Projections of global mortality and burden of disease from 2002 to 2030 — TEI-tagged view

Colin D MatherspersName, Dejan LoncarpersName (2006date). “Projections of global mortality and burden of disease from 2002 to 2030.” PLoS Medicinetitle·j.

Object of study — global and regional cause-specific mortalityterm and disease-burden trajectories from the 2002date baseline to 2030date

Research question — how global mortalityterm, population health, and DALYterm·def burden may change to 2030date under baseline, optimistic, and pessimistic socioeconomic and intervention scenarios

Contribution — updated country-level, cause-specific projections of deaths and DALYsterm·def to 2030date using revised inputs, calibrated structural regressions, separate disease models, and three explicit scenarios

Full text — tagged

The reference list is not part of the text we hold for this paper. Our copy was extracted from the publisher's page and the bibliography did not survive that extraction, so the article below ends before its references. Nothing has been removed by us.

Projections of global mortalityterm and burden of diseaseterm from 2002date to 2030date Abstract Global and regional projections of mortalityterm and burden of diseaseterm by causeterm·var for the years 2000, 2010, and 2030date were published by Murray and Lopez in 1996 as part of the Global Burden of Diseaseterm project. These projections, which are based on 1990 data, continue to be widely quoted, although they are substantially outdated; in particular, they substantially underestimated the spread of HIV/AIDS. To address the widespread demand for information on likely future trends in global health, and thereby to support international health policy and priority setting, we have prepared new projections of mortalityterm and burden of diseaseterm to 2030date starting from World Health Organization estimates of mortalityterm and burden of diseaseterm for 2002date. This paper describes the methods, assumptions, input data, and results.

Relatively simple models were used to project future health trends under three scenarios—baseline, optimistic, and pessimistic—based largely on projections of economic and social developmentterm, and using the historically observed relationships of these with cause-specific mortality ratesterm. Data inputs have been updated to take account of the greater availability of death registration dataterm and the latest available projections for HIV/AIDS, income, human capitalterm, tobacco smoking, body mass indexterm, and other inputs. In all three scenarios there is a dramatic shift in the distribution of deaths from younger to older ages and from communicable, maternal, perinatal, and nutritional causesterm·var to noncommunicable diseaseterm causesterm·var. The risk of death for children younger than 5 y is projected to fall by nearly 50% in the baseline scenarioterm between 2002date and 2030date. The proportion of deaths due to noncommunicable diseaseterm is projected to rise from 59% in 2002date to 69% in 2030date. Global HIV/AIDS deaths are projected to rise from 2.8 million in 2002date to 6.5 million in 2030date under the baseline scenarioterm, which assumes coverage with antiretroviral drugs reaches 80% by 2012. Under the optimistic scenarioterm, which also assumes increased prevention activity, HIV/AIDS deaths are projected to drop to 3.7 million in 2030date. Total tobacco-attributable deathsterm are projected to rise from 5.4 million in 2005 to 6.4 million in 2015 and 8.3 million in 2030date under our baseline scenarioterm. Tobacco is projected to kill 50% more people in 2015 than HIV/AIDS, and to be responsible for 10% of all deaths globally. The three leading causesterm·var of burden of diseaseterm in 2030date are projected to include HIV/AIDS, unipolar depressive disorders, and ischaemic heart disease in the baseline and pessimistic scenariosterm. Road traffic accidents are the fourth leading causeterm·var in the baseline scenarioterm, and the third leading causeterm·var ahead of ischaemic heart disease in the optimistic scenarioterm. Under the baseline scenarioterm, HIV/AIDS becomes the leading causeterm·var of burden of diseaseterm in middle- and low-income countriesterm by 2015.

These projections represent a set of three visions of the future for population health, based on certain explicit assumptions. Despite the wide uncertainty rangesterm around future projections, they enable us to appreciate better the implications for health and health policy of currently observed trends, and the likely impact of fairly certain future trends, such as the ageing of the population, the continued spread of HIV/AIDS in many regions, and the continuation of the epidemiological transitionterm in developing countries. The results depend strongly on the assumption that future mortalityterm trends in poor countries will have a relationship to economic and social developmentterm similar to those that have occurred in the higher-income countries.

Introduction As part of the groundbreaking Global Burden of Diseaseterm (GBD) study for 1990, Murray and Lopez [ 1 , 2 ] prepared projections of mortalityterm and burden of diseaseterm by causeterm·var to 2000, 2010, and 2020 under three alternative scenarios. These projections have been widely used and continue to be widely quoted to provide information on likely future trends in global health, for example in the Atlas of Heart Disease and Stroke [ 3 ]. However, these projections were based on the GBD 1990 estimates and on projections of HIV/AIDS, smoking, income, and human capitalterm from 1990 to 2020, and are now outdated. The HIV/AIDS projections in particular substantially underestimate the spread of the HIV epidemic and the level of HIV/AIDS mortalityterm around 2000.

We have thus prepared updated projections of mortalityterm and burden of diseaseterm from 2002date to 2030date using methods similar to those of the original GBD study and new input information. An earlier version of these projections has already been extensively used in the recently released World Health Organization (WHO) global report “Preventing chronic diseases: A vital investment” [ 4 ].

There is a substantial literature on the projection or forecasting of all-cause mortalityterm rates and mortalityterm rates for specific diseases. The methods used fall into two broad groups. First are those methods based on time-series analysisterm of historical trends in mortalityterm rates. These “aggregate modelsterm,” whether for all-cause mortalityterm or for specific causesterm·var, use the previous trend of the variable of interest as the basis for predicting its future value. By their data requirements, such methods are generally limited to high-income countries with good death registration dataterm (see, for example, Lee and Carter [ 5 , 6 ]). Second are the “structural modelsterm,” which are based on relationships between mortalityterm and a set of independent variablesterm, and are necessarily projections of those independent variablesterm. To the extent that the structural modelterm identifies the important components—and the relationships among them—of the “system” that determines the variable of interest, they offer the potential for more robust predictions. When the underlying system is complex and sensitive to one or more of its components, a shift in some of the system variables can introduce large changes in the outcome that may be missed by extrapolation (such as the discovery of antibiotics and infectious disease trends or the change in tuberculosis mortalityterm after the HIV epidemic). Aggregate modelsterm, in contrast, require considerably less knowledge of the system components and the relationships among them. These models can therefore provide more reliable estimates when such information is not available, especially when the system is not very sensitive to its inputs in timeterm·var intervals that are in the order of the prediction timeterm·var.

The GBD projections fall into the structural modelsterm class, as do a number of more sophisticated and data-demanding risk factorterm–based models for specific causesterm·var [ 7 – 10 ]. Methods also vary as to whether mortalityterm is projected for all age groupsterm simultaneously or separate models are developed for each age–sexterm·var group, and as to whether separate cause-specific projections are carried out. To our knowledge, the original GBD projections represent the only attempt to date to project cause-specific mortalityterm, rather than all-cause mortalityterm, for a complete set of causesterm·var at global and regional levels. The GBD projections use separate structural modelsterm for each of a small set of major causeterm·var groups and cause-compositional models for detailed causesterm·var within these groups. The latter are a special case of more general cause-compositional models that relate the proportional distributions of causesterm·var of death to the levels of all-cause mortalityterm [ 11 ].

Recently, both types of projection modelterm have been extended to allow incorporation of prior knowledge and enforce coherence in the resulting mortalityterm estimates across age groupsterm, causesterm·var, and regions. Czado et al. [ 12 ] extended the Lee-Carter Poisson log-bilinear model into a Bayesian model, in which the prior data impose smoothness of age and period effects in the projections. Girosi and King [ 13 ] developed a forecasting model that models future values of log-mortality as functions of covariates known to affect risk of death. They also used a Bayesian approach to allow different country–age–group cross-sections to “borrow strength” from each other in order to overcome the serious data limitations faced for many countries. Their model also constrains the projections to vary smoothly over timeterm·var and across age groupsterm in accordance with known age profiles for specific causesterm·var of death.

The detailed application of such methods may result in substantially improved forecasts of global mortalityterm and burden of diseaseterm trends in the next few years. In the meantime, these updated projections provide a comprehensive update of the original GBD projections, with some methodological improvements and extensions.

Methods The methods used in the present study for the projections of mortalityterm and burden of diseaseterm are similar to those used in the original GBD study [ 2 ]. Separate projections models for males and females and for seven age groupsterm (0–4, 5–14, 15–29, 30–44, 45–59, 60–69, and 70 y and older) were developed to produce parsimonious equations for ten cause-of-death clustersterm. Baseline, pessimistic, and optimistic projections of covariates in the regression equationsterm were developed and used to project mortalityterm rates for the causeterm·var clusters. Separate projection modelsterm were developed for HIV/AIDS, tuberculosis, lung cancer, diabetes mellitusterm, and chronic respiratory diseasesterm. Separate regression equationsterm were developed relating age- and sex-specific mortalityterm rates for 132 detailed causesterm·var to the age- and sex-specific mortalityterm rates from corresponding causeterm·var clusters, based on International Classification of Diseases Ninth Edition (ICD-9) coded vital registration data from 98 countries.

Whereas the original GBD study carried out projections for eight regions of the world, we carried out these projections at country level, but aggregated the results into regional or income groupsterm·var for presentation of results. Four income groupsterm·var were defined based on World Bank estimates of GDP per capitaterm in 2001 (see Table S1 for definitions of the regional and income groupsterm·var). Baseline estimates at country level for 2002date were derived from the GBD analyses for 2002date as published in the World Health Report 2004 [ 14 ]. Data sources and methods for the 2002date estimates were comprehensively documented by MatherspersName et al. [ 15 ], together with an analysis of uncertainty levels of the results [ 16 ]. Mortalityterm and burden of diseaseterm trends from 1990 to 2001 were recently analyzed by Lopez et al. [ 17 ].

Projections of burden of diseaseterm measured in disability-adjusted life yearsterm (DALYsterm·def) were then developed for the three scenarios based on projected mortalityterm rates and alternative assumptions for those causesterm·var with little or no mortalityterm. Population projections were developed based on United Nations Population Division projections of fertility and net migration and our projected mortalityterm rates. We describe the main features of the methods, assumptions, and input data in this section, with a particular emphasis on revisions and changes to the original methods used by Murray and Lopez. Protocol S1 and Tables S1 – S7 provide full details.

Regressions for Major-Cause Clusters Disease and injury causesterm·var of death are classified in the GBD using a tree structure in which the first level comprises three broad causeterm·var groups: Group Iterm·def (communicable, maternal, perinatal, and nutritional conditionsterm), Group IIterm·def (noncommunicable diseasesterm), and Group IIIterm·def (injuriesterm) [ 15 ]. We projected age- and sex-specific death rates at country level for ten major-cause clusters: Group Iterm·def conditions excluding HIV/AIDS (communicable, maternal, perinatal, and nutritional conditionsterm as defined in Table S2 ), malignant neoplasms (excluding lung cancer), diabetes mellitusterm, cardiovascular diseases, digestive disorders, chronic respiratory conditions, other Group IIterm·def diseases, road traffic accidents, other unintentional injuriesterm, and intentional injuriesterm. In contrast to the original GBD projections, diabetes mellitusterm was treated as a separate causeterm·var category from other Group IIterm·def diseases, since the available evidence suggests that its dominant risk factorterm, overweight and obesity, is becoming more prevalent over timeterm·var in both developed and developing regions and is projected to continue to rise [ 18 ].

Rather than attempt to model the effects of the many separate direct determinants or risk factorsterm for disease from the limited data that are available, the GBD methodology considered a limited number of socioeconomic variablesterm: (1) average income per capita, measured as gross domestic product (GDP) per capita; (2) the average number of years of schooling in adults, referred to as “human capitalterm”; and (3) timeterm·var, a proxy measure for the impact of technological changeterm on health status. This latter variable captures the effects of accumulating knowledge and technological development, allowing the implementation of more cost-effective health interventions, both preventive and curative, at constant levels of income and human capitalterm [ 1 ].

These socioeconomic variablesterm show clear historical relationships with mortalityterm rates, and may be regarded as indirect, or distal, determinants of health, although it is not necessary for the purposes of projection to know whether these associations are causal. In addition, a fourth variable, tobacco useterm, was included in the projections for cancers, cardiovascular diseases, and chronic respiratory diseasesterm, because of its overwhelming importance in determining trends for these causesterm·var. Tobacco useterm was measured in terms of “smoking impactterm”—that component of observed lung cancer mortalityterm attributable to tobacco smoking [ 19 ]. This indirect measure of the accumulated hazards provides a better measure than do current smoking rates for the overall health impact of tobacco, taking into account lag timesterm·var as well as important aspects of exposure such as duration, type, amount, and mode of smoking (see Protocol S1 for more details).

We used a regression equationterm of the same form as in the original GBD projections: where C a,k,i is a constant term, M a,k,i is the mortalityterm level for age groupterm·var a, sexterm·var k, and causeterm·var i, and Y, HC, and T denote GDP per capitaterm, human capitalterm, and timeterm·var, respectively. The log of the smoking impactterm SI is included in the equation only for malignant neoplasms, cardiovascular diseases, and respiratory diseases. This basic functional relationship makes no specific assumptions about the relationships between these more distal socioeconomic factors and more proximate determinants of mortalityterm rates such as environment, lifestyle, and physiological risk factorsterm. Nevertheless, our regression results indicate that a considerable proportion of the variance in age-, sex-, and cause-specific mortality ratesterm can be explained by this limited set of distal determinants.

Murray and Lopez used a variety of econometric approaches to estimate the equations (Equation 1), including methods that take into account auto-correlation and heteroscedasticity. Because these methods substantially restricted the subset of panel data that could be used, and in particular resulted in much loss of information for moderate- to high-mortality populations, they chose to use ordinary least-squares regressionterm based on the entire dataset for the final set of parameter estimates. We followed the same approach and estimated equations (of the form illustrated by Equation 1) separately for each age–sexterm·var–causeterm·var group.

Death registration dataterm from 106 countries for 1950–2002date [ 20 ] were used to estimate the regression equationsterm. In total, 2,605 country-years were available, almost double those used for the original projections by Murray and Lopez. Although the dataset is extensive, it does not include many observations from populations with high rates of mortalityterm. Whereas the original GBD projections applied a single set of models based on all observed death registration dataterm for projections in all regions, the new projections for low and lower-middle-income countries (income in international dollars less than $3,000 per capita in 2002date) were based on the observed relationships for a dataset consisting of 1,734 country-years of observation where income per capita was less than $10,000 per year. Additionally, observed regional trends in child mortalityterm from 1990 to 2002date were compared with those predicted by the projection modelterm for low-income countriesterm conditional on the observed timeterm·var series for the model covariates for that period. As a result, the regression coefficient for timeterm·var was set to zero for sub-Saharan Africa, and to 25% of its original value for other low-income countriesterm.

Final parsimonious regression results are listed in Tables S3 and S4 . The low-income regression results give somewhat more conservative declines for Group IIterm·def (noncommunicable) diseases in low-income regions. In general, the final regression equationsterm are broadly similar to those estimated for the original projections. Apart from injuriesterm, they explain a surprising proportion of the variance for many age–sexterm·var–causeterm·var categories. For males and females aged 70 y and above, the R 2 for many causeterm·var clusters was generally lower than for other age groupsterm, probably reflecting poorer quality of coding of causesterm·var of death or the smaller range of variation in mortalityterm rates between countries at these older ages, as well as the inherent reductions in the explanatory power of covariates with increasing death rates with age. The lower proportions of variance explained for injuriesterm may reflect lower levels of temporal variation in these causesterm·var as well as more heterogeneous injury causesterm·var that may be less well-correlated with income and human capitalterm.

For causeterm·var clusters in age–sexterm·var groups in which the R 2 was less than 10%, we carefully examined the consistency of the projections with other age–sexterm·var groups, and made a choice between using the final parsimonious regression equationterm, an alternative assumption such as stable (constant) rates, or a separate prediction model based on projections for the major relevant risk factorterm (diabetes mellitusterm and chronic respiratory diseasesterm).

Model predictions of age-, sex-, and cause-specific mortality ratesterm in 2002date for each of the ten clusters of causesterm·var were compared for each country with the results of the GBD study for that year. A series of scalars were then derived so that projected values for 2002date were identical to the 2002date GBD results. It was then assumed that these scalars would remain constant over the period 2002date–2030date.

Diabetes and Chronic Respiratory Conditions Initial regression analysis for diabetes mellitusterm found inconsistent trends between males and females, probably reflecting the large variations across countries and inaccuracies in recording diabetes as the underlying causeterm·var of death in many death registration systems. While overall trends for diabetes mortalityterm showed no consistent relationship across the sexesterm·var or levels of development, the regression analysis found significant betas for T (year) with death rates increasing with timeterm·var. Since a substantial proportion of diabetes mortalityterm is attributable to overweight and obesity [ 21 ], a separate projection modelterm for diabetes mortalityterm was developed using WHO projection of trends in body mass indexterm distributions from 2000 to 2010 (see Protocol S1 for more details).

Initial projections for chronic respiratory diseasesterm resulted in substantially increasing rates for high-income countries, although smoking is the main risk factorterm and smoking impactterm has been generally decreasing. It is likely that the initial projections may reflect increasing propensity to code chronic obstructive pulmonary disease (COPD) as the underlying causeterm·var of death with timeterm·var. Projections of mortalityterm for chronic respiratory diseasesterm were thus adjusted for projected changes in smoking impactterm (see Protocol S1 for more details).

Projections of Income, Human Capitalterm, and Smoking Impactterm Revised country-level projections for income per capita, human capitalterm, and smoking impactterm were developed for the baseline, pessimistic, and optimistic projections. Income per capita was measured using average GDP per capitaterm expressed in international dollars (purchasing power parity adjusted). Country-specific and regional income growth forecasts by the World Bank were used to project GDP per capitaterm for all WHO member states. Beyond 2015, projected growth rates for most regions approach 3% per annum under the baseline scenarioterm, with somewhat lower growth rates in sub-Saharan Africa, the Middle East, and high-income countries (see Protocol S1 for more details).

Revised estimates and projections of human capitalterm for WHO member states were prepared for the period 1950–2030date drawing on previous estimates by Barro and Lee for 98 countries [ 22 ] and observed relationships between growth in human capitalterm and growth in GDP per capitaterm.

Smoking impactterm was calculated for the historical mortalityterm country–year observations by subtracting nonsmoker lung cancer rates from observed total lung cancer mortalityterm rates in the data. Higher nonsmoker lung cancer mortalityterm rates were used in China and Southeast Asian countries, to reflect the higher levels of lung cancer in nonsmokers due to indoor air pollution from exposure to smoke from solid fuels [ 23 , 24 ].

Ezzati and Lopez developed projections of smoking intensity for 14 sub-regions of the WHO regions for the Comparative Risk Assessment (CRA) project, based on an analysis of past and current age–sex-specific smoking prevalence calibrated to regional characteristics of the tobacco epidemic for different sub-regions [ 23 , 24 ]. These regional projections were used to develop country-specific projections drawing also on more recent World Bank data on trends in adult per capita consumption of cigarettes and recent observed trends in age–sex-specific lung cancer mortalityterm (see Protocol S1 for more details). For four high-income countries, recent projections were used based on an age–period–cohort model estimated using historical data on age–sex-specific tobacco consumption, average tar content, and adult tobacco consumption per capita [ 25 ].

Regression Equationsterm for Detailed Causesterm·var To project death rates for specific causesterm·var within the major-cause clusters, we followed the original GBD methodology in estimating the relationship between trends in age–sex-specific mortalityterm for each specific causeterm·var with the trend in the age–sex-specific mortalityterm for the major-cause cluster to which the specific causeterm·var belongs. To avoid biasing the results due to changes in specific causeterm·var death rates associated with use of different versions of ICD, we restricted the analysis to country-years for which deaths were coded using ICD-9. The regression parameters were estimated on 1,357 country-year observations for 98 countries.

For each individual regression, observations were excluded from analysis if they related to fewer than 50 deaths. Regression results were used only where the relationship was reasonably strong, as measured by an R 2 greater than 0.25 and a p -value less than 0.001 for the regression coefficient. The resulting coefficients are shown in Table S5 .

The major-cause cluster regression equationterm for intentional injury predicted generally falling death rates in middle- and low-income countriesterm and rising death rates in high-income countries. For homicide, this is the opposite of the observed trends in the larger high-income countries. For this reason, homicide death rates were assumed to remain constant under the baseline scenarioterm, with some modifications for country-specific observed trends. Similarly, war deaths were assumed to remain constant under the baseline scenarioterm.

For countries with good death registration dataterm and populations of 5 million or more, trends in mortalityterm rates were estimated for ischaemic heart disease, cerebrovascular disease, tuberculosis, suicide, and homicide. These were used to adjust the initial years of relevant country-specific projections to match the recent observed trends.

Projections for HIV/AIDS and Tuberculosis We used separate projections for HIV/AIDS mortalityterm under several scenarios derived from existing models. The Joint United Nations Programme on HIV/AIDS (UNAIDS) and WHO [ 26 ] have prepared projections of HIV/AIDS mortalityterm under a range of assumptions about the future of the HIV epidemics in all regions and with varying treatment scale-upterm assumptions for both adult and children using a transmission model previously used to assess the impact of preventive interventions [ 10 , 27 ].

The model includes underlying regional demography, acquisition of HIV and other sexually transmitted infections, and progression from HIV infection to AIDS and death. For countries with generalized epidemics (mostly in sub-Saharan Africa), the model variables were estimated from sentinel site prevalence data. For countries with epidemics concentrated in groups with higher-risk behaviour, the size and HIV prevalence was estimated for each of these groups, and prevalence in low-risk populations was estimated by allowing for transmission from high-risk to low-risk groups via sexual mixing. Projections of these epidemics were based on assumptions about degree of saturation for each of the high-risk groups, timeterm·var to saturation, and spread from high-risk to low-risk populations over timeterm·var (see Protocol S1 for further information).

Our baseline projections for HIV/AIDS age–sex-specific mortalityterm rates were based on the UNAIDS and WHO “medium” scenario for treatment scale-upterm. Under this scenario, antiretroviral therapyterm (ART) coverage will reach 80% by 2012 in all regions, remaining constant beyond that year. The treatment scenarios assumed no effect of treatment on transmission and incidence ratesterm, and no additional prevention effortsterm resulting in reduction of transmission and incidence ratesterm.

Our pessimistic projections for HIV/AIDS age–sex-specific mortalityterm rates were based on the UNAIDS and WHO “slow” scenario for treatment scale-upterm. Under this scenario, ART coverageterm·var will reach 60% by 2012 in all regions except Latin America, where it reaches 70% in 2013. HIV/AIDS mortalityterm rates in high-income countries were assumed to remain constant as in the baseline scenarioterm.

Salomon et al. [ 28 ] have modelled scenarios for sub-Saharan Africa combining treatment and additional prevention effortsterm. In one scenario, ART coverageterm·var is scaled up and optimal assumptions are made about treatment impact on transmissibility and patient behaviour. A second scenario assumes that an emphasis on treatment leads to less effective implementation of prevention, resulting in only 25% attainment of the maximum potential impact of prevention effortsterm. For our optimistic projection of HIV/AIDS mortalityterm, we used a third unpublished scenario prepared by Salomon et al. that is approximately halfway between their two published mixed treatment-prevention scenarios. For low- and middle-income countries outside sub-Saharan Africa, we assumed similar proportional reductions in incidence ratesterm for HIV infection due to increased prevention, as estimated for sub-Saharan Africa (see Protocol S1 for more details).

Because of the powerful interaction between tuberculosis and HIV infections in regions such as sub-Saharan Africa, Murray and Lopez modified the original projections from 1990 to 2020 for tuberculosis death ratesterm. As part of the Global Plan to Stop TB covering the period 2006–2015, the Stop TB Partnership has prepared three projection scenarios for tuberculosis incidence, prevalence, and mortalityterm based on different assumptions about the pace of scale-up and coverage of interventions to achieve the UN's Millennium Development Goals for tuberculosis [ 29 ]. The “Global Plan” scenario assumes massive scale-up in tuberculosis control activities, achieving case detectionterm levels of more than 70% and using the WHO DOTS (directly observed therapy, short-course) treatment strategy to reach cure ratesterm of more than 85%. The “Sustained DOTS” scenario assumes that case detectionterm and treatment success rates increase until 2005 and then remain constant to 2015. A third “No DOTS” scenario assumes that the DOTS treatment strategy was never introduced in any region, so case detectionterm, treatment, and cure ratesterm would continue as they were pre-DOTS (see Protocol S1 for more details).

The projected annual regional trends in tuberculosis death ratesterm for HIV-negative cases under these three scenarios were used to project tuberculosis death ratesterm for countries in each region as follows. Annual projected trends under the “Sustained DOTS” scenario were used for the baseline projections, and those under the “Global Plan” scenario were used for the optimistic projections. For the pessimistic projections, annual trends in rates were estimated as halfway between those projected under the “sustained DOTS” scenario and those projected under the “No DOTS” scenario from 2006 onwards. For 2016–2030date, annual trends from 2015 to 2030date were assumed to converge on the regional projected trends for Group Iterm·def causesterm·var excluding HIV/AIDS under the baseline scenarioterm. For the optimistic and pessimistic scenariosterm, annual regional trends were assumed to remain constant from 2015 to 2030date.

Projections of Tobacco-Caused Deaths As part of the WHO's CRA project, Ezzati and Lopez [ 23 ] estimated the total mortalityterm attributable to tobacco smoking in 2000 for the world and for WHO sub-regions. Current levels of smoking impact ratioterm ( SIR ) were used as an indirect indicator of accumulated smoking risk based on excess lung cancer mortalityterm. SIR measures the absolute excess lung cancer mortalityterm due to smoking in the study population, relative to the absolute excess lung cancer mortalityterm in lifelong smokers of the reference population. SIR values were calculated from the projected smoking impactterm variables for individual countries for age groupsterm 30–44 y and older.

Projected mortalityterm attributable to tobacco was estimated using the CRA methods for lung cancer, upper aerodigestive cancer, all other cancers, COPD, other respiratory diseases, cardiovascular diseases, tuberculosis, and selected other disease causesterm·var.

Projecting the Burden of Diseaseterm and Injury The WHO has undertaken new assessments of the GBD for 2000–2002date, with consecutive revisions and updates published annually in WHO's World Health Reports. These assessments use a health gap measure, the DALYterm·def, developed by Murray and Lopez [ 30 ] to quantify the equivalent years of full health lost due to diseases and injury in WHO member states. The data sources and methods used for the GBD 2002date are documented elsewhere [ 15 – 17 ], and summary results for 14 regions of the world are published in the World Health Report 2004 [ 14 ] and available at http://www.who.int/evidence/bod .

The burden of diseaseterm estimates for 2002date were used as a base for projections of burden of diseaseterm to 2030date. Years of life lostterm (YLL) were calculated from the projections of mortalityterm by causeterm·var, age, and sexterm·var, according to the GBD method. To project DALYsterm·def, it is also necessary to project years lived with disabilityterm (YLD). Given the lack of good information on trends in disability and health state distributions, the approach used here is an elaboration of the methods and assumptions used by Murray and Lopez in the original GBD projections [ 1 ].

YLD projections were generally derived from the YLL projections by applying the ratio of YLD to YLL for 2002date. For ischaemic heart disease and stroke, future incidence ratesterm were assumed to decline at 50% of their mortalityterm rate declines reflecting declining case fatality rates as well as incidence ratesterm. For causesterm·var where there is little or no mortalityterm, age–sex-specific YLD rates per capita were generally assumed to remain constant into the future. For certain mental disorders, musculoskeletal conditions, and hearing loss, disability weights were assumed to decline somewhat with improvements in income per capita reflecting increasing treatment coverage. YLD rates for nonfatal communicable diseases and nutritional deficiencies were assumed to decline at between 50% and 100% of the mortalityterm rate declines for Group Iterm·def causesterm·var.

Projections of Population and Numbers of Deaths and DALYsterm·def Our projections of mortalityterm rates, together with UN medium variant assumptions for fertility rateterm·var projections and projected migration ratesterm·var [ 31 ], were also used to prepare consistent population projections for all regions. The projected global population in 2015 was 7.1 billion compared to the UN medium variant projection of 7.2 billion, reflecting somewhat higher adult death rates in our mortalityterm projections.

Projected death and DALYterm·def rates for 192 WHO member states for 2003–2030date were applied to the projected populations to generate projected numbers of deaths and DALYsterm·def for each of the three scenarios. The resulting country projections were added back into regional groups for presentation of results (see Table S1 ).The choices and assumptions incorporated in each of the baseline, pessimistic, and optimistic scenariosterm are summarized in Table S7 .

Results This section provides an overview of the projections. Detailed results for 2002date, 2015, and 2030date are presented in Dataset S1 for deaths and Dataset S2 for DALYsterm·def. These tables include baseline, pessimistic, and optimistic projected totals for each causeterm·var category, as well as a sexterm·var and age breakdown for the baseline category.

Figure 1 shows projected life expectancies at birth in 2030date under the three scenarios, for World Bank regions. Life expectancy at birth is projected to increase in all World Bank regions, with the largest increases in the African region and the South Asian regions. However, under the baseline scenarioterm, male life expectancy in the African region will still remain less than 55 y. In all regions except the European region, life expectancy increases are greater for females than for males. Life expectancy for women in the high-income countries may reach 85.0 y by 2030date, compared with 79.7 y for men. The highest projected life expectancy in 2030date is for Japanese women at 88.5 y (with a range of 87.7 to 89.2 across the pessimistic and optimistic scenariosterm). The female–male difference in life expectancy at birth is projected to narrrow from 5.9 y in 2002date to 5.3 y in 2030date in the high-income countries, whereas the gap will more than double in low-income countriesterm to 5.2 y in 2030date.

Projected Life Expectancy at Birth in 2030date by World Bank Region and Sexterm·var: Baseline, Optimistic, and Pessimistic Scenariosterm Compared with 2002date Estimates Projected global deaths in 2030date ranged from 64.9 million under the optimistic scenarioterm to 80.7 million under the pessimistic scenarioterm, variations of −11% to +10% relative to the baseline projection of 73.2 million. Figure 2 shows the projected global numbers of deaths in 2030date by age in the three scenarios, compared with the numbers of deaths by age in 2002date. In all three scenarios there is a dramatic shift in the distribution of deaths from younger to older ages. The risk of death for children younger than 5 y is projected to fall substantially in the baseline scenarioterm, by almost 25% between 2005 and 2015, and by more than 40% between 2005 and 2030date. These rates of decline are similar to those projected between 2000 and 2015 in the original GBD projections. The vertical bars attached to the points for 2030date in Figure 2 represent the range of deaths projected under the optimistic and pessimistic scenariosterm.

Global Numbers of Deaths by Age and Sexterm·var: Baseline, Optimistic, and Pessimistic Scenariosterm for 2030date Compared with 2002date Estimates Trends in Cause-Specific Mortalityterm Table 1 summarizes the projected annual average changes in age-standardized death rates for selected major causesterm·var for the baseline projections for the period 2002date–2020. For all the Group Iterm·def and Group IIterm·def causeterm·var groups in which the projections were based on the major-cause cluster regression equationsterm, age-specific and age-standardized death rates are projected to decline over the next 20 years. The average annual rate of decline is greater (at about 3%) for Group Iterm·def causesterm·var than for Group IIterm·def causesterm·var. The HIV/AIDS projections, discussed in more detail below, have a substantial projected average annual rate of increase of 3% for males and 2% for females. Other causesterm·var with projected increases in age-standardized rates include lung cancer, diabetes, chronic respiratory diseasesterm, road traffic accidents, violence, and war.

Projected Average Annual Rates of Change in Age-Standardized Death Rates for Selected Causesterm·var: World, 2002date–2020 The Group Iterm·def causesterm·var, excluding HIV/AIDS and tuberculosis, decline with average annual rates of change typically about one-third slower than the original GBD projections of Murray and Lopez. To some extent, this reflects the more conservative projections for low-income countriesterm, where the coefficient of the timeterm·var factor was reduced or set to zero. Average rates of decline for Group IIterm·def causesterm·var (excluding lung cancer and chronic respiratory conditions) are similar for females to those in the original projections. However, the differential between males and females in the original projections has disappeared in the current projections, with males having a greater average annual rate of decline for Group IIterm·def conditions than previously projected.

Figure 3 summarizes the contributions of major causesterm·var to global trends in numbers of deaths for the three major causeterm·var groups. Large declines in mortalityterm between 2002date and 2030date are projected for all of the principal Group Iterm·def causesterm·var, with the exception of HIV/AIDS. Under the baseline scenarioterm involving scale-up of ART coverageterm·var to 80% by 2012, but not additional prevention effortsterm, HIV/AIDS deaths increase from 2.8 million in 2002date to 6.5 million in 2030date. Total deaths due to other Group Iterm·def causesterm·var decline from 15.5 million in 2002date to 9.0 million in 2030date. Unfortunately, this is substantially offset by the projected rise in HIV/AIDS deaths. Under the optimistic scenarioterm involving additional HIV prevention activity, 3.7 million HIV/AIDS deaths are projected for 2030date, so that total deaths due to Group Iterm·def conditions would decline from 32% of all deaths in 2002date to 14% of all deaths in 2030date.

Baseline Projections of Deaths from Group Iterm·def, Group IIterm·def, and Group IIIterm·def Causesterm·var, World, 2002date–2030date Although age-specific death rates for most Group IIterm·def conditions are projected to decline, ageing of the population will result in significantly increasing total deaths due to most Group IIterm·def conditions over the next 30 years ( Figure 3 ). Global cancer deaths are projected to increase from 7.1 million in 2002date to 11.5 million in 2030date, and global cardiovascular deaths from 16.7 million in 2002date to 23.3 million in 2030date. Overall, Group IIterm·def conditions will account for almost 70% of all deaths in 2030date under the baseline scenarioterm.

The projected 40% increase in global deaths due to injury between 2002date and 2030date are predominantly due to the increasing numbers of road traffic accident deaths, together with increases in population numbers more than offsetting small declines in age-specific death rates for other causesterm·var of injury. Road traffic accident deaths are projected to increase from 1.2 million in 2002date to 2.1 million in 2030date, primarily due to increased motor vehicle fatalities associated with economic growth in low- and middle-income countries.

Projections of HIV/AIDS Mortalityterm and Other Selected Causesterm·var Figure 4 summarizes projected HIV/AIDS deaths by income groupterm·var for the three scenarios. The declining death rates for 2005–2010 in the baseline and pessimistic scenariosterm, followed by increasing death rates, reflect the effects of the assumed treatment scale-upterm scenarios. Rapidly increasing levels of ART coverageterm·var result in postponement of deaths for a number of years, but once the ART coverageterm·var plateaus at its final level, numbers of deaths continue to rise, reflecting largely the underlying growth in population. HIV incidence ratesterm essentially remain constant in the baseline scenarioterm for sub-Saharan Africa, and the global growth in incident cases and in deaths is largely driven by population growthterm in sub-Saharan Africa.

Projections of Total AIDS Deaths (Thousands) by Income Groupterm·var for Three Scenarios Scenarios are indicated: baseline (solid lines), optimistic (dotted lines), and pessimistic (dashed lines).

Under the baseline scenarioterm, the total deaths from HIV/AIDS over the 25-y period 2006–2030date are projected to be 117 million. Under the optimistic scenarioterm, in which additional prevention effortsterm result in a long-term 3% decline in incidence ratesterm, the projected total deaths over 2006–2030date are 89 million, a saving of 28 million lives.

The ranges defined by the optimistic and pessimistic projections differ substantially by causeterm·var ( Figure 5 ). For example, the range for cardiovascular disease before age 70 y is much wider than that for cancers before age 70 y or for road traffic accidents. Group Iterm·def deaths (excluding HIV/AIDS) have a wider range than most other causesterm·var, although the total deaths decline substantially in all three scenarios.

Projections of Global Deaths (Millions) for Selected Causesterm·var, for Three Scenarios: Baseline, Optimistic, and Pessimistic, 2002date–2030date Comparison with GBD 1990 Projections Figure 6 compares projected global deaths for each of the three major causeterm·var groups for 2002date–2030date with the corresponding projections for 1990–2020 from the original GBD study [ 1 ]. Projections for Group Iterm·def causesterm·var are substantially different, mainly because of the very large difference in projected HIV/AIDS mortalityterm. Total global deaths for Group IIterm·def and Group IIIterm·def causesterm·var are projected to increase at a somewhat slower rate than the original GBD projections, and in addition, the base levels for these causesterm·var in 2002date are somewhat lower than the levels projected by the original GBD study.

Comparison of Baseline Projections 2002date–2030date with the Original GBD Projections 1990–2020: Global Deaths for All Causesterm·var, and Major Causeterm·var Groups Despite these differences for all three major causeterm·var groups, and differences in projected global population numbers, the projections of total global all-cause deaths are almost identical to those in the original GBD study. Global deaths in 2020 under the baseline scenarioterm are 66.5 million, compared with 68.3 million projected by Murray and Lopez from the 1990 base, and the overall trend is almost identical.

Projected trends for global deaths due to most Group Iterm·def causesterm·var other than HIV/AIDS are broadly similar. The new baseline projections for lung cancer give lower global numbers and a lower rate of increase. Trends in global respiratory deaths are similar for the two sets of projections, but the projected rates of increase in global deaths due to cancers, cardiovascular disease, and digestive disorders are all somewhat slower for the current projections. Increases for road traffic accidents are somewhat slower in the current projections, whereas trends for other unintentional causesterm·var are quite similar. There are some substantial differences in trends for intentional causesterm·var, reflecting different assumptions and decisions concerning use of the regression parameters.

Leading Causesterm·var of Death Table 2 lists the 15 leading causesterm·var of death according to the baseline scenarioterm for males, females, and both sexesterm·var combined as projected for 2030date globally. Table 3 provides similar lists of the ten leading causesterm·var of death according to the baseline scenarioterm for the four income groupsterm·var of countries. The four leading causesterm·var of death in all scenarios are projected to be ischaemic heart disease, cerebrovascular disease (stroke), HIV/AIDS, and COPD, although HIV/AIDS moves from third to fourth position in the optimistic scenarioterm.

Changes in Rankings for 15 Leading Causesterm·var of Death, 2002date and 2030date (Baseline Scenarioterm) Ten Leading Causesterm·var of Death, by Income Groupterm·var, 2030date (Baseline Scenarioterm) Table 2 also summarises the changes in rank order of deaths between 2002date and 2030date for the 15 leading causesterm·var. Lower-respiratory infections, perinatal conditions, diarrhoeal diseases, malaria, and measles are all projected to decline substantially in importance. On the other hand, diabetes mellitusterm, lung cancer, stomach cancer, and liver and colorectal cancer are all projected to move up three or more places in the rankings.

Decomposition The results discussed so far have described projected changes in mortalityterm in terms of the absolute (and relative) numbers of deaths expected under the various scenarios. These changes may be due to changes in age-specific disease and injury mortalityterm risks, or due to demographic change that alters the size and age composition of the population, or both. Because mortalityterm risks are strongly age dependent for most causesterm·var, changes in the age structure of a population may result in substantial changes in the number of deaths, even when the age-specific risks remain unchanged.

We further analysed the relative impact of demographic and epidemiological changeterm on the projected numbers of deaths by causeterm·var by calculating three hypothetical alternatives. In the first, we calculated the expected number of deaths in 2030date given the 2030date projected age-specific rates under the baseline scenarioterm and the 2002date population. The difference between this and the 2002date mortalityterm estimates is a measure of the change in mortalityterm expected solely on the basis of changing age-specific mortalityterm rates, and is labelled “epidemiological changeterm” in Figure 7 .

Decomposition of Projected Change in Numbers of Deaths into Demographic and Epidemiological Components, by Broad Causeterm·var Group and Income Groupterm·var, 2002date–2030date Second, we calculated the expected number of deaths in 2030date by taking the 2002date age-specific death rates and applying them to the 2030date projected population. The difference between this and the 2002date mortalityterm estimates is a measure of the change in mortalityterm expected solely on the basis of changing demography (including size and age composition of the population). We then repeated this calculation, but applied the 2002date age-specific death rates to projected population numbers for 2030date that matched the total projected male and female population numbers for each country but retained the 2002date age distribution of the population. The difference between this and the 2002date mortalityterm estimates is a measure of the change in mortalityterm expected solely on the basis of population growthterm excluding changes in age composition. The difference between the total change in mortalityterm due to demography and this latter estimate gives a measure of the effect of the change in age composition of the population alone. These two components of demographic change are labelled “population growthterm” and “population ageingterm” in Figure 7 . The total projected change in numbers of deaths between 2002date and 2030date is the sum of the population growthterm, population ageingterm, and epidemiological changeterm components.

In almost all cases, demographic and epidemiological factors are operating in opposing directions in determining mortalityterm in 2030date. The major exception is HIV/AIDS, where demographic and epidemiological changeterm are acting in the same direction to increase total HIV/AIDS deaths to 6.5 million deaths in 2030date under the baseline scenarioterm. Demographic change dominates, as the majority of HIV/AIDS deaths are in sub-Saharan Africa, where population growthterm is highest and where HIV/AIDS incidence ratesterm are assumed to remain largely constant under the baseline scenarioterm.

For Group Iterm·def conditions other than HIV/AIDS for which substantial declines in mortalityterm rates are projected, the effect of these declines will be attenuated in most regions by demographic change leading to an increase in the child population most at risk for these conditions. Population growthterm and population ageingterm act in opposite directions for Group Iterm·def mortalityterm excluding HIV/AIDS in low-income countriesterm, but not in other income groupsterm·var. If future fertility ratesterm·var are higher than projected, then the higher child population numbers will further offset the projected reductions in death rates for Group Iterm·def conditions.

For Group IIterm·def (noncommunicable diseasesterm), demographic changes in all regions will tend to increase deaths substantially, with offsetting reductions in projected death rates in all regions. Population growthterm and population ageingterm both act to increase Group IIterm·def deaths in all regions, although the impact of population ageingterm is generally much more important than population growthterm. Population growthterm has the largest relative impact for low-income countriesterm, and the smallest for lower-middle-income countries. The latter group includes Eastern European populations such as Russia that will experience negative population growthterm.

For Group IIIterm·def (injuriesterm), demographic change similarly dominates the epidemiological changeterm. The latter is small at group level in most regions, because the projected increase in road traffic fatalities is offset by projected decreases in death rates for other unintentional injuriesterm.

Tobacco-Attributable Deathsterm Figure 8 summarises the projected number of tobacco-attributable deathsterm for the world, and for high-income and low- and middle-income countries. Under the baseline scenarioterm, total tobacco-attributable deathsterm will rise from 5.4 million in 2005 to 6.4 million in 2015 and 8.3 million in 2030date. Projected deaths for 2030date range from 7.4 million in the optimistic scenarioterm to 9.7 million in the pessimistic scenarioterm. Tobacco-attributable deathsterm are projected to decline by 9% between 2002date and 2030date in high-income countries, but to double from 3.4 million to 6.8 million in low- and middle-income countries.

Projected Numbers of Tobacco-Caused Deaths for the World and for High-Income and Middle- plus Low-Income Countriesterm, Three Scenarios, 2002date–2030date Table 4 divides the projected 2015 global mortalityterm due to smoking into leading causesterm·var: cancers are responsible for one-third of the deaths, followed by cardiovascular diseases and chronic respiratory diseasesterm, each responsible for 30% of the deaths. According to our baseline projection, smoking will kill 50% more people in 2015 than HIV/AIDS, and will be responsible for 10% of all deaths globally.

Projected Global Tobacco-Caused Deaths, by Causeterm·var, 2015 Baseline Scenarioterm Burden of Diseaseterm Global DALYsterm·def lost are projected to increase from 1.48 billion in 2002date to 1.54 billion in 2030date, an overall increase of only 3%. Since the population increase is projected to be 27% during the same period, there is actually a decrease in the global per capita burden. Unlike deaths, where the overall global death rate is projected to increase by 1%, the DALYterm·def rate decreases because the increasing number of deaths is offset by the shift in age at death to older ages, associated with fewer lost years of life. Even with the assumption that the age-specific burden for most nonfatal causesterm·var remains constant into the future, and hence that the overall burden for these conditions increases with the ageing of the population, there is still an overall projected decrease in the global burden of diseaseterm per capita of 19% from 2002date to 2030date.

The proportional contribution of the three major causeterm·var groups to the total disease burden is projected to change substantially, however. Group Iterm·def causesterm·var are projected to account for 30% of total DALYsterm·def lost in 2030date, compared with more than 40% in 2002date. In low-income countriesterm, the decline is even greater, from 56% in 2002date to 41% in 2030date, even including the doubling of the HIV/AIDS burden. The noncommunicable diseaseterm (Group IIterm·def) burden is projected to increase to 57% in 2030date, and to represent a greater burden of diseaseterm than Group Iterm·def conditions in all income groupsterm·var, including low-income countriesterm.

Table 5 lists the 15 leading causesterm·var of DALYsterm·def globally in 2002date and in 2030date according to the baseline scenarioterm, and Table 6 provides similar lists of the ten leading causesterm·var of burden of diseaseterm for high-, middle-, and low-income country groups. The three leading causesterm·var of DALYsterm·def in the baseline and pessimistic scenariosterm are projected to be HIV/AIDS, unipolar depressive disorders, and ischaemic heart disease. Road traffic accidents become the third leading causeterm·var under the optimistic scenarioterm, ahead of ischaemic heart disease and cerebrovascular disease. Perinatal conditions are the fourth leading causeterm·var under the pessimistic scenarioterm and the fifth leading causeterm·var under the baseline scenarioterm, after road traffic accidents. HIV/AIDS becomes the leading causeterm·var of burden of diseaseterm in middle-income countries, as well as low-income countriesterm, by 2015.

Changes in Rankings for 15 Leading Causesterm·var of DALYsterm·def, 2002date and 2030date (Baseline Scenarioterm) Ten Leading Causesterm·var of DALYsterm·def, by Income Groupterm·var and Sexterm·var, 2030date (Baseline Scenarioterm) Table 5 also illustrates the changes in rank order of DALYsterm·def between 2002date and 2030date for the 15 leading causesterm·var globally. Lower-respiratory infections, perinatal conditions, diarrhoeal diseases, malaria, measles, tuberculosis, and congenital anomalies are all projected to decline substantially in importance. On the other hand, ischaemic heart disease, diabetes mellitusterm, road traffic accidents, self-inflicted injuriesterm, COPD, hearing loss, and cataracts are all projected to move up three or more places in the rankings. Hearing loss is projected to be among the top ten causesterm·var of burden of diseaseterm in high- and middle-income countries, and Alzheimer disease and other dementias and alcohol-use disorders among the top four causesterm·var in high-income countries in 2030date. In low-income countriesterm in 2030date, Group Iterm·def conditions continue to account for five of the ten leading causesterm·var of burden of diseaseterm. These are HIV/AIDS, perinatal conditions, diarrhoeal diseases, malaria, and acute lower-respiratory infections.

Discussion We have addressed the need for updated projections of mortalityterm and burden of diseaseterm using methods similar to those of the original GBD study. Our intent was not to undertake major new methodological developments, but rather to provide updated projections for the first 30 years of the 21st century, using as much relevant new information as is available. We project that life expectancy will increase around the world for all three scenarios, fewer children younger than 5 y will die (a 50% decline under the baseline scenarioterm), and the proportion of people dying from non-communicable diseases will increase (from 59% in 2002date to 69% in 2030date under the baseline scenarioterm). Although deaths from infectious diseases will decrease overall, HIV/AIDS deaths will continue to increase; the exact magnitude of the increase will depend to a limited extent on how many people have access to antiretroviral drugs and much more on whether there are increased prevention effortsterm. Although a projected 6.5 million people will die from HIV/AIDS under the 2030date baseline projection, an even larger number will die from disease attributable to tobacco smoking (8.3 million). By 2030date, the three leading causesterm·var of burden of diseaseterm will be HIV/AIDS, depression, and ischaemic heart disease in the baseline and pessimistic scenariosterm. Road traffic accidents are the fourth leading causeterm·var in the baseline scenarioterm, and the third leading causeterm·var ahead of ischaemic heart disease in the optimistic scenarioterm.

These updated projections of mortalityterm and burden of diseaseterm have been prepared using a similar methodology to that of the original GBD study, but with some changes described above and with updated inputs and an updated base set of estimates for 2002date. We have incorporated a number of methodological improvements and changes. These include carrying out the projections at country rather than regional level, use of separate regression equationsterm for low-income countriesterm, incorporation of information from death registration datasets on recent observed trends for selected causesterm·var, and calibration of the regression equationsterm through a comparison of back-projections with observed child mortalityterm trends from 1990 to 2002date. Additionally, separate risk factorterm–based projection modelsterm were developed for diabetes mellitusterm and chronic respiratory diseasesterm. Population projections also included UN projections for net migration ratesterm·var.

Baseline estimates of deaths, disability, and DALYsterm·def lost in 2002date have been projected into the future based on an explicit set of assumptions and methods, and the results are nothing more than the numerical consequences of the assumptions and methods employed. In this paper and the accompanying Protocol S1 and Tables S1 – S7 , we have tried to explicitly describe and summarize all inputs and assumptions for the three scenarios to enable replication of these projections for different scenarios. Additional information and detailed datasets are available from the authors on request.

We do not claim that the scenarios presented here necessarily represent the best predictions of future global and regional health trends, and it is possible that more sophisticated causal models incorporating projections of important determinants may provide better predictions for specific diseases. However, we do believe that these projections, as with the previous GBD projections of Murray and Lopez, represent the most comprehensive and consistent set of health status projections available today.

Comparisons with Other Projections We are not aware of any projections of global and regional mortalityterm, apart from the previous GBD projections, that have included a comprehensive set of causesterm·var of death. A review of other published studies containing global and regional projections for specific single causesterm·var identified relatively few studies that did not either apply base estimates of age–sex-specific death rates (assumed not to change in the future) to UN population projections, or were based on the previous GBD projections, or were earlier versions of the projections by WHO and/or UNAIDS for HIV/AIDS and tuberculosis used here.

Our projected global population in 2015 under the baseline scenarioterm was 7.1 billion, compared with the UN medium variant projection of 7.2 billion [ 32 ], a difference of only −1.6% (see Protocol S1 for details of projection methods). By 2030date, the difference between WHO and UN baseline projections of population grew to −3.5%, and the range from pessimistic to optimistic scenariosterm to 7.75 billion to 8.07 billion. The largest difference in the population growthterm projections is for the European region, where we project the total population to decline at a faster rate than the UN projections.

Projected global deaths in 2030date ranged from 64.9 million under the optimistic scenarioterm to 80.7 million under the pessimistic scenarioterm, with a baseline projection of 73.2 million. Projected global deaths in 2030date under the UN medium variant projections were 1% higher, at 74.0 million [ 32 ]. Our global projections for all-cause mortalityterm are remarkably close to the UN projections, given that our projections are the sum of independent projections for 13 separate causeterm·var groups, whereas the UN projections are based on estimated trends in all-cause mortalityterm, with adjustments for projected HIV/AIDS mortalityterm.

Our baseline global projection for all-cause mortalityterm in 2020 (66.5 million deaths) is also remarkably similar to the original GBD baseline projection for 2020 of 68.3 million deaths, although the projected numbers of deaths for each of the three major causeterm·var groups differ quite significantly, as do projections for HIV/AIDS and tuberculosis. The congruence in numbers and trends in rates for all-cause mortalityterm is almost certainly a coincidence, since the global total death projections are derived from summing separate projections across 13 causeterm·var groups, and almost all of the inputs to those projections have changed significantly from those used in the original projections.

Under our baseline projection, mortalityterm of children younger than 5 y will decline globally by 29% from 2002date to 2015 (range 44% to 14% under the optimistic and pessimistic scenariosterm). For sub-Saharan Africa the decline is projected to be 25%. In contrast, child mortalityterm declined by only 12% globally and 7% in sub-Saharan Africa between 1990 and 2001 [ 17 ]. The more-than doubling of the decline in child mortalityterm for sub-Saharan Africa projected for the next decade is largely determined by the relatively high projected economic growth for this region according to the World Bank (about 2% annual increase in income per capita). Income per capita growth averaged only 0.5% in sub-Saharan Africa during the period 1990–2002date. Despite this, our back-projections to 1990 predicted a larger decline in child mortalityterm than observed. The major regression coefficients were adjusted for low-income countriesterm, and in particular for sub-Saharan Africa, to match observed trends (see Protocol S1 for details). Even with this adjustment, we are predicting more than double the decline in child mortalityterm in the next decade compared to that observed in the 1990s. This reflects the higher levels of economic growth projected for the next decade in sub-Saharan Africa.

Trends in tuberculosis death ratesterm projected by the Stop TB Partnership for deaths of HIV-negative persons were used here, as the GBD classifies deaths of HIV-positive persons as due to HIV/AIDS. The GBD base estimates of tuberculosis deaths for 2002date are higher than the Stop TB Partnership 2002date estimates for deaths in sub-Saharan Africa and substantially lower in Southeast Asia (see Protocol S1 ). As a result of these differences in base estimates, our baseline projection gives 960,000 tuberculosis deaths in 2015, compared to the 888,000 projected by the Stop TB Partnership under the “Sustained DOTS” scenario. Both these estimates are substantially lower than the approximately 2.2 million tuberculosis deaths projected for 2015 by Murray and Lopez in the original GBD study [ 1 , 2 ].

Bray and Møller have published global projections for cancer incidence in 2020 [ 33 ]. They projected that total cancer incidence would rise from 11 million new cases in 2002date to 16.5 million in 2020, assuming that 2002date age–sex-specific incidence ratesterm remained unchanged. Taking into account projected changes in age–sexterm·var–site-specific incidence ratesterm, we project global incidence for all types of cancer to be 15.5 million cases in 2020, 6% lower than Bray and Møller's estimate. They also estimated global incidence in 2010 of 1.5 million new breast cancer cases and 1 million new prostate cancer cases, taking into account current trends in incidence ratesterm. These are higher than our projections of 1.3 million and 0.85 million, respectively, mainly reflecting differences in the base incidence estimates for 2002date.

Wilde et al. [ 34 ] estimated that the global diabetes prevalence would rise from 171 million in 2000 to 366 million in 2030date, assuming that age–sex-specific diabetes prevalence rates remain unchanged within urban and rural populations in each region, but taking into account projected increases in urbanization to 2030date. Their estimate is somewhat higher than our projection of 328 million diabetes cases in 2030date. Our projections took into account projected trends for overweight and obesity, which may capture some of the impact of urbanization but did not explicitly account for trends in urbanization. Projections based on prevalence trends have been made for diabetes in Canada and the US [ 35 , 36 ]. These studies estimated that diabetes prevalence would reach 2.4 million for Canada in 2016, and 19.9 million for the US in 2025. Our projections of diabetes prevalence for the same years are one-third higher for Canada (3.2 million) and slightly higher for the US (20.9 million).

Projected tobacco-attributable deathsterm in 2030date under the baseline scenarioterm are lower than the 10 million predicted by Peto et al. [ 37 ] to occur “sometime in the 2020s or early 2030sdate.” This earlier prediction is fairly consistent with our projection of 9.7 million tobacco-attributable deathsterm under the pessimistic scenarioterm. Murray and Lopez projected 8.4 million tobacco-attributable deathsterm in 2020, also somewhat higher than our baseline projection of 7.0 million. Our lower projection reflects our downward adjustment of projected lung cancer mortalityterm to reflect recent trends in tobacco consumption. It must be emphasized that there is very substantial uncertainty in the tobacco-attributable mortalityterm projections, as the trends in apparent consumption of tobacco do not necessarily account for changes in duration, type, per capita amount, and mode of smoking.

A recent World Bank report projected deaths due to road traffic accidents from 1990 to 2020 using an economic model that related death rates to changes in fatalities per motor vehicle and motor vehicle ownership per person due to economic growth [ 38 ]. An overall growth of 66% was projected for road traffic fatalities from 2000 to 2020, somewhat higher than our projection of a 51% increase in global deaths due to road traffic accidents from 2002date to 2020 ( Figure 9 ). Projected regional increases were generally similar, although the World Bank projected a substantially greater increase in South Asia than we did, and a small growth for Europe and Central Asia, rather than a decline. In terms of absolute numbers of deaths, the projections are quite different, reflecting large differences in base estimates. The World Bank study estimated global road fatalities in 2000 as 723,000, compared with our estimate for 2002date of 1.2 million. The World Bank base estimates were derived from police statistics, unadjusted for under-reporting. Police statistics are known to generally substantially underestimate deaths due to road traffic accidents. For example, the World Bank estimate of 32,000 deaths for Europe and Central Asia in 2000 is 60% lower than the GBD estimate of 82,000 for 2002date, based on reasonably complete death registration dataterm.

Projected Growth in Road Traffic Fatalities, 2002date–2020: A Comparison of World Bank Projectionsterm with GBD Projections Uncertainties and Limitations There have been substantial improvements in the data and methods available for the assessment of global and regional mortalityterm by causeterm·var. The GBD 2002date estimates of deaths by causeterm·var, age, and sexterm·var were carried out separately for 226 countries and territories, drawing on a total of 770 country-years of death registration dataterm, as well as 535 additional sources of information on levels of child and adult mortalityterm, and in excess of 2,600 datasets providing information on specific causesterm·var of death in regions not well covered by death registration systems [ 15 ].

There remain, however, a large number of methodological and empirical challenges to more reliably estimating global, regional, and national disease burden. For regions with limited death registration dataterm, such as the Eastern Mediterranean region, sub-Saharan Africa, and parts of Asia and the Pacific, there is considerable uncertainty in estimates of deaths by causeterm·var. Uncertainty rangesterm for all-cause deaths in 2001 were estimated by MatherspersName et al. [ 16 ] to increase from around ±1% for high-income countries to around ±20% for sub-Saharan Africa. For most specific causesterm·var, uncertainty rangesterm are greater than those of the all-cause mortalityterm estimates, since there is additional uncertainty associated with causeterm·var attribution. For example, the relative uncertainty rangesterm for ischaemic heart disease in 2001 were estimated to range from about 12% for high-income countries to around ±30% for sub-Saharan Africa. The uncertainty rangeterm for HIV/AIDS deaths in sub-Saharan Africa was narrower at ±15%, reflecting the substantial data base for these estimates.

The projections of burden are not intended as forecasts of what will happen in the future but as projections of current and past trends, based on certain explicit assumptions. The methods used base the disease burden projections largely on broad mortalityterm projections driven by World Bank projectionsterm of future growth in income and WHO projections of increases in human capitalterm in different regions of the world, together with a model relating these to cause-specific mortalityterm trends based on the historical observations in countries with death registration dataterm from the past 50 years. The results depend strongly on the assumption that future mortalityterm trends in developing countries will have a similar relationship to economic and social developmentterm as has occurred in the more developed countries. If this assumption is not correct, then the projections for low-income countriesterm will be over-optimistic in the rate of decline of communicable diseases and the speed of the epidemiological transitionterm.

The predictions of the projections model were compared with historical trends in child mortalityterm from 1990 to 2002date, and as a result, certain regression coefficients were modified for low-income countriesterm. As a result, our revised projections assume that projected rates of change for cause-specific mortality ratesterm over timeterm·var, given levels of constant income and human capitalterm, will be slower than those observed in the mainly high- and middle-income countries with death registration dataterm from the past 50 years. This has reduced the projected rates of decline in Group Iterm·def conditions for low-income countriesterm compared to the original GBD projections, and it is entirely possible that this adjustment may be too conservative. On the other hand, the many problems facing low-income countriesterm in improving and sustaining access to effective health interventions, and in scaling up health systems to cost-effectively address these challenges, may mean that the low-income countriesterm do not experience the temporal pace of health improvement at constant levels of income and human capitalterm that have been seen in the high-income countries in the past 50 years.

The projections have also not taken explicit account of trends in major risk factorsterm apart from tobacco smoking, and, to a limited extent, overweight and obesity. If broad trends in risk factorsterm are for worsening of risk exposures with development, rather than the improvements observed in recent decades in many high-income countries, then again the projections for low- and middle-income countries presented here will be too optimistic. There is a need to develop much more comprehensive projection modelsterm that take explicit account of available information on trends in a wide range of risk factorsterm. The HIV/AIDS baseline projections in particular assume that transmission probabilities will remain largely unchanged in the future and there will not be substantial reductions in risk factorsterm for HIV.

A projections exercise such as this by its nature involves substantial assumptions about the similarity of future trends to past trends, and about the future trends in broad drivers of health improvement. There are thus wide uncertainty rangesterm around future projections. Nevertheless, there are some aspects of the projections that clearly involve more uncertainty than others. For example, the projections of HIV/AIDS mortalityterm are strongly affected by the assumptions made about the levels of additional prevention effortterm that occur over the next two decades. Additionally, there are substantial uncertainties about the future trends in chronic respiratory diseaseterm mortalityterm for non-smokers, and diabetes mortalityterm for persons not overweight. Also, the evidence on the associations of injury mortalityterm with income and human capitalterm was weaker than for Group Iterm·def and Group IIterm·def conditions, and stronger assumptions were thus required for injury projections for some external causesterm·var. In the absence of any realistic approach to forecasting future war deaths, rates for these were assumed to remain constant over timeterm·var in the baseline scenarioterm. This may be too conservative, given the substantial decrease in the number of wars and civil conflicts in the past decade or two.

Finally, as did Murray and Lopez, we recognize that the approach taken to projecting YLD is extremely crude, and that the projections of DALYsterm·def are likely to be even more uncertain than the projections of deaths. It may be the case that case fatality rates for many diseases decline during the next 30 years, so that YLD become an increasing proportion of the total DALYsterm·def for these causesterm·var. On the other hand, improvements in risk factorsterm and/or health interventions may lead to decreases in burden for some nonfatal conditions. Substantial research remains to develop robust and unbiased methods for measuring trends in case fatality rates, survival timesterm·var, and disability due to specific causesterm·var, let alone collecting such data across all regions of the world.

Conclusions Despite these uncertainties, projections provide a useful perspective on population health trends and health policies, provided that they are interpreted with a degree of caution. Projections enable us to appreciate better the implications for health and health policy of currently observed trends, and the likely impact of fairly certain future trends, such as the ageing of the population, and the continuation of the epidemiological transitionterm in developing countries. These projections represent a set of three visions of the future for population health, under an explicit set of assumptions and for specific projections of income, human capitalterm, and of future trends in tobacco smoking, HIV/AIDS transmission and survival, and overweight and obesity. If the future is not like the past—for example, through sustained and additional efforts to address the Millennium Development Goals, or through major scientific breakthroughs—then the world may well achieve faster progress than projected here, even under the optimistic scenarioterm. On the other hand, if economic growth in low-income countriesterm is lower than the forecasts used here, then the world may achieve slower progress and widening of health inequalities.

Supporting Information (37 KB DOC) Click here for additional data file.

(1 MB XLS) Click here for additional data file.

(1 MB XLS) Click here for additional data file.

(200 KB PDF) Click here for additional data file.

(69 KB DOC) Click here for additional data file.

(184 KB DOC) Click here for additional data file.

(334 KB DOC) Click here for additional data file.

(335 KB DOC) Click here for additional data file.

(765 KB DOC) Click here for additional data file.

(1.2 MB DOC) Click here for additional data file.

(67 KB DOC) Click here for additional data file.