Why we need at least 500,000 tests per day to open the economy — and stay open.
Partnering with The New York Times, we published a first projection of a nationwide testing target on April 18, estimating how many tests the entire nation should be performing by May 1, 2020.
Ubiquitous testing is essential to the nation’s ability to succeed with the planned phased opening of the economy and to stay open (we explain why we need testing in this post). In our analysis below, we estimate that the number of tests needed every day is, at a minimum, 500,000, though we likely need many more.
In order to know how many tests we need in the U.S., we have to have some idea of how many infected people there are. Let’s say you live in a country with 300 million people. If there are 1,000 infected individuals in the country today and the symptoms of COVID-19 are pretty distinct, we will need fewer tests than if there are 1 million infected people in that country. This, of course, assumes we are willing to miss a small proportion of infections. If we want to capture all the infections with complete certainty, we will need to test all 300 million people, which is impractical.
So, we can say, “Okay, I’m willing to miss a few percentage points of infected people on any given day.” The more we are willing to miss, the fewer tests we’ll need (but the more it will fuel the next outbreak). It’s a tradeoff.
We don’t know how many undiagnosed cases of COVID-19 there are in the U.S. today because we have such limited testing capacity.
We know that, on average, we’ve been identifying about 30,000 cases of COVID-19 every day through testing. We have been doing about 150,000 tests per day, meaning that our positive rate is about 20%.1
For context, in South Korea where testing has been extremely effective in limiting transmission, there is a 3% positive rate. In Germany and New Zealand, positive rates range from 6-8%. The World Health Organization has suggested that an adequate test positive rate should be between 3 and 12%; a test positive rate should above 10% likely reflects inadequate testing, meaning that testing should be increased to cast a wider net. (This Atlantic article explains why that is the case.)
As of now, much of the testing in the U.S. has focused on people with more severe symptoms. Mildly symptomatic people generally can’t get tests and asymptomatic people aren’t getting tested at all.
If we begin by assuming that all 30,000 patients identified every day in the U.S. have serious enough symptoms to be tested and that approximately 20% of patients with COVID-19 have serious symptoms, we estimate that this roughly implies that there are 150,000 new cases daily in the U.S right now, as the curve is flattening.
There is a different way to estimate the number of cases.
We can start with mortality projections from the Institute for Health Metrics and Evaluation (IHME). On April 12, when we built our estimates, IHME projected that on May 1 there would be about 1,150 deaths due to COVID-19. Given that the best estimate of the case fatality rate (CFR) of COVID-19 is around 1% and that on average there are two weeks between infection and death, this would mean that those 1,150 deaths arose from 115,000 new cases today. If you think the CFR is lower, that means there were many more cases on April 15. So, somewhere between 115,000 and 150,000 new cases per day. This is surely a large underestimate (CFR may be below 1% and we are likely undercounting deaths).
If we wanted to catch most of the 150,000 new cases, how many tests would we need? It depends, of course, on how effectively we can target those tests to those who are likely to have COVID-19. 150,000 new cases represents a tiny fraction of the U.S. population. We want to target our tests toward those who are likely to have it.
We also want to estimate how many tests we might we need as we enter into May, when many states are going to try to reopen. Using the IHME model, we estimate that there will be approximately 580 deaths due to COVID-19 on May 15 (remembering that median length from infection to death is about two weeks). Applying that same case fatality rate, we can estimate that on May 1, two weeks prior to these deaths (median time from infection to death is about two weeks),  there were approximately 58,000 new cases.
Again, those assumptions may be too low, but they are the best we have. We suspect that the number of cases in the beginning of May will be much higher because the number of true deaths will be higher than 580 on May 15 (or because we learn that CFR is much lower than 1%).
A high positive rate in testing suggests that many positive cases are being missed while a low rate suggests that there is a sufficient margin and enough individuals are being tested to capture all those who are infected. South Korea was able to use its testing/tracing strategy to get its outbreak under control, and they had a test positivity rate of 3%—so, they had cast a wide net.
What if we took South Korea’s approach? We would need to be testing 5 million people a day today and about 2 million people a day on May 1.
Right now, we are testing 150,000 people a day and have not been able to make that number budge for weeks.
We can’t do what South Korea has done. Our testing capacity is nowhere that good.
But what can we do? Can we do what WHO and others have recommended – reach a 10% test positivity rate operating under the assumption that we are not missing too many cases? That would mean that today, we would need to be testing about 1.5 million people per day and by May, we may need to be testing only about 580,000 people per day. And, of course, the more we can test, the more we can be confident that we are capturing all the infected people in the U.S.
There is no guarantee, however, that a 10% positive rate will ensure most cases are being identified.
So, let’s think about a different approach. Let’s assume that there are 58,000 positive cases and that about 75% have at least mild symptoms. That leaves about 45,000 new symptomatic cases of COVID-19 on May 1. The CDC Historical Influenza Surveillance numbers indicate that, on average, in early May in non-COVID-19 years, there are approximately 40,000 cases per day of influenza-like illness (ILI). If we add these to the 45,000 new symptomatic cases of COVID-19, then there are 85,000 individuals in the U.S. on May 1 with new influenza-like symptoms.
We should begin by testing all of them. That sounds good.
This strategy only captures people with ILI who seek out medical care and will miss all the asymptomatic people and those with symptoms who don’t seek care.
However, using this approach, we could then test contacts of all the positive cases. This strategy is likely to identify many of the asymptomatic and mild cases.
How many contacts?
Well, there is no gold standard for the number of contacts. During a shelter in place, you only need a few. But during normal times, some studies have estimated that the average number of contacts that must be identified and tested is 19. If we take a middle ground (when we open up again, we will maintain some social distancing), we might just need to test 10 contacts for each confirmed positive. This adds an additional 450,000 tests performed as part of contact tracing to the 45,000 COVID-19 symptomatic patients already confirmed. Adding this to our 85,000 tests of symptomatic individuals we find that as we enter May, we would need 535,000 tests, very similar to what we estimated through the method of pursuing a 10% test positivity rate. And as we continue through the month, with many states opening up and case numbers rising, we will likely need more tests.
In every way we can think about this, 500,000 tests per day is probably too low. We are likely substantially under-counting deaths and, therefore, the number of cases is likely higher. If, on the other hand, a case fatality rate of 1% is too high, that means there is a larger pool of asymptomatic individuals who are possible carriers of the coronavirus. Even if the CFR is 0.8%, that adds a lot more cases. A 10% test positivity rate would miss a lot of infected folks – as would the approach to trace and test on average only 10 contacts per positive.
Therefore, while we estimate that we should be testing between 500,000 and 600,000 people per day; this is clearly on the low side. We have, in trying to make these calculations, consistently tried to go low – make assumptions that lead us to need fewer tests. That’s why we are likely underestimating the number of tests needed.
If somehow we are able to achieve these goals and do excellent testing/tracing, we might (just might) get an R0 less than 1 (a very hopeful goal)—over time, we may need somewhat fewer tests. But that is very hopeful, and all the evidence suggests that the number of cases will rise as social distancing is relaxed, even with rigorous testing and tracing.
And so, we will likely need many more tests. But, if we can’t be doing at least 500,000 tests a day during May, it is hard to see any way we can remain open.
Ashish Jha, MD, MPH, is the Faculty Director of the Harvard Global Health Institute and K.T. Li Professor of Global Health at Harvard University.
Dr. Thomas Tsai, MD, MPH, is a surgeon at Brigham and Women’s Hospital in Boston and an assitant professor and health policy researcher at the Harvard T.H.Chan School of Public Health.
Benjamin Jacobson is a research assistant at the Harvard Global Health Institute.
Get important news, updates, and statistics sent to your inbox by sharing your email address here.