To know when the peak has passed, we need better COVID-19 modeling
Because we do not have COVID-19 testing of every person in the U.S., or a random sample, experts advising federal, state and local authorities are working with an invalid set of assumptions based on test-positive cases and deaths. Community test-positive cases have not reliably evolved into hospitalizations. For the vast majority of COVID-19 hospitalized patients, the diagnosis is established by hospital laboratory testing. In many ways, a hospitalization for COVID-19 is the real epidemiologic measure that should be used as the yardstick for this pandemic.
As a result of flawed epidemiology, the most influential experts attempting to model the COVID-19 pandemic have been off the mark in terms of the timing and size of the peak surge. The Murray model, most heavily relied upon by the administration, predicted the peak surge for New York state to occur on April 8 and require 16,492 beds for hospitalized patients. Unfortunately, New York State does not report COVID-19 hospitalizations to the public. However, the actual peak number of newly hospitalized cases in New York City occurred on April 1, at around 1,700, which based on population proportions would yield about 4,000 new COVID-19 hospitalizations and a statewide bed demand of about 20,000 on that date. Hence, the model was reasonably on target for the bed demand, but about a week late on the “surge.”
Outside of New York, the predictive modeling has been even farther off the mark. For Texas, the Murray model predicts a peak occurring on April 29, with an estimate of 2,824 hospital beds needed. The Dallas County Department of Health and Human Services, which lists COVID-19 as an illness requiring mandatory reporting from hospitals, reported a peak of 36 new hospitalized cases on March 27, more than a month earlier than the Murray model. Based on Dallas County population as a proportion of the state, this would yield a total of 401 new hospitalizations in the state at the peak, with about 2,000 Texas beds occupied.
What are the consequences of predicting the surge a month too late? At present, the Kay Bailey Hutchison Convention Center is set up as a military field hospital with a contingent of local and military personnel waiting for a wave of patients that will never arrive.
To be fair, Colorado is probably the best example of the Murray model delivering the goods. The Murray model anticipated the “surge” in hospitalized patients on March 31 and estimated 637 beds would be occupied; the surge actually happened on April 2 with 128 new cases hospitalized, and at a 1:5 ratio of incident to prevalent hospitalized patients, the model was early by three days but within 100 beds of reality. In the case of Colorado, the Murray model was still useless because Colorado was reporting the hospitalizations on a daily basis as the model was updating its predictions. Hence, Colorado had no need for the model since the current reporting was so well done for state and local leaders to respond appropriately and on time with needed shifts in resources.
Colorado Public Radio reported that Kathryn Colborn, an associate professor in the Department of Surgery on the University of Colorado’s Anschutz Medical Campus, said “hospitalizations — rather than predicted cases — is the number she plans to watch most closely. Statisticians have to guess the total number of infected coronavirus patients based on the number of people who’ve tested positive — likely a fraction of the real cases. … Hospitalizations will provide a clearer signal.”
For these working examples it is obvious we continue to need the real-time hospital census and numbers of new cases admitted in order to react to this pandemic on a daily basis. Because we did not have uniform hospital reporting of cases from the beginning, we had to rely on modeling based on community test-positive and fatal cases, which for most of the country is not sufficient to identify when the peak number of new cases will, or did, occur — and is only partially useful for identifying the need for hospital beds.
As an epidemiologist, I can only present this alternative view and demand better science. The daily case rate of hospitalized COVID-19 patients, and proper epidemiology applied to that case definition, will yield much better predictions for federal, state and local officials. It is very important for leaders to know when the peak has passed. It is not too late for an executive order for all hospitals to report new COVID-19 admissions, and additionally report on prevalent hospitalized patients in the medical wards and intensive care units.
Why guess from models when we could have the actual numbers on a daily basis, and be much more responsive and prudent with our resources in managing this pandemic and its recovery?
Peter A. McCullough, MD, MPH, is vice chairman of medicine at Baylor University Medical Center and a professor of medicine at Texas A&M College of Medicine in Dallas. An internist, cardiologist and epidemiologist, he is the editor-in-chief of “Cardiorenal Medicine” and “Reviews in Cardiovascular Medicine.” He has authored over 500 cited works in the National Library of Medicine.