Global, regional, and national comparative risk assessment of 84 behavioural, environmental and occupational, and metabolic risks or clusters of risks for 195 countries and territories, 1990–2017: a systematic analysis for the Global Burden of Disease Study 2017

1The environmental, behavioural, and metabolic risks that drive injury and disease are the mechanisms by which public health efforts can most efficiently and effectively prevent health loss. Effecting population health improvements, therefore, requires understanding of not only the injuries and diseases that drive health burdens, but also the risks that drive injury and disease. Through a constantly evolving collection of cohort studies, randomised trials, and case-control studies, decades of epidemiological research have worked to quantify the nature and magnitude of associations between risk exposures and outcomes in studied populations. Moving beyond individual studies of individual populations, this raw evidence can be synthesised to draw the comprehensive and robust conclusions that are necessary to inform good public health policy. The Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) comparative risk assessment (CRA) is a comprehensive and comparable approach to risk factor quantification that offers a useful tool for synthesising evidence on risks and risk–outcome associations. With each annual GBD, we update the GBD CRA to incorporate new data on risk–outcome pairs, risk exposure levels, and risk–outcome associations.

2Evidence before the studyPopulation-level estimates of individual risks have been produced periodically by both WHO and UNICEF, whereas independent scientific publications provide risk estimates that are limited in the number of risks assessed and population size evaluated. Since 2010, the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) has produced comprehensive assessments of risk factor burden by age, sex, cause, and location. The previous iteration of this study, GBD 2016, assessed 84 behavioural, environmental and occupational, and metabolic risks between 1990 and 2016, with major updates in the assessment of second-hand smoke, alcohol use, and diet. The GBD study remains the only peer-reviewed, comprehensive, and annual assessment of risk factor burden by age, sex, cause, location, and year that complies with the Guidelines for Accurate and Transparent Health Estimates Reporting.Added value of this studyGBD 2017 expands the scope of GBD 2016 with the estimation of one new risk factor—bullying victimisation—and 80 new risk–outcome pairs, with a total of 476 risk–outcome pairs. GBD 2017 incorporates 46 749 sources. We have expanded our estimation locations with the addition of subnational locations for Ethiopia, Iran, Norway, and Russia, and estimates for Māori and non-Māori populations in New Zealand. We implemented broad improvements to methods to better estimate risk factor exposures and relative risks. Notably, we have moved from total cholesterol to low-density lipoprotein cholesterol, implemented continuous measures of exposure for smoking, and updated the ambient particulate matter pollution model with new ground measurement data from almost 4000 sites. We expanded upon our decomposition analyses to investigate the drivers of risk-attributable burden and the changes in burden by country, and to decompose risk-attributable changes between broad categories of risks, thus providing deeper insight into changing patterns of risk-attributable burden and their underlying causes. We broadened our analyses of geographical and temporal trends in risk exposure and burden by estimating expected risk-weighted prevalence of exposures based on Socio-demographic Index. We explored the observed relationship between development status and risk exposure across all locations and years, and for the first time we described spatiotemporal patterns in the ratio of observed-to-expected levels of risk exposure.Implications of all the available evidenceDecomposing trends by their underlying drivers reveals improvements in risk-deleted burden (ie, burden not attributed to risks in the GBD analysis), and broadly, improvements in exposure to environmental and behavioural risks. Conversely, increasing exposure to metabolic risks is driving increases in burden, indicating a crucial need for risk mitigation policies in this area. By quantifying the relationship between development and risk exposure, we highlight which risks appear sensitive to development and, of those, which are likely to improve or worsen with development. This analysis highlights areas where countries are either overperforming or underperforming relative to their economic peers and provides insight into areas where risk-modification strategies might be the best targets to improve health.

3Evidence before the study

4Population-level estimates of individual risks have been produced periodically by both WHO and UNICEF, whereas independent scientific publications provide risk estimates that are limited in the number of risks assessed and population size evaluated. Since 2010, the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) has produced comprehensive assessments of risk factor burden by age, sex, cause, and location. The previous iteration of this study, GBD 2016, assessed 84 behavioural, environmental and occupational, and metabolic risks between 1990 and 2016, with major updates in the assessment of second-hand smoke, alcohol use, and diet. The GBD study remains the only peer-reviewed, comprehensive, and annual assessment of risk factor burden by age, sex, cause, location, and year that complies with the Guidelines for Accurate and Transparent Health Estimates Reporting.

5Added value of this study

6GBD 2017 expands the scope of GBD 2016 with the estimation of one new risk factor—bullying victimisation—and 80 new risk–outcome pairs, with a total of 476 risk–outcome pairs. GBD 2017 incorporates 46 749 sources. We have expanded our estimation locations with the addition of subnational locations for Ethiopia, Iran, Norway, and Russia, and estimates for Māori and non-Māori populations in New Zealand. We implemented broad improvements to methods to better estimate risk factor exposures and relative risks. Notably, we have moved from total cholesterol to low-density lipoprotein cholesterol, implemented continuous measures of exposure for smoking, and updated the ambient particulate matter pollution model with new ground measurement data from almost 4000 sites. We expanded upon our decomposition analyses to investigate the drivers of risk-attributable burden and the changes in burden by country, and to decompose risk-attributable changes between broad categories of risks, thus providing deeper insight into changing patterns of risk-attributable burden and their underlying causes. We broadened our analyses of geographical and temporal trends in risk exposure and burden by estimating expected risk-weighted prevalence of exposures based on Socio-demographic Index. We explored the observed relationship between development status and risk exposure across all locations and years, and for the first time we described spatiotemporal patterns in the ratio of observed-to-expected levels of risk exposure.

7Implications of all the available evidence

8Decomposing trends by their underlying drivers reveals improvements in risk-deleted burden (ie, burden not attributed to risks in the GBD analysis), and broadly, improvements in exposure to environmental and behavioural risks. Conversely, increasing exposure to metabolic risks is driving increases in burden, indicating a crucial need for risk mitigation policies in this area. By quantifying the relationship between development and risk exposure, we highlight which risks appear sensitive to development and, of those, which are likely to improve or worsen with development. This analysis highlights areas where countries are either overperforming or underperforming relative to their economic peers and provides insight into areas where risk-modification strategies might be the best targets to improve health.

9Previous GBD studies have assessed the relationship between development, as measured by the Socio-demographic Index (SDI), and both the magnitude and composition of disease burden.,,, The results of those analyses highlighted the dramatic declines in communicable, maternal, neonatal, and nutritional diseases (CMNNDs) that have generally occurred with increases in socioeconomic development as well as the subsequent increases in life expectancy and absolute burden of non-communicable diseases (NCDs)—a pattern referred to as the epidemiological transition. Previous GBD analyses also estimated the expected burden for each cause in every location and year, based on that location's SDI. The comparison of observed burden to the burden expected based on SDI offered insight into the relative performance of countries at similar levels of development. Here, we extend those methods to analyse epidemiological transition with regards to risk exposure and risk-attributable burden. This analysis allows the identification of risks that are positively associated with development, negatively associated with development, or independent of development status. By estimating the levels of risk exposure and risk-attributable burden on the basis of SDI, and comparing these expectations to observed levels, it is possible to identify locations that either underperform or overperform compared with similarly developed countries.

10The GBD 2017 CRA includes 84 risk factors and 476 associated risk–outcome pairs. We expanded the scope of GBD 2016 with the inclusion of 80 new outcomes for existing risks and one new risk factor: bullying victimisation. The study provides estimates of exposure and attributable deaths and disability-adjusted life-years (DALYs) for 195 countries and territories for 1990 through 2017, including new subnational estimates for Ethiopia, Iran, New Zealand, Norway, and Russia. We explored changes in risk-attributable DALYs by decomposing those changes into six main component drivers of change, explored the relationship between risk exposure and SDI, and estimated the ratio of observed-to-expected levels of exposure and risk-attributable burden by SDI. As with previous iterations of GBD, the GBD 2017 CRA results presented here supersede all previously published GBD CRA estimates.

11The Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2017 expands on GBD 2016 with the estimation of one new risk factor—bullying victimisation—and 80 new risk–outcome pairs, making a total of 476 risk–outcome pairs. The study further investigates the drivers of changes in risk-attributable burden and explores the relationship between development and risk exposure.In 2017, 34·1 million (95% uncertainty interval [UI] 33·3–35·0) deaths and 1·21 billion (1·14–1·28) disability-adjusted life-years (DALYs) were attributable to risk factors included in GBD 2017. All included risks combined contributed to 61·0% (59·6–62·4) of deaths and 48·3% (46·3–50·2) of DALYs worldwide.The five leading risks in 2017 were high systolic blood pressure, smoking, high fasting plasma glucose, high body-mass index, and short gestation for birthweight.DALY-based ranks for all metabolic risks increased between 1990 and 2017 for both males and females. Consequently, four of the five leading risks were behavioural risks in 1990, whereas three of the five leading risks were metabolic risks in 2017.Between 2007 and 2017, the absolute number of risk-attributable DALYs declined by 3·44% (95% UI 2·47–4·40). During that period, exposures to behavioural, environmental, and occupational risks declined (improved), but these gains were somewhat offset by increases in exposure to metabolic risks, population growth, and population ageing.Socioeconomic development was strongly associated with exposure levels for many risks. Among the leading risks, unsafe water, household air pollution, and child wasting show pronounced decreasing trends with development. Conversely, smoking, alcohol use, drug use, and high low-density lipoprotein cholesterol all show a pronounced increasing trend with development.

12The Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2017 expands on GBD 2016 with the estimation of one new risk factor—bullying victimisation—and 80 new risk–outcome pairs, making a total of 476 risk–outcome pairs. The study further investigates the drivers of changes in risk-attributable burden and explores the relationship between development and risk exposure.In 2017, 34·1 million (95% uncertainty interval [UI] 33·3–35·0) deaths and 1·21 billion (1·14–1·28) disability-adjusted life-years (DALYs) were attributable to risk factors included in GBD 2017. All included risks combined contributed to 61·0% (59·6–62·4) of deaths and 48·3% (46·3–50·2) of DALYs worldwide.The five leading risks in 2017 were high systolic blood pressure, smoking, high fasting plasma glucose, high body-mass index, and short gestation for birthweight.DALY-based ranks for all metabolic risks increased between 1990 and 2017 for both males and females. Consequently, four of the five leading risks were behavioural risks in 1990, whereas three of the five leading risks were metabolic risks in 2017.Between 2007 and 2017, the absolute number of risk-attributable DALYs declined by 3·44% (95% UI 2·47–4·40). During that period, exposures to behavioural, environmental, and occupational risks declined (improved), but these gains were somewhat offset by increases in exposure to metabolic risks, population growth, and population ageing.Socioeconomic development was strongly associated with exposure levels for many risks. Among the leading risks, unsafe water, household air pollution, and child wasting show pronounced decreasing trends with development. Conversely, smoking, alcohol use, drug use, and high low-density lipoprotein cholesterol all show a pronounced increasing trend with development.

13The Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2017 expands on GBD 2016 with the estimation of one new risk factor—bullying victimisation—and 80 new risk–outcome pairs, making a total of 476 risk–outcome pairs. The study further investigates the drivers of changes in risk-attributable burden and explores the relationship between development and risk exposure.

14In 2017, 34·1 million (95% uncertainty interval [UI] 33·3–35·0) deaths and 1·21 billion (1·14–1·28) disability-adjusted life-years (DALYs) were attributable to risk factors included in GBD 2017. All included risks combined contributed to 61·0% (59·6–62·4) of deaths and 48·3% (46·3–50·2) of DALYs worldwide.

15The five leading risks in 2017 were high systolic blood pressure, smoking, high fasting plasma glucose, high body-mass index, and short gestation for birthweight.

16DALY-based ranks for all metabolic risks increased between 1990 and 2017 for both males and females. Consequently, four of the five leading risks were behavioural risks in 1990, whereas three of the five leading risks were metabolic risks in 2017.

17Between 2007 and 2017, the absolute number of risk-attributable DALYs declined by 3·44% (95% UI 2·47–4·40). During that period, exposures to behavioural, environmental, and occupational risks declined (improved), but these gains were somewhat offset by increases in exposure to metabolic risks, population growth, and population ageing.

18Socioeconomic development was strongly associated with exposure levels for many risks. Among the leading risks, unsafe water, household air pollution, and child wasting show pronounced decreasing trends with development. Conversely, smoking, alcohol use, drug use, and high low-density lipoprotein cholesterol all show a pronounced increasing trend with development.

19The CRA conceptual framework was developed by Murray and Lopez, who established a causal web of hierarchically organised risks that contribute to health outcomes and facilitate the quantification of risks at any level in the framework. In GBD 2017, as in previous GBDs, we assessed a set of behavioural, environmental or occupational, and metabolic risks that were organised into five hierarchical levels (appendix 1 section 5). At Level 0, GBD 2017 reports estimates for all risk factors combined. Nested within Level 0, Level 1 includes three risk categories: environmental and occupational, metabolic, and behavioural risk factors. This hierarchical structure continues, with each subsequent level including more detailed risks factors that are nested within the broader category above it. There are 19 risks at Level 2, 39 risks at Level 3, and 22 risks at Level 4, for a total of 84 risks or risk groups, where all risks (Level 0) is included as a risk group. Although we have added bullying as a new risk factor, the total number of risk factors remains unchanged from GBD 2016 because of the merging of two risk factors: we previously estimated second-hand smoke and occupational exposure to second-hand smoke as two separate risks but have incorporated the two exposures into one second-hand smoke Level 3 risk for GBD 2017. Each risk factor is associated with an outcome or outcomes, and each combination of risk and outcome included in the GBD is referred to as a risk–outcome pair. Risk–outcome pairs were included on the basis of evidence rules (appendix 1 section 5). To date, we have not quantified the contribution of distal social, cultural, and economic risk factors; however, our analysis of the relationship between risk exposures and sociodemographic development, measured with SDI, offers insights into the relationship between economic context and risk factors.

20This analysis largely follows the CRA methods used in GBD 2016. Given the scope of the analysis, we offer a high-level overview of the study methods and analytical logic, detailing areas of notable change and innovation since GBD 2016 and include risk-specific details in appendix 1 (section 4). This study complies with the Guidelines for Accurate and Transparent Health Estimates Reporting statement (appendix 1 section 5).

21For GBD 2017, we have estimated risk factor exposure and attributable burden by age, sex, cause, and location from 1990 to 2017. GBD locations are arranged in a nested hierarchy: 195 countries and territories are within 21 regions and these 21 regions are within seven super-regions. Each year, GBD includes subnational analyses for a few new countries and continues to provide subnational estimates for countries that were added in previous cycles. Subnational estimation in GBD 2017 includes five new countries (Ethiopia, Iran, New Zealand, Norway, Russia) and countries previously estimated at subnational levels (GBD 2013: China, Mexico, and the UK [regional level]; GBD 2015: Brazil, India, Japan, Kenya, South Africa, Sweden, and the USA; GBD 2016: Indonesia and the UK [local government authority level]). All analyses are at the first level of administrative organisation within each country except for New Zealand (by Māori ethnicity), Sweden (by Stockholm and non-Stockholm), and the UK (by local government authorities). All subnational estimates for these countries were incorporated into model development and evaluation as part of GBD 2017. To meet data use requirements, in this publication we present all subnational estimates excluding those pending publication (Brazil, India, Japan, Kenya, Mexico, Sweden, the UK, and the USA; appendix 2). Subnational estimates for countries with populations larger than 200 million (measured using the most recent year of published estimates) that have not yet been published elsewhere are presented wherever estimates are illustrated with maps but are not included in data tables.

22Four components were used for the calculations to estimate the attributable burden for a given risk–outcome pair: (1) the estimate of the burden metric being assessed for the cause (ie, number of deaths, years of life lost [YLLs], years lived with disability [YLDs], or DALYs); (2) the exposure levels for the risk factor; (3) the counterfactual level of risk factor exposure or theoretical minimum risk exposure level (TMREL); and (4) the relative risk of the outcome relative to the TMREL. For a given risk–outcome pair, we estimated attributable DALYs as total DALYs for the outcome multiplied by the population attributable fraction (PAF) for the risk–outcome pair for each age, sex, location, and year. The same logic applies to estimating attributable deaths, YLLs, and YLDs. The PAF is the proportion by which the outcome would be reduced in a given population and in a given year if the exposure to a risk factor in the past were reduced to the counterfactual level of the TMREL. The PAF for each individual risk–outcome pair is estimated independently and incorporates all burden for the outcome that is attributable to the risk, whether directly or indirectly. For example, the burden of ischaemic heart disease attributable to high body-mass index (BMI) includes the burden resulting from the direct effect of BMI on ischaemic heart disease risk, as well as the burden through the effects of BMI on ischaemic heart disease that are mediated through other risks (eg, high systolic blood pressure [SBP] and high low-density lipoprotein [LDL] cholesterol). When aggregating PAFs across multiple risks we used a mediation adjustment to compute the excess attenuated risk for each of 205 mediation-risk-cause sets (appendix 1 section 5).

23Information about the data sources, estimation methods, computational tools, and statistical analyses used to derive our estimates are provided in appendix 1 (sections 1–4). The analytical steps for estimating the burden attributable to single or clusters of risk–outcome pairs are summarised in the appendix 1 (section 2). Table 1 provides definitions of exposure for each risk factor and the TMREL used. Although the approach taken is largely similar to GBD 2016, we have implemented improvements to methods and incorporated new data sources. Appendix 1 (section 4) details each analytical step by risk. Citation information for the data sources used for relative risks is provided in an online source tool.

24We report all point estimates with 95% uncertainty intervals (UIs). To ensure that UIs capture uncertainty from all relevant sources (uncertainty in exposures, relative risks, TMRELs, and burden estimates) we propagate uncertainty through the estimation chain using posterior simulation using 1000 draws, from which we derive the lower and upper bounds of the UI based on the 2·5th and 97·5th percentiles. Where reported, estimates of percentage change were computed on the basis of the point estimates for the timepoints being compared.

25For each risk, we produced a summary measure of exposure, called the summary exposure value (SEV). The metric is a risk-weighted prevalence of an exposure, and it offers an easily comparable single-number summary of exposure to each risk. SEVs range from 0% to 100%, where 0% reflects no risk exposure in a population and 100% indicates that an entire population is exposed to the maximum possible level for that risk. We show estimates of SEVs for each risk factor (table 2; appendix 2) and provide details on how SEVs are computed for categorical and continuous risks in the appendix 1 (section 2).

26Spatiotemporal Gaussian process regression has been used in previous versions of GBD to estimate exposure for many risks, typically those with rich age-sex-specific data. It synthesises noisy data by borrowing strength across space, time, and age to best estimate the underlying trends for a given risk. With sufficient data, spatiotemporal Gaussian process regression is a fast and flexible modelling strategy for fitting non-linear temporal trends. Although methods were detailed for previous iterations of GBD, we have implemented several improvements for GBD 2017. First, we have added a space-time interaction weight, which flexibly adjusts the spatial weight of datapoints as an inverse function of data density over time. Second, we refined our method for calculating model uncertainty to ensure that modelling CIs aligned better with observed data variance and were more resilient to parameter changes. Finally, we improved raking, a post-processing step that ensures internal consistency between nested locations (subnationals) and their parents. Specifically, we implemented an option to rake in logit space, ensuring that raked estimates of prevalence data are naturally constrained between 0 and 1. More details are given in appendix 1 (section 2).

27We decomposed temporal changes in DALYs into six main component drivers of change: (1) population growth; (2) changes in population age structures; (3) changes in exposure to environmental and occupational risks; (4) changes in exposure to behavioural risks, (5) changes in exposure to metabolic risks; and (6) changes due to all other factors, approximated as the risk-deleted death and DALY rates. The risk-deleted rate is the death or DALY rate that would be observed had we removed all risk factors included in GBD 2017. In other words, the risk-deleted rate is the rate that would be observed had we reduced exposure levels to the TMREL for all risk factors included in GBD 2017. Changes in risk-deleted rates might reflect changes in risks or risk–outcome pairs that are not included in our analysis, or changes in other factors like improved treatments. We used methods developed by Das Gupta and adapted in GBD 2016 to ensure that decomposition results are linear aggregates over time or risk. We did a decomposition analysis for the 10-year period of 2007–17, for individual risks and the all-risk aggregate, accounting for risk mediation at the Level 4 risk and cause level. The contribution of changes in exposure to the individual risks was scaled to the all-risk effect. The contribution of risk exposures at higher cause and risk aggregates (eg, all-cause attributable to Level 1 GBD risks), or for all ages and both sexes combined, were calculated as the linear aggregate of the effect of individual risks for each cause, age, and sex.

28SDI is a composite indicator of development status that was originally constructed for GBD 2015, and is derived from components that correlate strongly with health outcomes. It is the geometric mean for indices of the total fertility rate among women younger than 25 years, mean education for those aged 15 years or older, and lag-distributed income per capita. The resulting metric ranges from 0 to 1, with higher values corresponding to higher levels of development. SDI estimation methods and estimates are detailed in appendix 1 (section 2). We examined the relationship between SDI and SEV to understand the relationship between development status and risk factor exposure levels. For each risk factor, we fit a separate generalised additive model with a Loess smoother on SDI for each combination of age group and sex. Inputs to this model were age-sex-specific SEVs for all Level 4 risks in the GBD risk hierarchy, for all national GBD locations and years between 1990 and 2017. Using an analogous modelling framework, we estimated the expected age and sex structure by SDI and used these expected age and sex proportions to calculate age and sex aggregates of expected exposure. For each risk–outcome pair, we used the expected SEVs to calculate expected PAFs. Because the SEVs for a given risk are not cause specific, the expected PAF estimates were then corrected using cause-specific correction factors that were derived by calibrating expected PAFs against empirical PAFs. To estimate expected risk-attributable burden, we drew from the CRA methods, first calculating the joint adjusted expected PAF for all risks for a cause using mediation factors (appendix 1 section 2). We then drew from the methods for observed risk-attributable burden calculation, using expected YLLs, deaths, and YLDs (appendix 1 section 2) to generate expected burden for a given SDI.

29Bullying victimisation is a new risk factor for GBD 2017. We estimate two outcomes for bullying in the GBD analysis: anxiety disorders and major depressive disorder. Bullying is commonly conceptualised as the intentional and repeated harm of a less powerful individual by peers and defined in the GBD as bullying victimisation of children and adolescents attending school by peers. This does not mean that bullying occurs exclusively at school and includes bullying that might occur to and from school as well as cyberbullying. We developed inclusion criteria that were robust while adaptable to the heterogeneity in largely non-health literature. Prevalence data were sourced from multicountry survey series including the Global School-based Student Health Survey and the Health Behavior in School-aged Children survey, as well as peer-reviewed studies, and were available for 153 GBD locations, covering all seven GBD super-regions. To reflect the exposure data and the definition of bullying victimisation in GBD, we adjusted prevalence estimates for the proportion of young people attending school using data published by the UN Educational, Scientific, and Culture Organization. Because the effect of bullying on depressive and anxiety disorders has been reported to wane over time and because prevalence estimates were from surveys of young people reporting current bullying victimisation rather than estimates of past exposure at the time the outcomes occur (ie, retrospective estimates), we developed a cohort method in which the prevalence of bullying victimisation exposure was tracked for the cohort of interest and relative risks varied with time between exposure to bullying and the point of estimation.

30In GBD 2017, the modelling process for air pollution, including ambient, household, and ozone exposure sources, was substantially improved. We adjusted the risk hierarchy, retaining air pollution as a Level 2 risk, adding particulate matter pollution at Level 3, and moving both household air pollution due to exposure to smoke from solid cooking fuels and ambient particulate matter pollution to Level 4 of the hierarchy. Developed for risk attribution for particulate matter pollution, the integrated exposure response curves combine epidemiological data from ambient, household, second-hand, and active smoking sources to construct a risk curve for the full exposure range. We updated the integrated exposure responses to include studies on ambient air pollution cohorts that were published after we completed our literature review for GBD 2016, systematic reviews of all active smoking cohorts, and a systematic review of second-hand smoke and chronic obstructive pulmonary disease (COPD). We also developed a strategy to map cohort studies of household air pollution to exposure levels of particulate matter less than 2·5 μm in diameter (PM2·5) to incorporate them into the curves.

31We did a systematic search of the scientific literature of health outcomes resulting from long-term exposure to ambient particulate matter pollution and, consequently, included type 2 diabetes as a new outcome for both ambient and household air pollution. Evidence suggests that exposure to PM2·5 might be mechanistically linked to type 2 diabetes through altered lung function, vascular inflammation, and insulin sensitivity.

32We estimated ambient PM2·5 exposure by combining satellite data with a chemical transport model and land use information. We calibrated satellite measurements to ground measurements using the Data Integration Model for Air Quality (DIMAQ). We made three notable improvements as follows: (1) we expanded our database of ground measurements from approximately 6000 to 9700 sites; (2) we made updates so the calibration model varies smoothly over space and time in data-dense regions; and (3) we improved uncertainty estimation by sampling from the DIMAQ's poster distribution in each grid cell (appendix 1 section 4).

33For previous GBDs, we have calculated relative risks from the integrated exposure response curves to produce PAFs and attributable burden for ambient particulate matter and household air pollution using the same TMREL for both risk factors. However, were a population to reduce one of the component exposures (ie, either household or ambient pollution), the other would remain. To capture this, we used a proportional PAF approach in which the integrated exposure response is used to calculate a relative risk and PAF for exposure to particulate matter from both ambient and household sources, and these are then weighted by the proportion of individuals exposed to each source (appendix 1 section 4).

34In GBD 2016, we estimated the burden attributable to low intake of polyunsaturated fatty acids, where low intake was the result of polyunsaturated fatty acids being replaced by saturated fats. Considering that it is equally harmful to replace polyunsaturated fatty acids with either saturated fat or carbohydrates, we have redefined the risk factor as low polyunsaturated fatty acids intake where these were replaced by either saturated fatty acids or carbohydrates. In this approach, the TMREL for polyunsaturated fatty acids does not account for saturated fat intake.

35For estimating consumption of whole grains, we developed an approach to use UN Food and Agriculture Organization (FAO) data, notably increasing our data coverage across countries and through time. First, we separately estimated total grain and refined grain availability, where availability includes domestic production, adjusted for imports, exports, waste, and animal feed. With whole grains and refined grains representing the entirety of all grain available, we calculated the availability of whole grains as the difference between total and refined grains. Finally, we adjusted these estimates using 24-h dietary recall data to represent consumption.

36In past cycles of GBD, given the strength of the causal relationship between sugar-sweetened beverage intake and BMI compared with the association between sugar-sweetened beverages and disease endpoints, we estimated the disease burden of high intake of sugar-sweetened beverages through its effect on BMI. This decision was based on the observation that evidence supporting a causal relationship between sugar-sweetened beverages and BMI was stronger than evidence for a direct causal relationship between sugar-sweetened beverages and disease endpoints. In GBD 2017, we reassessed all existing evidence on causal relationships between sugar-sweetened beverages and disease endpoints, and found sufficient evidence for a causal relationship between sugar-sweetened beverages and ischaemic heart disease and type 2 diabetes. Therefore, we have updated our approach and quantified the burden of disease attributable to the direct effect of sugar-sweetened beverages on disease endpoints.

37We added four new outcomes for high BMI as follows: type 2 diabetes, liver cancer due to non-alcoholic fatty liver disease, subarachnoid haemorrhage, and intracerebral haemorrhage. We applied the relative risk of diabetes only to type 2 diabetes. Relative risks for the association between high BMI and all liver cancers were used for both liver cancer due to non-alcoholic fatty liver and liver cancer due to other causes. Similarly, relative risks for the association between high BMI and haemorrhagic stroke were used for both subarachnoid haemorrhage and intracerebral haemorrhage.

38We added five additional outcomes for high fasting plasma glucose (FPG) as follows: type 1 diabetes, type 2 diabetes, liver cancer due to non-alcoholic fatty liver disease, subarachnoid haemorrhage, and intracerebral haemorrhage. Because an increased FPG concentration is the hallmark of diabetes, we assumed the PAFs were 1·0 for FPG and both type 1 diabetes and type 2 diabetes. Relative risks for the association between high FPG and all liver cancers were used for liver cancer due to non-alcoholic fatty liver and liver cancer due to other causes. Similarly, relative risks for the association between high FPG and haemorrhagic stroke were used for both subarachnoid haemorrhage and intracerebral haemorrhage.

39We made four important changes related to the estimation of burden attributable to iron deficiency. First, the definitions of the GBD cause “dietary iron deficiency” and the risk factor “iron deficiency” are no longer identical. The GBD cause name was changed from “iron-deficiency anaemia” to “dietary iron deficiency” to clarify the focus on inadequate intake and exclusion of other causes that can manifest as absolute or functional iron deficiency. Second, although the GBD risk factor name remained “iron deficiency”, the exposure estimates were expanded to include all iron deficiency, irrespective of whether or not inadequate dietary intake is the underlying cause (appendix 1 section 4). This change was based on review of the Child Health Epidemiology Research Group (CHERG) Iron Report, whose component studies revealed no distinction as to the aetiology of iron deficiency. Third, on the basis of the studies included in the CHERG Iron Report, which only assessed overall maternal mortality as an outcome, we added all subcauses of maternal disorders as outcomes of iron deficiency (the risk), leading to higher estimates of the burden attributable to iron deficiency among women of reproductive age. Fourth, on the basis of the absence of evidence supporting dietary iron deficiency as a primary cause of death, dietary iron deficiency was removed from the GBD 2017 cause of death analysis, resulting in zero mortality burden and lower overall estimates of burden for dietary iron deficiency (the cause). Dietary iron deficiency (the cause) is expressed in terms of prevalence and YLDs, but the exposure to iron deficiency (the risk) remains expressed as the counterfactual haemoglobin concentration that would be present in a given population group in the absence of all causes of anaemia that manifest as iron deficiency.

40We made three major improvements to our analysis of low birthweight for gestation and short gestation for birthweight. First, we added individual-level linked birth and death cohort data from nearly 25 million births in Japan and Singapore to strengthen our analysis of the joint mortality risk surface. Second, we drew on the strong correlation between birthweight and gestational age that we identified in our microdata analysis and used birthweight data to inform exposure estimates of short gestation. We also strengthened the link between non-fatal and risk analyses to ensure estimates of preterm birth were fully consistent throughout GBD 2017. The addition of individual-level linked birth and death cohort data resulted in higher estimates for low birthweight prevalence, mostly in data-sparse locations, whereas the consistency changes resulted in higher exposure estimates for both low birthweight and for short gestation, particularly in the late neonatal period. Third, we corrected an error where the risk attributable to low birthweight was mistakenly attributed to short gestation and vice versa in GBD 2016. This correction has no effect on the aggregate risk of low birthweight and short gestation but is the chief driver of differences in each individually.

41We have moved from estimating total cholesterol in GBD 2016 to estimating LDL cholesterol for GBD 2017. During the past two decades, substantially more data have been collected on LDL cholesterol than total cholesterol concentrations. The strong statistical relationship between total and LDL cholesterol also allows us to model LDL cholesterol when other cholesterol subfractions, such as high-density lipoprotein, are reported, but LDL cholesterol concentrations are not. The use of LDL cholesterol improves the policy relevance of our estimates, because LDL cholesterol is the key target of cholesterol-lowering medications and is the most commonly used laboratory biomarker for clinical decision making. We applied this change to the full dataset, including data that were newly extracted for GBD 2017 and data that had been extracted in previous iterations of GBD.

42To estimate smoking-attributable burden for GBD 2017, we transitioned from using 5-year lagged daily smoking prevalence (ie, the prevalence of smoking 5 years before the date for which estimates are being produced) and the smoking impact ratio to using continuous measures of exposure that incorporate cumulative effects among daily, occasional, and former smokers for 47 smoking-attributable health outcomes. We continue to use 5-year lagged daily smoking prevalence as the measure of exposure for ten outcomes. We estimated exposure among current smokers for two continuous indicators: cigarettes per smoker per day, and pack-years. We estimated exposure among former smokers using years since cessation. We estimated non-linear dose-response curves using a Bayesian meta-regression model for each of these continuous exposures. For nine outcomes with significant differences in effect size by sex or age, we produced sex-specific or age-specific risk curves (appendix 1 section 5). We included all forms of smoked tobacco in our exposure estimates and, given data limitations, assume that the risk of non-cigarette smoked tobacco products is the same as the risk of cigarettes; given the scarcity of data, we do not include electronic cigarette or vaporiser use in our exposure estimates.

43We added two new outcomes for high SBP: subarachnoid haemorrhage and calcific aortic valve disease. For both outcomes, we estimated relative risks on the basis of data from a pooled cohort study of 1·2 million participants. We know of no large cohort that has reported age-sex-specific relative risks of either subarachnoid haemorrhage or calcific aortic valve disease due to increased SBP, and used proxy causes for each as follows: we estimated the relative risks for subarachnoid haemorrhage on the basis of all stroke and those for calcific aortic valve disease on the basis of other cardiovascular disease. For each cause, we estimated age-sex-specific relative risks associated with a 10 mm Hg increase in SBP using the DisMod meta-regression tool (appendix 1 section 2).

44We have improved the exposure-modelling framework for unsafe water and sanitation. We estimate exposure levels for unsafe water and sanitation using ordinal categories. For example, we estimate the prevalence of exposure to three levels of unsafe water: piped, improved, and unimproved drinking water. Previously, the prevalences of piped and improved water were modelled independently, and we derived the prevalence of unimproved water as one minus the sum of piped and improved water. For GBD 2017, we modelled the prevalence of piped water as before, but now explicitly model the prevalence of improved and unimproved water separately as proportions of the unpiped envelope. This approach enables us to use the exposure category for which we have the most data (ie, piped water access) while also ensuring that the three exposure categories sum to one. The modelling process for unsafe sanitation was revised in an analogous way.

45The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. All authors had full access to all data in the study and had final responsibility for the decision to submit for publication.

46We observed diverse temporal trends for levels of exposure to different risk factors between 1990 and 2017 (table 2). During that time, SEVs for two risks increased by more than 40%: SEVs for high BMI increased by 70·4% (95% UI 57·1–84·5) and SEVs for ambient particulate matter pollution increased by 41·2% (32·1–52·0). Conversely, SEVs for four risks decreased by more than 40%: SEVs for child underweight decreased by 44·4% (41·2–48·1), household air pollution by 45·8% (41·5–49·8), diet high in trans fatty acids by 46·6% (32·8–66·3), and unsafe sanitation by 47·8% (43·2–52·1).

47SDI and SEVs were strongly associated for many risks (figure 1; for all risks, see appendix 2). Unsafe water source, unsafe sanitation, household air pollution, lead exposure, child underweight, child wasting, child stunting, iron deficiency, vitamin A deficiency, low calcium, and intimate partner violence all show a pronounced decreasing trend with increasing SDI. Conversely, discontinued breastfeeding, smoking, alcohol use, drug use, high red meat consumption, high processed meat consumption, high sugar-sweetened beverage consumption, and high LDL cholesterol all show a pronounced increasing trend with increased SDI. Other risks show no clear association with SDI, including ozone, radon, low whole grains, and low fibre.

48Globally, 61·0% (95% UI 59·6–62·4) of deaths and 48·3% (46·3–50·2) of DALYs were attributed to the risk factors addressed in GBD 2017 (appendix 2). The largest percentage of deaths attributable to risk factors were among NCDs at 64·8% (63·1–66·6), followed by CMNNDs at 61·6% (59·0–64·4), and injuries at 24·4% (22·2–26·5). For DALYs, the largest risk-attributable proportion was among CMNNDs (64·0%, 61·8–66·0), followed by NCDs (45·6%, 42·8–48·4), and injuries (22·1%, 20·2–24·0). Risk-attributable DALYs declined 4·9% (3·3–6·5) and deaths declined 8·3% (6·9–9·5) between 2007 and 2017. The numbers of risk-attributable deaths and DALYs in 2007 and 2017 are provided for each cause in table 3.

49In 2017, NCDs had the largest risk-attributable burden, with 26·6 million (95% UI 25·8–27·5) deaths and 706 million (659–756) DALYs attributable to all risk factors combined (appendix 2). The five leading NCD causes of absolute risk-attributable DALYs were ischaemic heart disease, intracerebral haemorrhage, type 2 diabetes, COPD, and ischaemic stroke. In 2017, 95·0% (95% UI 93·4–96·3), and 93·8% (91·3–95·9) of ischaemic heart disease DALYs and deaths were risk attributable, resulting in 162 million (158–166) risk-attributable ischaemic heart disease DALYs, and 8·38 million (8·10–8·65) risk-attributable ischaemic heart disease deaths. After ischaemic heart disease, intracerebral haemorrhage had 57·9 million (55·2–60·6) risk-attributable DALYs (89·8% [87·0–92·3] of all intracerebral haemorrhage DALYs), type 2 diabetes had 57·4 million (45·0–71·9), COPD had 54·9 million (48·1–60·4; 67·3% [60·8–72·8] of all COPD DALYs), and ischaemic stroke had 47·8 million (43·4–52·3; 86·7% [82·7–90·8] of ischaemic stroke DALYs). Because we estimated a PAF of 1·0 for high FPG and type 2 diabetes, 100% of type 2 diabetes DALYs and deaths were attributed to risk factors addressed in GBD 2017.

50Among CMNNDs, 6·40 million (6·00–7·00) deaths and 446 million (419–475) DALYs were attributable to risks. The five leading causes of absolute risk-attributable DALYs among CMNNDs were lower respiratory infections, diarrhoeal diseases, neonatal preterm birth complications, neonatal encephalopathy due to birth asphyxia and trauma, and HIV/AIDS resulting in other diseases. In 2017, 80·5% (95% UI 77·4–83·4) and 64·0% (59·9–68·1) of lower respiratory infection DALYs and deaths were risk attributable, resulting in 76·9 million (66·1–92·8) risk-attributable lower respiratory infection DALYs and 1·64 million (1·51–1·77) risk-attributable lower respiratory infection deaths. Diarrhoeal diseases had 76·9 million (66·1–92·8) risk-attributable DALYs (94·9% [89·8–97·7] of all diarrhoeal diseases DALYs), neonatal preterm birth had 70·2 million (64·4–77·2; 100% of all neonatal preterm birth DALYs, since we estimate a PAF of 1·0 for this risk factor), neonatal encephalopathy due to birth asphyxia and trauma had 41·2 million (36·7–45·0; 72·9% [66·6–78·3] of neonatal encephalopathy due to birth asphyxia and trauma DALYs), and HIV/AIDS resulting in other diseases had 32·6 million (29·1–36·3; 77·5% [75·6–79·2] of HIV/AIDS resulting in other diseases DALYs).

51Among injuries, 1·09 million (0·99–1·19) deaths and 55·9 million (49·6–62·5) DALYs were attributable to risks in 2017. The five leading injury causes of risk-attributable DALYs were falls, self-harm by other specified means, motor vehicle road injuries, pedestrian road injuries, and motorcyclist road injuries. In 2017, 34·1% (95% UI 30·7–38·2) of DALYs and 43·4% (40·6–46·6) of deaths from falls were risk attributable, resulting in 12·2 million (9·93–15·0) risk-attributable DALYs and 302 000 (275 000–330 000) risk-attributable deaths from falls. Self-harm by other specified means had 8·07 million (5·50–10·7) risk-attributable DALYs (25·7% [17·6–33·9] of all DALYs from self-harm by other specified means), motor vehicle road injuries had 5·95 million (5·19–6·70; 24·2% [21·3–27·0] of all motor vehicle road injury DALYs), pedestrian road injuries had 5·50 million (4·76–6·28; 23·4% [20·8–26·1] of pedestrian road injury DALYs), and motorcyclist road injuries had 4·05 million (3·46–4·64; 29·0% [25·3–32·4] of motorcyclist road injury DALYs).

52In 2017, 34·1 million (95% UI 33·3–35·0) deaths and 1·21 billion (1·14–1·28) DALYs were attributable to GBD risk factors (table 3). Between 2007 and 2017, the number of all-age risk-attributable YLLs declined from 1·04 billion (1·02–1·07) to 944 million (922–968), and age-standardised YLL rates declined from 16 359 (16 005–16 748) per 100 000 to 12 509 (12 205–12 837) per 100 000. Age-standardised risk-attributable death rates declined from 531 (520–544) per 100 000 in 2007 to 448 (437–460) per 100 000 in 2017. Conversely, during that period, the absolute number of risk-attributable deaths increased from 31·5 million (30·9–32·2) to 34·1 million (33·3–35·0). During the same period, there was no significant trend in risk-attributable non-fatal burden because there was no statistically significant trend in non-fatal burden: age-standardised risk-attributable YLD rates were 3442 (2593–4374) per 100 000 in 2007 and 3357 (2528–4275) per 100 000 in 2017; and the absolute numbers of risk-attributable YLDs were 227 million (170–288) in 2007 and 263 million (198–336) in 2017. The largest percentage declines in the number of risk-attributable DALYs are for measles; the largest percentage increases are for osteoarthritis. Globally in 2017, high SBP was the leading risk factor, accounting for 10·4 million (9·39–11·5) deaths and 218 million (198–237) DALYs, followed by smoking (7·10 million [6·83–7·37] deaths and 182 million [173–193] DALYs), high FPG (6·53 million [5·23–8·23] deaths and 171 million [144–201] DALYs), high BMI (4·72 million [2·99–6·70] deaths and 148 million [98·6–202] DALYs), and short gestation (1·43 million [1·36–1·51] deaths and 139 million [131–147] DALYs; table 3).

53Behavioural risk factors accounted for 43·6% (95% UI 41·7–45·5) of all DALYs in 1990, followed by environmental and occupational risk factors at 17·4% (15·9–19·0) and metabolic at 10·3% (9·63–11·1). Between 1990 and 2017, these percentages declined by 16·2% for behavioural risk factors, which accounted for 36·5% (34·7–38·4) of all DALYs in 2017, and 29·3% for environmental and occupational risks, which accounted for 12·3% (11·5–13·3) in 2017. Proportions increased by 63·7% for metabolic risks, which accounted for 16·9% (15·6–18·3) of all DALYs in 2017.

54The proportion of all DALYs attributable to each Level 1 risk also varies with SDI (appendix 2). Metabolic risks accounted for increasingly large proportions of DALYs with increasing levels of SDI, up to high-middle SDI: metabolic risks accounted for 8·31% (95% UI 7·69–9·00) of DALYs in low SDI, 13·6% (12·6–14·7) in low-middle SDI, 21·3% (19·6–23·0) in middle SDI, 24·8% (22·7–27·2) in high-middle SDI, and 20·7% (18·4–23·1) in high SDI countries. Conversely, both behavioural and environmental and occupational risks accounted for smaller proportions of DALYs in higher SDI regions. Behavioural risks accounted for 38·8% (37·2–40·4) of DALYs in low SDI, 38·5% (36·8–40·2) in low-middle SDI, 35·2% (33·2–37·2) in middle SDI, 37·4% (34·9–39·7) in high-middle SDI, and 30·2% (28·1–32·4) in high SDI countries. Environmental and occupational risks accounted for 16·9% (15·6–18·3) of DALYs in low SDI, 14·1% (13·0–15·3) in low-middle SDI, 11·1% (10·2–12·0) in middle SDI, 9·61% (8·85–10·4) in high-middle SDI, and 6·56% (5·92–7·23) in high SDI countries (appendix 2).

55Between 1990 and 2017, high SBP was consistently responsible for the largest number of all-cause deaths, followed by smoking then high FPG. The fourth leading risk for mortality in 1990 was child wasting, which was in 21st position in 2017. The fourth leading risk for mortality in 2017 was high BMI, which increased in rank from 1990 when it was ninth. High LDL cholesterol remained the fifth leading cause of risk-attributable deaths in 1990 and 2017. These rankings and trends differ for DALYs. In 1990, the five leading risks for DALYs were child wasting, short gestation, low birthweight, smoking, and high SBP, whereas in 2017, the leading five risks were high SBP, smoking, high FPG, high BMI, and short gestation (figure 2A).

56Leading risks differ for males and females. For males, the leading risks in 2017 were (in order of descending rank) smoking, high SBP, high FPG, alcohol use, and short gestation; the leading risks for females were high SBP, high FPG, high BMI, short gestation, and low birthweight. Three of the five leading risks for males were behavioural risks, whereas three of the five leading risks for females were metabolic risks (figure 2B, C).

57DALY-based ranks for all metabolic risks increased between 1990 and 2017 for both males and females, whereas changes in ranks were more heterogeneous for environmental and occupational and behavioural risks. In 1990, six of the ten leading risks were behavioural risks, three were environmental, and one was metabolic. Of the ten leading risks in 2017 for both sexes, five were behavioural, four were metabolic, and only one was environmental. Broadly, in terms of their relative importance, metabolic risks rose in rank whereas environmental and occupational risks fell. Within behavioural risks, 13 of the 15 dietary risks increased rank. Appendix 2 shows trends in leading risk factors within each SDI quintile and trends in leading risk factors for deaths and DALYs by location.

58Changes in the absolute number of DALYs over time are the result of changes in six underlying components: (1) population growth; (2) population ageing; (3) changes in exposure to environmental and occupational risks; (4) changes in exposure to behavioural risks; (5) changes in exposure to metabolic risks; and (6) changes due to all other (ie, risk-deleted or residual) factors. Figure 3 shows the changes in these components for each Level 1 cause, and for all causes combined (for Level 2 causes, see appendix 2). Broadly, in the absence of demographic changes, changes in risk exposure and risk-deleted DALYs would have led to a 23·5% decline in DALYs between 2007 and 2017 in both sexes. Conversely, in the absence of changes in risk exposure and risk-deleted DALYs, demographic changes would have led to an 18·6% increase in DALYs during that period. Comparing drivers of change for males and females, females have generally had greater declines in risk exposure: the aggregate effect of changes in exposure across all risks would have led to a 6·93% decline in DALYs from all causes for males, versus a 9·08% decline for females (figure 3).

59Of all individual risk categories, changes in exposure to behavioural risks have driven the largest change in burden: declines in exposure to these risks would have resulted in a 7·41% decline in DALYs from all causes in both sexes combined. For both males and females, changing exposures to behavioural risks drove declines for NCDs and CMNNDs and increases for injuries (figure 3). Changes in exposure to environmental and occupational risks would have resulted a 2·59% decline in DALYs from all causes for both sexes, and it is the only risk category that was a driver of decline for all cause groups and both sexes. Conversely, in the absence of changes from other drivers, changes in metabolic risks would have resulted in a 2·15% increase in all-cause DALYs for both sexes.

60Among the three cause groups, CMNNDs are the only Level 1 cause group for which combined effects of improvements in risk exposure and risk-deleted burden have outweighed the effects of demographic changes for both males and females. Declines in risk exposure would have resulted in a 12·0% decline in DALYs from CMNNDs among males and a 14·8% decline in females. For NCDs, increasing exposure to metabolic risks has largely offset health gains from improvements in environmental and behavioural risks. In aggregate, changes in risk exposure would have resulted in a 2·83% decline in DALYs from NCDs in males and a 2·29% decline in females. Among both males and females, changes in exposure to environmental risks, behavioural risks, and all risks combined would have resulted in reduced injury burden.

61Drivers of changes in risk-attributable DALYs were spatially heterogeneous (figure 4). Broadly, population growth has driven increases in risk-attributable DALYs across most locations except for Eastern Europe, Cuba, Greenland, Guyana, Japan, and Portugal. Population ageing has driven increases in most locations, with notable exceptions in central, eastern, and western sub-Saharan Africa, and parts of the Middle East and south and southeast Asia. Conversely, all-cause risk-deleted DALY rates have declined across most locations; however, they have increased in a number of subnational locations in Brazil, China, the UK, India, Mexico, and the USA. Broadly, DALYs attributable to metabolic risk exposure have increased in most locations and DALYs attributable to environmental and occupational and behavioural risks have largely declined.

62Estimating the expected SEV for a given risk factor at a given level of SDI, a comparison can be made between observed exposure levels in a given place and time and their expected levels based on SDI. An observed to expected ratio (O/E ratio) of 1·0 indicates that observed exposure levels equal our expectation; O/E ratios less than 1·0 indicate that observed exposures are better than expected (ie, at levels associated with lower risk); and O/E ratios greater than 1·0 indicate that observed exposures are worse than expected (ie, at levels associated with greater risk).

63Across leading environmental risks, O/E ratios for ambient particulate matter pollution showed no consistent trends across regions, whereas there were nearly universal increases in O/E ratios for household air pollution and unsafe water (figure 5A). In 2017, O/E ratios for ambient particulate matter pollution were notably high in north Africa and Middle East (2·03) and south Asia (1·81). O/E ratios for household air pollution were notably high in southeast Asia, east Asia, and Oceania (3·05) and central Europe, eastern Europe, and central Asia (1·73). Between 1990 and 2017, O/ E ratios for unsafe water increased across all super-regions, with the largest changes in southeast Asia, east Asia, and Oceania (85·6% increase; ratio of 1·75 in 2017), north Africa and Middle East (57·8% increase; ratio of 1·02 in 2017), and south Asia (55·5% increase; ratio of 1·41 in 2017).

64Among the leading behavioural risks between 1990 and 2017, there were near universal declines in O/E ratios for both smoking and alcohol use, and inconsistent trends in O/E ratios for low birthweight (figure 5B). The range of O/E ratios between regions is especially broad for alcohol use. These ratios were consistently high in sub-Saharan Africa (O/E ratios declined from 2·10 to 1·49 between 1990 and 2017), Latin America and the Caribbean (ratios declined from 1·83 to 1·54), and central Europe, eastern Europe, and central Asia (ratios declined from 1·82 to 1·68). O/E ratios for alcohol use were consistently the lowest in north Africa and the Middle East, declining from 0·38 to 0·21 between 1990 and 2017.

65Trends in O/E ratios for leading metabolic risks were heterogeneous across super-regions (figure 5C). For high SBP, O/E ratios improved by more than 5% between 1990 and 2017 for two super-regions: 27·1% decrease in the high-income region and 8·05% in the central Europe, eastern Europe, and central Asia region. By contrast, ratios for high SBP increased by more than 5% in three super-regions: 22·0% in sub-Saharan Africa, 16·4% in southeast Asia, east Asia, and Oceania, and 9·67% in south Asia. In 2017, ratios for high SBP were highest in sub-Saharan Africa (1·19) and central Europe, eastern Europe, and central Asia (1·19) and were lowest in the high-income region (0·77). O/E ratios increased for both high FPG and high BMI across all super-regions between 1990 and 2017. The north Africa and Middle East region stands out as having the highest O/E ratios for both high FPG and high BMI across many years.

66Broadly, total expected risk-attributable burden declines with increasing SDI (figure 6). At an SDI of 0·1, we expect 101 000 risk-attributable DALYs per 100 000 males and 90 600 per 100 000 females; at an SDI of 0·5, we expect 20 200 per 100 000 males and 16 300 per 100 000 females; and at an SDI of 0·9, we expect 15 900 per 100 000 among males and 9640 per 100 000 among females.

67Increasing SDI was associated with dramatic declines in the expected burden attributable to environmental risks in both sexes: the expected burden attributable to environmental risks declines from 41 600 DALYs per 100 000 people (31·6% of total burden) with an SDI of 0·1 to 1870 DALYs per 100 000 people (10·6% of total burden) with an SDI of 0·9. The pattern for behavioural risks was more complex and heterogeneous. Although the expected burden attributable to malnutrition declined with increasing SDI, the expected burden attributable to tobacco use, drug use, alcohol use, and most dietary risks generally increased with increasing SDI. At an SDI value of 0·1, behavioural risks accounted for 65·0% of the total expected risk-attributable burden, at an SDI of 0·75 they accounted for 51·5%, and at an SDI of 0·9 they accounted for 52·9%. The expected burden attributable to metabolic risks increased with increasing SDI up to an SDI of 0·65, then declines with increasing SDI above that level. At an SDI value of 0·1, metabolic risks accounted for 3·44% of the total expected risk-attributable burden, at an SDI of 0·75 they accounted for 37·6%, and at an SDI of 0·9 they accounted for 36·4%.

68In males, the two leading risk factors were smoking and high SBP (in varying order) in four of the seven super-regions (figure 7; appendix 2). The three exceptions were sub-Saharan Africa (unsafe sex was the leading risk with an O/E ratio of 14·9 followed by child wasting with a ratio of 0·71), Latin American and the Caribbean (alcohol use with a ratio of 1·19 then high FPG with a ratio of 0·78), and the high-income super-region (smoking with a ratio of 0·65 then high BMI with a ratio of 0·83). In females, the leading risk factors were either high BMI or high SBP in four of the seven super-regions (figure 7; appendix 2). The three exceptions were sub-Saharan Africa (as for males, unsafe sex was the leading risk among females, with an O/E ratio of 5·51, followed by child wasting with a ratio of 0·63), south Asia (short gestation with a ratio of 0·79, followed by high SBP with a ratio of 0·79), and the high-income super-region (smoking with a ratio of 0·98 followed by high BMI with a ratio of 0·91). High FPG was the second leading risk factor in southeast Asia, east Asia, and Oceania (O/E ratio of 0·78) and Latin America and the Caribbean (O/E ratio of 0·84) for females. Notably, the O/E ratios exceed one for all five leading risk factors for males for the central Europe, eastern Europe, and central Asia regions, and for nearly all countries within those regions.

69The highest O/E ratios for environmental and occupational risks were in southern sub-Saharan Africa and south Asia, driven largely by air pollution and unsafe water, sanitation, and handwashing in both regions (figure 8). For behavioural risks, high O/E ratios were recorded in southern sub-Saharan Africa, driven largely by higher than expected unsafe sex and intimate partner violence, and eastern Europe, largely due to higher than expected alcohol and drug use. The burden attributable to metabolic risks was higher than expected in Oceania and much of the central Europe, eastern Europe, and central Asia super-region.

70Globally, bullying victimisation is the leading risk factor for mental disorder DALYs in 2017 (other risks factors are lead exposure, intimate partner violence, and childhood sexual abuse) and the 35th leading risk factor of all 54 risks for all NCDs. However, for major depressive disorder, bullying accounts for fewer DALYs than childhood sexual abuse or intimate partner violence. In 2017, 2·57 million (95% UI 1·54–3·97) DALYs were attributable to bullying victimisation, by peers, of children and adolescents attending school. This was distributed relatively evenly between anxiety disorders (1·36 million DALYs [0·81–2·13]) and major depressive disorder (1·21 million DALYs [0·69–1·94]). Globally, bullying victimisation accounted for 5·01% (3·40–6·83) of all anxiety disorder DALYs and 3·68% (2·37–5·21) of all major depressive disorder DALYs.

71In 2017, high SBP was the leading risk factor globally, accounting for 10·4 million (95% UI 9·39–11·5) deaths and 218 million (198–237) DALYs. Overall, 8·74% (7·78–9·68) of total DALYs were attributable to high SBP. Most of the burden attributable to high SBP was due to ischaemic heart disease and stroke, and high SBP accounted for 55·7% (48·1–63·1) of DALYs due to ischaemic heart disease and 57·1% (49·4–63·8) of DALYs due to stroke. Globally, age-standardised DALY rates attributable to high SBP declined between 1990 and 2017 (25·8% decrease [24·3–27·5]). Notable declines occurred in the high-income (54·5% [53·1–56·0] decline), north Africa and Middle East (32·6% [28·7–35·7]), and Latin America and Caribbean (32·4% [30·7–34·1]) super-regions.

72In 2017, smoking was the second leading risk for both deaths and DALYs. It was responsible for 7·10 million (95% UI 6·83–7·37) deaths (12·7% [12·2–13·2] of all deaths globally) and accounted for 182 million (173–193) DALYs (7·31% [6·81–7·80] of all DALYs globally). Although smoking-attributable age-standardised mortality rates declined 38·2% (36·4–40·0) between 1990 and 2017, the total number of smoking-attributable deaths has increased by 24·9% (21·3–28·6). In 2017, smoking accounted for a larger proportion of deaths among men in high-middle SDI countries (27·1% [26·3–27·9]) than in low SDI countries (10·2% [9·23–11·0]; appendix 2). Between 1990 and 2017, among men, the percentage of all-cause all-age deaths attributable to smoking in high SDI countries decreased, whereas it increased in all other SDI groups. Levels and trends in smoking prevalence and attributable burden vary by gender. As a risk factor for all-cause all-age deaths, smoking ranks first among men and eighth among women, reflecting both the lower prevalence and lower intensity of smoking among women. In 2017, the four Level 3 causes with the largest number of deaths attributable to smoking were ischaemic heart disease (1·62 million [1·54–1·69] deaths), COPD (1·23 million [1·12–1·35] deaths), tracheal, bronchus, and lung cancer (1·19 million [1·15–1·23] deaths), and stroke (887 000 [833 000–944 000] deaths), and together account for 69·4% of total smoking-attributable deaths (table 3). For lung cancer and COPD, changing from indirect estimation using the smoking impact ratio to direct estimation using measures of cumulative lifetime exposure resulted in large changes in smoking-attributable burden estimates in countries with scarce or low-quality data on lung cancer mortality. At the all-cause level, using measures of continuous and cumulative exposure that take into account risk among former and occasional smokers has resulted in increases in attributable burden in 71·8% of countries compared with previously published estimates.

73In 2017, low birthweight and short gestation was the third leading Level 3 risk factor globally for all-age DALYs. Age-standardised DALY rates attributable to this risk declined from 4558 (95% UI 4347–4739) DALYs per 100 000 in 1990 to 2680 (2543–2812) DALYs per 100 000 in 2017. Although exposure to low birthweight and short gestation decreased globally between 1990 and 2017, most of the declines in risk-attributable burden came from declines in mortality from causes attributable to this risk. For low birthweight, the biggest improvements in risk-attributable mortality were in Saudi Arabia, Maldives, and Singapore (appendix 2). Increasing SDI was associated with decreasing exposure and decreasing attributable burden. Exposure and attributable burden were highest in eastern, central, and western sub-Saharan Africa. For short gestation, the biggest improvements were in Saudi Arabia, Maldives, and Turkey. Similar to low birthweight trends, increasing SDI was associated with a decrease in exposure, as well as a decrease in attributable burden. Exposure and attributable burden were highest in south Asia and eastern, central, and western sub-Saharan Africa.

74High BMI accounted for 4·72 million (95% UI 2·99–6·70) deaths and 148 million (98·6–202) DALYs globally in 2017 (table 3). It ranked fourth among risk factors for mortality, primarily due to the effect of high BMI on cardiovascular disease. Globally, the SEV for high BMI for both sexes has increased 70·4% (57·1–84·5) since 1990 (table 2), and trends among children suggest that this will continue to rise. The resulting disease burden is large and increasing, with the absolute number of DALYs attributable to high BMI increasing by 36·7% (31·5–44·5) between 2007 and 2017 and by 127% (99·5–172) between 1990 and 2017, and age-standardised DALY rates attributable to high BMI increasing 6·84% (2·60–13·0) between 2007 and 2017 and 19·3% (4·83–43·2) between 1990 and 2017. In 2017, high BMI was the leading cause of premature mortality, as measured by YLLs, in seven countries and territories: Ecuador, Peru, Kuwait, Qatar, Saudi Arabia, United Arab Emirates, and American Samoa. Among risk factors assessed in GBD 2017, BMI prevalence continues to have one of the highest rates of increase over time. Trends in BMI prevalence and burden exhibit marked geographical variation, with the highest levels among wealthier countries.

75Globally, high FPG was the third leading risk factor for both mortality and DALYs in 2017, accounting for 6·53 million (95% UI 5·23–8·23) deaths and 171 million (144–201) DALYs. Cardiovascular disease was the leading Level 2 cause of death due to high FPG, accounting for 54·5% of deaths attributed to the risk. Diabetes and kidney diseases was the second ranking Level 2 cause of disability attributable to high FPG and accounted for 51·9% of DALYs attributable to this risk.

76In 2017, 5·25% (95% UI 4·49–6·01) of all deaths were attributable to ambient particulate matter pollution, making it the eighth leading risk for deaths, with a total of 2·94 million (2·50–3·36) deaths globally, and 1·05 million (0·88–1·22) deaths in southeast Asia, east Asia, and Oceania. It has increased from 15th leading risk for deaths in 1990, with 1·75 million (1·48–2·03) deaths, with a percentage change of 67·7% (56·6–79·3). It is among the top ten ranked Level 4 risk factors for deaths in 95 of the 195 countries and territories for which we produce estimates. Mortality rates attributable to ambient particulate matter pollution are notably high in Bahrain (11·9% [10·2–13·3] of all deaths were attributable to ambient particulate matter; fourth leading risk factor for deaths), Egypt (11·8% [9·07–14·4]; sixth leading risk factor), and Kuwait (10·6% [8·94–12·3]; sixth leading risk factor). The Level 3 causes with the largest proportion of burden attributable to ambient particulate matter pollution are COPD and lower respiratory infection: ambient particulate matter is responsible for 19·3% (12·6–25·4) of all COPD DALYs, 17·4% (13·7–21·6) of all lower respiratory infection DALYs, and 17·5% (13·2–22·7) of lower respiratory infection DALYs in children younger than 5 years. In GBD 2017, we added type 2 diabetes as an outcome of ambient particulate matter pollution, which contributes 10·5 million (6·70–13·9) or 12·6% of the 83·0 million (71·4–94·3) DALYs attributable to ambient particulate matter pollution in 2017 (table 3).

77In 2017, 4·32 million (95% UI 3·33–5·44) deaths and 94·9 million (78·8–112) DALYs were attributable to high LDL cholesterol. Overall, 3·80% (3·14–4·56) of total DALYs were attributable to high LDL cholesterol. The age-standardised DALY rate in males (1480 per 100 000 [1230–1750]) was higher than in females (880 per 100 000 [714–1070]). Globally, between 1990 and 2017, age-standardised DALY rates declined markedly (29·9% [28·1–31·7] decline). Notable reductions in age-standardised DALY rates occurred in both the high-income (61·8% [60·4–63·2] decline) and Latin America and Caribbean super-regions (40·7% [38·9–42·5] decline).

78Globally in 2017, dietary risks were the leading Level 2 risk factor for deaths and the second leading Level 2 risk factor for DALYs, accounting for 10·9 million (95% UI 10·1–11·7) deaths and 255 million (234–274) DALYs (table 3). Among dietary factors, high intake of sodium was the leading risk for mortality, accounting for 3·20 million (1·42–5·45) deaths. Low intake of whole grains was the leading risk factor for DALYs, accounting for 82·5 million (59·0–109) DALYs. Diet with high sugar-sweetened beverage consumption continues to show a marked increasing trend in overall attributable burden, with SEVs for this risk increasing 12·1% (7·02–18·2) between 2007 and 2017 and 17·1% (8·34–28·0) since 1990 (table 2).

79We estimated the burden of disease attributable to 84 metabolic, environmental and occupational, and behavioural risk factors or clusters of risks from 1990 to 2017 in 195 countries and territories, based on 46 749 data sources. In 2017, all included risks combined contributed to 61·0% (95% UI 59·6–62·4) of deaths and 48·3% (46·3–50·2) of DALYs worldwide, compared with 61·4% (59·9–62·7) of deaths and 52·4% (50·4–54·5) of DALYs in 1990. With each iteration of GBD, including GBD 2017, an increasing proportion of burden of disease has been attributed to risk factors. The contributions of risk factors to total deaths in the year 2010 (the most recent year for which we have comparable estimates from the most GBD iterations) were estimated to be 55·7% in GBD 2013, 57·3% in GBD 2015, 60·2% in GBD 2016, and 61·3% in GBD 2017. Similarly, in 2010, the contribution of risk factors to DALYs was 40·4% in GBD 2013, 41·5% in 2015, 46·5% in 2016, and 49·5% in 2017. This increase stems from the growing collection of risk–outcome pairs included in each iteration. With the addition of new risk factors and new risk–outcome pairs, future iterations of the GBD study should explain an increasing portion of the remaining unattributed burden of disease.

80The role of changes in risk factors in explaining changes in deaths and DALYs varies considerably across causes and ages. Since 1990, in the 64 individual or Level 4 risks, exposure levels increased significantly for 20 risks, did not change significantly for 14 risks, and decreased significantly for 31 risks. The risks with the highest increases in SEVs globally include high BMI, ambient particulate matter pollution, and high FPG; the risks with the largest decreases in exposure are unsafe sanitation, diet high in trans fatty acids, and household air pollution. These trends are consistent with what are expected with socioeconomic development.

81We found considerable heterogeneity across super-regions in the leading risk factors. Some notable patterns are the role of unsafe sexual practices as a driver of the HIV/AIDS epidemic in eastern and southern sub-Saharan Africa and the role of alcohol consumption in eastern Europe and central Asia. There are also marked spatial patterns for other risks such as high BMI in central America, north Africa and the Middle East, and Oceania. Interpretation of spatial patterns needs to take into account the fact that some risks have a strong relationship with socioeconomic development. The numbers of some environmental and behavioural risks, including unsafe water, sanitation, and handwashing, household air pollution, and childhood growth failure, decrease rapidly with increased development. Other risks tend to increase with development, including high BMI, high SBP, red meat consumption, high sugar-sweetened beverage consumption, and drug and alcohol use.

82Risks form clusters with regard to the nature of their association with development. In this regard, we observe four categories of risks as follows: (1) risks positively associated with SDI; (2) risks negatively associated with SDI; (3) risks with clear, but non-monotonic associations with SDI; and (4) risks that show little association with SDI.

83The category of risks that are positively associated with SDI includes smoking, high BMI, alcohol use, and high LDL cholesterol. This category includes predominately behavioural and metabolic risks that are associated with NCDs. The category of risks that are negatively associated with SDI includes child wasting, unsafe water source, diet low in fruits, household air pollution, unsafe sanitation, and no access to handwashing facility. This category includes predominately environmental and behavioural risks that are associated with CMNNDs. The category of risks that have non-monotonic associations with SDI includes high FPG and ambient particulate matter pollution, which both follow an inverted U-shaped association with development, and low birthweight, which has a U-shaped association with development. The category of risks showing little association with SDI includes high SBP and diet low in whole grains.

84Notably, for some risks, the trends of O/E ratios and SEVs are in opposite directions. For example, O/E ratios for household air pollution and unsafe water increased between 1990 and 2017, indicating that, for a given level of SDI, exposure levels have increased during that period. Conversely, SEVs for those risks declined during the same period. Thus, although exposure to these risks is broadly improving with socioeconomic development and time, improvements in development are occurring more rapidly than are improvements in the underlying risk structure in a population. For other risks, SEVs show temporal trends that run counter to trends expected on the basis of SDI. For example, although SDI was strongly positively associated with smoking, both SEVs and O/E ratios for smoking have declined despite concurrent increases in SDI. The marked declines in O/E ratios for smoking across all super-regions again suggest that the expected increase in smoking has not been fully realised with increasing SDI. In the case of smoking, the declining O/E ratios probably reflect benefits of smoking-reduction policies. Similarly, alcohol use was strongly positively associated with SDI, and trends in alcohol use SEVs did not change greatly, resulting in reductions in O/E ratios for alcohol use in all super-regions except south Asia.

85In addition to spatiotemporal variation in risk exposure, there were important differences between males and females within a given location. Females had notably higher SEVs for household air pollution, second-hand smoke, low bone mineral density, and diet low in fibre, legumes, calcium, and vegetables; whereas males had notably higher SEVs for lead exposure, smoking, alcohol use, ambient particulate matter pollution, red meat consumption, and most occupational risks.

86For smoking, SEVs were higher for males than for females at all SDI levels, and, for both sexes, SEVs increased with increasing SDI. However, although smoking SEVs for males increase nearly linearly with SDI, smoking SEVs for females remain very low at lower SDI, and then increase rapidly between SDI values of 0·6 and 0·9. This suggests that smoking interventions that target males might be most effective in lower SDI settings, whereas interventions in higher SDI settings should target males and females more equally. Similarly, alcohol use remains low among females across lower and middle SDI levels but increases with SDI at higher SDI levels, suggesting that interventions should target males in low to middle SDI settings, but be targeted more broadly in higher SDI settings.

87SEVs for high BMI increased for both sexes between 1990 and 2007, 2007 and 2017, and 1990 and 2017. Although this increase continues for males across the entire range of SDI, the increase for females slows and the curve flattens at an SDI of 0·66. Consequently, although females have higher BMI SEVs than males at low and middle SDI, they have lower SEVs than males at high SDI. Similarly, there was a crossover in sex differences for high SBP. Although high SBP SEVs for females decline slightly with SDI, they increase slightly with SDI for males. Consequently, expected SBP SEVs are lower for males than for females where SDI is 0·36 or below, and they are higher for males than for females where SDI is greater than 0·36.

88The number of all-cause DALYs declined for both males and females between 2007 and 2017, because the effects of improvements in both risk exposures and risk-deleted DALY rates have outpaced the opposing effects of population growth and ageing. Decreases in risk-deleted DALY rates are probably a function of improvements in treatments and improvements in risk factors that are not captured in the GBD analysis, as well as uncaptured social, cultural, and economic factors. Although changes in environmental and occupational and behavioural risks have driven declines in all-cause DALYs, increasing exposure to metabolic risks has had the opposite effect. This pattern is remarkably consistent across geographies. Consequently, increasing exposure to metabolic risks stands as one of the key drivers of increasing DALYs from NCDs. Although concerns about NCD burden have historically focused on high-income countries, previous iterations of the GBD study, as well as other studies, have clearly shown the large and growing problem of NCDs in low-income and middle-income settings., Similarly, although metabolic risk factors such as high LDL cholesterol, high SBP, and high BMI have historically been viewed mostly as challenges in high-income settings, these key metabolic risk factors are now commonly increased even in low-income settings. Between 1990 and 2017, 47 countries and territories had greater than 100% increases in SEVs for high BMI and, of these, 37 (79%) had an SDI that was less than the median value in 1990. Although these increases have occurred in connection with increasing SDI, increasing O/E ratios show that trends in these metabolic risks are outpacing the expectations based on development.

89High LDL cholesterol and high SBP are among the leading risk factors for all-cause risk-attributable burden in 2017. Both are commonly increased even in low-resource settings, and both are increased by obesity. Although both are also easily reduced or treated with inexpensive and cost-effective medications, these treatments remain inaccessible to broad segments of the world's population. Growing evidence from studies such as the SimCard Trial in rural India and China suggests that effective methods exist to deliver these kinds of important pharmacotherapy, even in the most resource-limited locations. A 2018 trial of a fixed low-dose triple-combination antihypertensive drug tested in urban Sri Lanka, consisting of amlodipine, telmisartan, and chlorthalidone, showed substantial improvement in control of blood pressure compared with usual care at 6 months, and no difference in adverse events. Another trial from 2016 showed that lowering LDL cholesterol with a high-potency statin was effective even among individuals only at intermediate risk of cardiovascular events. Efforts to improve universal health coverage must directly address how these therapies can be delivered even more effectively. Studies have further identified public health interventions that are cost-effective outside of high-income settings. Legislative interventions targeting sodium and trans fat, improvements in food labelling and advertising, and health-focused and diet-focused media campaigns are cost-effective interventions in south Asia, for example. Fortunately, philanthropic donors are beginning to identify cardiovascular risk as an important target for improving global health.

90O/E ratios offer a useful benchmarking tool for national policy makers who are hoping to compare the risk exposure against an empirical standard for countries at a given level of development. Although the burden attributable to a given risk indicates the risks for which exposure reduction can offer the greatest health gains, O/E ratios offer insight into what might be achievable given the resource constraints that exist at a certain level of development. For example, in the southeast Asia, east Asia, and Oceania super-region, high SBP and high FPG are among the leading risks, but both have SEV O/E ratios below 1·0. This suggests that, despite the high risk-attributable burden, the countries in this super-region show better than expected performance in managing exposure to these risks. Conversely, ambient particulate matter pollution and household air pollution are the tenth and 15th leading risks in the super-region, but the corresponding SEV O/E ratios are 1·32 and 3·05, suggesting that the countries in the super-region have greater exposure to these risks than other countries at similar levels of development. The degree of underperformance suggests that gains in managing these risks might be more easily achieved than would be gains from risks in which the super-region is already outperforming its peers.

91GBD 2017 marks the first time that bullying victimisation by peers of children and adolescents attending school has been included as a risk factor. It is one of only three psychosocial risk factors to be included in GBD, one of only two risk factors for major depressive disorder, and the first-ever risk factor for anxiety disorders in GBD. The substantial proportion of major depressive disorder and anxiety disorder YLDs attributable to bullying victimisation highlights the importance of considering psychosocial risk factors when assessing health outcomes. The small number of risk factors for mental disorders has been an ongoing criticism of GBD, especially considering that mental disorders are a leading cause of disability worldwide. GBD estimates suggest that the prevalence of mental disorders has changed little in the past few decades despite growing recognition of the associated burden. Psychosocial risk factors can be challenging to assess and quantify, which in turn can make them easy to overlook. However, bullying victimisation has an operationalised definition, can be measured in population surveys, and is consistently associated with risk of mental disorders., Scientifically assessed school-based interventions that address bullying show positive programme effects. A reduction in the prevalence of bullying victimisation could help reduce the prevalence, and therefore burden, of major depressive disorder and anxiety disorders later in life. Crucially, the inclusion of bullying victimisation as a risk factor in GBD might motivate investment in intervention programmes to reduce bullying. The inclusion of bullying victimisation in GBD also shows the importance of integrating cross-disciplinary research in the identification of the complex aetiology of mental disorders.

92Using continuous and cumulative measures of smoking, which capture both former and occasional smokers, has shown that the burden of smoking is higher than previously thought. The implications of the new methods on attributable burden varies across outcomes and locations. For 71·8% of countries, the burden attributable to smoking has increased as a result of the new methods. For the largest causes of burden (ischaemic heart disease, COPD, lung cancer, and stroke) there are differential trends across countries and by sex. By quantifying this dose-response relationship, we are able to highlight the potential health benefits of policies and programmes that reduce smoking. We have also produced a globally complete time series of estimates for age of initiation that shows that the average age of initiation has changed little across time, even in locations with strong and successful tobacco control. Targeting interventions to reduce the initiation rate in the critical age window of 13–23 years will yield substantial health benefits. Finally, the concept of capturing exposure over an individual's life course can and should be extended to other risk factors, for which the health effects are probably the result of cumulative exposure.

93The GBD study is the most comprehensive population-level CRA across countries and risks. For several risks, including ambient particulate matter pollution, household air pollution, intimate partner violence, unsafe water source,, breastfeeding, and lead exposure, GBD estimates are generally lower than published WHO estimates.,,,, These discrepancies can be attributed to different definitions, methods, granularity, and input data. For some findings, annual estimates might disagree, but regional patterns are consistent between WHO and GBD. UNICEF produces estimates for child stunting that are lower than GBD estimates, with some disagreement in which locations progress has been made. There is more consistency in estimates between UNICEF and GBD for child wasting and child underweight. GBD estimates for the prevalence of low birthweight and short gestation are slightly lower than the WHO estimates but show similar geographical patterns. Scientific literature reveals similar results to GBD for impaired kidney function and low birthweight and short gestation. Research published on iron-deficiency anaemia differs from GBD in methods and definitions, resulting in generally higher GBD estimates. GBD estimates were much lower than published research on occupational estimates,,, largely because of different cause–outcome pairs and GBD's application of the CRA approach.

94Improvements in our smoking estimation methods have resulted in higher estimates of smoking-attributable burden. Although smoking estimates from previous iterations of the GBD study were slightly lower than those published by WHO, our GBD 2017 estimates show marked similarities. Among the 142 countries and territories included in the WHO report and estimated in GBD, the correlation coefficient for 2017 smoking prevalence estimates for females was 0·91 and for males was 0·85. Where estimates diverge, these differences can be attributed to differing modelling methods or data sources. For example, the WHO model was fit on 1175 country-year data sources whereas the GBD model was fit on 2870 country-year data sources.

95As with previous iterations of GBD, we have continued to use the World Cancer Research Fund criteria of convincing or probable evidence as the threshold for including a risk–outcome pair in GBD 2017. The subjectivity of some aspects of these definitions might result in differing interpretations of the available evidence and different conclusions about the strength of the evidence supporting the inclusion of a given risk–outcome pair. Moving toward more objective and quantifiable criteria could help reconcile these different interpretations. Quantifiable criteria would also permit simplified evidence scores that could help policy makers who, faced with scarce resources, might choose to prioritise interventions that address risks supported by the strongest evidence base. Objective and quantifiable criteria would facilitate such comparisons, allowing stakeholders to more easily understand the variable strength of evidence supporting the causal connection for each risk–outcome pair. In an effort to summarise the evidence base for each risk–outcome pair in GBD 2017, we present the number of studies that are available, by type of study design. However, we recognise that quantifying the strength of evidence requires moving beyond simple quantification of the number of studies. To that end, we have begun to extract detailed information from all 3638 studies used across risk–outcome pairs. This effort includes details about study design, the study population and sampling, exposure measurement, outcome assessment, potential sources of bias and confounding, and analytic efforts to address bias and confounding. With this information, we can systematically analyse the degree to which different study limitations affect the resulting strength of associations, allowing us to objectively assess study quality. Looking at the studies for a given risk–outcome pair, we can then calculate an objective evidence score based on the number and quality of the supporting studies, and the potential for publication bias. Such a score would facilitate evidence-based decision making in the face of scarce political and financial resources, allowing attention to be focused on risks with both the greatest attributable burden and strongest supporting evidence.

96The inclusion of bullying as a new risk in GBD 2017 suggests potential areas for future work. We hope that its inclusion will help in setting a higher standard of measurement in future research on both the occurrence and health effects of bullying. It also provides a starting point for including additional risk factors related to violence against children, such as non-sexual forms of child abuse in addition to childhood sexual abuse, which is the only GBD 2017 risk factor pertaining to child maltreatment.

97A growing body of research has shown direct connections between ambient temperature and health outcomes. For example, the Multicountry Observational Study, the largest study of all-cause mortality risk and ambient temperature to date, estimated that 7·71% (95% CI 7·43–7·91) of deaths were attributable to suboptimal ambient temperature: 7·29% (7·02–7·49) of deaths were attributable to low ambient temperatures and 0·42% (0·39–0·44) were attributable to high temperatures. The growing evidence base, especially in the context of global concerns about the potential health effects of climate change, makes the inclusion of meteorological risk factors a priority for future iterations of GBD. We have already begun work to estimate the cause-specific burden directly attributable to effects of ambient temperature, with plans to include suboptimal ambient temperature as a risk in future estimations of GBD. The inclusion of exposure to suboptimal temperature is intended also to lay the groundwork for exploring other components of weather and climate in future GBD analyses.

98A study of this scope has many limitations. The new proportional PAF strategy that we have implemented for particulate matter pollution reflects a broader challenge with disentangling inherently connected risks with non-linear relationships between exposure and disease risk. In the case of particulate matter pollution, risk for a given outcome rises sharply with increased exposure at the low end of the exposure range, and the marginal increase in risk declines as the risk curve flattens at the upper end of the exposure range. Two risks compose particulate matter pollution: ambient particulate matter pollution and household air pollution. If exposure to each of these two sources were sequentially reduced to the TMREL, the reduction in burden per μg/m3 reduction in exposure for whichever source is removed first will be less than the reduction for the source that is removed second. This occurs because, for the first source, the exposure reduction occurs where the risk curve is flatter, whereas for the second source, exposure reduction occurs where the risk curve is steeper. No clear solution to this problem that does not raise a new problem exists within the CRA framework. Treating each exposure in isolation, as we did in previous GBD iterations, results in overestimates of attributable burden, because both risks will have their burden estimated from the steepest part of the risk curve. Considering both exposures together and treating each exposure as the first to be removed, in line with the CRA approach, results in underestimates of risk, because both risks will have their burden estimated from the flattest part of the risk curve. We have chosen instead to estimate burden for total particulate matter pollution and divide this total burden proportionately between the two sources. This accurately captures burden but deviates somewhat from the CRA approach. The challenge extends beyond ambient and household particulates. Smoking and second-hand smoke are additional sources of particulate matter exposure that we have not yet incorporated into our proportional burden strategy. This is a limitation of the CRA approach, and other areas probably have similar challenges with non-linear and inherently connected risks that have not yet been identified and corrected.

99Bullying is characterised by intention to harm, a power imbalance, and repetition; however, studies vary greatly with regard to how they specifically define these criteria. For example, we were unable to include many studies where the threshold for repetition was either insufficiently specific or not reported. Second, the relative risk data used in the calculation of PAFs were only from high-income countries. The magnitude of the association between bullying victimisation and outcomes might differ between high-income countries and low-income and middle-income countries.

100Our updated continuous measures of smoking better capture true risk, but also introduce new challenges. First, because of data limitations, we assume the risk of all smoked tobacco products is equivalent to the risk of cigarettes. Second, smoking histories were reconstructed on the basis of cross-sectional survey data, not longitudinal data. Third, we were not able to effectively adjust for illicit tobacco in our estimates of supply-side consumption. Fourth, we have not incorporated the effects of tobacco-control policies on various exposures. Consequently, if no survey data are collected after implementation of a tobacco-control policy, the impact of that policy will not be reflected in the modelled estimates. Fifth, insufficient evidence is available to estimate risk-reduction curves for individuals with different smoking histories, so the rate of risk reduction has been assumed to be the same for all levels of exposure. Sixth, we recognise that our exposure definitions are summaries of an individual's smoking history, and that more nuanced exposure definitions might perform better for individual-level risk prediction.

101This study provides a comprehensive and comparable assessment of 84 environmental and occupational, behavioural, and metabolic risks across locations and time. By quantifying levels and trends in exposures to risk factors and the resulting disease burden, this assessment offers insights into past programme and policy successes and highlights the current priorities for public health action. Our findings show that improvements in behavioural, environmental, and occupational risks, and decreases in risk-deleted DALY rates have largely outpaced the effects of population growth and ageing, in terms of absolute burden. Sustained efforts in reducing these risks are crucial to maintaining trends in declining burden from injuries and CMNNDs. Conversely, the combination of increasing metabolic risks and population ageing will probably continue to drive the increasing trends in NCDs at the global level, which presents both a public health challenge and opportunity. Working down from the global to national level reveals considerable heterogeneity in levels of risk exposure and risk-attributable burden. Although development underlies some of this heterogeneity, ratios of observed burden to the burden expected based on SDI reveal risks for which countries are overperforming or underperforming relative to their level of socioeconomic development. As such, these ratios provide a benchmarking tool to help focus local decision making. Our findings reinforce the importance of both risk exposure monitoring and epidemiological research to assess causal connections between risks and health outcomes, and they highlight the essential role of GBD in synthesising data to draw the comprehensive and robust conclusions that are necessary to inform evidence-based policy.

102Correspondence to: Prof Christopher J L Murray, Institute for Health Metrics and Evalution, Seattle, WA 98121, USA cjlm@uw.edu

103Correspondence to: Prof Christopher J L Murray, Institute for Health Metrics and Evalution, Seattle, WA 98121, USA cjlm@uw.edu