Data Comparisons

Wisconsin: Too early to tell.

14 Comments

Wisconsin’s Supreme Court through out their stay at home on May 13. Yesterday, some newspapers breathlessly reported that deaths and new positive cases hit a highs yesterday. What does this mean? I think it’s too soon to tell.

The Covid-19 death and new case rates per capita for Wisconsin are shown below. The average for the US per capita is shown to permit you to compare WI to the US as a while.


(Small symbols are daily values. Blue is every 7 days.)

The uptick in mean death might

  1. be deaths actually tickeing up relative to the baseline.
  2. be normal ‘noise’ added to the natural cycle in reported values that result from weekly lag in reporting.
  3. be enhanced weekly cycle due to the long weekend.

However, it is worth nothing that 15 days since the quarantine was lifted is a bit to short to attribute those deaths to increased circulation after the WI SCOTUS ruling. On average, we tend to expect them to take roughly 18 days from the time of infection to die (though that number is hardly firm.) So we would not expect people who were infected 15 days ago to be dying of Covid-19 yet. It the uptick in deaths is real, this would suggest the infections happened before the ruling.

The uptick in mean positive cases might be due to any of the same things that can cause an uptick in reported deaths. It could also be the result of a dramatically increased rate of detections. While the term “new cases” is used, it may be something of a misnomer. “New detected cases might be more descriptive. In the past, detection was limited by limitation in test availability; it’s been increasing rapidly recently:

We can see the test rate in Wisconsin below.

Having said that new detected cases could result from improved detection, it is worth nothing the ratio of positives to all tests was also high yesterday — see the small blue dots which hightlight yesterday and similar weekdays.

It is worth pointing out that the time to detect a case should be less than for people to die. So it’s somewhat more plausible an increase in cases could be due to lifting of the quarantine and people swarming to bars. There are news reports that people “swarmed” to bars. Google mobility does suggest a bump in visits to retail/recreation starting after the ruling.

News reports are news reports. In a week or two, we’ll have a better idea whether the up-ticks yesterday are more likely “noise” or real.

14 thoughts on “Wisconsin: Too early to tell.”

  3. It looks like there were 3 lows days before the spike. Likely a 3 day weekend thing. The media could have reported “lowest 3 day total since 2 months ago!”. From what I can tell there isn’t a lot of change since reopening overall in the US yet, and we are getting pretty far in to where we should start seeing it. I think this is primarily because not a lot of people are changing back to * risky * behavior even though they may be going out more.
    .
    There are few people doing idiotic things, and eventually one of those will produce a super-spreader event. But as long as these are rare and it gets contained in specific social circles then these can be overcome.
    .
    The media highlighting single day totals is borderline dishonest, they do need to report the data. Here is our paper handling a single day spike (which is perfectly OK):
    .
    “The numbers follow a lull in newly reported infections and deaths over the Memorial Day holiday. Since the state began its phased reopening, health officials have not reported a spike or surge in cases like critics and some doctors warned might come when people began to move around again. But health specialists say it’s too soon to draw solid conclusions”

  4. Lucia: Why do you think 18 days is the appropriate lag between infection and death?

    The lag infection and symptoms is 3-6 days (for those who get and recognize symptoms). If I remember correctly, the average time between infection and diagnosis is a week and has probably shrunk as testing has become available.

    In an earlier post, you used a lag of 7 days between confirmed diagnosis and death. When I varied the lag between confirmed cases and death, I found a lag of 7 days produced a more constant death rate (6.5%) than 4 days or 10 days or longer as the doubling time changed from 2.5 days to over 25 days.

    Doctors are placing far fewer patients on respirators – a dreadful experience – and have a better idea of which patients have a realistic change of ever going home if they are artificially kept alive. So it is possible that the time from infection to death has shrunk.

  5. 18 days is a number established early on, it isn’t hard to determine when someone died and the patients were asked when they started showing symptoms. This removes much of the ambiguity of testing delays and so forth. It’s possible this number has moved around some since then, but not very much. There is some additional time from exposure to showing symptoms, 2 to 5 days. So ~3 weeks from a change in behavior to it showing up in death counts is reasonable I think.

  6. The mean time from infection to death determined from Wuhan data is 23 days. 4-5 days from infection to symptoms plus 18-19 days from symptoms to death.
    .
    I have not seen updated numbers. Better treatment might well make it either longer or shorter.

  7. Frank

    Lucia: Why do you think 18 days is the appropriate lag between infection and death?

    The scientists in china monitored people more-or-less randomly and came up with an estimate. (I don’t have the ciatation right now. Annan cited it.) Anyway, that is a rough estimate.
    .

    In an earlier post, you used a lag of 7 days between confirmed diagnosis and death.

    Which only tells us the time between exposure and death is longer. After all, that lag between exposure and death includes
    *incubation time, (i.e time between exposure and actually being sick),
    * time between getting sick and being diagnosed. and
    * time between getting diagnosed and dying.

    Obviously the sum over all three is greater than the sum of only the final value. 18 days might not be correct, but it’s an order of magnitude.

  8. Mike M

    The mean time from infection to death determined from Wuhan data is 23 days. 4-5 days from infection to symptoms plus 18-19 days from symptoms to death.

    Well…that ok. Then I’m missing some days from my 18 day estimate. I’m reading different numbers. But 18 from exposure from death doesn’t seem low. It might be high.

  9. Tom Scharf

    It looks like there were 3 lows days before the spike.

    Agreed. But since we only need to wait a few days to see if the spike unspikes (or not), I didn’t belabor that!
    I’ll look at number tonight. (After watching netflix with Jim. 🙂 )
    .
    Had a good day tutoring today.

  10. I last looked for stuff on this a month ago.

    The CDC has a page on the subject:
    https://www.cdc.gov/coronavirus/2019-ncov/hcp/clinical-guidance-management-patients.html
    All the references seem to be based on Chinese data. It says 4-5 days from infection to first symptoms.
    .
    The go-to reference seems to be a paper published in the Lancet in March, cited on the CDC page, using data from China. Here is a really good summary of the findings:
    https://www.drugs.com/medical-answers/covid-19-symptoms-progress-death-3536264/
    Everything is measured as time from first symptoms. It says 10-12 days to improvement for people with “mild” cases, 18.5 days to death, and 22 days to hospital release for people who are hospitalized.

  11. Mike M

    It says 10-12 days to improvement for people with “mild” cases, 18.5 days to death

    That would be pretty much the ~18 days. It’s an average. But if using “method of eyeball”, much less than that, the increase in deaths is likely NOT due to an event. If much more… maybe. The “less than” is a stronger contraint than the “more than” because other factors can cause lags. But nothing can cause an “unlag”.

  12. The NYT has an article reviewing the fraction of seropositive people in the population in major cities. NYC and London were highest with 20% and 17%. Elsewhere 10% or less. Their conclusion is that we are not near the conventional threshold for herd immunity agrees with that reached on a podcast from the Hoover Institution discussing the same data with epidemiologist Jay Bhattacharya (a co-author on the paper discussed below). Both the host and Bhattacharya were strong advocates re-opening.

    https://www.nytimes.com/interactive/2020/05/28/upshot/coronavirus-herd-immunity.html?utm_source=Nature+Briefing&utm_campaign=3f86658169-briefing-dy-20200528&utm_medium=email&utm_term=0_c9dfd39373-3f86658169-43368041

    The NYT article contains a link to Bendavid study estimating 54,000 seropositive individuals general population in Santa Clara country, at a time when there were about 1,000 confirmed cases. The study’s raw measure of seropositivity was 1.5%, but that was adjusted downward to 1.2% to correct for false positive and negatives, and then upward to 2.5-4.2% to adjust for fact that the population of volunteers tested differed many ways the population of the country. Even worse, some may have volunteered because they later had questions about an illness that they now suspected was COVID! The authors have now revised their central estimate to 2.8% in a new version of their paper. Reading some of the details is enlightening. The test manufacture claimed a false positive rate of 2 in 371 negative controls, a false positive rate of 0.8% (0.1-1.9% 95% ci), but later revised that up to 3. The Santa Clara study involved 3330 samples and the 95% confidence interval will shrink (about 3X for a normal distribution) for a larger number of samples. The investigators performed only 30 negative controls (with no false positives), far too few to accurately define the false positive rate of the test in their hands. Later 3,000+ negative controls have been run at a variety of labs (to gain permanent approval to sell from the CDC), with an overall false positive rate of 0.5% (0.3-0.8%). However, the false positive rate varied 4 in 150 in one lab to 0 in 1102 in another. I suspect that the testers knew which samples were the negative controls in the latter lab. The authors assumed the false positive rate and confidence interval for all 3,000+ negative controls applied to their study, It is possible that the bulk of their 1.5% raw positives could have been false positives. The test had a false negative rate of 32% (18-50%) compared with an in-house ELISA assay and the manufacturer claimed 8% (3-17%) false negatives. The final result claims 2.8% (1.3-4.3%) seropositives, 54X (27X-91X) the number of confirmed cases at the time. Others investigators might make equally valid choices analyzing he same data and come up with a much wider confidence interval and a different central estimate. Who knows what claims will be allowed after peer review.

    There is every reason to place much more emphasis on the ratio of confirmed cases to seropositive cases where 10% or more aere seropositive and ignore the Santa Clara data, but the NYT isn’t publicizing this ratio. Does anyone have that information?

  13. lucia (Comment #185482): “That would be pretty much the ~18 days”
    .
    18.5 + 4.5 = 23 days from infection to death.

Comments are closed.