3 ms·

Seems like the larger correlation in deaths is related to societal status - farmers, laborers, and retirees have much higher death rates.

oefrha 7y ago
Farmers — likely crappy rural healthcare. Retirees — obviously age-skewed.
- dredmorbius 7y ago
  Quite possibly also: arriving sicker. Hypothesis: If you're a rural / non-urban farmer, under quarantine and travel restrictions, you don't go to a hospital unless you're already very sick. One of the real challenges here is in separating data artefacts from ground truth. Testing, surveillance, self-presentation, doctor-lobbying (by the patient or family) can all influence that. This makes all data already somewhat suspect, before any overt manipulations are considered (and makes those manipulations all the more toxic to a full understanding). And of course, there are numerous other factors. Farmers may well have poorer overall health (nutrition, behavioural risk factors (smoking, drink), preexisting conditions, latent or opportunistic infection, etc., etc.). All of which I'm very openly speculating on -- these are hypotheses which would have to be demonstrated through clinical assessment or systematic observation. Note that "systematic" and "complete" aren't the same -- if you are following a defensible sampling method, then 100% coverage isn't necessary. But you'd want to capture samples from other populations. The overall prevalence of COVID-19 within even Wuhan City and Hubei Province is still low, based on assessed cases (clinical symptoms or the more accurate PCR genetic screen for the SARS-nCoV-2 virus), still under 1%. Which means that population sampling would be unlikely to detect many cases without something on the order of many hundreds, possibly thousands, of samples. You're looking for an accurate central-tendency descriptor of a ~1:100 event. Knowing what the actual exposure rate has been would be useful. Whether it's sufficiently useful to come up with a comprehensive large-scale sampling methodology is another question. Note that it's not merely the sampling size, but the selection and successful measurement from each of the selected individuals for sampling which is necessary. (I've some statistics background, and some healthcare background. Enough to know basics of sampling, error, and bias. Epidemiological public health monitoring is beyond my pay grade however.)

oefrha 7y ago

Farmers — likely crappy rural healthcare. Retirees — obviously age-skewed.

dredmorbius 7y ago
Quite possibly also: arriving sicker. Hypothesis: If you're a rural / non-urban farmer, under quarantine and travel restrictions, you don't go to a hospital unless you're already very sick. One of the real challenges here is in separating data artefacts from ground truth. Testing, surveillance, self-presentation, doctor-lobbying (by the patient or family) can all influence that. This makes all data already somewhat suspect, before any overt manipulations are considered (and makes those manipulations all the more toxic to a full understanding). And of course, there are numerous other factors. Farmers may well have poorer overall health (nutrition, behavioural risk factors (smoking, drink), preexisting conditions, latent or opportunistic infection, etc., etc.). All of which I'm very openly speculating on -- these are hypotheses which would have to be demonstrated through clinical assessment or systematic observation. Note that "systematic" and "complete" aren't the same -- if you are following a defensible sampling method, then 100% coverage isn't necessary. But you'd want to capture samples from other populations. The overall prevalence of COVID-19 within even Wuhan City and Hubei Province is still low, based on assessed cases (clinical symptoms or the more accurate PCR genetic screen for the SARS-nCoV-2 virus), still under 1%. Which means that population sampling would be unlikely to detect many cases without something on the order of many hundreds, possibly thousands, of samples. You're looking for an accurate central-tendency descriptor of a ~1:100 event. Knowing what the actual exposure rate has been would be useful. Whether it's sufficiently useful to come up with a comprehensive large-scale sampling methodology is another question. Note that it's not merely the sampling size, but the selection and successful measurement from each of the selected individuals for sampling which is necessary. (I've some statistics background, and some healthcare background. Enough to know basics of sampling, error, and bias. Epidemiological public health monitoring is beyond my pay grade however.)