5 ms·

Sidebar, but I simply cannot figure out how the CDC estimates flu mortality. I've spent about an hour trying to find the info, and gave up. I was able to figur

Sidebar, but I simply cannot figure out how the CDC estimates flu mortality. I've spent about an hour trying to find the info, and gave up. I was able to figure out how they estimate flu deaths: they take the officially diagnosed deaths, and the multiply it by a factor based on presumed deaths. For example, they assume some % of pneumonia deaths are cause by the flu. But I simply cannot figure out how they estimate the number of flu infections per year, which is the other half of IFR. It obviously is a statistical model (they don't do millions of flu tests a year), but what is the model? Without this info, it's really not an apples-to-apples comparison when you take flu IFR and compare it to coronavirus IFR.

argonaut 6y ago
Doesn't this page answer all those questions? https://www.cdc.gov/flu/about/burden/how-cdc-estimates.htm https://www.cdc.gov/flu/about/burden/how-cdc-estimates.htm
- marcell 6y ago
  No! I dug through that page and the cited papers, and it absolutely does not have any of the answers to those questions. If you find the answers on that page, I'd love to read them. For example, they say: > The numbers of influenza illnesses were estimated from hospitalizations based on how many illnesses there are for every hospitalization, which was measured previously (5). Ok... "estimated from hospitalizations"? What does that mean? How are you deriving that estimate? If you read the study linked, they say this: > Multipliers were calculated as the simple inverses of the proportions at each step. We accounted for variability and uncertainty in model parameters by using a probabilistic (Monte Carlo) approach Ok... so it's a Monte Carlo simulation? What does that even mean in this context? Compare this to the Coronavirus IFR estimates. They are easy to understand. They test random samples of people in a population for antibodies, and use that to estimate the spread of the virus.
  - argonaut 6y ago
    I looked at the linked study. It's not very complicated. In fact one could argue their method was too simple. They estimated the number of illnesses by dividing the number of hospitalizations by a series of probabilities (the probabilities constitute a funnel from someone getting an infection to someone being hospitalized - for example the probability you actually seek hospital attention, the probability your test accurately detects influenza, etc). The probabilities were uniformly sampled from a range of plausible values (derived from their survey data). They did this process 10k times to arrive at an distribution over the number of illnesses (this is a Monte Carlo simulation).

argonaut 6y ago

Doesn't this page answer all those questions? https://www.cdc.gov/flu/about/burden/how-cdc-estimates.htm https://www.cdc.gov/flu/about/burden/how-cdc-estimates.htm

marcell 6y ago
No! I dug through that page and the cited papers, and it absolutely does not have any of the answers to those questions. If you find the answers on that page, I'd love to read them. For example, they say: > The numbers of influenza illnesses were estimated from hospitalizations based on how many illnesses there are for every hospitalization, which was measured previously (5). Ok... "estimated from hospitalizations"? What does that mean? How are you deriving that estimate? If you read the study linked, they say this: > Multipliers were calculated as the simple inverses of the proportions at each step. We accounted for variability and uncertainty in model parameters by using a probabilistic (Monte Carlo) approach Ok... so it's a Monte Carlo simulation? What does that even mean in this context? Compare this to the Coronavirus IFR estimates. They are easy to understand. They test random samples of people in a population for antibodies, and use that to estimate the spread of the virus.
- argonaut 6y ago
  I looked at the linked study. It's not very complicated. In fact one could argue their method was too simple. They estimated the number of illnesses by dividing the number of hospitalizations by a series of probabilities (the probabilities constitute a funnel from someone getting an infection to someone being hospitalized - for example the probability you actually seek hospital attention, the probability your test accurately detects influenza, etc). The probabilities were uniformly sampled from a range of plausible values (derived from their survey data). They did this process 10k times to arrive at an distribution over the number of illnesses (this is a Monte Carlo simulation).