Global and regional mortality from 235 causes of death for 20 age groups in 1990 and 2010: a systematic analysis for the Global Burden of Disease Study 2010

Summary

Introduction

Cause-specific mortality is arguably one of the most fundamental metrics of population health. The rates and numbers of people who die, where, at what age, and from what, is a crucial input into policy debates, planning interventions, and prioritising research for new health technologies. Trends in causes of death provide an important geographical summary of whether society is or is not making progress in reducing the burden of premature (and especially avoidable) mortality and where renewed efforts are needed. If a health information system is not providing timely and accurate information on causes of death by age and sex, major reforms are required to provide health planners with this essential health intelligence.

Despite the importance of tracking causes of death and the tradition since 1893 of standardisation of definitions and coding for causes of death in the International Classification of Diseases and Injuries (ICD), global assessments of causes of death are a major analytical challenge. Vital registration systems that include medical certification of the cause of death captured about 18·8 million deaths of an estimated annual total of 51·7 million deaths in 2005, which is the latest year for which the largest number of countries (100) reported deaths from a vital registration system. Even for these deaths, the comparability of findings on the leading causes of death is affected by variation in certification skills among physicians, the diagnostic and pathological data available at the time of completing a death certificate, variations in medical culture in choosing the underlying cause, and legal and institutional frameworks for governing mortality reporting.^1–5 For the remaining deaths that are not medically certified, many different data sources and diagnostic approaches must be used from surveillance systems, demographic research sites, surveys, censuses, disease registries, and police records to construct a consolidated picture of causes of death in various populations. Because of the variety of data sources and their associated biases, cause of death assessments are inherently uncertain and subject to vigorous debate.^6–8

Efforts to develop global assessments for selected causes began in the 1980s.^9–11 These efforts were motivated partly because the sum of various disease-specific estimates substantially exceeded the estimated number of deaths in the world, particularly for children.¹² Lopez and Hull¹¹ attempted to develop a set of estimates of mortality in children younger than 5 years (under-5 mortality) by cause consistent with all-cause mortality data in 1983. The Global Burden of Disease study 1990 (GBD 1990) was the first comprehensive attempt to do so, and included 134 causes covering all age groups. The GBD 1990 cause of death approach was applied with some refinements to yield estimates of causes of death for 1999, 2000, 2001, 2002, 2004, and 2008.^13–17 Over this period, special attention was paid to priority diseases such as malaria, HIV/AIDS, and tuberculosis. The Child Health Epidemiology Reference Group (CHERG) also produced estimates of under-5 mortality from 1G causes that summed to estimates of deaths in children younger than 5 years for 2000–03, 2008, and 2010,^18–20 partly using the GBD 1990 approach combined with other methods, and putting special emphasis on the use of verbal autopsy as a source of data in low-income settings. Additionally to these comprehensive approaches, the tradition of disease-specific analyses that began in the 1980s with global cancer mortality has continued and intensified. In the past 5 years, for example, articles and reports have been published on global mortality from maternal causes,^21–24 malaria,^25,26 tuberculosis,^27,28 HIV/AIDS,²⁹ road traffic accidents,³⁰ site-specific cancers,^31,32 and diabetes,³³ among others.^34,35 These assessments of individual causes are based on diverse epidemiological approaches of varying scientific rigour, and, moreover, are not constrained to sum to estimates of all-cause mortality from demographic sources.

Global cause of death assessments can be characterised in four dimensions: the universe of raw data identified and examined, efforts to evaluate and enhance quality and comparability of data, the statistical modelling strategy, and whether causes of death are constrained to sum to all-cause mortality. First, in terms of the universe of data, the various iterations of the GBD and CHERG analysis of deaths in children younger than 5 years have made substantial use of data on causes of death from systems that attempt to capture the event of death. Other single-cause analyses, such as the annual UNAIDS efforts to estimate HIV-related deaths, measles estimates,³⁴ the World Malaria Report,²⁶ the WHO Global TB Control Report,²⁸ and many others have used data on disease incidence or prevalence and on case-fatality rates combined in a model of natural history progression. Second, perhaps the area of greatest variation in the published studies is the efforts to assess and enhance the quality and comparability of available data. These efforts often include very specific steps undertaken for different data sources and are frequently poorly documented. Third, in the past two decades, efforts to develop statistical models for causes of death have become more sophisticated. Compositional models that estimate cause fractions for several causes at once were first applied to global health by Salomon and Murray³⁶ and have been used extensively by CHERG but only for a subset of causes. GBD revisions for 1999, 2000, 2001, 2002, 2004, and 2008 have used these compositional models to allocate deaths according to three broad cause groups: communicable, maternal, neonatal, and nutritional causes; non-communicable diseases; and injuries. More recently, the array of modelling strategies used for causes of death has been broadened to include spatial-temporal Gaussian process regression,^22,37 mixed effects hierarchical models, and ensemble models.³⁸ Given the profusion of statistical modelling options, an important innovation has been the reporting of out-of-sample predictive validity to document the performance of complex models.^22,38

Finally, in view of the developments in the field of mortality and cause of death estimation, for the GBD 2010 we completely re-evaluated all aspects of the GBD analytical strategy, including demographic estimation of all-cause mortality.^39,40 Because of the huge increase in published verbal autopsy studies and the availability in the public domain of cause of death data from government vital registration sources (130 countries), the universe of data has expanded substantially. Assessing and enhancing the quality and comparability of data can now take into account time trends in cause of death data from 1980 to 2010 that provide important insights into changes in certification and coding. Borrowing from other scientific fields, we have changed our analytical approach (see below) to an ensemble modelling strategy to generate more realistic uncertainty intervals (UIs) and more accurate predictions.³⁸ These innovations have been used in estimating mortality for an expanded GBD 2010 cause list of 291 causes compared with 134 in the GBD 1990 Study; of the 291 causes, 235 are causes of mortality, whereas the remaining causes account for years lived with disability (YLDs) but not deaths. We use a unified framework for all causes such that the sum of cause-specific estimates equals the number of deaths from all causes in each country or region, period, age group, and sex. This creates a link between the systematic analysis of data on all-cause mortality reported by Wang and colleagues⁴⁰ and results by cause presented here. In this Article, we provide a summary overview of the vast array of data and methods that have gone into this revision of the GBD, as well as what we believe are the key global and regional findings of importance for health priority debates.

Methods

Some general aspects of the analytical framework such as the creation of the 21 GBD regions and the full hierarchical cause list including the mapping of the ICD to the GBD 2010 cause list are reported elsewhere.³⁹ Although results are reported in this Article at the regional level for 1990 and 2010, the cause of death analysis has been undertaken at the country level for 187 countries from 1980 to 2010. Use of longer time series improves the performance of many types of estimation models; data from before 1980, however, are much sparser for developing countries so we restricted the analysis to 1980–2010.

Results

The GBD 2010 cause list divides causes into three broad groups. At the most aggregate level, communicable, maternal, neonatal, and nutritional causes account for 13·2 million (24·9%) of 52·8 million total deaths at all ages in 2010. Non-communicable causes account for 34·5 million or 65·5%. The third category, injuries, accounts for 5·1 million or 9·6%. The continued decrease in deaths from communicable, maternal, neonatal, and nutritional disorders is striking, if not surprising. The number of deaths from these disorders decreased by 2·7 million from 15·9 million in 1990 to 13·2 million in 2010, representing a 17% decrease from 1990 to 2010. The annual number of deaths from non-communicable diseases, by contrast, rose by just under 8 million, to 34·5 million, or two of every three deaths in 2010. The global fraction of deaths due to injuries increased slightly between 1990 and 2010 (from 8·8% to 9·6%), but this masks some important trends in mortality from these causes.

Table 1 decomposes these global trends into the contribution of total increase in population size, ageing of the population, and changes in age-specific and sex-specific rates. Global population growth alone would have been expected to increase deaths from all causes by 31·8% from 1990 to 2010. Because of the correlation between population growth rates and mortality rates from communicable, maternal, neonatal, and nutritional causes, population growth alone would have increased this category by 46·9%, non-communicable diseases by 22·9%, and injuries by 31·1% (table 1). Ageing of the world’s population such that the mean age of the world increased from 26·1 years to 29·5 years contributed to an 11·2% decrease in communicable, maternal, neonatal, and nutritional disorders, a 39·2% increase in non-communicable disease deaths, and a 9·2% increase in injuries (table 1). Declines in age-specific and sex-specific death rates have contributed to a 52·7% decrease in communicable, maternal, neonatal, and nutritional deaths, a 32·1% decrease in non-communicable disease deaths, and a 16·3% decrease in injury deaths. With decreasing age-specific death rates from all three groups of causes, including non-communicable diseases, the global shift towards non-communicable diseases and injuries as leading causes of death is being driven by population growth and ageing, and not by increases in age-sex-cause specific death rates.

At the second level of the GBD 2010 cause hierarchy, there are 21 major cause groups. Figure 1 (A–D) summarises the composition of causes of death for every age-sex group for male and female individuals separately in 1990 and 2010 at this second level of cause disaggregation. The structure of causes of death changed systematically with age. In the neonatal age groups, disorders arising during the neonatal period dominated, but with important contributions from the category of diarrhoeal disease, lower respiratory infections, and other infectious and non-communicable diseases, including congenital causes. By the post-neonatal period, causes of death were dominated by diarrhoea, lower respiratory infections, and other infectious diseases such as measles, among others. At ages 1–4 years, the category neglected tropical diseases and malaria were also an important contributor to global mortality. In the age group 5–14 years, infectious diseases, HIV/tuberculosis (HIV/TB), injuries, and some cancers predominated, although overall mortality at these ages was low. Important sex differences were evident from ages 15–34 years; among male individuals, injuries, HIV/TB, and some non-communicable diseases predominate. Among female individuals of the same age group, injuries were a smaller cause of death with maternal causes making an important contribution. In 2010, maternal causes accounted for 10·7% of deaths of women in this age group. By age 40 years, more than 50% of global deaths in 1990 were from non-communicable diseases—this transition age shifts to 45 years in 2010 because of the HIV epidemic. Beginning at age 50 years, circulatory causes begin a steady rise to become the largest cause of death.

Open in a separate window

Figure 1:

Percentage of global deaths for female and male individuals in 1990 and 2010 by cause and age

(A) Male individuals, 1990. (B) Female individuals, 1990. (C) Male individuals, 2010. (D) Female individuals, 2010. An interactive version of this figure is available online at http://healthmetricsandevaluation.org/gbd/visualizations/regional.

The comparison of the 1990 and 2010 plots shows various shifts in the cause structure by age and sex (figure 1). At younger age groups, neonatal disorders and other non-communicable causes, including congenital anomalies, predominate. The unfolding HIV/AIDS epidemic at the global level is clearly evident from the huge increase in the contribution of HIV/TB to cause of death patterns among young adult men and women. By 2010, for example, HIV/TB and injuries accounted for more than half of all deaths in the age group 20–39 years in men. Other important shifts, with age, are evident—namely, a rising fraction of deaths in many age groups from diabetes, chronic kidney diseases, blood and endocrine disorders, and cancers, along with a decrease in the fraction due to chronic respiratory deaths in the middle-aged and older groups. For women, the share of deaths at ages 20–39 years due to maternal causes notably decreased.

At a more disaggregated cause level, there is interest in a broad global overview of who dies of what, and how this is changing. Table 2 provides total numbers of deaths and age-standardised death rates for each cause in 1990 and 2010. Because there is substantial interest in causes of death for different age groups, we provide global deaths for the 20 GBD age groups, by sex, and including 95% UI for 2010 and 1990 (appendix pp 149–340). In addition to the numbers of deaths, we present the death rates by age for 2010 and 1990 in for those readers interested in comparing change in age-sex-specific death rates (appendix pp 341–533). There are many features of the tables that warrant discussion: we limit ourselves here to some general observations that we believe characterise the principal epidemiological trends around the turn of the millennium. Much of the decrease in the communicable, maternal, neonatal, and nutritional causes was due to the substantial reductions in diarrhoeal disease (from 2·5 to 1·4 million), lower respiratory infections (from 3·4 to 2·8 million), neonatal disorders (from 3·1 to 2·2 million), measles (from 0·63 to 0·13 million), and tetanus (from 0·27 to 0·06 million), reflecting the scaling up of effective treatments and technologies to combat these disorders generally associated with poverty (table 2). Not all diseases in this category decreased, however. Table 2 shows the massive increase in deaths between 1990 and 2010 from HIV/AIDS (from 0·3 to 1·5 million), despite the decrease after 2006, as well as a 19·9% rise in malaria mortality over the two decades.

Cancers claimed 8·0 million lives in 2010, 15·1% of all deaths worldwide, with large increases in deaths from trachea, bronchus, and lung cancers, twice the number of deaths from the next two most common sites for mortality (liver and stomach). Roughly half of the total liver cancer mortality was attributed to hepatitis B infection, with smaller fractions due to hepatitis C and alcohol (table 2). The largest cause fraction (13·3%) among all causes of death in 2010 was due to ischaemic heart disease, closely followed by stroke (11·1%), being roughly split equally between ischaemic stroke and haemorrhagic and other non-ischaemic stroke (table 2). Together, ischaemic heart disease and all forms of stroke killed an estimated 12·9 million people in 2010, a quarter of the global total, compared with one in five deaths worldwide 20 years earlier. Cirrhosis of the liver was the cause of a million deaths in 2010, 33% more than in 1990, roughly equally attributable to hepatitis B, hepatitis C, and alcohol. Diabetes deaths worldwide almost doubled, as did deaths from chronic kidney disease. Deaths from Alzheimer’s disease and other dementias rose more than three-fold, and deaths from Parkinson’s disease doubled. One of the few causes in this group to decrease was COPD, falling from 3·1 to 2·9 million. This is consistent with the decreases observed with development in countries such as the UK in the first part of the 20th century, only to be reversed as the effect of tobacco use becomes evident.^63,64 The massive increases in tobacco use since the 1970s, especially in men in less developed countries, might reverse this trend over the next decade or so.⁶⁵

One million more deaths from injuries occurred in 2010 (5·1 million) than two decades earlier, a 24% increase (table 2). This was driven primarily by a 420 600 increase in road traffic deaths, which claimed 1·3 million lives in 2010. Falls also claimed an additional 191 900 lives compared with 1990, with most other accidental causes being relatively constant, or decreasing, especially drowning. Deaths from intentional injuries increased for both self-harm and interpersonal violence. Deaths from forces of nature, war, and legal intervention were more than twice as common than two decades earlier. Given the huge annual fluctuation in deaths from forces of nature and war, trends must be interpreted with caution. The fact that deaths from injuries are rising, and account for one in ten deaths worldwide, argues for far greater policy action to prevent them.

Trends in numbers of deaths are of interest and importance for health services and health policies that are designed to reduce premature mortality from various causes. Yet numbers of deaths alone do not provide a clear indication of whether disease control strategies are working since they are heavily dependent on changes in population size and age structure. By computing age-standardised mortality rates, we effectively controlled for demographic changes across populations over time; however, age-standardised death rates are sensitive to the population standard used. Changes in age-standardised mortality rates between 1990 and 2010 are shown in the right hand panels of table 2. Death rates from all communicable, maternal, neonatal, and nutritional disorders have decreased by 30% since 1990, a much greater reduction than suggested from numbers of deaths alone (table 2). The age-standardised death rate from diarrhoeal diseases fell by 49%, whereas lower respiratory infections decreased by 34%. Interestingly, age-standardised death rates from trachea, bronchus, and lung cancers fell by 8% between 1990 and 2010, despite a 47% increase in numbers of deaths, due to continued decreases in mortality in developed countries and more modest increases in less developed countries where the full impact of smoking, especially in men, has yet to occur. Breast cancer mortality rates fell by 15%, even though numbers of deaths from the disease increased by more than a third.

Our findings suggest important decreases (20% or more) in age-standardised death rates from major vascular diseases, especially heart disease and strokes, for the world as whole, even though numbers of vascular disease deaths increased by a third to 15·6 million in 2010 (table 2). Death rates for COPD and liver cirrhosis also decreased, but almost doubled for Alzheimer’s disease, and rose for diabetes and chronic kidney disease. These represent important global health challenges that might, or might not, be evident from an assessment of trends in numbers of deaths alone. Globally, although the number of deaths from injury rose by 24% since 1990, death rates decreased modestly, although this masks variable trends for different injuries, with death rates from drownings and poisonings falling by about a third, less dramatically for self-harm and violence, and rising for transport injuries and, interestingly, from adverse effects of medical treatment. Death rates from forces of nature also massively increased (by 336%) comparing 1990 to 2010 because of the earthquake in Haiti in 2010.

Causes of death in children younger than 5 years are of particular interest because of the global focus on improving child survival over the past few decades that has been reinforced by the push to achieve Millennium Development Goal (MDG) 4 (to reduce by two thirds, between 1990 and 2015, the under-5 mortality rate) in recent years. The appendix (pp 149–340) provides a breakdown of deaths in children younger than 5 years into the early neonatal, late neonatal, post-neonatal, and 1–4 years age groups in 2010 and 1990. To facilitate an understanding of the leading causes at different ages under 5 years, we show in figure 2A, ​,B,B, and C the distribution of deaths in the neonatal periods, the post-neonatal period, and ages 1–4 years, respectively. Of 2·8 million early and late neonatal deaths, we estimate that 2·1 million were from neonatal disorders including preterm birth complications, neonatal sepsis, and neonatal encephalopathy, among others. A further 137 000 deaths were also due to disorders that arise in the neonatal period but which lead to death after the first month of life. Among the important neonatal disorders, preterm events accounted for 29% of global neonatal deaths, with nearly equal shares for neonatal sepsis and neonatal encephalopathy—17% each. Of the remaining 741 000 neonatal deaths, congenital anomalies accounted for 183 000 deaths, although most congenital deaths occured after the first month of life. Injuries accounted for 28 000 neonatal deaths; other non-communicable causes including haemoglobinopathies and haemolytic anaemias, some rare cancers, sudden infant death syndrome (SIDS), and other rare causes accounted for 45 000 deaths. Among communicable diseases, notably lower respiratory infections (194 000), diarrhoea (77 000), and meningitis (46 000) accounted for the remaining neonatal deaths.

Open in a separate window

Figure 2:

Pie chart of global neonatal, post-neonatal, and child deaths in 2010 for children of both sexes combined by cause

(A) Age 0–27 days (neonatal); 2 840 157 total deaths. (B) Age 28–364 days (post-neonatal); 2 031 474 total deaths. (C) Age 1–4 years; 1 969 567 total deaths. ETEC=enterotoxigenic Escherichia coli. EPEC=enteropathogenic E coli. Hib=Haemophilus influenzae type B. RSV=respiratory syncytial virus.

In the post-neonatal period (figure 2B), we estimated 2·0 million deaths. Nearly half of these deaths were due to three diseases: lower respiratory infections, diarrhoeal diseases, and malaria. Other important causes of death during the post-neonatal period included nutritional deficiencies, meningitis and encephalitis, injuries, whooping cough, measles, and HIV/AIDS. Causes that primarily lead to death in the neonatal period also contributed to 14·0% of deaths between 1 month and 11 months including neonatal disorders, congenital anomalies, and congenital syphilis. Our analysis of lower respiratory infections disaggregated by pathogen suggested that the most important identified causes of post-neonatal lower respiratory infections were respiratory syncytial virus (RSV), Haemophilus influenzae type B (Hib), and pneumococcus. For diarrhoeal diseases, the most important pathogen was rotavirus followed by cryptosporidium. In the age group 1–4 years (figure 2C), we noted 2·0 million deaths distributed across a wider array of causes. The most important cause globally in this age group was malaria, followed by lower respiratory infections, diarrhoeal diseases, and nutritional deficiencies. These four causes accounted for 52·4% of deaths in this age group. Four causes account for between 3% and 5% of deaths each: HIV/AIDS, meningitis or encephalitis, measles, and drowning. The specific pathogens causing lower respiratory infections substantially shifted in this age group compared with the age groups of under 1 year with a much more substantial part played by pneumococcal deaths. Just under 14% of deaths in this age group were from a long list of non-communicable causes, each of which accounted for a relatively small number of deaths.

Because of the focus of MDG 5 on maternal mortality, the composition of causes of death in women and men of reproductive age is of particular interest. Although deaths related to pregnancy and childbirth have been given special priority in the MDGs, any death in these young adult age groups is a major cause for concern. For women aged 15–49 years, we estimated 3·5 million deaths from all causes in 2010. Figure 3A shows that the leading cause was HIV/AIDS, followed by cardiovascular disease, maternal disorders, suicide, tuberculosis, breast and cervical cancer, and digestive diseases and cirrhosis. The top seven causes accounted for half of the deaths of women in these age groups. Although there is no MDG related specifically to male deaths in the reproductive age groups, disorders targeted by MDG 6 take an important toll on men in the age groups 15–49 years. The leading causes of death for men in this age group, however, were cardiovascular diseases, road traffic injuries, and HIV/AIDS, with other major causes including suicide and interpersonal violence.

Open in a separate window

Figure 3:

Global deaths in 2010 for individuals aged 15–49 years

(A) Female individuals, 3 496 480 total deaths. (B) Male individuals, 5 741 344 total deaths. CKD=chronic kidney disease.

Identification of more detailed causes is perhaps more important for priority setting and planning, since interventions are generally cause-specific. Figure 4 shows the top 25 causes of death in the world ranked in 1990 and 2010 with arrows connecting the causes between the two periods. Although the top four causes of death in 1990 remain the top four in 2010, the change in numbers of deaths is noteworthy, with ischaemic heart disease and stroke increasing by 26–35% over the interval, but lower respiratory infections and COPD declining by 7–18%. Lung cancer increased from the 8th cause to the 5th cause in the two decades because of a 48% increase in absolute number of deaths. The largest change was for HIV/AIDS, rising from the 35th cause to the 6th leading cause of death. Diarrhoea, tuberculosis, and malaria all dropped in the global league table. Large increases in absolute number of deaths and their relative importance can be seen for diabetes, liver cancer, falls, and chronic kidney disease. Each of these causes has increased by more than 50% over the two decades.

Open in a separate window

Figure 4:

Global death ranks with 95% UIs for the top 25 causes in 1990 and 2010, and the percentage change with 95% UIs between 1990 and 2010

UI=uncertainty interval. COPD=chronic obstructive pulmonary disease. *Includes birth asphyxia/trauma. An interactive version of this figure is available online at http://healthmetricsandevaluation.org/gbd/visualizations/regional.

Whereas the number of deaths from a given cause is a widely understood measure, its utility as a metric for informing public health priorities is limited since it gives equal weight to a death at age 90 years compared, for example, with a death at age 25 years or 5 years. Consequently, the predominance of non-communicable causes might be misleading. In the Global Burden of Disease studies, the computation of years of life lost based on the standard expectation of life at the age of each death quantified the amount of life lost due to premature mortality (YLLs) from each cause. By computing YLLs, we could aggregate information on deaths across all ages to summarise overall patterns of premature mortality. Figure 5 shows the demographic and epidemiological transition, with the change in the composition of YLLs by single year from 1990 to 2010 for both sexes combined. The effect of major mortality shocks such as the Rwanda genocide (1994) and the famine in North Korea (covering most of the 1990s with a peak in 1995), were evident even at the global scale. The fraction of YLLs due to diarrhoea, lower respiratory infections, meningitis, and other common infectious diseases decreased substantially from 27·3% in 1990 to 15·4% in 2010. The percentage due to HIV/TB has increased due to the HIV epidemic. There is a concomitant expansion of YLLs due to non-communicable diseases, particularly cardiovascular diseases. The share of YLLs from non-communicable diseases increased from 33·3% in 1990 to 42·8% in 2010. The appendix (pp 534–725) shows the global YLLs with 95% UIs for 2010 and 1990.

Open in a separate window

Figure 5:

Percentage of global years of life lost (YLLs) from 1990 to 2010 for all ages and both sexes combined by cause and year

Figure 6 provides a comparison of the top 25 causes of YLLs for both sexes combined, which gives an even more meaningful perspective on priorities for disease control than a simple ranking of causes of death according to the numbers of deaths from each cause. The leading cause of YLLs globally was lower respiratory infections in 1990 and ischaemic heart disease in 2010; in this period, the total number of YLLs from lower respiratory infections decreased by 45% but increased 28% for ischaemic heart disease. More generally, a number of communicable, maternal, neonatal, and nutritional causes declined in both absolute terms and in relative importance as causes of YLLs—most notably measles, tetanus, preterm birth complications, tuberculosis, meningitis, and protein-energy malnutrition. Conversely, several non-communicable diseases increased in importance over the two decades: ischaemic heart disease, stroke, lung cancer, cirrhosis, diabetes, liver cancer, and chronic kidney disease particularly, although COPD and congenital causes have declined in rankings of YLLs. Among injuries, road traffic, self-harm, and interpersonal violence increased substantially in both absolute and relative terms, whereas drowning decreased.

Open in a separate window

Figure 6:

Global years of life lost (YLLs) ranks with 95% UIs for the top 25 causes in 1990 and 2010, and the percentage change with 95% UIs between 1990 and 2010

YLLs=years of life lost. UI=uncertainty interval. COPD=chronic obstructive pulmonary disease. *Includes birth asphyxia/trauma. An interactive version of this figure is available online at http://healthmetricsandevaluation.org/gbd/visualizations/regional.

An important innovation in this study was the quantification of uncertainty by cause. UIs varied widely across causes. Figure 7 shows the 95% UIs for YLLs for each cause in 2010, ordered by the mean rank of every cause. The two leading causes—ischaemic heart disease and lower respiratory infections—have nearly overlapping UIs. These two causes are separated by a substantial gap with the next highest ranked cause, stroke. The 12th ranked cause, neonatal sepsis, has a UI that is nearly three times wider than the 11th ranked cause, COPD. Several causes have much larger UIs than adjacent causes in the rank list. Natural history models for whooping cough, measles, and syphilis have large UIs. This originates from considerable empirical uncertainty on the estimation of incidence and case-fatality rates. By contrast, the HIV/AIDS natural history model developed by UNAIDS has remarkably narrow uncertainty in many countries with large epidemics. Across the causes analysed with CODEm, where the validity of UIs were evaluated using out-of-sample performance, there was also substantial variation across causes reflecting both the coherence of the underlying data, and whether powerful explanatory covariates were identified.

Open in a separate window

Figure 7:

Global YLLs (millions) with 95 % uncertainty intervals versus rank by cause in 2010

YLLs=years of life lost. An interactive version of this figure is available online at http://healthmetricsandevaluation.org/gbd/visualizations/regional.

Figure 8 shows the composition of causes of death at the second level of aggregation (21 cause groups) for the 21 GBD regions in 1990 and 2010 for both sexes combined. The regions were ordered by the mean age at death, a crude but informative measure of the demographic and epidemiological transition.⁴⁰ At both time periods, there was substantial variation across regions in the relative importance of different causes, with communicable diseases and related causes being much more important in parts of sub-Saharan Africa and parts of Asia than in north Africa, and vascular diseases and cancer predominating in most other regions. By 2010, substantial progress was achieved, even in Africa, in reducing YLLs from communicable, maternal, neonatal, and nutritional causes particularly, although these still accounted for three out of four premature deaths in Africa. The predominance of vascular diseases as a cause of premature mortality in eastern Europe is clear from figure 8, especially compared with other developed regions, where cancer causes as much, if not more, premature death. The impact of the civil violence in Papua New Guinea in 1990 and the 2010 earthquake in Haiti led to notable shifts due to war and disaster. The combination of road injuries, other unintentional injuries, and intentional injuries ranges from a high of 23% of YLLs in 2010 in central Latin America to a low of 6% in the Caribbean, nearly a four-fold variation.

Open in a separate window

Figure 8:

Percentage of YLLs for all ages and both sexes combined by cause and region in 1990 and 2010

YLLs=years of life lost. (A) 1990. (B) 2010. An interactive version of this figure is available online at http://healthmetricsandevaluation.org/gbd/visualizations/regional.

Figure 9 shows the rank for every cause that was either in the global top 25 causes of YLLs in 2010 or appeared in the top 25 causes of YLLs for any region. The appendix (p 726) presents the same information for 1990. Different colours represent different bands of ranks (figure 9). Such heat maps help to visualise important variations in ranking of YLLs across regions. In both 1990 and 2010, a similar number of causes (60 or so) appeared in the rankings, but with very substantial regional variations. At the top of figure 9, causes highest in global rankings of YLLs, such as lower respiratory diseases, ischaemic heart disease, and stroke were the top ten causes of premature death in almost all regions in 2010, as were complications of preterm birth in all regions except Europe, high-income Asia Pacific, and east Asia. The massive effect of HIV/AIDS on mortality in most developing regions by 2010 was also clear, with north Africa and Middle East, east Asia, central Asia, and southern Latin America being notable exceptions. Malaria was a leading global cause but a minor cause in most regions outside sub-Saharan Africa and Oceania. Road injury was a remarkably consistent cause of YLLs; its lowest regional ranking is 19th in Oceania and it was in the top five causes in eight regions. All the neonatal causes and tuberculosis were important causes in some developing regions but minor causes in the more epidemiologically and demographically advanced regions. Figure 9 also highlights causes that were not in the top 25 global rankings but were important in selected regions. In some cases, these regional patterns gave a glimpse of future patterns and trends. Suicide was a top ten cause in the eight regions with the most advanced health transition. Other causes that seemed to be strongly related to the epidemiological and demographic transition included colon and rectum cancer, breast cancer, pancreatic cancer, brain cancer, non-Hodgkin lymphoma, Alzheimer’s disease, kidney cancer, and prostate cancer. Other diseases had a more focal regional pattern that was not directly related to the broad health transition. More notable examples highlighted by multicoloured rows include: cirrhosis, diabetes, interpersonal violence, sickle-cell disease, whooping cough, poisonings, oesophageal cancer, drug use, gallbladder cancer, malignant melanoma, and African trypanosomiasis. Generally, the distribution of ranks by cause for YLLs was much more heterogeneous than for YLDs.⁵² This was true for both time periods and suggests marked regional variation in disease and injury control priorities to improve survival.

Open in a separate window

Figure 9:

Regional ranking of leading causes of years of life lost (YLLs), 2010

Causes in the figure are ordered according to global ranks for causes. The figure shows all causes that are in the 25 leading causes in at least one region. Ranks are also colour-shaded to indicate rank intervals. *Includes birth asphyxia/trauma. An interactive version of this figure is available online at http://healthmetricsandevaluation.org/gbd/visualizations/regional.

Marked variation was noted in cancer rates by site and overall across regions in 2010 (figure 10). Figure 10 shows crude death rates to highlight the mixture of cancers recorded in health systems of every region but to remove the effect of variation in population size across regions. Crude rates are affected both by variation in age-specific and site-specific death rates and by population age-structure. In general, crude cancer death rates were higher in the regions with a more advanced demographic transition. But regions such as high-income Asia Pacific had a substantially different cancer profile from that in western Europe due to breast cancer, liver cancer, and stomach cancer along with a number of smaller cancers. At the other end of the epidemiological spectrum, crude cancer rates in three of the sub-Saharan African regions were the lowest. Central Latin America, tropical Latin America, and Andean Latin America have low cancer rates overall, whereas the Caribbean had higher rates than expected for its demographic transition.

Open in a separate window

Figure 10:

Cancer death rates in 2010 for all ages and both sexes combined by cause and region

Discussion

The GBD 2010 is the most comprehensive and systematic analysis of causes of death undertaken to date. The addition of time trends over 1980–2010 and quantification of uncertainty increases both the utility and the methodological rigour of the results. The global health community can now draw on annual estimates of mortality, by age and sex, for 21 regions of the world, for each year from 1980 to 2010, for 235 separate causes, each with 95% UIs to aid interpretation. These estimates of cause of death at the regional level are constructed from separate estimates of cause of death at the country level for 187 countries. No such resource for policy makers, donors, or scholars currently exists. At the most aggregate level, we have documented great changes in cause of death structure in regions such as central Latin America and tropical Latin America. We have also identified regions, such as eastern Europe and central Asia, where mortality has increased profoundly over the past two decades but the cause of death structure has not changed dramatically, at least for leading causes.

The shifting pattern of the number of deaths by cause across time, regions, and age groups is consistent with the three key drivers of change: rising total population, rising average age of the world’s population, and the broad epidemiological transition. For communicable, maternal, neonatal, and nutritional causes, the increase expected because of population growth alone is reversed by population ageing and decreases in age-sex-specific death rates. By contrast, both population growth and ageing are driving up the number of deaths from non-communicable diseases and injuries more than the decreases expected because of lower age and sex-specific rates. Overall, these factors are leading to a shift from a pattern dominated by the main infectious disease killers of children (such as lower respiratory infections, diarrhoea, malaria, and meningitis), and tuberculosis and maternal causes in younger adults, to one dominated by the non-communicable causes over age 40 years. Quite different regional stories are overlaid on this broad pattern. Injuries have very distinct regional patterns, with violent death being much more common in selected regions. The HIV epidemic has had massive effects in eastern and southern sub-Saharan Africa. Diabetes has a major effect in central Latin America, the Caribbean, north Africa the Middle-East, and Oceania. Theories of the epidemiological transition need to be nuanced to capture these distinct trends and patterns that suggest different health trajectories in different regions in the coming decades. Our findings are also different from many published studies on causes of death in some age groups or for specific causes; these differences warrant careful discussion.

The HIV epidemic has greatly changed the pattern of causes of death between 1990 and 2010 in eastern and southern sub-Saharan Africa. We estimated that, in 2010, 1·47 million deaths were due to HIV/AIDS compared with UNAIDS estimates of 1·77 million, which is a 21% difference at the global level. For specific countries and regions, the differences are much larger. We used UNAIDS estimates as the input to CoDCorrect for most countries, so the difference is almost entirely due to the juxtaposition of evidence on levels of all-cause mortality and natural history model estimates of HIV/AIDS deaths. We believe that this is an important strength of the burden of disease approach. Some differences are also due to the data from Thailand, where in the 2012 UNAIDS assessment mortality doubled compared with their 2010 assessment. Even the 2010 UNAIDS assessment was twice the magnitude recorded in two studies of national verbal autopsy and vital registration data.^66,67 We chose to use the 2010 UNAIDS revision estimates for Thailand because of the implausible nature of the 2012 revision estimates. Nigeria is another example of a country whose estimates contribute to a substantial part of the differences between global data from UNAIDS and GBD 2010. Since Nigeria is a country with a large population, a major HIV/AIDS epidemic, and poor data quality, when we fit our HIV/AIDS estimates for Nigeria into the country’s all-cause mortality levels our estimates are much lower (27·9% in 2010) than those of UNAIDS and this large difference in death is reflected at the global level. The uncertainty distributions for UNAIDS estimates of mortality in some countries in sub-Saharan Africa are implausibly narrow; for example, at the peak of the epidemic in Malawi, the UI for deaths at all ages provided by UNAIDS varied by plus 8·5% or minus 8·1%. Our UIs based on the UNAIDS figures were also implausibly narrow. Improved estimation of mortality from HIV/AIDS including uncertainty in the future will come both from continued progress in the estimation of the timecourse of the HIV epidemic by UNAIDS and from further data on the levels of adult mortality in some key countries such as Nigeria.

On the basis of a single-cause analysis, Murray and colleagues²⁵ reported 1·24 million malaria deaths in 2010 of which 42% were in individuals older than 5 years. These estimates had very large UIs; for example, the number of deaths in those older than 5 years in sub-Saharan Africa was estimated to range from 307 000 to 658 000. Large uncertainty in the results reflected both the sparse data available on causes of death in adults and the large variation in results across different models included in the final ensemble model. In these GBD 2010 results, where the sum of cause-specific estimates must equal the number of deaths from the demographic analysis for each country-age-sex-year, the number of malaria deaths in 2010 was estimated to be about 5% less or about 1·17 million. For deaths in children younger than 5 years, the change was from 714 000 to 676 000. For deaths in individuals older than 5 years, the change was from 524 000 to 494 000. Our finding of substantial deaths due to malaria in populations older than 5 years is driven by the evidence from verbal autopsy studies in endemic areas.²⁵ Validation studies suggest that verbal autopsy studies overestimate adult deaths at low malaria cause fractions and underestimate adult malaria deaths when malaria deaths are common.⁶⁸ The findings of adult deaths from malaria are consistent with data of hospital discharge and death in endemic areas but remain controversial.^69–72 Our UI for global deaths in individuals older than 5 years from malaria was 365 356 to 643 977, partly depicting the uncertainty in the underlying data sources.

An innovative dimension of the GBD 2010 has been the addition of estimates of deaths due to different diarrhoea and lower respiratory infection pathogens. These are important both for the prioritisation of existing treatments, such as rotavirus or pneumococcal vaccines, and for the development of future technologies. Making sense of the data by aetiology is extremely challenging. For diarrhoea, many pathogens can be cultured from the stool of individuals without diarrhoea. Studies such as the Global Enterics Multi-Center Study (GEMS)⁷³ are trying to estimate relative risks of diarrhoea in the presence of different pathogens. In the available observational data that do not use this relative risk approach, the relation between the prevalence of a given pathogen and the number of pathogens tested is strong. Studies testing only one pathogen effectively report higher fractions due to a pathogen than do studies that test for multiple pathogens. This finding is probably caused by the frequency of various pathogens in the same stool sample and rules for allocating shares of diarrhoea to pathogens such that the sum of the pathogen cause fractions total to 100%. In our analysis, we adjusted study results to be equivalent to studies reporting two to eight pathogens. Because of both the huge heterogeneity of results and the variation in the number of pathogens tested across studies, great caution should be used in interpreting our findings on the diarrhoea aetiologies. When large multicentre studies such as GEMS publish their results, their findings will be an important addition to this analysis; future revisions of the GBD should make use of these results as they become available. Nevertheless, our results are notably different from widely cited findings. For example, we reported 173 000 deaths due to rotavirus in children younger than 5 years and 78 000 deaths in individuals older than 5 years in 2010; these findings contrast with claims from WHO of 453 000 rotavirus deaths in children younger than 5 years alone in 2008.⁷⁴ Higher numbers were probably reported by WHO for three reasons: higher all-cause global deaths in children younger than 5 years than currently estimated by UNICEF or the GBD 2010; a much higher fraction of deaths in children younger than 5 years attributed at the time to diarrhoea; and a higher fraction of diarrhoea attributed to rotavirus. Because rotavirus remains an important cause of death in many countries, the GBD estimates by country, when released, will be an important method to assist in understanding where its burden is greatest.

For respiratory pathogens, even greater challenges exist. In many observational studies, no pathogen is identified in a substantial fraction of cases. Even in severe cases that lead to hospital admission, the case-fatality rate is likely to vary substantially by pathogen, which confounds the analysis. The substantial differences in our results from published assessments by O’Brien and colleagues³⁵ for pneumococcal lower respiratory infections, Nair and colleagues⁷⁵ for RSV, and Watt and colleagues⁷⁶ for Hib deserve exploration. Our findings of 168 000 deaths due to pneumococcal lower respiratory infection in children younger than 5 years in 2010 and 381 000 in 1990 contrast sharply with the 826 000 for the year 2000 published by O’Brien and colleagues.³⁵ in 2009, these investigators used an estimate of 10·6 million deaths in children younger than 5 years from all causes for the year 2000, which contrasts with our estimate of 9·4 million and UNICEF’s estimate of 9·6 million. They used a higher estimated fraction of deaths in children younger than 5 years due to lower respiratory infections than in our study, 27% and 18%, respectively. The fraction of lower respiratory deaths in children younger than 5 years due to pneumococcal pneumonia was 35·8% (UI 16·0–50·9) in 2000 compared with 19·8% (16·1–24·8) in the GBD 2010 for 2010. The 95% UIs for these cause fractions substantially overlap. Differences in the mean estimate result largely from the exclusive use by O’Brien and colleagues of the results of four vaccine trials in the Gambia, the Philippines, the USA, and South Africa to estimate an average fraction of pneumonia in children younger than 5 years. Although they reviewed the literature, they chose not to use the published observational studies. These observational data suggest substantial variation across regions; for example, pneumococcus might be less common in south Asia. We used both the trials and observational data to generate different breakdowns by pathogen by region giving extra weight to the trials; our findings, however, still showed variation by region and age. For RSV lower respiratory infections, our findings of 234 000 deaths in children younger than 5 years in 2010 and 520 000 in 1990 are notably higher than the 66 000 to 199 000 deaths for the year 2005 reported by Nair and colleagues⁷⁵ in 2010. They reviewed the published studies on RSV but chose, on the basis of expert opinion, to assume no RSV deaths in individuals older than 2 years;⁷⁵ the scientific literature, however, records deaths in those older than 2 years.^77,78 Data on hospital admission in the USA, for example, suggests substantial burden of severe RSV at least in the ages 3–5 years. Other investigators^79,80 argue that RSV is an important and often missed pathogen causing severe lower respiratory infections in elderly people. Studies, however, vary markedly in their sampling, culturing, and identification protocols, which might also account for some heterogeneity. Most studies show high fractions of neonatal lower respiratory deaths due to RSV, so the extent to which RSV is a major cause of post-neonatal lower respiratory deaths needs the most attention. For deaths due to Hib lower respiratory infection, Watt and colleagues⁷⁶ published in 2009 for the year 2000, an estimate of 292 000 deaths that compares well with our finding of 184 000 (95% UI 154 053–219 456) in 2010 and 447 000 (385 717–506 594) in 1990. The similarity of the result, however, is somewhat misleading. They used a much higher estimate of all-cause under-5 mortality and a higher cause fraction due to lower respiratory infection than we used. In other words, we estimated a larger fraction of lower respiratory infection deaths in children younger than 5 years due to Hib than did Watt and colleagues. The main reason for this difference is that Watt and colleagues assumed that Hib in individuals older than 2 years does not cause death; the available observational data,^81–84 however, show Hib mortality at all ages. For example, results from the study in Japan⁸⁴ showed severe cases of hospital-acquired and community-acquired pneumonia related to Hib, the differences in these assessments for all the respiratory pathogens is even more striking. New multicentre studies such as PERCH⁸⁵ will hopefully provide much needed data to strengthen the analysis of pathogens by region.

We estimate that in 2010 1·20 million deaths were due to tuberculosis, about 14% more than the WHO estimate of 1·05 million (2011 WHO TB Report). Our analysis and that of WHO use fundamentally different methods but do not yield, at the aggregate level, substantially different conclusions. The key difference lies in how the case-detection rate in every country was estimated; WHO used locally informed expert consultation to assess both under-diagnosis and under-reporting whereas we used a statistical model to try and estimate the case detection rate. Our models also captured more of the temporal and spatial correlation structure in the cause of death data yielding quite different estimates than did the WHO where data were strong (eg, Japan and Thailand). These differences result in important variations between our estimates and those from WHO at the country and regional levels that deserve further investigation. Despite these variations, for some countries, such as Ecuador, our estimates, WHO estimates, and the cause of death data were in close alignment. Assessments at the global and regional level point to substantial continued and sustained progress in reducing tuberculosis mortality. Our higher levels of mortality for tuberculosis in 2010 suggest that tuberculosis should remain a major priority.

Deaths due to maternal causes have been reported in various studies^86,87 in the peer-reviewed scientific literature and UN reports.⁸⁸ These studies, however, focused on the maternal mortality ratio. ICD 10 rules recommend that pregnancy-related deaths due to HIV should be included in the computation of the maternal mortality ratio but reported in cause of death tabulations for HIV. The GBD 2010 follows this convention so that our 255 000 maternal deaths in 2010 do not include 18 970 HIV-related deaths in pregnancy included in the HIV totals. We also revised the method used to estimate the fraction of maternal deaths related to HIV/AIDS in this study compared with Lozano and colleagues.²³ In GBD 2010, our estimates of the fraction of maternal deaths related to HIV/AIDS come from a review and statistical analysis of available data that provide a detailed breakdown of maternal deaths. Further, the deaths due to maternal causes reported here are after the application of the CoDCorrect algorithm, which is a strength of the comprehensive burden of disease approach. The global number of deaths due to maternal causes was reduced by 9·8% through the application of CoDCorrect. The revised approach for maternal mortality presented here also highlights the major causes of maternal death. The largest specific cause of maternal death is maternal haemorrhage accounting for 23% of deaths in 2010, followed by hypertensive disorders causing 18%, abortion causing 15%, sepsis causing 9%, and obstructed labour causing 4%. These results contrast with various estimates reported by WHO—one estimate for 2005, which is based on 35 197 deaths reported in various published studies⁸⁹ and one largely based on vital registration data for 2008.¹⁶ The 2005 WHO report estimated that abortion caused only 8% of maternal deaths but 14% in the 2008 study.¹⁶ There are also substantial differences for other maternal causes as well. Our findings are based on a larger set of published studies, including reports from 18 countries such as India, South Africa, Bangladesh, Ghana, and Pakistan, which contribute 22 943 additional maternal deaths on the detailed causes of maternal death.

For deaths in children younger than 5 years, our results differ from those published by CHERG²⁰ for the same year in several key ways: CHERG estimated 1·40 million deaths due to lower respiratory infection compared with 847 000 in our study; CHERG estimated 801 000 for diarrhoeal diseases versus 666 000 in the GBD 2010; for malaria CHERG estimated 564 000 versus 676 000 here. Within the neonatal causes, there are also striking differences in the composition of specific neonatal causes, with CHERG estimating much higher fractions due to preterm birth complications and lower fractions due to sepsis and other disorders. In exploring these differences, it is crucial to distinguish between the neonatal age group (under 28 days) and causes arising during the neonatal period that can cause mortality both under 28 days and in the post-neonatal period and less commonly over age 1 year. Understanding the source of the differences for neonatal causes, however, is challenging. Differences must arise from any or all of the following: the datasets used, the adjustments to the data, and the modelling strategies. In their analysis of post-neonatal child mortality (1–59 months), CHERG reported using vital registration data for 578 country-years between 1998 and 2009, whereas we used 1125 country-years for that period. They used 113 study years between 1980 and 2008; by contrast, we used 294 years. CHERG used a reduced number of studies because of their decision to use only studies that report on six major causes. The largest differences and the ones likely to explain the differing results lie in the modelling strategy. For a single cause such as diarrhoeal diseases or lower respiratory infections, we developed one ensemble model using all the data. The one ensemble model included component models with a wide variety of functional forms but all included age group fixed effects. CHERG tested and developed four different models with no relation between them: for under-5 mortality rates (UFMR) lower than 35 for babies younger than 1 month, for UFMR higher than 35 for babies younger than 1 month, for UFMR lower than 35 for children aged 1–59 months, and for UFMR higher than 35 for children aged 1–59 months. The relations between covariates in these different models were estimated as completely independent. Covariate selection was done for every cause of death independently for each of the four models. They used multinomial logistic models to make estimates for six causes of death simultaneously, a method found to perform worse than modelling causes individually and then scaling to all-cause mortality (appendix pp 146–148). Their model structure does not allow for spatial or temporal patterns in the residuals, probably substantially reducing model performance.²² Model performance is difficult to assess and compare with ours: they reported undertaking a cross-validation study but no metrics of model performance were reported such as the coverage of their UIs or a measure of prediction error. Further, in their cross-validation study they left out 10% of the data at random; as shown in previous studies,²³ this is the easiest task for prediction. A much harder task, which was used in the GBD 2010, is to leave out long sequences or all data for some countries and see how well the models perform. Additionally, leaving out data for one cause without also dropping it for the other causes in that country-year makes for an even simpler task in a multinomial logistic model such as the one CHERG used. The high levels of mortality for lower respiratory infections might be related to the bias in physician diagnosis of verbal autopsies, which was demonstrated in a rigorous validation study.⁶⁸ Although this study⁶⁸ used physician-certified verbal autopsy data on lower respiratory infections, by separating the data into four components and not including country random effects in their models, the bias towards higher such infections in verbal autopsy studies might have had a larger effect on their estimation procedure. Finally, the CHERG estimation strategy uses multinomial models for a subset of models and then adds on estimates for selected causes in selected countries such as HIV/AIDS, measles, tetanus, and malaria. The differential treatment of some causes and not others in the modelling strategy might also have caused distortions in the results. Interestingly, given the substantial difference in modelling approaches, the results for 2010 between the two approaches were actually surprisingly close.

The health-related MDGs place special priority on reducing under-5 mortality, maternal mortality, and deaths from HIV, tuberculosis, and malaria. These causes collectively accounted for 42·4% of YLLs in 2010. Even though the computation of YLLs heavily weights deaths in children younger than 5 years, more than half (57·6%) of global YLLs in 2010 were due to non-MDG diseases and injuries. Important global causes that are not included in the MDGs include ischaemic heart disease, stroke, COPD, road traffic injuries, and self-harm. The predominance of these causes in YLL rankings is not merely a volume issue: many of those who die from these causes do so at young adult ages. A more holistic view of preventable mortality within the MDG platform would argue that these causes ought to be included in any evidence-based framework for reducing avoidable deaths. Examination of the trends from 1990 to 2010 indicated that the MDG-related YLLs were declining at 2·0% per year, whereas the non-MDG related YLLs were increasing at 0·8% per year. Population ageing and the substantial if incomplete progress in reducing age-specific death rates from the MDG related causes all suggest that these trends will continue. Indeed, if they do, then non-MDG related causes are likely to account for more than two-thirds (67·6%) of YLLs by 2025. These findings highlight the importance of looking more critically and comprehensively at what are the leading causes of death and YLLs worldwide, and how these are changing. Our analyses, for the first time, allow such comparative assessments and are important inputs into discussions about goals and targets for the post-MDG era.⁹⁰ The rapid and global rise in premature death from leading non-communicable diseases argues strongly for inclusion of these disorders, and their principal causes, in this agenda, particularly in view of their close relation to poverty reduction goals.^91–96 It also stresses the need to understand the effective and affordable options for prevention of non-communicable diseases and injuries and treatment including both medical and surgical interventions.⁹⁷

Our study suggests that the number of deaths where chronic kidney disease is the ICD underlying cause of death increased by 82% from 1990 to 2010. In addition to these deaths, a reduced glomerular filtration rate (GFR) has been associated with an increased risk of death.⁹⁸ Even chronic kidney disease stages II–IV are associated with increased risk of death. The directly coded deaths due to chronic kidney disease that we estimated probably capture only those deaths due to end-stage renal disease. Other diseases such as diabetes are also associated with an increased risk of death from other causes. For diabetes, the risk factor analysis⁹⁹ provides an assessment of all mortality associated with hyperglycaemia. This number is substantially larger than the number of deaths directly coded to diabetes estimated here.

Those who study the health effects of war will be surprised by the 17 670 deaths related to direct conflict estimated for 2010. This number should be interpreted with great caution. First, in the ICD cause list, only direct deaths are assigned to this cause; 17 670 is not the total number of deaths related to conflict, which would include indirect deaths mediated through various mechanisms such as the destruction of health-care infrastructure.¹⁰⁰ Second, the number of direct deaths varies substantially from year to year. During 1990–2010, direct deaths peaked at 496 400 in 1994 with a low in 2001 of 14 700 and 17 700 in 2010. In 2011, because of the conflict in Libya, for example, direct deaths were likely to have been much higher, closer to 61 000. For episodic events such as wars or natural disasters, it is important to consider the burden of disease over longer periods of time to fully appreciate their effect on human populations.

For the first time in the GBD enterprise, we included deaths that were mainly related to hepatitis B, hepatitis C, alcohol, and other causes as disaggregated causes for cirrhosis and liver cancer. These categorical breakdowns are not counterfactual assessments but rather an attempt to assign deaths to the primary or dominant cause. Interactions exist between hepatitis and alcohol consumption, such that assessment of these conditions as a risk factor would give different results. Nevertheless, this categorical attribution provides a useful assessment of the magnitude of direct burden, particularly for guiding intervention priorities. The total number of deaths due to hepatitis B in 2010 was estimated to be 786 000 and those due to hepatitis C 499 000. If all deaths related to these diseases were directly counted in the main GBD 2010 cause list, hepatitis B would be the 15th ranked cause of death and hepatitis C would be the 25th ranked cause of death. Cirrhosis rates vary greatly across countries, with Egypt having the highest level related to an iatrogenic epidemic of hepatitis C that began as early as the 1920s.¹⁰¹ Some regions have high rates such as central Asia, Oceania, eastern and western sub-Saharan Africa, eastern Europe, and central Latin America. Not all this regional and country variation can be explained by hepatitis B, hepatitis C, or total alcohol consumption. For example, high rates in eastern Europe might be related to the content rather than volume of alcohol consumed. Theories on this distinction include hepatotoxic alcohol constituents in homemade poor quality alcohol, which is common in these regions.¹⁰² Given that much of the burden of cirrhosis can be preventable, its substantial global mortality deserves more policy attention.

The much more detailed categories of causes of death for injuries in the GBD 2010 provide some important insights into the global epidemic of road injuries. The number of deaths increased from 908 000 in 1990 to 1·329 million in 2010. These results are similar to the 1·237 million reported for 2007 by WHO.¹⁰³ The composition of this increase in road injuries, however, has differed by subcause. Road deaths to pedestrians were the major cause, rising from 284 000 in 1990 to 461 000 in 2010. Road deaths of occupants in motorised vehicles with three or more wheels and road deaths of riders of motorised vehicles with two wheels each have also increased by about 200 000 in the past two decades. Regional detail shows road deaths in east Asia, south Asia, and eastern and western sub-Saharan Africa rapidly escalating over the past two decades, whereas in high-income areas with a history of road safety programmes such as western Europe and high-income North America, road deaths have decreased.

Violence as a cause of death is one of the most heterogeneous across different regions. Crude death rates for violence are lowest in high-income Asia Pacific at one per 100 000. The rate in 2010 in high-income North America dominated by the USA of seven per 100 000 was nearly seven times higher than high-income Asia Pacific, western Europe, or Australasia. But in tropical Latin America, the rates are substantially greater still, at 30 per 100 000, and even higher in central Latin America and southern sub-Saharan Africa, at 33 per 100 000. The huge variation in violence raises important questions about the origins and sociopolitical context of violence, the drivers of change in violence-related mortality, and the effectiveness of public health strategies in reducing deaths from violence. In men from tropical Latin America, violence is the number one cause of YLLs. In 2010, men in the 20–24 years age group alone had 653 600 YLLs, three quarters the size of the 824 000 YLLs in men in high-income North America of all ages combined.

An important dimension to the GBD is the requirement that estimates of causes of death sum to estimates of all-cause mortality. The discipline of requiring this internal consistency has been a hallmark of burden of disease analysis since the GBD 1990. As quantified in the appendix (p 144) the effect is to reduce the number of deaths estimated for many causes compared with single-cause analyses. The correction factor is an indication of inconsistency at the country-age-sex-year level between demographic analysis and the cause-specific analyses. We believe that for causes for which the magnitude of these corrections is comparatively large, future research should be targeted to try to build a better understanding of the strengths and weaknesses of the various data sources, whether epidemiological or demographic. In some sense, the CoDCorrect ratios can help direct attention to settings in which the data are the most inconsistent and our knowledge the most uncertain. The substantial difference between single-cause-model estimates and those presented here raises questions about the value of publishing single-cause assessments. Some organisations such as WHO have in recent years been producing both types of assessments: WHO with CHERG produces estimates for 16 major causes of child death but also publishes single-cause estimates for tuberculosis, HIV/AIDS, malaria, maternal mortality, and other causes. Should leading scientific journals continue to review and publish studies on single-causes of death? Due consideration of the value of single-cause models to bring attention to neglected issues or stimulate innovation in methods or new data collection will need to be balanced against the greater robustness of more comprehensive assessments such as those presented here.

A study of this size and scope has many limitations. The ambition to estimate mortality from 235 causes with uncertainty for 187 countries over time from 1980 to 2010 means that many choices about data sources, quality adjustments to data, and modelling strategies had to be made. We highlight some key limitations. First, data on cause of death, even in settings with medical certification, might not always accurately capture the underlying cause of death. Results from autopsy studies^104–106 have shown that medically certified causes of death might be incorrect. Second, our approach to garbage code redistribution, although an improvement over past efforts, could benefit strongly from more empirical information on misclassification obtained in places where gold standard cause of death assignment is possible. We were not able to develop uncertainty distributions around garbage code redistribution algorithms; to the extent that this is poorly known, our UIs for some causes might be underestimated. Third, we made extensive use of verbal autopsy data, especially in low-income settings. Verbal autopsy validation studies⁶⁸ suggest that verbal autopsy is accurate for some causes such as breast cancer, drowning, or road injury, but less accurate for other causes. Verbal autopsy performance for some key causes of child death such as lower respiratory infections is particularly poor. Much could be learned about causes of death in countries where death certification is poor through the more widespread testing and application of recent advances in verbal autopsy methods, which greatly reduce heterogeneity in diagnostic practices across populations in which verbal autopsy is currently used.¹⁰⁷ Fourth, for some causes of death such as kidney cancer, poisonings, or paralytic ileus, only weak covariates have been identified that explain the spatial or temporal variation in the cause. Inevitably, model estimates for these causes will have wide UIs. The use of negative binomial models and fixed proportion models where data are extremely weak is another area for which better data and improved methods could strengthen the overall findings. Fifth, where natural history models have been used, their validation is extremely difficult at present. Natural history models are in principle used when concerns exist that data on direct cause of death are potentially biased. Improvements in data on cause of death for some causes such as HIV/AIDS might allow in the future opportunities to validate natural history models in selected countries. Where natural history models have been used, these approaches potentially will tend to yield higher estimates than those using more empirical strategies such as CODEm. Sixth, our use of CODEm for most major causes of death means that our UIs have in most cases been shown to be valid, but for causes where we have had to use other methods such as negative binomial, fixed proportion, or natural history models, the UIs have not been independently validated. Seventh, CODEm can be improved in the future by including an even broader set of model families. Ultimately, the greatest limitation is the availability of data on cause of death itself. Finally, in cases where expert opinion and the available data diverge, we tended to follow the available data. Examples of this practice include estimating deaths from malaria in individuals older than 5 years or deaths from Hib in those older than 2 years. Subsequent more detailed studies might affirm that expert opinion was correct and the available data substantially biased. Nevertheless, we believe that to follow the balance of the available data that meet our quality criteria is important.

Improving data collection on cause of death in the future is the most direct and obvious pathway to improved estimation of global, regional, and national causes of death with narrower UIs. Improved verbal autopsy methods^108–111 mean that it might soon be feasible to apply them routinely to generate comparable data on cause of death cost effectively in populations in which we are still substantially ignorant about the leading causes of death. Opportunities for strengthening death registration, cause of death certification, and the more widespread use of verbal autopsy exist. Some countries have civil registration systems that capture less than 70% of deaths; the priority in such cases is to improve certification and coding of cause of death. Other countries such as Saudi Arabia have functioning civil registration run by Ministries of the Interior that are not fully used by Ministries of Health. Collectively, the global health community would benefit enormously by placing much greater priority on strengthening vital registration systems to improve measurement of cause of death. This is now the key focus of the Health Metrics Network and it is reasonable to expect that substantial progress can be made with appropriate leadership, attention, and collaboration among global development partners.¹¹²

In the GBD 2010, a substantially new set of analytical approaches and methods have been developed and applied. These methods range from improved diagnostic redistribution methods to CODEm and CoDCorrect, drawing on information for almost a billion deaths and time series for hundreds of covariates that affect mortality. This is a massive endeavour, but, with appropriate investment and leadership, updating results as new data on causes of death or alternative covariates become available will be much more feasible than hitherto. Rather than huge periodic revisions of the GBD every decade, it is now feasible to conduct annual updates so that the consumers of health intelligence have the most recent and comprehensive information on comparative causes of disease burden available where and when it is required to help guide public health decision making. Public health priorities everywhere are changing, or soon will be, as large and avoidable causes of disease burden become more common with development. To not have strategically important and comparable health information available and used to inform the new health dialogue and disease control priorities, as we have shown here it can be, would be a massive missed opportunity for global health.

Supplementary Material

Acknowledgments

Footnotes

References