Monitoring trends and differences in COVID-19 case-fatality rates using decomposition methods: Contributions of age structure and age-specific fatality.

The population-level case-fatality rate (CFR) associated with COVID-19 varies substantially, both across countries at any given time and within countries over time. We analyze the contribution of two key determinants of the variation in the observed CFR: the age-structure of diagnosed infection cases and age-specific case-fatality rates. We use data on diagnosed COVID-19 cases and death counts attributable to COVID-19 by age for China, Germany, Italy, South Korea, Spain, the United States, and New York City. We calculate the CFR for each population at the latest data point and also for Italy, Germany, Spain, and New York City over time. We use demographic decomposition to break the difference between CFRs into unique contributions arising from the age-structure of confirmed cases and the age-specific case-fatality. In late June 2020, CFRs varied from 2.2% in South Korea to 14.0% in Italy. The age-structure of detected cases often explains more than two-thirds of cross-country variation in the CFR. In Italy, the CFR increased from 4.2% to 14.0% between March 9 and June 30, 2020, and more than 90% of the change was due to increasing age-specific case-fatality rates. The importance of the age-structure of confirmed cases likely reflects several factors, including different testing regimes and differences in transmission trajectories; while increasing age-specific case-fatality rates in Italy could indicate other factors, such as the worsening health outcomes of those infected with COVID-19. Our findings lend support to recommendations for data to be disaggregated by age, and potentially other variables, to facilitate a better understanding of population-level differences in CFRs. They also show the need for well-designed seroprevalence studies to ascertain the extent to which differences in testing regimes drive differences in the age-structure of detected cases.

Introduction

The novel Coronavirus disease 2019 (COVID-19), caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), has been spreading rapidly across the world, and on March 11 2020 was recognized as a pandemic by the World Health Organization.

COVID-19 outbreaks went along with mostly regular patterns of logarithmic increase of the number of confirmed cases, with a few notable exceptions. The number of deaths associated with COVID-19, however, have evolved considerably less regularly, and case-fatality rates (CFRs) differ substantially between countries [1, 2].

Examples of this discrepancy are shown in Fig 1. As of June 30, 2020, Germany had a total of around 195 thousand confirmed infections and 9 thousand deaths, resulting in a CFR of around 4.6%. Italy, on the other hand, up to the same day, had 240 thousand confirmed cases of infection, around 34 thousand deaths, and a CFR of 14.0%. On April 13, Italy had roughly the same number of cases as Germany on April 28, and a CFR of 12.9%. Thus, the outbreak in Italy is going along with a much higher CFR, which has also increased over time [2, 3]. Also shown in Fig 1 are trends for Spain (until May 21) and New York City (until June 30), which fall somewhere between Germany and Italy.

Fig 1

COVID-19 confirmed cases and deaths, and implied case-fatality rates (CFR) in Italy (since March 9, 2020), Germany (since March 1, 2020), Spain (since March 21, 2020), and New York City (since March 22, 2020).

Differences in the CFR could indicate that the risk of dying of COVID-19 among detected cases differs between countries or changes within a population over time. On the other hand, it could also imply compositional differences in the detected infections [1, 3]. Specifically, the risk of dying of COVID-19 is well-documented to increase with age. Thus, if the population of confirmed infected individuals is older in one country or time period than in another, the CFR will be higher, even if the age-specific risk of dying is the same.

Indeed, demographers have argued that age structure matters, and the age composition of the reported cases has been suggested as a potential explanation for differences in CFRs [1-5]. So far, however, there have been no assessments of the importance of the age structure of diagnosed cases versus the age-specific CFR.

In this paper, we analyze cross-country differences in observed CFRs and within-country time trends in CFRs. We use recent data on China, Germany, Italy, South Korea, Spain, the United States, and New York City to study cross-country differences, and we provide results on within-country time trends in Italy, Germany, Spain, and New York City. We use a standard demographic decomposition technique to disentangle two potential drivers of differences and trends: (1) the age structure of confirmed cases and (2) age-specific case-fatality rates [6].

We interpret our findings in light of the unfolding knowledge about data-driven biases affecting CFRs. Counts of confirmed cases and deaths might not be comparable across countries because of differences in case and death definitions; differences in the underestimation of cases and in their age structure as a consequence of the country-specific testing regime; reporting delays of case counts and death counts; and differences in delays between symptoms and death [1, 2]. These data-related issues might lead to over- or under-estimation of CFRs throughout the epidemic, and more reliable estimates will only be available after its conclusion. Currently, adjusting CFRs for all of these potential biases is challenging and beyond the scope of this paper. Nevertheless, the method described in this paper is also readily applicable to adjusted estimates of CFRs once they become available.

Decomposition approaches like the one used in this paper are commonly used to explain the role of age structure on changing incidence rates [7]. They have also been applied to differences in cancer fatality rates across regions with varying age structures [8]. We are not aware of any application to CFRs of infectious diseases in general and the COVID-19 pandemic in particular.

To facilitate the application of the approach described in this paper, we provide code and reproducibility materials for the open source statistical software R in a freely-accessible repository on the Open Science Framework: https://osf.io/vdgwt/. Moreover, we also provide some examples in an Excel spreadsheet in the same repository.

Data and methods

Data

We gathered data on the cumulative number of diagnosed infections and deaths attributable to COVID-19 for the following populations (in alphabetical order): China, Germany, Italy, South Korea, Spain, the United States, and New York City. An overview of the data is given in Table 1. For Italy, we cover the cumulative course of the epidemic over 8 periods in the analysis, starting on March 9 and ending on June 30. For all other populations, we use the data for the end of June. Results over time for Germany, Spain, and New York City are provided in the S1 File. For China, the most recent available age-specific data is from February, and for Spain it is cumulative to May 21. To provide additional context, also the cumulative number of tests for COVID-19 performed is shown in Table 1 [9], as the number of detected cases will depend on the testing regime.

Table 1

Populations covered in the analysis, and their cumulative detected cases, deaths, and number of performed tests.

Country	Date(s)	Detected cases (cumulative)	Deaths (cumulative)	Tests performed (cumulative)
China	February 11 2020	44,672	1,017	-
Germany	June 30 2020	194,983	9,051	5,881,908
Italy	March 9 -June 30 2020	240,455	33,736	5,390,110
South Korea	June 30 2020	12,800	282	1,252,957
Spain	May 21 2020	234,824	28,628	2,221,497
United States	June 27 2020	2,504,175	119,016	30,446,284
New York City (US)	June 30 2020	212,072	18,492	-

The age-specific data for China does not account for the retrospective correction of the number of deaths. The cumulative cases and deaths shown for Italy in this table are for June 30. For Germany, the number of total tests performed is from June 28, and thus from a slightly earlier date than the numbers of cases and deaths. For South Korea, the number of individuals tested is shown; i.e., the number without counting multiple tests for the same person.

All data is provided by the respective health authorities, and is collected as part of the COVerAGE database project which gathers and standardizes age-specific data on the COVID-19 pandemic [10]. The COVerAGE database is continuously updated and freely available online, but we also provide snapshots of the data used for the calculations in this paper together with the code. A complete list of sources is provided in the documentation of the database project [10].

For some of the countries (Germany, Italy, Spain, United States, and New York City) age is not available for some confirmed cases or deaths. The COVerAGE database project imputed the missing age using the observed age distribution of cases or deaths, respectively [10]. Removing these cases from the analysis altogether would have no substantive impact on the results for Germany, Italy, and New York City, as the age is only missing for few cases and deaths (less than 0.5 percent in all three cases). For Spain, however, where around 28% of cases and 44% of deaths have no recorded age, ignoring cases and deaths of unknown age would deflate age-specific case-fatality rates. For the U.S., for roughly 15% of confirmed cases the age is unknown, but age recording is relatively complete for deaths. As the age distribution of cases and deaths with unknown age might differ from those for which the age is known the imputation approach we use could potentially bias the results for Spain and the U.S. Currently, there is no indication that this is actually the case; nevertheless, the results for Spain and the U.S. need to be interpreted with more caution.

The original data is provided in different age groupings. For the decomposition, the age groups have to match. The COVerAGE database provides adjusted counts so that all countries conform with the age groups of South Korea, for which the age groups are 10-year age groups from birth to 80+. Specifically, counts were split using a recently proposed method tailored for this data situation [10, 11], based on the assumption that the age distributions of case and death rates are smooth; i.e., that there are no discontinuities or abrupt changes in rates over age. The S1 File show the original age groups of the data.

Case-fatality rates

The COVID-19 case-fatality rate (CFR) is defined as the ratio of deaths (D) associated with COVID-19 divided by the number of detected COVID-19 cases (N): CFR = D / N. In our application, the death and case counts are cumulative counts up to a certain date.

If case counts and death counts are available by age, which is our situation, the CFR can also be written as a sum of age-specific CFRs weighted by the proportion of cases in a certain age group. We use a as an index to denote different age groups. These age groups could, for instance, be 0 to 9 years, 10 to 19 years, and so on, but other groupings are also possible. We define age-specific CFRs as ; i.e., the number of deaths in age group a divided by the number of cases in the same age group. The proportion of cases in age group a is given by . Using this notation, the CFR can be written as a weighted average of age-specific CFRs:

We use the weighted expression and a mathematical decomposition approach introduced by Kitagawa to separate the difference between two CFRs into two distinct parts, one attributable to age-structure of cases and another to age-specific case-fatality [6]. The method attributes the total difference into these two components, leaving no residual. In other words, if we use i and j to index two different populations, then the decomposition approach splits the difference between their CFRs into where the α-component captures the effect of the age structure of cases, and the δ-component indicates the part of the difference attributable to age-specific case-fatality. The details of the method are described in the S1 File, which also provides a step-by-step walk-through of the decomposition and its interpretation.

Results

Country comparisons

Table 2 shows results for cross-country comparisons using the data from South Korea (June 30) as a reference, with countries sorted by increasing CFR. We chose South Korea as the reference because its CFR is arguably the closest match to its actual infection rate due to extensive testing relative to the number of confirmed cases and an earlier onset of the epidemic; moreover, the CFR was comparably low, and decompositions will estimate what factor leads other countries to differ from this low CFR setting, making results easy to interpret. For all other countries, we also use June 30 or the closest date available to us, as shown in Table 1. In the S1 File, we provide additional results using Germany (low CFR) and Italy (high CFR) as reference countries.

Table 2

Results of the cross-country decompositions of case-fatality rates (CFRs) using South Korea as a reference case.

Country (1)	CFR (2)	Difference (3)	Age (α) component (4)	Fatality (δ) component (5)	Age (α) component, relative (6)	Fatality (δ) component, relative (7)
South Korea	0.022			(Reference)
China	0.023	-0.001	-0.002	0.001	66.3%	33.7%
Germany	0.046	-0.024	-0.018	-0.006	74.7%	25.3%
USA	0.048	-0.025	-0.011	-0.014	43.6%	56.4%
New York City	0.087	-0.065	-0.015	-0.050	23.4%	76.6%
Spain	0.122	-0.100	-0.070	-0.030	70.1%	29.9%
Italy	0.140	-0.118	-0.077	-0.041	65.3%	34.7%

The third column shows the difference between each country and South Korea, and is calculated as the CFR of South Korea minus the CFR of the respective country. Data for all countries is for June 30, except China (February 11), Spain (May 21), and the United States (June 27).

Based on the cumulative data up to June 30, South Korea had a CFR of 2.2% (first line of the table, column “CFR”). For all countries the difference to the South Korean CFR is shown in the third column of the table (South Korea minus the respective country). The fourth and fifth column of the table show the absolute contributions of the case age distribution and age-specific fatality components, respectively. A negative number for the age structure indicates an older age structure of detected cases compared to South Korea, while a negative number for the fatality component indicates higher age-specific case-fatality rates compared to South Korea. The sixth and seventh column of the table indicate the relative contributions of the components.

All countries and regions have a higher CFR than South Korea, as indicated by the negative difference shown in column four of Table 2, and some of the differences are substantial. For instance, the Italian CFR is almost seven times as high.

In many cases, the (relative) contributions of the α-component (age structure) seem to be larger than the δ-component (fatality), and the α-component is always negative. This means that the age structure of cases is an important factor in explaining why most countries we study fare worse than South Korea. For instance, in the two cases with the highest CFRs—Italy and Spain—the relative contributions were similar with the α-component explaining around two thirds of the difference (Italy: 65.3%; Spain: 70.1%), and the δ-component explaining the remainder. In Germany, the case age structure also is the main driver of the difference in CFRs relative to South Korea, and explains close to 75% of the difference. The US and New York City seem to be an exception, and the high CFR compared to South Korea seems to be mostly due to higher mortality of diagnosed individuals.

Trends over time

For Italy we have a relatively long time series of data spanning several months. Table 3 documents how the Italian CFR evolved from March 9 to June 30, with selected dates presented in between. Similar analyses for Germany, Spain, and New York City can be found in the S1 File, and we briefly comment on the results below. The CFR of March 9 is used as a reference, and the decomposition shows which factor is driving the trend in the CFR. From the beginning to the end of the period under study the CFR tripled, from 4.3% to 14.0%. This increase over time is largely driven by worsening fatality of COVID-19 –the fatality component explaining more than 90% of the change in almost all time periods—and changes in the age structure of cases only played a minor role, with detected cases moving to a more favorable (younger) age distribution and slightly counteracting the effect of worsening fatality. As a robustness check we changed the reference period from March 9 to March 21 (CFR: 8.1%). This again resulted in the fatality component explaining more than 90% of the change in CFR. The results for Spain and New York City in the S1 File show that for these populations the increases in the CFR were also mostly driven by worsening fatality, although to a lesser extent than in Italy. In contrast, in Germany the case age component almost explained 99% of the more than twofold increase in CFR between March 21 (CFR: 1.8%) and June 30 (CFR: 4.6%).

Table 3

Development of the Italian case-fatality rate (CFR) over time.

Date (1)	CFR (2)	Difference (3)	Age (α) component (4)	Fatality (δ) component (5)	Age (α) component, relative (6)	Fatality (δ) component, relative (7)
09 March 2020	0.043			(Reference)
23 March 2020	0.087	0.044	-0.005	0.048	8.55%	91.45%
2 April 2020	0.118	0.075	-0.005	0.081	6.33%	93.67%
16 April 2020	0.126	0.083	-0.003	0.085	2.91%	97.09%
7 May 2020	0.131	0.088	0.000	0.088	0.02%	99.98%
26 May 2020	0.137	0.094	-0.001	0.095	0.56%	99.44%
16 June 2020	0.139	0.097	-0.001	0.098	0.99%	99.01%
30 June 2020	0.140	0.098	-0.001	0.099	1.25%	98.75%

The third column gives the difference between the CFR of the respective date minus the CFR of March 9.

Discussion

Case-fatality rates (CFRs) associated with COVID-19 vary strongly across countries and over time within countries. Our findings show that there is substantial variation in which factor explains the differences in CFRs. Differences in the age distribution of detected infections in some cases explain a substantial part of the total difference in CFRs. In particular, in many cases more than 50% of the difference in CFRs between countries with a low CFR and a high CFR can be explained by the age structure of detected infections. In contrast, in Italy, we observe a substantial increase in the CFR over time, mostly attributable to increasing age-specific case-fatality.

Ultimately, the approach discussed here does not directly explain why the age structure of confirmed cases or the age-specific case-fatality rates matter more in one case and less in another, and some expertise about the contexts which are being compared is required to interpret results. We discuss some potential explanations below, including potential data-related issues and biases.

Differences in the age structure of the populations which are being compared are unlikely to be a major driver of the age component that we estimated here, as the age composition of confirmed cases does not necessarily match the age composition of the population. For instance, according to Eurostat, the proportion of the population aged 80+ in 2019 was 7% in Italy and 6.5% in Germany, while in our data the proportion of reported infections in the same age range was 25% for Italy and only 11% in Germany.

Differences in testing regimes are a plausible mechanism driving both the different age structures of detected cases, as well as different age-specific fatality rates to the extent that denominators are underestimated in distinct degrees [3, 12, 13]. Results not shown here indicate that early in the pandemic in March the difference in the CFRs of South Korea and Germany—two countries with extensive and early tracing and testing of contacts of known cases—was largely driven by differences in fatality. The low contribution of the case age distribution component to the CFR disparity between South Korea and Germany suggests that these countries might have been more successful at catching the mild and asymptomatic cases among the younger population groups. Since then, the CFR of Germany has increased and the age structure of confirmed cases has shifted to higher ages, and the age structure has become more important in explaining the gap between South Korea and Germany, making test numbers alone an unlikely explanation for the different age structure of detected cases.

Differences in the COVID-19 transmission pathways might also be a factor. Depending on contact patterns and household structure, the elderly population might be affected earlier in some countries than in others, leading to a less favorable age distribution of infections [4, 14]. This could be relevant in explaining why the age distribution plays such a large role for the two countries with by far the highest CFR, Spain and Italy, which have a relatively large proportion of individuals living with their elderly parents or grandparents, and comparatively intensive intergenerational contact [15-18].

Disparities in age-specific case-fatality rates across countries may result from differences in age-specific prevalence of comorbidities, which exacerbate the risk of death from COVID-19 considerably [1, 19] or differences in quality or saturation levels of the healthcare system, among other potential factors [20]. The trend over time in the Italian CFR is an example where changes in age-specific case-fatality rates are driving trends, instead of changes in the case age distribution. This likely reflects the worsening situation in Italy over time as its health care system got under increasing pressure [12, 21]. However, an increase in CFR could also be expected once containment measures become effective, and newly confirmed cases increase at a slower pace than deaths from cases acquired prior to containment policies.

Only once an epidemic reaches its final conclusion and all cases have either resulted in recovery or fatalities, can the importance of the age difference in cases on CFRs be assessed with an acceptable degree of accuracy [22]. In this context a distinction should be made between CFRs, which are solely based on detected cases, and infection fatality rates (IFRs), which estimate the risk of dying once infected, including both confirmed and undiagnosed cases. Ideally policies for containing the spread of a virus would be designed on the basis of IFRs. However, particularly early on in an epidemic, the CFR is the only metric available until the extent of known data-driven biases can be assessed [1, 12, 13, 23–26].

Data quality can affect both the age composition of detected cases and age-specific case-fatality rates. For instance, counts may be affected by issues like reporting delays or censoring, or by inconsistent case definitions [1, 2, 23, 24, 27]. In many countries, only deaths occurring in hospitals are being reported in a timely manner [28], underestimating the full death count which would include deaths at home and in institutions. Deaths may be underestimated because of a lack of testing both before and after death. Countries might also differ in how they code deaths from underlying or contributory causes [28]. Excess all-cause mortality compared to a seasonal all-cause mortality baseline are suggestive that there is currently considerable underreporting of COVID-19 deaths, even if some of these deaths might be related to delayed or avoided medical treatment from other causes of death [29].

The relative importance of both the case age structure and mortality components could also be affected by comparing countries at different stages of the epidemic. This could result from cases not being detected at the beginning of the epidemic, or from differences in the lag between infection and death [12, 26, 30]. Generally, CFRs are highest at the beginning of an infectious outbreak, when the most serious cases are the most readily detected, and declines as testing capacity increases and less serious cases are identified [26]. This has notably not been the case for the COVID-19 epidemic, where the CFR has generally been increasing. Likely this reflects the success of widespread containment measures enacted in response to increasing caseloads. Newly identified cases are increasing more slowly than deaths, despite increases in testing capacity.

The application of the method we present in this paper is not limited to decomposing the current estimates of CFRs. It can also be applied to CFR estimates which have been corrected for biases, and to IFRs. It can, in principle, also be applied to excess all-cause weekly mortality counts, although this is not without challenges; we provide more discussion and some exploratory results on decomposing differences in excess mortality in the S1 File. Thus, while the data currently available as input for the decomposition approach might be of varying quality, this is not a flaw of the method itself. As data quality improves over time and adjustment methods become available our approach will continue to provide insights into differences and trends in mortality associated with COVID-19.

Finally, the choice of age groups may have affected our results. If ages were grouped too widely it might hide actual age-specific case-fatality differences. For instance, if the median age within the 10-year aggregated age groups that we used differed between populations, this would reduce the case-age structure explanation and inflate the age-specific mortality explanation. Finally, there are alternative decomposition techniques that might yield different results. However, differences are expected to be rather small; indeed, applying the method of Horiuchi and colleagues [31] to our data yields virtually the same results (results available upon request).

The results of this study add weight to recommendations for data to be disaggregated by age and potentially other variables to facilitate a better understanding of population-level differences in CFRs. Equally important will be well designed seroprevalence studies to ascertain the extent to which our findings are driven by differences in testing regimes, particularly in the diagnosis of mild and asymptomatic cases. To this extent we are encouraged by the recent start of such a study in Germany in line with official WHO recommendations [32, 33] and by first results from a large, population-representative studies from Italy and Spain [34, 35].

Overall, our results show that differences between countries with low and high CFRs can be driven to a significant extent by the age structure of confirmed cases. Decomposing differences in case-fatality rates over time or between countries reveals important insights for monitoring the spread of COVID-19. An accurate assessment of these differences in CFR across countries and over time are crucial to inform and determine appropriate containment and mitigation interventions, such as social confinement and mobility restrictions.

Details on methods and additional results.

(DOCX)

Click here for additional data file.

8 Jul 2020

PONE-D-20-12360

Monitoring trends and differences in COVID-19 case-fatality rates using decomposition methods: Contributions of age structure and age-specific fatality

PLOS ONE

Dear Dr. Dudel,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

As I mentioned by e-mail, it was quite difficult to secure reviewers for this article during this very busy period. Many candidate reviewers declined my invitation due to lack of time or conflicts of interest. Two reviewers have commented on your work, and I am very grateful for their dedication to read and carefully review your mauscript. Based on these reviews and my reading of the paper, I would like to open up the possibility of revising the article to strengthen it, taking care to answer the questions of the second reviewer, in particular her concern about the use of COVID notified death data, which represents only a portion of COVID-related deaths. I think that this point can be addressed through expanding the discussion or ideally through applying this method on weekly mortality death counts available in the HMD. I also agree with reviewer 2 about her suggestion to supplement the trends over time with other countries (the US is a candidate to consider). Could you also consider an update of the results to capture the decline in some countries and consequently a re-reading of some elements of the conclusion? The paragraph on seroprevalence surveys should also be updated in the light of the new studies available.

Please submit your revised manuscript by Aug 22 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

We look forward to receiving your revised manuscript.

Kind regards,

Bruno Masquelier, PhD

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Partly

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: This paper makes an important contribution by highlighting the association between population-level case fatality rates (CFRs) for COVID-19 and the age-structure of underlying infections as well as age-specific case fatality rates. The authors show that the age structure of confirmed cases can explain up to 2/3 of the differences in CFRs across countries.

The authors are to be commended for their demonstrated commitment to open science in making the data and code freely available, especially during the quickly changing landscape of research during the COVID-19 pandemic. The tool that is provided will be extremely valuable to scholars working to compare CFRs and the severity of COVID-19 across countries and time. Given the timely nature of the work and the clearly written contribution, I recommend publication as is.

One point of clarification: Could the authors clarify in the text if you also use data over time for Germany as in Figure 1.

Reviewer #2: This paper proposes an interesting method to compare and understand the impact of COVID19 on different countries. While this paper addresses an important and current topic with an intuitive and sound method, I found the analysis is not build upon robust data.

The authors use a mathematical decomposition method to separate the difference between two crude fatality rates (CFR) into an age-structure part and an age-specific case-fatality rate part, leaving no residual. They provide cross-country comparisons but also within-country comparisons over time. Countries included are China, Germany, Italy, South Korea, Spain, United States in addition to the city of New York. The method requires selecting a reference country (South Korea) or a reference date (9 March for Italy) to which they compare other CFRs. All included countries had a higher CFR than South Korea. According to the authors, age structure has the biggest relative importance in explaining the difference in CFRs with South Korea. The CFR of Italy has been decomposed over time and the authors highlight that the fatality component is the one driving the evolution of the CFR.

Strengths:

- The decomposition method is intuitive and well explained. In addition, it is proposed as a pertinent method to decompose the difference between other metrics (IFR, excess all-cause).

- Authors provide explanations for differences in CFR according to two important demographical factors.

- Data is subject to important limitations but these limitations are clearly highlighted in Introduction and are well discussed in Discussion.

- They reference the available literature which shed more lights on the topics and enrich the discussion.

Weaknesses:

- The variable used in their analysis is the ratio between reported COVID19 deaths and reported cases. Reported cases is highly determined by testing policies of a given country. Concerning reported COVID19 deaths, there is now a consensus in the demographic research community that the impact of COVID19 should be studied with all-cause death data. Reported COVID19 deaths are subject to multiple biases (national definition of COVID19 deaths, nurse houses not always included, reporting delays, ..). Having this in mind, it is hard to draw any robust conclusions from the results.

- Each country’s reported deaths might be subject to different biases. The fact that the method relies on the comparison to a reference country makes it even harder to understand in which direction biases go.

- For a majority of the analyzed country, the end date of the analysis does not consist of the end of the first wave. Thus, even when focusing only on the first wave, only part of the picture is studied.

Areas for improvement/Questions to the authors:

Major:

1) p.15, first paragraph: The authors mention that the method could be used on excess all-cause weekly mortality counts. This data is now available for some countries (see the Human Mortality Database for example). I would suggest that the authors use excess deaths that are less subject to bias and compare their results.

2) p.10: Trends over time were only studied for Italy. I would suggest the authors to analyze how trends differ between countries (i.e with Spain, the US) or at least address why they don’t.

3) The authors should present data on the amount of testing performed by each country, if possible, over the available age groups (and over time for Italy). It would be more transparent to read results having this information on the side.

4) p.4: “Counts of confirmed cases and deaths might not be comparable across countries because of differences in case and death definitions”. Despite rightly highlighting this fact, the purpose of the selected method is cross-country comparisons. Having this in mind, can the results support any conclusion?

5) Amount of missing data is only presented for Spain. What is the amount of missing data in the other countries? Could you add this information for all the countries?

6) p.6: “The database project imputed the missing age using the observed age distribution of cases or deaths, respectively”. Doesn’t this method exacerbate the bias if some ages are rarely reported (i.e deaths in nurse houses)?

Minor:

7) p.7 “Counts were split using a recently proposed method tailored for this data situation [9,10]”. For transparency, I would recommend to highlight assumptions required by this method (such as the underlying smoothness assumption of the counts).

8) In order to provide more descriptive statistics, could figure 1 be done for all the countries included in the analysis?

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

10 Aug 2020

For our replies to the reviewer and editor comments please see the attached documents.

Submitted filename: Response to reviewers.docx

Click here for additional data file.

27 Aug 2020

Monitoring trends and differences in COVID-19 case-fatality rates using decomposition methods: Contributions of age structure and age-specific fatality

PONE-D-20-12360R1

Dear Dr. Dudel,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Bruno Masquelier, PhD

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #2: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #2: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #2: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #2: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #2: This paper proposes an interesting method to compare and understand the impact of COVID19 on different countries. The paper addresses an important and current topic with an intuitive and sound method. The revision allows to contextualise better the different important results and improves transparency.

Thank you for addressing rigorously my concerns.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #2: No

1 Sep 2020

PONE-D-20-12360R1

Monitoring trends and differences in COVID-19 case-fatality rates using decomposition methods: Contributions of age structure and age-specific fatality

Dear Dr. Dudel:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Bruno Masquelier

%CORR_ED_EDITOR_ROLE%

PLOS ONE