Despair and Games with Models, 5/11/20 Data

COVID19-hospitalizationsandprojections-051120-mod-featured

Skipping the daily chart update for COVID-19 hospitalizations in Rhode Island didn’t deprive readers of anything, mostly because the line has been very close to the same, which means the model has, for these few days anyway, been predicting the correct decrease.

Across all of the various measures, COVID-19 is on the decline in Rhode Island.  New cases fell below a 2% daily increase in today’s data, which means less than a 0.5 infection rate over 14-days.  Under the current trend, Rhode Island would see its last new case reported on June 13.  Of course, that’s very unlikely because the government is — rightly — allowing more movement and activity.

This leads to our regular reminder that political leaders should be putting forward clear signals about what the targets are for our reopening, and we in the public should be debating it.  Clearly we don’t want to cause catastrophe by overwhelming our medical system, but how much are we, as a state, willing to encounter this disease in our regular lives, and at what cost are we willing to declare COVID deaths unacceptable?

To answer that question, we’ll have to acknowledge that our response is killing people, too.  A study by Well Being Trust, for example, estimates that the United States could see 75,000 “deaths of despair” — that is, fatalities due to suicide, overdose, substance abuse, and so on.

We also have to be clear about the deaths that we are witnessing.  In the 51 days of available data since the first Rhode Island COVID-19 death, only seven days saw more deaths in the hospital than out of it.  If people are passing elsewhere, it probably means that their case was hopeless or they were already in locations — like nursing homes — that are like hospitals.  In short, they are in identifiable locations that can be protected.

If we’re smart, reasonable, and a little bit courageous, it isn’t inconceivable that we could accelerate our reopening without spikes in the measures that are the most important (hospitalizations and deaths).

COVID19-hospitalizationsandprojections-051120-mod

 

One-day projections and actuals:

  • Cases:
    • Projection for 5/11: 11,427
    • Actual for 5/11: 11,450
    • Projection for 5/12: 11,602
  • Hospitalizations:
    • Projection for 5/11: 276
    • Actual for 5/11: 278
    • Projection for 5/12: 275
  • Deaths:
    • Projection for 5/11: 428
    • Actual for 5/11: 430
    • Projection for 5/12: 436


  • Mario

    I consider the experiment I made to project case growth to be a huge success. It’s complicated, and I ended up having to assume that the virus is actually quite a bit more fatal than the data out of New York suggested (~2%; it is largely a mathematical construct rather than a real assumption, it just makes the numbers work, but it could be evidence that the hit to RI was just to a much more vulnerable segment of the population than the less controlled spread in NY). This leaves me with an unaware, asymptomatic or pre-symptomatic population of only 6400 people walking around. Then you can take the level of testing and treat it like a poll, non-randomly sampling a known population.

    The level of testing is obviously unpredictable, but I figure if it were about 2400 it would find a little over 3.7% of the unknown infected, for 233 cases. At 3000 it would find 4.4%, at 1000 it would find only 2.3%. So that’s how I’m looking at it.

    So for tomorrow I think we’ll see about 219 new cases, 12 deaths, and 264 hospitalizations (+29, -37, -3.5). I can also project forward now, albeit with almost no confidence (especially as it comes to deaths), for a total of 650 deaths, with all cases drying up right at the end of June.

    Obviously, I don’t say this as an actual prediction (in many ways I’d love to be wrong) because things will change, like personal behavior, and that will change the end result. But having a solid, testable prediction will at least allow me to say how the outlook is changing as the real numbers come in.