Article Text
Abstract
Background Optimal imaging triage for intervention for large vessel occlusions remains unclear. MR-based imaging provides ischemic core volumes at the cost of increased imaging time. CT Alberta Stroke Program Early CT Score (ASPECTS) estimates are faster, but may be less sensitive.
Objective To assesses the rate at which MRI changed management in comparison with CT imaging alone.
Methods Retrospective analysis of patients with acute ischemic stroke undergoing imaging triage for endovascular therapy was performed between 2008 and 2013. Univariate and multivariate analyses were performed. Multivariate logistic regression was used to evaluate the effect of time on disagreement in MRI and CT ASPECTS scores.
Results A total of 241 patients underwent both diffusion-weighted imaging (DWI) and CT. Six patients with DWI ASPECTS ≥6 and CT ASPECTS <6 were omitted, leaving 235 patients. For 47 patients, disagreement between the two modalities resulted in different treatment recommendations. The estimated probability of disagreement was 20.0% (95% CI 15.4% to 25.6%). In a multivariate logistic regression, CT ASPECTS >7 (p=0.004) and admission National Institutes of Health Stroke Scale (NIHSS) score <16 (p=0.008) were simultaneously significant predictors of agreement in ASPECTS. The time between modalities was a marginally significant predictor (p=0.080).
Conclusions The study suggests that patients with NIHSS scores at admission of <16 and patients with CT ASPECTS >7 have a higher likelihood of agreement between CT and DWI based on an ASPECTS cut-off value of 6. Additional MRI for triage in patients with NIHSS at admission of >16, and ASPECTS of 6 or 7 may be more likely to change management. Unsurprisingly, patients with low CT ASPECTS had good correlation with MRI ASPECTS.
- Stroke
- CT
- MRI
Statistics from Altmetric.com
Introduction
Mechanical thrombectomy for acute ischemic stroke achieves much higher rates of recanalization than that of the initial experience, now ranging between 59% and almost 90% in recent trials.1–4 Favorable outcomes in successful thrombectomy cases therefore increasingly depends on the proportion of salvageable brain in affected territories, which probably decreases with time.5 Futile recanalization remains a concern, as reperfusion of large territories of irreversibly infarcted brain is unlikely to yield benefit to the patient. Indeed, in the recent group of randomized trials assessing IA thrombectomy, the study that yielded the highest proportion of patients with modified Rankin scores of 0–2 at rates of approximately 50% used a CT perfusion protocol to estimate “ischemic core” on the basis of RAPID software analysis (Stanford, non-commercial).4 ,6
The optimal imaging-based triage of patients with acute large vessel ischemic stroke remains unclear. In general, imaging paradigms estimate core infarct volume, which may help to predict potential treatment response. Regimens based on CT with tools such as the Alberta Stroke Program Early CT Score (ASPECTS),7 ,8 perfusion imaging, whether by CT or MRI,9–13 or primarily volumetric assessment of core by diffusion weighted imaging (DWI),14–17 are in use in varying proportions, as well as for prognostication for use during the administration of IV tissue plasminogen activator.18 Increasing the sophistication of imaging-based triage may yield better selection characteristics, but increasing imaging sophistication also increases imaging time, which may in turn delay treatment, resulting in poorer outcomes in some patients.19 ,20
This study is derived from the database of patients at an institution employing DWI-based triage paradigm for candidates for IA thrombolysis and thrombectomy, yielding a patient cohort that was imaged with both MRI and CT, allowing comparison of imaging characteristics and an analysis of current triage algorithms. It sought to ascertain what differences in CT and DWI ASPECTS in the same patient might be present in this patient population as a function of time from symptom onset. We hypothesized that the degree of mismatch might increase and/or decrease given that the imaging findings of ischemia become apparent on CT and MRI at different rates and that in some patients MRI might be unhelpful, and thus unnecessary.
Methods
Retrospective review of the institutional acute stroke endovascular database between 2008 and 2013, identified patients aged >17 years and considered for endovascular stroke treatment, in accordance with local institutional review board approval. Baseline clinical characteristics and treatment parameters were systematically collected, such as the National Institutes of Health Stroke Scale (NIHSS), stroke risk factors, and time at which last known well (LKW); timing and results of the imaging studies used to triage the patients, including MRI and CT and CT angiography, were also determined. All included patients showed evidence of a large vessel occlusion on non-invasive angiography. CT and MRI ASPECTS scores were retrospectively assessed by two fellowship-trained vascular neurologists, with arbitration by a board-certified neuroradiologist after review of the non-contrast enhanced CT images with appropriate window and leveling, and MRI diffusion-weighted images together with review of the apparent diffusion coefficient images.
MRI protocol
An abbreviated MR procedure was performed as previously described,12 which includes fluid attenuated inversion recovery; diffusion-weighted images were obtained. MR angiography of the head and neck was carried out at the discretion of the managing team.
The goals of this study were to (1) estimate the frequency of disagreement in ASPECTS between CT and DWI resulting in different management of patients—namely, whether or not they should receive IA therapy, and (2) identify any variables that predict when these two modalities will agree, such that DWI is not needed. Disagreement in patient management between DWI and CT ASPECTS was defined as CT ASPECTS≥61 and DWI ASPECTS <6. These goals were examined for all patients and for the subgroup of patients with ≤1 h between DWI and CT.
A 95% CI for the frequency of disagreement was constructed using the Wilson method.21
Univariate logistic regression was used to assess each potential predictor of disagreement. The following predictors were considered: age (continuous variable), gender, time between CT and MRI (treated as a continuous variable), time between LKW and MRI (treated as both a continuous variable, and then categorized as within the 0–3 h window or not), NIHSS at admission (ordinal variable and then categorized as <16 or ≥16), and CT ASPECTS (treated as ordinal variable and then categorized as ≤7 or >7). Linear and quadratic terms were evaluated for the continuous variables. In the univariate analyses, we looked at both continuous (or ordinal data) and a binary categorization for three variables: time between LKW and MRI, NIHSS at admission, and CT ASPECTS. For the other variables we did not consider categorization.
The CT ASPECTS categorization was based on a cut-off value used in a recent publication;1 the NIHSS categorization was based on a cut-off values of 16 separating moderate stroke from severe stroke; and the 0–3 h window for time between LKW and MRI was based on our institution's protocol, in which we try to carry out MRI of patients <3 h after LKW.
A multivariate logistic regression model was used to evaluate the effect of the time between MRI and CT on disagreement in ASPECTS, after adjusting for other variables associated with the disagreement in ASPECTS. Interactions with time between MRI and CT were considered. A significance level of 0.05 was used to evaluate the variables in the model.
Results
All patients
Data were available for 241 patients who underwent both DWI and CT. Six patients with DWI ASPECTS ≥6 and CT ASPECTS <6 were omitted, leaving 235 patients for analysis. The time interval between CT and DWI ranged from DWI performed 19.4 h before CT to DWI performed 18.8 h after CT. The median time interval was DWI performed 70 min after CT, with DWI performed between 44 and 147 min after CT in 50% of patients.
The patients comprised 125 women (53.2%) and 110 men (46.8%), with a mean (SD) age of 70.0 (14.6) years (range 26–100). The median time between LKW and CT was 224 min (range 0–1246). The median time between LKW and DWI was 305 min (range 7–1341 ). The mean NIHSS at admission was 15.5 (range 2–34); data were missing for two patients. The CT ASPECTS ranged from 1 to 10, with 50% of subjects having an ASPECTS of ≥8. The DWI ASPECTS ranged from 0 to 10, with 50% of subjects having a score of ≥7.
For 47 patients, disagreement between the two modalities resulted in different treatment recommendations. Thus, the estimated probability of disagreement overall was 20.0% ( 95% CI 15.4% to 25.6%).
Table 1 summarizes the results of the comparison between those with agreement and those with disagreement. Patients where the ASPECTS on CT and DWI agreed tended to have lower NIHSS scores at admission (p=0.009), and were more likely to have a CT ASPECTS >7 (p=0.002) than patients where the ASPECTS of the two modalities disagreed.
In a multivariate logistic regression model, CT ASPECTS >7 (p=0.004), admission NIHSS <16 (p=0.007), and time between LKW and CT (p=0.038) were simultaneously significant predictors of agreement in ASPECTS. The time between CT and MRI was a marginally significant predictor (p=0.092); the data suggest a marginal trend for increasing agreement in ASPECTS as the time between the two modalities decreases. There were no significant interactions with time between CT and MR (p≥0.44). Figure 1 illustrates the estimated probability of agreement (from the fitted model) in ASPECTS between CT and MRI as a function of the time between CT and MRI, holding NIHSS and CT ASPECTS constant.
Subgroup of patients with MRI and CT within 1 h
This subgroup was felt to best represent the CT/MRI discrepancy between these patients at the time of assessment, and is of particular interest in analysis. Table 2 summarizes the difference between CT and DWI ASPECTS for the 86 patients for whom the time between MRI and CT was within 1 h. The median difference in ASPECTS was 1, and the difference ranged from −3 (DWI ASPECTS > CT ASPECTS) to 5 (CT ASPECTS > DWI ASPECTS).
For 14 patients disagreement between the two modalities resulted in different treatment recommendations. Thus, the estimated probability of disagreement overall was 16.3% (95% CI 10.0% to 25.5%).
Table 3 summarizes the results for those with agreement and those with disagreement between DWI and CT. Patients for whom the ASPECT scores on CT and DWI agreed tended to have lower NIHSS scores at admission (p=0.040), and were more likely to have a CT ASPECT score >7 (p=0.028) than patients for whom the ASPECT scores of the two modalities disagreed. In particular, among the 14 patients with disagreement between the two modalities, nine of them had a CT ASPECT score of 6 or 7, and 11 of them had a NIHSS score ≥16. Owing to the small number of disagreements in the study sample, a multivariate analysis could not be performed for this subgroup.
Table 4 summarizes the six patients for whom the initial assessment of CT ASPECTS yielded a score lower than that of the DWI ASPECTS, including the probable reason for discrepancy. Figure 2 shows MR and CT images of a representative patient. One patient (patient 1 in table 4) had encephalomalacic changes in the region of new ischemia, which probably confounded analysis. The other five cases demonstrated similar degrees of ASPECTS abnormality on CT, and may represent variance in interpretation of ASPECTS severity.
Discussion
The basic premise of the imaging triage system employed at the study institution is to use MRI to identify and exclude patients for whom intervention would be futile,22 based on data that suggest that stroke volumes above certain thresholds correlate with adverse outcomes.15–17 ASPECTS may act as a proxy for stroke volume analysis.17 Most patients with acute ischemic stroke will receive CT examination as triage for IV tissue plasminogen activator triage, which can be used to triage for IA therapy as well. Changes in parenchyma that indicate ischemia may be more easily detectible on diffusion restriction as seen on MRI than from CT changes,23 and using MRI to assess ischemic core is generally accepted as a better method to determine the volume of irreversible infarction.
Initial analysis of the dataset examined whether there were some time spans between LKW and the time of imaging that might show a high rate of agreement between CT and MRI such that MRI would be highly unlikely to change management. No such period was found that demonstrated statistically significant differences from MRI/CT agreement in general. However, one subpopulation showed a higher degree of agreement (MRI ASPECTS findings leading to a decision to treat with IA)—namely, patients with NIHSS <16 and ASPECTS >7. In such patients, the agreement rate was 78.4% (29/37). For the subgroup in which CT and MRI were performed within 1 h—that is, the subgroup most closely representing CT and MRI results if obtained simultaneously, the MRI–CT agreement was 84.6% (11/13). Possibly, in MR-DWI-based centers, it might be better to avoid MRI and proceed directly to intervention.
Cost of imaging timing
It is generally accepted that ‘time is brain’ and, that while the rate of neuronal death varies, there is probably some progression of neuronal death as the ischemia time increases.24 ,25 Figure 1 shows the probability of agreement between CT ASPECTS and DWI ASPECTS with respect to the amount of time between acquisition of the CT and the MR images.
The decreased probability for agreement as the time between MR and CT acquisition increased showed a trend towards statistical significance (p=0.092). It is possible if not probable that this trend reflects continued progression of the ischemic core as time between acquisitions lengthen. While this finding appears intuitive, it argues against excessive delays in MR acquisition, which may be time consuming,26 ,27 as triage for IA therapy, and may confirm that ischemic core continues to expand in some subgroup of patients; excessive delays in MRI may result in some patients progressing to ischemic core volumes that discourage intervention.
Limitations
These data are derived from a single institution and acquired partially retrospectively. Additionally, patients who did not receive MRI on the basis of a large hypodensity on the initial CT were not included. The study does not include all outcome and demographic data, as it was deemed beyond the scope of this assessment. Thorough evaluation of the differences in the comorbidities and demographics might yield additional insight into the source of MRI and CT ASPECTS discrepancies.
Another limitation is the nature of ASPECT scoring, as the number is quantized, and assessment of abnormality, territory by territory is subject to interpreter error, which is certainly possible in clinical practice. The sample size in this study was too small to enable multivariate statistical models to determine the probability that MRI might alter patient management. Outcomes of these patients were not included in this study, as the focus of the study was to assess the relative value and tissue costs of CT and MRI during the course of ischemic stroke triage based on current data on ischemic core size. Interpretation of these data in light of new core size thresholds may yield different results. Additionally, the techniques used for thrombectomy are actively changing, and thus outcomes may change as new thrombectomy techniques evolve.28–30
Additionally, it is possible that DWI is not a definitive measure of core and some of the so-called ‘core’ might be salvageable with timely treatment, particularly at early time points,11 ,12 ,31 although this may be clinically uncommon. Animal data also support this suggestion,32 demonstrating reversal, as well as diffusion-weighted re-reversal, which might be qualitatively different from the result of prolonged ischemia.
Conclusion
The goal of imaging for candidates suitable for IA therapy is to assess which patients might benefit, and to exclude patients for whom intervention would probably be futile. MRI and CT, non-perfusion-based core assessments may afford different sensitivities to ischemic core volumes, with MRI probably being more sensitive. However, MRI acquisition requires time, which may lead to larger DWI lesions as the infarct extends.
Using a cut-off point of ASPECTS >6, the rate of agreement in the overall dataset was 80.0% (95% CI 84.6% to 74.4%), while patients who received CT and MRI within 1 h had an agreement rate of 84.7% (95% CI 90.0% to 74.5%). Whether such a level of agreement is sufficient to omit MRI is unclear.
The study suggests that patients with intermediate to high NIHSS scores at admission of <16 and patients with either low CT ASPECTS or CT ASPECTS >7 have a higher likelihood of agreement between CT and DWI in patient management recommendations based on an ASPECTS cut-off value of 6. Thus the addition of MRI for triage in patients with NIHSS at admission of >16 and CT ASPECTS of 6 or 7 may be more likely to result in a change of management. In institutions where MRI is routinely used, the possibility that MRI might be less useful for triage in other groups should be considered. Further study is warranted as an unnecessarily increased imaging time intuitively should have a deleterious effect on outcomes.
Acknowledgments
Thanks again for the wonderful editorial assistance of Christine Moore.
References
Footnotes
Contributors FKH: conceived and designed the research. DW, S-MM, SJ, and EC-C: acquired the data. FKH and DW: analyzed and interpreted the data. NAO: performed the statistical analysis. FKH: drafted the manuscript. FKH, SH, GT, KU, and IK: made critical revisions of the manuscript. FKH, SH, GT, KU, and DW: approved the final manuscript.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.