We have time to make analytically-sound public health decisions in California.
I began writing about COVID-19 out of frustration. In the early days of the pandemic, the reporting often placed panic-inducing headlines on inaccurate or incomplete data analysis. Sadly, history may be repeating itself. Over the past two weeks, the media has been citing increasing positive test results as evidence that the State has reopened too quickly and the virus is now spreading at such an alarming rate that we are nearing a crisis. Governor Newsom may share these concerns – over the past ten days, he has issued orders mandating masks in public and closing bars in certain counties.
As we adapt to living with the virus, it is critical that we base our public health policies on accurate and complete data assessments. If we were to do that today, we would see that our Statewide COVID-19 numbers are growing, but remain quite manageable. We would also see that the next week would provide us with the data we need to more accurately assess the true trajectory of the virus, the impact of the mask order, and the potential need for additional public health measures.
What numbers should we be using to monitor the trajectory of the virus? As I have stated repeatedly, the optimal approach to measuring the trajectory of the virus would be random testing of a sufficient number of people from different geographic areas at regular intervals. But we do not do this type of testing for active infections or antibodies.
What number should we not rely on? Daily positive test results. This number may not accurately reflect the spread of the virus for many reasons, including the fact that the volume of tests performed changes every day, the criteria for testing eligibility change often, and the mix of people being tested (e.g., hospital patients, symptomatic individuals, contact-traced individuals, asymptomatic individuals) varies every day and can impact the positivity rate. Indeed, depending on this mix, obtaining positive test results could be the functional equivalent of shooting fish in a barrel or trying to find a needle in a haystack. For these reasons, epidemiologists recommend tracking hospitalizations as the best measure of community spread, at least as an early warning mechanism. (Fatalities may also be a useful measure, but they can trail the date of infection by more than 30 days in some cases).
Unfortunately, the State does not report the most useful hospitalization metric – new daily admissions (although an incomplete record of this data is available through the State data portal). Instead, it reports a net hospitalization total each day, which is the result of new admissions minus daily discharges. For this and other reasons, changes in daily hospitalization totals signal, but do not precisely mirror, the rate of community spread. Nonetheless, they are the best available measure of the trajectory of the virus (and it is possible to model the relationship between net hospitalizations and viral growth rates to improve precision).
So what do the hospitalization numbers tell us about our reopening? For the State as a whole, they tell us to stay calm, recognize that the reopening went smoothly until June 6th, and be patient for another week before deciding whether to recalibrate public health policies (although this ship has started to sail).
To understand the data, it is critical to remember that hospitalizations reflect the spread of the virus several days earlier. The medical literature and my earlier modeling work suggest that most hospitalizations occur 7-11 days after a person is infected, and nearly all occur within 15 days.
Given that the State began reopening on May 8th, we should see a strong downward trend in hospitalizations until roughly 5/15 to 5/19. From that point forward, with a modest increase in the rate of spread, we should expect to see a much slower decline in hospitalizations, potentially followed by a leveling out or very gradual increase in hospitalizations as we expand our reopening.
As the chart below demonstrates, this is essentially what occurred during the first several weeks of our reopening. The steep decline in hospitalizations ended on May 16th – 8 days after the State reopened. From May 16th until June 16th, daily hospitalization levels hovered around 4500 – consistent with numbers from earlier in May and well below our peak of approximately 5800 on April 7th. Over the past two weeks, however, daily hospitalization levels have risen substantially and far more than we would have hoped. Why?
To correctly interpret the hospitalization data over the past two weeks, we need to take a much closer look at the data over the past 7 weeks and how it corresponds to different phases of the reopening.
Phase 1: Hospitalizations rose initially, but then slowly declined from May 16th to May 30th. This suggests that the virus continued to contract, albeit more slowly, during the initial phase of the statewide reopening, which lasted until roughly May 18th through May 22nd. During this latter 5-day window, many populous counties completed their attestation processes and began rapidly reopening more of their economies.
Phase 2: The increase in attestations on the eve of Memorial Day weekend, combined with another reopening round the following week, likely explains the subsequent rise in hospitalizations that began on June 1st and continued until June 8th. But hospitalizations then declined for 5 days. Interestingly, this rise and fall in hospitalizations suggests this may be pattern associated with major reopening rounds, particularly if they occur on Fridays and spur greater social interaction over the ensuing weekend. Indeed, my modeling strongly suggests there is such a pattern in the data going back to the beginning of May: Fridays through Sundays generate significantly more future hospitalizations than Mondays through Thursdays.
The George Floyd protests from May 29thth to May 31st appear to have had little or no impact on the spread of the virus. There was a miniscule increase in hospitalizations from June 6th to June 8th.
Phase 3: However, later rounds of the protests on June 6th and 7th likely contributed to the rise in hospitalizations that began on June 14th.
Phase 4: On June 12th, there was a substantial round of reopening across most of the State, which included hotels (for tourism), bars, wineries, gyms, and many other businesses. In addition, there were substantial Black Lives Matter protests on June 13th and June 14th. The combination of these events likely caused the significant increase in the hospitalization rate that began on June 19th and reached its greatest rate of daily increase on June 21st and June 22nd. My modeling supports this assessment.
Phase 5: With fewer protests since the 14th, the increase in the hospitalization rate from June 23rd to the present is largely the result of the ongoing economic reopening that began on June 12th.
The key question, therefore, is whether hospitalizations will flatten starting on June 24th and then decline as they have done with prior major reopening rounds. To date, the curve has not flattened nearly as much as it has after previous reopening rounds. This needs to be monitored carefully. In addition, we do not yet know whether the statewide order to wear masks in public will have any discernable impact on hospitalizations. Any evidence of its impact should start appearing in the data between June 26th and June 30th. Ultimately, by July 5th, we should have a far more accurate understanding of the trajectory of the virus.
What should we do while we wait to see the hospitalization data over the next week?
First, we should recognize that reopening has been largely successful. The reopening process had little impact on the trajectory of the virus until June 6th. This means that the State can handle a substantial amount of economic and other activity without meaningfully increasing the spread of the virus. This is very good news.
Second, we should not push the panic button. Statewide, we have ample hospital capacity even with the increases over the past two weeks (although I recognize certain counties are much closer to reaching baseline capacity and utilizing surge capacity). The State can, therefore, continue with our current level of reopening while it tracks new hospitalization data over the next week.
Third, we have plenty of time to make analytically-sound public health decisions. If the hospitalization rate does not level off, we will still have several weeks before the State will exceed its baseline hospital capacity. Under this scenario, our public health professionals would have ample time to develop and implement narrowly tailored “snap back” orders that further protect the most vulnerable and alter the activities most likely driving the growth of the virus.
In this regard, the mask order stands as a sensible and minimally-disruptive measure to slow the spread of the virus without altering the course of our reopening.
By contrast, it is hardly clear that the Governor needed to close bars in several counties this week. And although it is a fairly narrow response to the increased rate of transmission, it is also unclear whether there is any analysis to support the view that such closures will meaningfully reduce hospitalizations given the likely differences in the demographics of bar patrons and many of those most at risk for hospitalization from COVID-19.
It is clear, however, that we do not need to return to anything close to a full stay at home order under any foreseeable circumstances. Our experience from May 8th to June 6th proves this would be unnecessary to control the spread of the virus.
Fourth, please be prudent and protect yourself and others when in public. Living with the virus today does not mean returning to the way things were last November. We still need to take measures to ensure the virus does not spread at a rate that would overwhelm our healthcare system.