One has to be careful about drawing conclusions from the any one data point about COVID-19.
Recent numbers have shown signs of stabilization, only to be followed by what looks like a relapse. The number of new cases reported daily, globally, of COVID-19 hit a high on April 12th, at 99,100, then declined to 74,800 cases per day, on Wednesday, April 22nd, but then shot up again on April 24th to a new high of 102,200, according to figures maintained by Johns Hopkins University.
In the U.S., too, data shows daily cases plateauing recently, only to be followed by a subsequent new high in daily cases.
As some U.S. States re-open their economies, and as other locations around the world consider ways to re-open, some researchers are making the case that arbitrarily removing measures such as quarantine, and other "non-pharmaceutical interventions," could quickly bring a resurgence of the disease.
In particular, a study by scientists at the Los Alamos National Laboratory, part of the U.S. Department of Energy, posted this week on the medRxiv pre-print server, argues that interventions have helped to lower the incidence of the disease, and removing such measures arbitrarily could prompt a very rapid return of the disease.
"The number of cases averted in two weeks of intervention will be regained in only one week," if all measures are completely relaxed, according to the paper, titled, "Decline In Global Transmission Rates Of COVID-19."
A prominent takeaway from the work is the need for increased testing to more swiftly detect patterns in the disease's spread.
The paper is authored by Ethan Romero-Severson of the Theoretical biology and biophysics unit at LANL, along with colleagues Nick Hengartner and Ruian Ke of that unit, and Grant Meadors of the Computational physics unit of LANL. The paper has not been peer-reviewed, which should be kept in mind in considering its conclusions.
In response to an enquiry from ZDNet, lead author Romero-Severson, speaking for his personal view, and not on behalf of LANL, summarized the key takeaway from the report as he sees it.
"There is evidence that global prevention methods are working," wrote Romero-Severson. "However, extensive surveillance and caution should be taken while prevention measures are modified as the disease can quickly return to pre-prophylactic levels."
The study found evidence that non-pharmaceutical measures helped flatten the curve. Of 51 countries studied, the report "found evidence that the transmission of COVID-19 is decreasing in 24 countries, i.e. the effective reproductive number is less than 1, using data up to April 16, 2020."
"This suggests that, despite the highly heterogeneous populations represented by these countries, the growth of COVID-19 outbreak can be reverted."
There's an asymmetry, however: The disease can infect people much faster than infections can be reduced through quarantine and other measures.
"We estimated that in countries with decreasing transmission, the rate of decrease is in general less than 0.1/day," the authors observed.
"Based on data from 8 European countries, the US and China, we previously estimated that in the absence of intervention efforts, the epidemic can grow at rates between 0.19-0.29/day."
"This means that the outbreak can grow rapidly and avert public heath efforts made if social distancing measures are completely relaxed," they conclude.
This is not the first work the Los Alamos scientists have done about the disease. They issued a paper on February 11th which concluded that the infection rate of COVID-19 was much higher than believed at the time. And on April 15th they issued a report arguing that the disease is "highly transmissible" without "strong control measures," regardless of the geographic or social particularities of the country.
In the current study, the authors calculated the impact of things such as social distancing by performing a simulation that matched statistical modeling to actual data.
The authors begin with the standard epidemiological model that's been used frequently this year, the "SEIR" model, where a set of ordinary differential equations govern the relationship of four "compartments" of individuals: susceptible, exposed, infected and recovered.
They then connect that model to data via a statistical process that "models the random movement from infection to being either counted as an observed case or counted as an observed death." They then fit the statistical model to the actual reported case and death numbers. If the model fits the data, it suggests the model is capturing something about the nature of the disease.
They compared their model to reported data for 51 countries, including the U.S., and found that "The model can capture the data well, with a few exceptions." Iran, Spain, and Italy are trickier to model, and they hypothesize that this is because the time between a person becoming infected and that infection being detected is particular to those countries in a way that they haven't yet approximated correctly, and that needs more study.
The paper is not a causal model of the disease, and so it cannot predict with certainty what happens from different preventive measures. It is conceivable the infection rate, were lockdown to be removed, could be different from the infection rate before the lockdown was put in place. But Romero-Severson emphasized to ZDNet that the asymmetry between decline and growth, of whatever magnitude, is something basic to the model: prevention is slower than pandemic spread.
Given that asymmetry, the authors emphasize the information lag in detecting COVID-19. If the disease were to start rising again, it takes time for that to become apparent.
"Changes to policies should be made slowly because the signal of changing transmission can take weeks to fully propagate into current data streams as a result of the long lag between infection to case confirmation (as we estimated to be approximately 2 weeks)," the authors write.
Given the asymmetry, and the information lag, the authors observe that widespread testing is important to shorten that lag.
"We can shorten the delay by contact tracing and testing and employing a broader public health workforce," Romero-Severson explained to ZDNet, speaking, again, for his personal views on the work.
"The delay comes from two sources: time from exposure to testing, and time from positive test to reporting," he explained. "Testing more people and employing more people to get the reports processed would shorten the delay."
One possible implication, not addressed in the paper, is that non-pharmaceutical measures such as social distancing may have to be maintained at least intermittently, rather than relaxed entirely, as was suggested in a report by Harvard researchers in late March.
As with all models of COVID-19, it's important to remember that the model is a construction that is trying to approximate what goes on in the pandemic. It is not a proven assessment of what is going on, but rather an effort to use statistics to come to some understanding.
"I think that the most important message to get out is that we are in uncharted territory," Romero-Severson told ZDNet. "The fact that we are learning new things about this disease nearly daily is evidence of that."
"This kind of intense, real-time science is new for everyone and it's hard to resist our human need for (or comfort in) certainty," he added. "But we don't have that luxury now."