Distribution and Temporal Trends From 1993 to 2015 of Ischemic Stroke Subtypes
A Systematic Review and Meta-Analysis
Background and Purpose—Preventive strategies, together with demographic and socioeconomic changes, might have modified the worldwide distribution of ischemic stroke (IS) subtypes. We investigated those changes by means of a systematic review and meta-analysis.
Methods—We evaluated all population- and hospital-based studies reporting the distribution of IS etiologic subtypes according to the TOAST criteria (Trial of ORG 10172 in Acute Stroke Treatment). Studies were identified by searching articles indexed on PubMed and Scopus from January 1, 1993, to June 30, 2017. Two independent investigators extracted data and checked them for accuracy. Proportions of each etiologic subtype were pooled according to a random effect meta-analytic model weighted by study size; temporal trends were assessed using a mixed-effect meta-regression model.
Results—Sixty-five studies including patients from 1993 to 2015 were finally included. Overall, ISs were attributed to cardioembolism (22%; 95% confidence interval [CI], 20–23); large artery atherosclerosis (23%; 95% CI, 21–25); small artery occlusion (22%; 95% CI, 21–24); other determined cause (3%; 95% CI, 3–3); and undetermined cause (26%; 95% CI, 24–28). Cardioembolism was the leading IS etiologic subtype in whites (28%; 95% CI, 26–29) and large artery atherosclerosis in Asians (33%; 95% CI, 31–36). Meta-regression showed an increasing temporal trend for cardioembolism in whites (2.4% annually, P=0.008) and large artery atherosclerosis in Asians (5.7% annually, P<0.001), and a decrease for small artery occlusion in whites (−4.7% annually, P=0.001); there was considerable heterogeneity across all the analyses.
Conclusions—According to our systematic review and meta-analysis, cardioembolism in whites and large artery atherosclerosis in Asians are the leading causes of IS. The heterogeneous distribution of etiologic subtypes of IS may depend on the demographic and socioeconomic characteristics of the different populations. More extensive protocols should be adopted to reduce the persistently relevant proportion of undetermined cause IS.
A wide geographic variation of the burden of stroke between 1990 and 2010 has been described, with a decrease of stroke incidence and mortality in high-income countries paralleled by an increase of stroke incidence in low-to-middle-income countries.1
Ischemic stroke (IS) still represents the most common type of stroke2; from 1990 to 2010, its global burden significantly increased worldwide, with a decline of incidence and mortality in high-income countries, and a modest increase of incidence in low-to-middle-income countries.3
Effective preventive strategies, together with demographic and socioeconomic changes, might have modified not only the global burden of IS, but also the worldwide distribution of its etiologic subtypes in the last decades, including cardiac embolism, atherothrombosis, small vessel disease, and several less common etiologies. Accordingly, our aim was to describe the worldwide distribution and the temporal trends of etiologic subtypes of IS by means of a systematic review and meta-analysis.
The meta-analysis was performed according to the PRISMA guidelines (Preferred Reporting Items for Systematic Reviews and Meta-Analyses).4 The authors declare that all supporting data are available within the article and in the online-only Data Supplement.
Full articles were eligible for inclusion if they (1) reported the distribution of etiologic subtypes of IS, (2) used TOAST criteria (Trial of ORG 10172 in Acute Stroke Treatment)5 to define IS etiologic subtypes as cardioembolism, large artery atherosclerosis (LAA), small artery occlusion (SAO), other determined cause (OC), and undetermined cause (UND; Table I in the online-only Data Supplement), (3) had a population-based or hospital-based cohort design and consecutively recruited patients with IS, (4) fulfilled key methodological criteria and had a clear definition of study design and setting, and (5) were written in English. Exclusion criteria were (1) study design other than those reported above, (2) studies not reporting distribution of IS attributed to cardioembolism, LAA, and SAO, whereas studies presenting aggregate data from other determined cause and UND were potentially eligible, (3) studies not performing basic examinations to allow the diagnosis of the etiologic subtype according to the TOAST criteria, (4) patient selection according to other specific conditions (eg, comorbid diseases and performance of ancillary diagnostic tests), (5) study population limited to specific age groups (eg, young adults and exclusively pediatric study population), (6) patient selection according to specific stroke location (eg, anterior circulation strokes), (7) duplicate populations (in the case of duplicate or overlapping populations among studies we included the study reporting the largest number of patients), (8) pooled data for patients with IS and transient ischemic attack, (9) excessively high proportions (>45%) of UND IS, suggesting a high rate of incomplete investigations, and (10) studies reporting data about cases collected before the release of TOAST criteria in 1993.
PICOS criteria (Participants, Intervention, Comparator, Outcome, Study design) for inclusion and exclusion of potentially eligible studies are summarized in Table II in the online-only Data Supplement.
Search Strategy, Study Selection, and Data Collection
Studies were identified by searching articles indexed on PubMed and Scopus from January 1, 1993, to June 30, 2017. The following search terms were used in both databases: "brain ischemia" or “brain infarction” or “ischemic stroke” and (TOAST or atherothrombosis or atherosclerosis or “large vessel” or cardioembolic or lacunar or “small vessel”). Filters for English language were applied. We followed a 2-step procedure in the process of studies identification. In the first step, titles and abstracts were screened to verify study eligibility by 2 independent investigators (C.D.C. and L.P.). In the second step, full texts of selected studies were evaluated, where appropriate, by the same 2 independent investigators. Every disagreement was solved by consensus among all the study participants. The reference lists and Google Scholar citations for the selected articles were also searched. A manual search among references of selected articles and reviews was also performed. Duplicate publications were removed by manual check.
A data extraction sheet was developed. Data extracted from each eligible study included: country where the study was performed, study design, inclusion period, number of cases, sex distribution, age limits, mean age, race/ethnicity, proportion of patients undergoing diagnostic examinations, patient distribution according to TOAST criteria, and acute first or recurrent IS. Whenever the inclusion period was not clearly reported, missing information was obtained by contacting authors. Information was extracted by 2 independent investigators (C.D.C. and L.P.) and checked for accuracy with original articles. Race/ethnicity was reported in accordance with the 1997 Office of Management and Budget standards.6 The revised standards contain 5 minimum categories for race (American Indian or Alaska Native, Asian, black, Native Hawaiian or other Pacific Islanders, and white) and 2 categories for race/ethnicity (Hispanic or Latino and not Hispanic or Latino).6 Where available, race/ethnicity was considered at the patient level as reported in the studies. Where race/ethnicity was not reported in the study, it was inferred from the country-specific predominant race/ethnicity7 and applied to the whole study population. Country income level was classified as high- or low- to middle-income according to the World Bank’s country criteria.8
To perform the meta-analysis, we calculated proportions of each TOAST IS etiologic subtype from the crude numbers provided in the selected articles. We pooled the proportions of each etiologic subtype according to a random effect model9 weighted by study size. Pooled proportions were calculated separately for different ethnic groups and were compared using the Q test for between-group variability. Prespecified subgroup analyses were performed according to country income, population- or hospital-based study design, and acute first and recurrent IS. Temporal trends in the distribution of IS etiologic subtypes were assessed using a mixed-effect meta-regression model using the central year of the inclusion period of each study as the moderator variable; annual percent variations were estimated with the formula 100×[exp(B)-1], whereas 95% confidence intervals (CIs) were estimated with the formula exp(B±1.96×SE), where B is the regression coefficient and SE its standard error. We performed prespecified analyses for IS etiologic subtype distribution according to pragmatically established inclusion periods (1993–2003 and 2004–2017). In accordance with the Cochrane Collaboration Guidelines for systematic reviews,10 we assessed the clinical, methodological, and statistical heterogeneity of the included studies. Clinical heterogeneity was assessed by evaluating differences in the characteristics of study populations. Methodological heterogeneity was assessed by comparing the differences in study design and protocols. Statistical heterogeneity was assessed using the I2 statistic. A value of I2 of 0% to 40% was considered not important, 30% to 60% represented moderate heterogeneity, 50% to 90% substantial heterogeneity, and 75% to 100% considerable heterogeneity. Statistical significance for heterogeneity was set for P values <0.05. We used the SPSS version 20 and the R software for all the statistical analyses.
Characteristics of the Included Studies
A flow chart of study selection is reported in Figure 1. Out of 10 887 retrieved articles, we finally included 65 studies (References in the online-only Data Supplement), from 29 countries, referring to a total of 85 populations. Fourteen (21.5%) of the 65 included studies were population-based and 51 (78.5%) hospital-based; 23 (35.4%) studies included first-ever ISs and 42 (64.6%) first and recurrent ISs. Details of the included studies and of the diagnostic tests are reported in Tables III and IV in the online-only Data Supplement, respectively. White was the predominant race/ethnicity in 42 populations (reported in 8), Asian in 30 (reported in 1), Hispanic in 7 (reported in 5), black in 5 (reported in 4), and American Indian in 1 (reported in 1). The meta-analysis finally included 128 172 IS cases from 1993 to 2015. Mean age (SD) was 67.6 (4.6) years and 62.3% of the patients were males.
Worldwide Distribution of IS Subtypes
As reported in Figure 2, ISs were attributed to each of the following: cardioembolism (22%; 95% CI, 20–23); LAA (23%; 95% CI, 21–25); SAO (22%; 95% CI, 21–24); other determined cause (3%; 95% CI, 3–3); and UND (26%; 95% CI, 24–28). The pooled proportion of IS attributed to cardioembolism was 28% in whites (95% CI, 26–29), 20% in blacks (95% CI, 14–27), and 15% in Asians (95% CI, 13–17), whereas the proportion of IS attributed to LAA was 33% in Asians (95% CI, 31–36), 19% in whites (95% CI, 17–21), 17% in Hispanics (95% CI, 11–25), and 12% in blacks (95% CI, 9–18). The proportion of IS attributed to SAO was 20% in whites (95% CI, 18–22) and 24% in Asians (95% CI, 22–27). Compared with whites from low-to-middle-income countries, whites from high-income countries had a higher proportion of IS attributed to cardioembolism (29% versus 22%) and a lower proportion of IS attributed to LAA (18% versus 25%) and SAO (19% versus 26%). Asians from high-income countries had a higher proportion of IS attributed to cardioembolism compared with those from low- to middle-income countries (17% versus 10%). Data referring to non-white and non-Asian races/ethnicities, involving a few studies, are anyhow reported in Figure 2. Notably, proportions of each IS etiologic subtype were significantly different across race/ethnicity (cardioembolism, P<0.001; LAA, P<0.001; SAO, P=0.002; other determined cause, P<0.001; UND, P<0.001).
Temporal Trends of IS Subtypes
The temporal trends in the distribution of IS etiologic subtypes estimated by meta-regression are reported only for whites and Asians in the Table, as data referring to other ethnic groups were scarce. The analyses were performed separately for population- and hospital-based studies. Over 2 decades (1993–2015), whites had an increased proportion of IS attributed to cardioembolism (2.4% annually; P=0.008), and a decreased proportion of IS attributed to SAO (−4.7% annually; P=0.001), whereas whites from high-income countries had a decreased proportion of IS attributed to LAA (−3.1% annually; P=0.047; Table). In the same time period, Asians had an increased proportion of IS attributed to LAA (5.7% annually; P<0.001; Table).
From 1993 to 2003 to 2004 to 2015, as showed in Table V in the online-only Data Supplement, we found an increased proportion of cardioembolism in whites from 26% to 30% and of LAA in Asians from 28% to 37%; those changes were paralleled by a decreased proportion of SAO in whites from 24% to 16% and no changes in Asians. Statistical heterogeneity was considerable across all pooled analyses (Table VI in the online-only Data Supplement) and did not change when considering only first-ever ISs (Table VII in the online-only Data Supplement).
We performed this systematic review and meta-analysis to identify worldwide distribution and temporal trends of the etiologic subtypes of IS according to the TOAST classification. The knowledge of the distribution of etiologic subtypes and of their temporal trends is useful to apply the appropriate current therapeutic strategies and to identify future patterns of intervention.
According to our results, the most common etiologic subtypes were represented by cardioembolism in whites (28%) and LAA in Asians (33%), whereas the proportions of SAO were 20% in whites and 24% in Asians, and UND accounted for 26% of IS worldwide. Over the last 2 decades, we also observed a relative increase in the proportion of IS attributed to cardioembolism that was paralleled by a decrease in the proportion of SAO in whites and an increase of LAA in Asians. However, we cannot exclude that some patients with SAO and a concomitant cardioembolic source might have been categorized as UND and that in the LAA subtype might have been included mostly patients with extracranial stenosis among whites and patients with intracranial stenosis among Asians (Table I in the online-only Data Supplement). The observed worldwide geographic and temporal variations possibly depended on changes of the demographic structure of the resident populations, on genetic and environmental factors, on the local implementation of primary preventive measures, on the system of care, and on the extent of the diagnostic examinations that were performed to identify IS subtypes.
The higher relative proportion of IS attributed to cardioembolism that was found in whites with respect to Asians is in line with the reported higher proportion of IS attributed to cardioembolism in population-based studies for whites (26%) as compared with that reported in hospital-based studies for Chinese individuals (16%).11 The relative increased proportion of cardioembolic stroke might have depended on increased search for cardioembolic sources, paralleled by population aging, increased proportions of risk factors contributing to atrial fibrillation (AF),12,13 and better detection and control over time of atherothrombotic risk factors. Indeed, only one of the included studies reported an increasing prevalence of arterial hypertension over time among patients with IS, paralleled by an increasing use of antihypertensive treatments,14 whereas the prevalence of AF increased in whites from high-income countries.14–16 The increasing prevalence of AF among patients with IS was explained by the aging of the population and by the increased awareness of AF-related embolic risk.14–16 In our opinion, more attention should be paid to the treatment of the arrhythmia and of its concurrent risk factors, to slow the progression from paroxysmal to persistent or permanent AF.
The increased relative proportion of IS attributed to LAA in Asians was possibly favored by the progressive adoption of a Western lifestyle, whereas the decreased proportion of IS attributed to SAO in whites, and mostly in those living in high-income countries, might be attributed to improved control of arterial hypertension, in the absence of reliable data for the SAO subtype.17,18 In Asians, the not significant decrease of SAO was not in line with the results of studies that were performed in South Korea and Japan which attributed the decrease of SAO to improved control of arterial hypertension.19,20
Twenty-six percent of IS overall were attributed to UND, with proportions ranging between 22% and 38% across the different races/ethnicities, possibly reflecting differences in diagnostic protocols (Table IV in the online-only Data Supplement) impacting on the TOAST classification. Those proportions are a matter of concern, as the correct identification of the etiologic subtypes is of the utmost importance for effective secondary prevention.
Strengths and Limitations
To our best knowledge, this is the most extensive attempt, to date, investigating the worldwide distribution of the IS etiologic subtypes. The TOAST classification is widely used in research and in the real-world settings and allowed the inclusion of the largest number of the available studies. Indeed, we recognize that our systematic review and meta-analysis may experience some of the intrinsic limitations of the TOAST classification, including the possible oversizing of UND IS and misclassification of SAO and cardioembolism subtypes.21
However, some further limitations are unrelated to that classification. First, the extent of diagnostic protocols was heterogeneous among studies, as shown in Table IV in the online-only Data Supplement. Second, some studies only included subjects with a first-ever IS, whereas other studies included first-ever and recurrent IS, introducing a further source of clinical and methodological heterogeneity as recurrent IS may represent a substantial proportion of cases in populations with inadequate secondary stroke prevention. Third, the majority of the included studies were hospital-based and their referral populations possibly had different proportions of hospitalization, further increasing the heterogeneity of the results. Indeed, milder strokes and mostly SAO might have been under-represented in the hospital-based studies. However, should we have excluded hospital-based studies from our analyses, the results would have been deemed incomplete for the exclusion of all the studies that were performed in Asians (References in the online-only Data Supplement). Anyhow, it is worth mentioning that the majority of the population-based studies were performed in whites. To overcome at least some of the potential sources of heterogeneity, we also performed subgroup analyses according to study design, race/ethnicity, and income of the different study populations. Subgroup analyses performed according to country income and study design in non-white and non-Asian race/ethnicities did not allow any conclusive comparison. The I2 values of >90% indicating considerable statistical heterogeneity in both overall and subgroup analyses suggest caution in their interpretation.
The exclusion of studies not written in English might have contributed to a selection bias, even if the retrieved full-text articles did not contain any reference to studies not written in English fulfilling the inclusion criteria. Notably, all the time trends of the reported proportions of IS etiologic subtypes were obtained from relative measures and did not take into account the temporal trends of IS incidence.
Considering all the reported limitations, our findings suggest that the previously observed and reported a reduction in stroke incidence in high-income countries1,3 is paralleled by changes in the proportions of IS etiologic subtypes. In whites, our data suggest a relative increase in potentially preventable IS because of cardioembolism. Better prevention and treatment of AF, paying attention to its natural history, should be encouraged among the aging white populations and mostly in high-income countries. In Asians, preventive strategies should focus on IS attributed to LAA and SAO, improving the implementation of the best available treatments. Overall, more extensive protocols should be adopted to reduce the proportion of the UND etiologic subtype of IS, a result that is achievable also with long-term follow-up studies to unveil novel or still undetected risk factors. Last, standard diagnostic protocols for IS should be implemented worldwide to overcome disparities in the access to appropriate diagnostic investigations.
Sources of Funding
This study was supported by the ex 60% grant from the Italian Ministero dell’Istruzione, dell’Università e della Ricerca.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.117.020031/-/DC1.
- Received November 7, 2017.
- Revision received February 12, 2018.
- Accepted February 15, 2018.
- © 2018 American Heart Association, Inc.
- Feigin VL,
- Forouzanfar MH,
- Krishnamurthi R,
- Mensah GA,
- Connor M,
- Bennett DA,
- et al
- Bennett DA,
- Krishnamurthi RV,
- Barker-Collo S,
- Forouzanfar MH,
- Naghavi M,
- Connor M,
- et al
- Adams HP Jr,
- Bendixen BH,
- Kappelle LJ,
- Biller J,
- Love BB,
- Gordon DL,
- et al
- 6.↵The White House–President Barack Obama: Office of Management and Budget. OMB Standards. https://obamawhitehouse.archives.gov/omb/fedreg_1997standards. Accessed April 10, 2017.
- 7.↵Central Intelligence Agency (CIA). The World Fact Book. https://www.cia.gov/library/publications/the-world-factbook/fields/2075.html. Accessed April 10, 2017.
- 8.↵The World Bank. Data on the Income Level of Countries. http://www.worldbank.org/en/country. Accessed April 10, 2017.
- Higgins JPT,
- Green S
- Chugh SS,
- Havmoeller R,
- Narayanan K,
- Singh D,
- Rienstra M,
- Benjamin EJ,
- et al
- Wyse DG,
- Van Gelder IC,
- Ellinor PT,
- Go AS,
- Kalman JM,
- Narayan SM,
- et al
- Rosengren A,
- Giang KW,
- Lappas G,
- Jern C,
- Torén K,
- Björck L
- Jespersen SF,
- Christensen LM,
- Christensen A,
- Christensen H
- Jackson C,
- Sudlow C
- Jung KH,
- Lee SH,
- Kim BJ,
- Yu KH,
- Hong KS,
- Lee BC,
- et al