10 June 2020

Please define "cases" of coronavirus

I'm concerned that media outlets of all types are failing to define what they mean by a "case" of coronavirus.  Many articles and broadcasts correlate increasing "cases" with states' reopening efforts, which may be totally unfair if the cases they refer to are positive detection of the presence of the virus (or antibody evidence of past infection) in asymptomatic persons.  Even recognizing that asymptomatic persons can transmit the virus, comparing increased detection with increased public health danger is a false equivalence.

Politico today headlined that the outbreak is "spiking again."  Reporting it like this is probably valid:
Texas, for instance, has reported two straight days of record-breaking coronavirus hospitalizations — highs that come shortly after the state kicked off the third stage of its reopening plan.
But this -
Meanwhile, officials in at least 19 states have recorded two-week trends of increasing coronavirus cases, including spikes of more than 200 percent in Arizona and more than 180 percent in Kentucky...
- might just reflect increased testing.  Attention needs to focus on the prevalence of active disease (or perhaps a rising trend of %positive testing) or symptoms requiring sequestration.  Even increasing deaths is a potentially misleading statistic, since it may reflect events occurring weeks or months earlier.

This Washington Post article specifically discusses rising hospitalizations.


  1. there's evidence for and against your hypothesis:
    SO far cases are highly correlated with death rates. THere's a 2+ week lag but a tight correlation. So if your hypothesis is true then this should not be the case. or at least should deviate within the month.

    2. in many countries at a high enough case rate the new case rate seems to plateau beofre descending. This is puzzling as to why. Possibly it it related to health care saturation.

    3. there's a very simple way to sort this out-- partly. Look at the number of positive cases versus number of tests each week. Assuming there is not a major shift in testing strategy this should indicate if cases are increasing ro just more edge cases being found.

    4. But here is the big mystery in all of this:


    look at the "excess deaths" curves a ways down on the page. If the deaths per day isn't strongly diminished then how is it possible that the "excess" deaths is declining so fast. This is very hard to explain. it could relate to your point in that if more deaths are being attributed to corona due to test but the underlying death rate is declining it could mean that the death rate is on relative terms has fallen dramatically and is only being sustained by better attribution.

    1. I'm not sure what hypothesis I stated. What I may not have said clearly is that the definition of a "case" of coronavirus is important in interpreting data.

      Let's take the town of Backwater, East Dakota with 10000 residents. They have had 27 residents with positive pharyngeal swabs who were advised to self-sequester. They have had 4 people admitted to their local hospital. One resident has died. Those numbers have been stable or changing minimally for the past week.

      Today a philanthopist who was born there donates 10,000 testing kits and everyone gets tested tomorrow. The following day 165 people are found to be positive by pharyngeal swab. Public health investigation finds that 3 of those 165 have low-grade fever and 2 have coughs which they thought were due to "allergies."

      I don't believe the local newspaper or TV station should report a sudden spike of 165 new cases of coronavirus.

    2. I agree that the new cases number, by itself, is not very useful. But I disagree with charlie as to how good the "percent positive" metric is. If you go from testing only symptomatic people to giving a test to everybody who wants one (or is required to get one to return to work), the stat will change in a dramatic and misleading way. In other words, it suffers from the same problem as the new cases number in that it must be taken in context.

      The government of my home state of Nevada publishes stats for both new cases and number of new tests, so one can compare the trends. More importantly, in my opinion, it publishes a graph of hospitalizations. An early study called "Epidemiology and Transmission of COVID-19 in Shenzhen China: Analysis of 391 cases and 1,286 of their close contacts" found that, while deaths could occur five weeks or more after the onset of symptoms, almost all hospitalizations happened within 12 days, and half of those within four. Hospitalization numbers are clearly a reliable metric of severe illness, less vulnerable to the vagaries of testing and more timely than deaths.

      That's probably why the government of Florida stopped releasing the figures for hospitalizations in that state.

      Another statistic I would not worry about too much is emergency room visits. These are usually initiated by the patient, and may be unnecessary. Admittance to the hospital is determined by a doctor, and not done lightly these days.

  2. Just a fellow Wisconsinite testing Chrome vs my normal Safari. I'm using the post as Name and URL option

    1. Test successful. Congratulations.

      (Other readers can ignore this exchange, which was done for administrative purposes)


Related Posts Plugin for WordPress, Blogger...