Underestimated COVID-19 mortality in WHO African region – Authors' reply

We appreciate Debbie Bradshaw and colleagues’ comments on the findings of our modelling analysis.¹ We note the authors’ areas of concern, which are based on two imprecise assumptions. First, the authors misinterpret the SCORE assessment as a measure of specific institutional capacity, yet it is meant to be a measure of national capacity. Compared with other countries in the African region, South Africa modified its vital statistics system to produce more accurate statistics on the official rate of mortality from COVID-19, validating its high SCORE assessment value. Second, the authors’ comments conflate the number of excess deaths with the number of COVID-19 deaths, which is incorrect. Excess mortality is associated with the COVID-19 pandemic in causality, not in the magnitude of COVID-19 deaths. The two measures are not necessarily coupled,² and a strong evidence base exists to suggest that excess mortality is considerably driven by deaths other than COVID-19.³ In many instances, deaths not related to COVID-19 have been shown to account for more than half of excess mortality.⁴ For instance, evidence shows an increase in the documented number of deaths from diabetes by 96%, heart disease by 89%, Alzheimer's disease by 64%, and stroke by 35%. These additional deaths are due to multiple causes associated with avoidance of care, orders to stay at home during lockdown, decreases in surveillance, and other interlinked causes.⁵

Additionally, there are methodological challenges associated with estimates of excess mortality. Notably, there is no internationally accepted standard for calculating these estimates. Bradshaw and colleagues⁶ used data adjusted for completeness from national population registers and negative binomial regression methods to calculate the number of deaths from COVID-19 from May, 2020, to January, 2022. This approach is more robust with overdispersed data than is Poisson regression. However, data gaps remain a challenge for this approach in Africa. The approach requires accurate information and predictable trends relating to past deaths to produce useful counterfactual data (ie, the predicted number of deaths). Both assumptions are problematic in Africa due to data gaps, and the considerable potential for year-on-year changes in mortality arising from varying interventions to address the high rates of avoidable mortality. Therefore, counterfactual data would be associated with high estimation errors and wide confidence limits, making comparisons with other data problematic.

As a result, the inferences from Bradshaw and colleagues’ analysis⁶ are difficult to apply at a policy level, without knowledge of the causes of deaths. The authors additionally suggest an excess 296 000 natural deaths (>3·0 times the estimated number of COVID-19 deaths) in South Africa. Against an estimated number of 560 000 annual deaths in South Africa (crude death rate 9·4 deaths per 1000 people), such an increase in mortality would be visible, particularly in Africa where deaths are largely public and sociocultural events. We welcome further discussions on the methods used to estimate COVID-19 mortality, which formed the basis of our modelling analysis,¹ and the policy implications that arise from these findings.

We declare no competing interests.