Using mortuary and burial data to place COVID-19 in Lusaka, Zambia within a global context

Following the emergence of COVID-19 in 2019, SARS-CoV-2 spread globally, causing extensive social and economic disruption. Despite stringent non-pharmaceutical interventions (NPIs) and vaccines that reduced severe disease and transmission, Africa reported comparatively low cases and deaths per capita relative to global averages as of April 2023. Multiple hypotheses have been proposed to explain perceived low impact in Africa, including younger population structures, prior cross-reactive immunity, climate, effective NPIs, and genetic factors. However, substantial under-ascertainment due to limited testing and weak mortality registration complicates interpretation of reported metrics. Robust excess mortality data for African countries are scarce; WHO estimates for excess mortality in 2020 for most African countries have wide uncertainty due to data limitations, hampering inference about true burden and infection-fatality ratio (IFR). In Zambia, early 2020 data provided unique insights: a July 2020 population-based survey in Lusaka estimated 2.1% seroprevalence and 7.6% PCR positivity, greatly exceeding reported case numbers; post-mortem testing at the University Teaching Hospital (UTH) mortuary detected SARS-CoV-2 in 15% of deaths during June–October 2020. Burial registration data from mid-2017 to mid-2021 offer an opportunity to estimate excess mortality and transmission dynamics. The study asks whether age-patterns of COVID-19 severity in Lusaka during the first wave differed substantially from patterns elsewhere, contextualizes WHO excess mortality estimates by demographic vulnerability, develops a statistical framework to estimate pandemic impact using burial registration age distributions, and fits a transmission model to infer reproduction numbers, attack rates, and IFR, comparing Lusaka to global estimates.

The paper situates the research within a body of literature noting the apparent lower reported COVID-19 impact in sub-Saharan Africa. Proposed explanations include demographic advantages (younger populations), pre-existing immunity from other coronaviruses or endemic infections, climate and outdoor lifestyles, effective non-pharmaceutical measures, and genetic differences. Yet, under-ascertainment of infections and deaths is widely acknowledged, especially where testing capacity and civil registration systems are limited. Evidence from 2020 is notably sparse in Africa, with very few high-quality community serosurveys and limited excess mortality data. WHO’s global excess mortality assessment highlighted massive data gaps across the WHO Africa region, producing broad uncertainty bands for regional estimates. Prior Zambian studies during 2020 provided important context: a Lusaka community survey showed high PCR prevalence relative to reported cases, and UTH mortuary surveillance revealed substantial SARS-CoV-2 detection among the deceased. The study also references global IFR estimation efforts (e.g., Brazeau et al.) and notes challenges in IFR estimation in Africa due to incomplete ascertainment of infections and deaths and heterogeneous data quality across settings.

The study combines demographic, mortality, and epidemiological data with statistical and transmission modeling. Key components include: 1) Excess mortality estimation from burial registrations: Age-stratified all-cause weekly burial registrations for Lusaka (January 2018–June 2021) were analyzed. Due to high volatility in absolute weekly counts likely driven by registration service disruptions, the model focuses on changes in the age distribution of registrations. A Bayesian model (Metropolis-Hastings MCMC) was trained on 2018–2019 data to predict expected registrations in age groups 5+ during 2020–2021 using weekly <5 (U5) registrations as a baseline indicator of registration system function. Non-COVID-19 baseline registrations in each age-week group were modeled as Poisson-distributed. Predictions were cross-validated on pre-pandemic data and then compared to observed 2020–2021 registrations to estimate excess 5+ registrations. A supplementary analysis used 5–14 years as the baseline reference group to test robustness. To translate excess registrations into excess deaths, a weekly scaling factor was applied based on the ratio of U5 registrations to their 2018–2019 median, under the primary assumption that temporary declines in registration reflected process disruptions rather than true mortality changes. Assumed capture (registration coverage) of pre-pandemic deaths was set at 90% by default, with sensitivity at 80–100%. 2) Demographic vulnerability-weighted impact (DVWI): Age-specific global IFR estimates (Brazeau et al.) were weighted by Zambia’s population age structure to compute an overall IFR and a DVWI metric, defined as the uniform-by-age attack rate needed to produce observed excess mortality if directly attributable to COVID-19. This contextualizes excess mortality by demographic vulnerability. 3) Transmission modeling: An age-structured SEIR model parameterized for Lusaka demographics was fitted to multiple datasets during the first wave (June–October 2020): age-specific weekly burial registrations (limited to the first-wave window), UTH post-mortem PCR prevalence by age and week, and a July 2020 Lusaka population survey measuring PCR prevalence and seroprevalence. Due to lack of local contact data, a social contact matrix from peri-urban Nyanga, Manicaland, Zimbabwe, was used. The model estimated a time-varying reproduction number R0(t) at two-week intervals and epidemic start date. Likelihood components assumed Poisson-distributed burial registrations (sum of modeled COVID-19 deaths scaled to registry level plus estimated baseline) and binomial-distributed PCR/serology positives. MCMC sampling (eight chains, 30,000 samples each, with burn-in) was used to infer parameters and trajectories. 4) IFR inference and sensitivity analyses: Severity assumptions based on Brazeau et al.’s age-specific IFR curve were varied by adjusting the log-linear intercept (overall IFR level) and slope (age gradient) across wide ranges (e.g., 20–500% overall; 20–250% age-gradient). Model fit across combinations was compared via average posterior likelihood. Sensitivity analyses tested impacts of excluding outlying data (e.g., unusually high mortuary prevalence week), varying infection-to-death delays, alternative assumptions on U5 vs 5+ non-COVID mortality changes (±10% and ±20% relative shifts), removing the U5-based scaling (i.e., assuming changes reflected true mortality rather than registration disruption), and altering the assumed registration capture (80–100%). Ethics approvals and data sources are detailed, and code/data are provided via a public repository.

- Excess mortality: Relative to pre-pandemic patterns, the study estimates substantial age-dependent increases in mortality consistent with COVID-19 impact. Using scaled estimates and assuming 90% capture: 3220 excess deaths (95% Crl: 2256–4393) occurred in 2020 and 4419 (95% Crl: 3257–5783) during Jan 2020–June 2021, corresponding to 1256.2 (95% Crl: 880.1–1713.8) and 1723.9 (95% Crl: 1270.6–2256.0) per million population, respectively. This represents an 18.5% (95% Crl: 13.0–25.2%) increase in mortality relative to pre-pandemic levels during 2020 and 17.6% (95% Crl: 13.0–23.0%) during Jan 2020–June 2021. The abstract reports a comparable 3212 excess deaths (95% Crl: 2104–4591) during 2020. - Burial registration dynamics: Overall registrations declined across all ages during early NPIs (March–April 2020), likely due to registration process disruptions; age distribution and average age at death remained consistent with pre-pandemic during these declines. Surges in registrations during epidemic waves (mid-2020, early 2021, mid-2021) were driven by older ages, with weekly peaks 50% above pre-pandemic medians and average age at death peaking at 49.3 years (July 2020), 45.9 (Jan 2021, Beta wave), and 51.1 (June 2021, Delta wave). - DVWI: After adjusting for demographic vulnerability, Zambia’s 2020 DVWI based on WHO estimates allowed few conclusions due to wide uncertainty (median 0.251, 95% CI: 0–0.710). Using study excess mortality estimates, DVWI was markedly higher than WHO-based values: for 2020, DVWI median 1.149 (95% CrI: 0.805–1.568) under 90% capture; burial registration–only excess DVWI 0.589 (95% CrI: 0.431–0.742). - Transmission: The inferred R0(t) and Reff were around 3 before the first wave, falling below 1 in mid-July 2020. Estimated cumulative attack rate by October 2020 was ~24% (range across best-fitting models ~15–30%). R0(t) remained near Reff after July, suggesting limited immunity-driven reduction of transmission during the wave. - IFR patterns: Mortality and prevalence data are well explained by previously established age-specific IFR estimates (Brazeau et al.). Model fit was more sensitive to age-gradient than overall IFR scale; best fits centered near default assumptions. Plausible IFR ranges spanned roughly 50–250% of defaults across sensitivity analyses; Zambia’s demography-weighted IFR under default assumptions was ~0.11% (range with 80–167% overall severity: ~0.088–0.183%). - Consistency across data streams: The age-shift in mortality, mortuary PCR positivity (including 15% positive during June–October 2020), and community prevalence patterns were consistent with typical COVID-19 severity and spread, indicating substantial under-ascertainment of cases and deaths in official reports.

The findings indicate that Lusaka’s first COVID-19 wave had a substantial direct impact on mortality, with age distributions shifting toward older ages in a manner characteristic of COVID-19 severity globally. When contextualized for demographic vulnerability, excess mortality levels approached or exceeded those of heavily affected regions, contradicting the notion that Africa experienced intrinsically lower severity. Transmission modeling reconciled burial registration excesses with mortuary and community prevalence data using standard IFR-by-age assumptions, with no evidence of markedly different age-specific severity in Lusaka relative to global estimates. The inferred transmission dynamics—high transmissibility in May–June 2020 followed by declines below Reff=1 by mid-July—suggest that NPIs and behavioral changes mitigated the wave before widespread population immunity developed, leaving much of the population susceptible to subsequent waves (as reflected in early 2021 Beta and mid-2021 Delta waves). The study underscores that low reported cases and deaths were more likely due to under-ascertainment than truly low impact, highlighting the critical importance of robust mortality surveillance. The DVWI framework demonstrates that once demographic protection is accounted for, the burden in Lusaka aligns with, or exceeds, that in regions presumed to have higher impact. Policy implications include the need for equitable resource allocation and timely vaccine access; misconceptions about low severity in Africa may have hindered advocacy and uptake during 2021. Overall, methodological triangulation (burial registrations, mortuary surveillance, community surveys, and modeling) narrows uncertainty for Lusaka and challenges narratives of an ‘Africa paradox’.

The study integrates burial registry analysis, mortuary surveillance, community prevalence data, and transmission modeling to quantify COVID-19’s impact in Lusaka during 2020–mid-2021. It estimates large excess mortality, age-shifts consistent with COVID-19 severity, and transmission patterns similar to other global settings, finding no evidence for markedly different age-specific severity. After adjusting for Zambia’s youthful demographics, the burden is comparable to heavily affected regions, implying that low reported figures likely reflect under-ascertainment. The work demonstrates the feasibility and value of leveraging mortuary and burial data to infer epidemic dynamics where civil registration and surveillance are limited. Future research should: expand and strengthen vital registration and cause-of-death attribution systems; enhance integration of mortuary, community survey, and clinical data; refine local contact and care-access parameters; and develop standardized frameworks (e.g., DVWI) to contextualize excess mortality by demographic vulnerability, improving equitable decision-making in future pandemics.

Key limitations include: 1) Burial registration volatility and incomplete capture: registration systems experienced disruptions, and true capture rates are uncertain (assumed 90% in baseline; sensitivity 80–100%). 2) Assumptions about U5 registrations: the primary approach assumes changes in U5 registrations reflect registration process disruptions rather than true shifts in U5 mortality; although supported by correlation with other age groups and sensitivity analyses (including using 5–14 as baseline), residual bias is possible. 3) Attribution of excess mortality: excess deaths may include indirect effects (e.g., healthcare disruptions); DVWI does not distinguish direct from indirect causes. 4) External data and generalizability: lack of local contact data necessitated use of a Zimbabwean contact matrix; IFR inputs are largely derived from higher-income settings in early 2020 and may not fully reflect local care access or pathogen/variant differences. 5) Data sparsity and outliers: some weekly mortuary prevalence observations were high outliers; small age-stratified sample sizes increase uncertainty. 6) Timing and delay distributions: uncertainty in infection-to-death delays and epidemic start date can affect fits, though sensitivity analyses suggested limited impact. 7) Potential measurement biases in serology and PCR data and uncertainties in cause-of-death ascertainment remain. Together, these factors may affect the precise magnitude of estimates, though the main conclusions were robust across extensive sensitivity analyses.