Combining MRI with NIHSS Thresholds to Predict Outcome in Acute Ischemic Stroke: Value for Patient Selection

BACKGROUND AND PURPOSE: Selecting acute ischemic stroke patients for reperfusion therapy on the basis of a diffusion-perfusion mismatch has not been uniformly proved to predict a beneficial treatment response. In a prior study, we have shown that combining clinical with MR imaging thresholds can predict clinical outcome with high positive predictive value. In this study, we sought to validate this predictive model in a larger patient cohort and evaluate the effects of reperfusion therapy and stroke side. MATERIALS AND METHODS: One hundred twenty-three consecutive patients with anterior circulation acute ischemic stroke underwent MR imaging within 6 hours of stroke onset. DWI and PWI volumes were measured. Lesion volume and NIHSS score thresholds were used in models predicting good 3-month clinical outcome (mRS 0–2). Patients were stratified by treatment and stroke side. RESULTS: Receiver operating characteristic analysis demonstrated 95.6% and 100% specificity for DWI > 70 mL and NIHSS score > 20 to predict poor outcome, and 92.7% and 91.3% specificity for PWI (mean transit time) < 50 mL and NIHSS score < 8 to predict good outcome. Combining clinical and imaging thresholds led to an 88.8% (71/80) positive predictive value with a 65.0% (80/123) prognostic yield. One hundred percent specific thresholds for DWI (103 versus 31 mL) and NIHSS score (20 versus 17) to predict poor outcome were significantly higher in treated (intravenous and/or intra-arterial) versus untreated patients. Prognostic yield was lower in right- versus left-sided strokes for all thresholds (10.4%–20.7% versus 16.9%–40.0%). Patients with right-sided strokes had higher 100% specific DWI (103.1 versus 74.8 mL) thresholds for poor outcome, and the positive predictive value was lower. CONCLUSIONS: Our predictive model is validated in a much larger patient cohort. Outcome may be predicted in up to two-thirds of patients, and thresholds are affected by stroke side and reperfusion therapy.

R eperfusion therapy improves outcomes of patients with acute ischemic stroke. 1,2 The decision to treat is primarily determined by the time from symptom onset. Use of IV-rtPA is restricted to the 3-1 or 4.5-hour 2 time window, resulting in only 1%-7% of patients receiving IV-rtPA. 3 Advanced neuroimaging provides information about a patient's physiology that may be useful to guide treatment decisions, especially in an extended time window. Selecting patients for reperfusion therapy on the basis of the mismatch between lesions in diffusion-weighted and perfusion images has been proposed, 4 but this approach has not been uniformly proved to predict a beneficial treatment response. [5][6][7][8] On the basis of findings that patients with DWI infarct volumes of Ͼ70 mL have poor outcome regardless of treatment 9,10 and that the NIHSS score is a strong predictor of outcome, 11 we recently published a predictive model that combined DWI and mean transit time lesion volumes with NIHSS score thresholds to predict outcome in patients with anterior circulation acute ischemic stroke. 12 DWI lesion volume Ͼ 72 mL or NIHSS score Ͼ 20 predicted poor outcome (mRS, [3][4][5][6], while MTT lesion volume Ͻ 47 mL or NIHSS score Ͻ 8 predicted good outcome (mRS, 0 -2) with a high positive predictive value (PPV) when used in combination, in two-thirds of patients.
In this study, we sought to validate our predictive model in a larger, independent cohort of patients and investigated the effects of reperfusion therapy and side of involvement on these thresholds.

Patient Selection and Assessment
We retrospectively examined the clinical and imaging data of consecutive patients with acute ischemic stroke admitted to our comprehensive stroke center. Inclusion criteria were the following: 1) DWI and PWI obtained within 6 hours of the patient being symptom-free, 2) acute MCA stroke, 3) NIHSS score Ͼ 2, and 4) sufficient clinical follow-up to determine the mRS score at 90 days. This study was approved by our institutional review board. Records were maintained in compliance with the Health Insurance Portability and Accountability Act.
We identified 198 patients with acute ischemic stroke who underwent DWI and PWI within 6 hours. Seventy-five were excluded because of the lack of a follow-up mRS score (n ϭ 28), posterior circulation infarction (n ϭ 29), calculation of perfusion maps with a delay-sensitive deconvolution technique (n ϭ 9), or image quality insufficient for analysis (n ϭ 9). The remaining 123 patients were included for further analysis. This cohort was independent of the previously published cohort, 12 without overlap. In patients who received IV-tPA, treatment was started within 4.5 hours. In patients who received intra-arterial (IA) therapy, treatment was initiated within 8 hours after symptom onset.
Admission NIHSS score was determined by a stroke neurologist. Clinical outcome was evaluated by using the mRS score 90 days after stroke onset, which was determined by a stroke neurologist in 59 instances (48%) and was retrieved from a neurologic examination in the patient's medical record for the remaining patients. Patient outcomes were dichotomized into good (mRS 0 -2) versus poor (mRS 3-6).

Image Acquisition, Postprocessing, and Analysis
MR imaging was performed on a 1.5T Signa scanner (GE Healthcare, Milwaukee, Wisconsin). DWI was performed with a singleshot echo-planar spin-echo sequence with two 180°radiofrequency pulses. Three images per section were acquired at bϭ0 s/mm 2 , followed by 25 at bϭ1000 s/mm 2 . Imaging parameters were TR/TE, 5000/80 -110 ms; FOV, 22-cm; matrix, 128 ϫ 128 zero-filled to 256 ϫ 256; 5-mm thickness; 1-mm gap. PWI was performed by using a dynamic-susceptibility technique. Serial echo-planar gradient-echo images were acquired with TR/TE, 1500/40 ms; FOV, 22 cm; matrix, 128 ϫ 128; section thickness, 5 mm with a 1-mm gap. Twenty milliliters of gadopentetate dimeglumine, 0.5 mmol/mL (Bayer HealthCare Pharmaceuticals, Wayne, New Jersey), was injected intravenously at 5 mL/s, beginning 10 seconds after scanning started, followed by 20 mL of normal saline. PWI data were processed by using locally written, automated software. Signal-versus-time curves for each pixel were converted to ⌬-R2-versus-time curves. CBV was calculated by integrating the area under these curves; CBF, as the amplitude of the scaled residue function yielded by delay-insensitive singular-value deconvolution 13 ; Tmax, as the time at which the scaled residue function reached its maximum; and MTT, as CBV/CBF.
For quantitative measurement, visually detected DWI and MTT abnormalities were segmented in randomized order by 2 experienced neuroradiologists blinded to clinical information by using a semiautomated commercial analysis program (Analyze, Version 7.0; (AnalyzeDirect, Overland Park, Kansas). Tmax maps were thresholded to 6 seconds. Absolute mismatch was defined as MTT Ϫ DWI; percentage mismatch, as (MTT Ϫ DWI) / DWI ϫ 100%.

Statistical Analysis
Statistical analysis was performed by using MedCalc for Windows, Version 11.2.1 (MedCalc Software, Mariakerke, Belgium). P values Ͻ .05 were considered significant. Clinical and imaging data were compared by using the Student t test or Mann-Whitney U test, and categoric data, by using the 2 test. Receiver operating characteristics curves were calculated for DWI, MTT, Tmax, and the NIHSS score relative to clinical outcome. Areas under the curve (AUCs) were compared and stratified by reperfusion therapy and stroke side. Prognostic yield was defined as the proportion of patients fulfilling at least 1 threshold criterion.

Baseline Clinical and Imaging Characteristics, Treatment, Outcomes, and Predictors of 3-Month mRS
Sixty-eight patients (55.3%) had good clinical outcome 90 days after stroke onset (Table 1). There were no significant differences in age, sex, right-hemisphere involvement, relative mismatch volume, or time from symptom onset to MR imaging between patients with good-versus-poor outcome. Median admission NIHSS score was significantly higher (16 versus 7, P Ͻ .0001), and mean DWI, MTT, Tmax, and absolute mismatch volumes were significantly larger (63.4 versus 16.3, 189.9 versus 80.6, 217.3 versus 89.5, and 127.1 versus 64.9 mL, respectively; all P Ͻ .0001) in patients with poor outcome. More patients with poor outcome than with good outcome had proximal occlusions (ICA or MCA M1) versus MCA M2/M3 or no identifiable occlusion (P Ͻ .0001). Sixty-one patients (49.6%) received IV-tPA, and 25 (20.3%) received IA thrombolysis. Thirty-seven (30.1%) patients did not receive IV or IA therapy.

Side of Involvement Affects Clinical and Imaging Thresholds
The DWI threshold set at 95% specificity for poor outcome was lower with leftsided (Ͼ51.8 mL) than right-sided strokes (Ͼ98.5 mL), and NIHSS score thresholds were higher (Ͻ9 or Ͼ20 versus Ͻ5 or Ͼ16) when the left hemisphere was affected. Prognostic yield was higher with strokes involving the left side: For DWI, the yield was 40.0%-16.9% versus 20.7%-10.4%, and for NIHSS score, the yield was 66.2%-38.5% versus 29.3%-1.7% (Table 4). Furthermore, the AUC was higher for DWI with left-sided strokes (0.875 versus 0.706, P ϭ .043), and there was a trend toward a higher AUC with the NIHSS score (0.911 versus 0.808, P ϭ .077). The mean DWI volume in patients with good outcome was smaller in left-versus right-sided strokes ( Fig 3A). A probability plot for poor outcome demonstrates better predictability for left-sided strokes (Fig 3B). Scatterplots for good-versus-poor outcome show that with left-sided strokes, no patient with an NIHSS score of Ͼ20 (orϽ8) had good (or poor) outcome, while for right-sided strokes, no patients with NIHSS scores of Ͼ17 had good outcome, and some patients with NIHSS scores of Ͻ8 had poor outcome (Fig 3B). PPV of our model was 97.4% for left-sided strokes, and 82.1% for right-sided strokes (P ϭ .028).

Patients in Whom Imaging or Clinical Thresholds Predicted Clinical Outcome Incorrectly
Three patients had poor outcomes despite an admission NIHSS score of Ͻ8. All 3 had right-sided MCA strokes, and 2 had critical ICA stenoses with substantial infarct growth after initial evaluation (On-line Table and On-line Figure, patient 3). Three patients had MTT abnormalities of Ͻ50 mL, but poor outcome. All 3 had significant comorbidities and/or complicated recoveries. Also, patient 5 (On-line Figure) had an infarct in the precentral gyrus, a highly eloquent region strongly represented on the mRS score. Three patients had DWI abnormalities of Ͼ70 mL but  good outcome. These were relatively young patients (26 -45 years of age) who all underwent reperfusion therapy (twothirds had received IA thrombolysis and had uncomplicated recoveries without significant infarct extension). Two also had right-sided strokes.

DISCUSSION
We have validated our previously described predictive thresholds of DWI Ͼ 70 mL or NIHSS Ͼ 20 for poor outcome of patients with acute ischemic stroke and of MTT Ͻ 50 mL or NIHSS score Ͻ 8 for good outcome, with a high PPV in this much larger, entirely independent cohort. In the 65% of patients who met at least 1 of these criteria, outcome was predictable with an 89% PPV. Indeed, we identified only 9 of 123 patients in whom an imaging or clinical threshold predicted clinical outcome incorrectly. Furthermore, we extended our previous results by comparing treated (with IV or IA therapy) versus untreated patients and by comparing right-versus left-sided strokes. Factors involved in patient selection for reperfusion therapy are nuanced and are currently based on the time from onset and neurologic deficit. Advanced imaging may help to further refine these parameters and more appropriately select patients for intervention. Currently, the most frequently proposed advanced MR imaging method to select patients for intravenous reperfusion therapy is the mismatch between DWI and certain PWI lesion sizes. 4 Although some evidence supports this strategy, 14,15 it has not conclusively predicted a favorable treatment response. [5][6][7][8] In addition, relative mismatch measurements fail to account for the absolute sizes of the infarct core and abnormally perfused areas. 7 In the Echoplanar Imaging Thrombolytic Evaluation Trial (EPITHET), absolute Tmax and DWI lesion volumes influenced the response to reperfusion, but relative mismatch did not. 16 Finally, DWI/PWI mismatch does not reflect the patient's clinical status. Clinical status, as measured by the NIHSS score, is an important independent predictor of outcome 11 and neurologic worsening. 17 In general, in our predictive model, patients with DWI lesions of Ͼ70 mL or an NIHSS score of Ͼ20 have a high likelihood of poor outcome, and reperfusion therapy in these patients may offer minimal or no clinical benefit. Patients with MTT lesions of Ͻ50 mL or an NIHSS score of Ͻ8 have a high likelihood of good outcome. Aggressive treatments may unnecessarily expose this group to the risks of invasive   reperfusion therapy. The outcomes of the remaining 35% of patients, those with DWI volume of Ͻ70 mL, MTT volume of Ͼ50 mL, and an NIHSS score of Ͼ8 and Ͻ20, could not be predicted accurately by initial clinical and imaging assessments; this finding suggests that these patients may represent a group toward whom early invasive therapy may be directed. We hypothesized that patients who received IV thrombolytic and/or IA recanalization therapy would have higher DWI and NIHSS score thresholds for poor outcome than untreated patients. Indeed, the 100% specific DWI threshold for poor outcome was 103 mL in treated compared with 31 mL in untreated patients, and the 100% specific NIHSS score thresholds were similarly 20 versus 17, respectively, suggesting that treated patients may tolerate larger baseline infarcts. This suggestion is also supported by significantly larger DWI lesions and higher NIHSS scores in treatedversus-untreated patients with good outcome. An explanation for this finding might be that in patients treated with reperfusion therapy, infarcts grow less than in untreated patients. When predicting good outcome with the NIHSS score, MTT, or Tmax, we did not find conclusive evidence for differences by treatment. We did not further stratify patients by different treatment strategies and recanalization status (eg, IV-versus-IA thrombolysis) because recanalization status was unknown in the 61 patients treated with IV therapy only and our sample size was too small.
Because the dominant (usually left) hemisphere has important functions represented in the mRS, we hypothesized that patients with right-sided strokes would tolerate larger DWI lesions and that our model would be less accurate. Indeed, the AUCs were significantly higher for DWI (0.88 versus 0.71, P ϭ .04) with left-sided strokes. The difference in AUC for the NIHSS score in leftversus right-sided strokes also approached significance (0.91 versus 0.81, P ϭ .08). Moreover, patients with right-sided strokes could tolerate larger DWI lesions (100% specific DWI volume for poor outcome ϭ 103 versus 75 mL), and on average, patients with right-sided strokes and good outcome had significantly larger DWI lesions (20 versus 10 mL, P Ͻ .05). No patient with an infarct affecting the right hemisphere had an NIHSS score of Ͼ17, making outcome prediction in these patients less reliable.
In the 9 patients for whom the above clinical or imaging thresholds predicted outcome incorrectly, we identified 4 scenarios: 1) Young patients undergoing reperfusion may do well with DWI lesion volumes of Ͼ70 mL, in accordance with studies suggesting better outcome in patients younger than 40 years of age. 18 2) Factors delaying rehabilitation or causing infarct growth, such as comorbidities, complicated recovery, and irregular clinical course, might lead to poor outcome despite MTT lesion volumes of Ͻ50 mL. 3) Infarcts in areas affecting functions highly represented in the mRS score (eg, motor function in the precentral gyrus) may result in poor outcome despite small MTT lesion volumes. This possibility is underscored by a recent study, in which infarct location on MR imaging correlated highly with outcome. 19 4) NIHSS score and mRS are strongly biased toward dominant (usually left) hemisphere functions. All patients with false prediction based on an NIHSS score of Ͻ8 had right-sided infarcts, and two-thirds of patients with infarct volumes of Ͼ70 mL and good outcome had right-sided strokes. The challenges to prediction represented by these variables are important to keep in mind when evaluating patients for therapy. The limitations of our investigation are the consequences of the retrospective design and have been described elsewhere. 12 Briefly, we could not obtain data such as reperfusion status, final infarct volumes, and clinical status at day 30 for most patients. There were variable treatments and variable times from stroke onset to imaging and from imaging to follow-up clinical assessment. Potential selection bias was related to using MR imaging, the decision to offer thrombolysis, and our technical approach to performing IA therapy. While assessment of Tmax maps was based on a 6-second threshold, DWI and MTT lesions were outlined by visual estimation. Thresholds have been proposed to increase reliability, but most studies demonstrating a clinical benefit from perfusion imaging-based patient selection have used visual estimation. 14,15 No thresholding software is universally accepted, and our data suggest that both methods can be used in our model with similar yield and PPV. Finally, our method for measuring DWI and PWI lesion volumes is too time-consuming for routine clinical use. The faster ABC/2 volume estimation method based on 3 orthogonal measurements, easily obtained on an MR imaging scanner console, could be a feasible substitute. 20

CONCLUSIONS
The previously proposed predictive model combining DWI, MTT, and NIHSS score thresholds is validated in a much larger patient cohort. For acute ischemic stroke, thresholds applied to acute DWI and MTT lesion volumes and NIHSS scores can be used to predict good and poor clinical outcomes with a high PPV. Our prediction model shows higher PPV in left-sided strokes. Younger patients, those treated with IV and/or IA therapy, and those with right-sided strokes may have good outcomes despite larger DWI infarcts. Those treated with reperfusion therapy may also tolerate a higher NIHSS score. These variables, alone and in combination, may help guide decisions related to patient selection for therapy. Patients who do not meet these thresholds have variable outcomes and may represent a target population for more aggressive revascularization approaches.