Dr. Deborah Birx, White House coronavirus response coordinator, discussing projections in March from University of Washington researchers.
Photo: Alex Brandon/Associated PressThe near doubling of coronavirus death predictions in a closely followed model this week underscores a frustrating reality for officials weighing how and when to reopen society: Many basic facts about the new coronavirus remain unknown.
Epidemiologists have created many computer models to predict surge capacity in the health-care system and guide policy-making. These seek to predict how many people might be infected, how many will die, and when and how transmission might slow or speed up.
But the models are only as good as the underlying data and knowledge about the disease. Models are based on assumptions and estimates and shift with new information, often because of our own changing behavior but also because the scientific understanding of this newly emerged virus is still evolving. Researchers have strained to pin down basic bits of information about the disease, such as its infectiousness. Undercounted infections and deaths have blinded public-health authorities and modelers alike to the full scope of the pandemic.
That uncertainty plays out in the range of predictions from different modeling groups. As of the end of April, academic models projected anywhere from about 70,000 to nearly 170,000 Covid-19 deaths in the U.S. by mid-May, according to data compiled by the Reich Lab at UMass-Amherst.
On Monday, closely followed and scrutinized projections by University of Washington researchers nearly doubled to about 135,000 dead by August, based on the early easing of social-distancing measures and mathematical changes to the model from the university’s Institute for Health Metrics and Evaluation. The model faces increasing criticism from epidemiologists. The modelers contend that some of the critics have not been following all of the updates.
But the many projection models are valuable to public-health officials and policy makers, health experts say, especially when they are looking at the different projections collectively. The overall message, say epidemiologists: Social distancing has been working, but the virus likely isn’t going away soon.
“They are only as good as the underlying data. And the underlying data is flawed in a number of ways,” Jennifer Nuzzo, a senior scholar at the Johns Hopkins Center for Health Security, said at a media briefing last month. “That said, we can’t throw out the models. We still need to make decisions, and having tools to guide the decision-making is helpful.”
Models come in a variety of different flavors, but two of the major methods at the forefront of mapping the Covid-19 pandemic involve techniques called curve-fitting and SEIR modeling.
Curve-fitting, previously the method of the frequently cited IHME model from the University of Washington, looks at world-wide data on Covid-19 deaths and interventions already available—whether it is from Wuhan, Lombardy or New York—and attempts to match up the U.S. predictions with the patterns it sees elsewhere. When new data comes in, or social-distancing measures change, the model adapts.
The IHME model, however, has been criticized by other infectious-disease experts and epidemiologists, who said that it doesn’t take into account the dynamics of the disease and that it is likely too optimistic.
Modelers at the University of Washington originally picked the curve-fitting model based on deaths, partially because there was so little information on other variables, such as infections and transmissibility, said Ali Mokdad, a professor of health metric sciences at IHME.
But on Monday, the group updated their method and created a hybrid model, a combination of curve-fitting and SEIR, as the focus of policy makers has shifted from wanting to understand surge capacity and the timing of the peak to questions about loosening lockdown restrictions.
The SEIR method—short for “Susceptible,” “Exposed,” “Infected,” “Recovered”—is the more common epidemiological approach. The model places people into those categories and then simulates transmission based on the biology of the virus, along with many different variables, such as personal-interaction rates or potential seasonality.
“People in these different buckets are interacting with one another, and every time a susceptible person bumps into an infected person, there’s a probability of transmission,” said Ron Brookmeyer, a biostatistician and dean of the University California, Los Angeles Fielding School of Public Health. “This plays out in this simulation over time, day by day, and you can see the infection spreading.”
But public-health authorities still don’t know just how many people have been infected, a key number in figuring out fatality and hospitalization rates, as well as how many people are still susceptible. Death and case counts have often been underreported or lag other data, and aren’t measured consistently by different public-health authorities. Mathematically representing and predicting human behavior, such as how well people adhere to social distancing, presents another challenge.
The modelers at Columbia University, for instance, make their projections with three different social-distancing scenarios: 20%, 30% or 40% reduction of person-to-person contact in U.S. counties with at least 10 cases, and assume that social distancing increases with more new cases and remains in place indefinitely.
Each study has an R0 estimate ( | ) and 95% credible interval ( )
Some modeling groups use mobility data from Google and Facebook and cellphone providers.
And since there are still so many unknowns about SARS-CoV-2, even some of the most basic inputs can vary, including the estimate of the virus’s infectiousness. This number, called “R0” (pronounced “ARE-naught”) by epidemiologists, represents a simple idea: the average number of people that each infected person infects if everyone in a population is susceptible, as is believed to be the case with Covid-19.
A value less than 1 means that an outbreak is headed for control. Values that are just slightly higher can indicate significant spread. Studies suggest that the 1918-1919 influenza pandemic carried a value of about 1.8, the 1957-1958 influenza pandemic, about 1.65, and the 2009 influenza pandemic, about 1.46.
For Covid-19, estimates of this key value before social distancing was imposed ranged from 2 to 6.5 in 20 studies reviewed by The Wall Street Journal.
Estimates of infectiousness also vary because the rate itself varies by place. Early data came from Wuhan, China, a densely populated city of 11 million residents that is in the center of an urban region of 19 million. By comparison, most Americans live in the suburbs of midsize cities.
STAY INFORMED
Get a coronavirus briefing six days a week, and a weekly Health newsletter once the crisis abates: Sign up here.
Once a region begins to “flatten the curve” with extraordinary behavior changes, slowing the rate of transmission, a pure value for R0 under normal circumstances becomes harder to calculate. A team at the University of Virginia’s Biocomplexity Institute, analyzing just Virginia’s cases and controls, calculated that the state’s effective R0 dropped from 2.2 before March 15 to 1.1 after March 15, as three increasingly tough executive orders by Gov. Ralph Northam took hold between March 12 and March 24.
In addition to the variability of data points like R0, how each group of modelers decides to handle the inputs influences the outcome of the model.
“If you get really down deep into all of this math, they all make assumptions about the way the numbers behave,” said Jason Christie, a pulmonologist and epidemiologist at the University of Pennsylvania, where researchers created a model called Covid-19 Hospital Impact Model for Epidemics, or CHIME.
The value, say epidemiologists, is in looking at the models collectively to estimate what health-care systems might need in the near future and to gauge the impact of policies.
“We would have had much larger epidemics and deaths, if it wasn’t for the prime determinant of the trajectory, which is social distancing,” Christopher Murray, IHME director, said at a press briefing on Monday. “In an era where those mandates are being relaxed, people should be aware that the risk of infection is still there.”
Write to Brianna Abbott at brianna.abbott@wsj.com
Copyright ©2020 Dow Jones & Company, Inc. All Rights Reserved. 87990cbe856818d5eddac44c7b1cdeb8
"behind" - Google News
May 06, 2020 at 06:22AM
https://ift.tt/3cazyxo
The Tricky Math Behind Coronavirus Death Predictions - The Wall Street Journal
"behind" - Google News
https://ift.tt/2YqUhZP
https://ift.tt/2yko4c8
Bagikan Berita Ini
0 Response to "The Tricky Math Behind Coronavirus Death Predictions - The Wall Street Journal"
Post a Comment