Global, regional, and national incidence, prevalence, and years lived with disability for 354 diseases and injuries for 195 countries and territories, 1990–2017: a systematic analysis for the Global Burden of Disease Study 2017

1Measuring non-fatal health loss is one of the most complex endeavours in population health research. The evolution of modern health-care systems has led to an increasing number of diseases and injuries being diagnosed and treated in individual patients, and developments such as antihypertensive and statin medications, percutaneous coronary intervention, and antiretroviral therapies have led to averted deaths and longer lives. In parallel with the increasing complexity of clinical medicine in the past century, measuring non-fatal health loss has necessitated continuous refinement as diagnostic classification systems expand, new diseases emerge, and metrics of disability improve. Across the global landscape, increased non-fatal health loss paradoxically reflects both success in terms of diminishing rates of premature death but also failure in terms of maintaining health care for diseased and injured individuals. It is increasingly evident that differential access to care, economic inequality, and imbalanced risk factor profiles can and do challenge the ability of health systems to achieve equitable health outcomes in the face of complex and resource-draining diseases and injuries. Addressing such lapses in health equity can pose a burden to under-resourced health care systems and economies.

2Global progress in improving the burden of non-fatal health outcomes has been limited, in part by a predominant focus on mortality rates as a common metric of tracking global health progress.,, In the latter part of the 20th century, the global community focused on communicable diseases such as tuberculosis, HIV, malaria, and other conditions that cause premature mortality. In the past decade, it has become evident that measuring non-fatal health loss is important for tracking progress as the disease burden, in terms of disability-adjusted life-years (DALYs), evolves toward being dominated by years lived with disability (YLDs). Transitions in ageing populations and reduced mortality in many areas of the world have created dynamic temporal patterns, particularly within the past decade, and measuring such time patterns is important because advents such as developing a cure for hepatitis C, discovering new therapies for cancer, and improving treatments for HIV can rapidly transform the burden in populations with access to these developments, and as conditions such as diabetes and non-alcoholic fatty liver disease become increasingly prevalent in lower-income countries.

3Evidence before this studyThe Global Burden of Diseases, Injuries, and Risk Factors (GBD) study is a comprehensive study of health loss designed to capture complex patterns of disease and injury burden; for non-fatal health outcomes, these are measured in terms of incidence, prevalence, and years lived with disability (YLDs). Previous versions of the study have increased the estimation detail for conditions, locations, ages, and years. This study is a reassessment of the incidence, prevalence, and YLDs of diseases and injuries from 1990 to 2017 and updates results from previous GBD studies. There are no alternative measurements of non-fatal health loss that include the level of detail provided in the GBD study.Added value of this studyThis study adds new knowledge on non-fatal burden globally and improves upon previous iterations of the GBD study in the following ways. We expanded our database of non-fatal health outcomes by adding 2842 collaborator-provided data sources and incorporating new clinical data representing an additional 149 million admissions and 3·7 billion outpatient visits for use in GBD modelling. This resulted in a total of 68 781 sources being used in the estimation process for GBD 2017. We improved estimation methods including updating the calculation of the Socio-demographic Index (SDI), adding the ability to report the statistical differences in non-fatal health outcomes for males and females, using internally consistent GBD estimates of population and fertility, and adopting several cause-specific modelling improvements. Cause-specific improvements included the following; for diarrhoea, we added additional literature informing aetiological attribution; for HIV/AIDS, we updated absolute neutrophil count bias adjustments, antiretroviral therapy coverage data, and sex-specific survey estimates. For hepatitis, we added case fatality rates and hepatitis B vaccine coverage to viral hepatitis incidence models. For maternal, neonatal, and child health causes, we added in-facility delivery rates to the inpatient admission per-capita estimates to more accurately measure the denominator for incident cases and expanded the age range affected by protein-energy malnutrition. For cancer, we applied mortality-incidence ratios directly to cause-specific mortality rates to estimate incidence, and then calculated prevalence on the basis of incidence and survival estimates. For mental and substance abuse disorders, we adopted new covariates for opioid use and updated autism spectrum disorder designations to be consistent with the most recent Diagnostic and Statistical Manual of Mental Disorders. We also added 19 new causes to our cause hierarchy, including type 1 and type 2 diabetes, chronic kidney disease due to type 1 diabetes, and chronic kidney disease due to type 2 diabetes; cirrhosis due to non-alcoholic steatohepatitis (NASH); liver cancer due to NASH; invasive non-typhoidal salmonella; myelodysplastic, myeloproliferative, and other haemopoietic neoplasms; subarachnoid haemorrhage; non-rheumatic valvular heart disease including calcific aortic and degenerative mitral subtypes; aggregates of vision disorders and hearing loss; poisoning by carbon monoxide; poisoning by other means; and estimates for natures of injury (eg, fractures).Implications of all the available evidenceGlobal non-fatal burden is continuing to increase despite minor improvements in age-standardised rates. Three causes (low back pain, headache disorders, and depressive disorders) have prevailed as leading causes of non-fatal health loss for nearly three decades, while diabetes has emerged as the fourth leading cause of disability globally. The increase in YLDs reflects an ageing global population commensurate with declines in premature mortality across the development spectrum. Globally, patterns of non-fatal health loss vary dynamically by sex, age, location, SDI, and cause. The increasing burden of non-fatal diseases, injuries, and impairments could pose considerable challenges to health systems and economies not equipped to care for complex and expensive conditions.

4Evidence before this study

5The Global Burden of Diseases, Injuries, and Risk Factors (GBD) study is a comprehensive study of health loss designed to capture complex patterns of disease and injury burden; for non-fatal health outcomes, these are measured in terms of incidence, prevalence, and years lived with disability (YLDs). Previous versions of the study have increased the estimation detail for conditions, locations, ages, and years. This study is a reassessment of the incidence, prevalence, and YLDs of diseases and injuries from 1990 to 2017 and updates results from previous GBD studies. There are no alternative measurements of non-fatal health loss that include the level of detail provided in the GBD study.

6Added value of this study

7This study adds new knowledge on non-fatal burden globally and improves upon previous iterations of the GBD study in the following ways. We expanded our database of non-fatal health outcomes by adding 2842 collaborator-provided data sources and incorporating new clinical data representing an additional 149 million admissions and 3·7 billion outpatient visits for use in GBD modelling. This resulted in a total of 68 781 sources being used in the estimation process for GBD 2017. We improved estimation methods including updating the calculation of the Socio-demographic Index (SDI), adding the ability to report the statistical differences in non-fatal health outcomes for males and females, using internally consistent GBD estimates of population and fertility, and adopting several cause-specific modelling improvements. Cause-specific improvements included the following; for diarrhoea, we added additional literature informing aetiological attribution; for HIV/AIDS, we updated absolute neutrophil count bias adjustments, antiretroviral therapy coverage data, and sex-specific survey estimates. For hepatitis, we added case fatality rates and hepatitis B vaccine coverage to viral hepatitis incidence models. For maternal, neonatal, and child health causes, we added in-facility delivery rates to the inpatient admission per-capita estimates to more accurately measure the denominator for incident cases and expanded the age range affected by protein-energy malnutrition. For cancer, we applied mortality-incidence ratios directly to cause-specific mortality rates to estimate incidence, and then calculated prevalence on the basis of incidence and survival estimates. For mental and substance abuse disorders, we adopted new covariates for opioid use and updated autism spectrum disorder designations to be consistent with the most recent Diagnostic and Statistical Manual of Mental Disorders. We also added 19 new causes to our cause hierarchy, including type 1 and type 2 diabetes, chronic kidney disease due to type 1 diabetes, and chronic kidney disease due to type 2 diabetes; cirrhosis due to non-alcoholic steatohepatitis (NASH); liver cancer due to NASH; invasive non-typhoidal salmonella; myelodysplastic, myeloproliferative, and other haemopoietic neoplasms; subarachnoid haemorrhage; non-rheumatic valvular heart disease including calcific aortic and degenerative mitral subtypes; aggregates of vision disorders and hearing loss; poisoning by carbon monoxide; poisoning by other means; and estimates for natures of injury (eg, fractures).

8Implications of all the available evidence

9Global non-fatal burden is continuing to increase despite minor improvements in age-standardised rates. Three causes (low back pain, headache disorders, and depressive disorders) have prevailed as leading causes of non-fatal health loss for nearly three decades, while diabetes has emerged as the fourth leading cause of disability globally. The increase in YLDs reflects an ageing global population commensurate with declines in premature mortality across the development spectrum. Globally, patterns of non-fatal health loss vary dynamically by sex, age, location, SDI, and cause. The increasing burden of non-fatal diseases, injuries, and impairments could pose considerable challenges to health systems and economies not equipped to care for complex and expensive conditions.

10Estimates reported in recent iterations of the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) have also illustrated differential health outcomes in males and females in certain locations and conditions. This topic has received attention in terms of mortality rates for sex-specific conditions such as maternal causes,,,,, gynaecological and breast malignancies,,,, and long-term complications of obstructed labour, such as obstetric fistula.,,,, GBD 2016 also highlighted how global, age-standardised, all-cause YLD rates are approximately 10% higher in females than males, emphasising how there may be sex-specific characteristics of the non-fatal burden that have not been explored in detail, particularly with respect to the differences in sex-specific health outcomes. It is increasingly of interest to measure differences in male and female non-fatal health loss.

11This year's GBD study represents the continued effort of quantifying non-fatal health outcomes in terms of incidence, prevalence, and YLDs for a list of 354 GBD causes for the years 1990–2017. Because the study is remeasured and published on an annual basis, new estimates are provided not only for new estimation years but also for all previous estimation years and supersede any previous results. This year's study on non-fatal burden incorporates improvements in study design, estimation strategy, and data availability, and focuses on areas of non-fatal burden that are emerging as topical issues in measuring and improving health outcomes. We also explore the patterns of non-fatal health loss over time and estimate the statistical differences in non-fatal health loss for males and females.

12The GBD study provides a standardised approach for estimating incidence, prevalence, and YLDs by cause, age, sex, year, and location. The study aims to use all accessible information on disease occurrence, natural history, and severity that passes a set of inclusion criteria. Our objective is to maximise the comparability of data, despite different collection methods or case definitions; to find a consistent set of estimates between data on prevalence, incidence, and causes of death; and to predict estimates for locations and causes with sparse or absent data by borrowing information from other locations and covariates.

13The study conducts annual updates to incorporate new causes and data (including published literature, surveillance data, survey data, hospital and clinical data, and other types of data) and to improve demographic and statistical methods. In this study, we apply different methods to utilise available data and to measure specific epidemiological patterns of each cause of non-fatal burden. Our standard approach uses the Bayesian meta-regression tool DisMod-MR 2.1. Subsequently, we use data for severity and the occurrence of particular consequences of diseases, or sequelae, to establish the proportion of prevalent cases experiencing each sequela. There are several classes of alternative approaches for estimating non-fatal health outcomes, including for injuries, cancers, HIV/AIDS, other infectious diseases, and neonatal disorders. Presented below is a high-level description of our study methods; the supplementary methods (appendix 1 section 4) provide further detail on inputs, analytical processes, and outputs and methods specific to each cause in GBD 2017.

14Analyses were completed using Python version 2.7, Stata version 13.1, or R version 3.3. Statistical code used for GBD estimation is publicly available online. All rates are expressed as age-standardised based on the GBD reference population unless otherwise specified. This study complies with the Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER) recommendations (appendix 1).

15GBD 2017 is based on a geographical hierarchy that includes 195 countries and territories grouped into 21 regions and seven GBD super-regions (appendix 1). 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); given space constraints these results are presented in appendix tables and figures instead of in the main text (appendix 2). Subnational estimates for countries with populations larger than 200 million people (measured using our 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. Cause-specific results for non-fatal estimates for GBD 2017 cover the years 1990–2017. A subset of areas in this analysis focuses on 1990, 2007, and 2017 to show changes over time to better inform policy assessments.

16GBD 2017 is the first time that estimation of fertility and population has been done within the GBD framework. Previously, the GBD study used external sources, for fertility and population estimates, which affect estimates throughout the GBD study, particularly estimates expressed in terms of population rates. The purpose of using internally derived demographic estimates is to ensure internal consistency across all GBD estimates. That is, mortality rates and fertility rates have to match population rate change such that there should be no births, deaths, or migrations that are not accounted for in our population estimates.

17In GBD 2017, we further refined the existing cause list, and added 19 new causes, increasing the number of estimated causes in GBD to 359 with 282 causes of death estimated and 354 causes of non-fatal health loss estimated. In the GBD study, causes and their sequelae are organised into hierarchical levels. Level 1 contains three broad cause groups: communicable, maternal, neonatal, and nutritional diseases (CMNN); non-communicable diseases (NCDs); and injuries. For non-fatal health estimates, there are 22 Level 2 causes, 167 Level 3 causes, and 288 Level 4 causes. We also report estimates for 3484 sequelae, nine impairments, and seven nature of injury aggregates.

18In GBD 2017, we report on 381 Level 5 sequelae. We have opted to include aggregate sequelae for GBD 2017 to foster more nuanced interpretations of groups of health outcomes that are relevant to policy makers and clinical users of the GBD. In addition, this reporting list allows for more detailed evaluation of aetiologies and outcomes from GBD causes.

19For the first time in the GBD study, we present the burden of injuries in terms of nature of injury as well as external cause of injury. Previously, we reported the incidence, prevalence, and YLDs of injuries expressed only in terms of what caused the injury—eg, those caused by falls. However, the burden that results from falls is experienced in terms of the bodily harm that the fall itself causes—eg, spinal injury or skeletal fracture. We have grouped the 47 nature of injury sequelae into seven combined categories that represent 1410 sequelae. The supplementary methods (appendix 1) includes the full GBD 2017 non-fatal reporting hierarchy from Level 1 to Level 6.

20The process for non-fatal estimation begins with the compilation of data sources from a diverse set of possible sources, which include 21 possible Global Health Data Exchange (GHDx) data types ranging from scientific literature to survey data to epidemiological surveillance data. Our collaborator network provided 2842 data sources for GBD 2017. We analysed 21 100 sources of epidemiological surveillance data (country-years of disease reporting) for GBD 2017 and 4734 sources of disease registry data. For non-fatal estimation, we did systematic data and literature searches for 82 non-fatal causes and one impairment, which were updated to Feb 11, 2018. Search terms used for cause-specific systematic reviews, inclusion and exclusion criteria, preferred and alternative case definitions, and study methods detailed by cause are available in the supplementary methods (appendix 1 section 4). This search process contributed to the use of 15 449 scientific literature sources and 3126 survey sources used in non-fatal estimation, reflecting our updated counting criteria for GBD 2017. Household survey data archived in the GHDx were systematically screened together with sources suggested by country-level experts, surveys located in multinational survey data catalogues, and Ministry of Health and Central Statistical Office websites. Primary data sources containing disease prevalence, incidence, mortality risk, duration, remission, or severity were then combined in the estimation process. The supplementary methods section provides further details on gold standard data sources, adjustments, correction factors, and standardisations employed when incorporating these different types of non-fatal data (appendix 1 section 4).

21In addition to data sources based on primary literature, surveys, and surveillance, the GBD study has used an increasing number of hospital discharge records, outpatient visit records, and health insurance claims to inform various steps of the non-fatal modelling process. This year, we received hospital discharge records for an additional 30 country-years, specifically discharge records from India (3 country-years), Iran (10), Japan (6), Jordan (1), Nepal (1), Brazil (2), China (1), and Italy (6); inpatient and outpatient claims from Taiwan (province of China); additional years of inpatient and outpatient claims from the USA; and inpatient claims from Singapore, representing an additional 148 842 107 hospital admissions globally and bringing the total number of admissions that inform GBD estimation to more than 2·6 billion. Additionally, we received 10 years of outpatient visit records from Norway, representing a total of 153 351 282 outpatient visits over a 10-year period. Overall, the study now uses hospital data from 335 country-years, outpatient visit data from 45 country-years, and health insurance claims data from 33 country-years between the USA, Taiwan (province of China), and Singapore. These data inform multiple cause models in various ways, mainly by providing incidence and prevalence estimates adjusted for readmission, non-primary diagnosis, outpatient utilisation, or a combination of the above, but also by estimating parameters such as case fatality rates, remission rates, procedure rates, and distribution of disease subtypes. The supplementary methods provide a more detailed description of how the clinical data adjustments are calculated and how admission and outpatient visit data are processed and utilised (appendix 1 section 2).

22In the supplementary methods (appendix 1), we show the geographical coverage of non-fatal data, both incidence and prevalence, for GBD 2017. In addition, we illustrate the non-fatal data density and availability for GBD 2017 from 1990 to 2017 by GBD region and year for each of the three Level 1 GBD cause groups. The GHDx provides the metadata for all sources used for non-fatal estimation.

23For GBD 2017, we modelled non-fatal disease burden using DisMod-MR 2.1, a meta-analysis tool that uses a compartmental model structure with a series of differential equations that synthesise sparse and heterogeneous epidemiological data for non-fatal disease and injury outcomes. Estimation occurred at the five levels of the GBD location hierarchy—global, super-regional, regional, national, and subnational—with results of each higher level providing guidance for the analysis at the lower geographical level. Important modelling strategy changes from GBD 2016 to GBD 2017 for specific causes, as well as further details on these causes and their respective models, can be found in the supplementary methods (appendix 1 section 4).

24Custom models were created if DisMod-MR 2.1 did not capture the complexity of the disease or if incidence and prevalence needed to be calculated from other data, or both. Further details of these custom models can be found in the cause-specific methods sections of the supplementary methods (appendix 1 section 4).

25Prevalence was estimated for nine impairments, defined as sequelae of multiple causes for which better data were available to estimate the overall occurrence than for each underlying cause: anaemia, intellectual disability, epilepsy, hearing loss, vision loss, heart failure, infertility, pelvic inflammatory disease, and Guillain-Barré syndrome. Different methodological approaches were used for each impairment estimation process; these details are described in the supplementary methods (appendix 1 section 4).

26Severity splits apply a set of proportions that represent the distribution of cases of a given non-fatal cause by its underlying severities. Severity splits are typically categorised as asymptomatic, mild, moderate, and severe. This distinction is important for conditions such as asthma that have a broad spectrum of symptomatic severities. Severity splits for most conditions use the Medical Expenditure Panel Survey (MEPS) data or literature sources identified through systematic reviews. Further detail on the severity splits for each cause, including changes from GBD 2016, are available in the cause-specific modelling write-ups in the supplementary methods (appendix 1 section 4).

27Disability weight estimation is described in more detail elsewhere in the literature, but in summary, these represent the severity of health loss associated with a single given health state. The supplementary methods (appendix 1) provide a complete listing of the lay descriptions of all 234 health states used in the estimation of non-fatal results for GBD 2017.

28A combined disability weight is required to account for individuals with more than one condition. To calculate a combined disability weight, the health loss associated with two disability weights are multiplied together and then a weighted average of each constituent disability weight is calculated. The adjusted disability weight is proportional to the magnitude of the original disability weight. A simulation of 40 000 distinct individuals is done that calculates the distribution of comorbid conditions on the basis of the expected distribution of each condition's sequelae in the population. Then, the resulting distributions of comorbidity-adjusted disability weights are used to calculate YLDs. This process did not change from GBD 2016.

29YLDs were estimated as the product of prevalence estimate and a disability weight for health states of each mutually exclusive sequela, adjusted for comorbidity as described above. The GBD cause hierarchy also includes 35 residual disease categories to capture YLDs from conditions that lack specific estimation models.

30We apply the same technique for propagating uncertainty as used elsewhere in the GBD study design.,, The distribution of every step in the computation process is stored in 1000 draws that are used for every other step in the process. The distributions are determined from the sampling error of data inputs, the uncertainty of the model coefficients, and the uncertainty of severity distributions and disability weights. Final estimates are computed using the mean estimate across 1000 draws, and the 95% uncertainty intervals (UIs) are determined on the basis of the 25th and 975th ranked values across all 1000 draws.

31The Socio-demographic Index (SDI) is a summary measure that estimates a location's position on a spectrum of development. The SDI was originally constructed for GBD 2015 using the Human Development Index (HDI) methodology, wherein a 0–1 index value was determined for each of the original three covariate inputs (total fertility rate in women aged 15–49 years, educational attainment over the age of 15 years, and lag-distributed income per capita) using the observed minima and maxima over the estimation period to set the scales. In response to feedback from collaborators, we have refined the indicator with each GBD cycle. For GBD 2017, we replaced the total fertility rate with the total fertility rate in women under the age of 25 years. The GBD 2017 Population and Fertility analysis of age-specific fertility rates revealed that through the process of development, many countries exhibited a decline in age-specific fertility rates over the age of 30 years and then increased, creating a U-shaped pattern; however, age-specific fertility rates in ages 10–14 years, 15–19 years, 20–24 years, and total fertility under 25 years did not exhibit this pattern. Total fertility under 25 years remains highly correlated with mortality measures including under-5 mortality rates (Pearson's correlation coefficient r=0·873), and results from this revised method for computing SDI and results from GBD 2016 are also correlated (r=0·992). We computed the composite SDI as the geometric mean of the three indices for each location-year. The cutoff values used to determine quintiles for analysis were then computed using country-level estimates of SDI for 2017, excluding countries with populations of less than 1 million. These quintiles are used to categorise and present GBD 2017 results on the basis of sociodemographic status. The SDI values ranged from a low of 0·191 in Niger to a high of 0·918 in Denmark in 2017. Additional details on and results from the SDI calculation are available in the supplementary methods (appendix 1 section 2).

32The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing the report. All authors had full access to the data in the study and had final responsibility for the decision to submit for publication.

33Non-fatal estimates by cause for 354 causes and nine impairments for the years 1990, 2007, and 2017 are available by age and sex through the online results tool. Results and findings mentioned in the discussion can also be viewed interactively through an online data visualisation tool.

34Figure 1 shows the data density in terms of site-years by GBD region, cause group, and year. The figure shows how data density generally improves over time and how certain regions, particularly higher income regions, are more data dense than others. Additionally, the figure highlights how injuries data are generally less available than for the other two cause groups.

35Table 1 reports cause-specific global estimates of prevalence, incidence, and YLDs for causes at Levels 1–5 of the GBD hierarchy for 2017, as well as the percentage change in YLDs and age-standardised YLD rates between 1990, 2007, and 2017. Unless otherwise specified, all rates reported in this analysis are age standardised.

36For all ages and both sexes combined, globally, in 2017, the three most common causes at Level 3 of the GBD cause hierarchy in terms of all-age prevalent cases were oral disorders (3·47 billion, 95% UI 3·27–3·68), headache disorders (3·07 billion, 2·90–3·27), and tuberculosis including latent tuberculosis infection (1·93 billion, 1·71–2·20; table 1).

37Global age-standardised prevalence rankings remained unchanged for the top two Level 3 causes in the GBD hierarchy from 1990 to 2017, with oral disorders and headache disorders remaining the two most common causes. Tuberculosis including latent tuberculosis infection was the third leading cause in 1990 and became the fourth leading cause in 2017, whereas haemoglobinopathies were the fourth leading cause in 1990 and became the third leading cause in 2017. Between 1990 and 2017, the age-standardised prevalence decreased for oral disorders by 5·5% (95% UI 4·9 to 6·0) but increased for headache disorders by 0·3% (–0·2 to 0·9) and for haemoglobinopathies by 4·7% (4·3 to 5·1).

38Globally, in 2017, for all ages and both sexes combined, the three leading Level 3 causes in terms of incident cases were upper respiratory infections (17·1 billion, 95% UI 15·3 to 19·2), diarrhoeal diseases (6·29 billion, 5·81 to 6·82), and oral disorders (3·60 billion, 3·23 to 3·99; table 1). These case rankings remained unchanged for the top three causes between 1990 and 2017 despite a decrease in age-standardised incidence rates of upper respiratory infections of 2·6% (95% UI 2·0 to 3·1), from 232 815 new cases (95% UI 207 461 to 260 397) to 226 802 new cases (201 716 to 253 367) per 100 000, and in age-standardised incident rates of oral disorders of 0·3% (–1·1 to 0·6), from 48 423 new cases (43 233 to 53 971) to 48 276 new cases (43 109 to 53 919) per 100 000, and an increase in the number of new cases per 100 000 of diarrhoeal diseases of 11·7% (8·8 to 14·6), from 75 087 new cases (69 475 to 81 367) to 83 846 new cases (77 402 to 90 965) per 100 000.

39The global number of YLDs increased from 562 million (95% UI 421–723) to 853 million (642–1097) between 1990 and 2017, representing a 51·8% (50·2–53·5) increase and a 7·2% (6·0–8·4) increase in the all-age YLD rate, while age-standardised YLD rates decreased from 11 310 YLDs (8485–14 506) to 10 871 YLDs (8171–13 980) per 100 000, representing a 3·9% (3·1–4·6) decrease. CMNN causes accounted for 13·8% (12·5–15·1) of total YLDs in 2017, while NCDs accounted for 79·5% (77·8–81·1) and injuries for 6·7% (6·2–7·3), with a total of 118 million (86·7–154) YLDs for CMNN causes, 678 million (510–876) for NCDs, and 57·2 million (42·1–75·4) for injuries. The number of YLDs from CMNN causes increased from 1990 to 2017 by 13·6% (9·15–19·2), and the YLD rates from CMNN causes decreased by 14·8% (10·7–18·0) from 1846 YLDs (1343–2472) to 1573 YLDs (1159–2067) per 100 000 during the same period. The number of YLDs from NCD causes increased between 1990 and 2017 by 61·1% (60·0–62·4), and the YLD rate from these causes decreased by 1·2% (0·66–1·8) from 8684 YLDs (6540–11 223) to 8579 YLDs (6454–11 084) per 100 000. The number of YLDs from injuries increased between 1990 and 2017 by 52·7% (49·3–56·4), and the YLD rate from injuries decreased by 7·8% (6·27–9·28) from 779 (577–1023) YLDs to 719 YLDs (529–948) per 100 000. In 2017, the YLD rate for all causes ranged from 9120 YLDs (6877–11 622) per 100 000 in Columbia to 14 824 YLDs (11 080–19 203) per 100 000 in Yemen.

40Globally, in 1990, for all ages and both sexes, the leading Level 3 causes of YLDs were low back pain (42·5 million YLDs, 95% UI 30·2 to 57·2), headache disorders (35·1 million, 22·8 to 49·7), and dietary iron deficiency (31·7 million, 21·6 to 45·5). Between 1990 and 2007, the number of all-age YLDs attributed to low back pain increased by 30·0% (27·9 to 31·9) and those attributed to headache disorders increased by 34·0% (33·0 to 35·1), while the number of all-age YLDs for dietary iron deficiency decreased by 0·2% (–2·8 to 2·2). Between 1990 and 2007, the number of all-age YLDs attributed to depressive disorders increased by 33·4% (31·0 to 35·8), becoming the third leading cause of all-age YLDs in 2007, and shifting dietary iron deficiency to fourth; the rankings for low back pain and headache disorders did not change from 1990 to 2007. From 2007 to 2017, we observed further increases in the number of all-age YLDs attributable to the leading three causes: low back pain (17·5%, 95% UI 16·2–19·0), headache disorders (15·4%, 14·6–16·2), and depressive disorders (14·3%, 13·1–15·6).

41Figure 2 illustrates the leading Level 3 causes of YLD rates by GBD country and select subnational locations in 2017 for both sexes combined. The geographical variation in the leading Level 3 causes of YLD rates across countries is shown: low back pain was the leading cause in 126 of the 195 countries and territories whereas diabetes was the leading cause of YLD rates in Mexico, Equatorial Guinea, Congo (Brazzaville), Myanmar, Mauritius, and Gabon, as well as parts of the Caribbean and most of Oceania. Dietary iron deficiency was the leading cause of YLD rates in Yemen, India, Antigua and Barbuda, and in parts of western sub-Saharan Africa. Conflict and terrorism was the leading cause of YLDs in Afghanistan, Eritrea, Rwanda, and Burundi.

42The most common Level 3 causes of YLD rates by country and subnational locations by sex are shown in appendix 2. In both females and males, low back pain was the leading Level 3 cause by country. In females, the second most commonly leading Level 3 cause was headache disorders, which was the leading cause in 57 countries, followed by diabetes, which was the leading cause in 20 countries. In males, the second most commonly leading Level 3 cause was diabetes, highest in 39 countries, followed by drug use disorders and conflict and terror, each leading in five countries.

43Figure 3 shows rankings and trends of the top causes of YLDs by SDI quintile and by sex. SDI-based patterns varied by sex: the top cause of YLD rates between 1990 and 2017 in the lowest SDI quintile was dietary iron deficiency for females and low back pain for males, whereas in the highest SDI quintile it was low back pain for both females and males. In the lowest SDI quintile, for both sexes combined, the top three causes of YLD rates were low back pain, dietary iron deficiency, and headache disorders, while in the highest SDI quintile the top three causes were low back pain, headache disorders, and depressive disorders (appendix 2). The supplementary results show these results for the leading six impairments estimated in GBD: anaemia, heart failure, vision loss, hearing loss, epilepsy, and infertility (appendix 2). The immense burden of anaemia is shown, particularly for females in low-SDI regions. The differences in the burden of impairments by sex are also provided (appendix 2).

44Figure 4 shows the all-age and age-standardised YLD rates by SDI and GBD region between 1990 and 2017 for males and females combined for all causes and then separately for Level 1 causes. In general, many regions experienced decreases in all-cause age-standardised YLD rates as SDI increased (figure 4). However, there were important exceptions to this finding. First, some regions did not follow this trend consistently. Southern sub-Saharan Africa showed an increase in YLD rates for more than a decade before beginning a more precipitous decline coinciding with the apex of the HIV epidemic, whereas regions with a higher baseline SDI have generally experienced minimal changes or increases in age-standardised YLD rates over the past decade despite advances in SDI. We found that trends also varied over time depending on cause. As SDI increased, age-standardised and all-age YLDs improved for CMNN causes in most regions, but this relation was less reliable for NCDs. The NCD pattern also differed markedly between all-age and age-standardised rates for NCDs, with all-age rates increasing as SDI improved with relatively little change over time observed in age-standardised rates. For injuries, some regions initially experienced a declining burden as SDI increased, followed by an increasing burden as SDI continued to increase over time, as seen in Central Europe, for example.

45Table 2 shows global age-standardised prevalence for females and males for all Level 2 GBD causes as well as the relative difference between the sexes in 1990 and 2017. In both 1990 and 2017, the cause with the greatest relative difference between sexes was substance use disorders, which had a relative difference of 1·09 (95% UI 1·03–1·15) in 1990, which increased to 1·16 (1·09–1·23) in 2017. In 1990, 302 million (226–388) YLDs were in females and 260 million (195–336) were in males. The all-age YLD rate for females was 11 273·6 (8455·6–14 492·5) per 100 000 and the all-age YLD for males was 9571·4 (7160·9–12 356·1) per 100 000 in 1990. Among the global sum of YLDs, 53·6% (53·3–53·9) or 457 million (344–587) YLDs were in females and 46·4% (46·1–46·7) or 396 million (297–510) YLDs were in males. In the subset of causes that occur in both males and females, there were 761 million (565–987) total YLDs, with 53·0% (52·8–53·2) or 444 million (334–571) YLDs in females and 47·0% (46·8–47·2) or 395 million (296–510) YLDs in males. Between 1990 and 2017, all-age YLD rates increased for females and males, by 6·5% (5·4–7·7) to 12 007·7 (9036·7–15352·0) YLDs per 100 000 females and by 7·9% (6·6–9·2) to 10 328·1 (7744·1–13 306·9) YLDs per 100 000 males.

46Figure 5A illustrates the differences in global prevalence for females and males for 2017 by age group for the 22 Level 2 GBD causes of non-fatal health loss, calculated as the female-specific estimate subtracted from the male-specific estimate such that causes on the left side of the chart are more prevalent in females while causes on the right side of the chart are more prevalent in males in a given age group. Females experience overall higher prevalence in every age group and the highest age-specific global prevalence differences for female-predominant causes occur between the ages of 20 years and 49 years. From birth to 4 years of age, females have higher prevalence of other NCDs, which continues through all age groups, and of neoplasms than males, whereas neurological disorders are female predominant in all ages, becoming notably evident from age 5–9 years and continuing throughout all subsequent age groups. Similarly, HIV/AIDS and sexually transmitted infections are female predominant in all ages, a trend that becomes evident from age 10–14 years. Although nutritional deficiencies are more prevalent in males in the earliest age groups up to age 9 years, they are more prevalent in females in subsequent age groups up to age 74 years, at which point they once again become more prevalent in males up to the oldest age group. Females also have higher prevalence of self-harm and interpersonal violence starting from age 1–4 years and continuing until age 84 years, after which they become more prevalent in males. Digestive diseases are more prevalent in males in all age groups, whereas unintentional injuries emerge as a male-predominant cause from 5 years of age and remain as such until age 89 years, after which they become female predominant. Transport injuries are male predominant in all age groups starting from age 1–4 years. Respiratory infections and tuberculosis are male predominant in all ages. Cardiovascular diseases are male predominant starting from age 50–54 years, whereas chronic respiratory diseases emerge as a male-predominant cause from age 65–69 years.

47Figure 5B shows the sex differences in the global burden of YLDs for 2017 by age group for the 22 Level 2 GBD causes of non-fatal health loss. Before age 10–14 years, males have greater YLD rates than females, driven largely by higher rates of nutritional deficiencies. From age 10–14 years and older, females have greater overall YLD rates in every age group. Under the age of 1 year, females experience higher YLD rates due to neoplasms and maternal and neonatal disorders, whereas male infants experience higher YLD rates due to nutritional deficiencies and other NCDs. Starting at age 10 years, females experience a higher YLD rate due to other NCDs (until age 50–54 years), musculoskeletal disorders, mental disorders, neurological disorders, and chronic respiratory diseases. Males experience higher YLD rates due to unintentional injuries, transport injuries, and substance use disorders for most of life, although females older than 80 years of age experience a higher rate of unintentional injuries than do males. For self-harm and interpersonal violence, females experience higher rates than males until age 30–34 years, at which point males experience higher YLD rates. Males also experience higher YLD rates from neoplasms, cardiovascular diseases, and other NCDs starting in older age groups (≥60 years), and from digestive diseases from age 25–29 years.

48Figure 6 shows the extent to which males and females in region–cause combinations have diverged in terms of achieving equal change over time between 1990 and 2017. Each region–cause combination shows which sex has performed better over time either by decreasing more or by increasing less in terms of age-standardised YLD rates. Among the 462 region–cause combinations (excluding any “Global” or “All causes” combinations), females had more favourable outcomes over time for 260 combinations (56·3%), and males had more favourable outcomes for 202 combinations (43·7%). The Z score, that is, the number of SDs from equality, was 0·433 for the mean deviation of female-favourable causes and 0·313 for the mean deviation of male-favourable causes.

49Figure 7 shows the leading 20 causes of prevalence and YLDs for females and males separately for 1990, 2007, and 2017, with percentage change. In 1990, the most common causes for females were oral disorders, headache disorders, and haemoglobinopathies and haemolytic anaemias, whereas for males, the most common causes were oral disorders, headache disorders, and tuberculosis including latent tuberculosis. The number of prevalent cases in the top three leading causes for females increased by 23·1% (95% UI 21·8 to 24·5) for oral disorders, 31·5% (30·2 to 32·8) for headache disorders, and 29·9% (29·6 to 30·1) for haemoglobinopathies and haemolytic anaemias between 1990 and 2007. Between 2007 and 2017, the number of cases increased further by 13·5% (12·9 to 14·1), 14·5% (13·8 to 15·3), and 13·4% (13·2 to 13·6), respectively (figure 7A). Between 1990 and 2007, for males, the number of prevalent cases of oral disorders increased by 21·6% (20·1 to 23·1), the number of headache disorders increased by 31·3% (29·9 to 32·8), and the number of prevalent cases of tuberculosis including latent tuberculosis infection increased by 26·2% (21·7 to 30·3). Between 2007 and 2017, the number of prevalent cases for these three causes increased by 12·5% (11·9 to 13·1), 14·3% (13·6 to 15·2), and 1·1% (–0·67 to 3·01), respectively (figure 7B). The age-standardised prevalence of the top cause for females, oral disorders, decreased by 3·8% (3·3 to 4·3) between 1990 and 2007, and decreased by 1·3% (1·0 to 1·7) between 2007 and 2017. For the second and third top causes for females, the age-standardised prevalence decreased by 0·4% (–1·0 to 0·3) for headache disorders and increased by 4·2% (4·0 to 4·4) for haemoglobinopathies and haemolytic anaemias between 1990 and 2007 and increased by 0·3% (0·1 to 0·6) and 0·8% (0·6 to 1·0), respectively, between 2007 and 2017. In 2017, oral disorders, headache disorders, and haemoglobinopathies and haemolytic anaemias remained the top three leading Level 3 causes of global age-standardised prevalence for females. Between 1990 and 2007, the age-standardised prevalence for the top three leading causes for males decreased by 4·3% (3·8 to 4·9) for oral disorders, increased by less than 0·1% (–0·85 to 0·85) for headache disorders, and decreased by 3·1% (0·8 to 5·4) for tuberculosis including latent tuberculosis infection. Between 2007 and 2017, the age-standardised prevalence for males for oral disorders and tuberculosis further decreased by 1·6% (1·2 to 2·0) and 11·5% (10·1 to 12·8), respectively, whereas the age-standardised prevalence for headache disorders increased by 0·7% (0·4 to 0·9). In 2017, oral disorders, headache disorders, and tuberculosis including latent tuberculosis remained the three Level 3 causes with greatest global age-standardised prevalence for males.

50In terms of numbers of YLDs, in 1990, the leading causes for both females and males were low back pain, headache disorders, and dietary iron deficiency (figure 8). For both males and females, the top two leading causes of global YLDs remained consistent between both time periods, during which the total number of YLDs for each of these causes increased. The third top cause in 2017 was depressive disorders for females and diabetes for males. From 1990 to 2007, for females, the number of YLDs attributable to low back pain increased by 29·8% (95% UI 27·7–31·8) and headache disorders increased by 34·0% (32·8–35·3). From 1990 to 2007, for females, the number of YLDs attributable to depressive disorders increased by 32·2% (29·8–34·4), causing depressive disorders to become the third leading cause in 2007. The number of YLDs for the three leading causes for females continued to increase from 2007 to 2017, by 17·3% (15·8–18·8) for low back pain, 15·3% (14·4–16·2) for headache disorders, and 14·1% (12·8–15·5) for depressive disorders. From 1990 to 2007, for males, the number of YLDs attributable to the top two leading causes increased by 30·2% (28·1–32·2) for low back pain and 34·1% (32·7–35·4) for headache disorders. From 1990 to 2007, for males, the number of YLDs attributable to diabetes increased by 79·0% (75·8–82·2), causing diabetes to become the third leading cause. The number of YLDs for the three leading causes continued to climb from 2007 to 2017, increasing by 17·8% (16·5–19·3) for low back pain, 15·5% (14·5–16·5) for headache disorders, and 30·1% (24·8–36·1) for diabetes.

51Between 1990 and 2007, age-standardised YLD rates for females decreased for the three leading causes: low back pain (by 7·6%, 95% UI 6·2 to 8·9), headache disorders (by 0·1%, −0·80 to 0·69), and dietary iron deficiency (by 18·8%, 17·0 to 20·7). Between 1990 and 2007, the YLD rate for depressive disorders decreased by 3·0% (1·6 to 4·5) and became the third leading cause of YLDs in females in 2007 due to the simultaneous large decrease in dietary iron deficiency (figure 8A). Between 1990 and 2017, YLD rates for females decreased for low back pain (by 10·1%, 95% UI 8·8 to 11·3), depressive disorders (by 6·0%, 5·0 to 7·1), and dietary iron deficiency (by 30·7%, 27·4 to 33·7), while headache disorders increased by 0·6% (–0·19 to 1·42). Between 1990 and 2007, the YLD rates for males decreased for low back pain (6·8%, 5·6 to 7·8) and dietary iron deficiency (15·1%, 10·9 to 18·9), and increased for headache disorders (1·1%, 0·15 to 1·9). Between 1990 and 2007, the YLD rate for diabetes increased by 21·9% (20·0 to 24·0) and became the third leading cause of YLDs in males in 2007. From 2007 to 2017, the YLD rate for headache disorders increased by 1·5% (0·8 to 2·3) and that for diabetes increased by 4·2% (0·0 to 8·7), whereas the YLD rate for low back pain decreased by 1·3% (0·7 to 1·9). Between 1990 and 2017, the YLD rate for low back pain decreased by 8·0% (6·9 to 9·1) and that for dietary iron deficiency decreased by 26·7% (20·6 to 32·5), whereas the rates for headache disorders and diabetes increased by 2·6% (1·6–3·6) and 27·0% (22·2–32·4), respectively.

52While age-standardised all-cause global YLD rates decreased by less than 4% over the nearly three-decade period from 1990 to 2017, the number of total YLDs has increased by more than 50% during this time. This pattern is concerning given the lack of substantial improvement in age-standardised rates over time as well as the increased magnitude of total health loss. These patterns are probably related to population growth and ageing as well as increasing numbers of YLDs from conditions such as type 2 diabetes and opioid use disorders, which were less common in 1990. YLD increases, even when age-standardised rates are slightly improving, might pose a burden to economies and health-care systems that have not expanded proportionally to population growth or in populations where economic improvements have not been equitably distributed.

53Globally, over the 28-year period studied, three chronic NCDs (low back pain, headache disorders, and depressive disorders) have prevailed as three of the top four leading causes of YLDs, collectively causing 162 million (95% UI 118–216) YLDs in 2017 and representing nearly one in five YLDs globally. This substantial portion of global YLDs might be amenable to treatment and care access, because headache disorders and depressive disorders can be treated with low-cost therapeutics. The persistence of depressive disorders and low back pain is also concerning given the former's relation with self-harm and the latter's relation with potential loss of functional status in the work force.

54For the first time in the GBD study, we estimated the burdens of type 1 and type 2 diabetes separately. Our estimates illustrate the emergence of diabetes as a leading cause of disability globally, ranking as the fourth leading cause of age-standardised YLDs in 2017. This increased burden was observed across all levels of development. Age-standardised YLDs increased for both females and males across all SDI quintiles between 1990 and 2017 (figure 3). Diabetes poses complicated care challenges even in areas with reliable access to medical services. Preventive measures such as prediabetes screening, lifestyle modification, and treatment with low-cost medications such as metformin could avert worsening incidence rates, but as prevalence increases, health-care systems will also need to provide access to services such as wound care, surgical resources, medications including insulin, care for diabetic retinopathy, and increased focus on heightened cardiovascular disease risk.

55This is also the first year of GBD estimation for non-alcoholic fatty liver disease (NAFLD) and cirrhosis and liver cancer caused by non-alcoholic steatohepatitis (NASH). For cirrhosis and liver cancer caused by NASH, we estimated 892 million (95% UI 859–928) global cases for cirrhosis due to NASH and 97 400 cases (86 800–108 000) for liver cancer due to NASH, and an age-standardised prevalence of 11 061 cases (10 651–11 493) per 100 000 for cirrhosis due to NASH and 1·20 cases (1·07–1·33) per 100 000 for liver cancer due to NASH in 2017 (appendix 2), identifying NASH as an increasingly important cause of liver disease. We observed that these cases were distributed across all GBD regions, although more commonly in North Africa and the Middle East and in Oceania in terms of age-standardised rates. This expansion of NAFLD and liver disease due to NASH reflects the worsening burden of metabolic conditions globally. Given the complexity of chronic liver disease and the difficulty of effectively treating obesity, efforts to prevent obesity and to develop therapeutics for NAFLD will be increasingly important.

56After initial steep reductions in HIV incidence between 1998 and 2005, the global rate of decline in new HIV infections continues, although it has slowed in recent years and global prevalence has increased slightly since 2010, a trend which is probably driven by rapid expansion of antiretroviral therapy in high-prevalence countries and extended life expectancy of people living with HIV. These longer lifespans call for increased resources for treatment and continued prevention and treatment interventions to maintain declines in incidence. For tuberculosis, it is a notable success that drug-susceptible tuberculosis has declined in terms of age-standardised incidence since 1990. The burden of malaria has also declined since the mid-2000s in terms of age-standardised incidence rates. High age-standardised incidence rates of acute viral hepatitis have persisted, however, with age-standardised incidence rates of hepatitis A and B combined exceeding malaria incidence in 2017 despite availability of vaccines. The age-standardised incidence rate of acute hepatitis did not significantly change from 1990 to 2007, from 2007 to 2017, or from 1990 to 2017, suggesting that more proactive global initiatives are important, particularly with respect to vaccine coverage and cost-effective access to hepatitis C curative treatments.,,

57Incidence rates of lower respiratory infections (LRIs) declined among children younger than 5 years by 32·4% (95% UI 27·2–37·5) since 1990 and caused 83·0 million (66·4–101) episodes in children younger than 5 years in 2017, while the age-standardised incidence rate decreased by 5·8% (1·5–10·0) between 1990 and 2017 (appendix 1). Among all ages, pneumococcal pneumonia has prevailed as the leading aetiological subtype of LRI, accounting for more than five times more YLDs than the second leading cause (influenza). Diarrhoeal episodes have persisted as the second most common incident cause of health loss after upper respiratory infection among all ages, with 6·29 billion (5·81–6·82) total episodes in 2017, with rotavirus prevailing as the leading aetiology. The age-standardised incidence of diarrhoea varied nine-fold across GBD regions, with the lowest rates in Australasia and the highest rates in Oceania (appendix 2). The burdens of pneumococcal pneumonia and rotavirus diarrhoea are notable given the availability of vaccines for these pathogens, highlighting the need for continued expansion of vaccine coverage and delivery systems.

58While the burden of musculoskeletal disorders, vision loss, and hearing loss in older ages has been a consistent theme in past GBD studies due to their substantial contribution to non-fatal health loss, the burden of medically complex, high-resource conditions has also become increasingly concerning. Cardiovascular disease, chronic respiratory conditions, and neurological disorders sum to form more than 15% of age-standardised YLDs in every GBD region except two (southern Latin America and eastern sub-Saharan Africa; appendix 2). GBD risk factor estimates show that the drivers for atherosclerotic cardiovascular diseases—high systolic blood pressure,,, high cholesterol, poor diet, high fasting plasma glucose, and low physical activity—prevail in many geographies, indicating that this burden can be expected to persist and possibly expand as populations age. While the efforts of primary prevention approaches such as statin therapy and smoking cessation are evident in the gradual declines in age-standardised prevalence in high-income countries over the past 28 years, the prevalence of cardiovascular disease in these populations still remains high, which might also reflect improved survival from acute events due to improved response systems and interventions such as percutaneous coronary intervention, coronary artery bypass grafting, and rapid stroke treatment guidelines,,,, as well as secondary prevention with statins, antihypertensives, and smoking cessation.,,,,, As developing economies increasingly experience cardiovascular disease burden, investing in health system infrastructure to ensure access to these interventions as well as to primary prevention therapeutics could help to mitigate disability and improve survival into older ages.,, Chronic respiratory conditions also inflict considerable health loss in older people, contributing to more than 5% of YLDs starting in the 50–54 year age group, globally, who were exposed to risk factors such as smoking and indoor air pollution and similarly pose challenges to health systems with ageing populations, given the importance of oxygen therapy,,,, access to medical care,,, and the risk of respiratory infections among these patient populations., Neurological conditions continue to show large numbers of prevalent cases and YLDs, with consistent increases in recent years. We anticipate that this burden will also expand as populations age, creating challenges for health-care systems given the lack of modifiable risk factors for some neurological conditions and effective treatments for conditions such as Alzheimer's disease. Given the extensive care requirements these conditions—in isolation or as comorbidities—demand from both health systems and social and familial supports, their morbidity can reflect not only diminished health for an individual and population but also lost capital in an economy and among caretakers.,,,,,

59Opioid use disorders including opioid dependence have become an era-defining epidemic in the USA with more than 4 million prevalent cases in 2017. Outside the USA, opioid dependence was estimated to cause more than 1% of age-standardised YLDs in 135 countries in 2017, whereas in 12 out of 21 GBD regions, more than 5% of YLDs in the 15–49 year age group were caused by substance use disorders overall (appendix 2). Cumulatively, these results show pervasive health loss from substance use, which might also cause lost human capital given the concentration of the burden in working ages. The burden of mental disorders is present in both sexes and across all age groups after emerging in childhood with idiopathic intellectual disability and autism spectrum disorders and continuing into older ages with depressive disorders, anxiety disorders, and schizophrenia. Mental disorders have consistently formed more than 14% of age-standardised YLDs for nearly three decades, and have greater than 10% prevalence in all 21 GBD regions. These findings substantiate a global need for increased mental health and substance abuse treatment resources, improved judiciousness of opioid prescribing patterns particularly for chronic pain, and expanded harm-reduction strategies such as opioid substitution therapy and needle programmes.

60Among the 22 causes in the Level 2 GBD cause hierarchy, 14 causes had a relative difference between males and females greater than or equal to 10% in terms of prevalent cases in 2017, while 12 causes had a relative difference greater than 10% in terms of age-standardised prevalence. Notably, females had significantly higher prevalence of HIV/AIDS and sexually transmitted infections (driven primarily by higher burden from sexually transmitted infections) and self-harm and interpersonal violence, among other causes, whereas males had significantly higher prevalence of substance use disorders and transport injuries, among other causes. Figure 6 illustrates how divergent trends were nearly evenly distributed between males and females, despite the overall higher numbers of YLDs experienced by females. As indicated by the higher Z score in female-favourable causes, the cause-regions where females did better over time tended to have greater improvements in age-standardised YLD rates than did the cause-regions where males did better over time. There were important exceptions, however: for example, with HIV/AIDS and sexually transmitted infections in central sub-Saharan Africa, the age-standardised YLD rates for females increased by 30·0% (95% UI 6·8–68·1), while rates for males decreased by 43·0% (23·2–58·5) between 1990 and 2017. This particular trend reveals a striking lapse in health equity given the extensive knowledge of treatment and prevention methods for HIV/AIDS and many sexually transmitted infections.

61These observations imply that health systems might have achieved improvements in non-fatal burden that have not been shared across sexes. The underlying reasons for these imbalances are likely to be complex. For causes that are inherently more common in males or females, improvements in one sex over the other might reflect these causes differing in susceptibility to prevention, intervention, and treatment. For example, autoimmune conditions such as rheumatoid arthritis can affect both males and females but are ordinarily more common in females., However, such a cause still leads to increased disability that might also have ramifications not captured in a health study, such as income loss due either to disability at young ages or to the cost of modern immunotherapies. Similarly, females of childbearing age experience risk of pregnancy-related and maternal conditions, risks that could be pronounced in lower-resource settings without sufficient access to modern obstetric facilities and follow-up care., Other causes have firmly established causal risk factors such as smoking and chronic obstructive pulmonary disease (COPD), where the higher COPD rates in males are probably due to historically higher rates of smoking compared with females. Additionally, some causes may be susceptible to systematic misdiagnosis due to conventional clinical heuristics based on flawed or non-existent evidence. For example, in some areas, females presenting with acute coronary syndrome are more commonly misdiagnosed, whereas other causes, such as autism spectrum disorders, might be systematically underdiagnosed in females compared with males.,, These biases in sex-specific misdiagnosis emphasise the need for continued refinement in sex-specific burden estimation.

62While YLD rates are highest in older ages, globally, YLD counts are heavily concentrated in working-age males and females (ie, from 20–54 years), a pattern which is particularly evident among causes such as mental disorders, neurological disorders, and musculoskeletal disorders, which sum to more than 45% of all YLDs in these age groups. This finding is notable for two main reasons. First, these age groups have a considerable number of years to live that would otherwise be in full health, emphasising how conditions at these ages, even if having lower disability weights, can still contribute substantially to the non-fatal burden. Second, a disabling condition that occurs during this period of life could represent lost human capital. This loss will be increasingly important to characterise with human capital emerging as an important area of focus at the World Bank.,

63In GBD 2017, anaemia is classified as an impairment, with aetiologies ranging from rare genetic haemoglobinopathies to iron deficiency. Because these aetiologies alone could cause a relatively small proportion of all anaemia cases, the magnitude of anaemia's burden would be obscured by the granularity of these aetiologies if it was not also estimated in aggregate. In 2017, the sum of YLDs across all causes of anaemia was 58·2 million (95% UI 39·5–83·0), with a global age-standardised rate of 783·9 YLDs (531·3–1117·8) per 100 000 for both sexes, with sex-specific rates of 1012·8 YLDs (689·4–1436·5) per 100 000 females, and 555·9 YLDs (370·4–801·8) per 100 000 males. Globally, in terms of YLD rates in 2017, anaemia is the leading impairment for females, males, and both sexes combined. If anaemia were treated as a cause, it would rank as the fourth leading Level 2 cause and as the first leading Level 3 cause in terms of age-standardised YLD rates globally for females in 2017. The third leading impairment for males, females, and both sexes combined was blindness and vision impairment (1·34 billion [1·29–1·39] cases in 2017), which is notable given the economic consequences of vision loss,, the relative treatability of many vision loss conditions, and the increased risk of injury that stems from visual impairment., Summed together, the impairments included in the GBD cause hierarchy comprise a total of more than 5 billion prevalent cases worldwide in 2017, making it evident that health-care systems and policy makers should evaluate non-fatal burden not only for diseases and injuries but also for impairments and their aetiologies.

64Overview

65Overall, the most important systemic change to this year's study is the estimation of population and fertility in the GBD framework. In addition, we made changes to the GBD cause hierarchy (for example adding an aggregate Level 3 headache disorders cause), which limits certain cause-specific comparisons with past GBD cycles; this is discussed in more detail in the Limitations section below. Other notable cause-specific considerations, comparisons, and limitations that are new as of GBD 2017 are as follows. More details for each cause can be found in the supplementary methods (appendix 1 section 4).

66In GBD 2017, we added the Malnutrition and Enteric Disease (MAL-ED) study to our existing use of the Global Enteric Multicenter Study (GEMS),, to measure diarrhoeal aetiologies, which affected our aetiological distributions with increased norovirus and rotavirus attributions. Our overall diarrhoea incidence estimates were not affected. The Walker and colleagues' study estimated 1·7 billion episodes of diarrhoea among children younger than 5 years in 2010, a value that is greater than the GBD 2017 estimate in the same year (1·16 billion, 95% UI 1·00–1·34). For pneumonia, a study by Rudan and colleagues estimated a global pneumonia incidence of 0·22 episodes (IQR 0·11–0·51) per child-year among children younger than 5 years in 2010, accounting for 122·6 million episodes, a value that is greater than the GBD 2017 estimate of 94·5 million (79·5–112). These two studies are part of the Maternal and Child Epidemiology Estimation (MCEE). The MCEE methodology for diarrhoea and pneumonia incidence differs from GBD in that MCEE uses only community-based studies. By contrast, the GBD uses a greater volume of data including Demographic and Health Surveys, Multiple Indicator Cluster Surveys, and clinical data records. We also used diarrhoea-specific and LRI-specific summary exposure values as covariates in our models, which include location-specific and year-specific estimates of risk factors associated with diarrhoea or LRI. We observed that these covariates had significant relationships with diarrhoea and LRI incidence rates and therefore strengthened our estimates, particularly in data-sparse areas. Further modelling details are described in the supplementary methods (appendix 1 section 4).

67For HIV/AIDS, we used new methods to adjust for representativeness bias in ANC surveillance data, which led to our models predicting a lower peak and flatter trend of the epidemic in southern sub-Saharan Africa compared with GBD 2016. We updated our inputs of antiretroviral therapy coverage distribution informed by data from the International Epidemiology Databases to Evaluate AIDS and CD4+ progression parameters for children., We also improved our sex-specific modelling strategy in Spectrum by sex-splitting incidence based on a model fit to the sex ratio of prevalence observed in countries with representative surveys. Our estimated global prevalence results are similar to UNAIDS 2017 values, with some variation at the national level. Estimates in recent years are expected to be less stable due to improvements in HIV treatment coverage and prevention. Further modelling details are described in the supplementary methods (appendix 1 section 4).

68The hepatitis estimation process has changed to improve internal consistency between mortality and non-fatal estimates for GBD 2017. First, we included case-fatality rates from clinical data in our mortality models to improve the distribution of hepatitis deaths by virus. Second, we developed a time series of hepatitis B virus vaccine coverage in infants to use as a covariate. Third, the aggregate cause-specific mortality rate of acute hepatitis, cirrhosis, and liver cancer due to hepatitis B and hepatitis C was combined with virus-specific seroprevalence data to ensure internal consistency among incidence, prevalence, remission, and excess mortality rates. Fourth, the prevalence of chronic hepatitis B and C are now captured in the cirrhosis and other chronic liver diseases cause group rather than in the prevalence estimates for acute hepatitis B and acute hepatitis C. Overall, this caused prevalent cases of hepatitis B and C to be shifted from the acute hepatitis B and C causes to the cirrhosis and other chronic liver diseases cause. Specifically, in terms of global age-standardised prevalence rates for both sexes in 2016, acute hepatitis B decreased from 6295 cases (95% UI 5349–7436) per 100 000 in GBD 2016 to 219 cases (193–248) per 100 000 in GBD 2017, while its corresponding chronic cause, cirrhosis and other chronic liver diseases due to hepatitis B, increased from 170 cases (159–184) to 5625 cases (5159–6118) per 100 000. Similarly, acute hepatitis C decreased from 2152 cases (1927–2382) per 100 000 in GBD 2016 to 8·11 cases (7·37–8·98) per 100 000, while its corresponding chronic cause, cirrhosis and other chronic liver diseases due to hepatitis C, increased from 149 cases (136–164) to 1723 cases (1517–1969) per 100 000 (appendix 2).

69Several changes were adopted for GBD 2017. Adding clinical and claims records from outside the USA to our models affected congenital birth defects, haemoglobinopathies, and maternal disorders. Updating our inpatient admission per-capita estimates to include in-facility deliveries as a newborn admission affected neonatal disorders and congenital birth defects. Adopting in-facility delivery estimates for processing clinical data for maternal disorders led to lower estimates of pregnancy complications, especially in high-utilisation settings. We refined our method for estimating age-specific preterm birth and its complications to be consistent with GBD risk factor analysis, which resulted in preterm birth prevalence being higher than in previous GBD estimates. We enforced internal consistency among our estimates such that, for example, the sum of all specific types of congenital heart disease must equal the total number of cases of congenital heart disease. These methodological changes have lowered global estimates for many of the specific causes of birth defects, enforced the internal consistency of mortality and non-fatal estimates, narrowed the unexplained geographical variation in disease incidence, and strengthened the empirical relationship with known environmental, nutritional, and behavioural determinants of these conditions. For nutritional deficiencies, we extended our analysis of protein-energy malnutrition past age 5 years, incorporated cause-specific mortality rates, and included the prevalent cases of moderate wasting without oedema. Although the net result might seem to be an increase in the number of protein-energy malnutrition cases, it instead reflects a more comprehensive assessment of acute malnutrition than was estimated previously.

70Diabetes has been estimated in previous GBD studies, but this is the first year that type 1 and type 2 diabetes were estimated and reported separately. The estimation strategy for diabetes is provided in more detail in the supplementary methods (appendix 1 section 4), but in summary, we measured overall diabetes and type 1 diabetes. We subtracted estimates of type 1 from overall diabetes to produce estimates for type 2 diabetes. One of the main limitations to measuring type 2 diabetes is that surveys of diabetes in adults do not make a distinction between cases of type 1 and type 2 diabetes.

71In GBD 2017, for the first time we estimated NASH as a cause of liver cancer and cirrhosis and NAFLD as an asymptomatic health state. Our global age-standardised prevalence estimate of NAFLD and NASH that leads to cirrhosis or liver cancer was 10 935 cases (95% UI 10 522–11 365) per 100 000 in 2017, which was lower than a study by Younossi and colleagues, which estimated a global prevalence of 24%. This difference is due to how we adjust for alcohol use in the population, which helps to distinguish these conditions from alcohol-driven liver disease. Our study shows a similar geographical pattern to the Younossi study, with higher rates in North America and the Middle East, corresponding to higher rates of obesity.

72We split haemorrhagic stroke into subarachnoid haemorrhage and intracerebral haemorrhage as subtypes of stroke and added non-rheumatic valve disease as an additional cause and aetiology of heart failure. The overarching limitation in cardiovascular estimation is that low-income and middle-income locations are less likely to have diagnostic tests such as treadmill tests, ankle-brachial index measurement, and other modern diagnostics. This also limits the identification of preclinical atherosclerotic disease for individuals who have not had clinical events but who might have ischaemic changes with provocative cardiac testing. For stroke, there were few sources of incidence data in many low-income and middle-income countries, and many clinical records used the ICD code for “acute, unspecified stroke”, which does not differentiate between ischaemic stroke, intracerebral haemorrhage, and subarachnoid haemorrhage. Future incorporation of more clinical record data might help to address these limitations by providing more diagnostic detail for cause subtypes.

73We added several data sources that had lower estimates than the International Study of Asthma and Allergies in Childhood studies, which were the main source of asthma prevalence previously. Additionally, the adjustment factor for studies reporting on physician-diagnosed asthma without an additional question on wheezing changed between GBD 2016 and GBD 2017 after new data sources were added.

74For dementia, we added covariates for whether studies had common features present in their diagnostic protocol, such as a review of clinical records or a diagnosis by a clinician to try to correct for some of the heterogeneity between studies, because only a very small fraction of studies used the same methods. We do not have reliable covariates for prevalence or incidence of dementia and other neurological disorders, although in future studies, estimates of neurological disorders will benefit from the increased use of clinical and claims records.

75For GBD 2017, we added the new cause “myelodysplastic, myeloproliferative, and other haemopoietic neoplasms”, which were previously estimated as part of the “other neoplasms” group. In addition, we added new causes for three categories of benign and in-situ neoplasms: intestinal, cervical and uterine, and other.

76We estimated burden for a combined group of autism spectrum disorders consistent with the Diagnostic and Statistical Manual of Mental Disorders 5 designations in GBD 2017 as opposed to separately estimating autism and Asperger's syndrome and other autism disorders. Modelling improvements were made for opioid use disorders by incorporating new country-level covariates, including a measure of opioid analgesic consumption by country and the prevalence of injection drug use. Enhancements in the fatal modelling of opioid use disorders also contributed to these improved estimates. Improvements were also made in terms of data additions, particularly for cocaine and amphetamine dependence, where new data showed greater subnational variation in Mexico and Brazil.

77For GBD 2017, improvements were made to the computational framework required for injuries non-fatal measurement. These updates included incorporating GBD 2017 age groups and updating the differential equation solver for converting incidence to prevalence. Poisoning was divided into the two subcauses of poisoning by carbon monoxide and poisoning by other agents. Our estimate of 172 million (95% UI 152–194) falls per year resulting in any short-term or long-term disability was higher than those of WHO, which estimated 37 million falls that required medical attention per year—a difference that might be explained by GBD's inclusion of a wide spectrum of injury severities including minor injuries that result from falls. The WHO and GBD estimates on road injuries were more similar, with WHO estimating 20–50 million people experiencing non-fatal road injuries per year and GBD 2017 estimating 54·2 million (47·4–61·6) road injuries in 2017. For fires, WHO and GBD 2017 had similar estimates when comparisons were available—for example, in the USA, WHO reported 410 000 burn injuries in 2008 and GBD 2017 reported 360 000 (313 000–380 000) fire, heat, and hot substance injuries.

78As is evidenced in figure 1 and the supplementary methods, there is considerable heterogeneity in terms of data density for each Level 1 location-cause combination in the GBD study. Data availability does not consistently correlate with burden—that is, high-burden causes and locations can have relatively sparse data for non-fatal outcomes. This will also be increasingly problematic as more countries start to experience greater burden from NCDs.

79Currently, we inform most of the severity distributions that affect YLD calculation using the Medical Expenditure Panel Survey from the USA. This poses a limitation as the severities experienced in this population probably do not reflect global severity distributions or reflect that severity distributions are likely to vary over time, location, age, and by treatment access. As such, YLDs estimated might not reflect improvements in disability over time as SDI increases. Even for the limited diseases such as COPD, epilepsy, and stroke for which data from epidemiological studies allow us to differentiate severity by age, time, and location, the underlying data sources are still extremely limited in coverage by age, time, and location. Relying on published data on severity is unlikely to provide improved estimates in the near future. We also plan to access linked data between administrative records with diagnostic information and health surveys using health status measures to quantify severity of disease and how this changes over time and by age and sex.

80Comorbidity adjustment in GBD 2017 assumes independent distributions of comorbid conditions. This is a limitation because comorbidity distributions are known to vary by cause, age, sex, location, and time. For example, diabetes and cardiovascular disease are more likely to be comorbid than asthma and gynaecological disorders but currently our comorbidity adjustment does not capture these correlations.

81Clinical data records have known selection bias for subsets of the population that have access to health care. Some GBD causes rely heavily on clinical data given the lack of other sources, with efforts made to correct for this bias by using representative studies as a reference group. In recent years, the GBD study has used USA claims records to inform hospital data adjustments and various other elements of GBD estimation. Relying on data from a privately insured cohort in a high-income country poses limitations to the generalisability of these adjustments. This year, we also used tabulated claims records from Taiwan (province of China) for the first time in the GBD study, and in future publications we plan to continue incorporating claims data from additional countries to address this limitation. Acquiring further health insurance claims data will not only inform estimates in terms of incidence and prevalence measures but will also help to develop more accurate parameters that affect GBD estimates in other modelling steps.

82With regard to age-time patterns, in DisMod-MR 2.1, we start by estimating a global model that uses all the available data and covariates with hierarchical random effects for geographies to estimate a prior distribution that is passed down the geographical cascade. Time trends in our estimates result from where the covariates are changing over time, when year is included as a covariate or when data in different time periods lead to variation in the posterior estimates. This framework, however, does not capture true cohort effects. For example, we do not directly take into consideration prevalence in 30-year-olds in 1990 when estimating prevalence in 50-year-olds in 2010. This is a limitation for some diseases with known cohort effects such as the high incidence of hepatitis C in the USA in younger adults in the 1970s. We are continuing to develop a version of DisMod that explicitly incorporates a full age-time-period model in a Bayesian framework that will help to address this issue.

83While our cause list currently specifies 354 diseases and injuries, there is great interest in incorporating a greater level of detail. Going forward, we intend to become more granular in our cause list, but every new disease requires a substantial amount of new reviews and model building. In consultation with our Scientific Council, we will need to balance our available resources with this growing list of priorities for new additions.

84Similar to the cause reporting list, the hierarchy of the cause list and the method by which conditions are classified and grouped is a complex component of the GBD framework that has undergone nearly continuous improvement and refinement in years past. Grouping GBD causes into broad cause groups is a necessary part of the study's estimation and reporting process, but it can also cause limitations and challenges in the absence of broad consensus on objective frameworks for organising cause hierarchies. Other studies that measure burden of disease exemplify how disease groupings can affect a study's implications and interpretation. A recent study cited medical errors as the third leading cause of death in the USA but grouped all heart disease and all cancer as the first and second leading causes, respectively. A different aggregation or level of detail in terms of the disease grouping in such a study could change these rankings, which become important in the realm of health policy and resource allocation as well as popular and political perception of what causes death and disability in a population. In the GBD study, we have strived to retain as much detail as is methodologically possible given available data, and additionally we have incorporated input from the GBD collaborator network in terms of refining our cause hierarchy to represent a heterogeneous mix of medical specialties, research expertise, and location-specific knowledge. In addition, our framework allows us to revise the cause hierarchy for specific specialties such as the burden of neurological disorders.

85The methods and data that inform YLD estimation in the GBD study will continue to evolve. Several specific improvements have been suggested by the GBD collaborator network and the global research community, and the following future directions are planned to be incorporated in the coming cycles.

86Because the GBD cycle continues to be published on an annual basis, it will be important to incorporate updated data sources for older estimation years but also to add sources for the most recent estimation year. This necessitates data inclusion immediately after the end of the calendar year in order to be incorporated for the following year's estimation. This is challenging because some surveys, surveillance systems, and literature studies have a lag period between data collection and publication. One solution to this challenge would be adding, updating, and incorporating clinical data records from inpatient and outpatient visits from hospitals and clinics from more locations. Clinical data provide a level of detail, demographic information, and diagnostic accuracy that enhances the GBD study's ability to produce accurate estimates in the more recent estimation years. Relying more heavily on clinical data will require improved methods for adjusting for known representativeness biases. In the future, we plan to link medical records to representative survey results, which is possible with certain claims databases such as the Medicare Current Beneficiary Survey in the USA.

87In the next iteration of the GBD study, we plan to incorporate improved age-time estimation using a new tool, DisMod-AT, that simultaneously quantifies the relationships over time, by age, and by age cohorts. DisMod-AT will address the age-time measurement limitation described above.

88We plan to continue updating and refining our cause and location hierarchies including the causes described above. This is the first year that estimates for a non-geographical subnational grouping (ie, Māori and non-Māori populations in New Zealand) were modelled, and in future GBD studies it would be compelling to similarly estimate disease burden in groups within a location not strictly determined by geographies—for example, race groups within the USA or refugee populations in host countries.

89The global burden of non-fatal health loss in 2017 represents a complex culmination of nearly three decades of evolving disease patterns and risk factor profiles, improving health care, and dynamic populations. This study provides comprehensive updates from prior GBD studies and identifies topical themes that can be used to guide future research and discussion in the arena of non-fatal health loss. Continuing a trend from throughout the history of the GBD study, we show how YLDs continue to increase as countries improve along the development spectrum and emphasise how economies should anticipate difficulties in caring for older populations suffering from complex conditions such as diabetes, respiratory and cardiovascular diseases, and neurological conditions. Additionally, we report heterogeneity in outcomes experienced across different sexes, geographies, and income levels, and identify regions and causes where sexes have had divergent health trends over time. These findings should help to focus prevention and treatment efforts on groups and areas that have experienced inequitable health outcomes. Continuing the theme of health equity, we highlight the extensive amount of health loss experienced globally for conditions that are treatable or preventable such as viral hepatitis, emphasising the profound impact that health-care systems can produce with increasing investment in and access to health-care resources. Finally, we emphasise the burden of non-fatal health loss in younger and middle ages, where it is speculated that non-fatal health loss could ultimately lead to loss in human capital.

90Correspondence to: Prof Christopher J L Murray, Institute for Health Metrics Evaluation, University of Washington, Seattle, WA 98121, USA cjlm@uw.edu

91Correspondence to: Prof Christopher J L Murray, Institute for Health Metrics Evaluation, University of Washington, Seattle, WA 98121, USA cjlm@uw.edu