3 ms·

How do we know the doubling time? That seems like the biggest assumption that we're not sure about. I've seen 1 week, 5 days, and even recently 3 days.

raphlinus 7y ago
Look at the graphs on "our world in data": https://ourworldindata.org/coronavirus#trajectories-since-the-100th-confirmed-case https://ourworldindata.org/coronavirus#trajectories-since-th... Based on this, 3 days does indeed look like the doubling time. However, be careful when interpreting this data, as it conflates increases in testing coverage with actual viral spread. If you look at countries with excellent testing, such as South Korea, it was still in the 3 day range, though today they thankfully have transmission comparatively under control.
- JamesBarney 7y ago
  Thanks for linking to that graph. That's a great resource! At 3 days that changes both how much we have to slow it down, and how quickly we need to start slowing it down. Not to mention that if it's 15 days from infection to diagnosis then that's 5 doubling times. So by the time we realize we didn't do enough it's way too late to avoid a complete catastrophe.
  - raphlinus 7y ago
    I don't think 15 days is a good number for infection to diagnosis, though obviously it depends on how rapidly we can accurately diagnose cases. (In the subsection of the population without access to healthcare, it's a divide by zero error). I'm not qualified enough to provide an estimate myself, but happy to link to an authoritative review: https://www.uptodate.com/contents/coronavirus-disease-2019-covid-19 https://www.uptodate.com/contents/coronavirus-disease-2019-c... (see "Incubation period" for input on what you're looking for)
    - JamesBarney 7y ago
      You're right I misread PDF.
tunesmith 7y ago
The doubling time of deaths appears to be much slower than the doubling time of confirmed cases, and is probably a better comparison. It could be slightly faster than reports, if deaths are getting missed, or misattributed to other illnesses. But it still seems more accurate than judging the doubling time of confirmed cases, which is more just a measure of how quickly we are able to increase testing when there's a backlog of "actual cases".
- raphlinus 7y ago
  There's a much narrower range in reporting rate variation for deaths than for infection->test results, but (a) it lags considerably, and (b) there are many other things that conflate this data as well, including healthcare systems getting better at preventing Covid-19 deaths over time (learning from earlier cases, hopefully starting to see positive results from remdesivir and hydroxychloroquine). I am still comfortable with what I said elsethread, the best way to get your head around doubling time is to look at data from countries with very good test coverage, such as South Korea.
- malandrew 7y ago
  Even doubling times of deaths has issues as a measure. For example, the double time before you're forced to triage is going to be better than when your health system is overwhelmed.
- raphlinus 7y ago
  Where do you get that the doubling time of deaths is much slower than that of confirmed cases? I looked for data, and this graphic is not reassuring on the question: https://commons.wikimedia.org/wiki/File:Log-linear_plot_of_coronavirus_cases_with_linear_regressions.png https://commons.wikimedia.org/wiki/File:Log-linear_plot_of_c...
  - tunesmith 7y ago
    I was eyeballing various country deaths at worldometers.info, and they looked more like 8-11 days to me. Then again, the early South Korea stats looked steeper. But yeah, I recognize the larger point is we want to know doubling rate of "actual cases", and since that's currently unknowable, identifying the most reliable proxy. It just seemed to me that our testing rate occludes "confirmed cases" more than death rate occludes "actual cases" - at least with death rates we can identify the time points when mitigation rates are introduced, when hospitals get overwhelmed, etc.
bonzini 7y ago
2 to 3 days is what we've seen in most countries during the exponential phase. Some countries in Northern Europe have had values down to 2 days or less, but that was when they were ramping up their testing like crazy. (3 days was also the number I used one week ago to state "at least two of France, Germany, Spain and Switzerland will be in lockdown and with hundreds of deaths one week from now"[1], and boy was I right). [1] https://news.ycombinator.com/item?id=22528433 https://news.ycombinator.com/item?id=22528433

raphlinus 7y ago

Look at the graphs on "our world in data": https://ourworldindata.org/coronavirus#trajectories-since-the-100th-confirmed-case https://ourworldindata.org/coronavirus#trajectories-since-th... Based on this, 3 days does indeed look like the doubling time. However, be careful when interpreting this data, as it conflates increases in testing coverage with actual viral spread. If you look at countries with excellent testing, such as South Korea, it was still in the 3 day range, though today they thankfully have transmission comparatively under control.

JamesBarney 7y ago
Thanks for linking to that graph. That's a great resource! At 3 days that changes both how much we have to slow it down, and how quickly we need to start slowing it down. Not to mention that if it's 15 days from infection to diagnosis then that's 5 doubling times. So by the time we realize we didn't do enough it's way too late to avoid a complete catastrophe.
- raphlinus 7y ago
  I don't think 15 days is a good number for infection to diagnosis, though obviously it depends on how rapidly we can accurately diagnose cases. (In the subsection of the population without access to healthcare, it's a divide by zero error). I'm not qualified enough to provide an estimate myself, but happy to link to an authoritative review: https://www.uptodate.com/contents/coronavirus-disease-2019-covid-19 https://www.uptodate.com/contents/coronavirus-disease-2019-c... (see "Incubation period" for input on what you're looking for)
  - JamesBarney 7y ago
    You're right I misread PDF.

tunesmith 7y ago

The doubling time of deaths appears to be much slower than the doubling time of confirmed cases, and is probably a better comparison. It could be slightly faster than reports, if deaths are getting missed, or misattributed to other illnesses. But it still seems more accurate than judging the doubling time of confirmed cases, which is more just a measure of how quickly we are able to increase testing when there's a backlog of "actual cases".

raphlinus 7y ago
There's a much narrower range in reporting rate variation for deaths than for infection->test results, but (a) it lags considerably, and (b) there are many other things that conflate this data as well, including healthcare systems getting better at preventing Covid-19 deaths over time (learning from earlier cases, hopefully starting to see positive results from remdesivir and hydroxychloroquine). I am still comfortable with what I said elsethread, the best way to get your head around doubling time is to look at data from countries with very good test coverage, such as South Korea.
malandrew 7y ago
Even doubling times of deaths has issues as a measure. For example, the double time before you're forced to triage is going to be better than when your health system is overwhelmed.
raphlinus 7y ago
Where do you get that the doubling time of deaths is much slower than that of confirmed cases? I looked for data, and this graphic is not reassuring on the question: https://commons.wikimedia.org/wiki/File:Log-linear_plot_of_coronavirus_cases_with_linear_regressions.png https://commons.wikimedia.org/wiki/File:Log-linear_plot_of_c...
- tunesmith 7y ago
  I was eyeballing various country deaths at worldometers.info, and they looked more like 8-11 days to me. Then again, the early South Korea stats looked steeper. But yeah, I recognize the larger point is we want to know doubling rate of "actual cases", and since that's currently unknowable, identifying the most reliable proxy. It just seemed to me that our testing rate occludes "confirmed cases" more than death rate occludes "actual cases" - at least with death rates we can identify the time points when mitigation rates are introduced, when hospitals get overwhelmed, etc.

bonzini 7y ago

2 to 3 days is what we've seen in most countries during the exponential phase. Some countries in Northern Europe have had values down to 2 days or less, but that was when they were ramping up their testing like crazy. (3 days was also the number I used one week ago to state "at least two of France, Germany, Spain and Switzerland will be in lockdown and with hundreds of deaths one week from now"[1], and boy was I right). [1] https://news.ycombinator.com/item?id=22528433 https://news.ycombinator.com/item?id=22528433