Background: Because many cases of coronavirus disease 2019 (COVID-19) are asymptomatic, generalizable data on the true number of persons infected are lacking. Mortality rates therefore are calculated from confirmed cases, which overestimates the infection fatality ratio (IFR). To calculate a true IFR, population prevalence data are needed from large geographic areas where reliable death data also exist. Most previous IFR estimates came from non-U.S. populations, including a cruise ship, or were calculated by using simulation techniques (1–3). Previous estimates also are not age specific, are relatively ungeneralizable, and are unsuitable for making clinical or policy decisions.
…Discussion: By using SARS-CoV-2 population prevalence data, we found that the risk for death among infected persons increased with age. Indiana’s IFR for noninstitutionalized persons older than 60 years is just below 2% (1 in 50). In comparison, the ratio is approximately 2.5 times greater than the estimated IFR for seasonal influenza, 0.8% (1 in 125), among those aged 65 years and older (5). Of note, the IFR for non-Whites is more than 3 times that for Whites, despite COVID-19 decedents in that group being 5.6 years younger on average.
We are unaware of any similar IFR estimates by demographic group but recognize several limitations of our analysis. First, despite random selection and weighting for nonresponse, the potential for response bias remains. Second, imperfections in tests have the potential for false-positives, which may bias estimated infections upward. Separately, use of confirmed COVID-19 deaths may undercount the true number of deaths; both issues might result in lower IFRs. Third, because children and non–state tax filers were excluded, our estimates may lack generalizability to persons who were not studied. Fourth, we could not account for disease severity among random-sample participants with positive test results.
Although participants represented persons with less severe illness, some with positive test results may have later died of COVID-19, resulting in a potential underestimation of the IFR. However, accounting for right-censoring bias also might overestimate the IFR, because we cannot distinguish deaths among persons we randomly tested from those among patients who were hospitalized during the testing period. Race and ethnicity data for confirmed COVID-19 deaths may have been inaccurate, thus biasing these IFR estimates. Lastly, IFR is a population-based measure and should be interpreted cautiously as a measure of individual risk