Every day, the estimated death toll for Hawaii’s coronavirus outbreak fluctuates.

On Monday, it was 390 through April 4 of this year. As of Friday, it’s 372.

That’s according to the University of Washington’s Institute for Health Metrics and Evaluation, which updates a state-by-state analysis daily. The picture sketched for Hawaii is worrisome: the virus is expected to peak around May 3 and the state could be short 125 intensive care unit beds.

As Hawaii’s tourism industry grinds to a halt, record job losses mount and the COVID-19 virus spreads, many are wondering how bad it will get and when the coronavirus crisis will end.

According to the UW model, the answer — for today – is early May and late July. Ali Mokdad, a professor at UW, said he and other researchers hope their models help policymakers prepare for the worst of the pandemic by procuring necessary ventilators and readying hospital beds. To the public as a whole, the numbers offer something that feels concrete in a world that feels surreal and uncertain.

But estimating the trajectory of COVID-19 is an imprecise science.

Poor data collection means many analyses are wildly different, explains the news site FiveThirtyEight, in a recent story on the complexities of modeling the virus trajectory. Fatality rates, for example, might change depending on a community’s demographics. Limited testing and lack of clarity about how many people are symptomatic versus asymptomatic also muddies the waters.

Still, the public pores over the models, asking for more data, making their own graphs, grasping for a way to wrap their heads around what’s happening now and what could possibly come next.

As pressure grows to understand the virus, the universe of COVID-19 modeling is becoming more varied and growing.

Researchers at Stanford created an interactive tool that analyzes how social distancing and quarantine could affect the pandemic. Modeling from the University of Pennsylvania aims to help individual hospitals plan for an influx of COVID-19 patients. COVID Act Now offers county-level estimates of expected hospitalizations with and without interventions. Estimates of potential COVID-19 deaths from Imperial College London helped spur President Trump to take action.

Just this week, the White House announced a new estimate that between 100,000 to 240,000 could die in the U.S. from the virus. But already, experts are questioning what those estimates were based on.

The one thing everyone seems to agree on is it’s going to get much worse before it gets better.

“I would expect that you would see an increase in cases in the next few weeks at least or possibly longer,” Hawaii’s Department of Health director Bruce Anderson said Tuesday. “I hope it isn’t going to increase dramatically, but we will probably see … a significant increase in the curve over the next four weeks.”

Building Assumptions Into Projections

Every model is widely uncertain and relies on a lot of assumptions, says Victoria Fan, an associate professor of health policy at the University of Hawaii. Four years ago, she co-authored a paper estimating the cost of a pandemic, both in lives and for the global economy.

When it comes to COVID-19, research is changing rapidly about the virus and what interventions are effective.

“Pandemics are highly unpredictable. The basic facts about this disease we’re still debating,” she said.

Researchers may compare COVID-19 to previous pandemics like the Spanish flu, but she notes so much has changed — technology, better mortality rates, more wealth and better education may improve the world’s reaction to this crisis. Yet more urban density, increased travel and more elderly people add more risks.

Nick Redding of the Hawaii Data Collaborative says he likes the UW model’s state-by-state projections because it’s based partly on hospital data, which he sees as a stronger metric than infection rates.

Still, “there’s as much art as there is science to crafting these models because there are too many parameters where these assumptions have to be made,” Redding said. 

Mokdad from UW says their model used data from the American Hospital Association and also analyzed mortality rates, looking at what happened in China and South Korea. But data gleaned from other countries may be suspect: the New York Times reported this week that China may be downplaying the degree of the pandemic.

Stanford’s study also looks at hospitalizations and deaths and relies on a framework called SEIR that involves categorizing people based on whether they’re susceptible, exposed, infectious or recovered, said graduate student researcher Mallory Harris.

Harris says the research concluded that any sort of social distancing will bring down the number of cases and delay the peak, giving communities more time to prepare. But communities may have to adopt the “light switch method” of implementing social distancing measures intermittently until a vaccine is found.

Unlike the UW model, Stanford’s doesn’t make numerical estimates about how many people will die due to the limited data, Harris says. She emphasizes how well people adhere to social distancing can drastically change what will happen.

“We can make guesses and some of our guesses are really good. But it’s only going to be as good as the data that we have and a lot of it is dependent on what actions we end up taking collectively and how strongly people are following these shelter in place orders,” she said, adding that another factor is the availability of testing.

“Modeling diseases is a little different from modeling the weather because the weather is going to happen. But with diseases, there are things we can do to make things better but it depends on how a lot of people act.”

Limited Data In Hawaii

In Hawaii, the state has confirmed more than 300 infections and three deaths. Compared with national estimates, mapping the trajectory of the virus in Hawaii is particularly tough because of limited data, says Tim Brown, a senior fellow at the East-West Center who specializes in infectious disease and behavioral epidemiology.

“Right now any models that are being done are relatively ad hoc,” he said. “We’d probably need a more complete picture of what the situation is in the community which we don’t really have right now because the testing is all with symptomatic people.”

It will take another two to three weeks of data collection and more information about community transmission before the numbers will be significant enough to show a trend, Brown says.

Hawaii-specific data is necessary also because of how the state differs from others.

Population density, multigenerational households, an older population and wide income distribution combine to make the state vulnerable, Fan says. But she says the state’s general adherence to social distancing helps.

Mokdad at UW says Hawaii has an advantage because infected people can’t simply drive across state lines. But the state suffers from limited infrastructure capacity and hospital infrastructure tends to be concentrated in Honolulu and may be needed in more rural areas like the neighbor islands.

“In Washington and California, there’s more beds than what the state needs,” he told Civil Beat. “Hawaii doesn’t have that capacity.”

Gov. David Ige argued Tuesday that the UW model doesn’t do a good enough job of taking into account Hawaii’s travel quarantines and that Hawaii has too few cases to model off of. It’s unclear on what if any analysis state officials are basing their current understanding of the virus trajectory in Hawaii.

Still, the UW analysis suggests that Ige’s policies regarding social distancing in Hawaii are working, Mokdad says. A week ago, the day with the highest average deaths from COVID-19 was expected to hit in April. Now that’s been pushed back into May and fewer people are expected to die.

He’s seen a similar trend in California where the estimated death toll by Aug. 4 dropped from more than 6,000 to fewer than 5,000. To him, that’s encouraging.

“If we stay at home all of us for two weeks as a country, things will get better fast,” he said.

Hawaii companies are watching the modeling closely as they try to figure out when things could get back to normal.

Right now, the UW model suggests the curve flattens out in July, but it’s a moving target. Jim Kelly, spokesman for Hawaiian Electric, says the utility has seen a big drop in electricity usage as large hotels close and hopes that things will get back to normal by August or September.

But even if Hawaii gets control over the virus, the tourism economy may not go back to normal if the pandemic continues to rage elsewhere.

Fan from UH says some of the best models may be ones that the public will never see.

“The modeling world is highly proprietary,” Fan said, adding that it’s typical for insurance companies in particular to guard their models closely. “That’s very common because people make these very sophisticated models and that’s how they make money.”

She’s glad several models like UW’s are now public but hopes people understand their complexity.

When it comes to COVID-19 models, “don’t take it too seriously but don’t dismiss it either,” Fan said.