Thread by @biomed_data

Our analysis of pandemic excess deaths for the Nordic countries 🇩🇰🇫🇮🇮🇸🇳🇴🇸🇪is now published. #tldr:
-major dependencies on method and time-period used
-unusual 2018/2019 mortality years affect estimates
-concerns about @IHME_UW model. /1
doi.org/10.1093/ije/dy…

IHME @IHME_UW (doi.org/10.1016/S0140-…) is a major outlier, giving 2x total Nordic excess death vs. other models. Our estimate of IHME expected deaths (red) seem inconsistent with register data and substantially too low, causing too high excess deaths for 🇩🇰🇫🇮🇸🇪. /2

IHME also gives implausible reporting ratios and IFRs, e.g., 🇩🇰 & 🇫🇮 several-fold worse at identifying covid-deaths and much more lethal infections than 🇳🇴 and 🇸🇪 -errors explained by erroneous death estimates. doi.org/10.1016/S0140-… /3

IHME models have major influence –published this Spring in @TheLancet, already cited 100s of times and used in the Lancet Commissions’ report (doi.org/10.1016/S0140-…) as premise for all discussions, incl. country comparisons that we show are unreliable. /4

Other models that we analyzed, WHO, Economist (in two versions), WMD, and BME – had much more similar estimates, more consistent with trends derived from historic register data, and much more country-similar reporting ratios (below) and IFRs. /5

For the future, our study illustrates need for data-based critical post-publication review, comparison & sensitivity analysis of complex models whose uncertainties and assumptions are hard to interpret @TheLancet @richardhorton1 /6

Note: All estimates should be taken with caution as they miss detailed demographics, such as changes in the age group populations over the study period. We have a study soon to be submitted where we analyze population structure effects on mortality. /7