Improving the Comparative Quantification of Diseases, Injuries, and Risk Factors: The 2001 GBD Study
The 1990 GBD study represented a major advance in the quantification of the impact of diseases, injuries, and risk factors on population health globally and by region. Government and nongovernmental agencies alike have used its results to argue for more strategic allocations of health resources to disease prevention and control programs that are likely to yield the greatest gains in terms of population health. The results have also greatly increased understanding of the basic descriptive epidemiology of diseases and injuries worldwide.
Following publication of the initial results of the GBD study, several national applications of the methods it used have led to substantially more data on the descriptive epidemiology of diseases and injuries becoming available, as well as to improvements in analytical methods and mortality data in a number of countries. By emphasizing substantially more sophisticated approaches than in the past to the interpretation and presentation of population health data to policy makers, national burden of disease studies have stimulated efforts to improve and extend the collection of the health information data that are the basis for such analyses. A good example is the Islamic Republic of Iran where, over the last five years, the government has implemented a system of death registration with medical information on the cause of death that has been extended from four provinces initially to include 26, or almost all of the country's provinces. Another example is the government of Thailand's extensive verbal autopsy study aimed at addressing major coding deficiencies in Thailand's national mortality data (Choprapawon and others 2005).
Critiques of the original study's approach, particularly of the methods used to assess the severity weightings for disabling health states, have led to fundamental changes in the way that investigators incorporate health state valuations, that is, the use of population-based rather than expert opinion as used in the 1990 study, and to substantially better methods for improving the cross-national comparability of survey data on health status ( Murray, Tandon, and others 2002 ; Salomon and Murray 2004 ). Better methods for modeling the relationship between the level of mortality and the broad cause of death structure in populations that are based on proportions rather than rates have led to greater confidence in cause of death estimates for developing countries ( Salomon and Murray 2002 ). In addition, improved population surveillance for some major diseases such as HIV/AIDS, and the wider availability of data from verbal autopsy methods, particularly in Sub-Saharan Africa, have lessened the dependence on models for cause of death estimates, although substantial uncertainty in the use of such data persists. For more details on these and other methodological advances, see chapter 3 in this volume.
Perhaps the major methodological progress since the 1990 GBD study has been with respect to the quantification of the disease burden from risk factors. The initial study quantified the population health effects of 10 risk factors, but serious concerns exist about the comparability of the methods and estimates used. Different risk factors have different epidemiological traditions, particularly with regard to the definitions of hazardous exposure, the strength of the evidence on causality, and the availability of epidemiological research on exposure and hazard. As a result, comparability across estimates of the disease burden caused by different risk factors has been difficult to establish. In particular, much of classical risk factor research has treated exposures as dichotomous, with individuals either exposed or not exposed, with exposure defined according to an often arbitrary threshold value, for example, systolic blood pressure of 140 millimeters of mercury as the threshold for hypertension. Recent evidence for such continuous exposures as cholesterol, blood pressure, and body mass index suggests that such arbitrarily defined thresholds are inappropriate, because the hazards for these risks decline continuously across the entire range of measured exposure levels, with no obvious threshold ( Eastern Stroke and Coronary Heart Disease Collaborative Research Group 1998 ; Ezzati and others 2004 ; Rose 1985 ; WHO 2002 ).
For the 2001 GBD study, a new framework for risk factor assessment was defined that examines changes in the disease burden that would be expected under alternative population distributions of exposure to a risk factor or groups of risk factors ( Murray and Lopez 1999 ). Attributable fractions of disease due to a risk factor were then calculated based on a comparison of the disease burden expected under the current estimated distribution of exposure by age, sex, and region with that expected under a counterfactual distribution of exposure. One such counterfactual distribution was defined for each risk factor as the population distribution of exposure that would lead to the lowest levels of disease burden. Thus, for example, in the case of tobacco, this theoretical-minimum-risk counterfactual exposure would be 100 percent of the population being never-smokers, for overweight and obesity it would be a narrow distribution of body mass index centered around an optimal level of 21 kg/m 2 and so on. The distributions of the theoretical-minimum-risk exposure for the risk factors quantified in the World Health Organization's study of comparative risk assessment (the methodological and empirical basis for the 2001 GBD study) were developed by expert groups for each risk factor based on available scientific knowledge of risk factor hazard. The study also used systematic reviews and analyses of extant sources on risk factor exposure and hazard in an iterative process that increased comparability across risk factors ( Ezzati and others 2002 , 2004 ). These methods and results are described in more detail in chapter 4 in this volume.
Risk factors may affect disease and injury outcomes through other intermediate factors. For instance, some of the effects of diet and physical activity on cardiovascular diseases are mediated through changes in such intermediate factors as weight, blood pressure, and cholesterol. Risk factors may also affect disease and injury outcomes in combination with one another. For example, people who smoke and have elevated blood pressure and cholesterol have substantially higher probabilities of cardiovascular events. Finally, some risks have common social and behavioral determinants. For instance, members of poor households in rural areas are the most likely to be undernourished, use unsafe water sources, and be exposed to indoor smoke from solid fuels. Because of these epidemiological and social characteristics of risk factor exposure and hazard, policy-relevant analysis should include an assessment of the health benefits of simultaneous reductions in multiple risks. Multicausality also means that a range of interventions can be used for disease prevention, with the specific choices determined by such factors as costs, technology availability, infrastructure, and preferences. A novel aspect of the analysis of risk factors in the 2001 GBD study is the development and application of methods for estimating the disease burden attributable to the combined hazards of multiple risk factors ( Ezzati and others 2003 ).
The basic units of analysis in the 1990 GBD study were the eight World Bank regions defined for the World Bank's (1993) World Development Report 1993 . Designed to be geographically contiguous, these regions were nonetheless extremely heterogeneous with respect to health development, for example, the region referred to as Other Asia and Islands included countries with such diverse epidemiological profiles as Myanmar and Singapore. This seriously limited the applicability of these regions to comparative epidemiological assessments. Thus the 2001 GBD study followed a more refined approach. Estimates of overall mortality were first developed for World Health Organization member states using different methods for countries at different stages of health development. The choice of methods was largely determined by the availability of data ( Lopez and others 2002 ). Age- and sex-specific death rates for countries were essentially determined using one of three standard approaches: the use of routine life table methods for countries with complete vital registration; the application of standard demographic methods to correct for underregistration of deaths; or the application of model life tables where no vital registration or survey data on adult mortality were available ( Lopez and others 2002 ; Murray and others 2003 ).
The detailed methodological approaches adopted for estimating cause-specific mortality for countries and the descriptive epidemiology of nonfatal conditions for countries or subregions are described elsewhere (Mathers and others 2002; chapter 3 in this volume). This focus on individual countries as the unit of analysis, as well as the systematic application of standardized approaches for all countries in any given category of data availability, has vastly improved the cross-population comparability of disease and injury quantification.
A final major advance of the 2001 GBD study has been the systematic attempts to quantify some of the uncertainty in both national and global assessments of the disease burden (see chapter 5 in this volume). This uncertainty must be taken into account when making cross-national comparisons and needs to be carefully communicated to and interpreted by epidemiologists and policy makers alike.
