Oxfordshire Community Stroke Project Classification Poorly Differentiates Small Cortical and Subcortical Infarcts
Background and Purpose—The Oxfordshire Community Stroke Project (OCSP) is a common clinical stroke classification tool. We evaluated the accuracy of OCSP classification with a prospective magnetic resonance imaging (MRI) study.
Methods—Stroke/transient ischemic attack patients presenting within 48 hours of onset were included in the study (n=130). Following computed tomography scan, OCSP classification, total anterior circulation infarcts (TACI), partial anterior circulation infarcts (PACI), lacunar circulation infarcts (LACI), and posterior circulation infarcts (POCI) were performed by 3 independent examiners. All patients underwent diffusion-weighted MRI with planimetric volume measurement and classification into OCSP categories, organized by lesion location.
Results—Patients were clinically classified as TACI (12 patients), PACI (62 patients), LACI (38 patients), and POCI (18 patients). In 101 patients with diffusion-weighted MRI lesions, correct classification rates were: TACI (83.3%), PACI (83%), LACI (39%), and POCI (86%). OCSP had the following sensitivity (SE), specificity (SP), and positive predictive value (PPV): TACI (SE, 100%; SP, 98%; PPV, 83%), PACI (SE, 73%; SP, 78%; PPV, 83%), LACI (SE, 47%; SP, 83%; PPV, 39%), and POCI (SE, 92%; SP, 98%; PPV, 86%). Sixty-one percent of patients in the LACI group had radiographic appearances consistent with PACI, and 15% of those classified as PACI had lacunar infarcts. No differences in stroke severity existed between patients classified correctly (median National Institutes of Health Stroke Scale [NIHSS]=4; interquartile range [IQR]=7) or incorrectly (median NIHSS=3; IQR=3). Patients classified correctly had larger infarct volume (median=6.75 mL; IQR=33.2) than did those who were incorrectly classified (1.86 mL; IQR=5; P=0.008).
Conclusions—OCSP classification does not permit accurate discrimination between lacunar and small-volume cortical infarcts. Differential patterns of investigation for stroke etiology should not be based solely on clinical criteria.
Ischemic stroke encompasses a heterogeneous group of syndromes with different etiologies and prognoses. Patients with stroke and transient ischemic attack (TIA) often suffer early recurrence, and there is evidence that the risk of new ischemic events differs among clinical syndromes.1,2 Etiologic investigations and effective secondary stroke prevention are thus influenced by the correct categorization of stroke syndromes. A commonly used stroke classification system is the Oxfordshire Community Stroke Project (OCSP).3 This clinical tool categorizes stroke syndromes into 4 subtypes: total anterior circulation infarcts (TACI), partial anterior circulation infarcts (PACI), lacunar infarcts (LACI), and posterior circulation infarcts (POCI). The likelihood of individual stroke risk factors can be predicted by OCSP category.4 Furthermore, OCSP classification has value in predicting both the pattern of underlying vascular occlusion and clinical outcome.5–7
The accuracy of OCSP classification has previously been evaluated in neuroimaging studies that relied on computed tomography (CT) scans or magnetic resonance imaging (MRI) without diffusion-weighted sequences (DWI).8–12 Given the superior sensitivity of DWI for acute ischemia,13 we prospectively assessed the reliability of OCSP classification together with early CT scan findings in predicting the pattern of DWI lesions.
Between April 2008 and December 2009, we prospectively recruited patients with stroke or TIA within 48 hours of symptom onset who presented to the emergency department at our hospital (n=130). Informed consent was obtained in all cases. Patients were excluded if they were clinically unstable, had contraindications to MRI, or if CT revealed intracerebral hemorrhage/stroke mimics.
A CT scan was available in 128 of 130 patients before enrollment and was used to help refine the clinical classification. Patients were examined by 2 neurologists who independently classified the patient into 1 of 4 OCSP categories (Supplemental Table, http://stroke.ahajournals.org).3 Examiners took a history, completed a physical examination, and assessed CT scans. In cases of interrater disagreement, a third examiner independently classified the patient. All patients, including those ultimately diagnosed with TIA, were symptomatic at the time of enrollment. In thrombolyzed patients, clinical OCSP classification was performed before treatment.
Patients were imaged with DWI, using an 8-channel phased array radiofrequency head coil (MRI Devices) on a 1.5-Tesla whole-body Siemens Sonata MRI scanner (Siemens Medical Systems). DWI was acquired with single-shot, spin-echo diffusion echo-planar imaging, 220 mm field of view, 19 5-mm axial slices with 1.5 mm gap, b-value of 1000 s/mm2 along 3 orthogonal directions, repetition time/echo time 2600/86 msec, Generalized Autocalibrating Partially Parallel Acquisition (GRAPPA) R=2, matrix size of 128×128 zero-filled to 256×256.
MRI Classification and Analysis
MRI scans were anonymized and evaluated in blinded fashion without reference to CT or OCSP classification. Regional DWI analysis was performed using the Analyze software package (Biomedical Imaging Resource).14 DWI hyperintense lesion borders were defined using a semiautomated threshold intensity technique. Ischemic lesion volumes were measured with planimetric techniques. All measurements were performed by the same investigator (N.A.). Based on the acute DWI lesion topography, a “radiographic OCSP” classification was devised (Supplemental Table). The final radiographic classification was made by consensus agreement between 3 investigators (N.A., D.E., K.B.).
Statistical analysis was performed using SPSS 17.0 (SPSS Inc.). Using radiographic classification as the gold standard, sensitivity (SE), specificity (SP), positive (PPV), and negative predictive values (NPV) of each clinical OCSP category in patients with DWI lesions were calculated. Correct classification rates (CCR) were defined as the percentage of patients in each radiographic OCSP category (gold standard) that were localized correctly by clinical OCSP category. Differences in the frequency of DWI lesions in each category were tested using Pearsons χ2 test. Differences between non-normally distributed DWI lesion volumes were evaluated with the rank sum test. An interrater reliability analysis using the κ statistic was performed. We calculated 95% CI for the CCR if block size was sufficient (ie, n×p or n×(1−p) were both >5 for any particular proportion).
The clinical diagnosis was stroke in 105 patients, and was TIA, based on resolution of symptoms within 24 hours, in the remaining 25 patients. The clinical OCSP classification for all patients was as follows: TACI (12 patients), PACI (62 patients), LACI (38 patients), and POCI (18 patients); baseline characteristics are displayed in Table 1. Interrater reliability for the first 2 examiners was excellent (κ=0.93, P<0.001). Early infarct changes were seen on 31% of CT scans (40/128). Patients in the PACI group were not more likely to have early infarct changes on CT (16/60 scans; 27%) relative to the LACI group (7/38 scans; 18%); P=0.348. A total of 7/130 patients (5.38%) received thrombolytic therapy. Patients in the TACI subgroup had more severe neurological deficits, as defined by National Institutes of Health Stroke Scale (NIHSS) score, and received intravenous thrombolysis more frequently than did those in the other 3 categories.
MRI Lesion Frequency by Clinical Classification
Acute DWI lesions were present in 101/130 patients (78%). Patients diagnosed clinically with stroke were more likely to have DWI lesions (90/105 patients) than were those with TIA (11/25 patients; λ2=20.27; P<0.001). Patients with DWI lesions had higher NIHSS scores (median=3; interquartile range [IQR]=6) than did those without DWI lesions (median=1; IQR=3; P=0.0001). All patients with clinical TACI syndromes had DWI lesions versus 84% with PACI, 60% with LACI, and 78% with POCI syndromes. Patients clinically classified as PACI were more likely to have acute DWI lesions (52/62 patients) than were LACI patients (23/38 patients; λ2=4.53; P=0.03). The frequency of TIA in the LACI and TACI groups was similar (λ2=3.27; P=0.07). The frequency of DWI lesions did not differ between patients with PACI (52/62 patients) and POCI syndromes (14/18 patients; λ2=0.359; P=0.549).
SE and SP of Clinical OCSP Classification
Clinical TACI classification corresponded to the MRI classification with 100% SE and 98% SP (Table 2). All patients who had MRI lesions compatible with TACI were correctly categorized clinically; however, 2 patients classified clinically as TACI had a MRI classification consistent with PACI. Clinical TACI syndrome classification had a 100% NPV. The TACI group had the highest rate of thrombolysis (3/6 patients; 50%) as compared to the other subgroups received thrombolysis treatment. One patient classified clinically as TACI and thrombolyzed was found to have a PACI type infarct on DWI.
Clinical PACI classification was associated with moderate SE (73%) and SP (78%), and a PPV of 83%. Ten patients clinically classified as PACI did not have acute MRI lesions. The majority of misclassified patients in this group had lacunar infarcts on MRI (Table 3). One patient clinically misclassified in the PACI group had received intravenous thrombolysis treatment.
Clinical classification of LACI syndromes was insensitive (47%); of 23 DWI-positive patients classified clinically as LACI, only 9 had corresponding lacunar lesions. The remaining 14 patients had acute cortical infarcts consistent with PACI. Although clinical LACI classification was reasonably specific (83%), PPV was only 39%. Seventy percent of patients diagnosed with ataxia-hemiparesis lacunar syndrome had cortical MRI lesions (Figure).
Clinical POCI classification had 92% SE, 98% SP, and a PPV of 86%. Of 18 patients classified as POCI, 14 had a DWI lesion. Lesion topography was compatible with POCI in 12 patients. Two patients had MRI lesions consistent with LACI.
Correct Classification Rates and Positive Likelihood Ratios of Clinical OCSP
CCRs varied between subgroups. Overall, clinical OCSP classification correctly predicted infarct site in 74/101 patients with DWI lesions (CCR=73%; 95% CI, 65%–82%). Patients with LACI syndromes had the lowest CCR (39%; 95% CI, 19%–59%) and patients with POCI syndromes had the highest CCR (86%). Patients with TACI syndromes (CCR 83%) and PACI syndromes (CCR 83%; 95% CI, 72%–93%) were intermediate.
The positive likelihood ratio (odds that a clinically classified OCSP patient had a corresponding radiographic lesion) was only 2.8 for lacunar and 3.2 for PACI syndromes. The positive likelihood ratio was much higher for TACI (50.0) and POCI (46.0) syndromes.
Effect of Infarct Volume, Stroke Severity, and Laterality on Clinical Classification Accuracy
Patients who were misclassified by clinical OCSP classification had smaller infarct volumes (median=1.86 mL; IQR=5.0 mL) than did those who were classified correctly (median=6.75; IQR=33.3 mL; P=0.008; Figure B). The relationship between infarct volume and correct classification was independent of clinical severity. There were no differences in stroke severity between patients who were classified correctly (median NIHSS=4; IQR=7) and incorrectly (median NIHSS=3; IQR=3; P=0.117; Figure C). There were no significant differences between infarct volumes of clinically versus radiographically defined OCSP subcategories.
There was a trend toward higher median NIHSS scores in patients with dominant hemispheric infarction (4; IQR=6) compared with those with nondominant hemispheric lesions (2.5; IQR=7; P=0.08). Patients with dominant hemispheric infarcts were more likely to be correctly classified (83%) than were those with nondominant lesions (55%; λ2=8.66; P=0.003). Dominant hemispheric lacunar syndromes were more likely to be correctly classified (66.6%) relative to nondominant lacunes (9.1%; λ2=7.98; p-0.005). Correct classification rates were not affected by laterality in the other 3 categories.
This is the first prospective study evaluating the accuracy of OCSP clinical classification in conjunction with CT, relative to acute DWI. Although OCSP syndrome classification is generally reliable in patients with larger strokes, clinical and CT-based classification alone is associated with significant diagnostic inaccuracy in patients with smaller infarcts. Our results indicate that the probability of correctly distinguishing small-cortical from lacunar strokes is no better than chance.
Several studies have compared clinical OCSP classification with CT or MRI (Table 4).8–11,15,16 Visible ischemic changes were absent in up to 36% of patients in these studies. To estimate the sensitivity of OCSP classification, these patients were either excluded from the final analyses or “potential imaging abnormalities” were assumed to be present.9–11 This resulted in uncertainty in determination of sensitivity and specificity of clinical OCSP classification. In our study, the majority of patients in the most commonly misclassified OCSP categories (LACI and PACI) did not have early infarct changes on CT (Table 1), but most did have DWI lesions. Exclusion of these patients on the basis of a negative CT, as has been done in previous studies, would have resulted in falsely elevated estimates of clinical OCSP accuracy. In our study, clinicians had access to the CT scan, which likely increased the accuracy of clinical classification of TACI and larger-PACI patients, but did little to improve correct classification rates in those without early CT changes.
It has been previously demonstrated that OCSP classification has a particularly low PPV in predicting infarct location in subcortical strokes,8 and is least accurate in distinguishing between LACI and PACI syndromes.11 Our results confirm this, but also demonstrate that this inability to predict infarct distribution applies to all small-volume strokes, including those that are cortically based.
Most previous MRI investigations of OCSP utilized conventional MRI sequences, which are no more sensitive than are CT scans for hyperacute ischemic changes.17 To avoid diagnostic ambiguity, we used DWI, the modality most sensitive to ischemia.13 One other clinical/DWI comparative study has been performed, but radiographic classification was limited to lacunar versus nonlacunar infarct patterns.16 In this study, 40.5% of lacunar syndromes were associated with nonlacunar DWI lesions, consistent with our results. Despite exclusive use of DWI in our study, 14% of patients (15/105) with stroke did not have ischemic changes and remained unclassifiable by radiographic classification. A similar frequency of DWI negative strokes has previously been reported, and as in our study, these patients had subtle neurological deficits with lower NIHSS scores.18
The overall accuracy of clinical OCSP is partially dependent on the definition of the compatible imaging pattern. Unlike previous studies,11,15 we considered DWI infarct patterns to be compatible with only 1 clinical OCSP group, which may reduce the CCR. Despite this, the overall CCR of OCSP in our study (73%) was comparable to those in previous studies (Table 4).
Distinguishing between lacunar stroke and a small cortical infarct is a source of clinical uncertainty with management implications. Lacunar infarcts lying close to the cortex are more likely to produce cortical symptoms than subcortical lacunes.19 The pathophysiology, management, and prognosis of lacunar and cortical infarcts are different. Furthermore, stroke misclassification may result in inappropriate inclusion/exclusion of patients in clinical trials. Cortical infarcts misclassified as lacunes are more likely to have embolic risk factors, such as ipsilateral carotid stenosis and atrial fibrillation, relative to imaging-confirmed lacunar syndromes; they are also at higher risk of early ischemic recurrence.20 In our study, 61% of DWI-positive patients (14/23) originally classified as lacunes were found to have ischemic lesions consistent with cortical infarcts. In addition, 15% (8/52) of clinically defined PACI syndromes with MRI lesions had a lacunar DWI infarct pattern. Stroke misclassification may alter selection and timing of investigations (Figure), as physicians often delay or avoid investigation for a cardiac source of embolism in patients with lacunar syndromes. This does not imply that all patients with a normal CT scan require an MRI. Indeed, one could argue a more pragmatic approach is to investigate every stroke patient for proximal embolic sources, irrespective of clinical syndrome. Ultimately, our findings indicate that it is prudent for practicing clinicians to be less didactic in their approach to patients with negative CT scans.
Patients with TIA were excluded from previous comparative OCSP-imaging studies.8,9,11,15,16 In this study, although all patients were symptomatic at the time of presentation, 25 patients with minor neurological deficits had complete resolution of symptoms before 24 hours and were therefore classified as TIA. Despite this, 44% of these patients had DWI-positive lesions. As this was a prospective study, we did not exclude patients if their symptoms subsequently resolved. The disconnection between symptom resolution and imaging evidence of infarction further emphasizes the inaccuracy of clinical examination in isolation. For this reason, we favor a tissue diagnosis of stroke,21 and based our comparative analysis on the presence of DWI lesions, rather than clinical stroke/TIA diagnosis.
A limitation of this study is that this is not a consecutive sample of patients, as those in whom MRI could not be performed within 48 hours of onset were excluded. Also, the distribution of patients among the 4 OCSP categories differed from that in the original study by Bamford et al.3 The lower frequency of TACI patients was related to medical instability, making MRI difficult to perform in this group. Despite this, the present study is the largest prospective DWI analysis of clinical OCSP accuracy in combination with CT scan. Finally, a small percentage of patients (5.38%) in this study received thrombolytic therapy, which may have affected their final radiographic classification.
Reliance on clinical stroke syndrome classification may be misleading in a significant number of patients, particularly those with less severe neurological deficits and no CT changes. While routine MRI may not be feasible or practical in all centers, these findings do have implications for clinical research studies, in which stroke subtype and pathophysiology are relevant to inclusion/exclusion criteria and interpretation of results.
Sources of Funding
N.A. is supported by a Fellowship from the Canadian Institutes for Health Research (CIHR). K.B. is supported by salary and grant-in-aid awards from the Heart and Stroke Foundations of Canada, Alberta, Northwest Territories and Nunavut, the Alberta Heritage Foundation for Medical Research (AHFMR), and the Canadian Institutes of Health Research. J.A.M. is a University of Alberta clinical scholar.
The online-only Data Supplement is available at http://stroke.ahajournals.org/cgi/content/full/STROKEAHA.110.613752/DC1.
- Received January 13, 2011.
- Accepted February 24, 2011.
- © 2011 American Heart Association, Inc.
- Sacco RL,
- Foulkes MA,
- Mohr JP,
- Wolf PA,
- Hier DB,
- Price TR
- Mead GE,
- Wardlaw JM,
- Dennis MS,
- Lewis SC,
- Warlow CP
- Bamford J,
- Sandercock P,
- Jones L,
- Warlow C
- Mead GE,
- Lewis SC,
- Wardlaw JM,
- Dennis MS,
- Warlow CP
- Kobayashi A,
- Wardlaw JM,
- Lindley RI,
- Lewis SC,
- Sandercock PA,
- Czlonkowska A
- Lindgren A,
- Norrving B,
- Rudling O,
- Johansson BB
- Butcher K,
- Parsons M,
- Allport L,
- Lee SB,
- Barber PA,
- Tress B,
- et al
- Anderson CS,
- Taylor BV,
- Hankey GJ,
- Stewart-Wynne EG,
- Jamrozik KD
- Sylaja PN,
- Coutts SB,
- Krol A,
- Hill MD,
- Demchuk AM
- Mead GE,
- Lewis SC,
- Wardlaw JM,
- Dennis MS,
- Warlow CP
- Kidwell CS,
- Alger JR,
- Di Salle F,
- Starkman S,
- Villablanca P,
- Bentson J,
- et al