key: cord-280672-6x968dwk
authors: Fisman, David N.; Greer, Amy L.; Tuite, Ashleigh R.
title: Age Is Just a Number: A Critically Important Number for COVID-19 Case Fatality
date: 2020-07-22
journal: Ann Intern Med
DOI: 10.7326/m20-4048
sha: 
doc_id: 280672
cord_uid: 6x968dwk

In their article, Sudharsanan and colleagues show the importance of adjusting for the age distribution of cases of coronavirus disease 2019 before doing cross-country comparisons of case-fatality rates. The editorialists explore the effect of age distribution on these rates and other determinants of between-country variation in the severity of this disease.

A pandemic, by definition, represents worldwide, simultaneous epidemics caused by a novel pathogen. The multinational nature of such an event inevitably leads to cross-national comparisons of epidemic growth, impact, and public health response. Such comparisons lend themselves to ecological research: For example, the apparent slower epidemic growth rates in countries that use bacilli Calmette-Gué rin (BCG) vaccine has caused some researchers to assert that BCG vaccination may affect susceptibility to severe acute respiratory syndrome coronavirus 2 (1) . Others have made similar observations about higher mean temperature and slower growth of the coronavirus 2019 (COVID-19) epidemic (2) . These national-level comparisons are vulnerable to "ecological fallacy," or attribution of individual-level outcomes to aggregate exposures (3). However, they also represent "unfair comparisons" (3), because the countries in question differ fundamentally on a confounder known to be associated with COVID-19 severity: age.

The association among age, disease severity, and COVID-19 case recognition has been clear since February 2020 (4) . Older cases are more likely to be represented in surveillance data owing to greater severity and hence ascertainment. Failure to recognize younger, milder cases diminishes the denominator in case-fatality ratio (CFR) calculations (that is, deaths/cases), so between-country differences in age structure explain some fraction of observed between-country variation in epidemic severity and case-fatality. An analysis by Sudharsanan and colleagues (5) used data from 9 countries to demonstrate the importance of adjusting for the age distribution of cases before doing cross-country comparisons of CFR.

To ensure that between-jurisdiction comparisons are fair comparisons, the authors used the epidemiologic tool of standardization (6) . Direct standardization by age requires estimation of age-specific risk from different populations, which are then applied to a standard population, such that resultant differences in overall risk cannot be due to differences in population age structure. In their analysis, the authors show that adjusting for differences in population age structure substantially reduces the observed differences between country-specific CFRs.

To further explore the effect of age distribution on CFR, we can take the standardization approach used by Sudharsanan and colleagues and turn it on its head. We can apply age-specific risk from a single epidemic to other countries, to observe how an identical epidemic, from an age-specific attack rate point of view, might be perceived differently in different places, simply due to different age structures. Here, we use data from mainland China for the 66-day period from 7 December 2019 to 11 February 2020 (4). The analyses described below can be further explored in an associated app (https ://art-bd.shinyapps.io/time_to_outbreak_detection/). When the reported CFR for February 2020 in mainland China (2.3%) is age-standardized using population pyramids from other countries (7), standardization to a country with a younger population structure, such as Indonesia, markedly reduces observed CFR (1.7%), whereas adjustment to a country with an older population, such as Italy, increases the CFR (3.9%).

We can also estimate epidemic size using this approach but need to adjust for population size as well (larger countries, for a given attack rate, will have larger epidemics). We define the ratio of population in the other, comparator country (P O ) to the Chinese population (P C ) as R P = P O /P C . The ratio of the epidemic size in the other, comparator country (E O ) to observed Chinese epidemic size (E C ) is defined as R E = E O /E C . The "ratio of ratios" is R E /R P , which can be interpreted as the relative apparent outbreak size when an outbreak with identical age-specific attack rates occurs in a population with an age-structure that differs from that of mainland China. Just as CFR for an identical epidemic is expected to be higher in countries with older populations, the ratio of ratios, R E /R P , is greater than 1 for countries with older populations (1.15 for Italy) and less than 1 for countries with younger populations (0.81 for Indonesia). In other words, identical epidemics, adjusted for population size, appear smaller in countries with younger populations (shorter life expectancy) than in those with older populations (increased life expectancy), even with identical age-specific attack rates.

Age structure may also affect the time to recognize an epidemic. Countries with younger populations are likely to have more silent spread and be slower to identify epidemics. This may have been the root cause of a controversy that emerged early in the COVID-19 pandemic: Indonesia was predicted by models to have early importation of COVID-19 cases, but this was not consistent with Indonesian observations (8) . Critical illness and death associated with COVID-19 may result in initial outbreak identification and are more likely to occur in older individuals; we can arbitrarily define "older" as age greater than 59 years. We can calculate the incidence rate for observed infection, and the rate of transition to death, among susceptible older individuals in the mainland Chinese population in the early days of the epidemic by using an exponential failuretime model combined with published natural history data (4, 9) . When we simulate the mainland China epidemic in other countries, deaths accumulate more quickly in countries with high life expectancy (older populations) and more slowly in those with low life expectancy (younger populations). This is not to say that age distribution is the only determinant of between-country variation in epidemic severity. As Sudharsanan and colleagues (5) show, once age-related effects are removed, variability in CFR estimates remains. Differential outbreak responses are likely responsible for some of this variability (2): Weak public health responses that result in overwhelmed intensive care units will cause case fatality to inflect upward. Failure to adequately protect long-term care facilities from COVID-19 will swell CFR estimates as well.

Availability of testing is another key determinant of observed case fatality: A recent analysis found that more testing increases the case numbers in the CFR denominator, resulting in lower CFR, with residual variability in CFR explained by age structure and country per capita gross domestic product (10) . Finally, decisions about which deaths to classify as "COVID-19attributable" vary across countries. Serologic testing will ultimately help determine the true infection fatality ratio for COVID-19 and better quantify underrecognition of cases by age, but until such data are widely available, standardization provides a straightforward first step to ensure that between-country comparisons are fair comparisons.

Correlation between universal BCG vaccination policy and reduced morbidity and mortality for COVID-19: an epidemiological study. medRxiv. Preprint posted online 28

Impact of climate and public health interventions on the COVID-19 pandemic: a prospective cohort study

Ecological fallacy and aggregated data: a case study of fried chicken restaurants, obesity and Lyme disease

Vital surveillances: the epidemiological characteristics of an outbreak of 2019 novel coronavirus diseases (COVID-19)-China, 2020. China Centers for Disease Control Weekly

The contribution of the age distribution of cases to COVID-19 case fatality across countries. A 9-country demographic study

Standardization: a classic epidemiological method for the comparison of rates

R: a language and environment for statistical computing. R Foundation for Statistical Computing. 2020. Accessed at www.R-project

Using predicted imports of 2019-nCoV cases to determine locations that may not be identifying all imported cases. medRxiv. Preprint posted online 11

Mathematical modelling of COVID-19 transmission and mitigation strategies in the population of Ontario

Estimating the global infection fatality rate of COVID-19. medRxiv. Preprint posted online 11