Back

Use of Unofficial Newspaper Data for COVID-19 Death Surveillance

Ahamad, M. G.; Ahmed, M. U.; Talukder, B.; Tanin, F.

2020-09-11 health policy
10.1101/2020.09.09.20191569 medRxiv
Show abstract

ObjectiveTo highlight the critical importance of unofficially reported newspaper-based deaths from coronavirus disease 2019 (COVID-19)-like illness (CLI) together with officially confirmed death counts to support improvements in COVID-19 death surveillance. MethodsBoth hospital-based official COVID-19 and unofficial CLI death counts were collected from daily newspapers between March 8 and August 22, 2020. We performed both exploratory and time-series analyses to understand the influence of combining newspaper-based CLI death counts with confirmed hospital death counts on the trends and forecasting of COVID-19 death counts. An autoregressive integrated moving average-based approach was used to forecast the number of weekly death counts for six weeks ahead. ResultsBetween March 8 and August 22, 2020, 2,156 CLI deaths were recorded based on newspaper reporting for a count that was 55% of the officially confirmed death count (n = 3,907). This shows that newspaper reports tend to cover a significant number of COVID-19 related deaths. Our forecast also indicates an approximate total of 406 CLI expected for the six weeks ahead, which could contribute to a total of 2,413 deaths including 2,007 confirmed deaths expected from August 23 to October 3, 2020. ConclusionsAnalyzing existing trends in and forecasting the expected number of newspaper-based CLI deaths indicates yet-unreported COVID-19 death counts, which could be a critical source to estimate provisional COVID-19 death counts and mortality surveillance. Public Health ImplicationsConsidering unofficial newspaper-based CLI death counts is essential to identify COVID-19 death severity and surveillance needs to advance public health research efforts to prepare appropriate response strategies for low- and middle-income countries.

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.