The slowing pace of life expectancy gains since 1950

We obtained data for a set of 173 countries on life expectancy at birth and a set of key socioeconomic factors over a 60-year period, 1950–2009, all from publicly available databases. LEB data were from the Institute for Health Metrics and Evaluation (IHME) [6] for 1970–2009. Prior to 1970 LEB data were from: Human Mortality Database (HMD) [7‐9], World Population Prospects (WPP) [10, 11], and Gapminder [12], see Table A-1-1 of the Additional file 1.

Socioeconomic factors identified in the literature as contributing to life expectancy improvement were: Population density and GDP per capita, as proxies for urbanization and economic growth; CO2 emissions, as a proxy for environmental quality. Total Fertility Rate (TFR) and HIV prevalence are included because of their known correlation with population health. Summary statistics of the total sample of countries are in Table 1, stratified by decade.

Population Density (per square km) was retrieved from the United Nations Population Division [13]. Real GDP per capita is measured at 2005 constant prices, and was retrieved from Penn World Table of the University of Pennsylvania [14].

Average CO2 emissions are measured in metric tons per person, and were retrieved from the Carbon Dioxide Information Analysis Center (CDIAC) [15]. A country’s level of carbon dioxide is a proxy of the greenness of an economy’s production processes. CO2 is included in the model to control for confounding time trends in air pollution, which has a known negative effect on health.

TFR data for 160 countries were retrieved from the United Nations Population Division [16]. For the 13 remaining countries data were pulled from different sources, see Data Sources of the Additional file 1. Estimated HIV prevalence data were obtained from the Joint United Nations Programme on HIV/AIDS (UNAIDS) [17]. The global HIV/AIDS pandemic had a high impact on the overall number of deaths worldwide. In some low income countries, an estimated of 1–4 years of life expectancy are lost due to HIV/AIDS in those older than 5 years [18]. For the years prior to 1970, it was assumed that HIV prevalence was virtually zero, since the disease was first acknowledged in 1981 [19]. Some countries lacked complete data on HIV between 1970 and 1989. For these countries, country-specific trends after 1990 were used to back-cast HIV rates into the prior period, see HIV Imputation – Table A-4-1 of the Additional file 1.

LEB decadal gains was used as the dependent variable, defined as the difference between LEB_t + 10 and LEB_t. Analysis was performed separately within each LEB strata to confine comparisons of countries with roughly equivalent LEB starting points.

For the regression-adjusted analysis, a country fixed effects model was selected. This model controls for unobservable time-invariant features of each country [20, 21]. The final model included the following 6 covariates – motivated by theory, past evidence, and data availability: LEB, GDP per capita, population density, CO2 emissions, TFR, and HIV prevalence. The model also included time dummies for each decade from 1950 - 1959 to 2000–2009.

Each covariate was converted to a Z-score by centering it at its country-specific mean and dividing by its country-specific standard deviation. Z-scores make it possible to interpret effect sizes as the impact of a 1 SD change in a covariate on the decadal change in life expectancy for countries in each starting LEB stratum.

The equation of the fixed effects model for LEB decade gain is:

Where α_i represents the unknown intercept for each country indexed by “i”; X_k,it represents a vector of k independent variables (k = 6 factors), for i countries (i = 139 countries) in t times (t = 59 years); _tD_t represents binary time regressors for 5 decadal time-periods, with 1950–1959 excluded; and ε_it is the error term. Because each analysis is stratified to only include country-years of similar starting life expectancies, we would expect the benefits of accumulated biomedical technology over time to make the δ terms get progressively larger such that δ₅>δ₄>δ₃>δ₂>δ₁.

The main model includes 139 countries, because HIV prevalence data were missing for 34 countries. An additional model was specified omitting HIV prevalence, to ensure the results are not an artifact of a selective subsample, results are on Table A-2-9 of the Additional file 1. This model includes 173 countries, the lists of countries included in both analysis are provided on Table A-2-1 and A-2-2 of the Additional file 1.

It is possible to be misled by appending demographic data from different sources. Life expectancy estimates are inherently based on models, and the underlying age-specific mortality rates can be extensively modeled by demographers who are attempting to achieve best estimates consistent with all the evidence at hand. If the pre-1970 data came from models that overestimated life expectancy relative to the post-1970 source, our analysis should have found a step-off point only at 1970.

To assure that declining life expectancy is not simply the result of appending multiple databases, we repeated the analysis for 1970 onwards, using only IHME as source and continued to find statistically significant evidence of a slowdown in the pace of LEB gains, see Table A-2-8 of the Additional file 1. Thus, the effects are not solely due to the merging of life expectancy data from multiple sources.

It is possible that spillover effects from HIV impeded progress in improving health, even though our model controlled for HIV prevalence. It is also possible that HIV effects spill over to other decades and could have lagged effects for years into the future. This explanation is most relevant from 1990 onwards, and among countries from the lowest strata of starting LEB. From the 1980’s onward there was a slight improvement in HIV prevalence only among countries with a life expectancy between 61 and 71, decreasing from 0.86 (SD 2.45) in the 1980’s to 0.78 (SD 1.99) in the 2000’s. Countries with lowest LEB (LEB < 51) were the most affected by HIV, increasing their prevalence from 1.88 (SD 2.39) in the 1980’s to 17.03 (SD 7.83) in the 2000’s.

However, our analysis showed slowdowns in places and periods where HIV/AIDS was not prevalent enough to be a plausible explanation of LEB slowing pace. Additional file 1 Table A-2-10 confirms LEB growth deceleration since 1950 across regions where HIV prevalence was lower than Africa’s. We see evidence of decelerating rates of LEB growth in East Asia and Pacific, Europe and Central Asia, Latin America and the Caribbean, and Middle East and North Africa. Furthermore, LEB growth deceleration was occurring before and during the 1980s prior to the start of the HIV/AIDS era. Hence HIV may have contributed to slowing progress in LEB growth, but it cannot be the sole cause.

Stratification controls for LEB level in each country, and the fixed effects model can control for idiosyncratic features of every single country over time. However, our methods cannot adjust for unobservable features characterizing the shifting composition of countries within each grouping over time. For example, in the lowest group there are 635 country-years of observations on 70 countries (like Mali, Burkina Faso, and India) in the 1950s, and 60 country-years of observations on 11 countries (like Zimbabwe, Swaziland, and Lesotho) in the 2000s. It is plausible that the former larger set of countries in the 1950–59 is mainly composed of countries striving for better health, whereas the latter smaller set of countries that remain in the lowest group in 2000–09 include a higher proportion of fragile-states [22]. Fragile states have unique challenges to their health care systems, not only because of an increased burden of disease, but also because they have fragile governance and political conflicts, along with scarcity of health care workforce, and financial limitations.

While the shifting composition hypothesis is a compelling explanation for the declines in decadal life expectancy gains for the lowest stratum, it is somewhat less compelling for the highest stratum of life expectancy. Whereas the set of exceptional 12 countries achieving life expectancy above 70 by 1950–1959 would be enriched by high performers, all the 79 countries registering life expectancy above 70 in 2000–2009 would also have had to learn to strive to achieve higher life expectancies.

Explaining a declining pace of life expectancy gains around the world as reflecting a decline in the eagerness of societies and governments to improve population health should provoke critical reflection. Societies have shown great enthusiasm for signing declarations like the Millennium Development Goals and Sustainable Development Goals. Countries around the world have also shown a manifest willingness to increase governmental expenditure on delivering health personal services. They have had access to the very best policy advice on how to expand universal health service coverage to guarantee access to medical care for vulnerable groups. If the declining pace of progress indeed reflects a decline in political will among the least healthy countries, then the policy community is remiss in failing to harness any of the existing enthusiasm for higher government health spending into investments that will maintain historical rates of progress in the growth of life expectancy.

The final explanation is admittedly speculative, and would hold there has been a fundamental shift since the mid-twentieth century in the basic approach to improving population health in every country.

Infectious disease mortality reduced dramatically in high-income countries between 1880 and 1950, achieved as a result of what some call the “golden age of public health” [23, 24]. Improvements in housing, diet, the safety and quality of water, air, and food mediated by public health proponents were fundamental to these early twentieth century gains [25].

In the 1950s countries around the world began to have access to penicillin and other antibiotics. Following 1950, most countries prioritized efforts to scale-up health care services supply and life-saving medical treatments and vaccines. Following World War II, former colonies and less developed countries were challenged to simultaneously produce an effective public health workforce while meeting the mandate to introduce life-saving biomedical services.

In summary, a possible explanation for the slowdown in health progress is half a century of shifting emphasis away from covering populations with public health infrastructure to keep populations healthy and towards delivering clinical services to cure the sick. Unfortunately, the ideal evidence to test this conjecture would require data on each country’s expenditure on non-clinical public health related activities, and very few countries have this data [26]. The recent outbreaks of Ebola and Zika virus, and recurrent calls to strengthen public health, reflect a growing recognition of the unfinished agenda in building public health infrastructure around the world [27, 28].