Leave a comment

2.9 Assessing health needs: the Global Burden of Disease Study*

2.9
Assessing health needs: the Global Burden of Disease Study*

C. J. L. Murray and A. D. Lopez

Introduction
Measuring disease burden

Summary measures of population health

Constructing disability-adjusted life years: social value choices

Sensitivity analyses
Estimating mortality and disability

Classification

Estimating regional mortality patterns

Assessing disability
The global burden of disease in 1990: main findings

Regional imbalances in the burden of disease

Major causes of disease burden

Sex differences in disease burden
The global burden of disease in 1990: assessment of risk factors

Assessing risk factor burden

The contributions of risk factors to global burden

Diseases as risk factors for other diseases
The global burden of disease projections from 1990

Projection methods

Mortality projections

Leading causes of disease burden in 2020: projections

Recent health trends in the 1990s: implications for the Global Burden of Disease Study projections
Progress in refining the Global Burden of Disease Study approach

Measuring and valuing health

Comparative risk assessment
Conclusion
Chapter References

Introduction
The epidemiological transition and rapid changes in disease patterns have posed serious challenges to health-care systems and forced difficult decisions about the allocation of scarce resources. Epidemiological information is often required at all levels of health systems, and compilations of mortality and morbidity statistics at the national and subnational levels have been published by many countries for several decades. However, prior to the Global Burden of Disease Study which began in 1992, there had been no comprehensive efforts to provide comparable regional and global estimates and projections of disease and injury burden based on a common methodology and denominated in a common metric.
One of the major goals of the Global Burden of Disease Study was to facilitate the inclusion of non-fatal health outcomes into debates on international health policy, which had largely drawn on the mortality data available in countries, much of it referring to children. Secondly, there was a need to decouple epidemiological assessment from advocacy so that estimates of the mortality or disability from a condition are developed as objectively as possible. In addition, there was a need to quantify the burden of disease using a measure that could then be used for cost-effectiveness analysis. The Global Burden of Disease Study method quantifies not merely the number of deaths but also the impact of premature death and disability on a population, combining these measures into a single unit of measurement of the overall burden of disease in the population. The Study also presented the first global and regional estimates of disease and injury burden attributable to certain risk factors for disease, such as tobacco, alcohol, poor water and sanitation, and unsafe sex. Quantifiable estimates and projections of disease and injury burden from various exposures to specific disease and injuries, measured in a comparable fashion, are required if information on comparable assessments is to guide health policy debate.
Measuring disease burden
The incorporation of the burden of premature mortality and disability into one summary measure requires a common metric. Since the late 1940s, researchers have generally agreed that time is an appropriate currency: time (in years) lost through premature death, and time (in years) lived with a disability. A range of such time-based measures has been used in different countries, many of them variants of the so-called quality-adjusted life year. For the Global Burden of Disease Study, an internationally standardized form of the quality-adjusted life year was developed, called the disability-adjusted life year. The disability-adjusted life year expresses years of life lost to premature death and years lived with a disability of specified severity and duration. One disability-adjusted life year is the equivalent of one lost year of healthy life. Here, a premature death is defined as one that occurs before the age to which a person could have expected to survive if he or she were a member of a model population with a life expectancy at birth approximately equal to that of the world’s longest-surviving population—Japan.
To calculate total disability-adjusted life years for a given condition in a population, years of life lost and years lived with disability for that condition must each be estimated, and then summed. For example, to calculate disability-adjusted life years incurred through road traffic accidents in India in 1990, the total years of life lost in fatal road accidents must be added to the total years of life lived with disabilities by survivors of such accidents, weighted by the severity of the disability. For the Global Burden of Disease Study, 1990 was chosen as the base year for estimating disease burden.
Summary measures of population health
Over the past 30 or so years, several indicators have been developed to adjust mortality to reflect the impact of morbidity or disability. These summary measures of population health fall into two basic categories, health expectancy and health gap (Murray et al. 1999) (Fig. 1). Within the former category, Sullivan (1971) first suggested weighting life expectancy to measure the health of a population using a single indicator, disability-free life expectancy. Disability-free life expectancy incorporates a dichotomous weighting scheme, that is, it does not account for varying levels of severity. Wilkins and Adams (1983) suggested a more sensitive weighting scheme based on functional limitations, leading to the disability-adjusted life expectancy approach.

Fig. 1 A typology of summary measures.

As a summary measure of the burden of disability from all causes in a population, disability-adjusted life expectancy has two advantages (Murray and Lopez 1996a) over other summary measures. The first is that it is relatively easy to explain the concept of a lifespan without disability to a non-technical audience. The increasing popularity of health expectancy indicators among policy-makers has been documented (van de Water et al. 1996; Barendregt et al. 1998). The second is that it is easy to calculate disability-adjusted life expectancy using the Sullivan method which relies on prevalence data.
The disability-adjusted life year is an example of a particular type of health gap summary measure which allows the disaggregation of overall disease burden into the burden attributed to specific diseases, injuries, or exposures. In the Global Burden of Disease Study, the aim was to develop a measure based on explicit and transparent value choices that may be readily debated and modified. Overall, the disability-adjusted life year used in the Global Burden of Disease Study is an egalitarian measure in that it is built on the principle that only two characteristics of individuals that are not directly related to their health, their age and their sex, should be taken into consideration when calculating the burden of a given health outcome in that individual. Other characteristics, such as socio-economic status, race, or level of education are not considered.
Constructing disability-adjusted life years: social value choices
To assess premature mortality, the Study utilized a standard life table for all populations, with life expectancies at birth fixed at 82.5 years for women and 80 years for men. A standard life expectancy allows deaths at the same age to contribute equally to the burden of disease irrespective of where the death occurs. Alternatives, such as using different life expectancies for different populations that more closely match their actual life expectancies, violate this egalitarian principle. As life expectancy is rarely equal for men and women, the Global Burden of Disease Study assigned men a lower reference life expectancy than women. However, since much of the difference between men and women is determined by men’s higher exposure to various risks such as alcohol, tobacco, and occupational injury, rather than purely biological differences, this choice could be modified in future revisions of the Study.
If individuals are forced to choose between saving a year of life for a 2-year-old and saving it for a 22-year-old, most prefer to save the 22-year-old. A range of studies confirms this broad social preference to weight the value of a year lived by a young adult more heavily than one lived by a very young child or an older adult. Adults are widely perceived to play a critical role in the family, community, and society. It was for these reasons that the Global Burden of Disease Study incorporated age weighting into the disability-adjusted life year. It was assumed that the relative value of a year of life rises rapidly from birth to a peak in the early twenties, after which it steadily declines.
Individuals commonly discount future benefits against current benefits similarly to the way that they may discount future dollars against current dollars. Whether a year of healthy life, like a dollar, is also deemed to be preferable now rather than later, is a matter of debate among economists, medical ethicists, and public health planners, since discounting future health affects both measurements of disease burden and estimates of the cost-effectiveness of an intervention. There are arguments for and against discounting. In the Global Burden of Disease Study, future life years were discounted by 3 per cent per year. This means that a year of healthy life bought for 10 years hence is worth around 24 per cent less than one bought for now, as discounting is represented as an exponential decay function. Since the impact of discounting is significant, the findings of the Global Burden of Disease Study were published based on disability-adjusted life years with and without discounting. Discounting future health reduces the relative impact of a child death compared with an adult death. Another effect is that it reduces the value of interventions that provide benefits largely in the future, such as vaccinating against hepatitis B, which may prevent thousands of cases of liver cancer, but some decades later.
In order to quantify time lived with a non-fatal health outcome and assess disabilities in a way that will help to inform health policy, disability must be defined, measured, and valued in a clear framework that inevitably involves simplifying reality. There is surprisingly wide agreement between cultures on what constitutes a severe or a mild disability. For example, a year lived with blindness appears to most people to be a more severe disability than a year lived with watery diarrhoea, while quadriplegia is regarded as more severe than blindness. These judgements must be made formal and explicit if they are to be incorporated into measurements of disease burden.
Two methods are commonly used to formalize social preferences for different states of health. Both involve asking people to make judgements about the trade-off between quantity and quality of life. This can be expressed as a trade-off in time (how many years lived with a given disability a person would trade for a fixed period of perfect health) or a trade-off between persons (whether the person would prefer to save 1 life year for 1000 perfectly healthy individuals or 1 life year for perhaps 2000 individuals in a worse health state).
The Global Burden of Disease Study developed a protocol based on the person trade-off method. In a formal exercise involving health workers from all regions of the world, the severity of a set of 22 indicator disabling conditions—such as blindness, depression, and conditions that cause pain—was weighted between 0 (perfect health) and 1 (equivalent to death). These weights were then grouped into seven classes where class 1 has a weight between 0 and 0.02 and class VII a weight between 0.7 and 1. Subsequent valuation exercises carried out in various cultures have closely matched the results of the original Global Burden of Disease exercise (Table 1).

Table 1 Pearson’s correlation coefficients for median disability weights for 10 exercises, based on 14 conditions common to all exercises

In essence, the weight is set by the number of people with a given condition whose claim on a fixed health-care budget is equal, in the judgement of a participant, to that of 1000 healthy people. For example, if the participant judges that 1000 entirely healthy people would have an equal claim on the resources as 8000 people with some severe disability, the weight assigned to that particular disability is equal to 1 – (1000/8000), or 0.875. If 1000 entirely healthy people were judged to have an equal claim on the resources as 2000 people with a particular, less severe, disability, the weight assigned would be equal to 1 – (1000/2000), or 0.5.
For the Global Burden of Disease protocol, each participant is asked two versions of the person trade-off question: one about extending life for people in a given health state versus extending life for healthy people, and the second about giving health back to people in a given health state versus extending life for healthy people. Two questions are asked because people’s answers to each one are invariably inconsistent with the other, and the process of making them consistent forces the participant to think through the implications of their decision in greater depth.
The implications of choosing between the claims of different groups in a society are profound, so the process of setting weights cannot be undertaken lightly. The Global Burden of Disease Study protocol is a deliberative process in which a comparatively small group of participants (between eight and twelve) are confronted with the implications of their decision, encouraged to discuss their choices with their peers, and allowed to revise their initial choices. Once the 22 indicator conditions have been weighted, the participants assigned the remaining conditions across the seven classes.
Sensitivity analyses
To gauge the impact of changing these social choices on the final measures of disease burden, the Global Burden of Disease Study assessments were recalculated with alternative age weighting and discount rates, and with alternative methods for weighting the severity of disabilities. Overall, the rankings of diseases and the distribution of burden by broad cause group are largely unaffected by age weighting and only slightly affected by changing the method for weighting disability. Changes to the discount rate, by contrast, may have a more significant effect on the overall results. A higher discount rate results in an increased burden in older age groups, while a lower discount rate results in an increased burden in younger age groups. Changes in the age distribution of burden, in turn, affect the distribution by cause, as communicable and perinatal conditions are most common in children while non-communicable diseases are most common in adults. The most significant effect of changing the discount and age weights is a reduction in the importance of several psychiatric conditions.
Ultimately, however, the accuracy of the underlying basic epidemiological data from which disease burden is calculated will influence the final results much more than the discount rate, the age weight, or the disability weighting method. If, for example, estimates of the incidence of blindness are off by a factor of 2, then the results, whatever the social value choices used in the metric, will be substantially incorrect. We conclude that much more effort needs to be invested in improving the basic epidemiological data than in analysing the effects of what are eventually minor adjustments to the particular summary measure of population health employed.
Estimating mortality and disability
Classification
As most developing countries still have only limited information about the distribution of causes of death in their populations, a primary objective of the Global Burden of Disease Study has been to develop comprehensive internally consistent mortality estimates worldwide for each major cause in 1990. Deaths were classified using a tree structure, in which the first level of disaggregation comprises three broad cause groups as follows.
Group I comprised communicable, maternal, perinatal, and nutritional conditions (International Classification of Diseases (10th revision) (ICD-10) codes A00–B99, G00, N70–N73, J00–J06, J10–J18, J20–J22, H65–H66, O00–O99, P00–P96, E00–E02, E40–E46, E50, D50). (For an explanation of these codes in terms of disease and injury entities, see WHO (1992).)
Group II comprised non-communicable diseases (ICD-10 codes C00–C97, D00–D48, D51–D89, E03–E07, E10–E16, E20–E34, E51–E89, F00–F99, G03–G99, H00–H61, H68–H95, I00–I99, J30–J99, K00–K92, N00–N64, N75–N99, L00–L99, M00–M99, Q00–Q99).
Group III comprised injuries (ICD-10 codes V01–Y89).
Each group was then subdivided into categories: for example, cardiovascular diseases and malignant neoplasms are two subcategories of group II. Beyond this level, there are two further disaggregation levels such that 107 individual causes from the ninth revision of the ICD (ICD-9) can be listed separately.
Consistent with the goal of providing disaggregated estimates of disease burden to assist priority setting in the health sector, estimates were prepared by age and sex and for eight broad geographic regions of the world: Established Market Economies, Formerly Socialist Economies of Europe, China, India, Latin America and the Caribbean, Middle-Eastern Crescent, Other Asia and Islands, and sub-Saharan Africa.
Estimating regional mortality patterns
The Study arrived at mortality estimates by cause by drawing on the following four broad sources of data.
Vital registration systems
Cause of death data certified by a doctor have been assembled through vital registration systems for over 100 years in some European countries. Data were available for 1990 or thereabouts for about 70 countries.
Sample death registration systems
In China, a set of 145 Disease Surveillance Points, representative of both urban and rural areas, and covering about 10 000 000 people, provides useful mortality data. In India, Maharashtra State provides full medical certification for at least 80 per cent of urban deaths, while a rural surveillance system including more than 1300 primary health-care centres nationwide was used to assess broad rural patterns of mortality.
Epidemiological assessments
Epidemiological estimates exist for specific causes in different regions. These estimates combine information from surveys on the incidence or prevalence of the disease with data on case-fatality rates for both treated and untreated cases.
Cause of death models
These are based on the fact that the broad cause structure of mortality is closely related to the level of mortality in a population. Such models estimate the distribution of deaths by cause in a population from historical studies of mortality patterns in countries with vital registration. The models developed for the initial Global Burden of Disease Study drew on a data set of 103 observations from 67 countries between 1950 and 1991, and were used primarily to provide plausibility bounds on estimates derived from epidemiological assessments.
Vital registration data, corrected where necessary for under-registration, were used to construct regional model life tables for those regions where registration was complete or virtually complete. For other regions, sex/age-specific mortality rates were estimated from survey and census data using conventional demographic techniques.
Assessing disability
A disease or injury may have multiple disabling effects, or sequelae. For example, diabetes may result in diabetic vascular disease, retinopathy, or amputation. To estimate the total burden of disability, the Study measured the amount of time lived with each of the various disabling sequelae of diseases and injuries, in both treated and untreated states, and weighted for their severity, in each population. In all, 483 disabling sequelae of disease and injuries were analysed for the Study, for all regions and age groups, and for both sexes.
Calculating the number of years lived with a disabling condition requires information about its incidence, the average age of onset, the average duration of the disability, and the severity weight for the condition. Epidemiological experts were requested to estimate each of these variables for each condition based on an in-depth review of published and unpublished studies. For each sequela, prevalence, case-fatality, remission, and mortality were estimated. This information allowed correction of the preliminary estimates for internal consistency, that is, ensuring that the estimated prevalence was consistent with estimated incidence and vice versa. Consistency was validated using DISMOD software which was specifically developed for the Study (Fig. 2). (DISMOD is a computer model (DISease MODel) which allows for simultaneous estimation of age patterns of basic epidemiological parameters, such as incidence, prevalence, case-fatality, and duration, based on knowledge of a limited set of these variables.) When inconsistencies were detected, epidemiological experts were asked to revise their initial estimates. The final disability estimates were the result of several rounds of revision in a process lasting nearly 5 years.

Fig. 2 Basic relationships between susceptibles, cases, and deaths used in developing DISMOD.

The number of years lived with a given disability for each individual were calculated from the incidence of the disability, with the ‘stream’ of disability arising from it measured from the age of onset, for the estimated duration of the disability, multiplied by the condition’s severity weight. To calculate the years lived with disability due to a condition in any given population, the number of years lived with disability lost per incident case must be multiplied by the number of incident cases. A case of asthma, for example, carries a disability weight of 0.1 if untreated and 0.06 if treated. If the annual incidence of asthma in males aged 15 to 44 years is 1 million cases, the untreated proportion is 35 per cent, and the average duration is 7 years, then this sequela alone is estimated to cause 664 000 years lived with disability for that demographic group. Unlike the estimates of years of life lost, not all sequelae of all conditions could be explicitly assessed for years lived with disability. Estimates for conditions not explicitly considered were made on the basis of information about the ratio of total premature mortality to disability for each broad cause group.
The global burden of disease in 1990: main findings
The results of the Study demonstrate clearly that disability plays a central role in determining the overall health status of a population. Yet that role has until now been almost invisible to public health. The leading causes of disability are shown to be substantially different from the leading causes of death, which has considerable implications for the practice of judging a population’s health from its mortality statistics alone.
A key aim of the Global Burden of Disease Study was to quantify the burden of fatal and non-fatal health outcomes in a single measure, the disability-adjusted life year. This section presents the main results of the assessments of overall burden for each region. (More detail on the age–sex–cause and sequelae patterns can be found in Murray and Lopez (1996b).) To calculate disability-adjusted life years due to each disease or injury in a given year and population, the years of life lost through all deaths in that year were added to the years of life expected to be lived with a disability for all new cases of disease or injury occurring in that year, weighted for the severity of the condition.
Regional imbalances in the burden of disease
Sub-Saharan Africa and India together accounted for more than 40 per cent of the total global burden of disease in 1990, although they make up only 26 per cent of the world’s population. In contrast, the Established Market Economies and the Formerly Socialist Economies of Europe, with about a fifth of the world’s population between them, together bore less than 12 per cent of the total disease burden. China emerged as substantially the most ‘healthy’ of the developing regions, with 15 per cent of the global disease burden and a fifth of the world’s population. This means that about 579 years of healthy life were lost for every 1000 people living in sub-Saharan Africa, compared with just 124 for every 1000 people in the Established Market Economies (Fig. 3).

Fig. 3 Distribution of disability-adjusted life years by region, 1990.

In terms of the risk of dying, the Study found a sevenfold higher risk of child death (that is, a newborn dying before the age of 15 years) in sub-Saharan Africa compared with a newborn in the Established Market Economies (Fig. 4). This extraordinary excess mortality in many developing regions must remain a priority for global health programmes. Somewhat surprisingly, the risk of adult death in the Formerly Socialist Economies of Europe region, at least for males, was higher than in any other region of the world, except Africa (Fig. 5). This largely reflects the rapid increase in adult male death rates in Russia since 1987. In 1990, mortality at these ages (15 to 59 years) was still rising rapidly in Russia, reaching a peak in 1994. Since then, the probability of death between the ages of 15 and 60 has declined as rapidly as it rose. The trends for females are qualitatively similar, though less extreme.

Fig. 4 regional probabilities of death for males and females by age and group, 1990.

Fig. 5 Change in the rank order of disease burden for 15 leading causes worldwide, 1990 to 2000.

In addition, the Global Burden of Disease Study has provided support for the theory that people in high-income low-mortality populations not only live longer, but remain healthier for longer as well. In recent years, researchers have been divided between those who argue that ill health is compressed into the last few years of life in these populations, and those who argue that longer life merely exposes people to a longer period of poor health. The results suggest that older people in the developed world are healthier than their counterparts in developing countries. It was also found that babies born in sub-Saharan Africa could expect to spend about 15 per cent of their lifespan disabled, compared with just 8 per cent for babies born in the Established Market Economies. A 60-year-old person in sub-Saharan Africa can expect to spend about half of his or her remaining years with a disability, whereas the same person in the Established Market Economies is likely to spend just one-fifth of those years disabled. The results suggest that the proportion of the lifespan lived with a disability falls as life expectancy rises.
Major causes of disease burden
While the leading causes of disease burden in 1990, namely lower respiratory infections, diarrhoeal diseases, and perinatal causes, may come as no surprise, the fact that depression was the fourth leading cause was perhaps unexpected (Table 2). Indeed, the Study showed that the burden of psychiatric conditions had been heavily underestimated. Of the 10 leading causes of disability worldwide (in years lived with disability) in 1990, five were psychiatric conditions: unipolar depression, alcohol use, bipolar affective disorder, schizophrenia, and obsessive–compulsive disorder. Unipolar depression alone was responsible for more than 1 in every 10 years of life lived with a disability worldwide. Altogether, psychiatric and neurological conditions accounted for 28 per cent of all years lived with disability, compared with 1.4 per cent of all deaths and 1.1 per cent of years of life lost. The predominance of these conditions is by no means restricted to the rich countries, although their burden is highest there. They were the most important contributors to years lived with disability in all regions except sub-Saharan Africa, where they still accounted for 16 per cent of the total.

Table 2 Ten leading causes of disability-adjusted life years (DALY) worldwide for both sexes, 1990

Alcohol use was the leading cause of male disability, and the tenth largest in women, in developed regions. More surprisingly, perhaps, it was also the fourth largest cause in men in developing regions. The remaining important causes of years lived with disability were anaemia, falls, road traffic accidents, chronic obstructive pulmonary disease, and osteoarthritis.
The traditional causes of disease burden in developing societies—communicable diseases, maternal and perinatal conditions, and nutritional deficiencies—remain of major concern in the 1990s. Even though these group I conditions accounted for only 7 per cent of the burden in the Established Market Economies and less than 9 per cent in the Former Socialist Economies, they nevertheless made up more than 40 per cent of the total global burden of disease in 1990, and almost half the burden (49 per cent) in developing regions. In sub-Saharan Africa, 2 out of every 3 years of healthy life lost were due to group I conditions. Even in China, where the epidemiological transition is far advanced, a quarter of years of healthy life lost were due to this group. Worldwide, five out of 10 leading causes of disease burden (as measured by disability-adjusted life years) are group I conditions: lower respiratory infections (pneumonia), diarrhoeal disease, perinatal conditions, tuberculosis, and measles.
The burden of injury in 1990 was highest in the Formerly Socialist Economies of Europe, where almost 19 per cent of all burden was attributed to this group of causes. China had the second largest injury burden, followed by Latin America and the Caribbean with the third largest. Even in the Established Market Economies, however, the burden of injuries—dominated by road traffic accidents—was almost 12 per cent of the total. In almost all regions, unintentional injuries were a much greater source of ill health in 1990 than intentional injuries such as interpersonal violence and war. The only exception was the Middle-Eastern Crescent, where unintentional and intentional injuries took an approximately equal toll because of a particularly high burden of war in the region at the time.
Sex differences in disease burden
Although girls and boys suffer from broadly similar health problems in infancy and early childhood, striking sex differences emerge in adults. Firstly, and most obviously, women suffer disproportionately from their reproductive role. Although the burden of reproductive ill health is almost entirely confined to the developing regions, it is so great that, even worldwide, maternal conditions make up three out of the ten leading causes of disease burden in women aged 15 to 44. In developing regions, five out of the ten leading causes of disability-adjusted life years are related to reproductive ill health, including the consequences of unsafe abortion and Chlamydia infection. However, the Study revealed that poor reproductive health is not the only concern for women. In both developing and developed regions, depression is women’s leading cause of disease burden. In developing regions, suicide is the fourth. Thus, while programmes to reduce the burden of poor reproductive health among women must remain a high priority in the future, their psychological health deserves greater attention as well. For men aged 15 to 44, road traffic accidents are the greatest cause of ill health and premature deaths worldwide, and the second greatest in developing regions, surpassed only by depression. Alcohol use, violence, tuberculosis, war, bipolar affective disorder, suicide, schizophrenia, and iron-deficiency anaemia make up the remainder of the list in developing countries. Until only recently, road traffic accidents in developing regions have received relatively little attention from public health specialists.
The global burden of disease in 1990: assessment of risk factors
Exposure to particular hazards, such as tobacco, alcohol, unsafe sex, or poor sanitation, can significantly increase individual risk of developing disease. These hazards, or risk factors, are significant contributors to the total global disease burden and health policy-makers need accurate information on their impact in order to devise effective prevention strategies. Until recently, however, there have been few attempts to measure the burdens of these risk factors, or to express them in a currency that can be compared directly with the burdens of individual diseases. The Global Burden of Disease Study assessed, for the first time, the mortality and loss of healthy life that can be attributed to each of 10 major risk factors in each region. These risk factors are malnutrition, poor water supply, sanitation, and personal/domestic hygiene, unsafe sex, tobacco use, alcohol use, occupation, hypertension, physical inactivity, illicit drug use, and air pollution.
Assessing risk factor burden
The burden of disease or injury in a population that can be attributed to past exposure to a given risk factor is, essentially, an estimate of the burden that could have been averted in the population if that particular risk factor had been eliminated. More precisely, this is defined as the difference between the currently observed burden and the burden that would be observed if past levels of exposure had been equal to a specified reference distribution of exposure. In general, to calculate this, it is necessary to know: (a) the relative risk at different levels of exposure for each cause of death and disability linked to the factor; (b) the distribution of different levels of exposure in the population; and (c) the burden of disease or injury due to each of the causes linked to the factor. Depending on the nature of the risk factor, the reference distribution against which relative risk is compared could be zero exposure for the whole population, a population distribution of exposure from low to high levels based on observed populations, or an arbitrary distribution. The Study used, wherever possible, zero exposure as the reference, except for risk factors such as hypertension, where clearly no exposure was not an appropriate reference standard.
The contributions of risk factors to global burden
Of the 10 risk factors studied, the most significant were malnutrition, poor water, sanitation, and hygiene, unsafe sex, alcohol, tobacco, and occupation. Together, these six hazards accounted for more than one-third of total disease burden worldwide in 1990 (Table 3). Of the six, malnutrition and poor sanitation were the dominant hazards, responsible for almost a quarter of the global burden between them. Unsafe sex and alcohol each contributed approximately 3.5 per cent of the total disease burden, closely followed by tobacco and occupation hazards with just under 3 per cent each. These are comparable to the disease burden due to tuberculosis or measles. As might be expected, major inequalities exist between regions and between men and women in the burdens of most risk factors. For example, the ill health consequences of unsafe sex—which include both infections and the complications of unwanted pregnancy—are borne disproportionately by women in all regions. In young adult women in sub-Saharan Africa, unsafe sex accounts for almost one-third of the total disease burden. Tobacco and alcohol, owing to longer exposures, caused their heaviest burdens in men in the developed regions, where the two together accounted for more than one-fifth of the total burden in 1990. In Asia and other developing regions, the rapid increase in tobacco use over the past few decades is expected to kill many more people in the coming decades than have so far died in the developed regions.

Table 3 Contribution of 10 risk factors to the global burden of disease and injury in 1990

The impact of alcohol varies between regions not only because of different levels of use in each population, but also because of differences in the age structure of those populations. Alcohol has consistently been shown to provide some protection against death from ischaemic heart disease, but to increase the risk of several other diseases, such as pancreatitis, several cancers, cirrhosis of the liver, and many injuries. Because of its protective effect against ischaemic heart disease, in populations such as the Established Market Economies where this condition is common and injuries and violence are comparatively rare, alcohol may prevent about as many deaths as it causes. Nevertheless, alcohol causes a severe disease burden even in these rich countries, because it causes many injuries and premature deaths among younger adults and thus results in large numbers of years lived with a disability and years of life lost.
In sub-Saharan Africa, however, where ischaemic heart disease is relatively uncommon, the protective effect of alcohol is far outweighed by its harmful role in deaths and disability due to injuries. The contribution of alcohol to injuries is also extremely high in Latin America and the Caribbean, where alcohol use accounts for almost 10 per cent of total disease and injury burden, a figure surpassed only in the developed regions. Ultimately, alcohol is estimated to have caused about three-quarters of a million more deaths in 1990 than it averted, with more than four-fifths of the excess deaths in the developing regions.
Diseases as risk factors for other diseases
In addition to estimating disease and injury burden from risk factors such as tobacco and alcohol, it is important to recognize that several diseases may be factors for other diseases. The full impact of these conditions needs to be evaluated if public health priorities are to be appropriately guided. For example, diabetes mellitus strongly increases an individual’s risk of developing ischaemic heart disease and stroke, while infection with hepatitis B virus increases the risk of developing liver cancer and cirrhosis of the liver. Traditional methods of assessing deaths by the single underlying cause fail to capture these relationships. In the Global Burden of Disease Study, a short list of well-studied conditions were evaluated as risk factors. We estimated how much of the total disease burden would be averted in each region’s population if the condition were eliminated. The most dramatic differences between directly coded and total burden are for diabetes, hepatitis B, and hepatitis C (Table 4).

Table 4 Estimated impact of selected conditions viewed as risk factors, 1990

The global burden of disease projections from 1990
To plan health services effectively, policy-makers need to know how health needs might change in the future. To meet this need, the authors have developed projections of mortality and disability for each 5-year period from 1990 to 2020, by cause, for all regions and both sexes. The findings have considerable implications for public policy.
Projection methods
Rather than attempt to model the effects of the many separate direct, or proximal, determinants of disease from the limited data that are available, it was decided to model mortality change as a function of a limited number of socio-economic variables: (a) income per capita, (b) the average number of years of schooling in adults, termed ‘human capital’, and (c) time, a proxy measure for the secular improvement in health in the twentieth century that partly resulted from accumulating knowledge and technological development. These socio-economic variables show clear historical relationships with mortality rates; for example, income growth is closely related to the improvement in life expectancy that many countries achieved in the twentieth century. Because of their relationships with death rates, these socio-economic variables may be regarded as indirect, or distal, determinants of health. In addition, a fourth variable, tobacco use, was included because of its overwhelming impact on health, using information from more than four decades of research on the time lag between persistent tobacco use—measured in terms of ‘smoking intensity’—and its effects on health (Peto et al. 1992).
Death rates for all major causes based on historical data for 47 countries covering the period 1950 to 1991 were related to these four variables to generate the projections. A separate model was used for HIV and modifications for the interaction between HIV and tuberculosis. Three projection scenarios were developed using different projections of the independent variables.
Mortality projections
In all regions, life expectancy at birth is expected to increase for women. By 2020, infant girls born in the Established Market Economies may expect to survive to almost 88 years. For men, life expectancy will grow much more slowly, mainly because of the impact of the tobacco epidemic. Nevertheless by 2020, males born in sub-Saharan Africa, whose life expectancy at birth was below 50 in 1990, may expect to reach 58 years. (This projection, made before the massive effects of the HIV/AIDS epidemic were known, has since been revised downwards.) Males born in Latin America and the Caribbean, who in 1990 could have expected to live to 65, may expect to reach 71 years. However, for men in the Formerly Socialist Economies of Europe, life expectancy is not expected to increase at all between 1990 and 2020. This is partly due to the fact that life expectancy was falling in 1990, so that any positive change is likely to be merely recovering to the 1990 position.
In young children and adolescents under the age of 15 years, the risk of death is projected to decline dramatically in all regions, falling by about two-thirds in sub-Saharan Africa and India. In adult women, too, the risk of death is expected to fall in all regions. Men in the Formerly Socialist Economies of Europe and China, because of the tobacco epidemic, may expect a higher risk of dying between the ages of 15 and 60 than they do today. In other regions, the risk of death for men in this age group is expected to fall, but more modestly than in women. Remarkably, by 2020, men of this age group in the Formerly Socialist Economies of Europe could face a higher risk of death even than men in sub-Saharan Africa.
Deaths from communicable, maternal, and perinatal conditions and nutritional deficiencies (group I) are expected to fall from 17.3 million in 1990 to 10.3 million in 2020. As a percentage of the total burden, group I conditions are expected to drop by more than half, from 34 to 15 per cent. This projected reduction overall, despite increased burdens due to HIV and tuberculosis, runs counter to the now widely accepted belief that infectious diseases are making a comeback worldwide. It partly reflects the relative contraction of the world’s ‘young’ population; the age group under the age of 15 years is expected to grow by only 22 per cent between 1990 and 2020, whereas the cohort of adults aged between 15 and 60 is expected to grow by more than 55 per cent. In addition, the projection reflects the observed overall decline in group I conditions over the past four decades owing to increased income, education, and technological progress in the development of antimicrobials and vaccines. Even under the pessimistic scenario, in which both income growth and technological progress are expected to be minimal, deaths from these conditions are still expected to fall slightly to 16.9 million.
Clearly, it should not be taken for granted that the progress of the past four decades against infectious diseases will be maintained. It is possible, for example, that antibiotic development and other control technologies will not keep pace with the emergence of drug-resistant strains of important microbes such as Mycobacterium tuberculosis. If such a scenario were to prove correct, and if, in addition, case-fatality rates were to rise because of such drug-resistant strains, the gains of the present century could be halted or even reversed. The evidence to date nonetheless suggests that, as long as current efforts are maintained, group I causes are likely to continue to decline.
While group I conditions are expected to decline overall, deaths from non-communicable diseases are expected to climb from 28.1 million deaths in 1990 to 49.7 million in 2020, an increase of 77 per cent in absolute numbers. In proportionate terms, group II deaths are expected to increase their share of the total from 55 per cent in 1990 to 73 per cent in 2020. These global figures do not reveal the extreme nature of the change that is projected in some developing regions because they incorporate the projections for the rich nations, which show little change. In India, deaths from non-communicable diseases are projected to almost double, from about 4 million to about 8 million a year, while group I deaths are expected to fall from almost 5 million to below 3 million a year. In the developing world as a whole, deaths from non-communicable diseases are expected to rise from 47 per cent of the burden to almost 70 per cent.
The steep projected increase in the burden of non-communicable diseases worldwide is largely driven by population ageing, augmented by the large numbers of people in developing regions who are now exposed to tobacco. It is important to note that ageing will result in a rise in the absolute numbers of cases of non-communicable diseases and in their increased share of the total disease burden for the population as a whole, but not in any change to the rates of those diseases in any given age group. As studies in the Established Market Economies show, the age-specific rates of some important non-communicable diseases, such as ischaemic heart disease and stroke, have been falling steadily for at least two decades. Whether these rates are also falling in other regions is much less clear. However, any age-specific decrease in the rates of these diseases that may also emerge in low-income countries is likely to be outweighed by the large and demographically driven increase in the absolute numbers of adults at risk for these diseases, augmented by the tobacco epidemic. As with non-communicable diseases, deaths from injury are also expected to rise for mainly demographic reasons. Young adults are generally exposed to greater risks of injury.
Leading causes of disease burden in 2020: projections
When disability is taken into account as well as death, a different view of the future emerges—and one that emphasizes adult health problems still further. By 2020, the disease burden due to communicable diseases, maternal and perinatal conditions, and nutritional deficiencies is expected to fall to a fifth of the total. The burden attributable to non-communicable diseases, accordingly, is expected to rise sharply, and the burden from injuries is also expected to rise to equal that of group I conditions.
In 1990, the three leading causes of disease burden were, in descending order, pneumonia, diarrhoeal diseases, and perinatal conditions. The three conditions projected to take their place by 2020 are ischaemic heart disease, depression, and road traffic accidents. Pneumonia is expected to fall to sixth place, diarrhoeal diseases to ninth, and perinatal conditions to eleventh. Notably, measles, currently in eighth place, is expected to drop to twenty-fifth. However, not all infectious diseases are expected to decline, despite the projected overall decline of group I conditions. Tuberculosis is expected to remain at its current level of seventh place, a substantial source of disease burden for the foreseeable future. Of equally great concern is the finding that HIV, currently twenty-eighth in the ranking, will be in the top 10 by 2020 (Fig. 6).
As with the 1990 assessments, neuropsychiatric conditions emerge as a highly significant component of global disease burden when disability, as well as death, is taken into account. The projections show that psychiatric and neurological conditions could increase their share of the total global burden by almost half, from 10.5 per cent of the total burden to almost 15 per cent in 2020. This is a larger proportionate increase than that for cardiovascular diseases.
By 2020, the burden of disease attributable to tobacco is expected to outweigh that caused by any single disease. From its 1990 level of 2.6 per cent of all disease burden worldwide, tobacco is expected to increase its share to just under 9 per cent of the total burden in 2020, compared with just under 6 per cent for ischaemic heart disease (part of which is due to tobacco), the leading projected disease. This is a global health emergency that many governments have yet to confront.
The burdens of several important types of injury are also likely to increase because of the growth of the adult fraction of the population. For example, young men are the group most frequently involved in road traffic accidents, so if the proportion of young adults in the population increases sharply, road traffic accidents are likely to increase too. Indeed, according to the baseline projection, road traffic accidents could rise to third place from ninth worldwide. Violence, currently nineteenth, could rise as high as twelfth place and suicide could climb from seventeenth to fourteenth place.
Not surprisingly, these changes are not expected to be evenly dispersed worldwide. The total number of lost years of healthy life in the Established Market Economies is likely to fall slightly, while it will increase slightly in the Formerly Socialist Economies of Europe. Strikingly, however, sub-Saharan Africa’s future looks disturbingly poor despite the decline in the burden of group I conditions that currently dominate its health needs. Overall, the region faces an increase in the number of lost years of healthy life between 1990 and 2020, due mainly to a steep projected rise in the burden of injuries from road accidents, war, and violence.
Recent health trends in the 1990s: implications for the Global Burden of Disease Study projections
The original Global Burden of Disease Study projections were based on data and information about health conditions worldwide in the late 1980s/early 1990s. At that time, two major epidemics were affecting the health of large population groups: the HIV epidemic, particularly in Africa, which killed an estimated 300 000 people worldwide in 1990 but had, by that time, infected millions more, and the explosive increase in adult mortality rates in Russia and neighbouring countries, particularly from cardiovascular diseases and injuries, and particularly among men. Making projections in the context of such dramatic epidemiological trends is extremely hazardous, as recent trends have confirmed.
The 1990 Global Burden of Disease Study’s HIV/AIDS projections have severely underestimated the spread of the epidemic in sub-Saharan Africa, particularly southern Africa. By 1999, HIV/AIDS was estimated to have killed 2.2 million Africans, several times more than projected on the basis of what was known in 1990. Whether the disease burden will continue to rise, and how far, is uncertain and new projection methods are being developed to forecast the epidemic better, particularly in Africa.
The other large uncertainty in the projections, namely adult mortality in the Formerly Socialist Economies of Europe, has confounded epidemiologists with the dramatic change in mortality risks during the 1990s. Death rates are now falling markedly in most large countries in this region, and, if they continue to do so, the 1990 forecasts will prove to be unduly pessimistic. It is too early to decide whether or not the recent declines in mortality are the beginning of a long-term secular trend in mortality.
Progress in refining the Global Burden of Disease Study approach
Measuring and valuing health
One of the major innovations of the Global Burden of Disease Study was the attempt to measure and value states of health worse than perfect health in a comparable fashion across various societies. This presupposes a common conceptual framework and measurement strategy. In particular, the key domains of health that need to be assessed and the minimum number of items and response categories needed to measure them need to be decided. Self-report instruments currently in use lack cross-cultural comparability, with the result that the measurement of health in various populations is largely not comparable. The development and operation of a conceptual framework to measure and describe health in a way that improves comparability across populations is a key challenge for burden of disease research (Murray and Lopez 2000).
The issue of comorbidity is another measurement problem to emerge from the original Global Burden of Disease Study where further methodological work is required. In the Study, comorbid conditions were valued separately and time spent with these combined states was valued as the sum of the individual state valuations. This additive model is clearly problematic. More data are required on the prevalence of major comorbidities in order to avoid multiple attribution in health state valuations.
The process of valuing health states worse than perfect health entails a number of methodological choices ranging from which measurement methods to use to the choice of respondents. In particular, the empirical assessment of health state valuations from large population surveys in different countries would greatly increase the representativeness of disability weights applied in the assessment of the global burden of disease. Research on measuring health state valuations in more than a dozen countries in all regions has begun at the World Health Organization (WHO) and such population-based valuations are being collected for use in future iterations of the Global Burden of Disease Study.
Comparative risk assessment
The Global Burden of Disease assessment of the magnitude of risk factors illustrated the power of comparing risk factor burden with the burden of specific diseases and injuries. Such comparisons are an effective means of drawing the attention of decision-makers to the magnitude of health problems caused by various distal socio-economic, proximal, or physiological variables. In future efforts at comparative risk factor assessment, the number of variables examined will be expanded to include distal determinants. One clear method of improving future revisions of the Global Burden of Disease Study is to standardize the methods for risk factor assessment.
In order to standardize terminology and permit comparable assessments of disease burden due to various exposures, Murray and Lopez (1999) have proposed a series of criteria for future risk factor assessments. A key distinction is made between attributable burden (from past exposures) and avoidable burden (due to future exposures).
In addition, given the very different traditions in risk factor epidemiology, depending on the specific risk factor under study, recent efforts have focused on standardizing the approach to defining and measuring population distributions of exposure. In future, estimates of disease and injury burden will be assessed by comparing current exposure (and hazards) to a ‘counterfactual’ distribution of exposure strictly defined in the same fashion for all exposures being assessed. One obvious counterfactual is the population distribution of exposure which results in the theoretical minimum risk for the entire population. For most risk factors (for example, tobacco use, illicit drugs) this minimum risk counterfactual will be 100 per cent of the population having zero exposure, but for others (for example, cholesterol, hypertension), the theoretical minimum is not so obvious and may be derived from different perspectives (for example, lowest empirical observation, animal studies).
Other counterfactuals, defined in exactly the same way for all risk factors, could also be developed, such as reducing the population distribution of exposure to a level which is ‘feasible’ or even ‘plausible’ given current knowledge. Whatever distribution is chosen to assess risk factor burden, it is critical that the same definition is applied across all risk factors to improve comparability.
Most standardized approaches are also developed to ensure that relative risk estimates control, as far as possible, for obvious confounding, and where this has not been rigorously applied in the epidemiological literature, adjustment methods have been proposed similar to those used by Peto et al. (1992) for tobacco. The standardization framework being applied for risk factor burden also includes common guidelines for judging causality and for extrapolating relative risks from one population to another, and from younger to older ages.
Efforts to compile age–sex regional distributions of exposure and age–sex-specific relative risk are being led by the Burden of Disease Team and WHO. First results of this revised global Comparative Risk Factor Assessment project are expected in 2001.
Conclusion
The Global Burden of Disease Study has provided a new and much needed strategy to estimate current and projected health needs. In particular, it has shown that non-communicable diseases are rapidly becoming the dominant causes of ill health in all developing regions except sub-Saharan Africa, it has revealed the extent to which mental health problems have been underestimated worldwide, and it has shown the significance of injuries as a problem for the health sector in all regions. The findings pose new and immediate challenges to policy-makers and are certain to provoke debate. Ultimately, the Study’s impact will be judged in two ways: firstly, by the degree to which it stimulates other researchers to apply the same rigorous methods of measuring disease burden in all regions; secondly, to the extent that it changes priorities for public health in the decades ahead.

*The authors are extremely grateful to Brodie Ferguson for his assistance in preparing this chapter.
Chapter References
Barendregt, J.J., Bonneux, L., and van der Maas, P.J. (1998). Health expectancy: from a population health indicator to a tool for policy making. Journal of Aging and Health, 10, 242–58.
Murray, C.J.L. and Lopez, A.D. (ed.) (1996a). The Global Burden of Disease: a comprehensive assessment of mortality and disability from diseases, injuries, and risk factors in 1990 and projected to 2030. Global Burden of Disease and Injury Series, Vol. 1. Harvard University Press, Cambridge, MA.
Murray, C.J.L. and Lopez, A.D. (1996b). Global health statistics: a compendium of incidence, prevalence and mortality estimates for over 200 conditions. Global Burden of Disease and Injury Series, Vol. 2. Harvard University Press, Cambridge, MA.
Murray, C.J. and Lopez, A.D. (1999). On the comparable quantification of health risks: lessons from the Global Burden of Disease Study. Epidemiology, 10, 594–605.
Murray, C.J. and Lopez, A.D. (2000). Progress and directions in refining the global burden of disease approach: a response to Williams. Health Economics, 9, 69–82.
Murray, C.J.L., Salomon, J.A., and Mathers, C. (1999). A critical examination of summary measures of population health. GPE Working Paper Series. WHO, Geneva.
Peto, R., Lopez, A.D., Boreham, J., Thun, M., and Heath, C., Jr (1992). Mortality from tobacco in developed countries: indirect estimation from national vital statistics. Lancet, 339, 1268–78.
Sullivan, D.F. (1971). A single index of mortality and morbidity. HSMHA, Health Reports, 86, 347–54.
Van de Water, H.P., Perenboom, R.J., and Boshuizen, H.C. (1996). Policy relevance of the health expectancy indicator: an inventory of European Union countries. Health Policy, 36, 117–29.
WHO (World Health Organization) (1992). International statistical classification of diseases and related health problems (10th revision), Vol. 1. WHO, Geneva.
Wilkins, R. and Adams, O. (1983). Health expectancy in Canada, late 1970s: demographic, regional and social dimensions. American Journal of Public Health, 73, 1073–80.

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: