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

Background: The Global Burden of Diseases, Injuries, and Risk Factors Study 2017 (GBD 2017) includes a comprehensive assessment of incidence, prevalence, and years lived with disability (YLDs) for 354 causes in 195 countries and territories from 1990 to 2017. Previous GBD studies have shown how the decline of mortality rates from 1990 to 2016 has led to an increase in life expectancy, an ageing global population, and an expansion of the non-fatal burden of disease and injury. These studies have also shown how a substantial portion of the world’s population experiences non-fatal health loss with considerable heterogeneity among different causes, locations, ages, and sexes. Ongoing objectives of the GBD study include increasing the level of estimation detail, improving analytical strategies, and increasing the amount of high-quality data. Methods: We estimated incidence and prevalence for 354 diseases and injuries and 3484 sequelae. We used an updated and extensive body of literature studies, survey data, surveillance data, inpatient admission records, outpatient visit records, and health insurance claims, and additionally used results from cause of death models to inform estimates using a total of 68 781 data sources. Newly available clinical data from India, Iran, Japan, Jordan, Nepal, China, Brazil, Norway, and Italy were incorporated, as well as updated claims data from the USA and new claims data from Taiwan (province of China) and Singapore. We used DisMod-MR 2.1, a Bayesian meta-regression tool, as the main method of estimation, ensuring consistency between rates of incidence, prevalence, remission, and cause of death for each condition. YLDs were estimated as the product of a prevalence estimate and a disability weight for health states of each mutually exclusive sequela, adjusted for comorbidity. We updated the Socio-demographic Index (SDI), a summary development indicator of income per capita, years of schooling, and total fertility rate. Additionally, we calculated differences between male and female YLDs to identify divergent trends across sexes. GBD 2017 complies with the Guidelines for Accurate and Transparent Health Estimates Reporting. Findings: Globally, for females, the causes with the greatest age-standardised prevalence were oral disorders, headache disorders, and haemoglobinopathies and haemolytic anaemias in both 1990 and 2017. For males, the causes with the greatest age-standardised prevalence were oral disorders, headache disorders, and tuberculosis including latent tuberculosis infection in both 1990 and 2017. In terms of YLDs, low back pain, headache disorders, and dietary iron deficiency were the leading Level 3 causes of YLD counts in 1990, whereas low back pain, headache disorders, and depressive disorders were the leading causes in 2017 for both sexes combined. All-cause age-standardised YLD rates decreased by 3·9% (95% uncertainty interval [UI] 3·1-4·6) from 1990 to 2017; however, the all-age YLD rate increased by 7·2% (6·0-8·4) while the total sum of global YLDs increased from 562 million (421-723) to 853 million (642-1100). The increases for males and females were similar, with increases in all-age YLD rates of 7·9% (6·6-9·2) for males and 6·5% (5·4-7·7) for females. We found significant differences between males and females in terms of age-standardised prevalence estimates for multiple causes. The causes with the greatest relative differences between sexes in 2017 included substance use disorders (3018 cases [95% UI 2782-3252] per 100 000 in males vs 1400 [1279-1524] per 100 000 in females), transport injuries (3322 [3082-3583] vs 2336 [2154-2535]), and self-harm and interpersonal violence (3265 [2943-3630] vs 5643 [5057-6302]). Interpretation: Global all-cause age-standardised YLD rates have improved only slightly over a period spanning nearly three decades. However, the magnitude of the non-fatal disease burden has expanded globally, with increasing numbers of people who have a wide spectrum of conditions. A subset of conditions has remained globally pervasive since 1990, whereas other conditions have displayed more dynamic trends, with different ages, sexes, and geographies across the globe experiencing varying burdens and trends of health loss. This study emphasises how global improvements in premature mortality for select conditions have led to older populations with complex and potentially expensive diseases, yet also highlights global achievements in certain domains of disease and injury. The 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. The 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. Analyses 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 population19 unless otherwise specified. This study complies with the Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER)20 recommendations (appendix 1). GBD 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. GBD 2017 is the first time that estimation of fertility and population has been done within the GBD framework. Previously, the GBD study used external sources21, 22 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. In 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. In 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. For 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. The 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). In 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). In 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. For 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). Custom 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). Prevalence 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). Severity 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). Disability weight estimation is described in more detail elsewhere in the literature,23 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. A 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. YLDs 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. We apply the same technique for propagating uncertainty as used elsewhere in the GBD study design.19, 24, 25 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. The Socio-demographic Index (SDI) is a summary measure that estimates a location’s position on a spectrum of development.26 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 Fertility24 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).24 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). The 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.