Life expectancy disparities in the US, measured in relation to individual and community characteristics, are enormous by international standards. The Eight Americas encompass a large part of US life expectancy disparities, while forming easily identifiable subgroups of the US population. We have shown that a small number of preventable risk factors such as smoking, high blood pressure, elevated blood glucose, and adiposity are the leading risk factors for mortality in the US 
. The results from the current analysis suggest that these risk factors also contribute to the mortality disparities across the Eight Americas, especially for CVD and cancers. Therefore, had these risk factors been reduced to their optimal levels or even to the commonly used guidelines, there would be both aggregate health benefits and a considerable decline in life expectancy disparities. Our conclusions on the role of these risk factors on life expectancy disparities across the Eight Americas were not sensitive to the specific disparity metric used (for a discussion of disparity metrics see 
Analyses of disease-specific probabilities of death identified injuries, HIV/AIDS (especially for men), and selected noncommunicable diseases as those that accounted for disparities that remained after risk factors were reduced to optimal levels. Once we removed deaths from homicide, road traffic injuries, and HIV/AIDS in addition to the effects of risk factors, life expectancy further improved with larger benefits to the Americas that currently have lower life expectancy, especially for men. For example, reducing risk factors to their optimal levels and removing these three medical causes of death increased male life expectancy in Americas 5, 7, and 8 by 7.3–9.4 y (compared with 4.6–6.7 y when only risk factors are reduced). Yet even CVD mortality had a residual gradient, albeit substantially smaller, after four of its most salient risk factors were removed, with a clear survival advantage among Asians ( and ). The reasons for this residual advantage may be risk factors not included in our analysis (e.g., lower lipids as a result of dietary composition or use of statins, psychosocial factors, etc.) or disparities in health care access and quality of care. These factors could not be estimated in the BRFSS, but should be the subject of future data collection and research. Further, the benefits of reducing current exposure occurs over time and requires additional analysis of time-dependent effects. Most of the benefits nonetheless seem to occur within about 5 y for cardiovascular risk factors 
; even for the effects of smoking on cancers and chronic respiratory diseases, which have longer periods of risk reversibility, 75% or more of the benefits of cessation occur by about 15 y 
Our results on the effect of multiple preventable risk factors on life expectancy disparities at the national level are supported by those from analyses in specific cohorts, which were not nationally representative. For example, in the Atherosclerosis Risk in Communities (ARIC) Study, blacks and whites had nearly identical CVD incidence rates after adjustment for smoking, blood pressure, cholesterol and glucose 
. Analyses of the Multiple Risk Factor Intervention Trial found that adjustment for major CVD risk factors reduced the differences in CVD mortality although there was a statistically significant remaining difference 
. A recent reanalysis of the Whitehall follow-up study found that interventions for the same risk factors as the ARIC study were expected to reduce coronary heart disease mortality differentials between occupational classes by 86% 
. Finally, the Korean National Health and Nutrition Examination Survey follow-up study also found that absolute socioeconomic mortality inequalities could be substantially reduced if either behavioral (smoking, alcohol use, and physical inactivity) or metabolic risks (blood pressure, fasting serum glucose, and serum total cholesterol) were improved 
. Some of these studies found larger effects of risk factors on disparities that those in our analysis, possibly due to the inclusion of other risks (e.g., lipids, physical inactivity) and indicators such as income and education that either directly or through other metabolic, dietary, and lifestyle factors affect mortality disparities. National analysis in the US estimated that 58% of disparities in total mortality among young and middle-age men was due to smoking; there also seems to be some effect of risk factors on disparities in self-reported disease diagnoses and health status (noting that self-reported health status is commonly measured with bias, error, and incomparability) 
Beyond its innovation in assessing the effects of modifiable risk factors on the national US life expectancy disparities, our analysis has several strengths. Our PAF calculations incorporated multicausality and mediated effects, with parameters from systematic or comprehensive reviews of epidemiologic studies. We calculated PAFs for multiple risk factors using individual level exposure data, thereby eliminating the need to parameterize the joint distribution of risk factors and make strong assumptions about the shape of the distributions and their correlations. Our outcome variables were life expectancy and probabilities of death, which incorporate competing risks from other diseases using life table methods. Further, life tables were estimated separately for the Eight Americas and by sex because they have distinct patterns of competing risks. Finally, we quantified the uncertainty as a result of the sampling variability in exposure, RRs, mediated effects of BMI, and disease-specific mortality rates.
Population-level analyses like ours also have limitations. First, the BRFSS does not provide data or indicators of sufficient detail and quality on alcohol use, blood lipids, relevant dietary risk factors (e.g., salt and various fats), and physical activity in the Eight Americas. Therefore, these risk factors could not be included in our analysis, even though they may have significant variation by race and/or region 
. Using NHANES 2003–2006 data at the national level, the combined effects of LDL cholesterol and the four risk factors in our analysis on life expectancy would be 0.1 y higher for men and women than that of these four risk factors alone. The difference between the effects of the two clusters of risk factors is small despite the fact that LDL cholesterol is an important risk factor for CVD mortality; rather, because of multicausality, the combined PAF for the effects of multiple risk factors grows by progressively smaller amounts with each additional risk factor, even when its individual effect is relatively large. Further, mean age-standardized LDL cholesterol among blacks was only 1 mg/dl (<1%) higher than whites for men; black women had lower cholesterol than whites by 7 mg/dl (~6%), indicating that its contribution to disparities may be modest across groups other than Asians. Harmful alcohol use is an important risk factor for injuries and diseases such as cirrhosis, which had substantial disparity in the Eight Americas 
The second limitation of our analysis arises from the fact that the possibility of effect size modification by race cannot be ruled out, even though the current evidence is generally consistent with RRs being similar by race. RR differences may be especially relevant for smoking, for which factors such as smoking intensity and duration of smoking or smoking cessation may influence RRs. Third, because BRFSS does not measure SBP, FPG, weight, and height, we applied validated statistical models to NHANES and BRFSS data to predict these variables and correct for bias in self-reported data 
. While this is an innovative use of multiple data sources for subnational risk factor measurement, with relatively high predictive power, it could use only those predictors that were measured in NHANES and BRFSS using consistent or comparable questionnaires. There was unexplained variation in the model that could result in underestimating exposure variability across the Eight Americas 
. Hence our results should be considered conservative estimates of the effects of risk factors on mortality disparities. Further, these prediction models result in additional uncertainty beyond sampling error, making the reported uncertainty intervals in an underestimate of the true uncertainty. Fourth, the combined effects of the four risk factors included in our analysis may follow a model different fromthat presented in Text S1
. For example, a part of the effect of smoking on cardiovascular diseases may be mediated through blood pressure and/or glucose. A sensitivity analysis showed that 25% of the effects of smoking on cardiovascular outcomes were mediated through these factors, our estimated effects on lifeexpectancy levels and disparities would change by less than 0.06 y.
A key feature of our analysis is using the Eight Americas, which are based on race, and on county location and socioeconomic characteristics. As discussed in previous work 
, using county and race–county combinations as units of analysis has allowed us to examine mortality disparities in consistent, comparable, and easily identifiable units over decades, but does not capture within-county variations in risk factor exposure and mortality. Finally, we could not include individuals with Hispanic ethnicity as a separate America. Previous analyses have shown that Hispanic ethnicity is significantly under-recorded on death certificates, leading to implausibly high life expectancies when combined with population estimates from census using self-reported Hispanic ethnicity 
. Future analyses should attempt to adjust for this mortality undercount, or conduct analyses for Hispanics in regions where mortality undercount is likely to be small, e.g., in states with large Hispanic population 
Our results demonstrate that a small number of risk factors for chronic diseases account for a noticeable part of the disparities in life expectancy in the US, with the largest contributions from smoking and high blood pressure. These disparity effects influence young and middle-aged adults, as well as older adults, with the largest effects on CVD, diabetes, and some cancers. The report of the WHO Commission on Social Determinants of Health has called attention to distribution of money, power, and resources as the underlying sources of health disparities, but has also emphasized the need to focus on common risk factors for chronic diseases with known and effective interventions 
. Similarly the most recent House of Commons Health Committee Report in the UK identified three groups of causes for health inequalities: access to health care, socioeconomic factors, and lifestyle factors 
. The essential question is therefore how to use disease prevention to improve health and reduce health disparities together with policies that aim to reduce socioeconomic disparities, reform health care, and improve quality of care.
An essential first step in achieving the aggregate and disparity promises of prevention is to discard a dominant view in the US that behavioral, dietary, and metabolic risk factors are either personal choices and responsibilities or are in the domain of clinical practice and hence only a subject of health education and physician advice for individuals. Rather, we must identify, implement, and rigorously evaluate effective population-based and personal interventions that can reduce these preventable risk factors or mitigate their effects on disease outcomes (see, for example, the reviews commissioned by the Robert Wood Johnson Foundation on disparities in CVD and diabetes 
). Few or no current interventions have been effective in reducing overweight and obesity at the population level, emphasizing the need to develop and test new creative and ambitious interventions. Diabetes prevention through lifestyle and pharmacological interventions has been efficacious in trials 
but should be evaluated in community settings. Smoking and blood pressure both have efficacious and cost-effective interventions, and have successfully been lowered in the adult US population as a whole for decades. These interventions need to reach population subgroups and counties where smoking and blood pressure remain high. Salt intake is an important predictor of population blood pressure 
, and regulating and reducing salt in prepared and packaged food is an effective population-level intervention 
; screening and use of antihypertensives or combination therapy to reduce blood pressure and cardiovascular risk is also cost-effective 
and should be scaled up as a part of expanding and improving primary care in the context of US health reform. A recent systematic review of smoking interventions hypothesized that population-level interventions have the potential to reduce disparities in smoking 
. Yet in practice, US risk factor trends have at best had a mixed performance in terms of reducing exposure disparities 
. Therefore, both national versus local and aggregate versus disparity effects should determine the design, implementation, and evaluation of policies and programs that aim to reduce risk factor exposure.