All-cause mortality, GBD Learning

Is all-cause mortality increasing around the world? What about life expectancy? This Global Burden of Disease (GBD) Learning video presents the research and estimates from this GBD 2023 capstone, published in The Lancet: "Global burden of 292 causes of death in 204 countries and territories and 660 subnational locations, 1990–2023: a systematic analysis for the Global Burden of Disease Study 2023."

Video transcript

This transcript has been lightly edited for clarity

Is life expectancy increasing or declining? What about mortality rates around the world? How has the world recovered from the COVID-19 pandemic?

The most recent update from the Global Burden of Disease study (GBD) provides answers to these questions with estimates of global all-cause mortality and life expectancy estimates for males, females, and all sexes combined across 25 age groups and 204 countries and territories and 660 subnational locations from 1950 to 2023.

For GBD 2023, the entire time series of estimates from 1950 to 2023 is updated.

GBD 2023 uses brand-new methodology for estimating mortality. The biggest challenge for making mortality estimates is that many countries don’t have complete or reliable records of every death. Demographic research has historically relied on model life tables, which work by taking countries with high-quality data, say Sweden or Japan, and looking at their mortality patterns. You can then use those patterns as a template for a country that doesn’t have good data. This process has a fundamental challenge – it forces patterns from certain locations onto locations with less data, which can lead to incorrect mortality patterns in some locations.

In GBD 2023, we used a new statistical model called OneMod to improve the accuracy of mortality estimation. This model can estimate age patterns of mortality directly from the available data, borrowing insights from similar locations and important drivers of mortality, rather than imposing historical patterns from dissimilar populations. This method is used to estimate mortality rates and total deaths from all causes in a population, as well as life expectancy at birth.

Here’s how we make our estimates.

We start by preparing over 24,000 data sources, with more than 3,000 new data sources compared to GBD 2021. For the first time, we incorporated previously unused data sources, including complete birth history data for children ages 5 to 14, age-specific sibling history data for ages 15 to 49, and age-specific mortality data from health and demographic surveillance systems.

We then split the data into GBD age groups. Often, data comes in broad age and sex groups (for example, all adults or all children), which is not sufficiently detailed. We use a method of age-sex splitting to break data into males and females, and into 25 GBD age groups, ranging from early neonatal to 95 years and older.

We also need to address fatal discontinuities (also called shocks). In GBD, shocks are unexpected events that happen randomly and can’t be predicted in advance. Data with substantial shocks are initially excluded from the main mortality estimation process in

order to identify the underlying mortality patterns. Sudden, temporary jumps in mortality from large events would distort these long-term trends. Instead, shocks are incorporated later on to account for their impact on mortality estimates.

The next step in data preparation involves adjusting vital registration data for missing death registrations. The completeness of death registration varies significantly across locations and time, particularly in low-income and middle-income countries where high-quality vital statistics are scarce.

Once the data has been corrected for errors and bias, and made comparable and comprehensive, we can make our mortality and life expectancy estimates.

We do this using OneMod. This tool uses a single model to estimate death rates for every age, sex, and location simultaneously within a unified framework. It works by taking advantage of observed relationships in the data across different ages, time periods, and geographic locations, as well as information from drivers of mortality. The first step uses a generalized linear model that accounts for age-specific effects of drivers of mortality, such as the Socio-demographic Index (SDI). The second step uses a multivariate regression model that smooths across age, time, and location without overfitting. Unlike previous techniques that depended on model life tables, this data-driven method estimates mortality rates directly.

So, with this new, much more accurate methodology, what new stories have been revealed?

We estimate that there were 60 million deaths globally in 2023, an increase of 35% since 1950.

This increase is primarily driven by considerable global population growth and aging. The age distribution of deaths has shifted dramatically. In both 1950 and 1990, the largest number of deaths occurred among children under 1 year. By 2023, that had shifted to the 80–84-year age group.

While the number of deaths has increased, the age-standardized mortality rate has declined by 66.6% since 1950. That means that after controlling for increases in population and population aging, people are dying at substantially lower rates than in the past.

We also found some surprising results when looking at age-specific mortality rates. In sub-Saharan Africa, the death rate for children between 5 and 14 years old was 87% higher than our old models were telling us. We also saw the death rate increase by more than 60% for females aged 15 to 29 in this region, and decrease by 13% for people 50 and older. These changes are due mainly to new data sources that we were able to incorporate into our new methods.

In this figure, we can look at age-specific mortality rates for the different super-regions all at the same time. On the x-axis we have years from 1950 to 2023, and on the y-axis, age from 0 to 85+. Lower mortality rates are more blue, while higher mortality rates are more red. Age-specific mortality rates are mostly declining as time passes, with exceptions for shocks showing up as different-colored lines. One of the most substantial increases in mortality rates can be seen in sub-Saharan Africa in the 1990s and 2000s, due to the HIV/AIDS epidemic. The dark red in Southeast Asia, East Asia, and Oceania in 1960 was due to famine.

Global life expectancy has followed three distinct trends since 1950:

First, from 1950 to 2019, life expectancy rose from about 51 to 76 years for females and about 48 to 71 years for males.

The COVID-19 pandemic caused a dip between 2019 and 2021. Life expectancy fell about 1 to 2 years, to 75 for females and 69 for males in 2021. Three-quarters of countries experienced life expectancy declines during the COVID-19 pandemic.

In 2022 and 2023, we saw a rapid post-pandemic recovery. Life expectancy bounced back to pre-pandemic levels by 2023 in almost all countries.

We did see regional differences in recovery, with Latin America and the Caribbean showing the largest rebound post-pandemic, and Southeast Asia, East Asia, and Oceania showing the smallest

So, let’s review the main findings.

First, while the number of deaths increased, global age-standardized mortality rates have declined by two-thirds since 1950.

Second, our new modeling approach changed our estimates of age-specific mortality, with increases among young people in sub-Saharan Africa compared to GBD 2021.

And finally, life expectancy has risen dramatically since 1950, and the world has largely recovered from drops in life expectancy due to the COVID-19 pandemic.

GBD 2023 marks a major step forward in how we estimate mortality and life expectancy worldwide. This year, we introduced new methods which use data from over 24,000 sources – including many datasets we’ve never used before.

While overall death rates have declined – thanks to medical advances and improved living conditions – population growth and aging mean more people are dying each year. The COVID-19 pandemic caused a temporary setback, with a notable but uneven recovery.

Our latest estimates are more accurate and locally relevant, helping policymakers, researchers, and the public better understand health challenges and progress around the world.