Excess mortality during the COVID-19 pandemic in Aden governorate, Yemen: a geospatial and statistical analysis

Abstract

(ENGLISH) Background The burden of COVID-19 in low-income and conflict-affected countries is still unclear, largely reflecting low testing rates. In parts of Yemen, reports indicated a peak in hospital admissions and burials during May-June 2020. To estimate excess mortality during the epidemic period, we quantified activity across all identifiable cemeteries within Aden governorate in Yemen (population approximately one million) by analysing very high-resolution satellite imagery, and compared estimates to Civil Registry office records from the city. Methods After identifying active cemeteries through remote and ground information, we applied geospatial analysis techniques to manually identify new grave plots and measure changes in burial surface area over a period from July 2016 to September 2020. After imputing missing grave counts using surface area data, we used alternative approaches, including simple interpolation and a generalised additive mixed growth model, to predict both actual and counterfactual (no epidemic) burial rates by cemetery and across the governorate during the most likely period of COVID-19 excess mortality (from 1 April 2020), and thereby compute excess burials. We also analysed death notifications to the Civil Registry office during April-July 2020 and in previous years. Results We collected 78 observations from 11 cemeteries, of which 10 required imputation from burial surface area. Cemeteries ranged in starting size from 0 to 6866 graves. In all but one a peak in daily burial rates was evident from April to July 2020. Interpolation and mixed model methods estimated ≈ 1500 excess burials up to 6 July, and 2120 up to 19 September, corresponding to a peak weekly increase of 230% from the counterfactual. Satellite imagery estimates were generally lower than Civil Registry data, which indicated a peak 1823 deaths in May alone. However, both sources suggested the epidemic had waned by September 2020. Discussion To our knowledge this is the first instance of satellite imagery being used for population mortality estimation. Findings suggest a substantial, under-ascertained impact of COVID-19 in this urban Yemeni governorate, and are broadly in line with previous mathematical modelling predictions, though our method cannot distinguish direct from indirect virus deaths. Satellite imagery burial analysis appears a promising novel approach for monitoring epidemics and other crisis impacts, particularly where ground data are difficult to collect.

COVID-19: Is herd immunity the only option for fragile Yemen?

“…confirms this transmission trajectory in Yemen, suggesting a waning of the epidemic by September 2020 ( Fig. 1 ) ( Al Batati, 2020a , Besson et al, 2020 ). Since then, for more than five months, the WHO dashboard for COVID-19 in Yemen indicates only 322 of the total 2367 infections and 54 of the total 644 deaths till date ( WHO, 2021 ).…”

Categorizing the Status of COVID-19 Outbreaks Around the World

“…Such estimates have been made for many HICs and a few MICs – including Bolivia, Ecuador, Peru, Mexico and South Africa – by comparing total deaths in since the start of the COVID-19 pandemic with expected baselines based on deaths in recent years [9] . Estimates have been made using novel methods for several LICs, including for Sudan using a social media survey, a joint serological and molecular survey and modelling [10] , for Syria using community-uploaded obituary certificates and modelling [11] , for Yemen using geospatial analysis of burial surface areas in cemeteries [12] and for Zambia from a systematic post-mortem surveillance study [13] . By comparing the estimates of actual deaths with estimates of expected mortality rates, one can infer estimates for the share of population infected to date.…”

“…In this Annex, we present the data on total reported cases and deaths for each country. We also present the case fatality ratio (CFR), the ratio of reported deaths to reported cases, and the expected infection fatality ration (IFR), estimated from age-specific IFRs [12] and each country’s population by age. The data was downloaded on 2 February 2021.…”

“…Seroprevalence studies in Qatar [3] , Manaus (Brazil) [4] , Pakistan [5] , South Africa [6] and Nairobi (Kenya) [7] reported finding antibodies for SARS-CoV-2 in large percentages of the studied populations. Estimates of actual deaths due to COVID-19, combined with estimates of expected mortality rates, similarly suggest that SARS-CoV-2 has infected large shares of the populations in several L&MICs [8] , including Bolivia, Ecuador, Peru, Mexico and South Africa [9] , Sudan [10] , Syria [11] , Yemen [12] and Zambia [13] . In previous work, we showed that COVID-19 cases in some L&MICs could be described with the simplest of all epidemiological models, the Reed-Frost model, with disease outbreak parameters suggesting that the outbreaks had passed the point where cases were declining due to low susceptibility levels in the population, noting that control measures reduce the infection level at which cases start to decline [14] .…”

No excess mortality detected in rural Bangladesh in 2020 from repeated surveys of a population of 81,000

“…However, only a third of deaths are currently officially recorded within 45 days in Bangladesh (10). Considering the high excess mortality estimated from burial records in some countries without a functioning civil registry, such as Indonesia (11) and Yemen (12) for instance, the actual number of COVID-19 deaths could be considerably higher than the national data suggest. The possibility of under-reporting of a first wave of COVID-19 cases and fatalities in South Asia has been widely discussed but on the basis of limited direct evidence (5,13,14).…”