Area-level variation in infection fatality ratio in the US COVID-19 pandemic

By Dr. Liji Thomas, MDReviewed by Danielle Ellis, B.Sc.Dec 7 2021

The coronavirus disease 2019 (COVID-19), caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), led to a devastating pandemic that has taken over five million lives worldwide, besides closing down the global economy for months. Intensive research went into understanding the biology of the virus and modeling the impact of various non-pharmaceutical interventions and vaccines on the global infection and fatality numbers.

Study: Estimating area-level variation in SARS-CoV-2 infection fatality ratios. Image Credit: MartinsArts/Shutterstock

A new preprint describes the infection fatality ratio (IFR), a crucial indicator for planning public health programs and shaping policies.

*Important notice: medRxiv publishes preliminary scientific reports that are not peer-reviewed and, therefore, should not be regarded as conclusive, guide clinical practice/health-related behavior, or treated as established information.

Background

The IFR may be defined as the number of deaths due to an infection per 1,000 infections. It varies by age distribution, demographic group, the prevalence of comorbidities, healthcare availability, and other factors, making a flexible assessment method a necessity.

It is difficult to accurately estimate the number of infections in the case of a virus like the current one, which causes many asymptomatic infections. Moreover, many infected people do not get tested or report their infection. Different countries and regions may not have sufficient testing capacity or access to tests, while test accuracy may vary across regions and time.

Even the number of deaths may be under-counted or over-counted in some situations, but the case mortality data is used to provide the upper bound of the IFR.

The current study that appeared on the medRxiv* preprint server developed an inferential framework that could help understand the variation in testing, number of infections, and the IFRs across counties. The framework presented here can also be used to study area-level variation in other pandemics and other geographic settings.

What did the study show?

The researchers found that from March 1^st, 2020, to October 31^st, 2020, the IFRs ranged between 0 and 0.0273. the mean being 0:0023. The IFRs varied widely between regions, with the highest being in Arizona and Western Mexico, but the lowest in the Intermountain West and High Plains

The number of infections, meanwhile, was highest in the Southwest but lowest in Utah. Testing numbers were highest around the most populous centers. The accuracy of these trends is borne out by the similarity at the state level. Using Pearson correlation, the mean estimates for IFRs, number of infections, and tests were very accurate at county and state levels.

This was not the case with the IFR estimates based only on the numbers of cases and case fatality ratio, which were off by some orders of magnitude.

What are the implications?

The researchers described and tested an inferential framework that would enable them to estimate the number of tests, infections, and IFRs in the US at the county level. This is built on a noncentral hypergeometric model that adjusts for the differential probability of positive tests in sick rather than healthy individuals.

This model was designed to supplement seroprevalence studies and detailed case tracking studies. This depends on an assumed baseline IFR based on either of the above data sources that may be either non-available in some regions, or non-representative in their reach, or may not be adjusted for the delay from infection seroconversion.

To counter this, the scientists used testing efforts of reliable origin to understand and compensate for variation at area-level and over time.

Excess deaths are also used to estimate the total mortality that could be due to COVID-19, but this could show wide fluctuations depending on the comparison period in use. This can be useful only if detailed estimates of mortality at the area level can be traced directly to the infection. The current model may help fulfill this need, especially since it uses the best data at hand, estimating only where data is not available on the basis of the most high-quality data obtainable.

For instance, testing efforts data used in this model include only polymerase chain reaction (PCR) tests and not the more lately approved rapid antigen tests. The underlying assumptions of this model could be stretched if necessary to make it more applicable over several situations.

For instance, the odds of testing infected or uninfected individuals need not be the same overall regions. The assumed overall IFR value could range from 0 to 1 depending on the location, the infection, the period and so on, and the absolute value may be immaterial if the relative IFR is sought for various areas.

This versatile approach may be useful when studies are required over regions without the resources for extended seroprevalence studies since it can be changed to adjust for time and space differences. This allows for changing IFRs with differences in age, comorbidity status, healthcare availability and quality, and other factors.

The ability to estimate area-level IFR could help plan policies that target areas in need of greater intervention. Moreover, it could address the differences between the IFRs of various countries that remain unexplained by the population's age structure.

In conclusion, our approach can serve as an extension to other methods for estimation of pandemic burden to allow for greater precision around geographic and/or temporal population subgroups."

*Important notice: medRxiv publishes preliminary scientific reports that are not peer-reviewed and, therefore, should not be regarded as conclusive, guide clinical practice/health-related behavior, or treated as established information.