Back

Metrics for Accounting Deaths from COVID-19

Stear, B.; Hernandez, K.; Manian, V.; Dutta, S.; Taylor, D.; Conley, C.

2020-11-03 epidemiology
10.1101/2020.08.29.20184176 medRxiv
Show abstract

Efforts to mitigate the spread of coronavirus disease 2019 (COVID-19) in the United States require an accurate understanding of how the epidemic is progressing. The National Center for Health Statistics (NCHS) releases weekly numbers of deaths attributed to a set of select causes, including deaths from COVID-19 in the entire United States (US), by state, and cumulatively for individual counties. Comparing US and state level deaths from select causes recorded in 2020 with values from 2014-2019 identifies a number of differences that exceeded 95% confidence limits on historical mean values, including three states with deaths possibly from COVID-19 in December 2019. Comparing county-level NCHS datasets with county-level data on deaths from COVID-19 compiled by four public pandemic tracking sites suggests that a large number of COVID-19 deaths have not yet been reported to the NCHS. Dividing the numbers of COVID-19 deaths counted by the public tracking sites by the percentage of COVID-19 deaths reported to the NCHS suggests that approximately 20% of all US deaths from Natural Causes, as many as 200,000, may not yet have been reported to the NCHS. Evaluating changes in the fractions of deaths attributed to COVID-19 and other specific causes or nonspecific outcomes during the epidemic, relative to 2020 totals or historical mean values, can provide a valuable perspective on the public health consequences of COVID-19. Significance StatementEstimating total deaths from natural causes using the percentage of natural cause deaths from COVID-19 reported to the CDC and the number of COVID-19 deaths counted by public tracking sites suggests that up to 200,000 deaths from natural causes between 22 April and 15 August, 2020, around 20% of the total recorded as of 26 August, have not yet been reported to the CDC.

Matching journals

The top 6 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.