6 ms·

Overcounts are just as likely. Well incorrectly attributed deaths at least. Somebody who tests positive, and then dies in a car accident meets your above crite

Overcounts are just as likely. Well incorrectly attributed deaths at least. Somebody who tests positive, and then dies in a car accident meets your above criteria. For something less contrived, so does somebody who tests positive, and then subsequently dies of an unrelated heart disease. If you wanted to fix attribution, you’d need a coroner’s investigation into each death, which certainly isn’t happening. Even testing every death that you want to count isn’t feasible, because of test shortages and the potential for false negatives. A real world example of that is in New York, where a revision in their attribution methodology had their numbers jump up massively in one day [0]. What changed was they decided to add all deaths where Covid was a suspected cause to the count. Whether these shortcomings in methodology would lead to an overall overcount is probably unlikely. But methodology differences from one place to the next does make it difficult to compare statistics. With some factors applying downward pressure to the stats, and other factors applying upward pressure. One thing I’m going to be interested in looking up when this is all over is 2020s influenza deaths. I have a strong suspicion that 2020 is going to have record low flu deaths, with many of them being incorrectly attributed to Covid-19. [0]: https://www.theguardian.com/us-news/2020/apr/15/new-york-city-coronavirus-death-toll-jumps-revised-count https://www.theguardian.com/us-news/2020/apr/15/new-york-cit...

ska 7y ago
I can't see how they are as likely, and already posited that they are possible. I didn't give a criteria that says everyone who has tested positive should be counted as a covid-related death. New York pretty clearly did the right thing here. There isn't going to be a coroners investigation, but with somewhere that large you have decent statistics on expected rates and attributing the up tick they saw was methodologically better than excluding those deaths. After all if I understand it correctly "suspected deaths" here is counting the otherwise unaccounted for increase over baseline statistics, not the baseline cases. That's the correct way to do this, and afiak it is how NY did it in reporting but I haven't double checked. It's not perfect, but that's life. Agree that this years flu analysis will be interesting, but on the overall statistics nobody competent is going to conflate them with covid-19, there may just be larger error bars.
- AmericanChopper 7y ago
  It’s not better (or worse) imo, it’s simply another bias. It’s just one that moves the numbers in the direction counter to some of the other, more well discussed biases. > with somewhere that large you have decent statistics on expected rates I wouldn’t make that assumption. I can’t think of any place in the world that has the ability to collect this data with any sort of scientific rigour. It’s simply not possible at the moment, and without that you have no scientific basis to know what to expect. > Agree that this years flu analysis will be interesting, but on the overall statistics nobody competent is going to conflate them with covid-19 I wouldn’t make this assumption either. If you attribution criteria includes dying in hospital of Covid-like symptoms, then it also includes dying in hospital of flu-like symptoms, because a severe case of flu and a severe case of Covid present almost identical symptoms. The US has had ~55k Covid attributed deaths so far, in last years flu season, it had ~35k flu attributed deaths. The potential for misattribution on that basis alone is significant. The general point I was making was that not all of the confounding factors apply downward pressure to the stats, some of the provide upward pressure, and depending on methodology, you’re going to get a different set of confounding factors from place to place.
  - ska 7y ago
    I’m not disagreeing with you that their are confounding factors in both directions, this is clearly true. I’m disagreeing that at this moment - and with statistics like excess mortality in NYC, there is very good reason to believe that upward and downward pressure are not equally likely. As to the first assumption I disagree . We have pretty good ideas of all causes rates in somewhere like NYC, we have pretty good numbers on things like accidents etc. And then sort of factors you point out like the effect of a recession aren’t on effect yet for the most part. So you have a bunch of excess morbidity, and you have pretty good local statistics in the gaps, at least locally. You are right that this doesn’t go apples to apples in other places but we aren’t doing that so it does matter. There is one kind of interesting wrinkle, you may have stress related impacts that show up even on a short timeline. So there is a question if you can/should attribute those and you can pick reasonable ways, either choice . Flu isn’t likely going to be confounded too badly because people are going to be focussing in the counfounding impacts (e.g. of isolation). It just seems strange to me to assume this will be a problem , some people involved are good at this sort of analysis . Much more likely that a decent job will be done and the estimation errors will have wider bounds. I guess I just don’t see any data to support your pessimism , on that , but I could be wrong of course. I mean this sort of modelling isn’t easy but it is doable.
Isinlor 7y ago
If you look at deaths from Sweden and Belgium the excess mortality matches very well the deaths they attribute to COVID-19[0]. Both Sweden and Belgium count COVID-19 deaths in community outside hospitals. It would seem they over-count by some 5%. The example of these two countries suggests that the excess mortality can be basically fully attributed to COVID-19 [1]. [0]: https://www.nytimes.com/interactive/2020/04/21/world/coronavirus-missing-deaths.html https://www.nytimes.com/interactive/2020/04/21/world/coronav... [1]: https://www.euromomo.eu/graphs-and-maps https://www.euromomo.eu/graphs-and-maps
- ska 7y ago
  Right, the increase in excess mortality is almost certainly covid-19 related, and the potential for overcount is relatively small (and will probably be accounted for reasonably). By comparison not counting these will result in a large undercount, a much more significant effect.
  - AmericanChopper 7y ago
    You’re saying almost certainly, but you’re demonstrably making a lot of assumptions. For instance studies of the effect recessions have on mortality rates vary wildly, from having a positive effect (like the Great Depression), to having a hugely negative effect (Brazil 2012-2016). There’s other obvious confounding factors in this approach too. What is the impact on mortality rate in moving primary care to Tele-conferencing? The pandemic had prioritized Covid care, and stretched medical resources. What impact has that had on the mortality rate of non-Covid conditions? The answers to those questions are not obvious, and have not been rigorously studied. Even if you could answer them, that approach would imply redefining the Covid-19 death toll from being “deaths caused by Covid-19” to “deaths which may not have otherwise happened, during that particular period, if not for Covid-19”.
    - ska 7y ago
      On the time scale we are liking at so far, almost none of those terms contribute . I agree that over time it gets muddier , but right now in places like NYC it’s pretty clear .

ska 7y ago

I can't see how they are as likely, and already posited that they are possible. I didn't give a criteria that says everyone who has tested positive should be counted as a covid-related death. New York pretty clearly did the right thing here. There isn't going to be a coroners investigation, but with somewhere that large you have decent statistics on expected rates and attributing the up tick they saw was methodologically better than excluding those deaths. After all if I understand it correctly "suspected deaths" here is counting the otherwise unaccounted for increase over baseline statistics, not the baseline cases. That's the correct way to do this, and afiak it is how NY did it in reporting but I haven't double checked. It's not perfect, but that's life. Agree that this years flu analysis will be interesting, but on the overall statistics nobody competent is going to conflate them with covid-19, there may just be larger error bars.

AmericanChopper 7y ago
It’s not better (or worse) imo, it’s simply another bias. It’s just one that moves the numbers in the direction counter to some of the other, more well discussed biases. > with somewhere that large you have decent statistics on expected rates I wouldn’t make that assumption. I can’t think of any place in the world that has the ability to collect this data with any sort of scientific rigour. It’s simply not possible at the moment, and without that you have no scientific basis to know what to expect. > Agree that this years flu analysis will be interesting, but on the overall statistics nobody competent is going to conflate them with covid-19 I wouldn’t make this assumption either. If you attribution criteria includes dying in hospital of Covid-like symptoms, then it also includes dying in hospital of flu-like symptoms, because a severe case of flu and a severe case of Covid present almost identical symptoms. The US has had ~55k Covid attributed deaths so far, in last years flu season, it had ~35k flu attributed deaths. The potential for misattribution on that basis alone is significant. The general point I was making was that not all of the confounding factors apply downward pressure to the stats, some of the provide upward pressure, and depending on methodology, you’re going to get a different set of confounding factors from place to place.
- ska 7y ago
  I’m not disagreeing with you that their are confounding factors in both directions, this is clearly true. I’m disagreeing that at this moment - and with statistics like excess mortality in NYC, there is very good reason to believe that upward and downward pressure are not equally likely. As to the first assumption I disagree . We have pretty good ideas of all causes rates in somewhere like NYC, we have pretty good numbers on things like accidents etc. And then sort of factors you point out like the effect of a recession aren’t on effect yet for the most part. So you have a bunch of excess morbidity, and you have pretty good local statistics in the gaps, at least locally. You are right that this doesn’t go apples to apples in other places but we aren’t doing that so it does matter. There is one kind of interesting wrinkle, you may have stress related impacts that show up even on a short timeline. So there is a question if you can/should attribute those and you can pick reasonable ways, either choice . Flu isn’t likely going to be confounded too badly because people are going to be focussing in the counfounding impacts (e.g. of isolation). It just seems strange to me to assume this will be a problem , some people involved are good at this sort of analysis . Much more likely that a decent job will be done and the estimation errors will have wider bounds. I guess I just don’t see any data to support your pessimism , on that , but I could be wrong of course. I mean this sort of modelling isn’t easy but it is doable.

Isinlor 7y ago

If you look at deaths from Sweden and Belgium the excess mortality matches very well the deaths they attribute to COVID-19[0]. Both Sweden and Belgium count COVID-19 deaths in community outside hospitals. It would seem they over-count by some 5%. The example of these two countries suggests that the excess mortality can be basically fully attributed to COVID-19 [1]. [0]: https://www.nytimes.com/interactive/2020/04/21/world/coronavirus-missing-deaths.html https://www.nytimes.com/interactive/2020/04/21/world/coronav... [1]: https://www.euromomo.eu/graphs-and-maps https://www.euromomo.eu/graphs-and-maps

ska 7y ago
Right, the increase in excess mortality is almost certainly covid-19 related, and the potential for overcount is relatively small (and will probably be accounted for reasonably). By comparison not counting these will result in a large undercount, a much more significant effect.
- AmericanChopper 7y ago
  You’re saying almost certainly, but you’re demonstrably making a lot of assumptions. For instance studies of the effect recessions have on mortality rates vary wildly, from having a positive effect (like the Great Depression), to having a hugely negative effect (Brazil 2012-2016). There’s other obvious confounding factors in this approach too. What is the impact on mortality rate in moving primary care to Tele-conferencing? The pandemic had prioritized Covid care, and stretched medical resources. What impact has that had on the mortality rate of non-Covid conditions? The answers to those questions are not obvious, and have not been rigorously studied. Even if you could answer them, that approach would imply redefining the Covid-19 death toll from being “deaths caused by Covid-19” to “deaths which may not have otherwise happened, during that particular period, if not for Covid-19”.
  - ska 7y ago
    On the time scale we are liking at so far, almost none of those terms contribute . I agree that over time it gets muddier , but right now in places like NYC it’s pretty clear .