SEATTLE — After remaining fairly constant through the summer, novel coronavirus infections and deaths in the United States are likely to begin climbing again in September, marking the start of a second wave of the epidemic, according to a model from the University of Washington that has been widely cited but also criticized.

The UW’s Institute for Health Metrics and Evaluation (IHME) projects the national death toll from COVID-19 could reach nearly 170,000 by Oct. 1, with an uncertainty range of 133,201 to 290,222. That’s 57,000 more deaths in the coming months, in addition to the approximately 113,000 Americans killed by the virus so far.

The group’s forecast for Washington state follows a similar trajectory, though at a much lower level, with a projected total of 1,475 deaths by Oct. 1. As of Thursday, the state death toll stood at 1,194.

While the projections present one possible outcome, what actually happens can be altered through individual action like wearing masks and reducing contact with others, IHME Director Dr. Christopher Murray said Thursday in a briefing.

“We hope to see our model proven wrong by the swift actions individuals and governments take to reduce transmission,” he said.

But Murray said it’s possible that, in some parts of the country, schools that reopen in the fall will have to close, at least temporarily, if transmission spikes later in the year. Some cities may also reimpose lockdowns if serious illness begins to soar.

Despite the ongoing outbreak in the Yakima area, trends in Washington remain favorable, with cases in Snohomish and King counties generally leveling off, Murray said. But as the U.S. surpassed 2 million confirmed infections this week, many other states — including Arizona, California, Florida and Texas — are seeing cases climb. The upswings seem to be related to loosening of restrictions followed by increased mobility and social mixing, Murray said.

“We are seeing, across different states, pretty different epidemics,” he said.

There’s no evidence yet of a jump in new cases related to ongoing demonstrations in support of racial justice, Murray added. It’s possible their effect will be small, compared to the overall increase in mobility and contacts, or it might just take another week or so before infections are diagnosed.

Ruth Etzioni, a biostatistician at the Fred Hutchinson Cancer Research Center, urged against placing too much faith in any model prediction. “Never hang your hat on the exact numbers,” she said. “The basic message is that cases are going to go up — and we know they are already going up (in some places).”

Another model, from independent data scientist Youyang Gu, predicts a smaller second wave this summer, followed by a gradual decline in the fall. But because his model doesn’t show deaths dropping as rapidly, the total mortality projections are higher than IHME’s: More than 200,000 deaths nationwide by Oct. 1.

Gu expects the increase in infections and deaths in the summer will be a wakeup call, especially for states where the epidemic was not as severe. His model assumes people will be more careful and that testing and contact tracing will be expanded to more effectively detect and isolate new infections.

“I think people will begin taking greater precautions when they leave their homes, keeping social distancing, wearing masks,” he said. “I have a positive outlook for the fall and winter… (because) we can look to countries in Asia and Europe and see that the virus can be contained.”

Murray said the UW model has been upgraded to factor in a wider range of variables, including smoking rates and air pollution, both of which can increase susceptibility to the virus, along with more mobility data from cell phones and data on self-reported contacts from Facebook. The new model also does a better job of differentiating true upticks in cases from upticks that are a result of increased testing, he said.

Much of the projected resurgence in the fall is linked to a seasonal effect, which appears to closely track the annual pattern of pneumonia deaths, Murray added. If that pattern holds true, transmission would be expected to rise as the weather cools.

“We start to see a powerful increase starting early in September that will intensify through the first week of February,” Murray said. “Seasonality will be a very big driver of the second wave.”

But Gu cautioned that seasonal impacts are still unclear. Many of the states where cases are currently climbing, like Florida, South Carolina and Arizona, are far from cool.

The UW model has also started factoring in the effect of mask-wearing, which analyses show could cut transmission by 40% to 60%, Murray said.

Etzioni, who has been critical of the IHME model and the extent to which it was embraced by the White House and other decision-makers, said forecast models are less useful at this stage of the epidemic than they were early on, when no one had any idea how things would unfold.

Now, many of the trends are clear in the data. With the U.S. recording roughly 1,000 deaths every day, it doesn’t take a model to predict that — absent dramatic changes — mortality will exceed 160,000 before summer’s end, she said.

“When you hear big numbers like that, people’s eyes start to glaze over,” she said. “But every single one of those numbers is a total tragedy for somebody.”

What modelers should be focusing on now, Etzioni said, is evaluating different scenarios to help guide decisions: How much difference does mask-wearing make? What’s the impact of allowing large gatherings?

Murray said those are some of the questions his team will tackle next. They’re also preparing a global forecast for release next week, at a time when the number of infections worldwide recently surpassed 7 million and cases continue to rise across much of Latin America, South Asia and Africa.

“What we’re seeing on the global front is very sobering,” Murray said. “We’re entering a much more dangerous phase of the pandemic than we’ve been going through in the last three and a half months.”