Global, regional, and national comparative risk assessment of 79 behavioural, environmental and occupational, and metabolic risks or clusters of risks, 1990–2015: a systematic analysis for the Global Burden of Disease Study 2015

1Evidence before this studyThe most recent assessment of attributable deaths and disability-adjusted life-years (DALYs) at the global, regional, and national level was the Global Burden of Diseases, Injuries, and Risk Factors Study 2013, which covered 79 risk factors or combinations of risks from 1990 to 2013 in 188 countries.Added value of this studyThis study (the Global Burden of Diseases, Injuries, and Risk Factors Study 2015) incorporates recently published studies, newly acquired data for exposure to relative risks, and new risk-outcome pairs meeting study inclusion criteria. To enhance transparency of the supporting evidence, we provided an assessment of the strength of evidence supporting causality for all 388 risk-outcome pairs. For the first time, we separately assessed trends in risk exposure by computing a summary exposure value, which allows comparisons over time and across place for dichotomous, polytomous, and continuous risks. Quantification of exposure trends allowed decomposition of trends in attributable DALYs into the portion contributed by changes in population growth, population structure, exposure, and risk-deleted DALY rates. We found that reductions in exposure have been key drivers of change for only a small set of environmental risks, including sanitation, household air pollution, and behavioural risks (eg, undernutrition and smoking). For many risks, trends in attributable DALYs have been driven by the interplay between population growth, ageing, and declines in risk-deleted DALY rates. For some risks, including body-mass index, fasting plasma glucose, occupational exposure to carcinogens, and drug use, exposure is increasing and driving up attributable burden. Although an average risk transition has occurred as countries move through the development continuum, many risks initially increase and then decline at the highest development levels. We document leading risks for each country and territory included in the study.Implications of all the available evidenceRisk assessments allow identification of several groups of risk factors that deserve policy attention. Risks such as smoking, unsafe sanitation, and childhood undernutrition still cause many attributable DALYs, but recent trends show that exposure can be reduced. This assessment of risk also shows many large global risks for which changes in exposure are slow, such as high systolic blood pressure, ambient air pollution, diets high in sodium, high cholesterol, and alcohol intake, highlighting huge opportunities for intervention. Two large risks—high BMI and high fasting plasma glucose—have particularly large and concerning increases in exposure.

2Evidence before this study

3The most recent assessment of attributable deaths and disability-adjusted life-years (DALYs) at the global, regional, and national level was the Global Burden of Diseases, Injuries, and Risk Factors Study 2013, which covered 79 risk factors or combinations of risks from 1990 to 2013 in 188 countries.

4Added value of this study

5This study (the Global Burden of Diseases, Injuries, and Risk Factors Study 2015) incorporates recently published studies, newly acquired data for exposure to relative risks, and new risk-outcome pairs meeting study inclusion criteria. To enhance transparency of the supporting evidence, we provided an assessment of the strength of evidence supporting causality for all 388 risk-outcome pairs. For the first time, we separately assessed trends in risk exposure by computing a summary exposure value, which allows comparisons over time and across place for dichotomous, polytomous, and continuous risks. Quantification of exposure trends allowed decomposition of trends in attributable DALYs into the portion contributed by changes in population growth, population structure, exposure, and risk-deleted DALY rates. We found that reductions in exposure have been key drivers of change for only a small set of environmental risks, including sanitation, household air pollution, and behavioural risks (eg, undernutrition and smoking). For many risks, trends in attributable DALYs have been driven by the interplay between population growth, ageing, and declines in risk-deleted DALY rates. For some risks, including body-mass index, fasting plasma glucose, occupational exposure to carcinogens, and drug use, exposure is increasing and driving up attributable burden. Although an average risk transition has occurred as countries move through the development continuum, many risks initially increase and then decline at the highest development levels. We document leading risks for each country and territory included in the study.

6Implications of all the available evidence

7Risk assessments allow identification of several groups of risk factors that deserve policy attention. Risks such as smoking, unsafe sanitation, and childhood undernutrition still cause many attributable DALYs, but recent trends show that exposure can be reduced. This assessment of risk also shows many large global risks for which changes in exposure are slow, such as high systolic blood pressure, ambient air pollution, diets high in sodium, high cholesterol, and alcohol intake, highlighting huge opportunities for intervention. Two large risks—high BMI and high fasting plasma glucose—have particularly large and concerning increases in exposure.

8Analysis of the causes of poor health—specifically, the connections between risk factors and development of poor health—can provide insights into opportunities and priorities for prevention, research, policy, and development. One of the mainstays of modern epidemiology is quantification of elevated risks for particular diseases or injuries from exposure to a given risk factor for groups of individuals. Quantification of elevated risk for exposed groups of individuals from an array of risk-outcome pairs is important to inform decision making on individual health; however, public policy debates require the comprehensive metric of population-level risk, which is a function of elevated risk in the exposed population and the fraction of the population exposed to a given risk. Efforts to measure population risk have combined data for excess risk with the number of individuals exposed to provide comparative quantification of different health risks for populations that have been influential in establishment of policy priorities.,

9The comparative risk assessment (CRA) approach developed for the Global Burden of Diseases, Injuries, and Risk Factors (GBD) Study, provides an overarching conceptual framework for population risk assessment across risks and over time. The scale of the GBD Study required extensive work to develop exposure metrics, assess relationships, and compile health data from different parts of the world with differing levels of metadata and uncertainty, and the unique contribution of this work has been broadly recognised.,, A robust debate on specific risks and results emerged after publication of the Global Burden of Diseases, Injuries, and Risk Factors Study 2013 (GBD 2013). Inclusion and exclusion of particular risks and outcomes;,, the optimum targets for indicators such as high systolic blood pressure,, cholesterol,, diets high in sodium, and air pollution;, and the certainty of some dietary components of risk, were challenged, in addition to some details of methods. Underlying many of these discussions were heterogeneities in the strength of causal evidence for different risk-outcome pairs.

10The Global Burden of Diseases, Injuries, and Risk Factors Study 2015 (GBD 2015) CRA, in addition to updating data and methods, adds new transparency about the evidence supporting causal connections for each of the 388 risk-outcome pairs included in the analysis, allows the quantification and reporting of levels and trends in exposure, decomposes changes in attributable burden into population growth, ageing, risk exposure, and risk-deleted disability-adjusted life-year (DALY) rates, and examines how risks change with development. As with all iterations of the GBD Study, GBD 2015 results presented here supersede all previously published GBD CRA estimates.

11The CRA conceptual framework was developed by Murray and Lopez, who established a causal web of hierarchically organised risks or causes that contribute to health outcomes (methods appendix p 161), which allows quantification of risks or causes at any level in the framework. In GBD 2015, as in previous iterations of the GBD Study, we evaluated a set of behavioural, environmental and occupational, and metabolic risks, where risk-outcome pairs were included based on evidence rules (methods appendix p 161). These risks were organised into four hierarchical levels, described in table 1. To date, we have not quantified the contribution of other classes of risk factors (methods appendix p 161); however, using an analysis of the relationship between risk exposures and development, measured with use of the Socio-demographic Index (SDI), we provide some insights into the potential magnitude of distal social, cultural, and economic factors.

12Two types of risk assessments are possible within the CRA framework: attributable burden and avoidable burden. Attributable burden is the reduction in current disease burden that would have been possible if past population exposure had shifted to an alternative or counterfactual distribution of risk exposure. Avoidable burden is the potential reduction in future disease burden that could be achieved by changing the current distribution of exposure to a counterfactual distribution of exposure. Murray and Lopez identified four types of counterfactual exposure distributions: theoretical, plausible, feasible, and cost-effective minimum risk. In GBD studies to date and in this study, we focus on attributable burden using the theoretical minimum risk level (TMREL), which is the level of risk exposure that minimises risk at the population level, or the level of risk that captures the maximum attributable burden.

13Overall, this analysis follows the CRA methods used in GBD 2013. The methods described in this study provide a high-level overview of the analytical logic, with a focus on areas of notable change from the methods used in GBD 2013, with details provided in the methods appendix. This study complies with the Guidelines for Accurate and Transparent Health Estimates Reporting statement (methods appendix pp 177–79).

14In the GBD framework, geographies have been arranged as a set of hierarchical categories: seven super-regions, 21 regions nested within the seven super-regions, and 195 countries and territories nested in the 21 regions. Additionally, GBD collaborator interest and availability of data resulted in an expansion of countries for which we disaggregate our estimates at the subnational level. At the first level of subnational division, 256 geographic units are included in GBD 2015. For this study, we present results for the 195 national and territory-level geographies. We produced a complete set of age-specific, sex-specific, cause-specific, and location-specific estimates of risk factor exposure and attributable burden for 1990, 1995, 2000, 2005, 2010, and 2015 for included risk factors. Results presented in this study emphasise results for 1990, 2005, and 2015; online data visualisations provide access to results for all GBD metrics from 1990 to 2015.

15Four key components are included in estimation of the burden attributable to a given risk factor: the metric of burden being assessed (number of deaths, years of life lost [YLLs], years lived with disability [YLDs], or DALYs [the sum of YLLs and YLDs]), the exposure levels for a risk factor, the relative risk of a given outcome due to exposure, and the counterfactual level of risk factor exposure. Estimates of attributable DALYs for a risk-outcome pair are equal to DALYs for the outcome multiplied by the population attributable fraction (PAF) for the risk-outcome pair for a given age, sex, location, and year. A similar logic applies for estimation of attributable deaths, YLLs, or YLDs. Risks are categorised on the basis of how exposure was measured: dichotomous, polytomous, and continuous. The PAF represents the proportion of risk that would be reduced in a given year if the exposure to a risk factor in the past were reduced to a counterfactual level of exposure (methods appendix p 27).

16In this study, as in GBD 2013, we have included risk-outcome pairs that we have assessed as meeting the World Cancer Research Fund grades of convincing or probable evidence (methods appendix pp 12–13 contains definitions of these grades). Table 2 provides a summary of the evidence supporting a causal relationship between a risk and an outcome for each pair included in GBD 2015. For each risk-outcome pair, we used recent systematic reviews to identify independent prospective studies (randomised controlled trials, non-randomised interventions, and cohorts) that evaluated the putative relationship. For risk-outcome pairs for which no recent systematic review was available, we either updated reviews developed for GBD 2013 or did a new systematic search of literature (methods appendix pp 44–159). Table 2 summarises the evidence using multiple dimensions, which supports our assessment that each included risk-outcome pair meets the criteria of convincing or probable evidence (methods appendix [pp 12–13] contains a justification of the criteria presented to support causality). In this summary of evidence, we have focused on randomised controlled trials and prospective observational studies, along with supporting evidence, like dose-response relationships and biologically plausible mechanisms. Other evidence supporting causal connections, such as case-control studies, are not summarised in table 2.

17Information about the data sources, estimation methods, computational tools, and statistical analysis used in derivation of our estimates are provided in the methods appendix. The analytical steps for estimation of burden attributable to single or clusters of risk-outcome pairs are summarised in the methods appendix (p 162). Table 1 provides definitions of exposure for each risk factor, the TMREL used, and metrics of data availability. For each risk, we estimated effect size as a function of age and sex and exposure level, mean exposure, the distribution of exposure across individuals, and the TMREL. The approach taken is largely similar to GBD 2013 for each quantity for each risk. Some methodological improvements have been implemented and new data sources incorporated. The methods appendix (pp 44–159) provides details of each step by risk. Citation information for the data sources used for relative risks are provided in searchable form through a web tool. We estimate the joint effects of combinations of risk factors using the same methods as GBD 2013, namely using published studies to estimate the fraction of a risk that was mediated through the other risk (methods appendix pp 28–35). Relative risks by age and sex for each risk factor and outcome pair are provided in the methods appendix (pp 215–44).

18All point estimates are reported with 95% uncertainty intervals (UIs). UIs include uncertainty from each relevant component, consisting of exposure, relative risks, TMREL, and burden rates. Where percentage change is reported (with 95% UIs), we computed it on the basis of the point estimates being compared. In this study, we provide further methodological detail on new extensions to the CRA analysis.

19In previous GBD studies, we did not report comparable exposure metrics for the risk factors included because of the complexity of quantification of polytomous and continuous risks. Because of substantial interest in the trends in exposure, we developed a summary measure of exposure for each risk. This measure, called the summary exposure value (SEV), is the relative risk-weighted prevalence of exposure. Formally, it is defined as:

20where Pri is prevalence of category i exposure, RRi is relative risk of the category i, and RRmax is the maximum relative risk observed (between categories). This quantity is estimated for each age, sex, location, year, and outcome. For each risk factor, a single SEV is estimated by averaging of the outcome of specific SEV values for each age, sex, location, and year across outcomes. In the case of dichotomous exposure, SEV is equal to prevalence. For continuous risks:

21where P(x) is the density of exposure at level x of exposure, RR(x) is relative risk of the level x, and RRmax is the highest relative risk that is supported by data and reflects a level where more than 1% of the global population are exposed to that level or a higher risk.

22SEV takes the value zero when no excess risk for a population exists and the value one when the population is at the highest level of risk; we report SEV on a scale from 0% to 100% to emphasise that it is risk-weighted prevalence. We computed as the level for exposure with the highest relative risk supported by cohort or trial data and for which at least 1% or more of the global population is exposed. For comparison purposes, we have also computed age-standardised SEVs for every risk factor from the most detailed level using the GBD population standard.

23We did two related decomposition analyses of changes in DALYs from 1990 to 2015: decomposing changes in cause-specific DALYs due to changes in population growth, population age structure, exposure to all risks for a disease, and risk-deleted death and DALY rates; and decomposing changes in risk-attributable all-cause DALYs due to changes in population growth, population age structure, risk exposure to the single risk factor, and risk-deleted DALY rates. Risk-deleted rates are the rates after removal of the effect of a risk factor or combination of risk factors; in other words, observed DALY rates multiplied by one minus the PAF for the risk or set of risks. Our decomposition analyses draw from methods developed by Das Gupta to provide a computationally tractable solution to estimate the contribution of multiple factors to an outcome (methods appendix pp 36–37). For some risks where the PAF is 100%, such as fasting plasma glucose and diabetes, the methods have had to be further adapted. We were not able to include three outcomes in this analysis: cervical cancer, sexually transmitted diseases, and HIV/AIDS.

24We examined how changes in risk exposure were related to changes along the development spectrum. Drawing from methods used to construct the Human Development Index, we constructed the SDI, a summary measure of overall development based on estimates of lag-dependent income per capita, average educational attainment over the age of 15 years, and total fertility rate. In the SDI, we weighted each component equally and rescaled them from zero (for the lowest value observed during 1980–2015) to one (for the highest value observed) for income per capita and average years of schooling, and the reverse for the total fertility rate. We computed the final SDI score as the geometric mean of each of the components. For each risk, we calculated the average relationship between risk exposure, as measured by SEV, and SDI across all geography years by age and sex using spline regression (methods appendix pp 38–39). We then used this relationship to characterise how exposures to risk vary on the basis of SDI alone.

25The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The authors had full access to all the data in the study and had final responsibility to submit for publication.

26The SEV is a single, interpretable measure, which captures risk-weighted exposure for a population, or risk-weighted prevalence of an exposure. The scale for SEV spans 0% to 100%, such that an SEV of 0% reflects no risk exposure in a population and 100% indicates that an entire population has maximum possible risk. A decline in SEV indicates reduced exposure to a given risk factor, whereas an increase in SEV indicates increased exposure. Table 3 provides age-standardised SEVs for 61 risks at the global level, by sex, for 1990, 2005, and 2015 (results appendix pp 3619–4070 contains results for every geography). From 1990 to 2015, SEVs decreased by more than 30% for four risks: unsafe sanitation (38·3% [95% UI 36·1–40·5]), childhood underweight (34·2% [30·9–37·9]), childhood stunting (33·4% [30·3–37·4]), and household air pollution (30·2% [26·9–33·2]). The global SEV for smoking also decreased by 2015, decreasing by 27·5% (23·2–30·9) for men and 28·7% (20·2–34·1) for women; notably, smoking exposure among men still far exceeded that for women in 2015. Significant, although more moderate than for smoking reductions in global SEVs for both sexes occurred for second-hand smoke (12·2% [9·4–15·1]), unsafe water (9·4% [5·3–13·0]), and diet high in red meat (9·0% [7·6–10·3]) from 1990 to 2015. Risk exposure for high total cholesterol significantly declined for both men and women during this time, although this decrease was smaller among men (3·2% [2·2–4·4]) than among women (5·6% [4·6–6·7]). For a subset of occupational risk factors, such as ergonomic factors and asthmagens, global SEVs were reduced from 1990 to 2015.

27For a subset of risks, minimal changes in exposure occurred between 1990 and 2015. This finding was particularly evident among various dietary risks (eg, diet low in fruits) and behaviours related to nutrition (eg, non-exclusive and discontinued breastfeeding). Discordant trends emerged by sex for some risk factors, such as low physical activity, where global SEVs for men increased by 2·4% (95% UI 1·8–2·9), whereas the SEV for women declined by 1·5% (1·0–2·0). Global SEVs significantly increased for 27 risk factors for both sexes combined from 1990 to 2015; significant increases occurred for 24 risks for men alone and 23 risks for women alone. We recorded the most pronounced rises for various occupational exposures, such as diesel engine exhaust, silica, and benzene. Global SEVs for high body-mass index (BMI) increased by 38·7% (29·9–55·6) for men and 34·4% (27·7–45·7) for women. For both sexes, other risks with large increases included drug use (30·2% [23·3–39·1]), ambient ozone pollution (24·6% [15·0–31·9]), and high fasting plasma glucose (23·8% [22·4–25·4]).

28The proportion of deaths, YLLs, YLDs, and DALYs that could be jointly attributable to all risk factors combined differed by cause group and measure of health (table 4). Globally, 57·8% (95% UI 56·6–58·8) of deaths, 48·4% (47·4–49·3) of YLLs, 26·1% (25·0–27·1) of YLDs, and 41·2% (39·8–42·8) of DALYs could be attributed to the risk factors currently assessed as part of GBD 2015. Across health outcomes, attributable DALYs were highest for non-communicable diseases (NCDs), although the percentage of attributable burden ranged from 20·9% (19·7–22·0) for YLDs to 64·8% (63·3–66·2) for deaths in 2015. Among NCD cause groups, attributable DALYs were as high as 85·3% (84·0–86·6) for cardiovascular and circulatory diseases compared with low attributable DALYs, even among leading causes of disease burden (ie, 16·0% [13·9–18·2] for musculoskeletal disorders and 22·5% [20·1–25·4] for mental and substance use disorders). In 2015, approximately 40–60% of DALYs due to cancers, cirrhosis, and chronic respiratory diseases could be attributed to risk factors assessed in this study. Except for YLDs, less than 50% of disease burden for Group 1 causes—communicable, maternal, neonatal, and nutritional diseases—could be attributed to analysed risk factors. Risk factors accounted for less than 20% of early death and disability from maternal disorders (eg, 10·2% [4·1–16·8] of DALYs).

29Categories of risk factors—metabolic, environmental or occupational, and behavioural risks—often jointly contribute to disease burden. In 2015, 41·2% (95% UI 39·8–42·8) of global DALYs could be attributed to analysed risk factors, whereas 58·8% (57·2–60·2) of global disease burden could not be explicitly attributed to specific risk factors. In terms of individual risk categories, behavioural risk factors accounted for 30·3% (28·6–32·0) of attributable DALYs in 2015, followed by metabolic (15·5% [14·7–16·3]) and environmental or occupational risk factors (13·0 [11·9–14·0]). Regionally, total risk-attributable burden ranged from 59·0% (57·0–60·9) in southern sub-Saharan Africa to 33·5% (32·1–35·1) in north Africa and the Middle East; furthermore, ten regions had less than 40% of total DALYs attributable to risks being analysed (figure 1).

30Table 4 reports all-cause deaths and DALYs attributable to all risk factors from 2005 to 2015, including detail on attributable deaths and DALYs by risk-outcome pair (results appendix pp 3–2488 contains results for every geography). Globally, 32·2 million (95% UI 31·5 million to 33·0 million) deaths were attributable to all risk factors in 2015, a 4·9% (3·2–6·7) increase since 2005; however, age-standardised attributable deaths declined from 2005 to 2015 (a 17·9% decrease [16·6–19·2] to 497·5 deaths per 100 000 [485·2–510·0]). By contrast, total DALYs attributable to all risks decreased since 2005 by 5·6% (3·8–7·5) to 1·0 billion DALYs (0·96–1·09 billion) in 2015 and age-standardised DALYs attributable to all risks decreased since 2005 by 20·9% (19·5–22·5) to 14 412·9 DALYs per 100 000 (13 553·0–15 360·1) in 2015. Deaths and burden attributable to environmental and occupational risks significantly fell across measures, with age-standardised deaths falling by 22·5% (20·6–24·4) to 142·6 deaths per 100 000 (130·1–155·6) in 2015 and age-standardised DALYs decreasing by 24·8% (22·1–27·1) to 4500 DALYs per 100 000 (4164·6–4853·9) in 2015.

31Progress in environmental risks was mainly driven by sizeable reductions in mortality and disease burden attributable to unsafe water, sanitation, and hygiene, as well as to household air pollution. From 2005 to 2015, global deaths attributable to unsafe water and no handwashing with soap fell by more than 12%, whereas DALYs decreased by more than 20%. More rapid declines than in the above-mentioned risks in attributable deaths (27·5% [95% UI 22·3–32·7]) and burden (31·9% [25·7–37·4]) occurred for unsafe sanitation since 2005, to 807 904·2 deaths (727 439·6–895 462·5) in 2015 and to 46·3 million DALYs (41·1 million to 51·8 million) in 2015. Reductions in attributable mortality and DALYs due to diarrhoeal diseases (associated with unsafe water, sanitation, and hygiene) were particularly prominent. Attributable deaths due to household air pollution decreased by 13·0% (9·3–17·0) to 2·9 million deaths (2·2 million to 3·6 million) in 2015 and disease burden decreased by 20·3% (16·6–24·5) to 85·6 million DALYs (66·7 million to 106·1 million) in 2015; large declines also occurred in age-standardised rates of attributable mortality and DALYs. Occupational risk factors generally accounted for a smaller proportion of global deaths and disease burden than did environmental risk factors; nonetheless, attributable mortality and DALYS due to various occupational risk factors substantially increased from 2005 to 2015.

32Behavioural risks can be grouped into four main categories: generally large reductions for risk-attributable mortality and disease burden for risk factors associated with child and maternal malnutrition, mixed results for risk factors pertaining to alcohol and drug use, rising attributable deaths and DALYs due to dietary risk factors, and considerably varied trends for other behavioural risks, which span from sexual abuse and intimate partner violence to low physical activity. Attributable deaths and disease burden due to metabolic risks have increased since 2005, particularly for high fasting plasma glucose, for which all measures of attributable mortality and DALYs increased by more than 15% from 2005 to 2015. These increases in attributable burden from high fasting plasma glucose were led by increased deaths and DALYs from ischaemic heart disease, haemorrhagic stroke, chronic kidney disease, and diabetes. Attributable deaths and DALYs for high BMI also increased substantially, with 645 244 (95% UI 457 647–862 412) more attributable deaths in 2015 than in 2005. Attributable mortality and DALYs due to low glomerular filtration rates also significantly increased from 2005 to 2015, with these increases primarily associated with rises in attributable deaths and burden due to cardiovascular and circulatory diseases and chronic kidney disease.

33In 2015, the relative ranks and attributable burden due to Level 2 risk factors varied between men and women (figure 2). As the leading risk factor for both sexes, dietary risks accounted for 12·2% (95% UI 10·8–13·6) of total DALYs for men and 9·0% (7·8–10·3%) of total DALYs for women. These risks, which include diet high in sodium and diet low in fruit, contributed most to DALYs associated with three cause groups: cardiovascular and circulatory diseases, cancers, and diabetes and urogenital, blood, and endocrine diseases. In 2015, high systolic blood pressure also ranked among the leading risks for both sexes, contributing to 9·2% (8·3–10·2) of DALYs for men and 7·8% (6·9–8·7) of DALYs for women. Air pollution was the fifth-leading risk for both sexes, largely contributing to DALYs associated with cardiovascular and circulatory diseases, as well as lower respiratory infections, diarrhoeal diseases, and other common infectious diseases. Child and maternal malnutrition, the leading global risk factor in 1990, was the second-leading risk for women and the sixth-leading risk for men in 2015.

34Smoking was the second-leading risk factor for men in 2015, contributing to 9·6% (95% UI 8·5–10·7) of DALYs and a large proportion of male disease burden from cardiovascular and circulatory diseases, cancers, and chronic respiratory conditions. As the fifth-leading risk for men, alcohol and drug use was associated with 6·6% (6·1–7·1) of disease burden in 2015, primarily due to mental and substance use disorders, as well as cirrhosis and other chronic liver diseases; the burden attributable to these risk factors was far less for women (2·0% [1·8–2·2]) than for men. In 2015, high fasting plasma glucose was associated with 6·0% (5·4–6·6) of DALYs for men and 5·6% (5·1–6·2) for women. For women, 3·8% (3·4–4·3) of burden was attributable to unsafe sex, largely from HIV/AIDS and cervical cancer, whereas for men, 2·8% (2·5–3·0) was attributable to unsafe sex.

35Rising total attributable DALYs amid declines from 1990 to 2015 for age-standardised DALY rates were evident for various metabolic and behavioural risks, emphasising the need to parse out the effects of demographic and epidemiological factors on global risk profiles (figure 3). In 1990, childhood undernutrition, unsafe water, and high systolic blood pressure were the leading three risk factors for attributable DALYs. Of these risks, only high systolic blood pressure ranked among the leading three risks in 2015. Large reductions in both total attributable DALYs and age-standardised DALY rates from 1990 resulted in childhood undernutrition being ranked as the fifth-leading risk factor in 2015 and unsafe water being ranked as the 14th-leading risk factor.

36Environmental risk factors, including household air pollution and unsafe sanitation, decreased in terms of total attributable DALYs, age-standardised DALY rates, and relative ranks from 1990 to 2015. Over the period 1990–2005, attributable total DALYs for occupational risk factors, such as ergonomic factors, rose by more than 20% from 1990 to 2005, although age-standardised rates decreased by 9·5% (95% UI 7·3–11·7) over the same time period. Similar patterns occurred for most behavioural risk factors from 1990 to 2005, with significant increases in total attributable DALYs occurring for many of these risks; at the same time, age-standardised DALY rates significantly fell (eg, smoking, low physical activity, and most dietary risks, including diet high in sodium). Unsafe sex and drug use were exceptions, with each measure of attributable burden significantly increasing since from 1990 to 2005. For unsafe sex and drug use in particular, this rapid rise corresponded with the global HIV/AIDS epidemic. For most risk factors, the time period of 2005–15 resulted in an extension of earlier trends, with continued gains in reductions of attributable DALYs due to various environmental risk factors and more varied patterns for many metabolic and behavioural risks than for environmental risks. Yet, some important changes occurred between 2005 and 2015, including large reductions in attributable total DALYs (29·5% [26·8–32·0]) and age-standardised DALY rates (37·6% [35·2–39·8]) for unsafe sex and in attributable total DALYs (10·5% [2·8–17·8]) and age-standardised DALY rates (23·2% [16·8–29·1]) for intimate partner violence.

37A comparison of percentage change in risk exposure from 1990 to 2015 with the level of attributable DALYs in 2015 helps to identify large risks for which a long-term increase in global exposure has occurred (figure 4). Although disease burden attributable to unsafe sanitation, household air pollution, stunting, and underweight caused more than 10 million DALYs in 2015, exposure to these risks decreased for both sexes from 1990 to 2015 by more than 30%. Conversely, two risks caused more than 100 million DALYs and increased by more than 20%: high fasting plasma glucose and high BMI. Other risks with large increases in exposure but which caused less than 10 million DALYs include various occupational exposures, drug use, ambient ozone pollution, second-hand smoke, and diets low in polyunsaturated fatty acids (PUFAs). For a large group of risks at the global scale, exposure increased or decreased by less than 10% from 1990 to 2015. These included many components of diet, high systolic blood pressure, ambient particulate matter pollution, and alcohol use.

38Drivers of global changes in overall DALYs attributable to risk factors varied (figure 5). Across Level 3 risk factors, overall changes in all-cause attributable DALYs ranged from declines exceeding 50% for seven risk factors, including childhood undernutrition, suboptimal breastfeeding, and unsafe sanitation, to increases near to or exceeding 100% (ie, high BMI, occupational carcinogens, and drug use). Of these 46 Level 3 risk factors, attributable all-cause DALYs decreased significantly for ten from 1990 to 2015, whereas 34 increased significantly; two did not significantly change. Population ageing led to increased attributable all-cause DALYs for most risk factors, with a relative contribution that spanned from lower than 10% (for household air pollution from solid fuels and occupational injuries) to greater than 60% (for occupational carcinogens). Population ageing contributed to reductions in all-cause DALYs attributable to eight risk factors, namely environmental risks (eg, a 13·5% [95% UI 6·0–19·9] decrease for no handwashing with soap), those associated with nutritional deficiencies (eg, a 17·6% [12·6–24·0] decline for childhood undernutrition), and behavioural risks (eg, a 22·0% [10·9–31·4] decline for suboptimal breastfeeding). Changes in risk-deleted DALY rates since 1990 were primary drivers of reductions in all-cause, risk-attributable burden, with decreases in underlying DALY rates exceeding 50% for 13 risks by 2015. By contrast, changes in risk exposure varied markedly, contributing to declines in all-cause DALYs for ten risks (eg, 30·0% [29·0–38·0] due to declines in risk exposure for household air pollution and 21·7% [17·5–30·4] due to declines in risk exposure for iron deficiency); at the same time, change attributable to risk exposure increased for 16 risk factors to more than 25%, including high fasting blood glucose (25·1% [19·7–27·9]), ambient ozone pollution (37·5% [31·5–42·7]), occupational carcinogens (40·1% [28·5–50·8]), occupational injuries (41·1% [37·0–48·1]), high BMI (60·0% [54·9–69·2]), and drug use (70·0% [65·5–73·6]).

39Decreases in underlying cause-specific DALY rates—as opposed to other factors—were generally the main drivers of overarching reductions in cause-specific burden attributable to all risk factors (methods appendix pp 164–65). From 1990 to 2015, 36 causes decreased in terms of associated risk exposure, including a number of communicable causes, nutritional deficiencies, and chronic obstructive pulmonary disease. Notably, risk exposure was the only factor that improved from 1990 to 2015 for a subset of causes, with changes in population growth, ageing, and underlying cause-specific DALY rates all contributing to rising cause-specific disease burden; this finding was most evident for tracheal, bronchial, and lung cancer, as well as for cirrhosis and other chronic liver diseases due to alcohol use.

40Figure 6 shows the evolution of SEV by region for the ten leading global risk factors in terms of attributable DALYs as SDI changes and also provides the expected SEV level on the basis of SDI alone. Two main trends emerged: increasing and then levelling of SEVs for most metabolic and dietary risks and reductions in SEVs for environmental risks and those associated with childhood undernutrition as SDI approached mid-levels. For metabolic and dietary risks, above an SDI of approximately 0·8, expected levels of risk exposure either moderately dropped or remained fairly constant. An exception was alcohol use, for which SEVs increased with each increment of SDI. By contrast, SEVs for ambient particulate matter pollution did not substantially decline until above an SDI of 0·60, and the pace of SEV reductions for household air pollution accelerated above an SDI of about 0·40. These patterns reflect the complex shifts in risk exposure that accompany changes in development, which are further emphasised by regional SEV trends by risk factor.

41Two risk factors related to nutrition or diet—childhood wasting and diet high in sodium—reflected the nuances of changing risk exposure and levels of development. Particularly among regions with an SDI below 0·8, SEVs for childhood wasting decreased over time and with increasing SDI; nonetheless, a subset of regions, including south Asia and western sub-Saharan Africa, had consistently higher than expected SEVs for childhood wasting on the basis of SDI. For diet high in sodium, most regions saw minimal changes in exposure over time, even amid increases in SDI. In east Asia, SEVs for diet high in sodium were consistently above expected levels of exposure given SDI, whereas the opposite trend was found for Oceania and central Latin America. Across the development spectrum—including high-income Asia Pacific, southeast Asia, and eastern sub-Saharan Africa—exposure for diet high in sodium was at least moderately higher than expected on the basis of the SDI for a given region. Heterogeneous risk patterns occurred for two leading environmental risks—household air pollution and ambient particulate matter pollution—particularly in terms of the relationship between SEVs and increasing SDI. For smoking and alcohol use, strikingly different trends for SEV and SDI occurred. Although nearly every region recorded declines in SEVs for smoking, the rate at which these reductions took place alongside changes in SDI varied.

42Leading risk factors for early death and disability, as measured by attributable DALYs, varied by region, level of SDI, and sex in 2015 (figure 7). In high-income North America and the UK, smoking was the leading risk for attributable DALYs among both men and women, but for most of western Europe, smoking was the leading risk factor only for men, whereas high systolic blood pressure was the leading risk factor for women. A similar pattern emerged in east and southeast Asia, with smoking ranked as the leading risk factors for men in China, Thailand, Vietnam, and the Philippines, whereas metabolic risk factors—namely high systolic blood pressure—was the leading risk factor for attributable DALYs for women in these countries. Childhood undernutrition ranked as the leading risk factor for early death and disability for both sexes throughout western and central sub-Saharan Africa, as well as in a few countries outside of sub-Saharan Africa (eg, Laos and Tajikistan).

43In terms of the leading ten risk factors for both sexes, regional and country risk profiles showed both distinct patterns and heterogeneity. High systolic blood pressure was the leading risk for DALYs for 13 high-income countries and territories in 2015, high fasting plasma glucose was the leading risk for three, and smoking was the leading risk for 21. For a subset of geographies, including the USA, Canada, Australia, and the UK, drug use was a major risk for early death and disability in 2015. In 2015, high systolic blood pressure, high BMI, and high fasting plasma glucose were the leading risk factors for almost all geographies in Latin America and the Caribbean; Haiti was the primary exception, with unsafe sex as its leading risk for attributable DALYs in 2015. Across southeast Asia, east Asia, and Oceania, high systolic blood pressure was the leading risk factor for disease burden in nine countries and territories, ranging from China to Vanuatu. High BMI was the leading risk for DALYs in nine geographies, and high fasting plasma glucose ranked as the leading risk factor for three geographies, including Taiwan.

44High systolic blood pressure was among the leading two risk factors for all geographies in south Asia and, except for Pakistan and Bhutan, household air pollution remained among the leading four risk factors for attributable DALYs across geographies in this region. Ambient particulate matter pollution ranked as the third-leading risk factor in India and Nepal, whereas smoking was the second-leading risk factor for attributable burden in Bangladesh. In central Europe, eastern Europe, and central Asia, 28 of 29 countries had high systolic blood pressure as their leading risk factor for attributable DALYs in 2015; Tajikistan, where childhood undernutrition was the leading risk factor, was the only exception. Across central Europe, smoking was the second-leading risk factor for early death and disability, whereas alcohol use was among the leading four risk factors for attributable DALYs in geographies including Belarus, Moldova, and Russia.

45Throughout north Africa and the Middle East, except for Tunisia, high systolic blood pressure was among the three leading risk factors for disease burden in 2015, with nine geographies, including Egypt and Iran, recording this risk as the leading driver of early death and disability in that year. High BMI accounted for the highest attributable DALYs in eight countries, including Jordan and Saudi Arabia, and was also among the leading six risks for all countries in the region. For Afghanistan and Sudan, childhood undernutrition was the leading risk for DALYs in 2015, whereas it was the second-leading risk for DALYs in Yemen and the eighth-leading risk in Egypt. Unlike most of sub-Saharan Africa, several metabolic risks also emerged as leading drivers of attributable DALYs in southern sub-Saharan Africa by 2015; high BMI ranked as the second-leading risk factor in South Africa, and high systolic blood pressure was among the leading three risk factors for attributable DALYs in Botswana. In central sub-Saharan Africa, childhood undernutrition and unsafe sex were ranked first and second for attributable disease burden in all geographies except for Gabon. High systolic blood pressure ranked among the leading risk factors for attributable DALYs in most geographies in the region (eg, second in Gabon and third in the Congo).

46Drawing from 25 500 data sources, we estimated exposure to 79 metabolic, environmental and occupational, and behavioural risk factors or clusters of risks from 1990 to 2015 in 195 countries and territories and attributed deaths and overall disease burden to these risks. In 2015, all risks combined contributed to 57·8% (95% UI 56·6–58·8) of deaths and 41·2% (39·8–42·8) of DALYs worldwide. Since 1990, global risk exposure for both sexes combined increased significantly for 27 risks, did not significantly change for seven risks, and declined significantly for 27 risks. At the same time, that risk exposure increased for various leading risks, particularly metabolic risk factors associated with NCDs, and age-standardised risk-attributable deaths and DALYs declined for most risks. Globally, pronounced reductions in risk-deleted or underlying cause-specific DALY rates offset minimal changes in, or increased, risk exposure. These gains in risk-deleted DALY rates might not be large enough in the future to compensate for rising levels of risk exposure, such as high BMI or high fasting plasma glucose.

47Societal processes of urbanisation, the so-called westernisation of diets and lifestyles, and changes in employment activities, have all been viewed as primary drivers of changes in human health.,,, Such shifts have been thought to lead to deteriorating diets, rising obesity, decreased physical activity, and, ultimately, to worsening levels of metabolic risks, with associated higher rates of cardiovascular diseases and cancers., Results from the this study point to an ongoing risk transition, but with a trajectory complex and nuanced. The relationship between SEVs for most risks and SDI identified that poor water, poor sanitation, household air pollution, and micronutrient deficiencies and undernutrition decline steadily as countries develop. By contrast, some risks become worse as development proceeds, at least up to levels of SDI of about 0·8; these risks include low physical activity, high BMI, high total cholesterol, low PUFAs, partial breastfeeding, alcohol use, diet high in red meat, smoking, and diet high in sugar-sweetened beverages. Some risks that appear to worsen through early phases of development improved at the highest levels of SDI, such as smoking. As improvements in SDI continue and behavioural risks grow in dominance, an understanding of how to change behaviours effectively at both the individual level and for populations becomes increasingly relevant. Many other components of diet, occupational exposures, and some environmental risks do not show a marked relationship with development.

48The effects of the risk transition, at least globally, are often mitigated by trends in risk-deleted death or DALY rates. In this study, we identify likely candidates for specific drivers of improvements. Unmeasured risk factors could be driving these trends. Unlike cardiovascular diseases and some neoplasms, for other causes such as mental disorders or neurological disorders, the set of risks included in this study account for comparatively little of the observed burden; these causes make up an increasing share of the burden for geographies at high SDI. To date, we have not identified risk factors that meet our criteria of convincing or probable evidence, suggesting that research into unquantified risks is needed. In accompanying GBD 2015 analyses,,, we documented widespread improvements in overall development as measured by SDI. Gains in SDI are likely to operate through many pathways, including improved access to health care, public health programmes, and social and welfare policy. Advances in treatment are well documented for various causes, including HIV/AIDS,,, ischaemic heart disease,,,,, and various cancers,,, including breast,,, testicular, and Hodgkin's, yet for other causes, such as oesophageal cancer and interpersonal violence, the policies, programmes, and interventions responsible for declining risk-deleted death and DALY rates are less clear than for the aforementioned causes. Improvement of understanding of the risk-deleted rates in cause-specific mortality and disease burden will strengthen the evidence base for intervention effectiveness, the role of medical care access in addressing disease burden, and the importance of other social and welfare policies.

49We used our decomposition of drivers of attributable burden to identify four distinct groups of risks at the global level. First, for ten risks, attributable burden is declining, as is the exposure to the risk factor. This set of risks is dominated by the environmental risks, which are particularly common at low levels of SDI, and consist of vitamin A deficiency, undernutrition, zinc deficiency, suboptimal breastfeeding, poor sanitation, no handwashing, poor water, second-hand smoke, household air pollution, and occupational asthmagens. For these risks as a group, not only has exposure been declining, but risk-deleted DALY rates also declined, such as for diarrhoeal diseases. Furthermore, global shifts in population age structure contributed to decreases in both cause-specific and attributable burdens for risks that predominantly affect children. A second group of risks was characterised by declines in exposure exceeding 10% from 1990 to 2015, but increasing attributable burden, due in most cases to large increases driven by population growth and ageing. This group includes smoking, high systolic blood pressure, occupational ergonomic factors, childhood sexual abuse, and iron deficiency; for iron deficiency, the increase in attributable burden was quite small over the period 1990–2015. For a third group, attributable burden is increasing, and trends in exposure account for a less than 10% increase or decrease in attributable burden. This larger group of risks than the first two groups includes all components of diet except for low PUFA intake and diet high in sugar-sweetened beverages, some occupational exposures, residential radon, low glomerular filtration rate, alcohol use, high total cholesterol, intimate partner violence, lead exposure, low bone mineral density, and low physical activity. For these risks, attributable burden is increasing because of population growth and ageing. The degree of increase is driven by the extent to which declines in risk-deleted DALY rates compensate for the increases due to population growth and ageing. The final category is the risks that are perhaps the most concerning: those with increasing attributable burden and exposure contributing to an increase of at least 10% since 1990. This list includes occupational injuries, ambient ozone pollution, occupational carcinogens, diet low in PUFAs, diet high in sugar-sweetened beverages, drug use, high fasting plasma glucose, and high BMI. For these risks, declines in underlying rates were probably responsible for prevention of additional increases in attributable burden. Any risk for which attributable deaths and DALYs are increasing is a threat to both health systems and societies, but the risks that fall within the fourth group—risk factors with rising exposure and associated health loss—require immediate attention from policy makers and other stakeholders.

50The conventional approach to assessment of the contribution of unsafe water to health is to assess access to improved water sources as defined in the Millennium Development Goals., WHO estimated that, in 2015, 91% of populations living in low-income and middle-income countries had access to improved water and 68% had access to improved sanitation since 1990., Implicit in the focus on improved is the idea that the most important reductions in diarrhoea come from movement from an unimproved to an improved source of water or sanitation. However, findings from meta-analyses of intervention studies and retrospective cohorts show a wide variation of relative risks of diarrhoea within the category of improved water and improved sanitation. We found that the SEV for poor water was only 56·0% (95% UI 50·4–62·1) for men and 55·7% (50·2–61·8) for women in 2015, a small improvement of just 9·4% (5·3–13·0) since 1990. By contrast, for unsafe sanitation, the SEV decreased substantially, from 55·0% (53·8–56·5) in 1990 for men and 54·2% (52·8–55·7) for women to 33·7% (32·2–35·1) for both men and women in 2015. Our assessment of SEV for water showed that a substantial agenda is still needed across the world to achieve the overall goal of safe water and sanitation. The sixth Sustainable Development Goal includes targets for achievement of both universal and equitable access to safe drinking water and adequate and equitable sanitation for all by 2030., Given SEVs in many of these areas of the world, a large gap remains between current levels of safe water and sanitation and universal access. Moreover, our analyses show that progress in provision of safe water lags behind that of safe sanitation. This shift in focus to the lowest risk categories raises the bar considerably for what is needed for investment to reduce the risk of diarrhoea in all regions of the world.

51The SEV for smoking has decreased in many countries as well as globally. Global tobacco-attributable deaths and DALYs, however, have continued to rise because of increases in population numbers and ageing, which overwhelm declines in both exposure and risk-deleted rates of related disease burden. Given a known highly effective set of intervention strategies to reduce tobacco consumption,,,,,,, the challenge for tobacco is one of political priority for tobacco control. Despite the important developments of the Framework Convention on Tobacco Control, in many countries, progress has been slow or consumption has even increased. Continued close monitoring of tobacco consumption and the deaths and DALYs attributable to tobacco remains an essential aid to promotion of policies to reduce tobacco consumption.

52In GBD 2015, for long-term effects of smoking on lung cancer and chronic obstructive pulmonary disease, we used the Peto-Lopez method, which estimates the lifetime cumulative effect of cigarette smoking using a transformation of the observed lung cancer death rate. Although this method provides robust estimates of the burden of cancers related to tobacco, it is not fully consistent with the GBD approach of estimation of exposure independently of the outcomes affected by exposure. With a growing body of evidence for the association between smoking and other cancers and a good estimate of distribution of smoking, direct estimation of attributable burden is possible. Modelling of the direct relationship between smoking exposure in the past and present to cancers will also allow for exploration of counterfactual scenarios and other forms of CRA, such as estimation of avoidable burden. The set of outcomes that have been related to tobacco in pooled cohort studies includes many outcomes not quantified in this study, such as road traffic accidents, renal failure, and infectious diseases; careful assessment of which of these new outcomes meet the criteria of convincing or probable evidence is needed. Quantification of the full effects of tobacco will also require inclusion in future GBD studies of smokeless tobacco consumption. Regardless of the estimation method used or the scope of outcomes evaluated, tobacco remains a major global risk factor, despite more than 50 years of antitobacco efforts.

53Many aspects of dietary assessment remain controversial. Evidence for the effect of diet on NCDs mostly comes from prospective cohort studies using food frequency questionnaires (FFQs) with 1 year recall to establish diet at baseline. Findings from 24 h recalls or multiple-day diary records show a poor correlation with annual FFQs., Proponents of FFQs argue that the rank order of levels of intake across individuals in the annual FFQ is robust. Some authors have argued that measurement error for each diet component in the statistical analysis of the cohort data will tend to bias the findings towards the null and thus underestimate the effects of a diet component. However, the direction of bias in settings with measurement error in multiple independent variables is unknown in the presence of correlation between different variables., At the population level, single 24 h recall has been used in many nutrition surveys. 24 h recall probably underestimates, as shown in doubly labelled water and urinary sodium studies.

54Correlation between diet components is a crucial aspect of diet. Intake of beneficial dietary factors are generally, but not always, positively correlated with each other and inversely correlated with harmful dietary factors. This correlation could overestimate the relative risk of each dietary factor in cohorts as well as the total effect of dietary risks at the population level. Use of dietary pattern as the main exposure could potentially address this problem; however, several challenges exist in adoption of this approach for GBD. Concerns remain about the magnitude of the effect size of individual dietary risk factors on chronic diseases. Although many prospective cohorts have collected dietary data, published meta-analyses for most diet-disease pairs have included reports from only a fraction of these cohorts, indicating the potential for publication bias. Furthermore, most cohorts assessing the effect of diet on disease endpoints have adjusted for total energy intake in their statistical models. This practice emerged to address measurement error in dietary assessment tools and remove the effect of energy as a potential confounder. The adjustment for total energy intake means that diet components are defined as risks in terms of the share of diet and not as absolute levels of exposure. Because diet shares are analysed, increases in any component imply reductions in some other component, leading to the notion of replacement. Diet components are often analysed as pairs in which one component replaces another, further complicating analysis of cohort data. Given that many cohorts do not include the same dietary components in their analyses, the relative risks of dietary factors across cohorts might not be strictly comparable. This issue has been one of the main reasons for inconsistent findings in dietary meta-analyses., Future work to encourage more pooled analyses than at present of diet components for all the major cohorts would be beneficial; release of more data than available from the present major cohorts would stimulate various alternative analyses of diet, which would help strengthen the evidence for diet and attributable burden.

55Age-standardised DALYs attributable to diets high in sodium and exposure (as measured by SEV) for diets high in sodium increased slightly at the global level (7·2% [95% UI 0·7–19·0] for men and 4·0% [–0·3 to 12·1] for women) from 1990 to 2015. Reduction in sodium intake at the population level is one of WHO's nine global targets for NCDs. Many countries (eg, the UK, Finland, Japan, and Brazil) have already implemented or are considering implementing (eg, the USA) policies to reduce sodium intake.,,,, Reports from countries with high levels of sodium intake (eg, Japan) that have successfully implemented sodium reduction policies have argued for the beneficial effects of a lowering of sodium intake in these populations. Although multiple lines of epidemiological evidence support the harmful effects of very high levels of sodium intake, no scientific consensus has been reached on the optimal level of sodium intake. In GBD 2013, on the basis of findings of the Prospective Urban Rural Epidemiology (PURE) collaboration on sodium and cardiovascular mortality, and to incorporate the absence of scientific consensus on the optimal intake of sodium, we expanded the uncertainty for the TMREL for sodium to 1–5 g per day.,, PURE collaborators subsequently published a further analysis, which raised the possibility that, for those without hypertension, an inverse relationship might exist between sodium intake, all-cause mortality, and cardiovascular events. These findings challenge longstanding beliefs in the public health community with regard to the importance of modulation of sodium intake at the population level. Current policy is grounded in the evidence that links sodium intake and systolic blood pressure, which shows increases in systolic blood pressure with increases in sodium above 1 g per day., However, we have identified no prospective cohort studies that directly link sodium to disease endpoints that support reductions in the risk of outcomes at levels of intake below 3 g per day. Proponents of a low TMREL for sodium have argued that the cohort studies that generally show rising mortality at levels below 3–5 g per day might have the issue of reverse causation because ill individuals reduce sodium consumption., Given this continued debate, we chose not to change the uncertainty range of the TMREL from the current 1–5 g per day. If the findings from PURE and the pooled analysis are correct, an increased risk might exist for individuals without hypertension who consume less than 5 g per day. We have not included this potential increased risk in our quantification of uncertainty. Many studies have been completed or are underway examining the effects of sodium reduction in population groups or whole communities on systolic blood pressure, but none to date have been done that report the effects of sodium reduction on disease endpoints. Such studies, when and if they are done, could contribute to a resolution of outstanding questions about the TMREL for sodium. Regardless of the debate on the TMREL, we found that sodium accounted for at least 2·0% of global DALYs in 2015. The risk profile of increasing attributable numbers of DALYs, and no global progress in a reduction of exposure to diets high in sodium, places this risk among those of great concern for the management of health systems.

56Much of the diet policy debate has focused on the importance of reductions of sodium, sugar, and fat., Our assessment of the burden from diseases attributable to 14 dietary factors showed that, at the global scale, six factors each accounted for more than 1% of global DALYs, in order of importance: diets high in sodium, low in vegetables, low in fruit, low in whole grains, low in nuts and seeds, and low in seafood omega-3. Our findings suggest that, in addition to a policy focus on sodium, sugar, and fat, many other important components of diet should be minimised and promoted through education, subsidies, and other evidence-based programmes. If our findings are correct, a policy focus on the sugar and fat components of diets might have a comparatively smaller effect than that of promotion of increased uptake of vegetables, fruit, whole grains, nuts and seeds, and seafood omega-3. The disconnect between the diet policy debate and the evidence of which diet factors are most important globally can also be seen for red meat, where the harm of increased diabetes and colorectal cancer is smaller than that for all of the other 13 diet components considered under the GBD framework. Consideration of diet policy is made more complicated when cost-effectiveness, political feasibility, intensity of implementation, reactions of various stakeholders (eg, consumers and the food industry), and environmental effects, including climate change, are factored into national diet policy discussions. For example, promotion of some components of a healthy diet (eg, milk and dairy products) might have deleterious effects on environmental sustainability. These dimensions of diet policy should be added to the debates on how to transform national diets to be lower risk than at present.

57On the basis of findings from studies that showed the benefits from the reduction of LDL and total cholesterol to very low values, for GBD 2015, we revised the TMREL for total cholesterol downwards from 3·80–4·00 mmol/L to 2·78–3·38 mmol/L. This revision increased our estimate of the burden of total cholesterol for the year 2010 from 2·7 million (95% UI 1·9 million to 3·8 million) deaths in GBD 2013 to 4·0 million (3·1 million to 5·1 million) deaths in GBD 2015 and shifted our placement of cholesterol among leading risks from tenth (GBD 2013 ranking for the year 2010) to fourth (GBD 2015 ranking for the year 2010). Effective intervention options to influence population total cholesterol are available, including efforts to increase physical activity and diet and pharmacological interventions. Although consumption of dietary cholesterol is not linked to serum cholesterol, dietary interventions, such as increased consumption of fruits, vegetables, and fibre, and increased physical activity can influence LDL and total cholesterol and should be encouraged., Statins are highly effective, with few side-effects, justifying their use in many individuals, either alone or in combination with other medicines. In general, given the existence of proven and effective intervention strategies, the increase in the importance of cholesterol highlights a major opportunity for intervention.

58In this assessment, as in GBD 2013, we found that the most important Level 3 risk factor globally is elevated systolic blood pressure. The TMREL for this estimation, a systolic blood pressure of between 110 mm Hg and 115 mm Hg, is based on evidence from the pooling of prospective cohort studies that showed that individuals with a baseline blood pressure at this level have the lowest risk of future cardiovascular death., In addition to this observational evidence, two randomised clinical trials of blood pressure lowering have added substantially to the evidence in this area. Findings from the SPRINT trial showed that adults in the USA with elevated vascular risk and pre-existing hypertension (mean systolic blood pressure at baseline 139·7 mm Hg and on a mean of 1·8 medications) benefited from antihypertensive therapy that targeted a systolic blood pressure of 120 mm Hg. Findings from the HOPE-3 trial, a multinational study, showed that older adults (aged ≥55 years) with elevated vascular risk but without susbtantially elevated systolic blood pressure (mean systolic blood pressure at baseline 138·2 mm Hg) did not have any benefit when they received hydrochlorothiazide 12·5 mg plus candesartan 16 mg daily. However, neither study directly addresses selection of the GBD TMREL of 110–115 mm Hg since observational studies necessarily reflect the benefits accrued through maintenance of a healthy blood pressure throughout life rather than through use of blood pressure-lowering medications. The GBD estimate necessarily represents the effect of primary prevention and lifestyle modification, as well as the possibility of pharmacotherapy for achievement of the TMREL. The large potential gain in global health that we estimate if an optimal blood pressure was to be achieved suggests that further studies of blood pressure in people younger than 60 years of age remains an important area for investigation. A wide array of clinical and population strategies are available to reduce systolic blood pressure, including lowering population salt intake, increasing physical activity, reducing or slowing the rise of high BMI, and providing access to effective antihypertensives, which merit considerable attention in many countries.

59We report that the global SEV for alcohol decreased by 1·1% (95% UI −6·0 to 2·3) for men and 13·1% (−16·1 to −10·7) for women. Of note, however, because of the geographical distributions of alcohol consumption and background disease rates, alcohol consumption contributed to an increase in alcohol attributable DALYs over the same period. Assessment of the burden attributable to alcohol is complicated by potentially elevated risks in former drinkers compared with abstainers, potential protective effects of mild-to-moderate use of alcohol, and use of a TMREL of zero consumption for this study. Findings from meta-analyses have shown considerably elevated relative risks in former drinkers, equivalent to 30 g of pure alcohol per day or more for some outcomes,, although this finding could occur as a result of confounding by misclassification of former drinkers as lifetime abstainers.,,, Future GBD studies should carefully re-evaluate whether or not the excess risk in former drinkers is overestimated. If the relative risks for former drinkers have been overestimated, this overestimation will also affect our assessment of the alcohol SEV and the global trend in the SEV. The protective effect of mild-to-moderate alcohol use reported for ischaemic heart disease, diabetes in women, and reduced all-cause mortality is controversial;,,,,, some authors argue that this consistent finding is due to confounding., However, investigators of studies of all-cause mortality with certain types of quality exclusions find no overall mortality benefit of mild-to-moderate use. Various mendelian randomisation studies have contradicted previously claimed benefits of moderate alcohol consumption for several outcomes.,, Marked differences in male and female patterns of relative risk, such as for diabetes,, raise many questions about the biological pathways through which these effects might act. Finally, our alcohol analysis uses a TMREL of zero consumption; use of a higher TMREL than zero for alcohol use would increase our estimate of the global burden attributed to alcohol. Given the large burden of alcohol and the availability of effective options to reduce consumption at the population level, such as increased alcohol taxes, controls on outlet location and density, establishment or maintenance of limits on days or hours of sale, and screening and advice from health-care givers,,,, narrowing of the uncertainty in the alcohol assessment is important.

60Among the top five Level 3 risk factors, DALYs attributable to high BMI increased the most from 1990 to 2015 while SEV also increased. High BMI is mediated through increases in systolic blood pressure, cholesterol, and fasting plasma glucose (methods appendix pp 28–35). Our decomposition analysis suggests that the increase in risk-attributable burden due to high BMI is considerably smaller than it could have been if underlying rates, particularly for cardiovascular diseases, had not declined as much as they did over the past decade. The decline in the underlying rates is likely due to expanded access to preventive treatment and increased quality of care.,,,, If obesity continues to increase in the future, the consequences for health trends might be greater if the trend in underlying rates attenuates than if it continues to decrease. Although closely linked, we also estimate that the rate of increase in high fasting plasma glucose is slightly higher than that for obesity. The combined effect of rising obesity and rising fasting plasma glucose has consequences for various health outcomes and health-care delivery costs. For obesity, various policy options have been proposed,,, but few, if any, strategies have been proven to work at the population level. Analysis of obesity options needs to be closely linked to consideration of diet and physical activity. Although for most diet components we have reported on the effect of diet composition on health outcomes, a strong argument exists to explore the relationship of specific diet components with overall BMI.

61To move the GBD and CRA fields forward, we have reported, for the first time, details of the evidence that is available to support the causal relationship of each risk-outcome pair. The strength of the evidence of causality varies widely across risk factors. For the risk-outcome pairs where evidence from randomised controlled trials or high-quality prospective observational studies was not available, we considered the possibility of using other types of evidence, including mendelian randomisation, to establish causality. Although we have not used this approach in GBD 2015, this body of evidence seems to provide useful information that deserves a more detailed evaluation in future iterations of GBD than in this study. Likewise, the appropriate use of case-control data to support causality needs to be more clearly defined than in this study.

62Being explicit about the supporting evidence also shows that some risks have a similar body of evidence supporting them, even though, in some debates, the strength of evidence is thought to be very different. For this study, we focused on evidence mainly from prospective studies of exposure and disease endpoints. In select cases, we had to use prospective studies of intermediate outcomes, including trials of the effect of sodium on systolic blood pressure, prospective cohort studies of the effect of lead on systolic blood pressure, and trials and cohort studies of sugar-sweetened beverages' effect on BMI.,,, We did not include the strength of evidence uncertainty in our estimation of attributable burden, which requires more consensus than we had available on how different types of evidence can be combined to support a causal connection.

63In this study, as for GBD 2013, we have estimated the relative risk of exposure to particulate matter with an aerodynamic diameter smaller than 2·5 μm (PM2·5) over a broad range of daily doses by integrating relative risks from diverse sources of PM2·5, consisting of ambient air pollution, household air pollution, second-hand smoke, and tobacco smoking, using a single integrated exposure response (IER) curve., The premise behind this approach is that PM2·5 daily dose is a common indicator of risk from diverse sources of particulate pollution. Although assessments support an assumption of the equitoxicity of particulate matter mass from different sources for the purposes of burden estimation,, this issue remains an area of active research. We expect that the IER will continue to evolve in response to the latest evidence in the field of air pollution dose response; in keeping with the iterative nature of the GBD estimates, inclusion of additional cohort studies at lower levels of exposure to PM2·5 than those previously identified and included led to a revision of the TMREL distribution that is less than half of the level used in GBD 2013. Inclusion of new high-quality studies and the non-linear nature of the IER curve have increased our estimate of global burden. Methodological improvements to the curve-fitting algorithm that better capture the heterogeneity of component studies than do those in GBD 2013 produce UIs that are considerably wider than are those estimated in GBD 2013.

64To date, attributable burden has been the primary metric that was methodologically and computationally feasible for GBD risk assessments. As burden forecasts become routinely available, quantification of avoidable burden within the GBD analytic framework will become more tractable than at present. Development of avoidable burden estimates will require resolution of various important issues, such as the degree of reversibility for some risks and the type of counterfactual distribution most relevant to avoidable burden calculation. Arguably, feasible or cost-effective counterfactual distributions of exposure might be more relevant than is minimum risk.

65In the next iterations of the GBD CRA, we plan to incorporate new risks into two areas: distal socioeconomic factors, such as educational attainment, and the absence of effective health interventions, such as vaccination or seat belts. We will assess the evidence supporting new risk factors for inclusion and expand the scope of the analysis on the basis of both the evidence for causality and the availability of evidence for estimation of exposure levels. For behavioural, environmental and occupational, and metabolic risks, other candidate risks have been proposed for inclusion in GBD, including climate change, smokeless tobacco, added sugar, and access to health care. For existing risks, we will also carefully assess inclusion of new risk-outcome pairs. Continued close examination of evidence supporting causality might lead to modifications of the criteria for inclusion. For many distal factors, such as educational attainment, highly consistent evidence exists from prospective cohorts for all-cause mortality, but the relationships for cause-specific mortality might vary by context; for example, where access to treatment might modulate the effects of education on an outcome. For GBD 2015, we have made progress clarifying the evidence base for each risk-outcome pair. This type of in-depth examination will, we hope, lead to a simpler and more transparent approach than has been used before to define which risk-outcome pairs at any level of the causal web should be considered convincing or probable.

66Since estimation of the attributable burden of disease includes estimates of deaths, YLLs, YLDs, or DALYs, the limitations described elsewhere for those also apply to this analysis., We have developed our modelling strategies to quantify uncertainty given the available data. New data—particularly from countries where we currently lack data—might reveal levels of exposure that are outside the UIs that we have estimated. We assume that the joint effect for risk factors where we do not correct for mediation can be estimated with the multiplicative risk model. Although plausible, this model might not accurately capture how all risks interact. Analyses of the National Health and Nutrition Examination Survey III cohort suggest submultiplicative effects for some risks. A shift from a multiplicative model to a submultiplicative model has profound implications for all aspects of risk factor epidemiology, beginning at the estimation of relative risks and extending to use of relative risks as a generalisable construct.

67As a consequence of the absence of sufficient studies across all risk-outcome pairs, we did not systematically correct relative risks for publication bias. We did not correct relative risks for non-masking in studies; however, not all risks can be studied in a masked fashion (eg, tobacco smoking). Comparability would be compromised if we corrected some risks but not others. We generally assumed that relative risks were uniform across countries for a given age-sex group., Differences in relative risks with geography might exist, as has been found for the BMI relative risk curve and TMREL, but with the exception of breast cancer, we find that insufficient evidence exists to date to identify statistically significant differences.

68Development of robust models to estimate variation in SD is more difficult than for estimations of the mean. Because measurement error is necessarily included in SD from studies, they are an overestimate of the true SD. We did not correct SD for measurement error, with the exception of correction of observed systolic blood pressure to usual blood pressure. Measurement error also affects estimation of relative risks through regression dilution bias—the attenuation of the association between the level of risk and the incidence of disease outcomes. Because of the absence of detailed data for key cohorts, we have not corrected cholesterol or fasting plasma glucose relative risks for regression dilution bias. Our approach of estimation of the maximum relative risk on the basis of the reported cohorts and trials might underestimate burden if risk continues to rise at the highest levels of exposure beyond those reported. Although a log-linear function for relative risks and levels of exposure is adequate for the observed range of exposure, available data on the most extreme values are few. Given that a very small fraction of the population in any country is in these extreme exposure levels, the potential bias in our estimates is minimal. Additionally, estimation of burden for risks divided into polytomous risks might underestimate their burden compared with estimation of burden with a continuous risk variable.

69Although use of a log-normal distribution is the best of the parametric distributions that we have assessed for fasting plasma glucose, important deviations from the log-normal exist in survey data such as from the National Health and Nutrition Examination Survey. We have explored different mixtures of distributions, including two log-normal or a normal and log-normal, some of which provided better overall fit than the one that we used. However, we have so far not been able to solve the optimisation problem of estimation of two distributions given only a mean and SD reported from particular studies.

70We did not use the relative risk and exposure PAF calculation for unsafe sex, HIV risk from injecting drug use, and occupational injuries; we used direct evidence of the attributable fraction. Comparability of the results derived from these direct or categorical approaches to the risks estimated with the relative risk and exposure model is not certain.

71Too few studies exist to allow estimation of the contribution of household air pollution to ambient air pollution or vice versa. As a consequence, we might have underestimated the burden of household air pollution as a single risk factor; we might also have overestimated the burden of air pollution combined. Furthermore, our analysis of ambient air pollution has focused on PM2·5; other pollutants, including larger particulates than PM2·5 and nitrogen dioxide, might also be important to quantify. For cholesterol, we established our estimates of burden using total cholesterol; however, increasing evidence supporting the effects of LDL cholesterol—and improvements in data availability for serum LDL in different populations—suggest future estimations of risk from high cholesterol should use LDL cholesterol as the unit of estimation.

72Proxies for exposure to some risk factors are coarse, which is particularly the case for zinc deficiency, for which we estimated the balance between theoretical intake and physiological requirements from Food and Agriculture Organization food balance sheets for absorbable zinc. Although zinc deficiency can be estimated from the proportion of people estimated to have inadequate zinc intake, individual-level measurement of exposure truly needed to estimate the number of people at risk is not available. In a similar fashion, we used the proportion of the population in coarse occupational categories as a proxy for exposure to specific carcinogens and used fuel type as a proxy for household air pollution. Our ability to capture geographical variation and uncertainty in conversion of household solid fuel use to PM2·5 exposure improved the validity of our findings and UIs. Use of direct PM2·5 exposure measurement in households would be preferable to calibrate the widely available data for fuel use; however, such data have rarely been collected. As with all previous studies, we assessed the burden attributable to the availability of water and sanitation infrastructure, not the use of the infrastructure. Exposure for these risks has been defined in terms of availability and not use because that is what household surveys have collected. Our estimates are not biased, however, by this limitation because we derived relative risks from similar exposure definitions. Finally, we have not estimated intimate partner violence for men because of a scarcity of evidence of the health effects among men.

73Quantification of the health effects of a diverse set of largely avoidable risks is an important input into any national or global strategy to improve population health and make progress towards the Sustainable Development Goals. For the first time, we have been able to separately assess and report trends in risk exposure and decompose trends in attributable burden into the contribution of demographic change, risk exposure, and risk-deleted rates. Reductions in some risks strongly associated with low levels of development, such as poor water, poor sanitation, household air pollution, and undernutrition, have been important contributors to global progress. Some crucial risks for NCDs, particularly obesity, high fasting plasma glucose, and alcohol, are increasing. Deleterious trends have had a smaller effect than expected because of favourable declines in risk-deleted DALY rates for many causes, a component of which might be due to access to effective health care. In an era of rapid transition in societies and levels of health, tracking of and response to key risks will require constant monitoring at the local level. As the set of risk factors expands to encompass access to specific health-care interventions in future iterations of GBD, the GBD CRA will provide an even more comprehensive understanding than in this study of the drivers of health improvement and how they relate to the operation and priorities of health systems.

74Correspondence to: Prof Christopher J L Murray, Institute for Health Metrics and Evaluation, Seattle, WA 98121, USA cjlm@uw.edu

75Correspondence to: Prof Christopher J L Murray, Institute for Health Metrics and Evaluation, Seattle, WA 98121, USA cjlm@uw.edu

76This online publication has been corrected. The corrected version first appeared at thelancet.com on January 5, 2017

77This online publication has been corrected. The corrected version first appeared at thelancet.com on January 5, 2017