Global excess deaths associated with the COVID-19 pandemic

Excess mortality associated with the COVID-19 pandemic is used to quantify the direct and indirect impacts of the pandemic. Excess mortality is defined as the difference between the total number of deaths estimated for a specific place and given time period and the number that would have been expected in the absence of a crisis (e.g., COVID-19 pandemic). This difference is assumed to include deaths attributable directly to COVID-19 as well as deaths indirectly associated with COVID-19 through impacts on health systems and society, minus any deaths that would have occurred under normal circumstances but were averted due to pandemic-related changes in social conditions and personal behaviours.

Estimates of the excess mortality associated with the COVID-19 pandemic includes deaths from all causes. In certain locations, these estimates may include excess deaths associated with other crises such as extreme weather, disasters, or conflicts. 

While aggregate COVID-19 case and death numbers are being reported to WHO, they do not always provide a complete picture of the health burden attributable to COVID-19. In general, reported death numbers under-estimate the number of lives lost due to the pandemic, there are several reasons for this. They miss those who died without testing, they are contingent on the country correctly defining COVID as the cause-of-death and they miss the increases in other deaths that are related to the pandemic leading to overwhelmed health systems or patients avoiding care. A few countries have experienced lower than expected total deaths during the pandemic due to reduced contact and reduced mobility, which have led to reduced infectious disease related mortality as well as reduced transport and injury related fatalities. Reported COVID-19 death numbers do not account for this.

In light of the challenges posed by using reported data on COVID-19 cases and deaths, excess mortality is considered a more objective and comparable measure that accounts for both the direct and indirect impacts of the pandemic.

 

 

Negative excess deaths could be observed if deaths that would have happened in the absence of the pandemic were averted due to measures taken to deal with the pandemic. Some public health measures (e.g., lockdown, social distancing, mask wearing, working from home) have led to decrease in the number of deaths from causes other than COVID-19. For instance, a decreased number of deaths due to road traffic injuries and seasonal flu has been observed due to restricted movement of people. 

It is important to note that other groups and countries have produced excess death estimates. The difference in the estimates produced by WHO and those produced by other groups are due to key differences in the input data used and assumptions made including how the expected deaths are calculated, the statistical model used, and the variables used to predict deaths in locations where limited or no data is available. There might also be variations of the period/month/week assessed. To minimize these differences, uncertainty intervals are provided.

 

Mortality data to calculate actual deaths in real-time are available in only a subset of countries where reporting systems are functioning effectively, and historical datasets to calculate expected deaths are also often incomplete. Many countries do not have the mortality surveillance capacity to generate and collect data in a timely manner and these data gaps mean that excess mortality cannot be derived for all countries using standard methods.

The work of the TAG has been essential to establishing the methodology to model excess deaths where data has been unavailable and/or incomplete. This methodology is still under development and will likely be revised based on Member State feedback during the country consultation process.

 

The TAG is co-chaired by Professor Debbie Bradshaw (Chief Specialist Scientist, South African Medical Research Council), Dr Kevin McCormack (Head of Division for Sustainable Development Goals Indicators & Reports and Geographies, Irish Central Statistics Office) and Dr Oleg Chestnov (Fellow, Federal Research Institute for Health Organization and Informatics of the Ministry of Health of the Russian Federation). Member profiles and a list of observers can be found here.

 

 

WHO produces estimates using statistical modelling for all global health estimates, often with the advice of technical experts, to ensure robust statistical standards are followed and facilitate global comparability. This is the standard approach followed also by other UN agencies for global estimates. The model adopted for estimating excess deaths is not a ‘one-size-fits-all’ approach. WHO in collaboration with the TAG scrutinizes the relationships between excess mortality and certain contextually relevant variables in locations with good quality data to estimate excess mortality in countries with limited data.

This approach takes the specificity of countries (e.g., income level, reported COVID-19 deaths rate, test positivity rate, containment index) into consideration while ensuring global comparability. Countries may have their own approach to estimating excess deaths that may produce results that are different from those produced by WHO.

 

Tier classification is a simple grouping of countries based on mortality data availability. Countries are classified as Tier 1 countries if complete and nationally representative monthly all-cause mortality data for the specified period have been made available to WHO. Countries categorized as Tier 2 include countries for which WHO does not have access to the complete data and thus requires the use of alternative data sources or the application of scaling factors to generate the national aggregate. For these countries, WHO also utilizes subnational data where available. 

 

Tier classification is not a classification of health systems but one specific for and only applied to this current exercise. It is not a reflection of the public health advancement of a country nor of the legislation in place for mortality surveillance. It concretely takes into consideration clear inclusion and exclusion criteria based on data that have been made available to WHO.

 

 

In order to determine age-sex patterns of excess deaths, countries are categorized into eight groups using a K-means clustering approach. While a natural grouping would be geographically representative, using some regional identification (e.g., WHO regions), from a practical perspective it is not always possible owing to insufficient availability of empirical data in some countries or regions. Even within close geographic proximity, the impact of the pandemic, and its magnitude by sex and age can vary significantly. 

K-means is a method commonly used to characterize data and partition a data set into groups. Countries are grouped into clusters based on mean ages at death and of population, as well as the overall excess mortality rates and porportion of total deaths. The number of clusters is chosen to maximise the variation between clusters and to minimise the variation within clusters by age and sex. More details on the K-means clustering are provided in the methodology.

 

Every model is an approximation of reality. Models are subject to trade-offs, not least, the balance between comprehensiveness and comprehensibility. WHO’s estimates of the excess mortality associated with the COVID-19 pandemic are also approximations in a rapidly evolving pandemic with regard to transmissibility and severity. It is worth noting that WHO relies on models to compile a wide range of global health statistics. This is necessary as not all countries have high quality information systems, or those that do cannot always share their data in a timely manner. Therefore, WHO works with countries to strengthen their data and health information systems, while at the same time makes the best use of the data currently available to generate high quality evidence.

 

Specifically with regard to the excess mortality estimates, the model uses relationships that have been quantified using data from the countries with high quality systems for data reporting. We extrapolate these relationships to settings that are in many ways systematically different across multiple dimensions such as health systems capacity, underlying disease burden and age-structure. Countries have differed on when they experienced waves and how they responded, and quantifying this temporal relationships between COVID-19 surges or the effects of the emergence of variants is not trivial. In some settings the country responses were reactive but in many others they were proactive. As such, despite the effort to calibrate the model such that it is accurate in different settings, one of the more important limitations to note is that the input data have limited representation and this model generalizes the covariate relationships quantified to settings that may be systematically different from those observed. The final model is intentionally parsimonious. This reduction in complexity means it is not possible for such a reduced set of variables to explain all of the variation that is observed across countries. While the variables chosen are contextually relevant and are found to explain a significant component of the variation in excess, the underlying population characteristics that drove these differences are too complex to fully capture perfectly in a single global model and, hence, the trade-offs referred to above have had to be made. The estimates, therefore, are the best possible at this point in time for the first 24 months of the pandemic and have some degree of uncertainty around them.

 

In February 2021, in collaboration with the United Nations Department of Economic and Social Affairs (UN DESA), WHO convened a Technical Advisory Group (TAG) on COVID-19 Mortality Assessment to advise on the development of analytical methods for estimating excess mortality in all countries. The COVID TAG is comprised of leading demographers, epidemiologists, data and social scientists and statisticians from a range of backgrounds and geographies.

Under Working Group I of the TAG (Working Group I: Global excess mortality estimates including COVID-19), members considered several statistical models and after assessing performance, interpretability as well as extensibility, the TAG proposed a Poisson regression model (parameterized to account for over-dispersion) to predict the total number of deaths from all causes for the years 2020 and 2021, conditional on the monthly expected deaths over the period and a predicted relative rate parameter which is modelled using country-specific variables.

The model has been used by WHO to generate estimates for countries and regions for which adequate input data were available for reliable inference and to predict estimates for countries with no data available. In addition to determining levels of excess mortality associated with the COVID-19 pandemic for the years 2020 and 2021, the expertise of the TAG is also being leveraged to develop methods for disaggregating the estimated number of excess deaths by age and sex.

WHO’s estimates are produced following the Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER).

 

 

A death due to COVID-19 is defined for surveillance purposes as a death resulting from a clinically compatible illness, in a probable or confirmed COVID-19 case, unless there is a clear alternative cause of death that cannot be related to COVID disease (e.g. trauma). There should be no period of complete recovery from COVID-19 between illness and death. A death due to COVID-19 may not be attributed to another disease (e.g. cancer) and should be counted independently of preexisting conditions that are suspected of triggering a severe course of COVID-19. 

For any questions, please contact: [email protected]

 

The COVID-19 pandemic will have long lasting impacts on the structure of populations brought by changes in deaths and births. Substantial increases in mortality in many countries have affected premature mortality and life expectancy. WHO’s excess mortality estimates are critical inputs to WHO global health estimates and other UN partners’ work, including UN World Population Prospects.

The excess mortality estimates associated with the COVID-19 pandemic provide a valuable set of comparable country estimates to better understand the impact of the pandemic. It is thus of vital importance that these estimates are made available in a timely manner and updated periodically in order to identify inequalities and gaps in health information systems and civil registration and vital statistics (CRVS), which in turn will help in determining likely future vulnerabilities and target interventions.

For any questions, please contact: [email protected]