4 ms·

The hospitalizations in South Africa went from 19% to 1.7% but trust Imperial (whose highly inaccurate lockdown modelling last year caused who knows how much da

by d1a2n 5y ago

The hospitalizations in South Africa went from 19% to 1.7% but trust Imperial (whose highly inaccurate lockdown modelling last year caused who knows how much damage) to produce a study that gives significantly less favourable numbers. If anyone has the patience to read it, it would be interesting to know what statistical/definitional trickery they've used to make omicron maximally menacing.

rsfern 5y ago
How is 40% less likely to lead to 1+ day hospitalization “maximally menacing”? I haven’t read this imperial college paper yet, but yesterdays preprint from SA has it at 80% less likely to lead to hospitalization, with some uncertainty as to whether that’s due to intrinsic lower virulence vs underestimated prior immunity https://doi.org/10.1101/2021.12.21.21268116 https://doi.org/10.1101/2021.12.21.21268116 Edit: read through a good portion of it, the methodology seems pretty reasonable, including definitions for infection/reinfection, inclusion criteria, the covariates included in the model. I’d say the biggest uncertainty factors are (1) it’s still early and there aren’t that many hospitalizations yet (the study period ended like a week ago) and (2) for their correction accounting for undercounted prior infections, they assume undercounting of 1/3 based on some other papers. Maybe that’s reasonable, I’m not that familiar with testing in the UK. But they report both corrected and incorrect results
- disgruntledphd2 5y ago
  Unfortunately, Imperial's predictions for the UK government have become politicised, so it can be difficult to discuss them rationally. I'd be pretty concerned at the really small date range for this study, but otherwise it looks reasonable. The corrections of hazard ratios is the part that might, if you needed to, be hiding some trickery.

4 ms·

The hospitalizations in South Africa went from 19% to 1.7% but trust Imperial (whose highly inaccurate lockdown modelling last year caused who knows how much da

by d1a2n 5y ago

rsfern 5y ago
How is 40% less likely to lead to 1+ day hospitalization “maximally menacing”? I haven’t read this imperial college paper yet, but yesterdays preprint from SA has it at 80% less likely to lead to hospitalization, with some uncertainty as to whether that’s due to intrinsic lower virulence vs underestimated prior immunity https://doi.org/10.1101/2021.12.21.21268116 https://doi.org/10.1101/2021.12.21.21268116 Edit: read through a good portion of it, the methodology seems pretty reasonable, including definitions for infection/reinfection, inclusion criteria, the covariates included in the model. I’d say the biggest uncertainty factors are (1) it’s still early and there aren’t that many hospitalizations yet (the study period ended like a week ago) and (2) for their correction accounting for undercounted prior infections, they assume undercounting of 1/3 based on some other papers. Maybe that’s reasonable, I’m not that familiar with testing in the UK. But they report both corrected and incorrect results
- disgruntledphd2 5y ago
  Unfortunately, Imperial's predictions for the UK government have become politicised, so it can be difficult to discuss them rationally. I'd be pretty concerned at the really small date range for this study, but otherwise it looks reasonable. The corrections of hazard ratios is the part that might, if you needed to, be hiding some trickery.