Global, regional, and national burden of stroke, 1990–2016: a systematic analysis for the Global Burden of Disease Study 2016

1Globally, stroke is a leading cause of mortality and disability and there are substantial economic costs for post-stroke care. Results from the 2015 iteration of the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) showed that although the age-standardised death rates and prevalence of stroke have decreased over time, the overall burden of stroke has remained high. As populations age, and low-income and middle-income countries go through the epidemiological transition from infectious to non-communicable diseases as the predominant cause of morbidity, together with concomitant increases in modifiable risk factors, it is expected that the burden of stroke will further increase until effective stroke prevention strategies are more widely implemented.

2Although estimates of disease burden for stroke have been produced by other research groups by meta-analysing data in the literature on incidence and deaths,,, GBD is unique in its approach to generating estimates for all locations, including those with scarce or no epidemiological data, by using all available data from the literature, administrative hospital and medical claims records, and cause of death records. Additionally, the methods used by GBD allow unspecified stroke to contribute to both fatal and non-fatal estimates. These methods allow GBD to document disease burden from stroke in the most comprehensive way over time and to provide the necessary information for priority setting and planning of health services. The results provided here are the most up-to-date estimates of death, prevalence, incidence, and disability for overall stroke and the pathological types of ischaemic and haemorrhagic stroke, using the standard GBD metrics of deaths, prevalence, incidence, years of life lost (YLLs), years lived with disability (YLDs), and disability-adjusted life-years (DALYs).

3Evidence before this studyThe Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) produces the only comprehensive estimates of global, regional, and country-specific burden due to stroke from 1990 to 2016. Other sources of population-level estimates include reports from WHO and independent scientific publications of global or regional estimates of deaths or incidence. GBD is the only peer-reviewed, comprehensive assessment of stroke by age, sex, and location that is updated annually. This study quantifies stroke burden in terms of incidence, prevalence, deaths, years lived with disability, years of life lost, and disability-adjusted life-years, updating the estimates previously presented in the GBD 2013 and GBD 2015 studies to include results through 2016. The results presented here are also the source data for the recently published estimates of the lifetime risk of stroke.Added value of this studyThere were several important updates to this iteration of GBD, enabling generation of improved estimates. First, we developed new approaches for our inpatient hospital data processing, which allowed us to include data for several locations that had previously been excluded because of insufficient information about the catchment population. Second, we extended the terminal age group of 80 years and older into 80–84 years, 85–89 years, 90–94 years, and 95 years and older. We updated the previous systematic literature review performed as part of GBD 2013 (search terms “stroke” or “hemorrhagic stroke” combined with “incidence”, “prevalence”, “epidemiology”, or “mortality”) to capture any substantial new sources of data on stroke prevalence, incidence, and mortality. We also performed a literature review to inform the models used to generate severity information (search terms “stroke”, “cerebral infarction”, “cerebral hemorrhage”, or “subarachnoid hemorrhage” combined with “Rankin”). Furthermore, we generated expected values for all measures on the basis of sociodemographic development, allowing us to visualise comparisons between observed GBD results and these expected values.Implications of all the available evidenceThe findings presented in this manuscript provide crucial information that could serve as the basis for resource allocation for stroke prevention, evidence-based planning for acute stroke care, and stroke rehabilitation facilities. Additionally, we provide evidence that most of the burden of stroke can be attributed to modifiable risk factors and identified risk clusters that can be targeted to reduce the incidence of stroke. Because stroke has been identified as one of the priority areas for WHO and the UN in their actions to reduce the burden of non-communicable diseases, global estimates such as those generated by GBD are essential for appropriately targeting such efforts.

4Evidence before this study

5The Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) produces the only comprehensive estimates of global, regional, and country-specific burden due to stroke from 1990 to 2016. Other sources of population-level estimates include reports from WHO and independent scientific publications of global or regional estimates of deaths or incidence. GBD is the only peer-reviewed, comprehensive assessment of stroke by age, sex, and location that is updated annually. This study quantifies stroke burden in terms of incidence, prevalence, deaths, years lived with disability, years of life lost, and disability-adjusted life-years, updating the estimates previously presented in the GBD 2013 and GBD 2015 studies to include results through 2016. The results presented here are also the source data for the recently published estimates of the lifetime risk of stroke.

6Added value of this study

7There were several important updates to this iteration of GBD, enabling generation of improved estimates. First, we developed new approaches for our inpatient hospital data processing, which allowed us to include data for several locations that had previously been excluded because of insufficient information about the catchment population. Second, we extended the terminal age group of 80 years and older into 80–84 years, 85–89 years, 90–94 years, and 95 years and older. We updated the previous systematic literature review performed as part of GBD 2013 (search terms “stroke” or “hemorrhagic stroke” combined with “incidence”, “prevalence”, “epidemiology”, or “mortality”) to capture any substantial new sources of data on stroke prevalence, incidence, and mortality. We also performed a literature review to inform the models used to generate severity information (search terms “stroke”, “cerebral infarction”, “cerebral hemorrhage”, or “subarachnoid hemorrhage” combined with “Rankin”). Furthermore, we generated expected values for all measures on the basis of sociodemographic development, allowing us to visualise comparisons between observed GBD results and these expected values.

8Implications of all the available evidence

9The findings presented in this manuscript provide crucial information that could serve as the basis for resource allocation for stroke prevention, evidence-based planning for acute stroke care, and stroke rehabilitation facilities. Additionally, we provide evidence that most of the burden of stroke can be attributed to modifiable risk factors and identified risk clusters that can be targeted to reduce the incidence of stroke. Because stroke has been identified as one of the priority areas for WHO and the UN in their actions to reduce the burden of non-communicable diseases, global estimates such as those generated by GBD are essential for appropriately targeting such efforts.

10Methods used to generate estimates of stroke incidence, mortality, prevalence, YLDs, YLLs, and DALYs have been described in previously; additional details are in the appendix., Sources included in all models can be accessed via the GBD 2016 Data Input Sources Tool. For all models, point estimates were calculated from the mean of 1000 draws from the posterior distribution by age, sex, location, and year. 95% uncertainty intervals (UIs) were the 25th and 975th values of the ordered draws. The study was compliant with GATHER guidelines.

11Stroke was defined according to WHO criteria as rapidly developing clinical signs of focal (at times global) disturbance of cerebral function lasting more than 24 h or leading to death with no apparent cause other than that of vascular origin. Data on transient ischaemic attack were not included because of the very short period of disability and no associated mortality for these events. We modelled acute and chronic stroke separately. Stroke cases were considered acute from the day of incidence of a first-ever stroke through day 28 after the event. Stroke cases were considered chronic (prevalent) from 29 days after the occurrence of an event. Chronic stroke included the sequelae of an acute stroke and all recurrent stroke events. 28 days was selected as the cutoff between acute and chronic stroke because this corresponds to the period of early case fatality.

12Incident strokes were defined as the occurrence of first-ever stroke on the basis of a clinical diagnosis by a physician according to the WHO criteria described above. Ischaemic strokes were defined as all atherosclerotic and thromboembolic events that resulted in compromised blood flow to brain tissue and subsequent infarction. Haemorrhagic strokes were defined as all non-traumatic events due to subarachnoid or intracerebral haemorrhage identified by neuroimaging.

13Standard Cause of Death Ensemble modelling (CODEm) methods were used to estimate cause-specific mortality. The International Classification of Diseases (ICD) 9 and 10 codes that we used are listed in the appendix. For overall stroke, we included verbal autopsy data in addition to vital registration data; for the stroke type models, we used vital registration data only because accurate assessment of stroke type requires imaging studies that are not commonly available in populations where causes of death are ascertained by verbal autopsy. Covariates included in the models were chosen on the basis of an assessment of causal associations for the risk factors or markers of access to care (appendix).

14We used DisMod-MR 2.1, a Bayesian meta-regression tool, to model the non-fatal burden of stroke. Estimates were generated using a two-stage modelling approach. In the first stage, we ran four models (acute ischaemic, chronic ischaemic, acute haemorrhagic, and chronic haemorrhagic stroke) using only incidence, prevalence, and excess mortality data as inputs. We then used the ratio of acute to chronic cause-specific mortality estimated by these models to divide the ischaemic-specific and haemorrhagic-specific stroke deaths estimated in CODEm into acute and chronic proportions. The four models were then re-run using the same incidence, prevalence, and excess mortality data as well as the custom cause-specific mortality as input data, thus generating internally consistent fatal and non-fatal estimates.

15For the acute models, we used all available high-quality incidence and case fatality data from registries and published literature along with inpatient hospital data on incident events. Acute unspecified stroke was split according to the ratio of ischaemic to haemorrhagic stroke for each combination of sex, age group, and geographic location. The ICD codes used for data from inpatient hospital data sources are listed in the appendix. First-ever, type-specific (ischaemic vs haemorrhagic) data from stroke registries were the reference. Datapoints that did not meet our reference case definition, such as those that included recurrent stroke, did not report type-specific data, or only included hospital admissions, were adjusted in DisMod. Prevalence data from surveys, along with the incidence of those surviving the first 28 days calculated from the acute models were included as input data for the chronic models. Counts of data points and covariates and model settings for DisMod are in the appendix. Detailed descriptions of health states, lay descriptions, distributions of functional and cognitive disability, and disability weights for stroke sequelae in GBD 2016 are in the appendix.

16SDI was developed for GBD 2015 as a metric of overall development that positions all locations on a spectrum of socioeconomic development, using educational attainment, lagged distributed income, and total fertility rate. For GBD 2016, this index was updated such that minimum scores are the lowest observed level of GDP per capita or educational attainment or highest observed level of total fertility rate in known datasets. Maximum scores are now plateaus in the relationships between the component parts of the index and selected mortality or health outcomes, indicating no additional benefit to increases in education or lagged distributed income or decreases in total fertility rate. Gaussian process regression was used to establish the average relationship between cause-specific, age-standardised DALY rates and SDI for all locations from 1990 to 2016. These rates were used as the expected values for DALYs in comparisons between observed and expected rates.

17The comparative risk assessment framework developed for GBD was used to estimate levels and trends in attributable burden of stroke due to risk factors that satisfied the criteria of sufficient evidence of a causal relationship, availability of exposure data, and potential for modification. Four components were incorporated into estimating attributable burden using this approach: (1) burden estimates for stroke; (2) exposure levels for each risk factor; (3) relative risk of stroke as an outcome of exposure to the risk factor; and (4) theoretical minimum risk exposure level—ie, the level of exposure that minimises risk for each individual in the population. The population attributable fraction, estimated independently for each risk factor, is the proportion of the cause that would be decreased if the exposure to the risk factor in the past had been reduced to the counterfactual level of the theoretical minimum risk exposure level. Estimates of attributable burden for each risk–outcome pair were established by multiplying the relevant cause measure by the population attributable fraction. All estimates of attributable burden are generated at the most detailed level and estimates for risk groupings or all risk factors combined are generated via an aggregation process that accounts for the fact that the effect of one risk factor might be partly or completely mediated through the effect of another. This mediation analysis is informed by individual-level data from prospective cohort studies on the joint effects of combinations of risk factors.

18The funder of the study had no role in study design, data collection, data analysis, data interpretation, or the writing of the report. All authors had full access to the data in the study and had final responsibility for the decision to submit for publication.

19GBD stroke estimates for 1990–2016 are available for download from the GBD Results Tool at the Global Health Data Exchange. In 2016, stroke was the second largest cause of death globally (5·5 million [95% UI 5·3–5·7] deaths) after ischaemic heart disease (table). Fewer women died as a result of stroke (2·6 million [2·5–2·7] deaths) than did men (2·9 million [2·8–3·0] deaths). The number of global deaths due to ischaemic stroke (2·7 million [2·6–2·8]) was slightly lower than the number due to haemorrhagic stroke (2·8 million [2·7–2·9] deaths; appendix). Stroke was also the second most common cause of global DALYs (116·4 million [111·4 −121·4]), an increase from 1990 (95·3 million [91·6–100·6]). Women had fewer stroke DALYs (50·8 million [47·6–53·7]) than men (65·6 million [63·1–68·2]). The number of DALYs due to ischaemic stroke (51·9 million [47·9–55·6]) was lower than the number due to haemorrhagic stroke (64·5 million [62·6–66·5]; appendix). There were 80·1 million (74·1–86·3) prevalent cases of stroke globally in 2016: 41·1 million (38·0–44·3) prevalent cases in women and 39·0 million (36·1–42·1) prevalent cases in men. Of the total number of prevalent strokes, 84·4% (82·1–86·4) were ischaemic. There were 13·7 million (12·7–14·7) new stroke cases in 2016.

20The highest age-standardised incidences of stroke were observed in east Asia, especially China (354 [95% UI 331–378] per 100 000 person-years), followed by eastern Europe, ranging from 200 (181–218) per 100 000 person-years in Estonia to 335 (301–369) per 100 000 person-years in Latvia (figure 1). The lowest incidences were in central Latin America, especially El Salvador (97 [88–105] per 100 000 person-years). Age-specific stroke incidence was similar between men and women younger than 55 years, but significantly greater for men than women at ages 55–75 years (figure 2).

21Age-standardised incidence declined from 1990 to 2016 globally (−8·1% [–10·7 to −5·5]), in all SDI groups except the middle SDI group, and in most regions (table; appendix). The region with the largest decrease in age-standardised stroke incidence was southern Latin America (−33·3% [–36·4 to −29·7]) and the region with the largest increase was east Asia (4·9% [1·3 to 8·1]). For ischaemic stroke, the largest decrease was in southern Latin America (−38·0% [–39·4 to −36·6]), and the largest increase was in east Asia (17·5% [15·8 to 19·2]; appendix). For haemorrhagic stroke, incidence decreased in all regions. The largest decrease was in high-income Asia Pacific (−32·5% [–33·7 to −31·2]), and the smallest decrease was in southern sub-Saharan Africa (−5·1% [–6·2 to −4·0]; appendix).

22Globally, the age-standardised rate of deaths due to stroke decreased by 36·2% (−39·3 to −33·6) from 1990 to 2016, with decreases in all five SDI groups. These death rates also declined for all but one region from 1990 to 2016, with the largest decrease in the high-income Asia Pacific region (−66·3% [–68·8 to −63·4]) and no significant change in southern sub-Saharan Africa (−3·8% [–11·4 to 6·0]; table). These results were similar for ischaemic stroke and haemorrhagic stroke, with the largest decrease for both in the high-income Asia Pacific region (−70·2% [–72·3 to −67·8] for ischaemic stroke and −59·8% [–63·1 to −56·1] for haemorrhagic stroke; appendix). Death rates for neither ischaemic nor haemorrhagic stroke changed significantly between 1990 and 2016 in southern sub-Saharan Africa (0·6% [–7·9 to 11·9] for ischaemic stroke and −7·2 [–15·1 to 2·2] for haemorrhagic stroke).

23Age-standardised DALY rates for stroke also declined from 1990 to 2016 globally (34·2% [–37·2 to −31·5]) for all SDI quintiles, and for all regions, again with the largest change occurring in high-income Asia Pacific (−61·5% [–64·9 to −58·1]) and the smallest in southern sub-Saharan Africa (−7·3% [–14·7 to 1·8]; table). Southern Latin America was the region with the largest decrease for ischaemic stroke (−63·7% [–66·8 to −60·6]; appendix) and high-income Asia Pacific was the region with the largest decrease for haemorrhagic stroke (−59·9% [–63·4 to −55·8]; appendix). Southern sub-Saharan Africa had no change in ischaemic stroke (−1·9% [–9·9 to 7·4]) and a decrease in haemorrhagic stroke (−10·2% [–17·9 to −1·1]).

24Rates of YLLs and YLDs were very low for the younger age groups (<40 years) and then increased substantially with age, with YLLs increasing much more rapidly than YLDs (figure 3) because of the high mortality burden of stroke.

25After an increase in expected DALY rates at the lower end of the SDI scale, these rates decline rapidly for SDI values of 0·35 and higher (figure 4). For most regions, the burden of stroke decreased with increases in SDI over time. However, central and eastern Europe and central Asia saw increased DALY rates in the early 1990s after the dissolution of the Soviet Union, followed by subsequent decreases as SDI increased. Stroke DALY burden in southern sub-Saharan Africa showed a similar pattern, with an initial spike in rates with increasing SDI, followed by a steady decrease. DALY rates were higher than expected early in the time series for eastern and central sub-Saharan Africa and high-income Asia Pacific but have decreased with increasing SDI. Eastern Europe, central Europe, central Asia, Oceania, and east Asia had higher stroke-related DALY rates than would be expected on the basis of comparisons of SDI for all years. Conversely, rates for Latin America, western Europe, Australasia, south Asia, and southern and western sub-Saharan Africa were lower than expected for all timepoints. Although DALY rates in southeast Asia were initially lower than expected, they were in most recent years slightly higher than expected.

26Under the comparative risk assessment framework, most stroke DALYs (88·8% [95% UI 86·5–90·9]) can be attributed to risk factors measured in GBD; this percentage is similar for both stroke types (87·9% for ischaemic stroke [84·1–91·6] and 89·5% for haemorrhagic stroke [87·1–91·6]). Metabolic risks (high systolic blood pressure, high body-mass index, high fasting plasma glucose, high total cholesterol, and low glomerular filtration rate) accounted for 72·1% (66·4–77·3) of stroke DALYs. Behavioural factors (smoking, poor diet, and low physical activity) accounted for 66·3% (59·3–73·1) of attributable DALYs, and environmental risks (air pollution and lead exposure) for 28·1% (25·3–30·9). As the effect of many of these risk factors are mediated partly or wholly through another risk factor, the crude sum of the groups is expected to exceed 100%. The aggregation process to generate estimates of overall attributable burden accounts for joint effects of a combination of risk factors, thus the final estimate is less than 100%. The remaining burden is due to unknown or unmeasured risk factors, genetic factors, or the effect of gene–environment interactions. Population attributable fractions and UIs for the top ten risk factors for each stroke subtype by sex in 1990 and 2016 are in the appendix.

27Our estimates indicate that the global burden of stroke is high, with more than 80 million stroke survivors in 2016. Age-standardised death rates from stroke have decreased in all regions from 1990 to 2016, whereas incidence has decreased in most regions but increased in east Asia and southern sub-Saharan Africa. The overall burden of stroke, as quantified by age-standardised DALY rates, decreased from 1990 to 2016, but the absolute number of DALYs due to stroke increased over that same period. The increase in absolute numbers is largely due to population growth and ageing resulting in a greater number of people with stroke despite declining incidence and improved stroke survival leading to higher prevalence of chronic stroke.

28Studies have shown that much of the burden due to stroke can be attributed to modifiable atherosclerotic risk factors. INTERSTROKE, a case-control study done at 32 locations, found that the risk factors for stroke in low-income and middle-income countries were similar to those in high-income countries, although the relative contribution of each differed between regions. The high burden of stroke worldwide suggests that primary prevention strategies are either not widely implemented or not sufficiently effective. In addition to targeting behavioural risk factors, effective screening for conditions that increase stroke risk, such as hypertension, atrial fibrillation, and diabetes mellitus, is essential. Many screening strategies use the predicted absolute risk of cardiovascular disease to identify individuals at high risk of cardiovascular disease events and to define therapeutic thresholds for specific interventions., However, these approaches have limitations, including low efficiency and missing data for people with low to moderate cardiovascular disease risk, in whom about 80% of strokes occur., Preliminary evidence suggests that strategies via mobile technologies are effective for healthy lifestyle modification and primary stroke prevention., Treatment with statins and blood pressure medications has been shown to be effective and cost-effective for both primary and secondary prevention of stroke.,, Healthy lifestyle modification and better adherence to recommended medications via an affordable multidrug polypill containing blood pressure and lipid-lowering medications could potentially also enable cost-effective prevention of stroke globally, potentially halving stroke incidence and mortality.,,

29In addition to prevention efforts, appropriate acute and long-term treatment is essential, given the high recurrence rate of stroke. Highly effective treatments for stroke have been developed over the past few decades and are now considered the standard of care where available., To assist countries in identifying gaps in stroke care, a survey has been done in collaboration with the World Stroke Organization and WHO to obtain data on facilities and providers for acute stroke care and rehabilitation; results from the survey will be available soon. Studies are also underway to assess different approaches to treatment when the closest medical facilities do not have the resources to provide advanced stroke care (NCT02795962).

30Although the attributable burden for most of the risk factors identified for stroke has been quantified in GBD, the effect of atrial fibrillation has not yet been estimated. Atrial fibrillation increases the risk of stroke up to five fold, largely through an increased risk of thrombotic events leading to ischaemic stroke. Antithrombotic therapy with vitamin K antagonists, antiplatelet drugs, or novel oral anticoagulants have been shown to reduce this risk by up to 60% in a meta-analysis of clinical trials. Improved diagnosis of and treatment for atrial fibrillation is thus likely to have a substantial impact on stroke burden.

31GBD faces several measurement challenges for estimating cause-specific mortality and non-fatal burden of stroke. Although GBD employs spatiotemporal methods that use patterns across time and geographic regions to inform estimates for locations with sparse data, these approaches cannot completely overcome issues when data are missing for some large geographic regions. Although much stroke data is available for some regions, data on incidence, stroke type, and stroke severity is sparse in many low-income and middle-income countries. Adding new sources of data is an ongoing effort of GBD, but we are limited to locations where representative studies have been done or where there is access to administrative data. To ensure that inpatient hospital data capture all events, adjustments are made to these data using correction factors derived from medical claims data, to which we only had access for the USA in GBD 2016. The generalisability of claims data and the derived correction factors have been questioned.,, However, data from administrative sources provide essential information for capturing burden in many locations, and we are able to adjust the estimates during the modelling process using stroke-specific registries and studies as the reference. Using similar methods, we are also able to include data from the literature that do not meet our reference case definition of subtype-specific, first-ever stroke.

32Improvements in modelling strategy are made regularly for causes in GBD, including stroke. In GBD 2016, we did not estimate the burden due to subarachnoid haemorrhage and intracerebral haemorrhage separately; however, future iterations of GBD will. We also do not generate estimates for transient ischaemic attacks; however, incidence estimates for these events would be useful for health planning purposes. The new ICD 11 classifications, which include imaging criteria, could be used along with the original clinical WHO diagnostic criteria in future estimates.

33Our findings with regard to stroke burden and modifiable risk factors are consistent with those from GBD 2015 and emphasise the need for effective prevention strategies. Although age-standardised deaths due to stroke have been decreasing, the overall burden of stroke remains high, continues to increase, and is unlikely to decrease without interventions to address stroke risk factors. Stroke has been identified as one of the priorities for WHO and the UN in their actions to reduce the burden of non-communicable diseases; global estimates such as those generated by GBD are essential in appropriately targeting efforts.