3 ms·

> If you look at Fig1, describing the cohort selection, this study has two control groups. A historical as well as a contemporary control group. This gives an t

by timr 5y ago

> If you look at Fig1, describing the cohort selection, this study has two control groups. A historical as well as a contemporary control group. This gives an two different sample sources to reason about sampling bias. There's a clear difference in health between the cohorts. Supplementary table 1 breaks down the cohorts and summarizes co-variates. Disturbingly, the COVID cohort is significantly fatter than the contemporary control group: https://static-content.springer.com/esm/art%3A10.1038%2Fs41591-022-01689-3/MediaObjects/41591_2022_1689_MOESM1_ESM.pdf https://static-content.springer.com/esm/art%3A10.1038%2Fs415... 54% of the COVID cohort has BMI > 30, vs 42% of the contemporary controls, and 40% of the historical controls. They also have a higher percentage in long-term care (4% vs. 0.6%), a higher percentage with cancer (8% vs. 6%), chronic kidney disease (19.5% vs 16.7%), lung disease (15% vs. 11%), dementia (6% vs. 3%), diabetes (32% vs 22%) and hyperlipidemia (34% vs 25%). The Covid cohort is basically a much sicker population, in almost every way. Re-weighting the populations doesn't correct for this disparity. The authors have attempted to hide this by burying the data in the supplementary material, and presenting an "absolute standardized difference" that makes it look smaller than it is.

dv_dt 5y ago
No that figure is the study being transparent on how it controlled for imbalances of selected input conditions showing raw numbers then the post processed reweighting of the samples. How does that fail as a way to account for differences in those conditions? What would be your preferred technique of accounting for the differences of characteristics in those groups?
- timr 5y ago
  Look at supplementary figure 2. Their attempts to normalize the data do not eliminate the bias. The Covid-19 cohort is significantly sicker, even after adjustment. This paper is discussing small differences in small numbers, and the single biggest difference between groups is the baseline level of health for the study participants. About the only thing you can say from this is that old, sick people are sicker after Covid-19 severe enough to put them in the hospital.

by timr 5y ago

dv_dt 5y ago

No that figure is the study being transparent on how it controlled for imbalances of selected input conditions showing raw numbers then the post processed reweighting of the samples. How does that fail as a way to account for differences in those conditions? What would be your preferred technique of accounting for the differences of characteristics in those groups?

timr 5y ago
Look at supplementary figure 2. Their attempts to normalize the data do not eliminate the bias. The Covid-19 cohort is significantly sicker, even after adjustment. This paper is discussing small differences in small numbers, and the single biggest difference between groups is the baseline level of health for the study participants. About the only thing you can say from this is that old, sick people are sicker after Covid-19 severe enough to put them in the hospital.