Global Burden of Disease study at the World Health Organization: research methods for the most comprehensive global study of disease and underlying health policies

“World Development Report 1993: Investing in Health” is the first GBD study authored by Dr. Christopher Murray, director of the Institute for Health Metrics and Evaluation (IHME)[1]. The GBD 1990 study had a great impact on worldwide health policies and agenda settings. In particular, the study generated global interest in hidden or neglected health issues such as the burden of road injuries. The academic paper has been cited more than 11,000 times[3].

Since 1998, GBD standards regarding GBD work were made by the World Health Organization (WHO). The year 2010 saw significant progress in GBD research, with “Global Burden of Diseases, Injuries, and Risk Factors Study 2010” (GBD 2010), publishing new values for the entire time-series data from 1990 to 2010 and seven serial papers in The Lancet in December 2012.[4] In addition to researchers from Harvard University and the World Health Organization, a community of almost 5,000 experts across the world in statistics and epidemiology gathered. The GBD also began to be funded by the Bill & Melinda Gates Foundation[2].

In the GBD study 2013, IHME, the coordination center for GBD contributors' international networks, reflected the work of about 1,000 researchers in more than 100 countries. The results and related research were published in a broader range of journals[5]. The GBD 2015 reflected the work of nearly 2,000 researchers in more than 120 countries[6]. In 2016, starting from 133 countries and 3 regions, 2,518 collaborators participated[7], and it soon expanded to 3,676 collaborators from 146 countries and regions in the GBD 2017[8]. Finally, in the GBD 2019, the project increased to more than 5,000 collaborators in 152 countries and regions. This latest GBD, GBD 2019, was published in the Lancet[2, 9].

Until the GBD study 1990, 107 diseases and 483 non-fatal complications were researched through a dataset which covered 8 regions and 5 age groups[3]. Moreover, the GBD 2010 distributed a dataset of estimates for 291 diseases and injuries, 67 risk factors, 1,160 aftereffects in21 regions, 20 age groups, and 187 countries[4].

In the GBD 2013 study, 188 countries including more than 300 illnesses and injuries, 79 risk factors, and 2,300 aftereffects were covered, with many landmark studies conducted on smoking[5]. Researchers expanded the GBD to 315 diseases and injuries and 79 risk factors for 195 countries in 2015[6]. The total of locations became 774 in 2016[7], including 333 diseases and injuries, 84 risk factors, and 23 age groups[8]. The GBD 2016 study included research on alcohol, gun accidents, etc[10].

In the GBD 2017, the dataset found a significant increase in temporal coverage. Estimates of mortality and life expectancy were increased compared to 1950, and a total of 359 new causes of disease and injury were added to the list of fatal and non-fatal causes[8]. Contemporarily, GBD also added one new risk, bullying victimization, and 80 new risk-result pairs[8]. The latest GBD, GBD 2019, contained the most extensive data[2, 9]. Estimated mortality and life expectancy values have been even more increased to the most detailed level to a total of 990 locations, with new causes added to the list of fatal and non-fatal causes resulting in 369 illnesses and injuries[2, 9].

Compared to the early GBD, many methodological advances have been made in the modern GBD. The most representative of those advancements is the development of the Disease Burden Unit (DBU)[4]. WHO created the DBU which was able to generate GBD estimates in 2000, 2001, and 2002, and published estimates in the annual World Health Report[11].

The annual update to the entire series of GBD estimates began since GBD 2015[6]. These more frequent updates, containing the most useful and timely picture of population health, were provided to policymakers, donors, and other decision-makers. The 2015 update also introduced the Socio-demographic Index (SDI), a summary measure that extended the methodologies, datasets, and tools used in previous GBDs and identified where countries or other regions are in the development spectrum[6]. SDI, expressed through a scale from 0 to 1, is the composite mean of per capita income, average education level, and fertility ranking in all areas of the GBD study[6].

GBD 2017 was the first to provide independent estimates of 195 countries, territories, and populations worldwide using a standardized and replicable approach, as well as comprehensive updates on fertility rates. Through GBD 2017, reviewing many Sustainable Development Goal indicators and generating forecasts up to 2030 using prediction methods became possible[8, 12, 13]. Such development enabled researchers to assess the rate of change required to achieve Sustainable Development Goals.

Published in The Lancet in October 2020, GBD 2019 issued a comprehensive update on fertility rates[14]. The total fertility rate is a new summary measure update representing the average number of children a woman would deliver over her lifetime.

The fundamental aspect of the GBD data schema is the age, sex, country, and year of the diseases. Most important estimates can be derived from these core variables. In addition, GBD 2019 demographic data include population; the analogized rate of population change (2010 to 2019), total facility rate and live births (1950, 1980, and 2019), and net productive rate[2, 9].

More detailed data are provided on deaths. They include under-five mortality rate, change in under-five mortality rate (2010 to 2019), probability of death between ages 15 and 60 years and life expectancy at birth (by sex), healthy life expectancy, the total number of deaths, the total number of deaths among children under 5 years, and observed and expected life expectancy[14].

In addition, life habits and important causes of disease like alcohol consumption and daily smoking are included[10]. There are also the current status variable and previous status variable of alcohol and smoking, with the amount of consumption. Variables that indicate accidents and injuries (i.e., from firearms) can also be extracted from the dataset. Examples of these are explained in the final part of this article.

The GBD dataset includes a variety of diseases which are classified by level. The three diseases with the highest level (i.e., level 1) of GBD are (i) communicable, maternal, neonatal, and nutritional diseases; (ii) non-communicable diseases; and(iii) injuries. These are again classified as level 2 diseases as shown in the following table (Table 1).

Level 2 diseases are subdivided into all diseases, including 169 levels 3 and 165 level 4 diseases. These 351 causes and diseases can be matched with ICD-10 codes, and these matching tables are published by Wang et al., 2019[14].

Cause of death is obtained from vital registration, verbal autopsy, and cancer registration data. Particularly in cases of cancer data, pathology-based cancer registries with defined populations and hospital-based cancer registries exist.

The age-standardized rate is an operation to standardize mortality or the burden of death[15, 16]. The age-based distribution is a very important factor in the number of deaths or patients in a particular population. In most cases, the group with a large number of elderly people has a high mortality and prevalence of disease per population. Therefore, it is necessary to regularize these values by standardizing them as groups with the same age distribution. Using the standard population suggested by the WHO, the age-standardized rate can be computed based on adjusted age distribution differences in the population. This operation can be done by applying the observed age-specific mortality and prevalence for each population.

Mortality-to-incidence ratio is calculated by dividing the mortality rate for the year by the incidence rate. This is an indicator used to identify inequality in cancer prognosis. Mortality-to-incidence ratio is a higher-level comparative measure of a crude survival estimate than relative survival. Because of its simplicity, high-quality incidence and mortality data are available in most countries, so survival can be compared internationally[17].

Spatiotemporal Gaussian Process Regression (ST-GPR) is a time series model primarily used to estimate risk factor exposure. This smooths the value for each risk factor. Through this, interpolating non-linear trends are possible and only meaningful signals are extracted from noise.[2, 9]

The Healthcare Access and Quality (HAQ) Index measures the access and quality of medical benefits by the use of amenable mortality[18]. Deaths from avoidable causes in the presence of adequate medical care are calculated. HAQ Index approximates the national levels of personal health-care access and quality and is used as covariates in GBD.

The Cause of Death Ensemble model (CODEm)is an integrated cause of death modeling environment[19]. It enables trend estimation by exploring various possible models. In this process, a covariate selection algorithm is used, and this combination is executed through four model classes, including the spatial-temporal Gaussian process regression (ST-GPR). The algorithm proposes possible models by generating many plausible combinations of covariates. Finally, it combines with an ensemble with optimal non-sample validity.

Disability-Adjusted Life Year (DALY) is the number of years lost due to ill-health, disability, or early death. As a measure of overall disease burden, DALY can be calculated comparing the sum of the years of life lost (YLL) and the years lost due to disability (YLD)[20]. YLL is a product of the number of deaths (due to the specific condition) and standard life expectancy at age of deaths. YLD represents the loss of each full year of healthy life due to disability or ill-health. This can be calculated by the product of three terms: incident cases in the population, the weight of disability due to the specific condition, and the average year of the case until death (or remission).