Risk Factor Selection
The CRA project included a selected group of risk factors, presented in table 4.1. The criteria for selecting risk factors included the following:
-
they were likely to be among the leading causes of the disease burden globally or regionally;
-
they were not too specific, for example, every one of the hundreds of air pollutants or fruits and vegetables, or too broad, such as the environment or diet taken as a single exposure;
-
the likelihood of causality was high based on collective scientific knowledge;
-
reasonably complete data on exposure and risk levels were available or sufficient data were available to extrapolate information when necessary;
-
they were potentially modifiable.
[Table .]
The risks to health examined in the CRA project cover many of the important hazards to health addressed in various fields of scientific inquiry. Arguably, hundreds of risk exposures are harmful to health. We selected only a relatively small number of exposures for quantification, largely determined by the availability of data and scientific research about their level and health effects in different parts of the world.
We also had to make choices about the definition of each risk factor. Given the close interrelationships among diet, exercise, and physiological risks on the one hand, or among water, sanitation, and personal hygiene on the other, the exact definition of what a risk factor is requires careful attention. The absence of a particular risk factor like dietary fat intake from table 4.1 does not imply that it is of limited relevance. Similarly, the assessment of unsafe sex separately from that of non-use and use of ineffective methods of contraception does not override their close linkages. Rather, we focused the analysis on risk factors for which we were likely to be able to satisfactorily quantify their population exposure distributions and health effects using existing scientific evidence and available data and for which intervention strategies are available or might be envisioned.
Estimating Population Attributable Fractions
The contribution of a risk factor to disease or mortality is expressed as the fraction of disease or death attributable to the risk factor in a population and is referred to as the population attributable fraction (PAF), and is given by the generalized potential impact fraction in equation 4.1 (Eide and Heuch 2001; Walter 1980).
where RR(x) is the relative risk at exposure level x, P(x) is the population distribution of exposure, P (x) is the counterfactual distribution of exposure, and m is the maximum exposure level.
The corresponding relationship when exposure is described as a discrete variable with n levels is given by
PAFs obtained in this way estimate the proportional reduction in disease or death that would occur if exposure to the risk factor were reduced to the counterfactual distribution. The alternative (counterfactual) scenario used is the exposure distribution that would result in the lowest population risk, referred to as the theoretical-minimum-risk exposure distribution (Ezzati and others 2002; Murray and Lopez 1999). For risk factors for which the assumption of constant relative risk was not appropriate, we estimated PAFs by accounting for the determinants of hazard heterogeneity. For example, the PAFs for injuries as a result of alcohol use accounted for alcohol drinking patterns (moderate versus binge).
Because most diseases are caused by multiple risk factors, PAFs for individual risk factors for the same disease overlap and can add to more than 100 percent (Murray and Lopez 1999; Rothman 1976). For example, some deaths from childhood pneumonia may have been avoided by preventing exposure to indoor smoke from household use of solid fuels, childhood underweight, and zinc deficiency (which itself affects weight-for-age); and some cardiovascular disease events may be due to a combination of smoking, physical inactivity, and low fruit and vegetable intake. Such cases would be attributed to all these risk factors.
Attributable Mortality and Burden of Disease
For each risk factor and disease pair, we calculated PAFs for each age and sex group, and in each region, using the relationships in equations 4.1a and 4.1b, separately for mortality (PAFM) and incidence (PAFI) when the relative risks for mortality and incidence were different. For each of these age, sex, and region groups, we obtained estimates of mortality (AMij) and the burden of disease (ABij) from disease j attributable to risk factor i as follows:
where YLL denotes years of life lost because of premature mortality and YLD denotes years of healthy life lost as a result of disability.
Data on Exposure and Hazard
Between 1999 and 2002,for each risk factor, an expert working group conducted a comprehensive review of the published literature and other sources (government reports, international databases, and so on) to obtain data on the prevalence of risk factor exposure and hazard size (relative risk or absolute hazard size when appropriate, such as the effects of lead on blood pressure) (Ezzati and others 2004). The work included collecting primary data and undertaking a number of reanalyses of original data, systematic reviews, and meta-analyses. To increase comparability while acknowledging the fundamental differences in exposure and hazard quantification across risk factors, the criteria for using the scientific evidence included consistency of exposure variables used in exposure data sources with those used in epidemiological studies on hazard, population representativeness of exposure data, and study design for estimating the magnitude of hazardous effects (including minimizing the effects of confounders).
Data were initially presented separately for males and females and broken down into eight age groups (0-4, 5-14, 15-29, 30-44, 45-59, 60-69, 70-79, and 80 years old and older) and the 14 epidemiological subregions of the Global Burden of Disease (GBD) study (see chapter 3), which are based on a combination of World Health Organization regions and child and adult mortality levels, as described in the annexes of the annual World Health Report 2002 (WHO 2002). Data sources, models, and assumptions used to extrapolate exposure or relative risk across countries or regions are described in detail in chapters devoted to individual risk factors elsewhere (Ezzati and others 2004). External reviewers anonymously peer reviewed each risk factor chapter, including conducting re-reviews as appropriate.
In this reanalysis, estimates of mortality and disease burden attributable to risk factors were needed in World Bank regions (see map 1 inside the front cover). For six risk factors (childhood underweight, high blood pressure, high cholesterol, overweight and obesity, smoking, and indoor smoke from household use of solid fuels), country-level data were available and allowed reestimating exposure directly for World Bank regions. In such cases, we used newly available data sources on exposure to update CRA project estimations. For seven risk factors (unsafe water, sanitation, and hygiene; zinc deficiency; vitamin A deficiency; iron deficiency anemia; physical inactivity; low fruit and vegetable intake; and child sexual abuse), we estimated exposure in World Bank regions from the 14 GBD subregions using population-weighted averages. For another five risk factors (unsafe sex, urban air pollution, illicit drug use, non-use and use of ineffective methods of contraception, and contaminated injections in health care settings), where both exposure and hazards change across populations, we converted PAFs from GBD subregions to World Bank regions, with PAFs weighted by age-, sex-, and disease-specific mortality rates. The prevalence of alcohol use was converted from GBD subregions to World Bank regions and was used to estimate exposure and PAFs in World Bank regions for most disease outcomes, because relative risks did not vary across populations. For all injury outcomes, ischemic heart disease, depression, stroke, and diabetes, whose hazards varied across regions, PAFs were converted from GBD subregions to World Bank regions using mortality weighting.
Theoretical-Minimum-Risk Exposure Distributions
The theoretical-minimum-risk exposure distribution was zero for risk factors for which zero exposure could be defined and reflected minimum risk, such as no smoking. For some risk factors, zero exposure was an inappropriate choice, either because it is physiologically impossible, as in the case of body mass index (BMI) or high cholesterol, or because physical lower limits to exposure reduction exist, as for concentrations of ambient particulate matter. For the latter risk factors, we used the lowest levels observed in specific populations and epidemiological studies to choose the theoretical-minimum-risk exposure distribution. For example, counterfactual exposure distributions of 115 mmHg for systolic blood pressure and 3.8 mmol/L for total cholesterol, each with a small standard deviation, are the lowest levels at which meta-analyses of cohort studies have characterized dose-response relationships (Chen and others 1991; Eastern Stroke and Coronary Heart Disease Collaborative Research Group 1998; Law, Wald, and Thompson 1994).
Alcohol has benefits as well as causing harm for different diseases depending on the disease and on patterns of alcohol consumption (Corrao and others 2000; Puddey and others 1999). Rehm and others (2004) chose a counterfactual of zero for alcohol use. This was because despite its benefits for cardiovascular diseases in some populations, the global and regional burden of disease due to alcohol use was dominated by its impacts on neuropsychiatric diseases and injuries that are considerably larger than these benefits.
Finally, for factors with protective effects, namely, fruit and vegetable intake and physical activity, we chose a counterfactual exposure distribution based on a combination of levels observed in high-intake populations and the level to which the benefits may continue given current scientific evidence. Table 4.1 reports the theoretical-minimum-risk exposure distributions for the risk factors.
