Background
Thrombolytic therapy with intravenous tissue plasminogen activator (tPA) for acute ischemic stroke increases the risk of symptomatic intracerebral hemorrhage (SICH) [
1]. Factors associated with SICH include older age, higher baseline National Institutes of Health Stroke Scale (NIHSS) score, elevated blood glucose, prior antiplatelet use, presence of atrial fibrillation, congestive heart failure, renal impairment, and early ischemic changes on pretreatment brain imaging [
2]. Several risk score models incorporating these potential predictors have been constructed to determine the risk of tPA-associated SICH [
3‐
9]. Although the discriminatory abilities of such models appear good, there may be room for further improvement.
Stroke territory may help estimate the risk of post-thrombolysis hemorrhagic transformation because posterior circulation stroke might be associated with a low risk of SICH [
10,
11]. The concept was first implied in the Hemorrhage After Thrombolysis (HAT) score [
4], in which the presence of hypodensity in middle cerebral artery territory on computed tomography (CT) denotes an anterior circulation stroke. No risk prediction models have yet explicitly incorporated stroke territory. However, using neuroimaging study to determine stroke territory may not be applicable or timely in emergency settings. Even magnetic resonance imaging (MRI) detected acute lesions in only 46% of patients with acute ischemic stroke examined within 3 hours after symptom onset [
12].
The Oxfordshire Community Stroke Project (OCSP) classification, based on clinical syndromes alone, can predict the site and size of infarct on CT in patients with established or hyperacute ischemic stroke [
13,
14]. In a previous study [
15], we demonstrated that the OCSP classification could help evaluate the risk of post-thrombolysis SICH. We therefore examined whether and how the addition of the OCSP classification could improve current SICH risk scores.
Results
A total of 548 thrombolyzed patients were included in the study. Table
2 shows the characteristics. The rates of SICH were 7.3% per the NINDS definition, 5.3% per ECASS II, and 3.5% per SITS-MOST. The agreement of the OCSP classification was moderate (κ = 0.583) between the two initial assessors (Additional file
1: Table S2).
Table 2
Characteristics of the study patients
Demographics | |
Age, mean (SD) | 67 (12) |
Male, n (%) | 345 (63.0) |
Medical history, n (%) | |
Hypertension | 406 (74.1) |
Diabetes mellitus | 177 (32.3) |
Hyperlipidaemia | 303 (55.3) |
Atrial fibrillation | 169 (30.8) |
Congestive heart failure | 37 (6.8) |
Prior stroke/TIA | 116 (21.2) |
Current smoking | 183 (33.4) |
Antiplatelets | 129 (23.5) |
Warfarin | 9 (1.6) |
Clinical data | |
Baseline NIHSS score, median (IQR) | 13 (8–20) |
Body weight, mean (SD), kg | 65 (13) |
Systolic blood pressure, mean (SD), mm Hg | 161 (30) |
Glucose, mean (SD), mmol/L | 8.49 (3.72) |
Platelet count, mean (SD), ×109/L | 215 (70) |
Actual tPA dose, median (IQR), mg/kg | 0.86 (0.75–0.91) |
OTT, median (IQR), min | 125 (100–155) |
OCSP classification, n (%) | |
TACI | 207 (37.8) |
PACI | 162 (29.6) |
POCI | 48 (8.8) |
LACI | 111 (20.3) |
Uncertain | 20 (3.6) |
Both the SITS SICH risk score and the extended risk score reasonably predicted the occurrence of SICH and were well calibrated (Table
3). The AUCs were higher for the extended risk score than for the SITS SICH risk score across all definitions of SICH, with significant differences per NINDS (P = 0.015) and per ECASS II (P = 0.016). Reclassification of patients improved with an NRI of 22.3% (P = 0.011) for SICH per NINDS, 21.2% (P = 0.018) per ECASS II, and 24.5% (P = 0.024) per SITS-MOST (Additional file
1: Table S3 to S5). Given the designated risk strata, the SITS SICH risk score categorized 19% of patients at low risk, 66% at average risk, and 15% at elevated risk, whereas the corresponding values by the extended risk score were 37%, 44%, and 19%.
Table 3
Comparison of prediction performance between the SITS SICH risk score and the extended risk score
SICH per NINDS | | | | | | | |
SITS SICH risk score | 1.35 (1.11-1.64) | 1.72 | 0.887 | 0.624 (0.533-0.714) | - | - | - | - |
Extended risk score | 1.30 (1.15-1.46) | 4.30 | 0.745 | 0.704 (0.618-0.791) | 0.081 | 0.015 | 22.3% | 0.011 |
SICH per ECASS II | | | | | | | |
SITS SICH risk score | 1.34 (1.08-1.68) | 3.25 | 0.661 | 0.612 (0.503-0.721) | - | - | - | - |
Extended risk score | 1.30 (1.13-1.49) | 4.23 | 0.752 | 0.703 (0.611-0.796) | 0.091 | 0.016 | 21.2% | 0.018 |
SICH per SITS-MOST | | | | | | | |
SITS SICH risk score | 1.49 (1.14-1.94) | 2.10 | 0.835 | 0.678 (0.563-0.793) | - | - | - | - |
Extended risk score | 1.33 (1.12-1.58) | 6.45 | 0.488 | 0.723 (0.622-0.824) | 0.044 | 0.293 | 24.5% | 0.024 |
Discussion
We demonstrated that incorporation of the OCSP classification into the SITS SICH risk score significantly improved the performance in predicting SICH in our study population. The extended risk score reasonably predicted SICH across the three definitions with good calibration. A substantial proportion of patients without SICH were reclassified into lower risk category (NRI for these patients was about 14% across the three definitions of SICH). Overall, the extended risk score moved an additional 18% (37% minus 19%) of patients to the low risk category, and 46% (168 of 362) patients with average risk (3 to 5 points according to the SITS SICH risk score) were reclassified (Additional file
1: Table S3 to S5).
The improved discriminatory ability of the extended risk score is clinically useful and might facilitate thrombolytic treatment in acute ischemic stroke. Notably, many emergency physicians were unwilling to provide thrombolysis for acute ischemic stroke in fear of SICH [
24]. In Taiwan, perceived risk of SICH among emergency physicians, neurologists, and patients was a major barrier to implement thrombolytic treatment [
25]. Risk prediction models that could correctly identify patients at low SICH risk after thrombolysis are of great benefit to stroke patients because they help remove the psychological barriers to administering tPA.
Studies show that patients with posterior circulation stroke are unlikely to develop SICH or hemorrhagic transformations [
10,
11]. Therefore in our study the patients with POCI were assigned an extended risk score of zero. A potential mechanism underlying the low incidence of SICH in posterior circulation stroke is the infrequent permeability derangements detected on pretreatment MRI [
26]. Permeability derangements, which indicate blood–brain barrier disruption in ischemic fields, increased the propensity for hemorrhagic transformation in stroke patients treated with either recanalization therapy or conservative care [
27]. Another possible mechanism is better collaterals in the territory of the posterior cerebral artery than that of the middle cerebral artery [
28]. Patients with good collateral circulation might be less vulnerable to SICH after recanalization therapy [
29,
30]. Additionally, the size of ischemic brain tissue was associated with SICH after intravenous thrombolysis [
31,
32]. The small lesion volume in infratentorial strokes as compared to supratentorial strokes might partly explain the low incidence of SICH in patients with POCI.
Because the volume of the affected brain tissue is generally correlated with clinical stroke severity, models for predicting post-thrombolysis SICH usually included the NIHSS or the less commonly used Canadian Neurological Scale as a predictor [
3‐
9]. However, patients with right hemisphere strokes may have a low NIHSS score despite a substantial lesion volume [
33]. Consequently, patients with right hemisphere nonlacunar strokes would have a higher risk of post-thrombolysis hemorrhage than those with left hemisphere strokes and similar NIHSS scores [
34]. In contrast, the OCSP classification in the early hours of ischemic stroke correlated well with infarct size [
14]. Without the assessment bias between hemispheres, the OCSP classification complements the NIHSS score in predicting SICH risk.
Although the OCSP classification generally corresponded well to the radiological findings, the sensitivity and specificity were different among the four OCSP categories. Using lesion topography on diffusion-weighted MRI as the diagnostic standard, the likelihood of correct classification was low for PACI and LACI, whereas high for TACI and POCI [
35]. Therefore, we did not differentiate patients with PACI from those with LACI in designing the extended risk score. Theoretically, a prediction model combining imaging findings and clinical factors could improve the prediction of SICH. A recent study indicated that the blood Sugar, Early infarct signs, hyperDense cerebral artery sign, Age, and NIHSS (SEDAN) score [
6] had the highest predictive power among the existing risk prediction scores [
36]. Our previous study also found that the HAT score performed well [
16]. However, the interpretation of early infarct signs or hyperdense cerebral artery sign requires considerable radiological expertise, which may not be feasible in settings where neuroradiologists are not available on a 24/7 basis [
37]. In addition to indicating an anterior circulation infarct, the appearance of early infarct signs on CT also hints at longer elapsed time since stroke onset [
38]. The inclusion of total anterior circulation infarct and the onset-to-treatment time, which is a component of the SITS SICH risk score, in our revised risk score offers a clinical counterpart to the radiological findings of early infarct signs. Although the SITS SICH risk score in conjunction with the OCSP classification seems more complex, it provides an alternative when the local radiological expertise is limited. In particular, the OCSP classification could be easily and reliably determined using a standard symptom list [
39].
Methods of model assessment should depend on the intended use of the model. Intravenous thrombolysis should not be withheld for an otherwise eligible patient simply because of high anticipated risk of SICH. The demonstrated improvement in classification performance in the extended risk score should be used only to better comprehend the risk associated with thrombolytic therapy. In particular, reclassification performance is sensitive to the number of clinically relevant risk categories, and the OCSP classification renders applying the risk scores largely unnecessary in patients with POCI.
Awareness of the applications and potential limitations of the risk scores will aid the clinicians in daily clinical decision making. Patients and their family could be better informed of the risk of SICH before making a shared treatment decision with their physicians. Those at high risk of SICH may benefit from more intensive monitoring, such as blood pressure and blood glucose. Another potential use of the risk scores is for case-mix adjustment in light of the fact that post-thrombolysis SICH might be used to measure performance of acute stroke care [
40].
Our study has limitations. First, the OCSP syndromes were assessed based on medical records, rather than personal examination. Although the interrater agreement in the clinical classification of syndromes was moderate, the accuracy of classification might be compromised. Further studies with prospective OCSP classification are needed to confirm the clinical implications of our findings. Second, the number of SICH events was small, which precluded multivariable analyses, such as reweighting the predictors of the SITS SICH risk score. The limited number of events might also explain the failure to show superior discrimination (higher AUC) of the extended risk score in predicting SICH per the SITS-MOST definition. Third, a proportion of our patients were treated with a lower dose of intravenous tPA. Whether the prediction performance of risk models is subject to dosage remains to be explored. Finally, our study should be viewed as hypothesis generating. Potential patient selection bias could impact the model performance. Further validation is needed to strengthen generalizability of our findings.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
SFS, conception and design, analysis and interpretation of data, drafting the article; SCCC, analysis and interpretation of data, revising the manuscript critically for important intellectual content; HJL, conception and design, analysis and interpretation of data, revising the manuscript critically for important intellectual content; CHC, conception and design, analysis and interpretation of data, revising the manuscript critically for important intellectual content; MCT, analysis and interpretation of data, revising the manuscript critically for important intellectual content; CSW, analysis and interpretation of data, revising the manuscript critically for important intellectual content; YCH, analysis and interpretation of data, revising the manuscript critically for important intellectual content; LCH, analysis and interpretation of data, revising the manuscript critically for important intellectual content. YWC, conception and design, analysis and interpretation of data, revising it critically for important intellectual content, final approval of the version to be published. All authors read and approved the final manuscript.