Hospitalization data signal that COVID-19 has been contracting for weeks. It is time to plan for a less tumultuous fall.
In early April, I published a series of data analyses demonstrating that California had flattened and then bent the hospitalization curve. Those conclusions proved to be correct, although it took another month for the State to acknowledge its success and begin modifying the stay at home order.
For the past several weeks, it has been “deja vu all over again.” In the midst of widespread concern over COVID-19 case counts, the best available data – hospitalizations – once again signal that the State has altered the trajectory of the virus. There is no question that the virus was spreading rapidly during the first half of June. But shortly after the Governor issued his statewide mask order, the data show that trajectory of the virus flattened. By early July, the virus was contracting, albeit unevenly, with apparent slowing during the week and spikes on the weekends. But the overall downward trend in the data is unmistakable. With the impact of the Governor’s last round of business closures now appearing in the data, it is increasingly obvious the virus is once again in decline.
This is very good news for California. But rather than celebrate, we need to take stock. We need to learn critical lessons from the way we have managed the pandemic over the past three months so that we can do better in the weeks and months ahead.
The trajectory of the virus changed after the statewide mask order was issued.
The best available measure for assessing the recent trajectory of the virus is hospitalizations. (Fatality data is the best measure of spread, but it reflects the trajectory of the virus roughly a month earlier). To correctly interpret hospitalization data, it is critical to recognize that there have been several reopening “phases” based on the amount of economic activity permitted in various counties as well as the issuance of the statewide mask order on June 18th. We must therefore consider the data phase-by-phase (something the State does not do in its reporting) to understand the actual trajectory of the virus since reopening. Building on my prior analysis, I am focusing on the six most recent reopening phases:
Phase 3: The significant statewide increase in hospitalizations began in this phase. It reflects the economic and other activity from June 4th to June 11th, including increasing Black Lives Matter protests.
Phase 4: This brief phase reflects the major economic reopening on June 12th as well as the large Black Lives Matter protests that occurred on June 13th and 14th.
Phase 5: This phase reflects economic activity before the statewide mask order was issued.
Phase 6: This phase reflects the impact of the statewide mask order, which was issued on June 18th, as well as the reopening of personal care services in many counties across California on June 19th.
Phase 7: This “snapback” phase corresponds to three rounds of closures that began on June 28th (bars), expanded on July 1st (indoor business closures), and included beach closures over the July 4th weekend.
Phase 8: This snapback phase began on July 14th and corresponds to the fourth round of business closures (personal care services) announced on July 13th. It also corresponds to a period where the State changed its hospital data reporting. On July 23, the State issued a cryptic note stating that it lacked “historical data from 39 non-reporting facilities” in that day’s “update, leading to lower numbers.” By July 30, the note had changed to acknowledge a lack of data from “8” non-reporting facilities. At this point, it is not clear what data, if any, is not reflected in the hospitalization numbers reported over the past few days, why the number of non-reporting facilities has changed, or when the State will finish updating numbers from these facilities. I will update this Post if the data related to Phase 8 change in a material way.
Hospitalizations typically occur 7-11 days after an infection. The chart below maps daily net total hospitalizations (using a 5-day moving average) to their various phases using a 9-day lag (e.g., Phase 3, which corresponds to June 4th to June 11th, produces hospitalizations from June 13th to June 20th).
As this chart demonstrates, the hospitalization curve was effectively flat by July 19th. It has been trending downward since July 23rd. Although it is not as obvious from this chart, this flattening actually began when the Governor issued the statewide mask order.
To illustrate the impact of the mask order and subsequent closures more clearly, it is helpful to consider the daily change in net new hospitalizations.
When examining the daily data, it is important to keep in mind that a spreading virus would increase new admissions by an increasing amount each day (e.g., 50, 55, 60), a flat virus would produce the same number of new admissions each day (e.g., 50, 50, 50), and a contracting virus would produce a declining number of new admissions each day (e.g., 50, 45, 40). Because the chart below reflects net daily new admissions (new admissions – new discharges), it is also important to recognize that there may be short periods of time (depending on the number and duration of prior admissions) when net admissions may not mirror the actual trajectory of the virus (i.e., new admissions). Over a longer period of time, however, a sustained decrease in net daily new admissions signals the virus is contracting (as we saw from late March into April).
The Chart below maps daily net hospitalization admissions to their relevant Phase nine days earlier:
The increase in daily net admissions peaked on June 30th. Daily net admissions have been dropping ever since. So what does this sustained drop signal? Initially, the mask order (Phase 6) substantially reduced the rate of spread even though it was accompanied by another round of reopening. Over time, it effectively flattened the trajectory of the virus. This impact was visible in the data beginning July 1st, just as I previously predicted, when I urged policy makers not to push the panic button. This, in and of itself, is very good news for California. So please keep wearing your mask and being sensible on the weekends. These efforts are making a difference.
What is equally clear is that the virus began contracting, albeit unevenly, sometime in late June, producing the downward trend in hospital numbers that began in early July. Following the 4th of July weekend, the substantial declines in net daily admissions, followed by brief spikes, have repeated themselves on a continuous basis, creating an unmistakable overall downward trend in the trajectory of the virus (even without the data from Phase 8). This is the same pattern we saw from late March into early April. The virus has been contracting.
The snapback measures implemented by the Governor in Phase 8 have almost certainly reinforced this downward trend. The daily net hospitalization numbers have been negative for more than a week.
This drop in hospitalizations is occurring at the same time as a substantial increase in COVID-19 fatalities. This is not evidence that the virus is currently expanding or the hospitalization data is unreliable. Rather, the fatality numbers we have seen over the past week most likely correspond to the period where daily increases in new hospitalizations peaked at the end of June. We should expect to see fatalities start decreasing in the days ahead, just as daily new hospital admissions began decreasing in early July.
What does the current decline in the virus mean for California going forward?
We are three months into our effort to live with the virus. It has been a tumultuous roller coaster ride for the last two. With hospitalizations declining, now is the time to take stock. There are several important lessons we should learn from our effort to manage the pandemic as we prepare for the fall, the reopening of schools, and the return of more working parents to the workforce.
First, we can accommodate a significant amount of economic activity without increasing the spread of the virus.
Second, we can flatten and bend the trajectory of the virus with sensible public health measures, social distancing on weekends, and targeted business closures. We do not need another stay at home order.
Third, if we are going to live with the virus until we achieve herd immunity or widespread vaccination, our decision making and policy making must be driven by the best available data. For the purpose of assessing the recent trajectory of the virus, this is not our current testing data – case counts or positivity rates.
Our current testing protocols were not designed to measure community spread. Instead, they were developed to facilitate “medical evaluation of persons with symptoms of COVID-19” as well as “efforts by public health agencies and essential employers to prevent and control the spread of COVID-19.” As a result, there are three challenges with attempting to repurpose testing data to measure the spread of the virus across the State. First, the people prioritized for testing are not a cross-section of our statewide population. Instead, they are a narrow cross-section of the people most likely to have the virus. We currently do not prioritize testing for the following people unless they have likely been in contact with an infected person: children, people who work in professional services, retirees, stay-at-home parents, bankers, postal workers, transportation workers, or construction workers (and the list goes on). Second, even if the prioritized high-risk groups somehow represented a valid cross-section of the state population, the State does not randomly select members of each group for testing. Instead, it is necessarily looking for positives (to prevent and control the spread) and devoting testing resources to those most interested in being tested because they self-identify as high-risk. Third, even if these problems could be overcome, testing eligibility has varied during the pandemic and there is no system for ensuring consistency in testing among each eligible demographic over time, which makes the comparison of daily results problematic. We should, therefore, ignore our current testing data (and the headlines it generates) as a measure of community spread.
Not only is the testing data unreliable as a measure of spread, but the lag in reporting test results often makes testing data no more timely than hospitalization data. If different data sets are providing the same historical perspective, we should rely on the one that is demonstrably more accurate – hospitalizations. Although early in the pandemic the State may have been forced to rely on symptomatic case counts as its only potential measure of spread in the absence of timely hospitalization data, that period ended in March. Today, there is simply no justification for favoring our current testing data over hospitalization data when assessing the trajectory of the virus.
This is not to say that testing should not be performed. To the contrary, it is essential for improving workplace safety, identifying hot spots, and contact tracing. And random testing across the entire statewide community, which the State does not perform, would be very useful for assessing the trajectory of the virus.
But in the absence of random community-wide testing, the State should stop using its current testing data for the purposes of monitoring counties (case counts per capita and positivity rates) and deciding whether to open or close businesses. Instead, these decisions should be based on hospitalization data. Ideally, the State would track and publish daily new admissions for confirmed and suspected COVID-19 patients as standalone metrics to measure and monitor the trajectory of the virus. These data are the best near-real-time measures of viral spread.
Fourth, the State needs to be more disciplined and patient in making opening and snapback decisions. The initial impacts of these policy decisions are not visible in hospitalization data for 7-11 days, and their full impact is often not visible for 15 days. When the State makes decisions in shorter intervals, it risks making mistakes. For example, we reopened personal care businesses on June 19th. By June 21st, however, the data clearly demonstrated that the openings on June 12th and the Black Lives Matter protests on June 13th and 14th had caused the virus to continue its surge. More recently, by July 1st, the data showed that the mask order had significantly slowed the spread of the virus. But before this impact could be fully demonstrated, the Governor imposed 3 rounds of closures. Going forward, the State should ensure 14-21 day gaps between opening and closing decisions so that the actual impacts of a decision can be accurately measured before the next decision is made and implemented.
After a tumultuous start to summer, we have the breathing room and time needed to plan for a better fall. If we learn the right lessons from our initial efforts to manage the pandemic, we should be able to chart a smoother course as we learn how to best live with the virus.