Question

Visit "Global: Both Sexes, All Ages, 2016, DALYs" on the Institute for Health Metrics and Evaluation GBD Compare Data Visualization Hub website. Compare the primary causes of disability-adjusted life years (DALYs) from countries in two different socio-demographic index levels or economic regions. Identify three social or political-economic differences that help explain the differences you observed. Discuss the utility of the disability-adjusted life year (DALY) measure as a composite measure of health. Why is the DALY helpful given the different categories of Communicable, Noncommunicable, and Injury when it comes to comparing mortality and morbidity?

Answer 1

Global, regional, and national disability-adjusted life-years (DALYs) for 333 diseases and injuries and healthy life expectancy (HALE) for 195 countries and territories, 1990–2016: a systematic analysis for the Global Burden of Disease Study 2016

• Background

Measurement of changes in health across locations is useful to compare and contrast changing epidemiological patterns against health system performance and identify specific needs for resource allocation in research, policy development, and programme decision making. Using the Global Burden of Diseases, Injuries, and Risk Factors Study 2016, we drew from two widely used summary measures to monitor such changes in population health: disability-adjusted life-years (DALYs) and healthy life expectancy (HALE). We used these measures to track trends and benchmark progress compared with expected trends on the basis of the Socio-demographic Index (SDI).

• Methods

We used results from the Global Burden of Diseases, Injuries, and Risk Factors Study 2016 for all-cause mortality, cause-specific mortality, and non-fatal disease burden to derive HALE and DALYs by sex for 195 countries and territories from 1990 to 2016. We calculated DALYs by summing years of life lost and years of life lived with disability for each location, age group, sex, and year. We estimated HALE using age-specific death rates and years of life lived with disability per capita. We explored how DALYs and HALE differed from expected trends when compared with the SDI: the geometric mean of income per person, educational attainment in the population older than age 15 years, and total fertility rate.

• Findings

The highest globally observed HALE at birth for both women and men was in Singapore, at 75·2 years (95% uncertainty interval 71·9–78·6) for females and 72·0 years (68·8–75·1) for males. The lowest for females was in the Central African Republic (45·6 years [42·0–49·5]) and for males was in Lesotho (41·5 years [39·0–44·0]). From 1990 to 2016, global HALE increased by an average of 6·24 years (5·97–6·48) for both sexes combined. Global HALE increased by 6·04 years (5·74–6·27) for males and 6·49 years (6·08–6·77) for females, whereas HALE at age 65 years increased by 1·78 years (1·61–1·93) for males and 1·96 years (1·69–2·13) for females. Total global DALYs remained largely unchanged from 1990 to 2016 (–2·3% [–5·9 to 0·9]), with decreases in communicable, maternal, neonatal, and nutritional (CMNN) disease DALYs offset by increased DALYs due to non-communicable diseases (NCDs). The exemplars, calculated as the five lowest ratios of observed to expected age-standardised DALY rates in 2016, were Nicaragua, Costa Rica, the Maldives, Peru, and Israel. The leading three causes of DALYs globally were ischaemic heart disease, cerebrovascular disease, and lower respiratory infections, comprising 16·1% of all DALYs. Total DALYs and age-standardised DALY rates due to most CMNN causes decreased from 1990 to 2016. Conversely, the total DALY burden rose for most NCDs; however, age-standardised DALY rates due to NCDs declined globally.

• Interpretation

At a global level, DALYs and HALE continue to show improvements. At the same time, we observe that many populations are facing growing functional health loss. Rising SDI was associated with increases in cumulative years of life lived with disability and decreases in CMNN DALYs offset by increased NCD DALYs. Relative compression of morbidity highlights the importance of continued health interventions, which has changed in most locations in pace with the gross domestic product per person, education, and family planning. The analysis of DALYs and HALE and their relationship to SDI represents a robust framework with which to benchmark location-specific health performance. Country-specific drivers of disease burden, particularly for causes with higher-than-expected DALYs, should inform health policies, health system improvement initiatives, targeted prevention efforts, and development assistance for health, including financial and research investments for all countries, regardless of their level of sociodemographic development. The presence of countries that substantially outperform others suggests the need for increased scrutiny for proven examples of best practices, which can help to extend gains, whereas the presence of underperforming countries suggests the need for devotion of extra attention to health systems that need more robust support.

• Implications of all the available evidence

The epidemiological transition continues apace globally, with a shift from DALYs attributable to communicable, maternal, neonatal, and nutritional diseases to those attributable to non-communicable diseases. This progression is concomitant with improvements in SDI and thus improvements in education, fertility rates, and economic status. A more detailed analysis than in this study of the epidemiological changes that have occurred in countries that have consistently exceeded expectations could provide improved insights into good practice in public health policy, which might be emulated elsewhere. A similarly detailed appraisal of countries that are lagging in DALYs and HALE relative to expectations on the level of SDI alone will help identify countries in most need of domestic and international attention across the development continuum.

As the second in a series of now annual updates, the Global Burden of Diseases, Injuries, and Risk Factors Study 2016 (GBD 2016) is the most comprehensive and current source of summary health metrics. The Global Burden of Disease (GBD) is based on development of the largest available database of health outcomes, risk factor exposure, intervention coverage, and sociodemographic factors related to health. We applied analytical techniques to reduce data biases and support comparability, propagated the uncertainty in these estimates, and provided insights at the highest temporal and spatial resolution afforded by the data.

The purpose of this study is to present the results of GBD 2016 for DALYs and HALE, building on updated estimates of mortality, causes of death, and non-fatal health loss7, 10 to identify nations with notable variation in health performance from that expected on the basis of SDI. Approaches to the analysis have been previously described.GBD 2016 improvements include addition of newly available retrospective data, refined analytical methods (such as improvement to mortality to incidence ratios [MIRs] for cancers to better reflect lower survival in low-income and middle-income countries based on SDI), new subnational estimation for England and Indonesia, disaggregation of certain cause groupings to capture greater detail, and expansion of older age groups to enhance relevance for a wider range of health policy decisions.

• Estimation of mortality and non-fatal health loss

To estimate all-cause and cause-specific mortality, the GBD study first systematically addressed known data challenges—such as variation in coding of causes or age group reporting, misclassification of deaths from HIV/AIDS, or methods for incorporation of population-based cancer registry data—using standardised methods described in detail in the GBD 2016 mortality and causes of death publications. As noted in other GBD publications, each death is attributed to a single underlying cause in accordance with the ICD. We take steps to standardise cause of death data to address the small fraction of deaths that are not assigned an age or sex, deaths assigned to broad age groups that are not 5 year age groups, and various revisions and national variants of the ICD. Additionally, we identify and redistribute deaths assigned to ICD codes that cannot be underlying causes of death, are intermediate causes of death rather than the underlying causes, or lack specificity in coding. We estimated cause-specific mortality using standardised modelling processes—most commonly, the Cause of Death Ensemble model, which uses covariate selection and out-of-sample validity analyses and generates estimates for each location-year, age, and sex. Additional detail, including model specifications and data availability for each cause-specific model, can be found in the appendix of the GBD 2016 mortality and causes of death publications. We used the all-cause mortality estimates to establish a reference life table from the lowest death rates for each age group among locations with total populations greater than 5 million.From this reference life table, we multiplied life expectancy at the age of death by cause-specific deaths to calculate cause-specific YLLs. We then used the GBD world population age standard to calculate age-standardised rates for deaths and YLLs. The GBD world population age standard and the standard life expectancies are available in the appendix of the GBD 2016 mortality publication.

Changes implemented since the Global Burden of Diseases, Injuries, and Risk Factors Study 2015 (GBD 2015) for cause-specific mortality include incorporation of substantial sources of new mortality data; important model improvements for HIV, malaria, tuberculosis, injuries, diabetes, and cancers; disaggregation of specific causes into subgroupings to provide additional detail (the following were all estimated separately for the first time: alcoholic cardiomyopathy; urogenital, musculoskeletal, and digestive congenital anomalies; Zika virus disease; Guinea worm disease; self-harm by firearm; sexual violence; myocarditis; and the following types of tuberculosis: extensively drug-resistant tuberculosis, multidrug-resistant tuberculosis without extensive drug resistance, drug-susceptible tuberculosis, extensively drug-resistant HIV/AIDS-tuberculosis, multidrug-resistant HIV/AIDS-tuberculosis without extensive drug resistance, and drug-susceptible HIV/AIDS-tuberculosis); modelling of antiretroviral therapy (ART) coverage for each location-year by CD4-positive cell count at initiation; breakdown of terminal age groups from 80 years and older to 80–84 years, 85–89 years, 90–94 years, and 95 years and older; expansion of the GBD location hierarchy; and changes in the calculation of SDI.10 The database for GBD 2016 now includes data for the 333 causes estimated for DALYs and new subnational units for Indonesia (n=34) and England (n=150). For GBD 2016, we included substantial amounts of additional data sources from new studies and our network of collaborators; details of the types of data added can be found in the GBD 2016 cause of death10 and non-fatal8 publications. Additionally, research teams did systematic reviews to incorporate literature data into fatal and non-fatal models. Further details on search strings are available in the GBD 2016 non-fatal8 and cause of death publication appendices. The Registrar General of India provided improved verbal autopsy data collected through their Sample Registration System, enabling a more detailed and thorough analysis of subnational data for India than in GBD 2015. The methods for constructing the SDI, initially developed for GBD 2015,15 were revised for GBD 2016 to account for expansion in the number of subnational estimates and the effect of a growing time period of estimation given fixed limits for index components.10 The components of SDI—total fertility rate (TFR), educational attainment in the population aged older than 15 years, and lag-distributed income (LDI)—are based on new systematic assessments of educational attainment, LDI, and fertility, and each component is scaled relative to maximum effect on health outcomes.

In most cases, we estimated non-fatal health loss using the Bayesian meta-regression tool DisMod-MR 2.1 to synthesise variable data sources to produce internally consistent estimates of incidence, prevalence, remission, and excess mortality. Cause-specific data availability and epidemiological characteristics required additional analytical techniques in some cases (details are available in the appendix of the GBD 2016 non-fatal these causes include many neglected tropical diseases (NTDs) such as dengue, as well as injuries, malaria, and HIV/AIDS.

We estimated each non-fatal sequela separately and assessed the occurrence of comorbidity in each age group, sex, location, and year separately using a microsimulation framework. We distributed disability estimated for comorbid conditions to each contributing cause during the comorbidity estimation process. Although the distribution of sequelae—and therefore the severity and cumulative disability per case of a condition—can be different by age, sex, location, and year, previous studies have found that disability weights do not substantially vary across locations, income, or levels of educational attainment.In the GBD study, disability weights were based on population surveys with respondents and held invariant between locations and over time.20 Additional details, including model specifications and data availability for each cause-specific model and development of disability weights by cause and their use in the estimation of non-fatal health loss, are available in the appendix of the GBD 2016 non-fatal publication.

For non-fatal estimation, several methodological changes were made for GBD 2016. New data for the main causes of YLDs were identified through our collaboration with the Indian Council of Medical Research and the Public Health Foundation of India. For particular risk factors and diseases, the volume of available data increased substantially, such as child growth failure (stunting, wasting, or underweight), anaemia, congenital anomalies, schistosomiasis, intestinal helminths, and lymphatic filariasis. We have improved our analysis of total admissions per person by country, year, age, and sex, which facilitated incorporation of additional hospital data sources that were previously excluded because of incomplete knowledge of catchment population size. We extended our analyses of linked USA medical claims data to impute age-specific and sex-specific ratios for multiple admissions per illness episode, ICD code appearance in the non-primary position, and inpatient versus outpatient use.8 We applied each of the three ratios sequentially to non-linked hospital inpatient data from elsewhere that only had a single ICD code per visit to adjust prevalence and incidence data. We have incorporated more predictive covariates into our non-fatal disease models to better predict variation in disease levels rather than measurement error as the source of variation, and we improved our analysis of the MIRs for cancers, resulting in considerably higher ratios in lower SDI quintiles and thus substantially lower YLD estimates for cancer.

We calculated DALYs as the sum of YLLs and YLDs for each cause, location, age group, sex, and year. The same estimates of YLDs per person for each location, age, sex, and year from 1990 to 2016 are used to establish HALE by age group within abridged multiple-decrement life tables with use of methods developed by Sullivan.

For all results, we report 95% uncertainty intervals (UIs) derived from 1000 draws from the posterior distribution of each step in the estimation process. Unlike confidence intervals, UIs capture uncertainty from multiple modelling steps, as well as from sources such as model estimation and model specification, rather than from sampling error alone. Uncertainty associated with estimation of mortality and YLLs reflects sample sizes of data sources, adjustment and standardisation methods applied to data, parameter uncertainty in model estimation, and uncertainty within all-cause and cause-specific mortality models. For estimation of prevalence, incidence, and YLDs, UIs incorporated variability from sample sizes within data sources, adjustments to data to account for non-reference definitions, parameter uncertainty in model estimation, and uncertainty associated with establishment of disability weights. Because direct information about the correlation between uncertainty in YLLs and YLDs was scarce, we assumed that uncertainty in age-specific YLDs was independent of age-specific YLLs or death rates.