Relationship between Lesion Topology and Clinical Outcome In Anterior Circulation Large Vessel Occlusions

Abstract

Background and purpose

Diffusion-weighted imaging (DWI) ASPECTS, a surrogate of infarct volume, predicts outcome in anterior large vessel occlusion (LVO) strokes. We aim to determine whether topological information captured by DWI ASPECTS contributes additional prognostic value.

Methods

Adults with intracranial ICA, M1 or M2 MCA occlusions who underwent endovascular therapy were included. The primary outcome measure was poor clinical outcome (3-month modified Rankin Scale score 3–6). Prognostic value of the 10 DWI ASPECTS regions in predicting poor outcome was determined by multivariable logistic regression, controlling for final infarct volume, age and laterality.

Results

213 patients (mean age 66.1±14.5 years, median NIHSS 15) were included. Inter-rater reliability was good for DWI ASPECTS (Deep regions: Kappa=0.72, Cortical regions: Kappa=0.63). All DWI ASPECTS regions with the exception of the putamen were significant predictors (p<0.05) of poor outcome in univariate analyses. Statistical collinearity among ASPECTS regions was not observed. Using penalized multivariable logistic regression, only M4 (OR=2.82 95%CI 1.39–5.76) and M6 (OR=2.45, 95%CI 1.15–5.3) involvement were associated with poor outcome. M6 involvement independently predicted poor outcome in right hemispheric strokes (OR=5.8, 95%CI 1.9–20.3) whereas M4 (OR=4.3, 95%CI 1.3–15.0) involvement predicted poor outcome in left hemispheric strokes adjusting for infarct volume. Topologic information modestly improved the predictive ability of a prognostic score that incorporates age, infarct volume and hemorrhagic transformation.

Conclusions

Involvement of the right parieto-occipital (M6) and left superior-frontal (M4) regions impact clinical outcome in anterior LVOs over and above the effect of infarct volume and should be considered during prognostication.

INTRODUCTION

Ischemic strokes that involve cerebral cortex result in neuro-cognitive deficits such as aphasia, neglect, visuospatial and cognitive dysfunction in addition to motor disabilities that contribute to long-term disability.^1–4 In LVO stroke patients, 3-month modified Rankin Scale (mRS) score 0–2, i.e. slight disability with preserved independence, is a commonly used clinical metric and study endpoint to measure good outcome while mRS 3–6 reflects poor outcome.^{5, 6} Despite being heavily influenced by motor impairment, mRS also captures functional limitations and disabilities due to non-motor deficits.⁷ Final infarct volume (FIV) is one of the most robust predictors of clinical outcomes in anterior circulation LVO (aLVO) stroke and is incorporated in the Pittsburgh Outcomes after Stroke Thrombectomy (POST) score that predicts clinical outcome with excellent discriminative power. ^8–10 Yet, significant deviation from predicted outcomes in not uncommon in clinical practice even with expert clinical judgement or the use of validated prognostic scores.¹¹ It is possible that infarct topology explains some of this variability.

Previous efforts to assess the relationship between infarct topology and outcome have utilized CT and MRI-based approaches.^{12, 13} The Alberta Stroke Program Early CT Score (ASPECTS) captures infarct location by dividing the anterior circulation into 10 regions: three deep regions (Caudate [C], Putamen [P], Internal capsule [IC]) supplied by the lenticulostriate perforators from the M1 MCA segment, four cortical regions at the level of the basal ganglia (M1, M2, M3, Insula [I]) and three supra-ganglionic cortical regionFs (M4, M5, M6). Hypodensity or loss of gray-white distinction on CT is scored ‘0’ while absence of hypodensity is scored ‘1’.¹⁴ A higher score suggests preserved brain parenchyma and an early CT ASPECTS ≥7 predicts better outcomes with thrombolytic and endovascular reperfusion therapy.^{15, 16} A study of the NINDS tPA trial cohort used individual regions on pre-treatment CT ASPECTS to predict long-term clinical outcomes in stroke patients and found that M6 region involvement predicted poor outcomes in older patients.¹³ This supports the hypothesis that infarct topology impacts clinical outcome, however, whether lesion location impacts outcome after controlling for infarct volume has not been evaluated.

Use of pre-treatment CT to assess the impact of individual ASPECTS regions on clinical outcome suffers from two limitations: (1) low/moderate-sensitivity for core infarct and (2) inability to account for stroke expansion. Magnetic resonance imaging (MRI) with diffusion weighted imaging (DWI) sequences is significantly more sensitive than CT with a higher inter-rater reliability for the detection of early ischemia.^17–19 DWI ASPECTS on MRI performed more than 12 hours from time stroke onset captures both topologic information and the completed infarct size. This study utilized DWI ASPECTS measured 12–72 hours post-treatment in an endovascular cohort of aLVOs to evaluate the association between lesion topology, infarct volume, stroke laterality and clinical outcome.

METHODS

Data Source and Subjects

The study cohort was derived from a prospective endovascular registry at University of Pittsburgh Medical Center (UPMC) and contained consecutively treated adult patients (age≥18 years) with aLVO (ICA-terminus, M1 and M2 MCA) between January 2007 and January 2014. Patients were required to have had a follow-up MRI scan 12–72 hours from treatment with a documented 3-month (±4 weeks) mRS score. Institutional review board approval for the maintenance of the institutional endovascular stroke database was obtained.

Measurements

Baseline NIHSS scores were calculated by the treating physicians. ASPECTS on pre-treatment non-contrast head CT scan upon arrival at the stroke center was determined by the treating neurologist prior to endovascular treatment. All patients underwent CT or MR angiography prior to endovascular treatment and level of occlusion was confirmed by conventional angiography. Reperfusion (Thrombolysis-In-Cerebral-Infarction [TICI] grade) was determined by the operating physician after the procedure.²⁰ DWI ASPECTS was calculated retrospectively by two investigators using the DWI B-1000 sequence from a 12–72 hour follow-up MRI scan. Each DWI ASPECTS region was scored 1 (preserved) or 0 (restricted diffusion >1 mL). Punctate foci of restricted diffusion were not considered in the ASPECTS calculation under 1 mL except when involving the deep regions or insula. Deep DWI ASPECTS (caudate, putamen and internal capsule) and Cortical ASPECTS (M1–6) and Total DWI ASPECTS were calculated. DWI ASPECTS information for the opposite hemisphere was also collected. FIV (mL) was determined by measuring area of restricted diffusion (DWI B-1000 hypersignal) on each slice followed by summation over the number of slices.^{10, 21} Three-month mRS was measured by the treating physician at follow-up. Those who died earlier than 3 months were assigned 3-month mRS of 6. Patients without MRI or 3-month outcomes were excluded from the analysis.

Statistical analysis

All DWI ASPECTS regions were evaluated in univariate analysis and then simultaneously assessed by multivariable logistic regression using Firth’s penalized likelihood method to predict poor outcome (3-month mRS 3–6). All DWI ASPECTS regions were included in the final model regardless of statistical significance. Multi-collinearity was assessed first by creating a correlation matrix for all DWI ASPECTS regions. Effect of collinear pairs of variables (Pearson’s correlation coefficient ≥0.5) was assessed by dropping one variable of each pair from the regression model and assessing for changes in the overall model and of individual regression coefficients. Collinearity was also evaluated by the variation inflation factors (VIF), VIF≥5.0 being considered significant. Statistical interaction between laterality (right versus left hemisphere) and significant DWI ASPECTS regions was assessed. Because interaction parameters reached significance, we evaluated DWI ASPECTS regions in separate penalized regression models for right and left hemispheric strokes adjusting for FIV. We also simultaneously assessed 20 DWI ASPECTS variables (10 on the side of occlusion and 10 contralateral) by penalized logistic regression to predict poor outcome. Four of 213 patients had contralateral hits resulting in abnormal contralateral DWI ASPECTS (range 8–10).

DWI ASPECTS regions of predictive value were assessed in a model that included the POST score which incorporates age, FIV and parenchymal hemorrhagic transformation.¹⁰ Models incorporating the POST score with or without significant topological variables were compared using −2log Likelihood and the likelihood ratio test. Receiver Operating Characteristic (ROC) area under the curve (AUC) was used to compare discriminative power of the POST score for poor outcome, with and without topological variables in the model.²² We also determined the Net Reclassification Improvement (NRI), Integrated Discrimination Improvement (IDI) and relative IDI (rIDI) to test whether addition of significant DWI ASPECTS regions to a model containing the POST score improved the predictive ability of the model.²³ All statistical analyses were performed using IBM SPSS Statistics Version 22 and SAS software Ver. 9.4.

RESULTS

Patient characteristics

Out of 532 aLVO patients treated in the study period, 250 were excluded due to lack of MR imaging within the 12–72 hour time window and 69 were excluded due to lack of 3-month mRS data (Supplemental Figure I). The 213 patients included in the analysis had mean age of 66.1 (±14) years, median NIHSS 15 [IQR 11–18], median initial CT ASPECTS 8 (IQR 7–9) and mean time from last-normal-to-treatment of 603 min (±58 min); 51% had right hemispheric involvement; 18.3% (n=39) had ICA terminus and 67.6% (n=144) had M1 MCA occlusions. Successful reperfusion (TICI2B/3) status was achieved in 66.3% (132/199) cases (Table 1). The median FIV was 74 mL (IQR 16–97) and the mean POST score was 105.5 (± 3.2). 53.05% (n=113) of patients achieved poor outcome (3-month mRS 3–6).

Table 1.

Patient characteristics and outcome measures

Variable (N=213)	Measure	Value
Age (years)	Mean(SD)	66.1 (14.5)
Baseline CT ASPECTS	median(IQR)	8 (7–9)
Baseline NIHSS score	median(IQR)	15 (11–18)
Time-to-Treatment (minutes)	mean(SEM)	603 (58)
Level of Occlusion
M1 MCA occlusion	n(%)	144 (67.6%)
Intracranial ICA occlusion	n(%)	39 (18.3%)
TICI2B/3 reperfusion	n(%)	123 (57.8%)
Right Hemisphere	n(%)	109 (51%)
Outcome measures
FIV (mL)	median(IQR)	74 (16–97)
POST score	median(IQR)	92.2 (74–120.1)
3-month mRS	median(IQR)	3 (1–6)
Poor outcome (mRS 3–6 at 3 months)	n(%)	113 (53.05%)

Impact of lesion topology on 3-month clinical outcomes:

Median total DWI ASPECTS was 5 (IQR 2–7) for the study population and the distribution of individual DWI ASPECTS regions is summarized in Table 2. Post-treatment DWI ASPECTS showed a stronger correlation (Pearson’s coefficient=0.49, p<0.001) with 3-month mRS as compared to pre-treatment CT ASPECTS (Pearson’s coefficient=0.22, p=0.003). Inter-rater reliability for two observers was good for DWI ASPECTS (Deep ASPECTS regions: Kappa=0.72, Cortical ASPECTS regions: Kappa=0.63). In univariate analyses, all ASPECTS regions, with the exception of the putamen (p=0.089), were significantly associated with poor outcome (p<0.05). However, only M4 (OR=2.82 95%CI 1.39–5.76, p=0.005) and M6 (OR=2.45, 95%CI 1.15–5.3, p=0.02) involvement were significantly associated with poor outcome in multivariable penalized regression analyses that included all DWI ASPECTS regions (Table 3). Highest correlation was observed between C and P (Pearson’s coefficient=0.53), and I and M2 (Pearson’s coefficient=0.52) DWI ASPECTS regions suggesting multi-collinearity (Supplemental Table I). Exclusion of putamen or M2 regions from the full model in order to account for possible collinearity between C and P regions and between I and M2 regions did not alter the results i.e. only M4 and M6 had significant regression coefficient estimates (Supplemental Table II). No significant statistical collinearity was observed for any of the ten variables (Variance Inflation Factors < 5).

Table 2.

Distribution of MRI DWI ASPECTS regions

DWI ASPECTS Region**	Measure	Value
Total ASPECTS	median(IQR)	5 (2–7)
Deep (C, P, IC)	median(IQR)	1 (0–3)
Cortical (I, M1-M6)	median(IQR)	3 (1–5)
Caudate (C )	n(%)	122 (57.5%)
Putamen (P)	n(%)	68 (32.1%)
Internal Capsule (IC)	n(%)	126 (59.4%)
Insula (I)	n(%)	53 (25%)
M1	n(%)	118 (55.7%)
M2	n(%)	70 (33%)
M3	n(%)	109 (51.4%)
M4	n(%)	118 (55.7%)
M5	n(%)	49 (23.1%)
M6	n(%)	135 (64%)

^**Each DWI ASPECTS region scored as 1 if spared, 0 if involved.

Table 3.

Multivariable penalized regression analysis to predict poor outcome (3-month mRS 3–6)

DWI ASPECTS Region	Odds Ratio (OR)	95% Confidence Interval	p value
C	1.494	0.699–3.210	0.309
P	1.276	0.557–2.924	0.571
IC	1.029	0.483–2.160	0.942
I	1.361	0.581–3.177	0.485
M1	1.069	0.492–2.267	0.866
M2	1.901	0.829–4.375	0.138
M3	0.857	0.396–1.812	0.696
M4	2.819	1.394–5.763	0.005
M5	1.812	0.830–4.075	0.151
M6	2.445	1.148–5.300	0.024

ORs compare "Involvement" to "Sparing" of individual DWI ASPECTS regions.

Since our goal was to estimate the topologic influence of infarction, over and above the influence of infarct volume, we controlled for final infarct volume in the model and still found that M4 (OR 2.27, p=0.045) and M6 (OR 2.67, p=0.016) involvement remained significant predictors of poor outcome while all other ASPECTS regions were not. Since complete MCA infarcts result in cortical ASPECTS of 0, inclusion of these patients in our analysis may have biased the estimates. In our cohort, 17.8% (38/213) of patients had involvement of all cortical regions. After excluding these patients, we still found that M4 (OR 6.52, p=0.011) involvement was a significant predictor of poor outcome and M6 (OR 2.93, p=0.087) involvement did not reach statistical significance.

Hemispheric laterality influences the impact of infarct topology on clinical outcome

Cortical symptoms observed in dominant and non-dominant hemispheric strokes are distinct and the association between infarct topology and clinical outcome is likely to be influenced by the side of the lesion (dominant vs. non-dominant). Since the majority of individuals are left hemispheric dominant, we used laterality (right versus left) as a surrogate of hemispheric dominance. In models that incorporated infarct volume, M4 and M6 DWI ASPECTS regions, we found that the interaction terms M4*Laterality (p=0.038) and M6*Laterality (p=0.013) reached statistical significance (Supplemental Table III) confirming the presence of effect modification. Therefore, the effect of topology on clinical outcome was assessed separately in left and right hemispheric strokes. In left hemispheric strokes, we found that M4 involvement was the only independent predictor of poor outcome (OR=5.46, 95% CI, 1.47–20.34, p=0.012) while all other ASPECTS regions and FIV were not. In right hemispheric strokes, only M6 involvement independently predicted poor outcome (OR=7.47, 95% CI 2.07 – 27.02, p=0.002) while other ASPECTS regions and FIV were not significant predictors of poor outcome (Table 4). As an alternative approach, we used penalized regression with 10 right and 10 left DWI ASPECTS regions simultaneously in one model and confirmed that right M6 (OR=6.25, 95%CI 2.0–22.1) and left M4 (OR=5.63 95%CI 1.9–18.6) regions were the only significant predictors of poor outcome (Supplemental Table IV).

Table 4.

Infarct laterality, topology and 3-month poor outcomes (mRS 3–6)

Laterality	Significant predictors	OR	95% Confidence Interval	p value	Non-significant covariates
Right (N=105)	M6 region	5.77	1.87–20.3	0.005	FIV, all other DWI ASPECTS regions
Left (N=103)	M4 region	4.31	1.34–15.03	0.022	FIV, all other DWI ASPECTS regions

Penalized logistic regression was used to estimate ORs which compare "Involvement" to "Sparing" of individual DWI ASPECTS region (p>0.05 considered non-significant).

Topologic information may improve predictive power of the Pittsburgh Outcomes after Stroke Thrombectomy (POST) score in aLVOs

We previously found that age, FIV and parenchymal hemorrhage are among the strongest predictors of clinical outcome in aLVOs and based on these observations, developed the POST score as a reliable tool for prognostication.¹⁰ We used the POST score in the model to control for these variables and found that M4 involvement in left hemispheric and M6 involvement in right hemispheric strokes, in addition to the POST score, were independently associated with poor outcome (Table 5). A model incorporating the POST score along with M4 and M6 involvement status had a significantly better model fit as compared to the POST score alone (−2log Likelihood 231.32 vs. 214.66 respectively, p<0.005). Addition of M4 and M6 involvement to the POST score seemed to modestly improve the discriminative power of the prediction model when compared to the POST score alone in right (AUC 0.90 vs. 0.84) and left hemispheric strokes (AUC 0.89 vs. 0.86). NRI was 4.55% (5/110) among patients with observed poor outcome and 0% (0/99) among those without poor outcome. Overall NRI of 4.55% suggests that addition of M4 and M6 information improved the classification accuracy of the model (Supplemental Table V). IDI, a measure of separation in predicted probabilities for events (mRS3–6) and nonevents (mRS0–2), was 0.038. The relative IDI (rIDI), a measure of incremental contribution of M4 and M6 variables to the model, was 13.84% (p<0.05).²³

Table 5.

Multivariable Logistic Regression: Topological Differences between Right and Left Hemispheric infarcts

	Right Hemisphere				Left Hemisphere
Variable	OR†	95% C.I.		p value	OR†	95% C.I.		p value
POST score*	0.97	0.95	0.99	.001	0.96	0.94	0.98	.001
M4 involved	0.67	0.22	2.10	.494	4.47	1.52	13.14	.007
M6 involved	4.58	1.51	13.83	.007	0.90	0.29	2.79	.855

^*POST score incorporates Age, FIV and PH1/PH2 hemorrhage and predicts clinical outcome in aLVOs.

^†All OR estimates compare "Involvement" with "Sparing" of individual DWI ASPECTS regions.

DISCUSSION

The ASPECTS method, originally designed for use with non-contrast head CT scans, measures infarcted regions in the anterior circulation.¹⁴ ASPECTS can also be determined on DWI MRI and like CT ASPECTS, DWI ASPECTS of ≥7 is associated with smaller infarct volumes and favorable clinical outcomes.¹⁸ We found that post-treatment DWI ASPECTS in an endovascular aLVO cohort can reliably capture information regarding infarct burden and topology.¹⁸ The inter-rater agreement for DWI ASPECTS in our study (Kappa≈0.7) was significantly higher than previously reported for CT-ASPECTS.¹⁹ This topological information captured by DWI ASPECTS also provided additional prognostic value in aLVO patients. We found that involvement of specific regions, namely the left M4 (superior frontal) and the right M6 (superior parietal) regions, have independent effects on the 3-month clinical outcome. This finding is further supported by observations made by Phan et. al. who assessed the utility of individual pre-treatment CT ASPECTS regions in an analysis of the NINDS tPA patient cohort using penalized logistic regression, and found that M6 involvement was associated with poor outcomes in the elderly.¹³ A limitation of pre-treatment CT ASPECTS as a predictor of final outcome is that many patients progress to larger infarcts especially if early recanalization of the occluded vessel is not achieved. This can be overcome by using post-treatment imaging in the first 72 hours to account for infarct growth, hemorrhagic transformation without the effect of cerebral edema. In patients who have undergone endovascular therapy, rapid recanalization prevents core infarct expansion and improves long-term clinical outcomes.^{5, 24} Partial or complete recanalization may also result in lower collinearity amongst ASPECTS regions, especially if ASPECTS is based on a follow-up scan at which time no further infarct expansion is anticipated. Therefore, our use of post-treatment MRI DWI ASPECTS after 12–72 hours may have overcome the problem of collinearity. Indeed, we did not observe multicollinearity amongst the 10 DWI ASPECTS regions in our analysis.

Our observations suggest that not all ASPECTS regions have equal clinical importance. Right parietal lobe involvement or left superior frontal lobe involvement seem to be independently associated with poor outcome. Furthermore, neither motor strip nor internal capsule region was found to be independently associated with clinical outcome. Since we did not capture information regarding hemispheric dominance in our cohort, we performed our analyses using laterality as a surrogate of hemispheric dominance. Since a large majority of individuals are expected to have left hemispheric dominance, ²⁵ we expect that most left hemispheric strokes in our study affected the dominant hemisphere while most right hemispheric strokes affected the non-dominant hemisphere. Therefore, the prognostic importance of the left M4 and right M6 regions likely reflects the importance of the dominant M4 and non-dominant M6 regions, respectively.

Our findings also illustrate the impact of cortical symptoms such as aphasia or hemineglect on patient outcomes, even when the outcome measure (mRS) is heavily influenced by motor deficits that affect ambulation.⁷ This is not surprising because hemineglect has been found to impact motor independence as measured by the Functional Independence Measure (FIM) and aphasia has been found to impact cognitive FIM in stroke patients.¹⁷ Severity of the disability and need for help or assistance distinguishes between patients with mRS 2 (significant disability, yet independent) and 3 (moderate disability, needing some help). This explains how cortical symptoms may impact the clinical outcome when the mRS score, dichotomized as 0–2 (good outcome) and 3–6 (poor outcome) is used in stroke research involving aLVOs.^{5, 6}

Our results may also be relevant to the interpretation of pre-treatment ASPECTS during patient selection for endovascular treatment. Involvement of the left M4 or right M6 regions prior to endovascular treatment may represent a poor prognostic factor especially if moderate core infarcts are already visible on the pre-treatment scan. On the other hand, reperfusion strategies could be considered even in patients with moderately sized core infarcts or intermediate pre-treatment CT ASPECTS (5–7) if the right M4 or left M6 regions are still preserved.

Robust prediction tools have been validated for use in aLVOs but do not utilize information regarding infarct topology. One example is the POST score which incorporates age, final infarct volume and presence of parenchymal hematoma to predict clinical outcome with high accuracy (AUC >0.8).¹⁰ The importance of lesion location on clinical outcome explains why a model that integrates DWI ASPECTS variables (M4 or M6) with the POST score seems to be superior to the POST score by itself. Since the patient cohort used in this analysis is a subset of the cohort used in the validation of the POST score, independent validation in other datasets is needed to further assess whether topological variables improve the predictive power of the POST score.

Our study has several limitations. Firstly, our database did not contain information regarding other measures of stroke-related disability such as FIM or stroke-related quality of life measures which may have a higher sensitivity than mRS for minor disabilities.²⁶ Although only M4 and M6 ASPECTS regions were significantly associated with clinical outcome in our analysis, other cortical and deep regions may also influence long-term clinical outcome but may not have been detected in our analysis possibly due to moderate sample size or clinical impacts that are underestimated when dichotomizing clinical outcomes as mRS 0–2 and 3–6. We also did not measure the volumes of individual DWI ASPECTS regions and are therefore unable to address the relationship between lesion location and lesion volume for individual ASPECTS regions. Such a study using ASPECTS may be limited by the arbitrary boundaries that separate various ASPECTS regions. Instead, we used the total FIV to control for the effect of infarct size on clinical outcome. Although we did not find statistical evidence for multi-collinearity in our study as measured by the VIF, this measure may have underestimated the effect of anatomy and shared vascular supply by various DWI ASPECTS regions. The moderate correlation observed between C and P, and I and M2 regions, suggest some degree of collinearity but these did not impact the overall results of our analysis. Voxel-based assessment of infarct location and volume may overcome the above limitations in our study and may reveal other cortical regions that are of prognostic significance.¹² Next, our study is retrospective and based on a single institutional endovascular stroke cohort. A large proportion of patients (319/532) were excluded due to lack of MRI data in the 12–72 hour time frame (46.9%) or missing follow-up information (12.96%). Since patients who did not have follow-up MRI scans were excluded from the study, this may have introduced selection biases due to the exclusion of patients who suffered early fatal complications, had contraindications for MRI scans, had strokes that were very large or had no residual deficits that an MRI was not felt to be necessary. Therefore, confirmation of our results in other aLVO populations and prospective cohorts is needed before generalizing our results.

In conclusion, we have demonstrated that topological information captured by MRI DWI ASPECTS on post-treatment imaging is of prognostic value in anterior circulation LVOs. Specifically, involvement of the right parietal region (M6) or the left superior frontal (M4) regions independently increase the odds for poor outcomes, even after controlling for other robust predictors of clinical outcome including age and final infarct volume. Our results support the hypothesis that infarct location and laterality, and not just infarct volume, impact clinical outcomes and provide additional prognostic value in aLVO patients.