Please Stop Comparing COVID-19 Deaths with Traffic Fatalities

You may have recently heard some people comparing the number of deaths from COVID-19 with the annual number of traffic fatalities. This comparison is then followed by the statement “But we don’t close down the highway system because there are 37,000 traffic fatalities per year.” That’s true; we don’t close down the highways. And we also don’t close down all the bathrooms in the country even though an estimated 234,000 people per year go to the emergency room after suffering an injury in a bathroom.

But this is a false comparison. First, neither traffic fatalities nor bathroom injuries are growing at an exponential rate; if they were, you would certainly hear calls for drastic action. Second, the persons making the comparisons are using different measures of risk for COVID-19 and traffic fatalities. Third, no one possessing a heart says that the current level of traffic fatalities is acceptable and that we shouldn’t try to reduce it. In the last 40 years, tremendous progress has been made toward making cars and roads safer, reducing drunk driving, increasing seat belt use, and other safety enhancements. And fourth, the U.S. has had detailed data on nearly all fatal traffic accidents since 1975. The inadequacy of data on COVID-19 has made it difficult to anticipate needs or track progress, although research institutions are now starting to assemble and disseminate better data that can be used in the fight against this disease.

Exponential Growth

How many times have you heard a TV pundit say something like this: “The number of X is three times higher this year than last year — it’s growing exponentially!” After which you shout at the television, “No, it isn’t! That’s linear growth, not exponential growth! Stop using ‘growing exponentially’ as a synonym for ‘bigger’!”

Figure 1. Number of confirmed cases in the U.S. Source: Johns Hopkins University, downloaded 10:48 am EDT March 25, 2020.

But COVID-19 cases and deaths really are growing exponentially. Right now, the number of cases (and number of deaths) are doubling in a fixed time period. This is like the doubling-pennies problem, where if you’re given one penny on June 1, two pennies on June 2, four pennies on June 3, and continue doubling the number of pennies you’re given on each successive day, at the end of June you have almost 11 million dollars.

As of March 25, the U.S. has the fastest rate of growth among the countries studied by Vox.com, and the number of deaths in New York is doubling every two days. Figure 1 shows the number of confirmed cases in the U.S. from January 21 through March 23. This is what exponential growth looks like. And this is why public health experts are telling people to do everything they can to slow the growth of number of cases. Even though eventually the exponential growth would stop on its own (after everyone has had the disease), millions might die while waiting for the “natural” process to end.

An analogous situation for car crashes would occur only if every car crash causes a chain reaction. Each car involved in a collision draws two more cars into the pileup, those two cars each bring two more in, and continuing like the doubling-pennies problem. Soon, there are no medical resources left to help new victims but crashes keep occurring. Wouldn’t any responsible person shut down the highway in that situation?

Statistics on Traffic Fatalities

Traffic fatalities are not growing exponentially. In fact, they have decreased over time. In the 1970s, there were about 45,000-50,000 fatalities from traffic crashes every year; since 2010 there have been about 32,000-38,000 fatalities per year. These numbers come from the Fatality Analysis Reporting System of the National Highway Traffic Safety Administration (NHTSA), a census of vehicle crashes involving fatalities in the United States including information on driver/passenger/pedestrian status, alcohol/drug use, seat belt use, road and weather conditions, vehicle makes/models/years, type of collision, date and time, and other characteristics for each crash.

Figure 2. Fatalities from traffic crashes per 100,000 population, from 1966 to 2017. Data source: NHTSA Traffic Safety Facts, Table 2.

Just looking at the number of traffic fatalities each year does not give the full picture because the population and number of vehicle miles traveled have both been increasing while traffic fatalities have been decreasing. Figure 2 displays number of fatalities per 100,000 population. This is a population mortality rate; it estimates the risk that a person in the population will die in a traffic accident. (Often people use the number of fatalities per 100 million vehicle miles traveled to estimate the population risk, and that time series shows an even sharper decline over time than Figure 2. But in this post I want to stick to statistics that are used to estimate fatality rates from diseases.)

Remember, though, the fatality rates being quoted for COVID-19 are not population fatality rates. They are case fatality rates, and are currently computed as (number of persons who have died from the disease) divided by (number of confirmed cases of the disease). The case fatality rate estimates for COVID-19 are, in general, quite poor: they vary greatly from country to country and from state to state simply because the number of tests that have been administered vary so much and many persons with the disease are not tested. Germany, for example, has one of the lowest estimated case fatality rates for COVID-19 at 0.5%, but that may be partly because its extensive testing program has uncovered cases that would be undetected in countries that have not tested as widely.

But even if the case fatality rate estimates for COVID-19 were accurate, it’s not valid to compare these with the population mortality rates for traffic fatalities, because the statistics measure different things.

Traffic Safety, COVID-19, and Data

In fact, society has not accepted carnage on the highways as unavoidable. The story of decreased traffic fatalities highlights how science, statistical data collection, and action have saved lives (although there is still much progress to be made).

Fatalities attributable to drunk driving have decreased (see Table 13 of Traffic Safety Facts, but again, more improvement is needed), and use of seat belts has increased. Moreover, cars manufactured today are much safer than those from 40 or even 20 years ago. Our 2018 Subaru Forester has a safety rating from theautoprofessor.com that is a full grade higher than the 2000 model it replaced (see my earlier blog post about auto safety statistics), and other makes and models of cars have shown similar improvements over time.

But we know that these improvements have occurred for traffic and car safety because data exist demonstrating them. There aren’t yet data of similar quality for COVID-19 in the United States as a whole. Some states and cities, however, are now publishing detailed information on cases.

New York City, for example, presents updated statistics on COVID-19 cases twice a day. The website also provides guidance for how to slow the spread of the virus. The March 18 report issued by the CDC on severe outcomes gives statistics about hospitalizations, ICU admissions, and deaths by age group, but it does not give breakdowns by gender or pre-existing health condition. The New York City data has that information, and more. They report that, as of March 25, 57% of the cases and 60% of the deaths have been males. Moreover, nearly all of the deaths (231 out of the 241 deaths whose investigations have been completed) were people who had an underlying illness such as diabetes, lung disease, or cancer. These patterns are similar to what had been seen in Italy, where researchers are analyzing the data that have been collected to help fight this disease.

And more data resources have been opened up in the fight against this virus. The data science competition platform Kaggle has made available a data set containing information about the country, gender, age, and summary of exposure and outcomes for a set of individuals who have had the disease. The Allen Institute for Artificial Intelligence has partnered with other organizations to provide the COVID-19 Open Research Dataset, a collection of more than 40,000 research articles (updated weekly) relevant to COVID-19. The statistical and scientific communities are rising to the challenge.