Stroke Subtype Classification to Mechanism-Specific and Undetermined Categories by TOAST, A-S-C-O, and Causative Classification System
Direct Comparison in the North Dublin Population Stroke Study
Background and Purpose— Reliable etiologic classification of ischemic stroke may enhance clinical trial design and identification of subtype-specific environmental and genetic risk factors. Although new classification systems (Causative Classification System [CCS] and ASCO [A for atherosclerosis, S for small vessel disease, C for cardiac source, O for other cause]) have been developed to improve subtype assignment, few comparative data exist from large studies. We hypothesized that both CCS and ASCO would reduce the proportion of patients classified as cause undetermined compared with the Trial of ORG 10172 in Acute Stroke Treatment (TOAST) scheme in a large population-based stroke study.
Methods— A single rater classified all first-ever ischemic strokes in the North Dublin Population Stroke Study, a population-based study of 294 529 North Dublin residents. Published algorithms for TOAST, CCS, and ASCO were applied.
Results— In 381 first-ever ischemic stroke patients, CCS assigned fewer patients as cause undetermined (26.2% versus 39.4%; P<0.000001), with increased assignment of cardio-aortic embolism (relative increase 6.9%; P=0.004), large artery atherosclerosis (relative increase 44.1%; P=0.00006), small artery occlusion (relative increase 27.3%; P=0.00006), and other causes (relative increase 91.7%; P=0.001) compared with TOAST. When ASCO grade 1 evidence was applied, fewer patients were classified as small artery disease (relative decrease 29.1%; P=0.007) and more as large artery/atherothrombotic (relative increase 17.6%; P=0.03). ASCO grade 1 did not reduce the proportion of cause undetermined cases compared with TOAST (42.3% versus 39.4%; P=0.2). Agreement between systems ranged from good (κ=0.61 for TOAST/ASCO grade 1 small artery category) to excellent (κ=0.95 for TOAST/CCS and ASCO grade 1/CCS cardio/aorto-embolism category). Application of ASCO grades 1 to 3 indicated evidence of large artery/atherosclerosis (73.3%), cardio-embolism (31.3%), small artery (64.7%), and other cause (12%) in TOAST-undetermined cases.
Conclusion— Both CCS and ASCO schemes showed good-to-excellent agreement with TOAST, but each had specific characteristics compared with TOAST for subtype assignment and data retention. The feasibility of a single combined classification system should be considered.
Key outcomes such as disability, fatality, and recurrence after ischemic stroke differ according to subtype defined by stroke mechanism.1,2 Identification of the underlying cause of stroke is an important element of daily clinical practice, guiding treatment decisions and prognosis for individual patients. Accurate and reproducible assignment of the likely mechanism of ischemic stroke is also important in clinical trials investigating benefit in specific patient groups3 (eg, cardioembolic or atherosclerotic stroke) and in epidemiological studies investigating associations between environmental and genetic risk factors and specific stroke phenotypes.4,5
The Trial of ORG 10172 in Acute Stroke Treatment (TOAST) classification system was initially developed to improve standardization of subtype assignment in a multicenter randomized clinical trial.6 It has since been used in settings such as epidemiological and genetic association studies for which it was not originally intended.5 Since development of the TOAST system, advances in diagnostic technology have resulted in an increasing proportion of stroke patients for which multiple potentially contributory mechanisms are identified.7 Under TOAST, such patients are classified as cause undetermined, along with those who have been investigated incompletely. This subgroup is often excluded from additional analysis, leading to suboptimal use of available information.
Newer stroke classification schemes such as the Causative Classification System (CCS) and A-S-C-O (A for atherosclerosis, S for small vessel disease, C for cardiac source, O for other cause) Phenotypic System (ASCO) may improve our ability to identify the most likely cause where multiple potential mechanisms are found,8 accounting for the relative importance of each and the extent of the diagnostic evaluation.9 These schemes have considerable potential to refine subtype assignment and improve the efficiency of data use in clinical stroke studies. However, they have not been compared directly, and little information exists concerning their application in independent patient cohorts. Additional data are needed regarding the relative frequencies of subtype assignment under each scheme, differences in subtype assignment by CCS and ASCO compared with TOAST, and the characteristics of each in unselected population-based compared with hospital-investigated stroke cohorts.
We hypothesized that both CCS and ASCO would reduce the proportion of patients classified as cause undetermined and thus improve subtype assignment compared with TOAST when applied in a population-based stroke study. We aimed to address this hypothesis and to describe the frequencies of assigned subtypes in all (community plus hospital) and hospitalized ischemic stroke patients.
We analyzed patients in the North Dublin Population Stroke Study, a population-based prospective cohort study conducted over 1 year of stroke and transient ischemic attack (TIA) among the 294 529 inhabitants of North Dublin city, defined by the Irish local government district electoral boundaries.10
Multiple validated overlapping sources of hospital and community case ascertainment were used, according to recommended criteria for “ideal” studies of stroke incidence.11 Hot and cold pursuit methods were used.12 All 4 acute and the 9 nonacute hospitals in North Dublin city participated. Investigators undertook daily review of hospital admission records and consultation requests to relevant specialist services and daily visits to hospital wards and specialist units (intensive care, neurology, coronary care, cardiothoracic, and neurosurgery) to ascertain hospitalized and in-hospital events. All vascular and cerebral imaging data from participating institutions were reviewed twice weekly. Staff at specialist outpatient clinics (ophthalmology, geriatric, neurology, and vascular surgery) and physicians in nonacute hospitals were contacted regularly. More than 95% of North Dublin general practitioners and nursing homes participated directly and were contacted fortnightly to ascertain community-treated cases. A daily community-based minor stroke/TIA clinic was established to facilitate assessment of suspected cases from community health personnel, clinics, and emergency departments in the region. Review of pathology department records, death certificates, and coroner reports allowed identification of both community and hospital stroke fatalities.
All potential cases were reviewed for eligibility by medical file/imaging review by a study physician and verified by an experienced stroke physician. All patients with possible TIA, possible stroke for whom eligibility was unclear, or recurrent events were reviewed in person by an experienced stroke physician.
Each recruited case was assessed for the presence of specific vascular risk factors: hypertension, diabetes mellitus, hyperlipidemia, cigarette smoking, and peripheral and coronary arterial disease. Hypertension was recorded if patients were on antihypertensive therapy prestroke or had persistently elevated systolic blood pressure of >140 mm Hg or diastolic >90 mm Hg. Diabetes mellitus was recorded in patients with a prestroke diagnosis, 2 fasting venous plasma glucose measurements ≥7.0 mmol/L, or plasma glucose ≥11.1 mmol/L 2 hours after 75-g oral glucose load. Hyperlipidemia was defined as total cholesterol ≥5.0 mmol/L or low-density lipoprotein ≥3.5mmol/L or if patients were on dietary or medical therapy prestroke. Coding of atrial fibrillation and carotid stenosis required diagnosis by ECG or extracranial vascular imaging, either before stroke or within 6 months of the qualifying event.
Ethics committee approval was obtained from all participating institutions and the Irish College of General Practitioners. All participants (or authorized surrogates) provided informed consent.
Eligibility Criteria for Subtype Analysis
The following prespecified inclusion criteria were applied: (1) ischemic stroke identified by an appropriate clinical syndrome (according to World Health Organization definition) in whom brain imaging or pathological examination was available to exclude primary intracerebral hemorrhage; (2) resident of North Dublin city during the 1-year study ascertainment period (December 1, 2005, through November 30, 2006); and (3) first-ever ischemic stroke in a lifetime.
Exclusion criteria were: (1) qualifying event of TIA, hemorrhagic stroke, or recurrent stroke; and (2) unavailability of brain imaging or pathology reports.
The TOAST system classifies ischemic stroke into 5 categories: cardio-embolism, large artery atherosclerosis, small artery occlusion, other determined etiologies, and stroke of undetermined etiology. The undetermined category is a heterogeneous group comprising cases with ≥2 causes identified, no cause found despite appropriate investigation, and incomplete evaluation.
The CCS scheme also assigns patients into 5 mechanism-based categories: cardio-aortic, large artery atherosclerosis, small artery occlusion, other cause, and undetermined. However, unlike TOAST, the CCS system assigns the most likely subtype based on updated estimates of stroke risks associated with specific cardiac and vascular pathologies or clinical or imaging parameters known to be more commonly associated with particular stroke mechanisms.
The A-S-C-O (phenotypic) classification assigns a graded level of certainty (range 1 to 3) for the presence of each of 4 stroke mechanism categories: atherothrombosis (A), small vessel disease (S), cardio-embolism (C), and other causes (O). This combines etiologic information for individual patients in a single code, which can be grouped according to most likely mechanism (eg, high-risk [C1] cardioembolic source) or any occurrence of a shared phenotype (eg, all cases with evidence for atherosclerosis [A1+A2+A3]). ASCO also includes information on extent of diagnostic evaluation (eg, A0 denotes that no evidence of atherosclerosis was found despite appropriate investigation, whereas A9 signifies that appropriate investigations were not undertaken). Cases with ASCO grade 1 (ASCO1) evidence in a single domain were grouped for comparison with TOAST and CCS subtypes. Comparisons were repeated for ASCO1 and ASCO2 evidence. For these analyses, patients for whom a high risk source in a single domain was identified (ie, A1, S1, C1, or O1) were grouped together irrespective of whether a full diagnostic evaluation was completed or not. In clinical practice, not all patients undergo all etiologic investigations, especially when a known high-risk mechanism has already been identified (eg, severe ipsilateral carotid artery stenosis or metallic cardiac valve in the setting of subtherapeutic anticoagulation therapy). To assess possible effects of this method of grouping ASCO1 cases, we repeated the analysis in patients who underwent a diagnostic evaluation comprising all of brain imaging, ECG, ambulatory monitoring, echocardiography, extracranial vascular imaging, and other investigations as indicated.
A single trained stroke physician (M.M.) used published algorithms to perform data abstraction and subtype classification.6,8,9,13 Clinical, imaging, or pathological data relevant to subtype assignment were abstracted from the medical file at the time of recruitment (within 1 week of stroke onset) and again at 6 months after recruitment. Subtypes were assigned ≥6 months after a qualifying event to include the complete diagnostic evaluation in each case.14 For CCS, the web-based automated algorithm was used (http://ccs.martinos.org). In general, classification using the CCS computerized interface was completed more rapidly than ASCO or TOAST subtyping. In particular, automatic disabling and enabling of dependent elements in CCS focused the data abstraction process, and incorporation of pop-up definitions in the interface obviated the need for repeated reference to the original papers.
The primary analysis comprised comparisons of nonindependent proportions to examine the success of different methods (CCS versus TOAST, ASCO versus TOAST, and CCS versus ASCO) for categorization of stroke subtype. The McNemar test was used to compare the probabilities of discordant categorization using each approach with Bonferroni correction for multiple comparisons. Agreement between systems was measured using the κ statistic, with values interpreted as moderate (0.41 to 0.6), good (0.61 to 0.8), or very good (0.81 to 1) agreement.15 The 95% CIs for the relative difference in paired proportions were calculated as described by Newcombe.16 Independent proportions were compared using χ2 or Fisher test as appropriate. Multiple-group comparisons of independent means were compared using one-way ANOVA. Analyses were performed in STATA (Version 9.0).
A total of 568 patients with new stroke events were identified over the 1-year ascertainment period. Ischemic stroke was confirmed in 454 patients based on imaging (98.9%; 449 of 454) or pathological data (1.1%; 5 of 454). Of these, 83.9% (381 of 454) had first-ever ischemic stroke and were included for comparison of subtype classification systems. A total of 354 of 381 (92.9%) were admitted to hospital, whereas 27 of 381 (7.1%) were treated in the community. Demographic, risk factor, and investigation profiles of the cohort are presented in Table 1. Patients with cardio-embolism were older, more likely to have preexisting coronary disease, and least likely to smoke (P<0.01 for all; Supplemental Table I, available online at http://stroke.ahajournals.org). Those with large artery disease were more likely male and had highest smoking rates (P<0.01). Diabetes mellitus (but not hypertension) was more common in small artery disease by CCS (P=0.05) and ASCO (P=0.08).
Comparison of TOAST and CCS in Population-Based Cohort
A total of 39.4% (150 of 381) were classified as undetermined cause by TOAST (Table 2). This was because of identified competing mechanisms in 36.7% (55 of 150), no cause identified despite adequate evaluation in 35.3% (53 of 150), and incomplete evaluation in 28% (42 of 150). Compared with TOAST, the CCS scheme increased the assignment of patients to a defined subtype across all categories (P<0.01 for all comparisons; Bonferroni significance threshold 0.0125). This coincided with a 33.3% reduction of patients assigned to the cause undetermined category by CCS compared with TOAST (P<0.000001). Substantial relative increases in subtype assignment were observed across defined categories by CCS. This increase was most apparent for the other determined cause category (relative increase 91.7%; P=0.001), with increases in patients classified as large artery (relative increase 44.1%; P<0.0001), small artery (relative increase 27.3%; P<0.0001), and cardioembolic/cardioaortic (relative increase 6.9%; P=0.004) categories.
Agreement between TOAST and CCS ranged from good (for other determined and undetermined) to excellent (for cardio-embolism/cardio-aortic; Table 3).
Comparison of TOAST and ASCO in Population-Based Cohort
When cases with ASCO1 evidence for a defined mechanism were compared with TOAST, a reduction of patients assigned to the small artery group was observed (relative decrease 29.1%; P=0.007; Table 2). A small increase in assignment to large artery/atherothrombotic was observed (relative increase 17.6%; P=0.03). However, this did not meet statistical significance after Bonferroni correction. ASCO1 did not change the assignment of patients to a defined subtype in other categories.
When comparing patients with ASCO1 or ASCO2 evidence with TOAST, a reduction in patients designated as cause undetermined was observed (relative decrease 18.7%; P<0.0001). This coincided with an increase in those assigned to small artery (relative increase 27.3%; P<0.0001). Smaller increases were observed in large artery/atherothrombotic (relative increase 20.6%; P=0.02) and other causes categories (relative increase 50%; P=0.03), but these did not meet Bonferroni-adjusted significance.
Agreement between TOAST and ASCO1 is presented in Table 3. Similar levels of agreement were observed when TOAST was compared with ASCO1 and ASCO2, with the exception of the small artery category, in which agreement increased substantially (from κ=0.61 to κ=0.86) by application of ASCO2 criteria.
CCS and ASCO in Population-Based Cohort
Compared with ASCO1, CCS increased assignment of patients to small artery (relative increase 79.5%; P<0.000001) and cardio-embolism/cardio-aortic categories (relative increase 11.2%; P=0.004), with a trend toward increased large artery assignment (relative increase 22.5%; P=0.06), and a corresponding decrease in assignment to the undetermined category (relative decrease 37.9%; P=0.0005; Table 2). Agreement between CCS and ASCO1 ranged from good to excellent (Table 3).
Additional Information Provided by ASCO
We examined the additional data provided by application of ASCO in the entire cohort (n=381) and the subgroup designated as undetermined cause by TOAST (n=150).
Ninety-seven (out of a possible 625) unique codes resulted from application of the ASCO algorithm in the population cohort. These codes were grouped according to the presence or absence of disease to show the frequency of ASCO phenotypes (Table 4).
Cardiac investigations (echocardiography and ECG or ambulatory monitoring) were performed in 104 of 150 (69.3%) and vascular imaging (extracranial or intracranial) in 122 of 150 (81.3%) of TOAST-undetermined cases. A high-risk cardioembolic source (C1) was identified in 15.3% (23 of 150), whereas 31.3% (47 of 150) had evidence of any potential (C1/C2/C3) cardioembolic source. An ASCO-defined high-risk ipsilateral atherosclerotic source (A1) was identified in 16.7% (25 of 122), whereas 73.3% (110 of 122) had evidence of any large artery or aortic atherosclerosis (A1/A2/A3). For small artery disease, a high level (S1) of evidence was found in 15.3% (23 of 150) and some evidence (S1/S2/S3) in 64.7% (97 of 150). For the other causes category, a high level (O1) of evidence was found in 8% (12 of 150) and some evidence (O1/O2/O3) in 12% (18 of 150).
Comparison of Classification Schemes in Hospital-Admitted Subgroup
We repeated the comparisons of CCS and ASCO with TOAST in the hospital-admitted subgroup (n=354) because patients identified and treated in the community may have fewer investigations (Table 1).
The proportion of patients assigned by each classification scheme increased in the hospital subgroup compared with the population-based cohort for cardioembolic/cardioaortic, large artery, and other determined subtypes (Table 5). Slight reductions were observed in patients classified as small artery and undetermined in the hospital-admitted compared with the population cohort by TOAST, CCS, and ASCO systems. The absolute differences between specific subtype assignments in hospital and population cohorts were minimal (absolute difference range −1.6 to +0.8%).
We repeated our analysis in patients who underwent at least the combination of brain imaging, ECG or ambulatory monitoring, echocardiography, and extracranial vascular imaging (n=259). Again, CCS (but not ASCO1) reduced assignment to cause undetermined, with increased assignment to large and small artery groups (P<0.001 for all) and trends observed for other causes (P=0.06) and cardioembolic (P=0.125) groups compared with TOAST. Agreement between systems remained unchanged.
Assessment of Inter-Rater Reliability
Although not a primary aim of our study, we performed a secondary analysis to evaluate inter-rater reliability for CCS and ASCO1 assignment. Subtype assignments were retrospectively compared between a trained stroke physician (M.M.) and a nonstroke specialist (C.D.) based on original case record review in subgroups of ischemic stroke patients. Although the stroke physician was not blinded to assignments of the nonspecialist physician, all cases were classified based on the rater’s application of CCS and ASCO algorithms to abstracted data.
Thirty-eight CCS and 100 ASCO cases were classified by each rater. CCS inter-rater reliability was good to excellent for cardioaortic (κ=0.87), large artery atherosclerosis (κ=1.0), and other causes (κ=0.65), but only moderate for small artery occlusion (κ=0.48) and undetermined cause (κ=0.53). Similarly, ASCO1 inter-rater reliability was good to excellent for cardioembolic (κ=0.88), atherothrombosis (κ=0.79), and other causes (κ=0.66), but only moderate for small vessel disease (κ=0.48) and undetermined cause (κ=0.59).
Our study describes the practical application of the CCS and ASCO subtyping systems compared with TOAST in a large population-ascertained cohort of ischemic stroke patients. Previous reports8,9,17 have described the derivation of both CCS and ASCO, but to our knowledge, neither has yet been applied in a large stroke study. Our findings support the validity of both classification systems but identify some points that may require additional refinement before there is wider use in epidemiological studies and clinical trials.
We found high levels of agreement between CCS and TOAST for classification of all stroke subtypes. Agreement was highest for cardioembolic/cardioaortic, small artery, and large artery atherosclerosis subtypes but remained good for other determined and undetermined cause categories. This is not unexpected because the CCS may be considered a refinement of TOAST based on updated evidence of the association of specific risk factors with ischemic stroke. However, the CCS substantially improved classification across all defined subtype groups compared with TOAST. No patient assigned to a TOAST category was assigned to a different category by CCS (Figure), indicating that disagreements between the systems related to greater assignment to each subtype by CCS. Marked improvements in classification were observed for other causes, large artery, and small artery subtypes (Table 2) and are further reflected by a one-third reduction in patients assigned to the undetermined group compared with TOAST. A similar pattern was observed when CCS and ASCO were compared. The CCS criteria that most frequently led to changes in TOAST and ASCO subtype assignments included: (1) presence of systemic embolism or complex aortic atheroma (cardioaortic); (2) TIA or stroke in the same vascular territory within 1 month before the qualifying event or presence of ipsilateral internal watershed infarctions or multiple, temporally separate, infarctions exclusively within the territory of the affected artery (large artery atherosclerosis); (3) presence of stereotypic lacunar TIAs within 1 week or classic lacunar syndrome (small artery occlusion); and (4) specific disease process occurring in close temporal or spatial relationship to the onset of brain infarction, such as arterial dissection or cardiovascular interventions (other causes).
High levels of agreement were observed when ASCO1 (corresponding to high likelihood of causal association) classification was compared with TOAST, particularly for cardioembolic, large artery, and other causes categories. However, application of ASCO1 criteria did not change the proportion of patients assigned to these subtypes or to the cause undetermined group. Agreement with TOAST was lowest (κ=0.61) for classification of small vessel disease by ASCO1 criteria, associated with a substantial (29%) reduction in patients assigned to the small vessel disease category by ASCO. This probably relates to the more stringent requirements of ASCO1 for small artery subtype classification. ASCO1 criteria require demonstration of a small, deep infarct on imaging corresponding to symptoms and the presence of ≥1 additional feature, such as lacunar infarction in another territory, recent lacunar-type TIAs in the same territory, or imaging evidence of leucoariosis, microbleeds, or dilatation of perivascular spaces. In comparison, the TOAST small artery category requires a traditional clinical lacunar syndrome and either normal CT/MRI brain or a small brain stem or subcortical infarct in the relevant vascular territory. Similar to TOAST, ASCO2 criteria permit assignment of small artery subtype if there is imaging evidence of a single deep branch artery stroke or an appropriate clinical syndrome. When we applied ASCO1 and ASCO2 criteria, agreement with TOAST increased substantially, and patients assigned to the small artery category increased by 27%. In addition to the less restrictive criteria outlined above, this increase is also partially attributable to the inclusion by ASCO1 and ASCO2 of other clinical syndromes suggestive of lacunar stroke, such as isolated dysarthria, hemichorea and hemiballism.9
Although the ASCO system did not greatly increase the assignment to specific subtype categories, it provided additional information not available under the TOAST and CCS schemes. For example, although only 10.5% of patients met ASCO1 criteria for large artery disease, an additional 68.5% (261 patients) had some evidence of atherosclerosis by ASCO criteria (grades 1 to 3). Similar large increases were observed in cardioembolic and small artery categories. ASCO also provided insights into patients coded as undetermined by TOAST, a group frequently excluded from analysis in clinical studies. For example, 15% to 17% of TOAST-undetermined patients had ASCO1 evidence of a cardioembolic, large, or small artery subtype. An additional 31% to 73% had some evidence (ASCO1 to ASCO3) of these subtypes. It is possible that these characteristics of the ASCO system may improve current methods for refinement of phenotype in stroke genetic studies or for patient selection for clinical trials of therapeutic agents targeting specific disease mechanisms. For example, it is possible that statin therapy may be of greater benefit to stroke and TIA patients with ASCO1 to ASCO3 evidence of atherosclerosis compared with other ASCO phenotypes. Phenotypic classification systems such as ASCO may also facilitate studying genetic/environmental risk markers in such groups. However, we emphasize that these hypotheses relating to the utility of ASCO3 information in research and clinical practice settings remain to be proven.
Inter-rater reliability for both CCS and ASCO1 was excellent or good for cardioaortic/cardioembolic, large artery atherosclerosis/atherothrombosis, and other determined cause categories, reflecting clarity in the definitions of causation used in both systems. For small artery disease, inter-rater reliability was only moderate, with resultant reduction in inter-rater reliability for the undetermined cause category. This finding could be explained by ambiguities in original case files or the effect of differing degrees of experience of the raters in adjudicating on lacunar stroke syndromes. We caution that these results require confirmation because this analysis was retrospective, and the sample size (particularly in the CCS analysis) was small. Kappa values are influenced by the number of observations which should be borne in mind when interpreting reliability analyses with low numbers.
Strengths of our study include the large sample size and population-based design, which adheres closely to recommended criteria for rigorous stroke epidemiological studies.11 A limitation is that not all investigations were performed in some patients. This is perhaps inevitable in observational studies of real-world practice and is reflected by rates of investigation similar to our study in other population and hospital studies.2,18 In some cases, investigations for stroke mechanism may cease when a high-risk source is found (eg, echocardiography may not be performed if ipsilateral high-grade carotid stenosis is detected). In others, investigating stroke mechanism may be considered of secondary importance (eg, in severe stroke when death is imminent). To allow comparison of ASCO with TOAST and CCS subtypes, we grouped cases according to a reduction rule (ie, all patients with ASCO evidence in a single domain were classified in that domain, including some patients who may not have had diagnostic investigations for other potential etiologies). This approach reflects clinical decisions often made in daily practice (eg, a patient with ischemic stroke and subtherapeutic anticoagulation for a metallic cardiac valve prosthesis may not have cerebrovascular imaging). However, it is possible that fewer patients may be designated as cause undetermined by this method. We addressed this issue by repeating our analysis on hospital-treated patients and on those for whom vascular imaging and cardiac studies were completed. Our findings remained unchanged, suggesting that the inclusion of patients without a full diagnostic evaluation was not a major contributor, and that they may be generalized to a broad range of study settings.
Our study clarifies several characteristics of both CCS and ASCO systems. Both scales displayed high agreement with TOAST, with lowest agreement identified for ASCO1 classification of small artery disease. The CCS improved assignment to specific stroke categories and reduced the proportion of patients classified as undetermined cause, suggesting an advantage when a high degree of specificity is required for subtype assignment. Other advantages of the CCS system include its derivation based on best-available evidence for association of stroke with identified risk markers and relative ease of application of the CCS algorithm via a computerized interface. The ASCO system retained information with potential relevance to stroke etiology not captured by either TOAST or CCS schemes, which may have utility in some study settings. By setting more restrictive criteria, ASCO may also increase the specificity of classification of small artery disease but with a reduction in sensitivity when grade 1 criteria are applied.
Given the specific strengths of both systems and the desirability of a single agreed subtyping scheme for clinical stroke research, a harmonized classification scheme that integrates the strengths of both systems should be considered. This might incorporate the evidence-based algorithm of CCS to improve the overall specificity of subtype assignment, with additional levels of evidence (similar to ASCO2, ASCO3, and ASCO9 categories) to maximize retention of information where relevant in specific study settings. Consensus would be required on other points, such as categorization of aortic arch atherosclerosis.19 The feasibility of such an integrated system should be examined in future research.
We sincerely thank the patients, families, general practitioners, and hospital colleagues who participated.
Sources of Funding
M.M. is supported by awards from the Mater College of Postgraduate Education and Research, the Irish Heart Foundation, and an unrestricted educational grant from Pfizer Healthcare. O.C.S. is the recipient of Mater Foundation and Health Research Board of Ireland awards. P.J.K. is the recipient of awards from the Irish Heart Foundation and a Clinician Scientist Award from the Health Research Board of Ireland. L.K. is supported by a Health Research Board of Ireland Clinician Scientist Award. Funding was also received from the Health Services Executive and National Lottery of Ireland.
- Received December 8, 2009.
- Revision received April 12, 2010.
- Accepted April 20, 2010.
Lovett JK, Coull AJ, Rothwell PM. Early risk of recurrence by subtype of ischemic stroke in population-based incidence studies. Neurology. 2004; 62: 569–573.
Kolominsky-Rabas PL, Weber M, Gefeller O, Neundoerfer B, Heuschmann PU. Epidemiology of ischemic stroke subtypes according to TOAST criteria: incidence, recurrence, and long-term survival in ischemic stroke subtypes: a population-based study. Stroke. 2001; 32: 2735–2740.
Jackson C, Sudlow C. Are lacunar strokes really different? A systematic review of differences in risk factor profiles between lacunar and nonlacunar infarcts. Stroke. 2005; 36: 891–901.
Adams HP Jr, Bendixen BH, Kappelle LJ, Biller J, Love BB, Gordon DL, Marsh EE III. Classification of subtype of acute ischemic stroke. Definitions for use in a multicenter clinical trial TOAST Trial of Org 10172 in Acute Stroke Treatment. Stroke. 1993; 24: 35–41.
Lee LJ, Kidwell CS, Alger J, Starkman S, Saver JL. Impact on stroke subtype diagnosis of early diffusion-weighted magnetic resonance imaging and magnetic resonance angiography. Stroke. 2000; 31: 1081–1089.
Ay H, Benner T, Arsava EM, Furie KL, Singhal AB, Jensen MB, Ayata C, Towfighi A, Smith EE, Chong JY, Koroshetz WJ, Sorensen AG. A computerized algorithm for etiologic classification of ischaemic stroke: the Causative Classification of Stroke System. Stroke. 2007; 38: 2979–2984.
Central Statistics Office Ireland, Census 2006, http://www.cso.ie/census/.
Feigin VL, Carter KV. Standards for an ideal stroke incidence study. Stroke. 2004; 35: 2045–2047.
Rothwell PM, Coull AJ, Giles MF, Howard SC, Silver LE, Bull LM, Gutnikov SA, Edwards P, Mant D, Sackley CM, Farmer A, Sandercock PA, Dennis MS, Warlow CP, Bamford JM, Anslow P; Oxford Vascular Study. Change in stroke incidence, mortality, case-fatality, severity, and risk factors in Oxfordshire, UK from 1981 to 2004 (Oxford Vascular Study). Lancet. 2004; 363: 1925–1933.
Goldstein LB, Jones MR, Matchar DB, Edwards LJ, Hoff J, Chilukuri V, Armstrong SB, Horner RD. Improving the reliability of stroke subgroup classification using the Trial of ORG 10172 in Acute Stroke Treatment (TOAST) criteria. Stroke. 2001; 32: 1091–1098.
Madden KP, Karanjia PN, Adams HP Jr, Clarke WR. Accuracy of initial stroke subtype diagnosis in the TOAST study. Trial of ORG 10172 in acute stroke treatment. Neurology. 1995; 45: 1975–1979.
Grau AJ, Weimar C, Buggle F, Heinrich A, Goertler M, Neumaier S, Glahn J, Brandt T, Hacke W, Diener HC. Risk factors, outcome, and treatment in subtypes of ischaemic stroke: the German stroke data bank. Stroke. 2001; 32: 2559–2566.