I am (what is now called) a data scientist. I routinely write programs to glean actionable data from terabyte-sized databases. (Note that others work with petabytes ...)
I played around with publicly available Covid data and eventually gave up. It's too unreliable and inconsistent to admit of firm conclusions -- and this is what gives Wei…
I am (what is now called) a data scientist. I routinely write programs to glean actionable data from terabyte-sized databases. (Note that others work with petabytes ...)
I played around with publicly available Covid data and eventually gave up. It's too unreliable and inconsistent to admit of firm conclusions -- and this is what gives Weinstein (and others) room to play around in, and justifiably so.
The three biggest issues in Covid data are:
1. Deaths and hospitalizations "with" vs. "from" Covid.
2. The extraordinary sensitivity of the PCR test most often used to "find" Covid. Note that the fellow who won the Nobel Prize for developing PCR said that anyone who knows what he's doing could find anything in any sample.
3. The conflation of IFR and CFR. That is, while it's of some interest to track infections (those who have Covid antibodies), cases (those who are hospitalized because of a Covid infection) is the more important criterion. Cf. the flu, which has a very low IFR and a significantly higher CFR.
These issues are rarely mentioned (much less addressed) in journalistic coverage of Covid. While the scientific community is (mostly) aware of them, this knowledge rarely makes it into policy decisions or WHO or CDC guidelines.
Yeah; it's the latest (and probably last) stop in my checkered career.
I'm looking forward to CDC mortality data for 2021 and 2022 (available, probably, in 2024) to help sort things out.
The basic problem is this: Say you have serious heart disease, get Covid, and die. That could be either "with" or "from," and different physicians may well come to different conclusions. As a data guy, I would like to see a nice quantization of symptom severity to make that decision in software, but that may not be possible.
Many who die have co-morbidities, often several. If someone who has 4 chronic progressive illnesses has a massive brain hemorrhage, sepsis from a kidney infection, a fatal heart attack, those are what we list first- as required- as the cause of death. There is no motive for treating MD's to over-represent COVID as cause of death. You may as well rely on a counterfactual proposition- "If this patient didn't have COPD, ESRD, chronic CHF, diabetes and was much younger than 85, they wouldn't have died when they contracted COVID and went into severe respiratory failure".
I am (what is now called) a data scientist. I routinely write programs to glean actionable data from terabyte-sized databases. (Note that others work with petabytes ...)
I played around with publicly available Covid data and eventually gave up. It's too unreliable and inconsistent to admit of firm conclusions -- and this is what gives Weinstein (and others) room to play around in, and justifiably so.
The three biggest issues in Covid data are:
1. Deaths and hospitalizations "with" vs. "from" Covid.
2. The extraordinary sensitivity of the PCR test most often used to "find" Covid. Note that the fellow who won the Nobel Prize for developing PCR said that anyone who knows what he's doing could find anything in any sample.
3. The conflation of IFR and CFR. That is, while it's of some interest to track infections (those who have Covid antibodies), cases (those who are hospitalized because of a Covid infection) is the more important criterion. Cf. the flu, which has a very low IFR and a significantly higher CFR.
These issues are rarely mentioned (much less addressed) in journalistic coverage of Covid. While the scientific community is (mostly) aware of them, this knowledge rarely makes it into policy decisions or WHO or CDC guidelines.
I didn't know you were a data scientist, Fred.
Do you have a resource that you've seen that best untangles the "with" vs. "from" issue?
Yeah; it's the latest (and probably last) stop in my checkered career.
I'm looking forward to CDC mortality data for 2021 and 2022 (available, probably, in 2024) to help sort things out.
The basic problem is this: Say you have serious heart disease, get Covid, and die. That could be either "with" or "from," and different physicians may well come to different conclusions. As a data guy, I would like to see a nice quantization of symptom severity to make that decision in software, but that may not be possible.
Many who die have co-morbidities, often several. If someone who has 4 chronic progressive illnesses has a massive brain hemorrhage, sepsis from a kidney infection, a fatal heart attack, those are what we list first- as required- as the cause of death. There is no motive for treating MD's to over-represent COVID as cause of death. You may as well rely on a counterfactual proposition- "If this patient didn't have COPD, ESRD, chronic CHF, diabetes and was much younger than 85, they wouldn't have died when they contracted COVID and went into severe respiratory failure".