Excess deaths are a proxy for the mortality effects of SARS-CoV-2 infection. The key assumption is that increases in all-cause mortality during peak weeks of COVID-19 compared to pre-pandemic periods are nearly all due to the infection, even if SARS-CoV-2 infection was not confirmed. The validity of this method is supported in part by documenting modest reductions (negative excess) in overall mortality in selected east Asian countries that effectively prevented the original wave from March to June, 2020.

The difference of 3 million deaths across the three models is far from trivial. However, given that WHO imprimatur carries substantial influence on countries, a more relevant question is whether WHO estimates are credible. About half of WHO’s estimate is derived from observed data, the other half from modelled data. IHME combines six different approaches and applies complex methods to create estimates for various countries. Yet IHME’s method yields implausibly narrow uncertainty intervals. The Economist applies machine learning, using many covariates, and has appropriately wider uncertainty intervals; like WHO, it makes its model fully open source. In high-income countries, much of the discrepancy between excess and reported COVID-19 deaths occurred during the first viral wave, from March to June, 2020, when SARS-CoV-2 infections and COVID-19 deaths swept through nursing homes. Italy, for example, has robust and rapid reporting of COVID-19 deaths, and WHO estimated 161 000 excess deaths, as did Italian researchers.

By contrast, IHME

and The Economist estimated 259 000 deaths and 192 000 deaths, respectively. That at least 9 million COVID-19 deaths were missed by official reports raises a few key issues.