Bay Area deaths from the new coronavirus would at least double by June and could multiply 10 to 36 times under modeling the state of California uses to project the pandemic’s potential toll.
It may sound astounding that in 10 Bay Area counties, where 241 COVID-19 deaths were reported as of Friday, the state model projected as many as 8,727 by the beginning of June. But how could that be?
Experts warn such projections can be wildly misleading. Even so, the state model’s dire outlook offers a window into the thinking of Gov. Gavin Newsom, the first to order a statewide stay-home order to check the virus’s spread, as well as Bay Area officials who issued the first county-level shelter-in-place order. It may explain their reluctance to say when those restrictions will end.
“It gives you some idea what the potential results are,” said Santa Clara County Executive Jeff Smith, who has a medical degree and whose administration led the Bay Area push for a regional shelter-in-place order in mid-March. The state model is one of three the county uses to guide its response. “All of them are helpful in the sense that they make it clear that the virus is spreading in a way that is very dangerous.”
But such models can be controversial if their projections appear to be off the mark. Critics say that can lead officials toward a response that doesn’t fit the threat and undermine public confidence.
And the modeling data the state shared with counties April 14 show just how suspect such projections can be.
“This is just way, way, way off from the observed data,” said Dr. George Rutherford, professor of epidemiology and biostatistics at the UC San Francisco, after reviewing the data. He noted that the model’s projections for deaths in San Francisco by June range from 59 to 960, with a median of 248. As of Friday, there were 21 reported deaths from the coronavirus in the city.
“Projections are always difficult to do,” said Rutherford, adding that UCSF’s own modeling for short-term projections of hospital bed days proved reliable enough for them to send staff to help with outbreaks in New York and the Navajo Nation, the largest U.S. reservation. “I think it’s important to get this stuff right. Overestimates I think are hard to justify.”
The state uses an open-sourced Johns Hopkins University model that Smith said is widely used and respected. Rodger Butler, spokesman for the California Health and Human Services Agency, said the state “reviewed multiple models to inform our decision-making but only used this one.”
The state model used data as of March 24 and factored in the statewide stay-home order and other distancing and hygiene measures in place since March 20.
Asked whether the month-old projections are still relevant, Butler said that “California is focusing its response on the actual case and hospitalization data.” But he said the counties had requested the modeling data.
Among the parameters the state plugged into the model were an infectious period of 2.6 to 6 days, a transmission rate of 2-3 people per confirmed case, a mean incubation period of 5.2 days and a 1% fatality rate. They also assumed 10% of the infected would be hospitalized, 32% of that group would be admitted to intensive care and 15% of those would require ventilators to help them breathe.
If any of those prove to be off, it would affect the projection’s accuracy.
“Every model is based on presumptions you put in,” Smith said.
The state model’s projections span a wide range, and even at the low end appear to overstate hospitalizations, intensive care needs and fatalities.
At the statewide level, the model’s mid-range, median projection was that hospitalizations would reach 10,711 by April 12 and peak at nearly 44,000.
Actual confirmed and suspected hospitalizations April 12 were less than half that figure, 5,048, and sat Friday at 4,929, below even the model’s low-end projection of 7,506, with trend lines suggesting they already have peaked. Friday’s 1,531 cases requiring intensive care treatment were well below the low-end 2,885 peak projection.
The model projected statewide deaths by June would range from 2,521 at the low end to 39,713 at the high end. There were 1,529 COVID-19 deaths statewide as of Friday, in a week that saw the state’s biggest single-day jump.
California would have to see a 65% increase in deaths by June to reach even that low-end projection.
In Santa Clara County, which had 94 deaths as of Friday, the projected total by June ranged from 117 to 1,938. For Alameda County, which had 45 deaths as of Friday, the projection ranged from 119 to 1,977. For Contra Costa County, which had 22 deaths as of Friday, the projected range was from 80 to 1,311.
By contrast, in Contra Costa County, the top health official in late March said 2,000 to 14,000 could die by the end of the year.
The state’s high-end projection of Santa Clara County fatalities is similar to San Jose officials’ “best-case” forecast presented publicly in late March that 2,000 would die in the county by June, even with the shelter-in-place order. That prediction angered Smith, who said it was “absurd” and would erode public trust as people see far lower figures, which he said is a result only of restrictions that have prevented new infections.
“It scared a lot of people,” Smith said. “It gives people a false sense that things have gotten better, and that’s not true.”
Santa Clara County also uses a popular University of Washington model and a custom model from Stanford University, Smith said. The University of Washington model predicts California deaths will total 1,719 by June, fewer than the low end of the state projection.
But Smith said the recent discovery that a Santa Clara County woman who died Feb. 6 had COVID-19 changed the picture and throws earlier forecasts into question.
“There’s a large reservoir of virus ready to attack,” he said, “and a large susceptible population ready to be infected.”