The Washington PostDemocracy Dies in Darkness

Opinion Why it’s so hard to pin down the risk of dying from coronavirus

A worker prepares a coronavirus scanning and surveillance system kit in Ahmedabad, India, on March 6. (Sam Panthaky/AFP/Getty Images)
Placeholder while article actions load

Marc Lipsitch is a professor of epidemiology and director of the Center for Communicable Disease Dynamics at Harvard T.H. Chan School of Public Health.

This week, World Health Organization Director General Tedros Adhanom Ghebreyesus stated an undisputed fact: 3.4 percent of people with reported covid-19 infections worldwide have died. Or the fact was undisputed until misinterpretations set in, almost immediately. Some in the news media mistook the statement as meaning an alarming 3.4 percent mortality risk for coronavirus infections overall. President Trump weighed in on Wednesday, saying of the WHO statistic, “I think the 3.4 percent is really a false number.”

The latest updates on the coronavirus

Yet the WHO director general was simply describing the percentage of those with reported infections who had died. The risk of dying if you become infected is another matter.

For a variety of reasons that are common to epidemics, the covid-19 mortality rate remains elusive. My colleagues in infectious-disease epidemiology and I know the challenges of estimating fatality rates in the middle of outbreaks from almost two decades of experience, including with SARS in 2003 and H1N1 flu in 2009. To estimate risk, we need to know the infection-fatality rate, or IFR — the chances that a person who contracts the infection will die. Also useful is the so-called symptomatic case-fatality rate, or sCFR, the risk that a person who contracts the infection and becomes symptomatic will die.

Contracting an infection is one thing; showing symptoms is another. An unknown number of people — unknown, but maybe considerable — might be infected with covid-19 without becoming symptomatic.

There are two main sources of bias that make estimates of the IFR or sCFR so challenging early in an epidemic — and we are, alas, still early in the spread of covid-19.

One source of bias arises because, initially, we tend to see the most severe cases. In Wuhan, China, where the epidemic began, care and testing were prioritized for the sickest patients. In other places, such as Iran, the first covid-19 tests were administered because individuals were unexpectedly dying of pneumonia. Either way, the cases we know about are not a random sample of all cases, but a sample of the sickest — so the risk of dying is higher in the people we know about than in typical cases.

By contrast, the other source of bias can make us underestimate the risk of dying. At any moment in a growing epidemic, most cases are people who were infected recently — that’s what it means for an epidemic to grow: There are more new infections this week than there were last week. Many of the people with these new infections will recover, and some will die. We don’t yet know the fates of those who were infected, say, just yesterday or the day before. To count appropriately, we need to know how many of the current cases will die, not just how many have died.

Epidemiologists have devised statistical ways to estimate that number, to correct the simple calculation to reflect current and expected future deaths among the known cases. Another approach to avoid this problem is to use data from a population where the epidemic has subsided, and fatal case counts have had time to catch up with total case counts.

The coronavirus infection-fatality rate — the number everybody wants to know — is still undetermined. First, we need to figure out the proportion of infected people who show symptoms. That will ultimately require performing serologic testing (blood tests measuring antibodies) in a large cohort of an exposed population. Once we do that, the infection-fatality rate can be estimated from the symptomatic case-fatality rate times the proportion of infections that are symptomatic. This is the most useful number, because mathematical models and prior experience with flu can give us estimates of how many people might be infected — and with this number, how many are likely to die.

Serologic testing — to detect an antibody indicating whether a person is now or has been infected — is underway in some populations in China. That will help to give us a picture of the whole spectrum of covid-19 infection, from completely unrecognized, to symptomatic, to severe, to fatal.

Several estimates have suggested that the risk of dying, for those infected with covid-19 and showing its flu-like symptoms, is around 1 or 2 percent. Elderly adults have a considerably higher risk of both becoming infected and dying, as do people with compromised immune systems. The estimates might change as new data arrive, but the range of 1 to 2 percent for fatalities among the symptomatic seems to be the consensus for now. The overall fatality rate for people infected with covid-19 will be lower — possibly much lower — when we know how many people are infected but asymptomatic.

We’ve learned a lot about how to make these estimates from past outbreaks. The first lesson is that the estimates will change and be refined as scientists do their best to make sense of imperfect data.

Read more:

Richard E. Besser: As coronavirus spreads, the bill for our public health failures is due

Robert Gebelhoff: An epidemic expert: Are we past the point of containment for coronavirus?

John M. Barry: Can this virus be contained? Probably not.

Greg Sargent: Trump’s latest coronavirus lies have a galling subtext

Frida Ghitis: Trump’s coronavirus response puts his authoritarian instincts on full view

William Pesek: How the coronavirus is shaking up Asia’s political order