At STAT’s request, Jha and his colleagues at the institute calculated the number of tests that each state would need to be doing as of May 1. STAT then compared those numbers to each state’s daily testing totals in an average week in mid-April, using data collected by the Covid Tracking Project.
The results show that states with few COVID-19 cases and deaths so far will need to perform relatively few tests: between 68 and 145 per day in Alaska and between 31 and 156 in Montana, for instance. States harder hit by the pandemic face a much heavier lift: New York would have to do 130,000 to 155,000 tests every day, New Jersey 75,000 to 90,000, and both Massachusetts and Illinois about 30,000 to 35,000.
Many hard-hit states are not even close to their goals. New York, for instance, has been averaging barely more than 20,000 tests per day since mid-April. New Jersey has been doing about 7,000, on average. Neither has announced reopening plans or dates, giving them time to ramp up testing. Massachusetts and Illinois are in no better shape, conducting just under 7,000 a day. Michigan, Connecticut, and Colorado are all about 15,000 tests a day below their May 1 targets. Texas, with more than 9,000 tests a day, and Washington state, with more than 3,000, are already doing enough.
The more worrisome gap involves states that, despite having thousands of COVID-19 cases, are easing mitigation strategies by, for instance, allowing more businesses and public spaces such as beaches to reopen. To catch hot spots before they turn into wildfires of disease, Georgia must do 9,600 to 10,000 tests per day; it has been averaging around 4,000. Florida will need 16,000; in the last week it has been hitting just above 10,000. South Carolina is a rare bright spot: It will need 1,200 to 1,600 tests per day and has been averaging close to the low end of that, with at least 1,500 tests on several recent days.
In the last week, the U.S. as a whole conducted 1.6 million tests, according to the Covid Tracking Project. The Harvard team says twice that many tests will be needed — at minimum.
Performing enough tests is only one of the essential steps before states can reopen, experts say. Test results also have to be returned more quickly, public health workers must identify and contact potentially exposed people, and hospitals and nursing homes require adequate amounts of personal protective equipment and other supplies in case a new wave of seriously ill patients crashes over them.
The Harvard institute based its calculations on best-case scenarios. If the goal were not to miss a single new infection, “you would need more than 300 million tests a day,” said Jha — testing nearly every resident of every state every day (or every few days). “That’s an interesting theoretical exercise but since it’s not going to happen, it’s policy-irrelevant. If you say that, people stop listening.”
Instead, he said, “we tried to come up with numbers that wouldn’t make governors gag.”
Jha and the institute’s Ben Jacobson crunched the numbers two ways for each state. Both start with the number of deaths projected for May 15 by Los Alamos National Laboratory, whose COVID-19 model the Centers for Disease Control and Prevention consults. Los Alamos projects that, nationally, the death toll on that day will be 545 — but more if states keep easing up on social distancing. For comparison, on April 25, the U.S. reported just over 2,065 new COVID-19 deaths.
The researchers then assume that deaths on May 15 will reflect the number of cases on May 1, since two weeks is the average time between infection and death. If 1 out of every 100 people diagnosed with COVID-19 dies (for a “case fatality rate” of 1%), then on May 1 there would be 100 times as many new cases as there would be deaths on May 15; call it 54,500 (545 times 100) cases on May 1 for the country as a whole. The case fatality rate for COVID-19 remains unknown, so if it is lower, perhaps 0.5%, then the number of cases would be greater: 109,000. Jha’s choice of 1% reflects his decision to make recommendations that are within reach of realistic testing capacity, and 1% is the rough consensus of experts now.
To control the epidemic, public health workers would need to identify those 54,500 new cases on or around May 1 in order to trace and test their contacts, quarantining those who also test positive. (The “around” is because cases don’t have to be identified right away; probably within two or three days of infection is OK.) The chances of picking precisely the right 54,500 people to test are nil.
It’s therefore necessary to test many times more than 54,500 people to find that number of cases. How many more?
Jacobson and Jha calculated that using two different methods, which serve as a check on each other.
One method is based on the fact that people most likely to be infected with the new coronavirus are also the most likely to be tested. But because of inadequate testing capacity, and because many people don’t even show symptoms, tens of thousands of cases have been missed.
Roughly 20% of U.S. tests are positive for the virus. Epidemiologists estimate that for infectious diseases such as influenza and tuberculosis, if more than 3% of people test positive, then the net is not being cast wide enough. South Korea has a 3% positive rate for its coronavirus testing. To achieve that, the U.S. would have to increase its number of daily tests to 2 million by May 1. (It should be 5 million now, but that ship has sailed.) Even 2 million seems beyond reach.
But the World Health Organization says a positive rate below 10% “reflects adequate testing,” Jha said. Using that rate, the U.S. would have to be testing 545,000 people per day by May 1 and every day thereafter, until projected deaths two weeks ahead fall.
At STAT’s request, Jacobson and Jha then applied the 10% positive rate to each state individually. That yielded an enormous range, from 155,000 a day in New York to just 18 a day in Wyoming. Is it feasible? That will vary by state, but last week California said it aims to conduct 25,000 tests a day by the end of April, up from an average of about 14,000 in recent days. The Harvard team calculated the Golden State needs about 26,000, suggesting that if California hits its goal it could well be on track to safely reopen.
As a check on their work, the researchers did the calculation a second way.
They again started with Los Alamos’ 545 projected deaths on May 15 and inferred that there were 100 times as many new cases, 54,500, on May 1. That’s a starting point for tracing contacts, which offers the best shot at containing both the current outbreak and any that threaten to erupt after shutdown orders are lifted: “Reopened” cities and states must catch and quarantine cases, then identify and test their contacts before they infect more people.
Without social distancing, each case has an estimated 19 close contacts, Jha said. With social distancing, each might have 10, again suggesting roughly half a million tests per day.
A lot of assumptions — case fatality rate, test positivity rate, and more — went into these calculations. But other researchers have come up with approximately the same number via different reasoning.
Last week, an analysis by the Rockefeller Foundation concluded that the U.S. should test 3 million people a week. Vital Strategies, the nonprofit headed by former CDC Director Tom Frieden, recommended a minimum of 450,000 tests per day. Researchers arrived at that figure by counting the number of people who have the highest priority for testing: people who are sick and their contacts, of course, but also nursing home and shelter residents, prison inmates, and vital workers in health care and public transit and other infrastructure, said epidemiologist Cyrus Shahpar of Vital Strategies.
As in the Harvard calculation, the national testing goal of 450,000 would not be evenly distributed according to state (or city) population, he said.
The White House has said that individual states and cities need to do roughly 30 tests per 1,000 people per month, as Deborah Birx, the White House’s COVID-19 response coordinator, explained at a recent briefing, citing New Orleans’ 27 tests per 1,000. Birx said that all but three states — Oregon, Maine, and Montana — had the ability to do that many tests and that the administration is working with states to ensure all the potential for testing is “brought to bear.”
Both Shahpar’s and Jha’s calculations, however, say the number of needed tests is closer to 45 per 1,000 people per month, but with significant regional variation.