THE LANCET

Alternative projections of mortality and disability by cause

1990–2020: Global Burden of Disease Study

Christopher J L Murray, Alan D Lopez

Summary depression, road-traffic accidents, cerebrovascular disease,

chronic obstructive pulmonary disease, lower respiratory
Background Plausible projections of future mortality and infections, tuberculosis, war injuries, diarrhoeal diseases,
disability are a useful aid in decisions on priorities for and HIV. Tobacco-attributable mortality is projected to
health research, capital investment, and training. Rates increase from 3·0 million deaths in 1990 to 8·4 million
and patterns of ill health are determined by factors such as deaths in 2020.
socioeconomic development, educational attainment,
technological developments, and their dispersion among Interpretation Health trends in the next 25 years will be
populations, as well as exposure to hazards such as determined mainly by the ageing of the world’s population,
tobacco. As part of the Global Burden of Disease Study the decline in age-specific mortality rates from
(GBD), we developed three scenarios of future mortality communicable, maternal, perinatal, and nutritional
and disability for different age-sex groups, causes, and disorders, the spread of HIV, and the increase in tobacco-
regions. related mortality and disability. Projections, by their nature,
are highly uncertain, but we found some robust results with
Methods We used the most important disease and injury
implications for health policy.
trends since 1950 in nine cause-of-death clusters.
Regression equations for mortality rates for each cluster by
Lancet 1997; 349: 1498–1504
region were developed from gross domestic product per
person (in international dollars), average number of years
of education, time (in years, as a surrogate for
Introduction
technological change), and smoking intensity, which shows Future health scenarios that are likely, or probable, or
the cumulative effects based on data for 47 countries in
merely possible can have an important role in shaping
public-health policy. Studies on health projections1–3
1950–90. Optimistic, pessimistic, and baseline projections
provide an indication of the strong interest shown by
of the independent variables were made. We related
scientific and public-health communities in the definition
mortality from detailed causes to mortality from a cause
and quantification of scenarios of future health. There
cluster to project more detailed causes. Based on
have, however, been few comprehensive efforts to project
projected numbers of deaths by cause, years of life lived
health scenarios for a population4 and none for the entire
with disability (YLDs) were projected from different relation
world or for major regions. In this paper, the last of four
models of YLDs to years of life lost (YLLs). Population on the Global Burden of Disease Study (GBD) (see
projections were prepared from World Bank projections of Lancet 1997; 349: 1269–76; 1347–52; and 1436–42) we
fertility and the projected mortality rates. describe how we created three scenarios of future
Findings Life expectancy at birth for women was projected mortality and disability by cause, which may have
to increase in all three scenarios; in established market important public policy implications.
economies to about 90 years by 2020. Far smaller gains in Our scenarios were based on future health status as a
male life expectancy were projected than in females; in function of projected changes in key socioeconomic
formerly socialist economies of Europe, male life variables, which influence health states. Uncertainties in
expectancy may not increase at all. Worldwide mortality
the projections arise from the validity of these relational
models, assumptions about their invariance over time,
from communicable maternal, perinatal, and nutritional
and, of course, uncertainty about the future rates and
disorders was expected to decline in the baseline scenario
distribution of factors that currently influence health and
from 17·2 million deaths in 1990 to 10·3 million in 2020.
survival. Despite these uncertainties, several robust
We projected that non-communicable disease mortality will
projections—due largely to demographic change and the
increase from 28·1 million deaths in 1990 to 49·7 million
future effects of current smoking patterns—emerge from
in 2020. Deaths from injury may increase from 5·1 million our analysis. Further detail on the methods used to
to 8·4 million. Leading causes of disability-adjusted life estimate causes of death, to develop epidemiological
years (DALYs) predicted by the baseline model were (in profiles of each disabling sequela, to assess the burden
descending order): ischaemic heart disease, unipolar major attributable to major risk factors, and to project the
burden of premature mortality and disability has been
published.5

Methods
Harvard School of Public Health, Boston, Massachusetts, USA Projection methods
(C J L Murray DPhil), and World Health Organization, Geneva, We used 12 separate analytical or computational steps to
Switzerland (A D Lopez PhD) construct a baseline scenario, and optimistic and pessimistic
Correspondence to: Prof Christopher J L Murray, Harvard School of scenarios. Separate projection models for both sexes and seven
Public Health, 9 Bow Street, Cambridge MA 02138, USA age-groups—0–4, 5–14, 15–29, 30–44, 45–59, 60–69, and 70

1498 Vol 349 • May 24, 1997

 THE LANCET

Women older than 25 (data available for 98 countries since 1950).9 For
90 smoking intensity we used Peto and colleagues’ method10 to
calculate observed lung-cancer rates minus non-smoker lung-
85 cancer rates for each age-sex group to measure cumulative
tobacco exposure. The fourth independent variable, time, was
80 used to cover the impact of technological change on health
status.
75 Our regression equations took the form:
LnMa,k,i=Ca,k,i+b1LnY+b2LnHC+b3T,
70
where Ca,k,i is a constant term; Ma,k,i is the mortality rate for age-
65 group a, sex k, and cause i; and Y, HC, and T denote gross
domestic product per person, human capital and time,
60 respectively. For cancers, cardiovascular diseases, and chronic
respiratory disorders for age-groups older than 30 years, smoking
55 intensity was added to this equation. The detailed econometric
analysis that underlies our results is reported elsewhere.11
50 In general, these equations explained most of the variance in
Life expectancy (years)

Life expectancy at birth in 1990

group 1 mortality and much of variance in group 2 morality, but
45 Baseline life expectancy projected for 2020
explained little of the variance in intentional injuries (and road-
traffic accidents for children). The R2 for many cause clusters
was generally lower for men and women aged 70 or older than
Men for other age-groups, which probably reflects inaccurate coding
90 of causes of death or the narrower range of variation in mortality
rates between countries for this age-group. The R2 for
85 cardiovasucular diseases among women ranged from 48% to
69% between the ages of 15 and 69, whereas for men in the
80 same age-groups, the R2 ranged from 12% to 63%.
For causes that had R2 of less than 10%, we chose not to use
75 the parsimonious regression equations, and mortality rates
specific for age and sex were assumed to stay constant. For
70 intentional injuries in all age-sex groups, constant rates for
1990–2020 were assumed.
65 Predictions for 1990 based on these equations were compared
with the GBD estimates of mortality rates by cause in 1990. A
60 series of specific scalars for age, sex, cause, and region were
calculated, which were the observed death rate divided by the
55 predicted death rate. To make projections, these scalars were
assumed to remain constant from 1990 to 2020. A consequence
50 of the large scalars for group 1 disorders in regions such as sub-
Saharan Africa was that we may have overestimated the future
45 group 1 mortality.
For income per person and average years of schooling,
EME FSE CHI LAC OAI MEC IND SSA baseline, pessimistic, and optimistic projections were generated
from World Bank forecasts for 1995–2004 and the empirical
Region record 1950–90.5 Projections of smoking intensity were based on
Figure 1: Life expectancy at birth in 1990 and in baseline, the observed relation between the number of cigarettes
optimistic, and pessimistic scenarios in 2020
Abbreviations for regions are: EME=established market economies; Disorder Ranking Change in
FSE=formerly socialist economies of Europe; IND=India; CHN=China; ranking
OAI=other Asia and islands; SSA=sub-Saharan Africa; LAC=Latin 1990 2020
America and the Caribbean; MEC=the middle eastern crescent. (baseline
Vertical bars indicate upper and lower limits of life expectancy model)
projected for 2020 defined by optimistic and pessimistic scenarios. Within top 15
Ischaemic heart disease 1 1 0
years or older—were developed to produce parsimonious Cerebrovascular disease 2 2 0
equations for nine cause-of-death clusters: all group 1 Lower respiratory infections 3 4 ↓1
(communicable, maternal, perinatal, and nutritional disorders); Diarrhoeal diseases 4 11 ↓7
Perinatal disorders 5 16 ↓11
malignant neoplasms, cardiovascular diseases, digestive diseases, Chronic obstructive pulmonary disease 6 3 ↑3
chronic respiratory diseases, and other group 2 (non- Tuberculosis 7 7 0
communicable) disorders; road-traffic accidents, other Measles 8 27 ↓19
unintentional injuries, and intentional injuries (group 3 Road-traffic accidents 9 6 ↑3
disorders). We used these nine cause groups to find total Trachea, bronchus, and lung cancers 10 5 ↓5
Malaria 11 29 ↓18
mortality in each age-sex group.
Self-inflicted injuries 12 10 ↑2
We developed regression equations from a dataset based on Cirrhosis of the liver 13 12 ↑1
vital registration data from 47 countries for the years 1950–90. Stomach cancer 14 8 ↑6
We used four independent variables in the analysis: income per Diabetes mellitus 15 19 ↓4
person, average years of schooling per adult, smoking intensity, Outside top 15
and time. Income per person, measured in international dollars Violence 16 14 ↓2
(adjusted for differences in purchasing power not covered by War injuries 20 15 ↑5
official exchange rates), was the general proxy for many indices Liver cancer 21 13 ↑8
HIV 30 9 ↑21
of development.6 Research has shown consistently that education
also influences health status.7,8 Level of education was measured Table 1: Changes in ranking for most important causes of
as the average number of years of schooling for the population death from 1990 to 2020 in baseline scenario

Vol 349 • May 24, 1997 1499

 THE LANCET

Tuberculosis HIV
5·0 5·0
4·5 4·5
4·0 4·0
3·5 3·5
3·0 3·0
2·5 2·5
2·0 2·0
1·5 1·5
1·0 1·0
0·5 0·5
Deaths (x106)

0·0 0·0

Chronic obstructive pulmonary disease Diarrhoeal diseases

5·0 5·0
4·5 4·5
4·0 4·0
3·5 3·5
3·0 3·0
2·5 2·5
2·0 2·0
1·5 1·5
Baseline model
1·0 Optimistic model 1·0
0·5 Pessimistic model 0·5
0·0 0·0
1990 2000 2010 2020 1990 2000 2010 2020

Year
Figure 2: Projected increase in mortality from 1990–2020 for tuberculosis, HIV, chronic obstructive pulmonary
disease, and diarrhoeal diseases in baseline, optimistic, and pessimistic scenarios
consumed per adult in the UK from 1900 to 1990 and calculated Programme on AIDS for WHO, with some modifications.21 The
smoking intensity 30 years later. Estimates for per-person predictions suggested that, by 2020, there will be essentially few
cigarette consumption for men and women for each region were or no new cases of HIV in any region. Based on infectious-
used with this relation to estimate sex-specific future smoking disease modelling,22,23 this was too optimistic for our baseline
intensity. Since the future course of the smoking intensity scenario. The Global Programme on AIDS’ projections were,
variable is largely determined by current smoking patterns, we therefore, modified, and for the baseline scenario, we assumed
did not generate projections for the optimistic and pessimistic that the number of incident cases per year would stabilise once
scenarios. incidence fell to half of the peak incidence. The equilibrium
To generate projections of specific disorders within the nine incidence was assumed to be 75% of peak incidence for the
cause clusters, the relation between the age-sex-specific mortality pessimistic projections and 20% for the optimistic projections.
rate from a disorder and the age-sex-specific mortality rate from Because of the powerful interaction between tuberculosis and
the cause cluster to which the disorder belonged were used.5,12,13 HIV infection in regions such as sub-Saharan Africa, we
The variables that defined these relations were estimated from a predicted that about a third of HIV-positive individuals will die
restricted dataset of ICD-9 data from only 67 countries from from tuberculosis in high-prevalence regions.24 The pace of
about 1990. decline in tuberculosis death rates was modified to reflect the
The regression results for specific disorders were used only projected regional HIV seroprevalence.
when the relation was reasonably strong, shown by an R2 greater Based on projected death numbers for each model, years of life
than 0·25 and an X-coefficient with p<0·001. Otherwise, the lost (YLLs) were calculated by the GBD method.5 To project
proportionate distribution of the remaining cause cluster was disability-adjusted life years (DALYs) it was also necessary to
assumed to stay constant. project years lived with disability (YLDs). We divided causes of
We used the cause-cluster regression estimates and alternative death and disability into three categories, each of which had a
assumptions for selected causes, the projections of the different method to project YLDs: for those causes in which the
independent variables, and the equations for specific disorders to age-sex-specific ratio of YLDs to YLLs, as estimated in the 1990
project mortality rates by cause for eight regions and 14 age-sex GBD results, was less than ten in all regions, YLDs were
groups from the base year 1990 to 1995, 2000, 2005, 2010, estimated with the assumption that these ratios for age, sex, and
2015, and 2020. To screen for data-entry errors or mistakes in region would be constant from 1990 to 2020; for selected
the computer code, we examined graphs of each of these age- disabling disorders that are likely to decline the epidemiological
specific rates from 1990 to 2020 (10 976 graphs in total). Any transition, rates of change in the age-specific YLD rates were
errors were corrected. A final set of projected specific rates by indexed on the group 1 death rate; and specific YLD rates for
age, sex, cause, and region was then generated. age and sex for the remaining causes, such as bipolar disorder,
Alternative projections of incidence and mortality from the unipolar major depression, drug dependence, schizophrenia,
HIV epidemic are well documented.14–20 For the purpose of the alcohol dependence, Alzheimer’s disease and other dementias,
GBD, we used the projections prepared by the Global Parkinson’s disease, multiple sclerosis, post-traumatic stress

1500 Vol 349 • May 24, 1997

 THE LANCET

Disorder group Projected DALYs

EME FSE IND CHN OAI SSA LAC MEC Developed Developing World
Group 1
Total group 1 5·2 3·0 24·4 4·3 16·5 39·8 12·6 19·9 4·3 22·2 20·1
Infectious and parasitic diseases 2·9 1·0 17·3 1·4 9·7 28·5 7·4 8·4 2·1 14·3 12·9
Respiratory infections 1·2 0·8 3·2 1·1 2·8 5·4 1·5 4·6 1·0 3·4 3·2
Maternal disorders 0·0 0·1 0·3 0·1 0·3 0·6 0·2 0·4 0·1 0·3 0·3
Perinatal disorders 0·7 0·7 2·4 0·9 2·2 3·7 2·3 4·6 0·7 2·7 2·5
Nutritional disorders 0·4 0·3 1·2 0·9 1·4 1·5 1·1 1·9 0·4 1·3 1·2
Group 2
Total group 2 84·7 79·7 56·5 79·3 66·3 31·9 68·1 59·6 82·7 56·7 59·7
Malignant neoplasms 17·3 16·1 7·1 18·7 11·6 4·5 8·5 5·3 16·8 9·0 9·9
Other neoplasms 0·6 0·3 0·1 0·3 0·2 0·2 0·4 0·2 0·5 0·2 0·2
Diabetes mellitus 2·1 0·7 0·8 0·4 0·9 0·2 1·6 1·0 1·5 0·7 0·8
Endocrine disorders 0·9 0·2 0·0 0·2 0·2 0·4 1·0 0·6 0·6 0·4 0·4
Neuropsychiatric disorders 25·4 16·4 12·6 15·4 17·4 8·5 21·6 14·9 21·8 13·7 14·7
Sense organ disorders 0·1 0·1 2·7 1·7 2·3 1·3 1·2 1·3 0·1 1·8 1·6
Cardiovascular disorders 19·4 26·1 18·4 16·3 15·6 6·0 13·2 17·7 22·0 13·8 14·7
Respiratory disorders 5·3 8·1 6·4 16·3 4·3 4·5 6·3 6·6 6·4 7·4 7·3
Digestive disorders 5·2 3·9 2·5 3·5 6·5 1·8 4·6 3·7 4·7 3·4 3·5
Genitourinary disorders 1·2 1·1 0·8 1·1 1·3 0·9 1·3 2·1 1·2 1·1 1·2
Skin disorders 0·1 0·1 0·0 0·1 0·1 0·3 0·1 0·1 0·1 0·1 0·1
Musculoskeletal disorders 5·0 4·5 1·1 2·6 2·6 0·8 5·5 1·2 4·8 1·9 2·2
Congenital anomalies 0·9 1·2 3·2 1·9 1·8 2·2 1·7 3·3 1·0 2·4 2·2
Oral disorders 1·0 0·8 0·8 0·7 1·5 0·3 1·3 1·8 0·9 0·9 0·9
Group 3
Total group 3 10·1 17·4 19·1 16·4 17·2 28·3 19·3 20·5 13·0 21·1 20·1
Unintentional injuries 6·9 11·6 16·4 11·0 13·6 15·4 13·2 9·8 8·8 13·6 13·0
Intentional injuries 3·2 5·7 2·8 5·4 3·6 12·9 6·2 10·7 4·2 7·5 7·1
*Abbreviations for regions as in figure 1. Developed=established market economies and formerly socialist economies of Europe. Developing=all other regions.
Table 2: Percentage distribution of projected DALYs for specific two causes

disorder, panic disorder, obsessive-compulsive disorder, communicable diseases (group 2) is expected, with a rise
rheumatoid arthritis, osteoarthritis, benign prostatic hypertrophy, in annual mortality from an estimated 28·1 million deaths
dental caries, periodontal diease, edentulism, and glaucoma were in 1990 to 49·7 million (48·0 and 53·0) in 2020. The
assumed to remain constant from 1990 to 2020. projected increase in group 2 deaths is expected to be
Population projections for each region were developed from
larger for males (91%) than for females (61%), consistent
World Bank projections of fertility and the estimations in our
mortality projection scenarios. with the increase observed in industrialised countries
The projected rates of YLLs and YLDs that incorporated the during epidemiological transition. Deaths from injuries
projections of HIV and modifications for tuberculosis were (group 3) are also projected to increase dramatically from
applied to these projected populations to generate projected 5·1 million in 1990 to 8·4 million (8·2 and 8·4) in 2020.
numbers of deaths, YLLs, YLDs, and DALYs for each of the Increases in the absolute number of deaths due to group
three scenarios. 3 causes are largely determined by the projected changes
in population size and age structure, and, in particular,
Results by an increase in the number of men in the 15–29 years
Life expectancy at birth (figure 1) for women in all three age-group, in which the risk of death from injury is
scenarios was projected to increase in all regions, with the highest.
largest gains expected in sub-Saharan Africa, India, and Another way to describe the projected changes in
other Asia and islands. Life expectancy for women in mortality is to examine the changes in the leading causes
established market economies may reach close to 90 of death. Table 1 shows the predicted change in the
years—this is especially plausible since Asian women in ranking of various causes of death from 1990 to 2020 in
the USA already have a life expectancy at birth of over 86 the baseline projections. Diarrhoeal diseases, perinatal
years.25 The smallest gain for women is projected for disorders, measles, and malaria are all projected to decline
formerly socialist economies of Europe. Comparison of substantially in importance, whereas lung cancer, stomach
women’s and men’s projected life expectancy highlights cancer, war injuries, liver cancer, and HIV are predicted
the much lower life expectancy in men in 1990 in all to move up five or more places by 2020. The range
regions except India, and the far smaller gains projected defined by the optimistic and pessimistic projections
in all regions for men than women. The much smaller differs substantially by cause (figure 2). For example, the
improvements in male mortality are due largely to the trend for deaths from tuberculosis rises in the pessimistic
greater impact of tobacco use in men. Only in men in projection and falls in the optimistic projection, whereas
formerly socialist economies of Europe is there no for diarrhoeal diseases the trend is downwards in all
improvement in life expectancy projected between 1990 models.
and 2020, although, since life expectancy dropped in In 1990, an estimated 1·38 billion DALYs were lost
1995 in parts of the region such as Russia by as much as 5 due to disease and injury occurring in that year. The total
years,23 some gain between 1995 and 2020 is implied. number of DALYs in 2020 worldwide is expected to be
Worldwide annual mortality from communicable similar at about 1·39 billion in the baseline model (1·30
maternal, perinatal, and nutritional disorders (group 1) is billion and 1·69 billion in the optimistic and pessimistic
predicted to decline from 17·2 million in 1990 to 10·3 models, respectively). The proportionate contribution
million in 2020 in the baseline model (8·2 million in the from the three groups of disorders, however, is expected
optimistic and 16·9 million in the pessimistic scenarios). to change significantly (table 2). Therefore, in 2020,
Conversely, a very large increase in deaths from non- group 1 causes are projected to account for 20·1% (17·2,

Vol 349 • May 24, 1997 1501

THE LANCET

Rank Worldwide Developed regions Developing regions

Disease or injury DALYs Cum % Disease or injury DALYs Cum % Disease or injury DALYs Cum %
(3106) (3106) (3106)
·· All causes 1388·8 ·· All causes 160·5 ·· All causes 1228·3 ··
1 Ischaemic heart disease 82·3 5·9 Ischaemic heart disease 18·0 11·2 Unipolar major depression 68·8 5·6
2 Unipolar major depression 78·7 11·6 Cerebrovascular disease 9·9 17·4 Road-traffic accidents 64·4 10·8
3 Road-traffic accidents 71·2 16·7 Unipolar major depression 9·8 23·5 Ischaemic heart disease 64·3 16·1
4 Cerebrovascular disease 61·4 21·1 Trachea, bronchus, and lung 7·3 28·0 Chronic obstructive pulmonary 52·7 20·4
cancers disease
5 Chronic obstructive pulmonary 57·6 25·3 Road-traffic accidents 6·9 32·3 Cerebrovascular disease 51·5 24·6
disease
6 Lower respiratory infections 42·7 28·4 Alcohol use 6·1 36·1 Tuberculosis 42·4 28·0
7 Tuberculosis 42·5 31·4 Osteoarthritis 5·6 39·5 Lower respiratory infections 41·1 31·4
8 War injuries 41·3 34·4 Dementia and other degenerative 5·5 43·0 War injuries 40·2 34·6
and hereditary CNS disorders
9 Diarrhoeal diseases 37·1 37·1 Chronic obstructive pulmonary 4·9 46·0 Diarrhoeal diseases 37·0 37·6
disease
10 HIV 36·3 39·7 Self-inflicted injuries 3·9 48·4 HIV 34·0 40·4

Table 3: Ten projected leading causes of DALYs in 2020 according to baseline projection

29·4) of DALYs worldwide, compared with 43·9% in traffic accidents. Despite a 30-year decrease projected for
1990. The contribution from group 2 is projected to rise total group 1 death and DALYs, four group 1 disorders
from 40·9% to 59·7% (61·4 and 53·6). The relative are predicted to remain in the ten leading causes of
contribution from injuries is also expected to rise from DALYs in 2020: lower respiratory infections,
15·2% to 20·1% (21·4 and 17·0). tuberculosis, diarrhoea, and HIV. In the developing
DALYs due to all group 1 disorders are expected to regions, these four causes are also the only group 1
decrease substantially by 2020. This fall is expected for disorders expected to remain in the ten leading causes of
infectious and parasitic diseases, which accounted for DALYs. In developed regions, osteoarthritis, dementia,
22·9% of DALYs worldwide in 1990; the proportion is and breast cancer are all expected to be in the ten leading
expected to decrease to 12·9% (10·4 and 18·0) in 2020. causes of burden for women in 2020.
DALYs due to maternal disorders are expected to fall By the same method as we used to estimate mortality
from 2·2% to 0·3% (0·3 and 0·9) and those from attributable to tobacco in 1990,5 we estimated mortality
respiratory infections to fall from 8·5% to 3·2% (3·0 and and disability attributable to tobacco in all three models.
4·8) of the worldwide total. Conversely, major increases The number of deaths attributable to tobacco was
in DALYs are expected for some of the leading non- projected to increase from 3·0 million in 1990 to 8·4
communicable diseases. DALYs from cancers are million (in the baseline scenario) in 2020 (table 4). In the
expected to rise from 5·1% to 9·9% (10·5 and 18·0) of developed regions, the number of deaths attributable to
the worldwide total in 2020. The proportionate share of tobacco was projected to rise from 1·6 million in 1990 to
the global burden of disease due to neuropsychiatric 2·4 million in 2020. The largest increases in the epidemic
disorders is projected to rise from 10·5% in 1990 to of tobacco-related mortality will be in India, China, and
14·7% (15·7 and 12·2) in 2020, and that due to other Asia and islands, where attributable deaths will
cardiovascular diseases, to rise from 11·1% to 14·7% increase from 1·1 million to 4·2 million by 2020. In terms
(15·4 and 13·7). Chronic respiratory infections are also of DALYs, the contribution of tobacco in the baseline
likely to move to a higher rank, rising from 4·4% in 1990 model is projected to increase to nearly 9% of worldwide
to 7·3% (6·5 and 6·7) in 2020. Both unintentional and burden in 2020, by which time tobacco is projected to
intentional injuries are projected to increase from 11·1% cause more deaths than any single disease worldwide.
to 13·0% (13·8 and 11·3) and from 4·1% to 7·1% (7·6 Figure 3 shows the projected changes in population size
and 5·7) of the worldwide total, respectively. by age for the world, developed, and developing regions
Table 3 shows the ten leading causes of DALYs for used for our models.
both sexes together for the developed and developing
regions, and the world in the baseline scenario.
Discussion
Worldwide, the top three contributors to the burden of
In all three scenarios, substantial changes were predicted
disease in 2020 are predicted to be ischaemic heart
in regional patterns of mortality and disability. Some
disease, followed by unipolar major depression and road-
major themes are worth emphasising. The distribution of
deaths by age will shift from younger to older ages. A
Region* Deaths3106 (% of worldwide total)
major decline in the mortality and disability from
communicable, maternal, perinatal, and nutritional
1990 2020
disorders was predicted in all three scenarios, although
FME 1·1 (36·7%) 1·3 (15·5%)
FSE 0·5 (16·7%) 1·1 (13·1%)
the declines in the optimistic and baseline scenarios were
IND 0·1 (3·3%) 1·5 (17·9%) much greater than in the pessimistic scenario. Deaths and
CHN 0·8 (26·7%) 2·2 (26·2%) DALYs due to group 2 disorders were projected to
OAI 0·2 (6·7%) 0·7 (8·3%)
SSA 0·1 (3·3%) 0·3 (3·6%)
increase by 77% and 47%, respectively, in the baseline
LAC 0·1 (3·3%) 0·4 (4·5%) scenarios, and by similar amounts in the pessimistic and
MEC 0·1 (3·3%) 0·8 (9·5%) optimistic scenarios. The most striking increases in the
World 3·0 8·4 burden of neuropsychiatric disorders were projected for
*Abbreviations for regions as in figure 1. other Asia and islands, the middle eastern crescent, sub-
Table 4: Worldwide deaths attributable to tobacco use in 1990 Saharan Africa, and India. The largest change in DALYs
and 2020 due to injuries is expected in sub-Saharan Africa.

1502 Vol 349 • May 24, 1997

 THE LANCET

200 populations and increasing in others, the various

% change from 1990 to 2020
World
projections are extremely uncertain. The future course of
Developed
150 HIV in Asia is probably the most uncertain. The HIV
Developing
epidemic is an unprecedented reversal of human health
100 progress.
Perhaps the most important determinant of human
50 health trends is the increase in tobacco-related mortality
and disability. As regions have developed, age-specific
0 group 2 mortality rates have tended to decline; therefore a
15-year-old boy in sub-Saharan Africa has a higher risk of
–50 dying from a non-communicable disease than does a 15-
year-old in the established market economies. The
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70
pervasive effects of tobacco, however, are the major
factors that counter this decline in group 2 mortality.
Age (years)
Tobacco-related deaths were projected to increase from
Figure 3: Projected change in global population by age from 3·0 million in 1990 to 8·4 million in 2020. Tobacco use
1990 to 2020
accounted for 2·6% of global DALYs in 1990 and was
Female life expectancy at birth was projected to projected to account for 9·0% in 2020, which would
increase in all regions—by 4·1 years in formerly socialist make it the largest single health problem in 2020.
economies of Europe, and by 13·1 years in sub-Saharan National and international policy responses to this public-
Africa. Predicted gains in male life expectancy, by health challenge need to be intensified.
contrast, were much smaller, ranging from no change in The final factor that influenced our projections was the
formerly socialist economies of Europe to an increase of expected decline in group 1 mortality, particularly in
10·2 years in sub-Saharan Africa. The gender gap in life children and women of reproductive age. This forecast—
expectancy was, therefore, also projected to increase in all which is contrary to the rise of new and re-emerging
regions. Even in established market economies, a low- diseases—is based on the empirical record in the 47
mortality region, the male-female gap in life expectancy countries for which reliable data were available in
could well continue to widen, far beyond biological 1950–90. There was no convincing empirical evidence
differences in life potential. that, at the population level, the dramatic gains of the past
We attributed these dramatic trends to four main four decades in reduction of group 1 mortality will not
factors: ageing of the population, the spread of HIV, the continue in the future. The prediction of a decline in
rise of tobacco-related mortality and disability, and the group 1, which we forecast in all three scenarios was
decline in group 1 death rates. The average age of the implicitly based on the presumption that socioeconomic
world’s population will increase, largely because of the development will decrease disease incidence and severity,
demographic transition caused by declining fertility rates and that research and development will guarantee the
(figure 3). In the developed regions, the population aged availability of antibiotics that are effective against resistant
15–44 is predicted to decline, and the populations aged strains of major pathogens.
45–64 and 65 and older will increase by about 26% and The baseline, optimistic, and pessimistic scenarios we
71%, respectively. By contrast, in the developing world, designed offer three possible visions of future mortality
the growth rate in the adult population will be dramatic. and disability. There are, however, many other possible,
By 2020, the number of children and adolescents younger albeit less probable, scenarios that could be envisaged,
than 15 will increase by 25% but the number of adults particularly for group 1 disorders. For example, new
aged 45–59 years will increase by 140%. Even if all age- antibotics may not be discovered, and drug resistance
specific death and disability rates were to remain may become so common for tuberculosis, malaria,
constant, the differential rate of increase in different age- pneumococcus, and other pathogens, that mortality rates
groups would lead to major changes in the burden of could increase. Such a scenario is possible, but, we
disease. If 1990 death and DALY rates were to remain believe, not highly probable. In-vitro drug resistance of
constant but the expected changes in population age certain bacteria—for example, methicillin-resistant
structure were to occur, the increase in group 2 burden Staphylococcus aureus—has been recognised for many
expected in 2020 would be even larger than in our three years. Because of many factors, however, including
scenarios. In other words, age-specific group 2 death and differences between in-vitro resistance and treatment
DALY rates are projected to decline (with a few failure, and the continued development of novel
exceptions, such as some adult male age-groups in India, antibiotics, there has been no increase in mortality
China, and formerly socialist economies of Europe. directly attributable to drug resistance in any population
HIV, which 20 years ago was not a significant that had reliable data on mortality trends. Possible but
determinant of global mortality or disability, will become, improbable scenarios may still, however, be important for
soon after the year 2000, one of the ten leading causes of policy formulation. Societies should be willing to reduce
mortality and disability. According to the modified Global the risks or consequences of unlikely, but real and
Programme on AIDS projections, HIV will cause 1·7 potentially disastrous outcomes.
million deaths in 2006—in the optomistic scenario Any projection exercise is, by its nature, an exercise in
mortality peaks at 1·7 million deaths, and in the conjecture with wide confidence intervals. Nevertheless,
pessimistic model mortality peaks at 1·9 million deaths in some features of our study introduce more uncertainty
2012. In these projections, HIV will still cause 1·2 million than others. First, the timing and magnitude of the peak
(0·8 and 1·8 million) deaths and 2·6% (1·8 and 3·0) of all in HIV incidence was based on estimates developed by
DALYs in 2020. Clearly, with a disease such as HIV, for the Global Programme on AIDS. Alternative projections
which the incidence of new infections is declining in some with higher or lower peak incidences that occur sooner or

Vol 349 • May 24, 1997 1503

THE LANCET

later are also credible. Second, death and YLD rates for 2 Garrett MJ. A way through the maze: what futurists do and how they
do it. Futures 1993; 25: 254–74.
intentional injuries were assumed to remain constant from
3 Taket A. Health futures in support of health for all: report of an
1990 to 2020, since we were unable to indentify international consultation convened by the World Health
significant relations between income, education, and time, Organization, Geneva, 19–23 July, 1993. Geneva: WHO, 1993.
based on the datasets available. The panel dataset, 4 Ruwaard D, Kramers PGN, van den Berg Jeths A, Achterberg PW.
Public health status and forecasts: the health status of the Dutch
however, included some examples of countries, such as population over the period 1950–2010. Bilthoven: National Institute of
Hungary, where injury death rates have changed Public Health and Environmental Protection (RIVM), 1994.
significantly over a short time. Injury death rates in a 5 Murray CJL, Lopez AD, eds. The Global Burden of Disease: a
comprehensive assessment of mortality and disability from diseases,
given region may, therefore, increase or decrease by 2020. injuries, and risk factors in 1990 and projected to 2020. Cambridge:
For war injuries in particular, the assumption of constant Harvard University Press, 1996.
rates combined with substantial adult-population growth 6 Summers R, Heston A. The Penn World Tables (Mark 5·0): an
gives projections that are highly uncertain. Third, there expanded set of international comparisons 1950–1988. Q J Economics
1991; 106: 327–68.
are specific relations between group 1 disorders and group 7 Cleland JG, van Ginneken J. Maternal education and child survival in
2 disorders, such as hepatitis B and liver cancer, and developing countries. Soc Sci Med 1988; 27: 1357–68.
human papillomavirus and cervical cancer, that were not 8 Caldwell JC. Education as a factor in mortality decline: an examination
modelled directly. We predict that those group 2 of Nigerian data. Popul Stud 1979; 41: 5–30.
9 Barro RJ, Lee JW. Economic growth in a cross-section of countries:
disorders that are related to group 1 disorders may NBER working paper. Cambridge: NBER, 1993.
decrease at a faster rate than other group 2 disorders. As a 10 Peto R. Mortality from tobacco in developed countries: indirect
result, our projections for such disorders may be estimation of national vital statistics. Lancet 1992; 339: 1268–78.
overestimated. Finally, the entire set of YLD projections 11 Acharya A, Murray CJL. An econometric model for forecasting age-
and cause-specific mortality rates. Cambridge, Massachusetts: Harvard
was based on either fixed-relational models or arbitrary Center for Population and Development Studies, 1996.
assumptions about rates of decline. 12 Preston SH. Mortality patterns in national populations. New York:
Our results provide a comprehensive view of the future Academic Press, 1976.
and may aid health planners in many different fields of 13 Bulutao R. Mortality by cause 1970–2015. In: Gribble JN, Preston
SH, eds. The epidemiological transition: policy and planning
public health. However, the strength of these projections implications for developing countries, workshop proceedings.
it is also their major disadvantage. Focused, detailed Washington DC: National Academy Press, 1993.
projections of a specific disorder or injury in a given 14 Anderson RM, May RM. Epidemological parameters of HIV
transmission. Nature 1988; 333: 514–19.
population are likely to be more reliable than the generic 15 Anderson RM, May RM, Boily MC, Garnett GP, Rowley JT. The
methods employed here, simply because more spread of HIV-1 in Africa: sexual contact patterns and the predicted
information about the disease or injury would usually be demographic impact of AIDS. Nature 1991; 352: 581–89.
available to guide projections. 16 Anderson R. AIDS: trends, predictions, controversy. Nature 1993;
363: 393–94.
We recognise the uncertainties in our projections and 17 Brookmeyer R, Damiano A. Statistical methods for short-term
we urge those who use them to exercise caution in their projections of AIDS incidence. Stat Med 1989; 8: 23–24.
application. We emphasise that the projections are simply 18 Karon J, Dondero T, Curran J. The projected incidence of AIDS and
three visions of the future—the numerical consequences estimated prevalence of HIV infection in the United States. AIDS
1988; 1: 542–50.
of the set of assumptions and methods used to generate 19 Chin J, Lwanga S. Estimation and projection of adult AIDS cases: a
these visions. Perhaps our method and assumptions will simple epidemiological model. Bull World Health Organ 1991; 69:
prove to be reasonable, perhaps not. In making the results 399–406.
20 Chin J, Romenyi MA, Morrison F, Bulutao R. The global
available to the public-health community, we hope that epidemiology of the HIV/AIDS pandemic and its projected
the methods and assumptions will be challenged, demographic impact in Africa. World Health Stat Q 1992; 45:
rethought, and reapplied to enhance their usefulness. 220–27.
There is a clear need for projections of this nature, and 21 Low-Beer D, Stoneburner R, Mertens T, Berkley S. HIV. In: Murray
CJL, Lopez AD, eds. Health dimensions of sex and reproduction: the
we are aware that the research reported here is, or should global burden of sexually transmitted diseases, HIV, maternal
be, seen as the first step in an exhaustive effort to predict conditions, perinatal disorders, and congenital anomalies. Cambridge:
the future of health. Harvard University Press, 1996.
22 Bailey C. The mathematical theory of epidemics. London: Charles
This work is supported by the Edna McConnell Clark Foundation, the Griffin, 1957.
Rockefeller Foundation, the World Bank, and WHO. The views expressed 23 Anderson RM, May RM. Epidemiological parameters of HIV
are entirely those of the authors and do not reflect the opinions, policies, transmission. Nature 1988; 333: 514–19.
or standards of WHO. WHO considers that DALYs and the burden of 24 Kumaresan J, Raviglione MC, Murray CJL, Solomon J. Tuberculosis.
disease approach discussed in these papers are potentially useful for health In: Murray CJL, Lopez AD, eds. The global epidemiology of infectious
situation assessment but require further research. diseases. Cambridge: Harvard University Press, 1996.
25 Hahn RA, Eberhardt S. Life expectancy in four US racial/ethnic
populations: 1990. Epidemiology 1995; 6: 350–55.
References 26 Mesle F, Shkolnikov VM, Hertrich V, Vallin J. Tendances recentes de
1 Bezold C. Health care: thinking ahead. World Health Forum 1994; 15: la mortalité par cause en Russie 1965–1994. Paris: INED Données
189–92. Statistiques, 1996.

1504 Vol 349 • May 24, 1997