Re:Beware second waves of Covid-19 if lockdowns eased early:...
by thu2111 on Friday March 27, 2020 @06:07AM
(#22701580)
It's possible for you both to be right.
The Professor is correct that his study essentially recommended (and caused) the nationwide house arrest now implemented, and that this is why he's now estimating far fewer deaths.
At the same time I wouldn't pick that study as a hill to die on. It has flaws that are severe and obvious to any layman and probably more that aren't, e.g.
• The assumption of mass deaths was based on constant NHS capacity for a year. That assumption is obviously false, the NHS managed to double ICU capacity within a week or so just by converting wards and now they're building emergency hospitals, China style.
• They used hospitalisation and fatality rates calculated from Italy, although the problems with biased sampling there are well discussed.
• They applied no discount rate learned from prior SARS outbreaks, where the final CFR turned out to be far lower than initial estimates, even though it's clear that the same will happen this time.
• When people asked for their source code the good professor announced it was of such low quality he couldn't release it until Microsoft had rewritten it for him.
• It mixes up excess mortality with incidental mortality. The UK has the same problem as Italy, where deaths that involve COVID-19 are being reported and counted as caused by it (because it's on the 'notifiable diseases' list). A shift in distributions is being reported as deaths that would never have happened otherwise, which isn't the case. And the ICL team know this because the professor blithely stated that of the 20,000 predicted deaths, probably 2/3rds would have happened anyway. Why was this rate not applied in the study given that excess mortality is obviously what people care about?
Right now it's really hard to feel confidence in any epidemiological output. This field feels more like psychology or economics than physics or microbiology.
The Professor is correct that his study essentially recommended (and caused) the nationwide house arrest now implemented, and that this is why he's now estimating far fewer deaths.
At the same time I wouldn't pick that study as a hill to die on. It has flaws that are severe and obvious to any layman and probably more that aren't, e.g.
• The assumption of mass deaths was based on constant NHS capacity for a year. That assumption is obviously false, the NHS managed to double ICU capacity within a week or so just by converting wards and now they're building emergency hospitals, China style.
• They used hospitalisation and fatality rates calculated from Italy, although the problems with biased sampling there are well discussed.
• They applied no discount rate learned from prior SARS outbreaks, where the final CFR turned out to be far lower than initial estimates, even though it's clear that the same will happen this time.
• When people asked for their source code the good professor announced it was of such low quality he couldn't release it until Microsoft had rewritten it for him.
• It mixes up excess mortality with incidental mortality. The UK has the same problem as Italy, where deaths that involve COVID-19 are being reported and counted as caused by it (because it's on the 'notifiable diseases' list). A shift in distributions is being reported as deaths that would never have happened otherwise, which isn't the case. And the ICL team know this because the professor blithely stated that of the 20,000 predicted deaths, probably 2/3rds would have happened anyway. Why was this rate not applied in the study given that excess mortality is obviously what people care about?
Right now it's really hard to feel confidence in any epidemiological output. This field feels more like psychology or economics than physics or microbiology.