Risk Stratification for Early Neurological Deterioration After Mechanical Thrombectomy and Development of a 24-h Postprocedural Reassessment Model in Patients With Acute Anterior-Circulation Large-Vessel Occlusion Based on Computed Tomography Perfusion Mismatch Volume

Figures

• 

  Fig. 1.. Patient selection flowchart. Of 612 consecutively screened patients, 174 were excluded: posterior-circulation occlusion (n = 68), no preprocedural computed tomography perfusion (CTP) or unanalyzable images (n = 39), prestroke modified Rankin Scale (mRS) score >2 (n = 22), severe comorbidity or terminal malignancy (n = 15), concomitant intracranial hemorrhage/tumor or another severe intracranial lesion (n = 10), onset-to-puncture time >24 h (n = 7), and inability to ascertain both the 72-h National Institutes of Health Stroke Scale (NIHSS) score and survival status (n = 13). The primary analysis included 438 patients; all 9 deaths within 72 h were retained and classified as early neurological deterioration (END). The 24-h landmark cohort excluded 51 patients who had developed END at or before 24 h and included 387 patients, of whom 36 developed delayed END after 24 h and through 72 h. LVO, large-vessel occlusion.

• 

  Fig. 2.. Schematic workflow for CTP image postprocessing. (A–G) show image input, motion correction, arterial input function (AIF)/venous output function (VOF) selection, perfusion maps, threshold-based segmentation, quantitative output, and variables entered into the risk model, respectively. The illustrative values are an ischemic core volume of 31.4 mL, a time-to-maximum (Tmax) >6-s hypoperfusion volume of 126.7 mL, a mismatch volume of 95.3 mL, and a mismatch ratio of 4.0. NCCT, noncontrast computed tomography; CTA, computed tomography angiography; CBF, cerebral blood flow; CBV, cerebral blood volume; MTT, mean transit time; rCBF, relative cerebral blood flow; R, right; L, left. In panel (E), the red area represents the ischemic core (rCBF <30%), and the blue area represents the hypoperfused region (Tmax >6 s).

• 

  Fig. 3.. Nonlinear association between mismatch volume and END risk and exploratory stratification. (A) shows the restricted cubic spline (RCS) curve with 95% confidence intervals (CIs). The horizontal dashed line indicates OR = 1, and the two vertical dashed lines indicate the exploratory cut points of 75 and 105 mL; (B) shows the observed END rates in the <75-mL, 75–105-mL, and >105-mL groups.

• 

  Fig. 4.. Scatterplots and regression analyses of CTP metrics and clinical and treatment-related variables. (A) ischemic core volume versus admission NIHSS score; (B) ischemic core volume versus mismatch volume; (C) mismatch volume versus admission NIHSS score; (D) mismatch volume versus collateral score; (E) ischemic core volume versus HT; (F) mismatch volume versus successful recanalization. (A–D) show scatterplots with fitted linear regression lines, and (E,F) show scatterplots for binary outcomes with fitted logistic regression curves; shaded areas represent 95% CIs. ρ, Spearman rank correlation coefficient; OR, odds ratio. A: ρ = 0.46, p < 0.001; B: ρ = −0.35, p < 0.001; C: ρ = −0.21, p < 0.001; D: ρ = 0.29, p < 0.001; E: OR per 10 mL = 1.21, p < 0.001; F: OR per 10 mL = 1.03, p = 0.217.

• 

  Fig. 5.. Summary of internal performance for the preprocedural baseline and 24-h landmark reassessment models. (A) shows the apparent area under the receiver operating characteristic curve (AUC) with 95% CIs and the bootstrap-corrected AUC. (B) shows sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and Hosmer–Lemeshow (H–L) test results.

• 

  Fig. 6.. Sensitivity analyses. (A) shows ORs with 95% CIs for the association between mismatch volume <75 mL and END across sensitivity analyses. (B) shows the apparent and bootstrap-corrected AUCs.

• 

  Fig. 7.. Summary of least absolute shrinkage and selection operator (LASSO) sensitivity analyses. (A) shows the variables retained at λmin and λ1se; (B) shows the corresponding AUCs with 95% CIs and bootstrap optimism. λmin, penalty parameter associated with the minimum cross-validation error; λ1se, more parsimonious penalty parameter selected using the 1-standard-error rule.