Global, Regional, and National Burden of Cardiovascular Diseases for 10 Causes, 1990 to 2015

Background The burden of cardiovascular diseases (CVDs) remains unclear in many regions of the world. Objectives The GBD (Global Burden of Disease) 2015 study integrated data on disease incidence, prevalence, and mortality to produce consistent, up-to-date estimates for cardiovascular burden. Methods CVD mortality was estimated from vital registration and verbal autopsy data. CVD prevalence was estimated using modeling software and data from health surveys, prospective cohorts, health system administrative data, and registries. Years lived with disability (YLD) were estimated by multiplying prevalence by disability weights. Years of life lost (YLL) were estimated by multiplying age-specific CVD deaths by a reference life expectancy. A sociodemographic index (SDI) was created for each location based on income per capita, educational attainment, and fertility. Results In 2015, there were an estimated 422.7 million cases of CVD (95% uncertainty interval: 415.53 to 427.87 million cases) and 17.92 million CVD deaths (95% uncertainty interval: 17.59 to 18.28 million CVD deaths). Declines in the age-standardized CVD death rate occurred between 1990 and 2015 in all high-income and some middle-income countries. Ischemic heart disease was the leading cause of CVD health lost globally, as well as in each world region, followed by stroke. As SDI increased beyond 0.25, the highest CVD mortality shifted from women to men. CVD mortality decreased sharply for both sexes in countries with an SDI >0.75. Conclusions CVDs remain a major cause of health loss for all regions of the world. Sociodemographic change over the past 25 years has been associated with dramatic declines in CVD in regions with very high SDI, but only a gradual decrease or no change in most regions. Future updates of the GBD study can be used to guide policymakers who are focused on reducing the overall burden of noncommunicable disease and achieving specific global health targets for CVD.

C ardiovascular diseases (CVDs) are a leading cause of death in the world and a major barrier to sustainable human development (1).
In 2011, the United Nations formally recognized noncommunicable diseases, including CVDs, as a major concern for global health and set out an ambitious plan to dramatically reduce the effect of these diseases in all regions (2). An increased awareness of these global noncommunicable disease goals has expanded attempts to track and benchmark national efforts at reducing CVD and other noncommunicable diseases (3,4).
The third Sustainable Development Goal recognized the importance of CVD by targeting a one-third reduction in premature mortality due to noncommunicable diseases (5). Countries that take the SDG goals seriously will have to contend with a wide range of barriers limiting their ability to improve health care and reduce CVD risks. In many regions of the world, the relative position of CVD as a health problem remains unclear or is based on limited data.
Many low-and middle-income countries have implemented health examination surveys that have improved measurement of CVD and its associated risk factors (6).
Systematic evaluation of data collected in death certificates, verbal autopsy, health surveys, prospective cohort studies, health system administrative data, and disease registries is needed to appropriately guide efforts to reduce the health burden of CVD. The GBD (Global Burden of Disease) study is an effort to continuously improve our understanding of the burden of diseases by integrating the available data on disease incidence, prevalence, and mortality to produce consistent, transparent, and up-to-date global, regional, and national estimates (7).
The global number of CVD deaths and regional patterns of total CVD mortality were previously reported from the GBD 2013 study (8). The  vital registration. Models of disease incidence and prevalence now uniformly include estimates of excess mortality and, for stroke, cause-specific mortality, so that they are better informed by the available mortality data. For each incidence or prevalence data point, we matched the age-sex-location-year-causespecific mortality rate to produce a ratio conceptually equivalent to an excess mortality rate. Because of implausibly rapid increases in deaths reported due to atrial fibrillation, we have developed a unified model of atrial fibrillation that makes use of prevalence, case fatality, and mortality data to estimate both the nonfatal and fatal burden due to this condition. DEFINING DISEASE CATEGORIES. CVD was estimated overall and separately for the 10 most common global causes of CVD-related death. These causes were ischemic heart disease (IHD), ischemic stroke, hemorrhagic and other stroke, atrial fibrillation, peripheral arterial disease (PAD), aortic aneurysm, cardiomyopathy and myocarditis, hypertensive heart disease, endocarditis, rheumatic heart disease (RHD), and a category for other CVD conditions. The GBD cause list is a hierarchical, mutually-exclusive, and collectively exhaustive list of causes of death. The 3 level 1 GBD causes consist of communicable, maternal, neonatal, and nutritional disorders; noncommunicable diseases; and injuries. Level 2 causes consist of 21 cause groups, such as neoplasms and CVD. Levels 3 and 4 consist of disaggregated subcauses (Online Methods Appendix Table 1).
Cause of death was defined by international standards governing the reporting of death certificates, in which a single underlying cause is assigned by a physician. For example, IHD was defined as an underlying cause of death across International Classification of Diseases (ICD) revisions (most recently ICD-10 I20 to I25, ICD-9 410 to 414) (12). The leading causes included as "other cardiovascular and circulatory diseases" were nonrheumatic valvular disorders and pulmonary embolism. A proportion of deaths that were assigned on death certificates to nonfatal, undefined, or intermediate causes (e.g., cardiac arrest, heart failure, or hypertension) were redistributed using statistical regression methods or fixed proportions (9). Redistribution of deaths coded to heart failure was accomplished using a regression model that accounted for the variable use of these codes by age, sex, and location. This approach improves upon methods that either exclude deaths coded to an intermediate cause or reassign them using a fixed proportion that ignores variation by age, sex, or location. Deaths due to unspecified types of stroke (ICD-10 I64) were distributed using the ratio of ischemic to hemorrhagic stroke deaths in a country's region or, for South Asia, the global ratio, stratified by age. A Bayesian noise reduction algorithm was applied to death data to improve estimation of the underlying mortality rate (see the Online Appendix for details). This noise reduction algorithm was adopted to improve upon prior methods in which 0 counts were excluded, an approach that leads to an upward bias in estimates. Verbal autopsy, a method in which a standardized interview collects information from household members on symptoms preceding death, was included as a data input only for total CVD, ischemic heart disease, and stroke deaths, and was excluded for other CVD causes of death.
Disease prevalence was estimated at a more granular level of specific disease sequelae, using input data from systematic reviews of the published scientific which accounts for out-of-hospital sudden cardiac death (13,14). Adjustments were made for the advent of troponin-testing technology for diagnosis of acute coronary syndromes during the years covered by the study using meta-analysis of its increased sensitivity (compared with prior markers) to adjust pre-2000 incidence rates upward by 56%. Stable angina was defined according to the Rose Angina Questionnaire, which was adjusted to account for the observed differences in survey and administrative data found in the United States. Cerebrovascular disease relied on a case definition developed by the World Health Organization and was estimated separately for 2 subcategories: 1) ischemic stroke; and 2) hemorrhagic or other nonischemic stroke (15). Stroke data was adjusted to match our case definition of subtypespecific first-ever incident events, and was used to separately estimate acute and chronic stroke. PAD was defined by an ankle brachial index (ABI) <0.9, and symptomatic PAD was defined as self-report of claudicatory symptoms among those with ABI <0.9 (16).
Atrial fibrillation was defined by electrocardiogram and included atrial flutter. The prevalence of symptomatic heart failure was estimated using both health system administrative and population-based registry data, and was then attributed to specific underlying heart failure etiologies (some of which were not CVD).
Hypertensive heart disease was defined as symptomatic heart failure due to the direct and long-term effects of hypertension, with its nonfatal burden derived from the model of heart failure. Cardiomyopathy was defined as symptomatic heart failure due to primary myocardial disease or toxic exposures, such as alcohol, with its nonfatal burden derived from the model of heart failure (17). Acute myocarditis was estimated as an acute and time-limited condition due to myocardial inflammation using health system administrative data. Endocarditis and RHD were defined by their clinical diagnosis. Estimates of RHD include cases identified by clinical history and physical examination, including auscultation or standard echocardiographic criteria for definite disease.  Table 1 summarizes data sources used to estimate CVD burden. Table 1 also shows the data representativeness index for nonfatal estimates, which is the proportion of age-sex-location strata with available data for nonfatal modeling shown by cause and over time.
Online Methods Appendix Tables 2 and 3 are tables of all data sources. Data sources for models are also available online from the Global Health Data Exchange (18). National income, metabolic and nutritional risk factors, and other country-level covariates were estimated from surveys and published systematic reviews. Analysis of mortality used Cause of Death Ensemble modeling (CODEm), an approach that incorporates country-level covariates, including agesex-country-year-specific estimates of CVD risk factors, national income, and other causal factors (Online Appendix). CODEm borrows strength across space, time, and age groups using a variety of geospatial model types, and weighs the results using tests of out-of-sample predictive validity. Analysis of disease prevalence used epidemiological state-transition-based disease modeling software, DisMod-MR, which accounts for study-level differences in measurement method (9). Disease-specific incidence, prevalence, case fatality, and mortality rates were   Least squares regression of death rates on SDI was used with a smoothing spline and dummy variables for outlier regions that skewed fit to capture the average relationship for each age-sex-cause group.

RESULTS
All results of the GBD 2015 study, including prevalence, mortality, YLL, YLD, and DALYs, for all country-years are available for download from the    The relationship between SDI and the agestandardized CVD death rate at the global level is shown in Figure 3. As SDI increases beyond 0.25, the highest CVD mortality rates shift from women to men. Relationship between age-standardized mortality rate for CVD and SDI over time. Each colored line represents a time trend of the relationship for the specified region.
Each point represents a specific year for that region. The black line represents the overall global trend for age-standardized death rate of CVD in relation to SDI.
Roth et al.   Figure 5A). IHD accounted for almost one-half of all CVD cases in Central Asia and Eastern Europe, but a smaller proportion in Central Europe, where other cardiovascular and circulatory diseases made up a larger proportion of total cases (Figure 4).
Eastern sub-Saharan Africa, the Middle East/North The death rate due to IHD rose steeply above age 40 A m e r ic a C a r ib b e a n O c e a n ia E a s t A s ia

S o u t h e a s t A s ia S o u t h A s ia N o r t h A f r ic a a n d M id d l e E a s t C e n t r a l S u b -S a h a r a n A f r ic a E a s t e r n S u b -S a h a r a n A f r ic a S o u t h e r n S u b -S a h a r a n A f r ic a W e s t e r n S u b -S a h a r a n A f r ic a H ig h -in c o m e A s ia P a c if ic A u s t r a l a s ia
This figure displays the relative distribution of age-standardized prevalence by CVD cause for 21 GBD world regions. Abbreviations as in years, increasing from an estimated 33 deaths per 100,000 (95% UI: 32 to 35 per 100,000) for those 40 to 44 years of age to 1,050 per 100,000 (95% UI: 1,025 to 1,076 per 100,000) by ages 75 to 79 years (Online Figure 4B). Above 80 years of age, the IHD death rate was estimated to be more than twice that rate (2,671 per 100,000; 95% UI: 2,600 to 2,738 per 100,000) and was by far the leading global cause of death.
The estimated age-standardized IHD death rate was highest in Central Asia (336 per 100,000; 95% UI: 326 to 347 per 100,000) and Eastern Europe (326 per 100,000; 95% UI: 319 to 333 per 100,000), followed by Oceania, South Asia, and the Middle East/North Africa ( Table 3). These regions, as well as high-income North America and Latin America, had particularly high proportions of total CVD deaths that were due to IHD   1,248 to 1,502 per 100,000) for those >80 years of age.
The prevalence was highest in Western sub-Saharan Africa, followed by Central and Eastern sub-Saharan Africa, tropical Latin America, and the Caribbean.
The lowest rates were estimated for Western and Eastern Europe.
There were 962,400 deaths (95% UI: 873,600 to 1,024,500 deaths) due to hypertensive heart disease in 2015. The mortality rate rose for ages >60 years, peaking at 296 per 100,000 (95% UI: 257 to 315 per 100,000) for age >80 years. Death rates due to hypertensive heart disease followed a similar pattern as the condition's prevalence.     This death rate was almost 5Â higher, 116 per 100,000 (95% UI: 90 to 145 per 100,000) above 80 years of age than below that age. Atrial fibrillation mortality rates  per 100,000 (95% UI: 13 to 15 per 100,000), which was twice the rate of the next-highest region, Southern to 24 per 100,000) older than 80 years of age.
Age-standardized mortality rates were greatest in   A m e r ic a C a r ib b e a n O c e a n ia E a s t A s ia

S o u t h e a s t A s ia S o u t h A s ia
N o r t h A f r ic a a n d M id d l e E a s t C e n t r a l S u b -S a h a r a n A f r ic a E a s t e r n S u b -S a h a r a n A f r ic a S o u t h e r n S u b -S a h a r a n A f r ic a W e s t e r n S u b -S a h a r a n A f r ic a H ig h -in c o m e A s ia P a c if ic A u s t r a l a s ia This figure displays the relative distribution of age-standardized prevalence by CVD cause for 21 GBD world regions. Abbreviations as in

CHANGES IN THE DECLINE OF CVD MORTALITY.
Of particular concern is that CVD age-standardized mortality shows less decline in the past 5 years than over the past 25 years. This trend, which is most obvious for IHD and aortic aneurysm, is observed not only in high-income countries, but also in Central Latin America for men. Regions with very high rates of CVD that have declined rapidly, such as Central Asia and Eastern Europe, also see moderation in that decline. Our use of the most recently available mortality data (through 2013 in many high-income countries) may explain why our findings differ from a recent analysis of CVD trends (29). Although an explanation of stagnation in declining CVD mortality is beyond the scope of this analysis, several possibilities can be considered. Rising rates of obesity may be increasing CVD risk over a short period of time (30). Interventions that reduce CVD mortality rates may have maximally diffused to the population able to access them, whereas interventions to address obesity are more challenging to implement. Some CVD risk factors, in particular air pollution or changes in average temperature, may account for larger increases of CVD mortality than previously suspected (31,32). Improving methods for estimating the most likely future trajectories for CVD is an important area for further research. PERIPHERAL ARTERIAL DISEASE. We estimated that PAD is the most prevalent cardiovascular condition globally, although low estimated rates of claudication and mortality made it a minor contributor to DALYs.
The high prevalence of PAD in comparison to IHD is a notable finding that may reflect the ease of its diagnosis using ABI, compared with more complex diagnostic testing required for IHD. Further attention should be paid to the use of ABI or palpation of foot pulses as a screening tool for overall vascular risk in low-income settings (41,42). There is some evidence that even those with asymptomatic PAD would benefit significantly from inexpensive medications, such as an angiotensin-converting enzyme inhibitor or antiplatelet agents (43,44 (48,49). The GBD study takes several steps to improve the reliability and comparability of vital registration data, including redistribution of garbage codes, but some systematic bias due to regional patterns in the use of diagnosis codes may remain. For example, the relatively large number of deaths coded to cardiomyopathy in Balkan countries may lead to an underestimate of the true number of IHD deaths. Many stroke deaths are coded to nonsubtype-specific stroke codes, which, when combined with lack of access to computed tomography scanners to aid in the acute management of stroke, adds additional uncertainty to the breakdown of stroke into ischemic and hemorrhagic subtypes. Uncertainty regarding stroke subtype is of particular concern in South Asia, where no subtype-specific mortality data were found, and the global proportion of ischemic and hemorrhagic stroke was used instead. The rapid rise, globally, in death certificates coded to atrial The GBD study has reported subnational estimates for a growing number of countries; however, the current report is limited to country-level estimates (50)(51)(52). There is substantial small-area variation in CVD burden within countries, and our national estimates represent only an average level for an entire country. Even these national estimates are important starting points for improving evidence available to policymakers.
The GBD study accounts for comorbidity using a simulation method that assumes an independent probability of having any disease state. A sensitivity analysis found that taking dependent and independent comorbidity into account changed overall estimated YLDs by a small amount that increased with age, ranging from 0.6% to 3.4%. Because CVD is more common at older ages, the assumption of independence may have a larger effect on this group of causes.
Unfortunately, data on the full correlation structure of prevalent CVD conditions remains limited.
GBD includes an estimate of measurement error, reported as a 95% UI, for each result. Our ability to detect significant trends over time is limited for those regions where UIs are wide, as seen on our global map of change in prevalence over time. Some countries may have experienced a rise or fall in CVD burden that we cannot detect because of limited data.
Although inclusion of measurement error is a strength of the GBD study, nonsampling error has not been quantified. GBD uses a wide range of validation methods, but relies on in-sample and out-of-sample validity testing to guide model selection. Additional sources of error in the GBD study may include regional patterns of clinical diagnosis, death code redistribution, selection of data sources, covariate selection for selected models, and measurement of the SDI. For example, measures of wealth, rather than income per capita, could potentially capture additional aspects of the epidemiological transition for some countries. Limited data on PAD with increased ABI due to noncompressible arteries may lead to underestimates of its true burden. The burden due to Chagas cardiomyopathy has not been included, but is estimated by the GBD study as a sequelae of Chagas disease. Chronic kidney disease and congenital heart disease are also estimated by the GBD study and have been reported separately.
Overall, the GBD study is most likely to underestimate uncertainty for those geographies where few data sources are available.