Global age-sex-specific mortality, life expectancy, and population estimates in 204 countries and territories and 811 subnational locations, 1950–2021, and the impact of the COVID-19 pandemic: a comprehensive demographic analysis for the Global Burden of Disease Study 2021

1Understanding mortality and population trends over time and across locations, age groups, and sexes is crucial for planning population-specific public health policies. Age-specific mortality rates can indicate the emergence of new adverse health risks in specific locations, while population counts can inform resource allocation and aid in planning future development. The COVID-19 pandemic has highlighted the importance of demography in understanding disease and injury burden and the roles health policy and infrastructure have in health and demographic outcomes., As the COVID-19 pandemic enters an endemic phase in some locations, demographic indicators can provide important context for understanding and addressing COVID-19, long COVID-19, and the interaction between COVID-19 and other diseases and injuries. Furthermore, demographic trends in the decades before the COVID-19 pandemic and reversals in those trends during the first 2 years of the COVID-19 pandemic (2020–21) can provide insights into potential long-term effects of the pandemic. These shifts in demographic patterns, including in population growth and age distribution, can help policy makers and public health experts better understand how the pandemic has impacted different groups within society and inform strategies for future pandemic preparedness and health-care planning.

2The Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) is an evolving research effort that quantifies the state of global health. The scope of the study has historically included estimating key demographic metrics and comprehensive health metrics for a set of national and subnational locations that has expanded over time. Mortality has been estimated as part of GBD since the first GBD estimates were published in the 1993 World Bank World Development Report, and mortality estimates have been included in each update since GBD 2010.,,,,, A comprehensive, internally consistent modelling strategy for estimating population and fertility was introduced in GBD 2017, greatly improving the consistency of results. Previously, GBD drew on population estimates from the UN Population Division of the Department of Economic and Social Affairs (UNPD)., In GBD 2019, the demographic analysis used population, fertility, and mortality estimates to produce a typology that better helped to specify phases of demographic transition. The GBD demography framework is part of the greater GBD enterprise; thus, it differs from other demographic research initiatives by using estimates of disease and injury burden to inform population and mortality estimates, and vice versa. Attempting to estimate the effects of the pandemic is now a major focus of GBD and other demographic research efforts.,,,

3The GBD 2021 demographic analysis improved on GBD 2019 by using additional data sources and refined methods to generate updated estimates of mortality, life expectancy, and population size at the global, regional, national, and subnational levels for each year from 1950 to 2021. GBD 2021 is the first round to incorporate the COVID-19 pandemic into the modelling process through the estimation of excess mortality due to the pandemic, defined as the net difference between the number of deaths that occurred between 2020 and 2021 and the number of deaths that would be expected over the same period based on previous trends in all-cause mortality. The unified approach to estimate all-cause mortality and excess mortality in GBD 2021 is an innovation in current demographic research methods. This facilitates analysis of the interplay between wider demographic processes and the COVID-19 pandemic. In this iteration of the GBD demographic analysis, we aim to provide policy makers and the public with the information needed to gain a better understanding of the demographic context of disease and injury burden since 1950 and during the COVID-19 pandemic in 2020–21 specifically.

4Evidence before this study The UN Population Division of the Department of Economic and Social Affairs (UNPD) produces estimates and projections of global, regional, and national demographic metrics that are updated biannually. Their latest findings, published in the World Population Prospects 2022 revision, incorporated WHO estimates of excess mortality due to the COVID-19 pandemic in 2020 and 2021. Estimates of excess mortality during the pandemic have also been generated by the Institute for Health Metrics and Evaluation and the World Mortality Dataset. The International Database of the US Census Bureau reports population estimates and projections for more than 200 countries and areas, of which a subset are updated every year. Organisations including WHO, the Organisation for Economic Co-operation and Development, and the European Union release demographic estimates less regularly and typically only for select metrics or locations. Some national statistics offices also produce their own demographic indicators. The Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) generates regularly updated and globally comparable health metrics, including mortality, life expectancy, and population estimates for past years, and forecasts up to the year 2100. The current GBD 2021 cycle is directly preceded by GBD 2019, which reported demographic estimates for 204 countries and territories for each year from 1950 through 2019. While each of these studies represent important efforts to provide insights into demographic estimates and the COVID-19 pandemic, only GBD estimates comply with the Guidelines for Accurate and Transparent Health Estimates Reporting, which identifies best practices for reporting global health estimates. Added value of this study GBD 2021 is one of the first studies to fully evaluate demographic trends in the context of the first 2 years of the COVID-19 pandemic. The study employed a unified framework to calculate excess mortality rates due to the COVID-19 pandemic along with a comprehensive set of demographic metrics including all-cause mortality, life expectancy, and population counts for 204 countries and territories and 811 subnational locations. This allowed estimates of all-cause mortality to inform estimates of excess mortality due to the pandemic, and vice versa. In contrast, the demographic estimates published by UNPD for 2020 and 2021, although based on data available during the pandemic, did not use a unified framework for all-cause and excess mortality. Additionally, while the US Census Bureau published population estimates for 2020 and 2021, the estimates were adjusted to reflect the effects of the pandemic for only a subset of locations. GBD 2021 utilised a suite of customised and validated data processing and modelling tools, systematically analysing thousands of data sources to produce global, regional, national, and subnational demographic estimates by age, sex, and Socio-demographic Index (SDI) level for each year from 1950 to 2021. Compared with GBD 2019, GBD 2021 utilised 5296 additional data sources. Additionally, the model life table system used in GBD 2021 was improved to provide more accurate mortality estimates for older age groups. All estimates are packaged within freely accessible data-sharing and visualisation tools. Implications of all the available evidence Our study highlights the impact of the first 2 years of the COVID-19 pandemic at a novel level of granularity, demonstrating unprecedented reversals in adult mortality and life expectancy trends at the global, regional, and national levels. Furthermore, globally comparable measures of excess mortality due to the pandemic show substantial variation in the burden experienced by different countries and territories. Our comprehensive set of demographic estimates provides a rich description of evolving long-term trends in mortality and life expectancy across age groups, sexes, and SDI levels, and our population analyses reveal changing dynamics and age structures with implications for the future of health-care systems, economies, and societies. Collectively, the estimates reported here provide an integrated demographic framework for GBD and a valuable foundation for policy evaluation, development, and implementation around the world.

5Evidence before this study

6The UN Population Division of the Department of Economic and Social Affairs (UNPD) produces estimates and projections of global, regional, and national demographic metrics that are updated biannually. Their latest findings, published in the World Population Prospects 2022 revision, incorporated WHO estimates of excess mortality due to the COVID-19 pandemic in 2020 and 2021. Estimates of excess mortality during the pandemic have also been generated by the Institute for Health Metrics and Evaluation and the World Mortality Dataset. The International Database of the US Census Bureau reports population estimates and projections for more than 200 countries and areas, of which a subset are updated every year. Organisations including WHO, the Organisation for Economic Co-operation and Development, and the European Union release demographic estimates less regularly and typically only for select metrics or locations. Some national statistics offices also produce their own demographic indicators. The Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) generates regularly updated and globally comparable health metrics, including mortality, life expectancy, and population estimates for past years, and forecasts up to the year 2100. The current GBD 2021 cycle is directly preceded by GBD 2019, which reported demographic estimates for 204 countries and territories for each year from 1950 through 2019. While each of these studies represent important efforts to provide insights into demographic estimates and the COVID-19 pandemic, only GBD estimates comply with the Guidelines for Accurate and Transparent Health Estimates Reporting, which identifies best practices for reporting global health estimates.

7Added value of this study

8GBD 2021 is one of the first studies to fully evaluate demographic trends in the context of the first 2 years of the COVID-19 pandemic. The study employed a unified framework to calculate excess mortality rates due to the COVID-19 pandemic along with a comprehensive set of demographic metrics including all-cause mortality, life expectancy, and population counts for 204 countries and territories and 811 subnational locations. This allowed estimates of all-cause mortality to inform estimates of excess mortality due to the pandemic, and vice versa. In contrast, the demographic estimates published by UNPD for 2020 and 2021, although based on data available during the pandemic, did not use a unified framework for all-cause and excess mortality. Additionally, while the US Census Bureau published population estimates for 2020 and 2021, the estimates were adjusted to reflect the effects of the pandemic for only a subset of locations. GBD 2021 utilised a suite of customised and validated data processing and modelling tools, systematically analysing thousands of data sources to produce global, regional, national, and subnational demographic estimates by age, sex, and Socio-demographic Index (SDI) level for each year from 1950 to 2021. Compared with GBD 2019, GBD 2021 utilised 5296 additional data sources. Additionally, the model life table system used in GBD 2021 was improved to provide more accurate mortality estimates for older age groups. All estimates are packaged within freely accessible data-sharing and visualisation tools.

9Implications of all the available evidence

10Our study highlights the impact of the first 2 years of the COVID-19 pandemic at a novel level of granularity, demonstrating unprecedented reversals in adult mortality and life expectancy trends at the global, regional, and national levels. Furthermore, globally comparable measures of excess mortality due to the pandemic show substantial variation in the burden experienced by different countries and territories. Our comprehensive set of demographic estimates provides a rich description of evolving long-term trends in mortality and life expectancy across age groups, sexes, and SDI levels, and our population analyses reveal changing dynamics and age structures with implications for the future of health-care systems, economies, and societies. Collectively, the estimates reported here provide an integrated demographic framework for GBD and a valuable foundation for policy evaluation, development, and implementation around the world.

11For each new GBD iteration, recently available data and improved methods are used to update the full time series of demographic estimates from 1950 to the latest year of analysis; GBD 2021 demographic estimates therefore supersede all previous estimates.

12The GBD 2021 demographic methods closely followed those used in GBD 2019. Improvements for GBD 2021 centred on a single framework to estimate both all-cause mortality and excess mortality due to the COVID-19 pandemic. The analytical process for computing internally consistent demographic estimates included six main components: (1) estimating age-specific fertility rates; (2) estimating under-5 and adult (age 15–59 years) mortality rates; (3) estimating age-specific mortality rates using a relational model life table system with HIV adjustments; (4) estimating excess mortality due to the COVID-19 pandemic and adjusting all-cause mortality estimates accordingly; (5) accounting for fatal discontinuities such as wars, famines, and natural disasters; and (6) estimating population sizes. To resolve discrepancies due to the inherent interdependent nature of population, mortality, and fertility estimates, the estimation process was run twice: first to generate preliminary numbers, and second to refine all estimates and ensure internal consistency. A detailed description of all methods and analytical flowcharts for all-cause mortality, fertility, and population estimation are available in appendix 1 (sections 2–6, 8).

13This study complies with the Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER); a completed GATHER checklist is provided in appendix 1 (section 8). Python (version 3.8.17 and 3.10.4), Stata (version 15.1), and R (version 3.5 and 4.2) were used for statistical analysis This manuscript was produced with the GBD Collaborator Network and in accordance with the GBD Protocol. An international network of collaborators provides, reviews, and analyses the available data to generate health metrics; the 2021 GBD round drew on the expertise of more than 11 000 collaborators across more than 160 countries and territories.

14The GBD 2021 analysis used a range of data types for mortality and population estimation that were identified from a systematic search of available data from government websites, statistical annuals, demographic compendia, large-scale surveys, and collaborator input; comprehensive details on the sources of input data are available online via the GBD 2021 Sources Tool. Under-5 mortality rates (U5MRs), defined as the probability of death from birth to age 5 years, were estimated using 30 526 location-years of vital registration data (3179 new location-years for GBD 2021 compared with GBD 2019), 237 location-years of sample vital registration data, and 1445 other sources (including 57 new surveys, one new census, and ten other new sources; appendix 1 section 8). Adult mortality, defined as the probability of death before age 60 years assuming survival to age 15 years, was estimated using 30 207 location-years of vital registration data (3150 new location-years for GBD 2021 compared with GBD 2019), 1435 location-years of sample vital registration data, 75 censuses, 280 surveys (including 65 sources of household death data and 167 sources of sibling history data), and 24 other sources (appendix 1 section 8). Age-specific mortality was estimated using 43 758 empirical life tables for 1950–2021 (compared with 35 406 in GBD 2019; appendix 1 section 8). Prevalence surveys, antenatal clinic serosurveillance, and vital registration were used to adjust for the impact of the HIV epidemic due to its exceptional impact on age-specific mortality. Fatal discontinuities were accounted for using 2235 location-years from vital registration and 237 other sources (compared with 1812 from vital registration and 174 other sources in GBD 2019). Estimation of excess mortality due to the COVID-19 pandemic utilised an additional 146 139 datapoints of all-cause mortality data at either weekly or monthly intervals from vital registration and surveillance reports that were assessed for completeness of registration (compared with our previous excess mortality estimation, GBD 2021 used 1389 additional weeks or months of data).

15Population estimates utilised national and subnational censuses (1277 overall; 25 new), population registries (749 location-years of data), and post-enumeration surveys (161 in total). Additionally, migration data on refugee movements from the UN High Commissioner for Refugees and datasets for select countries (primarily Gulf States and nations in the EU) were used to inform migration estimates.

16GBD 2021 all-cause mortality estimation followed the analytical framework for mortality analysis used in GBD 2019. Point estimates from surveys were generated using both direct and indirect estimation methods for U5MR, while for adult mortality, they were generated from sibling history data with methods that correct for inherent biases such as zero-survivor and recall bias. Time series estimates of the completeness of adult vital registration data were generated using the same modelling process as GBD 2019, which used a combination of five death distribution methods, and point estimates were adjusted accordingly.

17Time series of under-5 and adult mortality without fatal discontinuities were estimated using spatiotemporal Gaussian process regression (ST-GPR), including a bias-adjustment process for U5MR, to correct for systematic differences in the data sources and smooth results across time and location. Education, HIV, and lag-distributed income were included as covariates, along with U5MR for adult mortality. These estimates were used as inputs for the GBD relational model life table system with adjustments for older-age mortality to estimate HIV-free age-specific mortality rates. HIV mortality was modelled with a combination of ST-GPR, the Estimation and Projection Package Age-Sex Model, and Spectrum, and subsequently used to produce life tables that included HIV mortality. These abridged life tables were used to generate full life tables by single year age groups with further detailed age groups under the age of 1 year. Sex-redistributed and age-redistributed fatal discontinuities by cause were aggregated by age and sex and added to the estimated mortality from the previous step to generate the final all-cause mortality life tables by location, year, sex, and age. We recalculated abridged life tables, including fatal discontinuities for each location, year, and sex combination, and then calculated the final envelope from these abridged life tables. Detailed methods for estimating each mortality component are available in appendix 1 (section 2).

18Excess mortality due to the COVID-19 pandemic in 2020 and 2021 is defined as the observed all-cause mortality minus the mortality that would be expected had the pandemic not occurred, based on historical trends. Excess deaths are those attributed to the COVID-19 pandemic as a whole, both from SARS-CoV-2 infection and from other pandemic-related factors such as deferred care seeking., Excess mortality was calculated using similar methods as in Wang et al (2022), with several key improvements. We included yearly observed deaths from vital registration to supplement daily, weekly, and monthly observed death data. We then used five variants of the spline for weekly seasonal patterns that set the second-to-last knot at 18, 24, 36, 48, or 60 months to allow for more stable trends. To select covariates, we used Rover, a method developed at the Institute for Health Metrics and Evaluation based on Bayesian model averaging. Rover is conceptually similar to the Bayesian model averaging method, which is widely used to explore the parameter space and aggregate estimates across candidate models based on performance metrics. The main difference is that while Bayesian model averaging uses marginal likelihood, Rover focuses on out-of-sample performance. We included covariates pertaining to the COVID-19 pandemic, such as seroprevalence, and background population health metrics, such as the Healthcare Access and Quality Index. With the best model selected, we ran a prediction process using 100 draws for each covariate and 100 draws of estimated coefficients and residuals, estimated from the regressions run at the draw level using draw-level input data on both excess mortality and covariates. Mean values and 95% uncertainty intervals (UIs) were then generated at national, regional, and global levels. Out-of-sample predictive validity testing was conducted based on our final model specification. Complete excess mortality methodology is detailed in appendix 1 (section 2.8).

19To determine age-specific and sex-specific excess mortality, we estimated all-cause mortality twice: once with data from during the pandemic in 2020 and 2021 included and once without. For location-years with vital registration data from during the pandemic, we computed the difference in estimated age-sex-specific mortality between the two sets of estimates. We then applied this distribution to our excess mortality estimates to calculate age-specific and sex-specific excess mortality. Due to instability in age-sex distributions and implausible patterns, we used the global age-sex distribution for locations with fewer than 75 000 excess deaths, unless otherwise noted (appendix 1 section 2.8). Other pandemic-related mortality (OPRM) was estimated by calculating the difference between excess mortality and the sum of deaths due directly to COVID-19 infection and indirect deaths due to lower respiratory infections, measles, and pertussis. For locations with a negative OPRM, we adjusted the non-pandemic mortality estimates downward accordingly. We redistributed small discrepancies that remained between the mortality estimates that used vital registration age-sex-specific data from during the pandemic and the non-pandemic mortality estimates plus age-sex-specific excess mortality to ensure that the final mortality estimates including mortality shocks were consistent with observed high-quality vital registration data.

20We used the Bayesian hierarchical cohort component model for population projection (BCCMP) from GBD 2019 to produce age-specific population estimates. This method used age-specific fertility estimates from GBD 2021 (appendix 1 section 3), the previously described age-specific mortality estimates, and available census and registry data as inputs. Auxiliary refugee and migration data were used to inform the prior distribution on net migration in countries with substantial migration or reliable data. The model estimates an age-specific 1950 baseline population, age-specific net migration, and age-specific population estimates that are fully consistent with the input fertility and mortality estimates. Complete population estimation methodology is in appendix 1 (section 4).

21We analysed the relationship between age-specific log mortality rates and SDI using MR-BRT (meta-regression-Bayesian regularised trimmed), a meta-regression programme (appendix 1 section 6.1). SDI is a composite indicator of a country's lag-distributed income per capita, average years of schooling, and the total fertility rate in females younger than age 25 years (appendix 1 section 5). MR-BRT defines a linear mixed-effects model with a B-spline specification for the relationship between outcomes of interest and SDI. We used a cubic spline with five knots between 0 and 1, with left-most and right-most spline segments enforced to be linear, and with slopes matching adjacent interior segments. To ensure that the results were not sensitive to the choice of spline knots, we used a model ensemble of over 50 cubic spline models, as described above. For each model, interior knot placement was randomly generated to be between 0·1 and 0·9, with minimum inter-knot distance of 0·1 and maximum inter-knot distance of 1·0. The final predictions were obtained using the ensemble aggregate over these 50 models. This model was performed separately for each GBD age-sex group. Expected mortality rates for each age-sex group based on SDI were used to estimate expected life expectancy. A similar analysis was done for excess mortality rates due to the COVID-19 pandemic, with the exception that two-degree splines were used.

22We produced estimates for each demographic metric by age-sex-location-year for 25 age groups: early neonatal (0–6 days), late neonatal (7–27 days), 1–5 months, 6–11 months, 12–23 months, 2–4 years, 5–9 years, every 5-year age group up to 95 years, and 95 years and older (fertility estimated for 5-year age groups between ages 10 years and 54 years); for males, females, and all sexes combined; for 204 countries and territories grouped into 21 regions and seven super-regions; and for every year from 1950 to 2021. We also included subnational analyses for 21 countries and territories (Brazil, China, Ethiopia, India, Indonesia, Iran, Italy, Japan, Kenya, Mexico, New Zealand, Nigeria, Norway, Pakistan, the Philippines, Poland, Russia, South Africa, Sweden, the UK, and the USA) and estimates by SDI quintile. All countries and territories were assigned an SDI value ranging from 0 (lowest income and educational attainment and highest fertility) to 100 and then grouped into quintiles from low SDI to high SDI.

23Uncertainty was propagated throughout the estimation process. For under-5 and adult mortality, ST-GPR generated 1000 draws for every location, year, and sex combination; 1000 draws were also produced for the crude death rate associated with HIV estimates. The 100 draws of excess mortality due to the COVID-19 pandemic were repeated ten times to generate 1000 draws. These draw-level inputs were then used to create 1000 draws of all-cause mortality estimates and draw-level estimates of fatal discontinuities. Mean estimates and 95% UIs (the 25th and 975th ranked values from the 1000 draws) were generated for all demographic metrics using the draw-level estimates. The uncertainty associated with fertility and mortality estimates was included as inputs in the BCCMP model to produce 1000 draws of population estimates.

24The funders of this study had no role in study design, data collection, data analysis, data interpretation, or the writing of the report

25This section presents global, regional, and national-level results for key demographic metrics; given space constraints, estimates at the subnational level are presented in appendix 2 and are also available in downloadable form through the GBD Results tool. All subnational locations are listed in appendix 1 (section 8).

26This section presents global, regional, and national-level results for key demographic metrics; given space constraints, estimates at the subnational level are presented in appendix 2. All subnational locations are listed in appendix 1 (section 8).

27The proportion of deaths registered in vital registration systems increased substantially at the global level during the study period, from 30·3% in 1975 to a peak of 61·1% in 2016, before declining in subsequent years due to lags in reporting (figure 1). Completeness of death registration in vital registration systems varied markedly between regions, however, most progress in completeness was observed in China (where completeness peaked at 71·2% in 2018) and India (where completeness peaked at 80·1% in 2019; appendix 2 table S1). The Indian Sample Registration System is considered complete for the sample population it covers. Outside of China and India, progress in death registration has been slow, with only a 10·3 percentage point increase observed in the rest of the world between 1975 and the peak in 2016. This increase was concentrated in north Africa and the Middle East, which improved from 20·6% completeness in 1975 to a peak of 56·0% in 2016. While registration has been complete (defined as >95%) since 1975 for nearly all countries in the high-income super-region and central Europe, eastern Europe, and central Asia, in sub-Saharan Africa peak completeness of only 8·7% was reached in 2008 and completeness has declined since then. Death registration in Latin America and the Caribbean was more variable: countries such as Costa Rica, Cuba, and Argentina have been complete for many years; registration in countries such as Peru and Ecuador has remained around 60–90% complete, and others, such as Bolivia, continue to lack registration data. At the national level, 96 countries and territories had at least 1 year of complete death registration between 2010 and 2021; 29 countries and territories without complete death registration had at least 1 year of registering more than 75% of deaths; and 47 countries and territories had no vital registration data in the GBD 2021 mortality database. Registration was incomplete or non-existent in many countries with large numbers of deaths in 2021, especially in sub-Saharan Africa, including Nigeria and Democratic Republic of Congo. In the 2020–21 period, super-regions had varying degrees of lowered completeness indicative of lags in reporting (figure 1).

28Between 1950 and 2019, global age-standardised all-cause mortality rates per 100 000 population broadly declined, from 1980·5 age-standardised deaths (95% UI 1855·5–2115·0) in 1950 to 736·1 (700·1–772·8) in 2019 (appendix 2 table S3A), which equates to a 62·8% (60·5–65·1) decline in mortality during the entire period. Global all-cause mortality rates across the human lifespan for the younger than 15 years and older than 40 years age groups broadly improved for both females and males between 1950 and 2019 (figure 2). This pattern was relatively consistent across super-regions, with the exception of increased mortality in sub-Saharan Africa during the HIV epidemic and a fluctuating pattern in the central Europe, eastern Europe, and central Asia super-region. However, substantial variation in mortality levels and trends across super-regions and over time were observed in the 15–39-years age group. This age group was particularly susceptible to mortality shocks such as famine in China between 1959 and 1961; conflicts in the Middle East during multiple time periods; war in India, Pakistan, and Bangladesh and genocide in Bangladesh in 1971; war and genocide in Cambodia in the 1970s; the Rwandan genocide in 1994; and the earthquake in Haiti in 2010 (figure 2). Conflict and war had a larger impact on mortality rates in males than females. Furthermore, the HIV epidemic had an especially large impact on this age group in sub-Saharan Africa and a lesser impact in southeast Asia, east Asia, and Oceania, with a larger impact on females than males. Additionally, male mortality rates increased in Latin America and the Caribbean during the 2000s, to varying extents in countries such as El Salvador, Peru, Guatemala, Honduras, Mexico, Venezuela, and Brazil (appendix 2 figure S5). An increase in male and female mortality was observed in the high-income super-region during the late 2010s, which was most notable in the USA, Canada, and Spain (appendix 2 figure S5).

29During the COVID-19 pandemic in 2020 and 2021, global age-standardised all-cause mortality rates increased by 21·9% (95% UI 13·6–31·1) for males aged 15 years and older compared with 2019 and 16·6% (10·0–23·4) for females in the same age group and time period, reversing trends in mortality observed before the pandemic (appendix 2 table S3). In contrast, during 2020 and 2021, global mortality rates for both males and females generally remained constant or further decreased for age groups younger than 15 years (figure 2). In particular, between 2019 and 2021, global U5MR decreased by 7·0% (2·3–11·1). This continued reduction in child mortality was consistent across all super-regions (figure 2).

30All-cause mortality rates differed between sexes, and the extent of this difference varied across age groups and by location. Female mortality was generally lower than male mortality in all age groups, with substantial heterogeneity across countries and territories (figure 3). The highest variability in the ratio of male to female mortality rates across countries and territories was found in the 15–39 age groups; although little change in the mortality sex ratio has been observed between locations over time, the ratio generally increased between 1970 and 2021, indicating that the gap between male and female mortality has been increasing, generally driven by mortality rates among females decreasing at a faster rate than among males. Globally in 2021, the mortality rate for males aged 15–39 years was 65·9% (95% UI 56·8–74·7) higher than for females. The widening gap between males and females was also observed for nearly all age groups aged 40 years and older. In the neonatal age groups, the ratio of male to female mortality rates declined slightly over time towards 1, while the variability among countries and territories remained similar. Individuals aged 40 years and older had a consistent pattern of an increasing ratio of male to female mortality rates over time, with increased variability observed among those aged 65 years and older across countries and territories from 1970 to 2000, followed by little change in variability from 2000 to 2021.

31Despite declines in age-standardised all-cause mortality rates during the study period, the global number of deaths due to all causes combined increased from 44·0 million (95% UI 40·3–47·7) in 1950 to 50·3 million (49·3–51·4) in 2000 and 57·0 million (54·9–59·6) in 2019, largely reflecting a growing population and changing age structures. Global deaths further increased to 63·1 million (60·6–65·9) in 2020 and 67·9 million (65·0–70·8) in 2021, a notable spike attributable to the COVID-19 pandemic (table 1). Since 1970, the number of global deaths in the 25 years and older age group had increased steadily, until an unprecedented increase in 2020–21 (figure 4). This increase was observed across all GBD super-regions, with the exception of central Europe, eastern Europe, and central Asia, from 2000 to 2019. In contrast, deaths in children under 5 years declined over the entire study period, including during the COVID-19 pandemic period, with death counts of 20·0 million (17·2–23·0) in 1950, 9·21 million (8·73–9·73) in 2000, 5·21 million (4·50–6·01) in 2019, 4·89 million (4·19–5·71) in 2020, and 4·66 million (3·98–5·50) in 2021 (appendix 2 table S1). Initially, most of this decline could be attributed to declines in both U5MR and the under-5 population in southeast Asia, east Asia, and Oceania (especially China) until a tapering off around the year 2000. After this, the share of the decline attributed to sub-Saharan Africa began to increase, and this pattern continued during 2021 (figure 4). The largest number of under-5 deaths was observed in south Asia and sub-Saharan Africa during the pandemic, with south Asia accounting for 25·7% (24·1–27·2) of all deaths in children under 5 years in 2020 and 25·3% (24·0–26·6) in 2021, and sub-Saharan Africa accounting for 55·5% (53·2–57·7) in 2020 and 56·3% (54·1–58·4) in 2021. The number of global deaths in the intermediate age group (ages 5–24 years) demonstrates large yearly variability with no clear patterns, since deaths in this age group were heavily impacted by mortality shocks such as the Rwandan genocide in 1994 and natural disasters such as the earthquake in Haiti in 2010. Deaths in this age group increased slightly during 2020 and 2021 in most super-regions, but these increases were minimal compared with previous years, and in comparison to the increase observed in ages 25 years and older.

32Historically, global life expectancy at birth has increased steadily; between 1950 and 2021, global life expectancy at birth increased by 22·7 years (95% UI 20·8 to 24·8), from 49·0 years (46·7 to 51·3) to 71·7 years (70·9 to 72·5; table 1; appendix 2 table S4). Life expectancy improved for females from 51·6 years (49·4 to 53·8) in 1950 to 76·0 years (75·2 to 76·7) in 2019 and for males from 46·7 years (44·3 to 49·2) in 1950 to 70·8 years (69·9 to 71·7) in 2019 (figure 5). At the super-region level, the largest increases in life expectancy occurred in south Asia and north Africa and the Middle East, while at the national level, some of the largest increases were in South Korea and Iran (appendix 2 table S4). During this time period, the smallest gains in life expectancy occurred in the central Europe, eastern Europe, and central Asia and high-income super-regions and, at the national level, in Ukraine and Lesotho. Increasing life expectancy was generally consistent across all super-regions over the entire period, with the exception of mortality shocks in several locations, stagnation in sub-Saharan Africa during the HIV/AIDS epidemic, and slow progress in central Europe, eastern Europe, and central Asia before the mid-2000s. In 2020 and 2021, however, these trends reversed. Between 2019 and 2021, global life expectancy declined by 1·6 years (1·0 to 2·2); all super-regions had decreases in life expectancy during this period, ranging from a 3·7 year (3·4 to 4·1) decline in Latin America and the Caribbean to a 0·3 year (–1·9 to 1·3) decline in southeast Asia, east Asia, and Oceania (appendix 2 table S4). An increase in life expectancy during this period was only observed in 32 (15·7%) of 204 countries and territories.

33We estimated 5·89 million (95% UI 5·48–6·44) excess deaths globally attributable to the COVID-19 pandemic in 2020 and 9·97 million (9·26–10·9) excess deaths in 2021 (table 1). The GBD super-regions with the highest all-age excess mortality rates in 2020 and 2021 combined were central Europe, eastern Europe, and central Asia (269·7 excess deaths per 100 000 population [250·0–289·6]) and Latin America and the Caribbean (199·0 [184·7–215·4]). The super-regions with the lowest all-age excess mortality rates during this time period were southeast Asia, east Asia, and Oceania (23·8 [8·9–44·1]) and high-income (90·2 [87·2–93·2]; appendix 2 figure S2). At the national level, in 2020 and 2021 combined, all-age excess mortality rates were highest in Bulgaria (520·8 [382·0–630·0]) and Lesotho (447·0 [379·3–514·0]), the highest rate in 2020 was in Peru (413·4 [410·3–416·1]), and the highest rate in 2021 was in Bulgaria (697·5 [532·4–830·5]; appendix 2 figure S2). For seven countries and territories (Taiwan [province of China], Mongolia, Japan, New Zealand, Iceland, Antigua and Barbuda, and Barbados), the all-age excess mortality rate for 2020 and 2021 combined was negative, indicating that fewer deaths occurred in these locations during the first 2 years of the pandemic than what would be expected based on past trends. In 2020, 20 countries and territories had negative excess mortality, while in 2021, only New Zealand and Barbados had negative excess mortality (table 1).

34Additionally, we computed age-standardised excess mortality rates to compare the impact of the pandemic across countries and territories while controlling for different population age structures. Age-standardised rates and all-age rates differed substantially, with the highest age-standardised excess mortality rates observed in nations in sub-Saharan Africa, Latin America, and the Middle East (figure 6). The lowest age-standardised rates were found in some countries and territories in the Caribbean, east Asia, and Oceania, and some high-income nations. There was substantial variability within all super-regions. The countries or territories with the highest age-standardised rates during 2020 and 2021 combined were Eswatini (992·5 age-standardised excess deaths per 100 000 population [95% UI 745·5 to 1173·2]), Lesotho (874·3 [734·7 to 1009·4]), and Somalia (715·6 [549·3 to 912·7]); the nations with the lowest rates were Barbados (–61·5 [–111·6 to –13·1]), Mongolia (–32·9 [–209·6 to 131·0]), and Antigua and Barbuda (–13·7 [–55·5 to 27·9]).

35Between 1950 and 2021, longer life expectancies at birth were generally associated with higher SDI levels (figure 7; table 2). For females in 2021, the super-regions with the largest proportion of nations with a life expectancy higher than expected based on SDI were high-income (31 of 36 nations), south Asia (three of five nations), and Latin America and the Caribbean (16 of 33 nations), while central Europe, eastern Europe, and central Asia (23 of 29 nations), sub-Saharan Africa (35 of 46 nations), and north Africa and the Middle East (14 of 21 nations) had the highest proportion of nations with a lower life expectancy than expected based on SDI. For males in 2021, the GBD super-regions with the largest proportion of nations with a life expectancy greater than expected based on SDI were high-income (31 of 36 nations), south Asia (three of five nations), and north Africa and the Middle East (11 of 21 nations); the super-regions with the highest proportion of nations displaying a life expectancy lower than expected based on SDI were central Europe, eastern Europe, and central Asia (24 of 29 nations), sub-Saharan Africa (34 of 46 nations), and southeast Asia, east Asia, and Oceania (24 of 34 nations). Between 1950 and 2021, an increase in both life expectancy at birth and SDI was observed in all countries and territories. For females in 2021, the five countries or territories with the largest positive difference between estimated life expectancy and expected life expectancy based on SDI were Somalia (13·9 years), Niger (10·0 years), Spain (6·5 years), Portugal (6·0 years), and Singapore (5·6 years); the five countries or territories with the largest negative difference were Lesotho (–19·6 years), Eswatini (–17·9 years), Botswana (–12·8 years), Equatorial Guinea (–12·5 years), and Zimbabwe (–12·5 years; table 3). For males in 2021, the five countries or territories with the largest positive difference between estimated life expectancy and expected life expectancy based on SDI were Somalia (12·2 years), Niger (10·6 years), the Maldives (8·4 years), Bhutan (7·1 years), and Singapore (6·7 years); the five countries or territories with the largest negative difference were Lesotho (–21·2 years), Eswatini (–18·7 years), Zimbabwe (–13·4 years), South Africa (–12·8 years), and Botswana (–12·4 years; table 4).

36In 2020 and 2021 combined, lower age-standardised excess mortality rates due to the COVID-19 pandemic were broadly associated with higher SDI levels, but the association was not consistently strong (figure 8). The GBD super-regions with the largest proportion of countries and territories with an excess mortality rate higher than expected based on SDI were central Europe, eastern Europe, and central Asia (26 of 29 nations), Latin America and the Caribbean (21 of 33 nations), and south Asia (three of five nations); the super-regions with the largest proportion of nations with an excess mortality rate lower than expected based on SDI were southeast Asia, east Asia, and Oceania (33 of 34 nations), high-income (33 of 36 nations), and sub-Saharan Africa (27 of 46 nations). At the national level, the five countries or territories with the largest positive difference between estimated excess mortality and expected excess mortality based on SDI (ie, higher mortality than expected) were Bulgaria, North Macedonia, Lesotho, Peru, and Bolivia; the five nations with the highest negative difference between estimated excess mortality and expected excess mortality based on SDI (ie, lower mortality than expected) were Barbados, Mongolia, New Zealand, Antigua and Barbuda, and the Marshall Islands.

37The global total population increased annually over the study period, from 2·52 billion (95% UI 2·48–2·58) in 1950 to 6·10 billion (5·98–6·22) in 2000 and 7·89 billion (7·67–8·13) in 2021 (table 5). Annual growth in total population fluctuated over the study period, from an annual increase of 46·9 million (41·0–52·7) from 1950 to 1951 with the highest annual increase of 92·5 million (75·7–106·6) observed between 2008 and 2009 (figure 9). After 2009, population growth plateaued, and in 2017, the annual increase in population began to decline. Between 2019 and 2021, this decline accelerated, with annual gains of just 77·0 million (49·4–95·6) from 2019 to 2020 and 69·0 million (50·8–93·2) from 2020 to 2021. These reduced gains include the impact of excess deaths due to the COVID-19 pandemic, therefore the magnitude might not persist as excess mortality declines. The majority of global population growth during the study period is attributed to three GBD super-regions: sub-Saharan Africa; south Asia; and southeast Asia, east Asia, and Oceania. The population of sub-Saharan Africa grew at a steadily increasing rate throughout the study period, contributing 9·1% (7·3–11·0) of the total global population growth from 1950 to 1951, 23·3% (19·4–27·6) from 2000 to 2001, and 39·5% (28·4–52·7) from 2020 to 2021. South Asia contributed 17·1% (13·8–20·6) of the total global population growth from 1950 to 1951, rose to a peak contribution of 32·9% (28·4–37·8) from 1999 to 2000, and remained relatively constant in more recent years, with a contribution of 26·3% (9·0–44·7) from 2020 to 2021. In contrast, the annual growth of the population fluctuated in southeast Asia, east Asia, and Oceania. The contribution of this super-region to annual global population growth was relatively stable up to a peak of 37·3% (30·4–41·8) from 1956 to 1957 and then subsequently decreased, contributing 14·1% (0·0 to 30·2) from 2020 to 2021. Central Europe, eastern Europe, and central Asia contributed little to global population growth, and in fact experienced a decline in population over some periods, with growth from 1950 to 1992, a decline from 1993 to 2006, growth from 2007 to 2018, and a return to population decline in 2019. Population growth was relatively stable in Latin America and the Caribbean and north Africa and the Middle East at the super-regional level during the previous three decades, whereas population growth in the high-income super-region began to decline starting around 2015.

38The majority of countries and territories (154 [75·5%] of 204 countries and territories representing all seven super-regions) had a positive rate of natural increase (calculated as the number of births minus the number of deaths divided by person-years) between 2000 and 2009 followed by a smaller positive rate between 2010 and 2019 (figure 10). 26 countries and territories had a rate of natural increase that was positive during both decades and that was larger between 2010 and 2019 than between 2000 and 2009 (figure 10). Of these countries and territories, nine were in sub-Saharan Africa, eight were in central Europe, eastern Europe, and central Asia, and five were in the high-income super-region. Seven countries and territories had a positive rate of natural increase between 2000 and 2009 followed by a negative rate of natural increase between 2010 and 2019: Bosnia and Herzegovina, Greece, Japan, North Macedonia, Poland, Portugal, and San Marino (figure 10). The countries and territories of Belarus, Estonia, Latvia, Russia, and Ukraine experienced a negative rate of natural increase between 2000 and 2009 and continued to have a negative rate of natural increase between 2010 and 2019, but to a smaller extent (figure 10). The rate of natural increase was negative between 2000 and 2009 in Bulgaria, Croatia, Germany, Hungary, Italy, Lithuania, Moldova, Monaco, Romania, and Serbia, and to an even larger extent between 2010 and 2019 (figure 10). Of the 204 countries and territories, peak population was reached between 1950 and 1969 in three countries and territories, between 1970 and 1989 in eight countries and territories, between 1990 and 2009 in 23 countries and territories, between 2010 and 2021 in 22 countries and territories, and the peak population had not yet been reached as of 2021 in 148 countries and territories.

39The age structure of populations changed substantially across the globe between 1950 and 2021, with a general shift in the distribution away from younger ages and towards older ages (table 5). From 2000 to 2021, the proportion of the population aged younger than 15 years decreased in 196 of 204 countries and territories, with some of the largest declines observed in Saudi Arabia (from 36·0% to 20·1%) and Syria (41·5% to 26·1%). The eight countries in which the proportion of the population aged younger than 15 years did not decline were Angola, Chad, Kazakhstan, Mali, Niger, Nigeria, Russia, and Somalia. During this same period, the proportion of the population aged 65 years and older increased in 175 of 204 countries and territories; some of the largest increases were observed in Japan (from 17·2% to 28·9%) and Puerto Rico (from 11·0% to 22·0%). Three of 204 countries and territories had an increase in the proportion of the population aged younger than 15 years combined with a decline in the proportion of the population aged 65 years and older; these nations (Mali, Nigeria, and Chad) are all located in sub-Saharan Africa. The ratio of the population aged 65 years and older to the population aged less than 15 years increased between 2000 and 2021 in 188 of 204 countries and territories, including all nations within the high-income; Latin America and the Caribbean; south Asia; and southeast Asia, east Asia, and Oceania super-regions (figure 11). Some of the largest increases occurred in Japan, Puerto Rico, and South Korea. The countries and territories in which this ratio did not increase were Afghanistan, Benin, Burkina Faso, Burundi, Cameroon, Chad, Democratic Republic of Congo, Guinea, Guinea-Bissau, Kyrgyzstan, Liberia, Mali, Mozambique, Nigeria, Sierra Leone, and South Sudan.

40Our comprehensive set of updated demographic metrics indicate profound changes in the global health landscape during the first 2 years of the COVID-19 pandemic relative to historical trends. Long-term trends of decreasing mortality were superseded by marked increases in mortality rates in age groups older than 15 years during 2020 and 2021; in contrast, mortality in children under 5 years remained largely unaffected by the pandemic and continued to decrease globally. Global life expectancy declined sharply during 2020 and 2021, reversing the longstanding trend of life expectancy improvement. Age-standardised rates demonstrated the pandemic was disproportionately severe in countries within sub-Saharan Africa, the Middle East, south Asia, and Latin America. The COVID-19 pandemic has also highlighted the need for timely and comprehensive data collection and reporting. The development of high-quality civil registration and vital statistics systems has stagnated in many parts of the world due to multifaceted societal, financial, logistical, legislative, and political reasons, with notable exceptions including China, India, and some countries in north Africa and the Middle East. Population growth has slowed globally since 2017, although future declines might not persist at rates similar to those in 2020 and 2021 as the pandemic eases. In contrast, population growth is steady in south Asia and accelerating in sub-Saharan Africa. Increasing populations in many low-income and middle-income locations, combined with a shift in the age distribution away from younger ages and towards older ages, is likely to lead to new social, economic, and political challenges.

41Although the proportion of registered deaths has continuously increased at the global level since 1950, we observed marked variability across GBD super-regions and individual countries and territories. Civil registration and vital statistics are particularly scarce in sub-Saharan Africa; investment in vital registration system development in these nations is recommended to improve the availability of data necessary for accurate health measurements and policy evaluation. The COVID-19 pandemic highlighted the need for accessible and up-to-date health data when trying to understand and track emerging global health events. Much uncertainty remains about the true extent of the effect of the pandemic on mortality in countries and territories with minimal to no vital registration data available, which is particularly concerning considering that these countries are potentially the most negatively impacted by the pandemic. With the exception of China, India, and some countries in north Africa and the Middle East, progress in improving the extent of global death registration has slowed—perhaps due to a focus on cheaper but less permanent and systematic data collection efforts, such as small-scale and large-scale surveys. Although surveys are an invaluable source of demographic information, investing in more expensive yet comprehensive civil registration and vital statistics systems is crucial to monitor and improve population health.

42Beyond creating and improving civil registration and vital statistics systems, countries and territories without data during the past decade would also benefit from collecting additional data from other sources, such as censuses and nationally representative surveys. 30 countries and territories had no available data on child mortality for the period 2015–21, and 62 countries and territories had no available data on adult mortality. 41 countries and territories had no usable census data between 2010 and 2021, but census data were available before 2000 for these countries. Furthermore, the COVID-19 pandemic interrupted many data collection efforts, such as the USAID Demographic and Health Surveys Program, and national censuses, which are now resuming. Impacts such as these must be resolved to improve future data availability.

43The COVID-19 pandemic had differential effects on mortality across the lifespan. Life expectancy decreased in every GBD super-region and 84% of countries and territories from 2019 to 2021, but younger age groups were minimally affected. This finding is a welcome contrast to early warnings about potentially devastating impacts of the pandemic on child mortality. Conversely, increases in mortality rates in populations aged 25 years and older were observed on a scale not seen in the previous 70 years. Although the burden of excess deaths and all-age excess mortality rates due to the pandemic was largest in countries in central and eastern Europe, and Latin America, our analysis of age-standardised mortality rates highlights the relative severity of the pandemic's effects on mortality in certain countries within sub-Saharan Africa, the Middle East, south Asia, and Latin America. There was a general association between higher SDI and lower excess mortality, but this association was not particularly strong, and many countries were exceptions to this association, suggesting that at the population level, SDI was not always a strong predictor of excess mortality due to the COVID-19 pandemic in 2020 and 2021. Excess mortality was particularly high in nations such as Bolivia and South Africa when compared with other countries and territories with a similar SDI, which some have argued was in part due to relaxed containment strategies and vaccine hesitancy. Conversely, excess mortality was particularly low in countries such as the Solomon Islands and Bhutan, which might be a reflection of delayed transmission in more isolated nations and of high vaccination rates. These findings emphasise that mortality outcomes during the COVID-19 pandemic were not solely determined by SDI and that vaccination efforts, public policies, and individual behaviour changes likely influenced the severity of the pandemic across countries and territories at all levels of SDI.,,,, Reports published as recently as 2023 have shown that since 2021, mortality due to the pandemic has declined,, presumably driven by vaccination efforts, public policies, individual behaviour changes, and the emergence of new SARS-CoV-2 variants with lower case-fatality ratios., However, mortality has increased in some locations, which might be due to lifting of protective restrictions.

44In the era of the UN Sustainable Development Goals (SDGs), there has been a decline in the global U5MR, which continued during the COVID-19 pandemic. However, progress has varied substantially between countries, and many continue to lag behind SDG targets. Based on the trajectory of U5MR between 2010 and 2021, 38 countries will not reach SDG target 3.2 of a U5MR at least as low as 25 deaths per 1000 livebirths by 2030 (appendix 2 table S2A). To eradicate preventable under-5 deaths, more equitable global strategies—intensified in regions with the highest rates—are imperative. Compared with child mortality, reductions in adult mortality have not been as consistent globally. Historically, increased adult mortality was observed in the 1990s in countries spanning eastern and southern Africa, eastern Europe, and central Asia. During the late 2010s, some high-income nations, including the USA, have had mortality spikes, particularly among the 15–39-years age group, which reflect mortality patterns associated with increased drug and alcohol misuse and mental health disorders., The 15–39-years age group is particularly volatile globally, and is the age group most affected by fatal discontinuities such as conflict. Sex differences in mortality vary widely across the globe. The global ratio of male to female mortality has generally increased, although it has differed as a function of age. The largest variability in the ratio of male to female mortality was in the 15–39-years age group with much less variability observed in younger and older age groups. These differences go beyond biological explanations and highlight the importance of future efforts to address mortality risks to which males are particularly susceptible due to behavioural factors, war and conflict, occupational hazards, homicide, and suicide., The substantial differences among countries show, however, that it is also important to address mortality risks that predominantly affect women, such as maternal mortality, gender-based violence, and economic disparities., We also found that life expectancy was consistently higher in countries in the Americas, east Asia, and western Europe than countries in sub-Saharan Africa, and this effect was strongly associated with SDI. Although we did not establish causal effects, this finding is supported by many studies showing that social determinants of health are key drivers of mortality,,,,, and improving education, economic prosperity, and gender inequalities is vital for continual progress in health outcomes globally. However, notable exceptions regarding the relationship between mortality and SDI indicate that other factors are also involved.

45Although the rate of global population growth has plateaued and started to decline since 2017, in lower income countries—primarily in sub-Saharan Africa—rapid population growth has continued. Thus, much of future population growth will likely occur in the poorest regions. Resource scarcity and rapid infrastructure expansion will be crucial issues to address., These factors, and a history of colonialism, can contribute to political instability., These challenges will require responses from governments and the global community. Furthermore, the concentration of population growth has shifted to locations with the poorest health—ie, locations with the highest child mortality rates. This might lead to challenges in continuing improvement of health outcomes.

46Outside of these locations, slowing of population growth is widespread. Although most countries and territories had not reached a peak population as of 2021, in 171 of 204 countries and territories a lower rate of natural increase was observed between 2010 and 2019 than between 2000 and 2009. Furthermore, our analysis of population age structures over time indicated a prominent shift towards older ages in most regions and nations. As older populations expand and reduced younger populations reach working-age, nations could encounter economic and social challenges requiring updated policies related to health care, retirement, reproduction, childcare, and migration.,,, The shift towards a higher ratio of older people to younger people will require greater attention to be paid to labour shortages, health systems strengthening, and evaluation of government policies on retirement and health care.,, However, beneficial consequences such as the so-called second demographic dividend of greater personal wealth and investment in human capital might offset some of these challenges. Future research on these topics must seek to understand how changing population dynamics impact health outcomes and systems, and how health interventions can be tailored to address the unique challenges posed by these demographic shifts. Migration is particularly relevant to these challenges. Voluntary emigration from locations with younger adult population bulges to locations in need of more labour to support ageing populations is an open public policy discussion., The level of migration needed to support older age populations is dynamic and is likely to change over time with technological innovations and new public policies. Furthermore, environmental constraints in some high-income countries might limit immigration possibilities. Migration of skilled workers out of lower-income countries might consequently worsen these economies., Global cooperation is necessary, and guidelines such as the UN Global Compact for Safe, Orderly and Regular Migration can help lead this work.

47There are numerous differences in data processing and statistical modelling assumptions between the GBD 2021 estimates reported here and those from other demographic studies that provide important advantages. Excess mortality estimates for 2020 and 2021 have been previously reported in the GBD study and by other institutes. Our previous excess mortality estimates reported 18·2 million (95% UI 17·1–19·6) excess deaths in this study. Estimating mortality during the COVID-19 pandemic was particularly difficult due to many factors including delays in reporting, differing granularity of available data, and political will to provide accurate data. Although our earlier estimates were based on the best available data and methodology at the time, we have made data and modelling improvements that resulted in this lower estimate. We updated to more reliable data sources in some countries that corrected errors in reporting, and included more data up to the end of 2021. Methodologically, we modelled data at the yearly level, and additionally included age-specific detailed projections from our GBD mortality modelling process to inform our non-pandemic counterfactual, which generally led to higher estimates of expected non-pandemic mortality and thus lower excess mortality.

48Our current estimate of global excess mortality during 2020 and 2021 is comparable to the WHO estimate of 14·9 million (95% UI 13·3–16·6) excess deaths, with our mean estimate falling within the uncertainty interval of the WHO estimate and vice versa. Our estimates tend to be higher than those of WHO for sub-Saharan Africa, with the largest differences being 233 000 more deaths in Nigeria and 177 000 more deaths in Ethiopia; and south Asia, with the largest differences being 262 000 more deaths in Pakistan and 171 000 more deaths in Bangladesh. However, our estimate for India was 1·3 million deaths lower than that of WHO, which is the largest discrepancy in this direction. We also estimated 123 000 more excess deaths in China—our results indicated positive excess, whereas WHO estimated negative excess. The largest differences occur in locations for which little or no all-cause mortality data were available for the pandemic period, and thus estimates relied on predictive models. These differences reflect different covariates used for predictions models. Additionally, WHO models and predicts all-cause mortality rates in locations without data, whereas we predict excess mortality rates directly, which leads to different assumptions and functional forms for statistical models. Differences in locations with all-cause mortality data are driven by different data processing steps and different models for expected non-pandemic mortality.

49The latest estimates from UNICEF, published in 2023, reported a global U5MR of 38·1 deaths (95% UI 36·1–42·2) per 1000 livebirths in 2021, which is consistent with our estimate of 35·7 deaths (30·5–42·0) per 1000 livebirths. The mean relative difference at the national level between our 2021 U5MR estimates and those provided by UNICEF is –2·6%, ranging from –58·4% to 111·9%. Similar to our estimates, the UNICEF estimates show a continued decreasing trend in child mortality during the COVID-19 pandemic. Between 1950 and 2019, the mean relative difference between our estimates and UNICEF estimates across countries and territories was –2·0%, ranging from –64·3% to 154·6%. These differences primarily reflect differences in data inclusion, processing, and synthesis. For example, our estimate of mortality in Iran in 2021 is 58·4% lower than that of UNICEF. We included vital registration data from 2021 and our estimates closely match this observed mortality, whereas UNICEF does not include these data, leading to higher estimates. Using the most recent available data suggests our estimates are more reliable.

50Adult mortality estimates at the country level from the 2022 UN World Population Prospects (WPP) report are on average 11·1% lower than our 2021 estimates, which range from 41·8% lower to 289·5% higher. Between 1950 and 2019, the mean relative difference between our adult mortality estimates and those from WPP 2022 was –4·3%, ranging from –64·0% to 229·6%. Differences between WPP 2022 estimates of national life expectancy at birth and those from GBD 2021 are primarily driven by these differences in adult mortality estimates, and variability in child mortality estimates. While location-years with complete death registration show substantial agreement between estimates, with a mean relative difference of 1·3%, our estimates for 2021 range from 7·8 years lower to 10·1 years higher, and our estimates for years before the COVID-19 pandemic range from 20·4 years lower to 38·4 years higher. The largest discrepancies were due to location-years with large fatal discontinuities or scarcity of high-quality vital registration data. Furthermore, discrepancies between 2021 estimates are highly influenced by the differences in estimation of excess mortality due to the COVID-19 pandemic. As one of the largest differences, our life expectancy estimate for Nigeria in 2021 is 10·1 years higher than the WPP estimate, driven by our estimated 41·8% lower adult mortality. Our adult mortality estimates more closely follow the bulk of the data from sibling-survival histories, and our age-specific mortality estimates rely on a database of 43 758 empirical life tables as opposed to the Coale-Demeny north model life table used by WPP 2022, which has been shown to underperform compared with other modern model life table methods.,

51For further comparison with WPP and as a model validation exercise, we compared estimated age-specific mortality rates and death counts from our analysis and from WPP with those calculated directly from all location-years of vital registration data deemed to have complete death registration. When comparing our results, we used our population estimates as the denominator to calculate mortality rates from vital registration; similarly, we used WPP population estimates as the denominator for that comparison. Across all location-year-age-sex mortality rates, our estimates had mean absolute error of 0·024, indicating a good fit to the data, along with root mean squared error (RMSE) of 0·52. These were lower than the respective 0·033 and 0·53 calculated for WPP. Similarly, our death count estimates had a mean absolute error of 84·8 and RMSE of 365 compared with a mean absolute error of 222 and RMSE of 1032 for WPP estimates.

52Estimates of the global population from WPP 2022 are similar to that of this study, with an estimated global population of 7·91 billion in 2021, compared with our estimate of 7·89 billion (95 % UI 7·67–8·13). On average in 2021, country-level population estimates were 0·2% lower in GBD 2021 than WPP 2022 and ranged from 34·2% lower to 82·2% higher. For specific ages, differences in the younger than 15 years age group ranged from 48·0% lower to 75·3% higher, while differences in the 65 years and older age group ranged from 36·0% lower to 39·5% higher. The largest relative differences were for locations in which no recent census data were available, and those with substantial net in-migration from other countries.

53This research has several limitations. First, estimates continue to be limited by data source availability and scope. COVID-19 showed the crucial need to create more robust vital registration systems that can highlight the differential effects of disease and injury across population subgroups in a timely manner. 93 of 204 countries and territories had no available all-cause mortality data to estimate excess mortality due to the COVID-19 pandemic, which means our estimates in these areas are solely driven by associations with covariates. These locations were largely in regions where the effects of the pandemic were most severe. Furthermore, the scarcity of high-quality civil registration and vital statistics systems to produce reliable data in many low-income and middle-income countries introduces large-scale uncertainty in all demographic estimates. Additionally, population estimates in certain countries rely on modelled projections due to no available recent censuses. Future development of reliable data sources is crucial because estimates improve as the quality of underlying data improves. Subsequent GBD cycles will provide revised estimates after additional data for recent years become available.

54Second, analysis of more granular subpopulations such as subnational areas or by other population characteristics was restricted by data availability. Although our effort represents the most comprehensive global analysis of mortality and population, the estimates presented in this research mask substantial heterogeneity in smaller geographies. This limits the utility of our estimates to provide insights for more targeted interventions, for example, understanding occupational hazards in industrial regions. Improving this aspect of the research requires more comprehensive and detailed data, such as by race, ethnicity, socioeconomic status, and smaller administrative levels,,, and future work will aim to produce more comprehensive health metrics.

55Third, the GBD demographics approach has not developed an encompassing model to estimate migration together with population, mortality, and fertility. Estimating migration in a model that jointly informs population, mortality, and fertility will not only improve accuracy of population estimates, but also allow assessing and improving corrections for death registration completeness and census coverage. This is crucial in locations with large migration flows, such as the United Arab Emirates and Qatar, where current methods for these corrections might not perform well., The increased importance of migration at present and in the future, especially considering the shifting age structure in many populations, places renewed importance on producing reliable migration estimates.

56Fourth, we assumed a binomial distribution when calculating data variance and did not evaluate other models of distribution. Some of our input data might be overdispersed, resulting in inaccurate estimates of data variance. However, we do not expect that changing our assumptions on the distribution would have a sizeable impact on estimates since the sampling errors on vital registration and civil registration mortality and fertility data are likely to be much smaller than non-sampling errors. In the future, we will consider testing such assumptions.

57Fifth, computational resources did not permit propagation of uncertainty for all covariates throughout the analytical process. While uncertainty from model estimation was accounted for at each stage, such as U5MR, adult mortality, and age-specific mortality rates, uncertainties for some covariates such as lag-distributed income and education were not. Similarly, estimates of coefficients in the COVID-19 excess mortality prediction model did not include uncertainty. Future iterations of GBD will investigate computationally more efficient implementation of current methods and development of new methods to allow for all sources of uncertainty to be included in modelling.

58The COVID-19 pandemic will likely continue to impact estimates of demographic trends in future years due to reporting lags and the persistent effects of the pandemic. Future research should focus on understanding the full demographic impact of the pandemic in 2022 and beyond. Methodologically, we aim to improve our incorporation of excess mortality and COVID-19 direct mortality estimates into the GBD mortality estimation process, rather than post-hoc unification of two separate modelling endeavours. We also plan to develop a standalone migration model and integrate this model into the GBD demographic estimation process. Along with this, we aim to simultaneously estimate mortality and population rather than the current sequentially iterative approach. This would allow the uncertainty in mortality estimates to inform population estimates and vice versa, helping address issues in age, period, and cohort trends that might otherwise arise.

59Tracking long-term health trends and evaluating the impact of the COVID-19 pandemic require accurate global, regional, and national estimates of mortality, life expectancy, and population, because these crucial demographic indicators foundationally underpin our understanding of population health. The comprehensive demographic metrics reported in this study show that marked reversals in adult mortality and life expectancy trends occurred during 2020 and 2021, leading to increased mortality and reduced life expectancy worldwide. This increased mortality did not occur in younger populations: mortality rates in children under 5 years continued to decline globally during the first 2 years of the pandemic, although more equitable and intensified investment is needed to achieve SDG targets in many locations. While global population growth is slowing, geographical distributions and age structures are undergoing fundamental shifts—low-income countries and territories continue to grow, and population structures across the globe are ageing. Nations in the post-pandemic world will need to address emerging health-care, economic, and social challenges with new policies and practices. The development, implementation, and evaluation of these health policies and practices in diverse locations around the world can be informed and guided by the GBD 2021 demographic estimates. Accurate mortality, life expectancy, and population estimates might be even more important to informing policy and practice in a post-pandemic world than in the past. Collectively, the extensive set of demographic estimates reported here represent a valuable global tool for policy evaluation, development, and implementation in diverse locations around the world.

60To download the data used in these analyses, please visit the GBD 2021 Sources Tool. The statistical code used in GBD 2021 is available online.

61For the Mortality Visualisation Tool see https://vizhub.healthdata.org/mortality/

62For the Mortality Visualisation Tool see https://vizhub.healthdata.org/mortality/