Global mortality, disability, and the contribution of risk factors: Global Burden of Disease Study Christopher J L Murray, Alan D Lopez THE LANCET taken into account, our list differs substantially from other lists of the leading causes of death. DALYs provide a common metric to aid meaningful comparison of the burden of risk factors, diseases, and injuries. Lancet 1997 349: 1436���42 Introduction In this, the third of a four-article series on the Global Burden of Disease Study (GBD) (see Lancet 1997 349: 1269���76 and 1347���52, and the next issue) the primary indicator used to summarise the burden of premature mortality and disability (including temporary disability) is the disability-adjusted life year (DALY). The burden of 107 disorders is compared with the burden attributable to ten major risk factors and to selected diseases as risk factors for other conditions. More extensive detail on the estimation of causes of death and development of epidemiological profiles of each disabling sequela have been published.1 DALYs are the sum of life years lost due to premature mortality and years lived with disability adjusted for severity. The value choices incorporated into DALYs and the basis of their selection have been extensively debated and discussed.2���4 In this article we give more details on the methods used to estimate attributable burden. Methods Estimation of attributable burden For the GBD, assessments of the burden attributable to each of the ten major risk factors were made by specialists on each topic: tobacco,5 alcohol,6 illicit drugs,7 occupation,8 air pollution,9 poor water supply, sanitation, and personal and domestic hygiene,10 hypertension,11 physical inactivity,12 malnutrition,13 and unsafe sex.14 Attributable burden in this study has been defined (for a specific risk factor, population, and time) as ���the difference between 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���. Definition of the reference distribution of exposure varies by risk factor (table 1). Malnutrition���Mason and colleagues13 estimated the burden of disease attributable to malnutrition with data from 55 studies on the relative risk of mortality as a function of the SD of nutritional status. The proportion of the population aged 0���4 years with a weight-for-age lighter than 2 SDs below the National Center for Health statistics��� reference population mean (based on extensive survey data in almost every country in the world) was used to estimate the attributable fraction of child mortality in each region. Poor water supply, sanitation, and personal and domestic hygiene���Huttly10 estimated attributable fractions for diarrhoea, ascariasis, trichuriasis, and dracunculiasis, based largely on the theoretical effects of interruption of the faecal-oral route of transmission. Unsafe sex���Berkley and colleagues14 estimated the burden attributable to unsafe sex based on 100% of the burden of Summary Background Prevention and control of disease and injury require information about the leading medical causes of illness and exposures or risk factors. The assessment of the public-health importance of these has been hampered by the lack of common methods to investigate the overall, worldwide burden. The Global Burden of Disease Study (GBD) provides a standardised approach to epidemiological assessment and uses a standard unit, the disability- adjusted life year (DALY), to aid comparisons. Methods DALYs for each age-sex group in each GBD region for 107 disorders were calculated, based on the estimates of mortality by cause, incidence, average age of onset, duration, and disability severity. Estimates of the burden and prevalence of exposure in different regions of disorders attributable to malnutrition, poor water supply, sanitation and personal and domestic hygiene, unsafe sex, tobacco use, alcohol, occupation, hypertension, physical inactivity, use of illicit drugs, and air pollution were developed. Findings Developed regions account for 11��6% of the worldwide burden from all causes of death and disability, and account for 90��2% of health expenditure worldwide. Communicable, maternal, perinatal, and nutritional disorders explain 43��9% non-communicable causes 40��9% injuries 15��1% malignant neoplasms 5��1% neuropsychiatric conditions 10��5% and cardiovascular conditions 9��7% of DALYs worldwide. The ten leading specific causes of global DALYs are, in descending order, lower respiratory infections, diarrhoeal diseases, perinatal disorders, unipolar major depression, ischaemic heart disease, cerebrovascular disease, tuberculosis, measles, road-traffic accidents, and congenital anomalies. 15��9% of DALYs worldwide are attributable to childhood malnutrition and 6��8% to poor water, and sanitation and personal and domestic hygiene. Interpretation The three leading contributors to the burden of disease are communicable and perinatal disorders affecting children. The substantial burdens of neuropsychiatric disorders and injuries are under- recognised. The epidemiological transition in terms of DALYs has progressed substantially in China, Latin America and the Caribbean, other Asia and islands, and the middle eastern crescent. If the burdens of disability and death are 1436 Vol 349 ��� May 17, 1997 Harvard School of Public Health, Boston, Massachusetts, USA (C J L Murray MD), and World Health Organization, Geneva, Switzerland (A D Lopez PhD) Correspondence to: Dr Christopher J L Murray, Harvard Center for Population and Developmental Studies, 9 Bow Street, Cambridge, MA 02138, USA

THE LANCET sexually transmitted diseases and fractions for HIV, hepatitis B, and cervical cancer (caused by human papillomavirus). In addition, the burden of maternal disorders based on the proportion of ���unwanted births��� was estimated from the data of various contraceptive-demand surveys. Tobacco���To estimate the burden attributable to tobacco, the method proposed by Peto and Lopez15 was used. Relative risks of death from lung cancer, upper aerodigestive cancers, other cancers, chronic obstructive pulmonary disease, cardiovascular diseases, and other medical causes were taken from the American Cancer Society Cancer Prevention Study, second round, a prospective study with follow-up in 1984���88. To correct for potential confounding of the estimated relative risks for smokers, the excess risk due to tobacco for all diseases other than lung cancer was halved.15 A smoking impact ratio defined as: C���N Smoking impact ratio=������ S���N where C is the observed lung-cancer rate in a given age-group of a population N is the non-smoker lung-cancer rate observed in the Cancer Prevention Study population and S is the smoker lung-cancer rate in the Cancer Prevention Study. The smoking impact ratio can be used as a surrogate for the prevalence of cumulative exposure in the attributable fraction formula. Because non-smoker lung-cancer rates are higher in China and other Asia and islands than in the USA,16 alternative non-smoker lung- cancer rates were used for these two regions. Preliminary results from a large case-control study in China were used to estimate the attributable fractions for China and other Asia and islands (Peto and Liu Boqu, personal communication). Deaths from tobacco-chewing among women in India were also estimated with attributable fractions reported by Notani and colleagues.17 Alcohol���The attributable burden of alcohol includes alcohol- related injuries and alcohol-related disease, and the protective effect of alcohol on ischaemic heart disease. The reference distribution for the burden of alcohol from all three components was no consumption. Attributable fractions for injuries in the established market economies were based on review of published data of the effects of alcohol on diseases and injuries,18 and for disease, attributable fractions were based on a meta-analysis.18 For other regions, attributable fractions were scaled to reflect different regional rates of consumption and estimated differences in drinking patterns for injuries, and consumption per person by country, cirrhosis death rates (excluding those attributed to hepatitis B), and deaths coded to alcohol dependence for disease. Large-scale prospective studies consistently show that, after correction for smoking, alcohol consumption exerts a protective effect on cardiovascular death at all levels of consumption.22���25 The protective effect of alcohol was estimated for each region from the relative risk of death from ischaemic heart disease and scaled estimates of the proportion of the population that abstains from drinking. Occupation���Leigh and colleagues8 used direct reports on occupation-related injuries in Scandinavia to arrive at minimum occupational injury rates for each region. The incidence rates were scaled, based on small-scale, published studies and other registration sources. For occupational diseases, data from reporting systems were available for the USA, Canada, Australia, Sweden, Denmark, the UK, Switzerland, Luxembourg, Hungary, Mexico, and China (selected causes only). For most of the working population in countries without registration systems, the reported rates from Canada and Australia were used to estimate occupational-disease death rates. Hypertension���Nichols and Elliott11 reviewed more than 50 population-based studies (including multicentre studies) to estimate the distributions of systolic and diastolic blood pressure by region, age, and sex. Relative risks of death for different blood pressures were estimated by logistic regression analysis of data from 18 studies. The reference distribution used was a systolic blood pressure of 110 mm Hg. Estimates of attributable burden due to hypertension are based on their attributable fractions for cerebrovascular and ischaemic heart disease. Burden from other causes due to hypertension was not estimated. Physical inactivity���Pratt and Koplan12 estimated relative risks specific for age and sex for ischaemic heart disease, colon cancer, and diabetes, based on a review of published studies. To control for confounding, the excess risk from inactivity was halved in developing regions. The prevalence of inactivity was estimated from a review of population-based surveys of physical activity for each region. The reference of exposure was a population in which 100% of individuals are regularly physically active. Illicit drugs���Donoghoe and colleagues7 developed attributable fractions for HIV, hepatitis B, perinatal disorders, protein-energy malnutrition, drug use, road-traffic accidents, poisonings, self- inflicted injuries, and violence, based on a review of published data about illicit drug use18 and, in the absence of local prevalence studies, on the estimated number of illicit drug users in each region. Air pollution���Hong and colleagues9 analysed the burden attributable to total suspended particulates and sulphur dioxide. Exposure levels in urban areas were available for all regions except the middle eastern crescent and sub-Saharan Africa. The reference distribution used was the WHO air quality guidelines. Vol 349 ��� May 17, 1997 1437 Risk factors Measure of exposure Reference distribution of exposure Time lag from exposure to burden Malnutrition Population with weight-for-age less than 2 SDs below average Population weight-for-age more Intermediate based on extensive national surveys than minus 2 SDs Poor water, sanitation, and hygiene Based on the theoretical faecal-oral route of transmission ���� Short Unsafe sex Based on the theoretical model of transmission of STDs and on ���� Short to long contraceptive-demand surveys for maternal disorders Alcohol* Disease Indexed on alcohol consumption, non-hepatitis B cirrhosis, and ���� Long alcohol-dependence syndrome Injury Indexed on estimate of consumption patterns based on ���� Short small-scale studies Occupation* Disease Registration data for developed regions��� and constant rates ���� Long for all other regions Injury Registration data for established market economies and constant ���� Short rates for all other regions Tobacco Indexed on lung cancer ���� Long Hypertension Population surveys of blood pressure Systolic blood pressure of 110 mm Hg Long Physical inactivity Population surveys of activity patterns Regular physical activity Long Illicit drugs Small-scale studies ���� Short to intermediate Air pollution Monitoring systems in urban areas for most regions WHO guidelines Short to long *Different methods and characteristics apply depending on whether burden arises from a disease or injury. ���Established market and formerly socialist economies of Europe, and Latin America and the Caribbean. Table 1: Summary of procedures used to estimate attributable burden from ten major risk factors