key: cord-315536-fgjhli0p
authors: Bignami, Simona; Ghio, Daniela; Van Assche, Ari
title: Estimates of COVID-19 case-fatality risk from individual-level data
date: 2020-04-22
journal: nan
DOI: 10.1101/2020.04.16.20067751
sha: 
doc_id: 315536
cord_uid: fgjhli0p

When calculated from aggregate data on confirmed cases and deaths, the case-fatality risk (CFR) is a simple ratio between the former and the latter, which is prone to numerous biases. With individual-level data, the CFR can be estimated as a true measure of risk as the proportion of incidence for the disease. We present the first estimates of the CFR for COVID-19 by age and sex based on event history modelling of the risk of dying among confirmed positive individuals in the Canadian province of Ontario, which maintains one of the few individual-level datasets on COVID-19 in the world.

The severity profile of a novel pathogen is one of the most critical issues as it begins to spread, when assessing disease course and outcome is crucial for planning health interventions (1) . For this reason, an important epidemiological indicator to monitor during the current outbreak of COVID-19 is the case-fatality risk (CFR), the proportion of confirmed cases who result in fatalities. When calculated from aggregate data on confirmed cases and deaths, the CFR is a simple ratio between the former and the latter, which is prone to numerous biases (1) (2) (3) (4) (5) . With individual-level data, the CFR can be estimated as a true measure of risk as the proportion of incidence for the disease.

There are only two accessible individual-level datasets on COVID-19 confirmed positive cases that include their clinical outcome (death or recovery), which are maintained by the Canadian provinces of Ontario and Alberta 1 . Ontario is currently the second province in Canada for number of COVID-19 infections and fatalities. We present the first estimates of the CFR for COVID-19 by age and sex based on event history modelling of the risk of dying among confirmed positive individuals in Ontario. 2 

Between January 1 and April 15, 2020, Ontario recorded 8,447 confirmed positive cases of COVID-19, of which 3,902 had recovered and 385 had died. The individual-level dataset we analyse contains basic information on age, gender, clinical outcome (death or recovery) and name of the reporting health facility for these cases. Individuals' age is recorded at the time of 1 The dataset for Ontario is available at: https://data.ontario.ca/dataset?keywords_en=COVID-19. The dataset for Alberta at: https://covid19stats.alberta.ca. 2 The number of COVID-19-related deaths recorded in Alberta (48 as of April 15) is too small for this type of analysis.

All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 22, 2020. . testing in ten-years groups, with information missing for 5 cases (of which 2 are recoveries and 3 have an unresolved outcome). Information on gender is missing for 46 cases (of which 5 are fatalities, 12 are recoveries, and 29 have an unresolved outcome). These observations were excluded, so that the analysis is based on 8,394 cases, of whom 56.2% (4,716) are females.

Following the guidelines of the World Health Organization, in Canada only individuals symptomatic with fever, cough and/or difficulty breathing are tested for COVID-19. In Ontario, case definition refers to these symptomatic individuals who tested positive for the virus and also to their close contacts even if they were not tested for the disease. In our dataset, this latter category included 1,429 (17%) of cases.

We estimate the CFR through event history modelling in order to take into account censoring that arises because, at the time of observation, the outcome is unknown for a nonnegligible portion of infected individuals. In this framework, the CFR coincides with the cumulative incidence (CI) of mortality, with recovery as a competing risk. The CI is estimated after fitting the competing risk regression model, controlling for gender, with stcrreg in Stata/SE (version 12.0, StataCorp, LLC). The CFR (cumulative incidence of mortality vs recovery) for COVID-19 confirmed positive cases in Ontario is presented in Table 1 . It can be seen that it increases exponentially with age, All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 22, 2020. . https://doi.org/10.1101/2020.04.16.20067751 doi: medRxiv preprint reaching 18.7% for those aged 90+. By comparing the estimated cumulative incidence of death for COVID-19 with the crude case-fatality risk calculated from aggregate data on the total number of cases and deaths, we can appreciate to what extent the latter over-estimates the risk of COVID-19-related mortality in each age group. This is especially the case in the 70-79 and 80-89 age groups, when the difference between the two estimates is 6.2% and 5.0%, respectively. At any age, female positive cases have a risk of mortality that is 29 percent lower than male cases (SHR=.706; CI=.581, .858 ; p=.000). The comparison of the cumulative incidence of COVID-19-related mortality for males and females is presented in Figure 1 . 

Methods for Estimating the Case Fatality Ratio for a Novel, Emerging Infectious Disease

Potential Biases in Estimating Absolute and Relative Case-Fatality Risks during Outbreaks

The many estimates of the COVID-19 case fatality rate

A demographic adjustment to improve measurement of COVID-19 severity at the developing stage of the pandemic

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

Building an International Consortium for Tracking Coronavirus Health Status

Beware of COVID-19 projections based on flawed global comparisons

No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted