May 6, 2020 12:34PM

Six Models Project a Drop in COVID-19 Deaths as States Open Up

“Closely watched and scrutinized projections by University of Washington researchers nearly doubled the expected number of COVID-19 deaths to about 135,000 by August [4th], based on the easing of social distancing measures,” reports The Wall Street Journal. Not really. That’s just a convenient excuse. The estimate was doubled because it was obviously wrong.

At the time of my April 6 blog, “Tracking the White House’s Favorite Epidemic Curve,” the “Murray Model” from The Institute for Health Metrics and Evaluation (IHME) was projecting 81,766 deaths by August 4. I wrote that “If COVID-19 deaths [per day] do not stop climbing soon after mid‐​April, then Chris Murray’s IHME model may begin to appear too optimistic.”

As it happens, daily deaths did seem to peak at 2216 on April 16 and were then flat or down for few days, hitting a low of 1965 on April 26 (before rising back to 2343 on May 1). In the first week of April, while the curve appeared to be flattening, the Murray Model reduced projected deaths to total only 60,000, which would have required daily deaths to drop like a stone – more like a cliff than a curve. It did not take long for reality to outrun that estimate.

An April 30 news story in Stat noted, “On Wednesday, April 29, the country blew past 60,000, more than three months before the Institute for Health Metrics and Evaluation projected. IHME, whose model has been criticized by many epidemiologists, now says the most likely death toll on Aug. 4 will be 72,433, though it could be as low as about 60,000 (obviously impossible) and as high as 115,000.” That still‐​incredible scenario, suggesting there would be only 12,000 more deaths ahead over the next three months (about 133 per day), is the one that had to be doubled to 134,475 through August 4. It wasn’t because of any easing of mandatory social distancing rules, whose efficacy is assumed rather than proven, but because 72,433 was no more realistic than 60,000.

The Wall Street Journal theory – that the IHME “nearly doubled the expected number of COVID-19 deaths … based on the easing of social distancing measures”– is easy to disprove. The online projections are available state‐​by‐​state, making it easy for anyone to see that daily deaths are still projected to decline in states that eased social distancing rules.

The IHME modelers may have encouraged the Journal writers’ incorrect interpretation of why their death estimates had to be fixed by fabricating a flawed state grading scheme for social distancing mandates (which I criticized here). In reality, the reason the IHME’s total projected deaths have increased is not because daily deaths are projected to rise (on the contrary, deaths still fall quickly) but because deaths are now expected to remain relatively higher for dozens more days.

Focusing on daily deaths, state infection reproduction rates and hospitalization admissions is more informative than “confirmed” cases, which are the tip of an invisible iceberg (because nobody tests those with no symptoms or symptoms to mild to seek medical attention) and sensitive to the intensity of testing.

The recent increase in IHME’s cumulative death estimate over six months does not mean COVID-19 deaths rates have gotten worse, only that their projections have gotten better.

None of the epidemiological models can claim to have much predictive value for more than a month ahead. So, what do they say about COVID-19 deaths in May? Ryan Best and Jay Boice gathered six model predictions of total U.S. deaths through May 31 ­–from IHME, MIT, Columbia, Northeastern, Los Alamos and the University of Texas. Aside from the just‐​revised IHME model, the average prediction of the other five ranged from 93,000 to 114,000, with a median of 100,000.

Having already reached 72,050 by May 5, reaching 100,000 by the end of the month would leave 27,950 more to go in 26 days or 1,075 deaths per day. That sounds worse that it is. Since there were 2,343 deaths on May 1, all models predict daily deaths must fall sharply in order to average less than half that many daily deaths over the whole month. The newest mid‐​range estimate from the IHME, for example, shows daily deaths down to 924 by May 31. Contrary to misdirected news stories about one model expecting more deaths than it did before, all six of these leading models expect a very significant drop in the pace of COVID-19 deaths this month.