Development of a registration interval correction model for enhancing excess all-cause mortality surveillance during the COVID-19 pandemic

Background Estimates of excess deaths provide critical intelligence on the impact of population health threats including seasonal respiratory infections, pandemics and environmental hazards. Timely estimates of excess deaths can inform the response to COVID-19. However, access to timely mortality data is challenging due to the time interval between the death occurring and the date the death is registered and available for analysis (‘registration interval’). Development Using data from the New South Wales, Australia, Births Deaths and Marriages Registry, we developed a Poisson regression model that estimated near-complete weekly counts, for a given week of death, from partially-complete death registration counts. A 10-weeks lag was considered, and a 2-year baseline of historical registration intervals was used to correct lag weeks. Application Validation of estimated counts found that the root-mean-square error (as a percentage of mean observed near-complete registrations) was less than 7% for lag week 3, and <5% for lag weeks 4–9. We incorporated this method utilizing an existing rapid weekly mortality surveillance system. Counts corrected for registration interval replaced observed values for the most recent weeks. Excess death estimates, based on corrected counts, were within 1.2% of near-complete counts available 9 weeks from the end of the analysis period. Conclusions This study demonstrates a method for estimating recent death counts to correct for registration intervals. Estimates obtained at a 3-week lag were acceptable, while those at greater than 3 weeks were optimal.