Alternative epidemic indicators for COVID-19 in three settings with incomplete death registration systems

Abstract

Not all COVID-19 deaths are officially reported, and particularly in low-income and humanitarian settings, the magnitude of reporting gaps remains sparsely characterized. Alternative data sources, including burial site worker reports, satellite imagery of cemeteries, and social media-conducted surveys of infection may offer solutions. By merging these data with independently conducted, representative serological studies within a mathematical modeling framework, we aim to better understand the range of underreporting using examples from three major cities: Addis Ababa (Ethiopia), Aden (Yemen), and Khartoum (Sudan) during 2020. We estimate that 69 to 100%, 0.8 to 8.0%, and 3.0 to 6.0% of COVID-19 deaths were reported in each setting, respectively. In future epidemics, and in settings where vital registration systems are limited, using multiple alternative data sources could provide critically needed, improved estimates of epidemic impact. However, ultimately, these systems are needed to ensure that, in contrast to COVID-19, the impact of future pandemics or other drivers of mortality is reported and understood worldwide.

Fig. 1.. Mortality reporting and seroprevalence in Addis Ababa, Ethiopia in 2020.

(A) Reported COVID-19 deaths compiled from the Ethiopian Public Health Institute and estimated excess mortality using cemetery surveillance from (12) using data from 2015 to 2019 to derive the baseline. Dashed line indicates the end of the early peak in excess mortality featured in the sensitivity analysis. (B) As in (A), but with excess mortality derived from 2019 data only. (C) Estimated seroprevalence from model fit to excess mortality under 2015–2019 baseline (median and 95% credible intervals) with the first peak compared to reported values by Abdella et al. (17). Gray shaded area highlights the sampling period of the serosurvey, with weighted reported estimates both corresponding to this entire period. (D) Seroprevalence estimated under models fit to either reported COVID-19 or different estimates of excess mortality with different baselines (median and 95% credible intervals) compared to reported seroprevalence in (17).

Fig. 2.. Mortality reporting and seroprevalence in Aden, Yemen in 2020.

(A) Reported COVID-19 deaths and estimated excess mortality from satellite surveillance of burials from (13). (B) Estimated seroprevalence of combined IgG/IgM (positive for either IgG and/or IgM) from the model fitted to excess mortality using the default IFR (= 0.3%) under different IgG seroreversion half-lives compared to observed values from (19) [reported seroprevalence: 27.4% (95% CI, 25.6 to 29.3%)]. Dashed vertical line indicates point in time when seroprevalence dynamics divert due to difference in IgG half-life modeled. (C) As in (B) but for IgG antibodies [reported seroprevalence: 25.0% (95% CI, 23.2 to 26.9%)]. (D) Combined negative log-likelihood of models of IgG and combined IgG/IgM antibodies under varying assumptions of the IFR and IgG seroreversion half-life. In (B) and (D), the IgM seroreversion half-life is held constant at 50 days. Values associated with the highest log-likelihoods (shown here in light blue) indicate the best fit of the parameter values to the observed data.

Fig. 3.. Estimates of under-ascertainment of deaths in Khartoum.

(A) Daily and weekly mean reported COVID-19 deaths in Khartoum. We fit models to the reported COVID-19 deaths in (A) under different assumptions for what proportion of the true number of COVID-19 deaths these represent (reporting fractions). In (B), the resultant model fits were used to estimate the proportion of individuals aged over 15 years that would have experienced a symptomatic COVID-19 infection by 3 June 2020. The points and vertical bars show the median and 95% CI for each model fit and are compared against the observed cumulative number of symptomatic cases in Khartoum estimated from a social media–distributed survey (14), suggesting that 4% of COVID-19 deaths were detected. A cross-sectional household mortality and serosurvey conducted in Omdurman in March to April 2021 estimated seroprevalence to be 54.6% (95% CI, 51.4 to 57.8%) after adjusting for specific test performance (21). This estimate is depicted in (C) by point and whiskers and is compared against the adjusted seroprevalence (median and 95% CI shown in orange) predicted by a model fit with an assumed mortality reporting fraction of 4%. On the basis of the same survey, 20,766 (95% CI, 14,641 to 27,750) excess deaths are estimated to have occurred across Khartoum state. This estimate is depicted in (D) by point and whiskers and is compared against the cumulative number of COVID-19 deaths (median and 95% CI shown in purple) predicted by a model fit with an assumed mortality reporting fraction of 4%. In (C) and (D), the gray shaded area highlights the sampling period of the serosurvey and mortality survey in Omdurman.