and a very serious issue in terms of the validity of these results. It should be discussed further or highlighted in the limitations.
Additionally,
Line 27 – 30: The sentence is too long.
Line 37: The “covid” should be written in upper case.
Line 39: Delete “that.”
Line 57 – 60: The sentence needs to be rephrased.
Methods
Line 67: What is the current name of ICES? Why use the former name?
Lines 96 & 97: What is the rationale for having a positive PCR test before 30th September 2020? Was the PCR test not conducted after the date? Is it not possible to have a positive PCR test in October and still die the same month?
Lines 98: What informed the 60 days?
Lines 115 & 116: The sentence should not be here; it is meant for the result section. What kind of descriptive analysis was conducted? Mean and SD? OR frequency and percentage?
Lines 116 – 199: Was binary logistic regression the only inferential statistical analysis conducted? The authors should discuss the other inferential statistics.
Lines 127 & 128: The sentence seems not to be the focus of this paper. The crux of the paper is the number of COVID-19-caused deaths within 60 days of a positive PCR test, which is shown in Table 4.
Line 140: It appears difficult to understand the Chi-square analysis presented in Table 4. I guess the location was the independent variable. But what was the dependent variable? Is it age or death?
Was another analysis done using age as the independent variable? Did the authors combine the results of two analyses in one table? If yes, how did they do it? I cannot determine how the degree of freedom is 5 from the table.
What is the meaning of “test + n = 5073,” “test + n=3310,” and “38187 test + n =38187?”
Table 5: The authors need to report the model summary of the binary logistic regression.
Lines 144 & 145: The “… for long-term care home versus other residential setting (X2 = 6185.5641) …” information is not in the table. The table contains the Wald statistics and p-values of location as a whole without those of the location categories.
Lines 145 & 146: “next most significant variable” needs to be removed.
Discussion
Line 178 & 179: Truly, more people received formal care in their homes than in care facilities. However, the results of the analysis did not suggest they were older adults. The demographic table did not separate the age groups receiving care in any of the locations. It will be good for the authors to discuss within the scope of their results. Better still, they can remove those below 65 years from the analysis since the manuscript focuses more on older adults.
Lines 187 & 188: Which results did the authors discuss in this paragraph? The discussion appears to be anchored on the binary logistic regression, which has been discussed in the previous paragraph. It is recommended that the authors skilfully merge the information in both paragraphs for a good read.
Lines 191 & 192: What do the authors mean by contributory but not explanatory? Age and chronic disease burden also explained the COVID-19 death.
Line 217: Remove the “is” that is between “than” and “funding.”
Comments on the Quality of English LanguageThe quality of the English Language used is average.
Author Response
Comments and Suggestions for Authors
I thank the authors for putting together the manuscript titled “At greatest risk from COVID-19: older age, chronic disease burden, place of residence? Descriptive analysis of population-level Canadian data”. While the paper tends to pass an important message, there are some concerns with the results and the writing style.
There are some concerns with the use of terms and symbols. COVID or Covid should be replaced with “COVID-19” throughout the manuscript.
done
There are places “+” sign was used to mean “and above” (P1. L17), while in other places, E.g. Table 5 it meant “positive”.
Both replaced with words
There are conjectures and imprecise results/data throughout the manuscript. For instance, in P1. L12, state the exact population of Ontario using the most recent sources.
done
P5. L 145, & 159 and P6. L 182 and throughout the manuscript, stated the exact OR results.
done
P2. L45, “populace province” check grammatical error
Changed to Canada’s province with the largest population
and the use of terms throughout the paper, especially when the test, virus, or disease is to be described. Check P2. L59, “SARS-C0V-2” is not figure “0”, it should be the letter “O”.
corrected
And so many other places with grammatical errors.
Have tried to find and correct all
In the text, use “80 years and older” instead of “>80”.
done
Be consistent with “%” or “per cent” or “percent”.
Changed to % throughout
The statistical methods section needs to be developed further. How were missing variables, univariate and multivariate outliers treated and other relevant assumptions of logistics regression, were they tested and met?
Already in paper: Outliers were included in all analyses.
To the best of our knowledge, all independent variables were available for anyone who died of COVID-19, so there were no missing variables.
Why were comorbidity categories 3 and 4 collapsing together for inferential analysis – any citation for the rationale?
We have added the following to explain this: As the proportion of those with 4 or more chronic diseases was small we combined this category with the preceding one, using the categories of 0, 1, 2, and 3 or more for analyses.
Please mention all the statistical tools used, it seems Peason’s chi-square was also used.
Info added
Result section: check the tables for proper rounding. The total percentages for the residence distribution (Table 1) were 100.1%.
corrected
Similarly in Table 1, it is better to calculate the percentages of chronic disease burden across the residence as a proportion within each type of residence, than the current format which is a ratio of the entire population. This is because the sample of each residence type differs greatly.
This is what we have calculated
I am not sure of the best choice of statistics reported from the SPSS binary logistics table. The output often contains the standardized regression coefficient (beta) which is interpreted as the odd ratio (OR). It appears the authors are interpreting something else based on the magnitude of the figures, perhaps “Wald”. If so, this is not a correct use of Wald estimates. This observation may impact the reported results significantly. Where 95% CI was used in a table, please create an extra column for p-value.
We used SAS to do the logistic regression and have corrected this in the paper. The point estimate is the odds ratio in this setting. As stated before, all p values were highly significant – added to notes below table.
P5. L143-145, the authors reported that the chi-square of large sample sizes often turns significant. This is true and a very serious issue in terms of the validity of these results. It should be discussed further or highlighted in the limitations.
done
Additionally,
Line 27 – 30: The sentence is too long
fixed.
Line 37: The “covid” should be written in upper case.
fixed
Line 39: Delete “that.”
Oops! – thanks and deleted.
Line 57 – 60: The sentence needs to be rephrased.
We have rewritten the paragraph to make it more clear
Methods
Line 67: What is the current name of ICES? Why use the former name?
There is a long answer that probably isn’t relevant to readers – it is now just known as ICES. To avoid confusion we have changed this to read: ICES (the Institute for Clinical Evaluative Sciences)
Lines 96 & 97: What is the rationale for having a positive PCR test before 30th September 2020? Was the PCR test not conducted after the date? Is it not possible to have a positive PCR test in October and still die the same month?
We have added information in the methods and discussion (limitations) as follows:
“To account for a known lag time in PCR test recording in OLIS we included those dying after testing positive between 30 August 2020 and 30 September 2020. This, therefore, included those testing positive in September 2020 and dying between 30 and 60 days after testing. As LTC residents were routinely tested even if asymptomatic, while those living in the community could only access PCR testing if they had clinical signs of COVID-19, that is, those who were already sick, the timing just described might underestimate COVID-19 deaths among LTC residents relative to community dwellers who might be several weeks further along the trajectory of illness when tested.”
“Tracking deaths for less than 60 days after positive PCR testing in September 2020 might have increased the apparent deaths among those in LTC (see methods for more re this) relative to those living in the community. Correcting this, however, would have increased the magnitude of the disproportionate deaths in LTC.”
Lines 98: What informed the 60 days?
This was a government decision – anyone dying within 60 days of a positive PCR test was considered to have died of COVID-19.
Lines 115 & 116: The sentence should not be here; it is meant for the result section.
removed
What kind of descriptive analysis was conducted? Mean and SD? OR frequency and percentage?
Frequencies and percentages – added
Lines 116 – 199: Was binary logistic regression the only inferential statistical analysis conducted? The authors should discuss the other inferential statistics.
We have added info about Chi Square testing.
Lines 127 & 128: The sentence seems not to be the focus of this paper. The crux of the paper is the number of COVID-19-caused deaths within 60 days of a positive PCR test, which is shown in Table 4.
You are correct – the main outcome appears later in the results. However, we thought it necessary to describe the demographics of the population and of those testing positive prior to describing the main outcome. To make the structure of the results section more clear we have added subtitles that identify this sequence of reporting.
Line 140: It appears difficult to understand the Chi-square analysis presented in Table 4. I guess the location was the independent variable. But what was the dependent variable? Is it age or death?
It is death – as stated in the title of the table.
Was another analysis done using age as the independent variable? Did the authors combine the results of two analyses in one table? If yes, how did they do it? I cannot determine how the degree of freedom is 5 from the table.
Table 4 shows frequencies and percentages for each of the 9 categories – eg age <65 and living in LTC, age 65-74 and living in LTC, etc. We have tried to make this more clear.
What is the meaning of “test + n = 5073,” “test + n=3310,” and “38187 test + n =38187?”
Corrected – each n is the number of people in the location who tested positive for COVID-19 and as documented in Table 3
Table 5: The authors need to report the model summary of the binary logistic regression.
Have added the term binary. Beyond that - not sure what you mean here – do you mean that we should report that all variables were entered at once rather than in a stepwise fashion?
Lines 144 & 145: The “… for long-term care home versus other residential setting (X2 = 6185.5641) …” information is not in the table. The table contains the Wald statistics and p-values of location as a whole without those of the location categories.
These stats are in Table 5 in the upper portion of the table
The key findings reported in the table are the ORs and for these the 3 locations are reported separately
Lines 145 & 146: “next most significant variable” needs to be removed.
done
Discussion
Line 178 & 179: Truly, more people received formal care in their homes than in care facilities. However, the results of the analysis did not suggest they were older adults. The demographic table did not separate the age groups receiving care in any of the locations. It will be good for the authors to discuss within the scope of their results. Better still, they can remove those below 65 years from the analysis since the manuscript focuses more on older adults.
We no longer can access the raw data to do any re-analysis. The regression analysis adjusts for all other independent variables when looking at the effect of a particular one. Therefore, if we understand your concern correctly, differences in ages for those at each location will have been adjusted for when looking at the impact of location on death from COVID-19 because there were 3 age categories entered as independent variables (each >65 years in incremements).
Lines 187 & 188: Which results did the authors discuss in this paragraph? The discussion appears to be anchored on the binary logistic regression, which has been discussed in the previous paragraph. It is recommended that the authors skilfully merge the information in both paragraphs for a good read.
You are correct – paragraphs have been merged.
Lines 191 & 192: What do the authors mean by contributory but not explanatory? Age and chronic disease burden also explained the COVID-19 death.
Edited to read: . . . or why age and chronic disease burden were much weaker explanatory variables
Line 217: Remove the “is” that is between “than” and “funding.”
done
Reviewer 2 Report
Comments and Suggestions for AuthorsThank you for giving me the opportunity review this important paper which examined age, chronic disease load, sex, or place of residence as predictors of COVID-19 deaths in Canada.
Overall, the paper is written very nicely. I have the following comments before it can proceed further.
Introduction: Provide some findings from the systematic reviews related to your objectives in other countries.
Methods: Provide details for case selection from register. Inclusion and exclusion criteria. Were there any excluded cases?
Analysis: What assumptions were met for logistic regressions.
Results: Please provide Odds ratio for the results of logistic regression.
Discussion: This needs a bit work. Provide specific recommendations for each key finding.
Author Response
Comments and Suggestions for Authors
Thank you for giving me the opportunity review this important paper which examined age, chronic disease load, sex, or place of residence as predictors of COVID-19 deaths in Canada.
Overall, the paper is written very nicely. I have the following comments before it can proceed further.
Introduction: Provide some findings from the systematic reviews related to your objectives in other countries.
Details added
Methods: Provide details for case selection from register. Inclusion and exclusion criteria. Were there any excluded cases?
Those not captured by the OLIS data system as being PCR+ would have been excluded. We have added info re this in the methods (2.3). Other than that there were no exclusions, and this has been added in section 2.1.
Analysis: What assumptions were met for logistic regressions.
Added in methods (2.5)
Results: Please provide Odds ratio for the results of logistic regression.
done
Discussion: This needs a bit work. Provide specific recommendations for each key finding.
We have added to the discussion but are cautious about extrapolating too far beyond our findings. Our main policy recommendation remains the same – to put in place structures and resources that enable older adults to age in place rather than being warehoused (we don’t used this word in the paper as it seems a bit inflammatory for a medical article) in residential care facilities.
