It doesn’t matter that the United States surpassed China this week in reported COVID-19 cases because those numbers (83,507 and 81,782 respectively as of March 26) don’t tell us how many people actually became infected in either country. Nor do they tell us how fast the disease is spreading, since only a tiny portion of the population in the United States has been tested.
“The numbers are almost meaningless,” says Steve Goodman, a professor of epidemiology at Stanford University. There’s a huge reservoir of people who have mild cases, and would not likely seek testing, he says. The rate of increase in positive results reflect a mixed-up combination of increased testing rates and spread of the virus.
We will need more complete data, smarter data and more coordinated data to communicate something meaningful about the extent of COVID-19 in the United States, how many people are likely to die, which hospitals are likely to be swamped and whether drastic changes in the way Americans live will start to slow down the spread of the virus.
Random sampling would help too, to estimate the number of mild or asymptomatic cases and get at the true total. And then there’s the promise of widespread antibody testing, which could reveal how many people in a given sample had been infected in the past.
With a population of 1.5 billion people, China’s some 80,000 cases look like a rounding error, says Nigam Shah, an assistant professor of biomedical statistics at Stanford. And India’s claim of some 754 cases probably reflects a severe lack of tests — not that the disease there is still so rare. The positive tests say little about how many people are dying or will die, since most cases are mild.
What should we be watching instead? One possibility is hospitalisation. That idea was put forward by statisticians Jacob Steinhardt, an assistant professor from UC Berkeley, and Steve Yadlowsky, a graduate student at Stanford who specialises in analysing health care data. They argue that rate of increase in hospitalisation could reflect the growth of the disease without being distorted by changes in the testing rate.
Measuring death rates can eventually track the speed with which COVID-19 is spreading — as deaths represent a fraction of cases. But there’s a lag of some three weeks between infection and death. Hospitalisation give an intermediate point, as Steinhardt and Yadlowsky explain: They estimate that it takes between 11 and 14 days for someone to get sick enough to show up at the hospital. Rates of increase in COVID-19 patients admitted to the ICU can provide additional useful data.
How to forecast the number of patients
These numbers might not accurately reflect the growth of the disease, however, if the hospitals or their ICUs become overwhelmed, start turning people away or raise the threshold for how sick you have to be to be admitted.
But collecting this kind of data can help prevent that from happening, said Stanford’s Shah.
If we all behave responsibly, he says, then we can turn what would have been a hospital capacity problem into a logistics problem. Once you have a handle on the rate of new COVID-19 patients admitted to hospitals and ICUs, you can start to forecast how many more will arrive in coming days.
Stanford’s Goodman said that he’s confident scientists will eventually collect the data we need to understand this pandemic and how it’s playing out in the United States. “Right now we are floundering in a sea of ignorance about who is infected and the fate of people who are infected,” he says.
Though death rate figures of around 1 per cent have been tossed around, Goodman says he’s sceptical that anyone knows the death rate of this disease since we don’t know the true rates of infection.
And we can’t identify the most vulnerable groups. “There’s this delusion being disseminated that it’s all about age,” he says. He thinks that since 95 per cent of deaths to date in New York City were of people who had pre-existing conditions, this is the bigger risk factor. But since age is a risk factor for many of those conditions, the two are correlated.
He could figure it out if he could get data on pre-existing conditions broken down by age, but says the New York health department won’t release that data. It matters a lot, he says, since we’re shaping policies around who is most vulnerable. We should find out who they are. They should know who they are.
Some other useful data could easily be collected at testing sites. As doctors Farzad Mostashari and Ezekiel Emanuel pointed out recently in STATnews, health departments should tally not just positives but total tests, and record demographic and symptom information on all the test takers. Much of that isn’t collected or coordinated.
Random sampling would help too, agree both Shah and Goodman, to estimate the number of mild or asymptomatic cases and get at the true total. And then there’s the promise of widespread antibody testing, which could reveal how many people in a given sample had been infected in the past.
With attention to the right kinds of data, scientists can soon assess whether lockdowns and social distancing efforts are slowing the rate of spread in the United States. Any dent we’ve made in new infections should start to show up in data on hospital admissions in a week or two.
Trump promised Americans we could ease up on restrictions by Easter, while most scientists would like to wait until they can base such changes on evidence. Goodman says at this point, figuring out what to do next is like building an aeroplane in the air. In a later phase of the pandemic, we might be able to focus more on mass testing and quarantining people known to be sick or exposed. We probably can’t responsibly stop lockdowns by Easter, but we may know enough by then to start to think about the timing and nature of an exit strategy.
Faye Flam is a Bloomberg Opinion columnist.