After much back and forth in the early months of the Covid-19 pandemic, a consensus is emerging that the overall risk of dying for those infected with the disease — at least so far, in a population with an age distribution roughly similar to that of the US or Europe — is about 6 or 7 in 1,000.
The Centers for Disease Control and Prevention upwardly revised its “best estimate” of the fatality rate in July to 0.65% from 0.26%. An occasionally updated “meta-analysis” by Australian researchers Gideon Meyerowitz-Katz and Lea Merone of all relevant studies on the disease has it at 0.68%.
This isn’t much below the approximately 1% estimated in a Feb. 10 study by the Covid-19 disease-modeling group at Imperial College London, which was adopted as a provisional consensus by many in the epidemiology and public health communities. It’s also within the range of 0.05% to 1% proposed in a March 17 op-ed article by Stanford Medical School professor John Ioannidis, although at the time the skeptical Ioannidis intimated that the true fatality rate was likely to come out toward the low end.
Ioannidis has his own running meta-analysis, based only on studies of the prevalence of coronavirus antibodies, that puts the median fatality rate at 0.24% but acknowledges that in the areas hardest-hit by Covid-19 it’s 0.9%.
For a variety of reasons, fatality-rate estimates from places with high Covid-19 prevalence are more likely to be accurate than those from places with low prevalence, plus if you’re trying to estimate a population-wide fatality rate then areas where the disease is widespread ought to weigh more heavily than those where it isn’t, so this doesn’t seem incompatible with an overall rate of 0.6% or 0.7%.
The varying results from different antibody studies do suggest that the disease may be less deadly where it is less prevalent — because health-care systems aren’t overburdened, and possibly because viral loads are lower — which seems like an important factor in deciding whether interventions to slow the spread of the disease are worth the effort. Improved medical techniques and treatments also appear to be reducing the severity of the disease, which could cut the fatality rate over time while also increasing the rewards to delaying tactics.
Some of these delaying tactics have of course come at huge economic cost, raising questions about whether the dangers posed by Covid-19 really merited such a drastic response. I am not here to offer entirely satisfactory answers to those questions! But I thought it might be useful to take a closer look at how Covid-19’s risks stack up against those posed by the most comparable menace, influenza — both the seasonal variety and the occasional global pandemics. And though I realize that catching an infectious disease brings lots of other risks short of death, fatality rates do seem like the best metric for comparison available at the moment.
Assigning an infection fatality rate to influenza turns out to be harder than one might think. That’s partly because there are multiple varieties of influenza, but also because until quite recently the terms “infection fatality rate” or “infection fatality risk” weren’t really a thing (in a search of the National Institutes of Health’s PubMed database of medical articles I found no uses of the former from before 2020 and only three of the latter). Instead researchers calculated what they usually called a “case fatality rate,” which in scientific articles I’ve perused has described:
1. The percentage of hospitalizations that result in deaths.
2. The percentage of laboratory-confirmed cases that result in deaths.
3. The percentage of symptomatic cases that results in deaths.
4. The percentage of infections that result in deaths.
That’s confusing, right? To be sure, definition No. 3 seems to be the most commonly used, and for many diseases it’s roughly equivalent to No. 4 because almost all infections result in symptoms. But as you’ve surely heard, a high percentage of infections with the new coronavirus — probably around 40% to 45% — don’t ever cause significant symptoms. Influenza infections are if anything even more likely to be asymptomatic. One meta-analysis found that 65% to 85% may be; 50% to 75% is another frequently cited range.
When I first went looking in early March for a rough estimate of seasonal influenza’s case fatality rate, 0.1% was what I most often encountered. This appears to be according to definition No. 3: If you divide the CDC’s estimates of deaths from influenza by its estimates of symptomatic cases over the past nine flu seasons, the resulting fatality rates range from 0.1% in 2018-2019 to 0.18% in 2010-2011, and average out to 0.13%.
To compare influenza fatality rates with the Covid-19 infection fatality rate, though, one really needs to factor in those asymptomatic infections. A while back, University of Oxford infectious disease epidemiologist Christophe Fraser suggested on Twitter that doing so with seasonal influenza would deliver an infection fatality rate of about 0.04%. When I ran this estimate by influenza expert Lone Simonsen, a professor at Roskilde University in Denmark who used to work at the CDC and NIH, she endorsed it. One can also just take the 0.13% symptomatic-case fatality rate from the CDC and the low-end estimate that 50% of infections are asymptomatic and conservatively calculate an infection fatality rate of 0.065% — exactly one-tenth what the CDC currently estimates for Covid-19.
That’s a tidy little result. Perhaps too tidy, given that seasonal influenza does appear to pose a greater danger to infants and toddlers than Covid-19 does. Only 25 Americans age 4 and younger have died from Covid-19, according to the CDC, while in most recent flu seasons the estimated fatalities for that age group have been in the low hundreds. Apart from that, though, the risk profile by age is quite similar for seasonal flu and Covid-19, with those 65 and older accounting for about 80% of US deaths from both. Maybe Covid-19 isn’t exactly 10 times more dangerous than seasonal influenza, but it’s probably in that ballpark.
This may seem hard to square with the CDC’s tally of 61,000 fatalities in the worst recent US flu season, that of 2017-2018 — more than one-third of the 158,268 deaths attributed so far to Covid-19. Part of the explanation is of course that Covid-19 isn’t done with us yet. The CDC estimates that there were 45 million symptomatic influenza cases in the US in 2017-2018, which if 50% of infections were asymptomatic would mean that 90 million Americans, or 28% of the population, were infected. Data scientist Youyang Gu’s handy infections tracker, which is based mostly on mortality data, estimates that just 9.2% of Americans had been infected with Covid-19 as of July 15.
A bigger issue may be that the CDC’s annual influenza numbers are statistical-model-based estimates that include “influenza-like” respiratory illnesses and deaths for which influenza was just one of multiple causes, while the Covid-19 numbers are based on an actual count of deaths attributed to the disease by state authorities (although of course other underlying causes are often involved as well). Harvard and Emory medical school professors Jeremy Samuel Faust and Carlos del Rio suggested in May that a better comparison would be with the number of influenza deaths counted in the CDC’s National Healthcare Safety Network reports, which have ranged from 3,448 to 15,620 in recent years. In the CDC’s “underlying cause of death” data based on death certificates, the highest influenza fatality total in the past two decades was 11,164 in 2018.
Another approach is simply to look at the overall number of weekly deaths and compare them with expectations based on past years’ data. According to the CDC there were 28,200 excess deaths over the course of the 2017-2018 flu season, and since late March of this year there have been an estimated 191,011. That’s almost seven times more and, again, the coronavirus isn’t done yet. Saying it’s roughly 10 times deadlier than the seasonal flu does not seem to be a great exaggeration. It may prove to be an understatement.
Influenza pandemics are another matter. They occur when a strain of influenza to which most people have never been exposed, and for which vaccines are not immediately available, sweeps the world. Those 65 and older are often less affected by these pandemics than younger cohorts, possibly because they were exposed to similar influenza strains in their youth. During the 2009-2010 H1N1 pandemic, for example, about 80% of the deaths were among those under 65. But only an estimated 12,469 Americans died of the disease. Its overall case fatality rate was a very low 0.02% and the infection fatality rate surely lower than that, so there’s really no comparison between it and Covid-19.
The pandemic influenza strains of 1968, 1957-1958 and 1918 were much more dangerous. It’s tough to sort out the infection fatality rates from the case fatality rates in the literature, but there were antibody surveys conducted during and after the 1968 and 1957-1958 pandemics, and World Health Organization fatality-rate estimates of a bit under 0.2% for both seem to take at least some asymptomatic infections into account. There was no antibody testing in 1918 (the influenza virus wasn’t identified till 1933), but a 2011 paper co-authored by the aforementioned Christophe Fraser concluded on the basis of transmission patterns that there were few asymptomatic infections, meaning that the case fatality rate, usually estimated between 2% and 3%, may represent something close to the infection fatality rate. Another way of measuring is just to count estimated US deaths from the three pandemics, which were, going back in time from 1968, 100,000, 116,000 and 675,000. As a share of the population, that’s equivalent to 164,000, 221,000 and 2.1 million deaths today.
A fuller comparison would adjust for changes in the age distribution of the population. US Army epidemiologist John F. Brundage did that a few years ago and concluded that an influenza strain with a virulence equivalent to 1918’s would have killed 1.3 million people in the US in 2006, far less than a straight share-of-population calculation would predict. In his accounting, the aging of the US had lessened the risk from a disease with mortality rates that were highest for infants and those in their 60s and 70s, but also quite high for young adults. Advances in medicine since 1918 surely would reduce the modern toll as well. Then again, a disease with the age profile of Covid-19 might have proved less deadly for the far-younger US populace of 1918 than it is for us now.
No matter what adjustments one makes, Covid-19 appears to be markedly less dangerous than the 1918 influenza, especially when you factor in that the latter wasn’t nearly as deadly in the US as in some other countries. On the other hand, the current pandemic seems certain to pass 1968’s in population-adjusted US fatalities — it probably already has, if you go by the excess mortality data — and quite likely to pass that of 1957-1958. If the coronavirus were to infect circa 30% of Americans, as those two are estimated to have done, at current fatality rates it would cause almost four times as many deaths (adjusted for population) as the 1968 pandemic and almost three times as many as 1957-1958. And because it is more infectious than influenza, Covid-19 might not stop at 30% in the absence of control measures. In one hard-hit area of Peru a recent antibody survey found that 71% of those tested had been infected; some neighborhoods of New York City may have infection rates nearly that high.
Again, such comparisons are complicated by medical advances, changing demographics and the differing age profiles of the diseases. A 2016 study of excess mortality from the 1957-1958 pandemic, co-authored by the aforementioned Lone Simonsen, found that 44.1% of the excess deaths across 39 countries were among those 4 and younger, versus 32.5% among those 65 and older. There were, to be sure, a lot of small children in those days (11.3% of the US population was 4 or younger, versus 6% now) and a lot fewer 65-plussers (8.9% then versus 16.5% now). Still, even in the absence of a quality-adjusted-life-years comparison that reflects the greater cost inherent in the loss of younger people’s lives, I think it’s fair to say that the 1957-1958 pandemic merits being at least mentioned in the same breath as Covid-19.
Yet the reaction to it was nothing like that we’ve seen this year. There were school closures, but they were far from universal. There was also a sharp if short recession, with the percentage drop in US real gross domestic product in the first quarter of 1958 the biggest on record until the second quarter of this year, but only a few contemporary observers appear to have attributed it to the pandemic.
So why did life temporarily grind to a halt in much of the world for Covid-19 when it did not for a new strain of H2N2 influenza in 1957? One possibility is that public health officials, political leaders, the news media and many others have overreacted this time around. Another is that they underreacted in 1957 and 1958. Also, the US and most of the rest of the world are much healthier and wealthier than they were in the 1950s and 1960s, meaning that a threat such as Covid-19 stands out more than it would have then, and that the resources available to fight it are greater. Again, I don’t think comparisons like those I’ve offered here tell us what the correct pandemic policies are. But they do at least help us better understand how Covid-19 stacks up, which is to say that it’s definitely the worst such pandemic to come along in 60 years and probably the worst in a century.
Bloomberg
