During the COVID-19 pandemic, surveillance systems worldwide underestimated mortality in real time due to longer death reporting lags. In Spain, the mortality monitor “MoMo” published downward biased excess mortality estimates daily. I study the correction of such bias using polynomial regressions in data from January to March 2021 for Spain and the Comunitat Valenciana, the region with the highest excess mortality.
This adjustment for real-time statistics consisted of (1) estimating forthcoming revisions with polynomial regressions of past revisions, and (2) multiplying the daily-published excess mortality by these estimated revisions. The accuracy of the corrected estimates compared to the original was measured by contrasting their mean absolute errors (MAE) and root mean square errors (RMSE).
Applying quadratic and cubic regressions improved the first communication of cumulative mortality in Spain by 2–3%, on average, and the flow in registered deaths by 20%. However, for the Comunitat Valenciana, those corrections improved the first publications of the cumulative mortality by 36–45%, on average; their second publication, by 23–30%; and the third, by 15–21%. The flow of deaths registered each day improved by 62–63% on their first publication, by 19–36% on the second, and by 12–17% on the third.
It is recommended that MoMo's estimates for excess mortality be corrected from the effect of death reporting lags by using polynomial regressions. This holds for the flows in each date and their cumulative sum, as well as national and regional data. These adjustments can be applied by surveillance systems in other countries.