Patients included in this study received treatment between November 1, 2021, and February 8, 2023. Our study was a subanalysis utilizing data from a nationwide, prospective registry in China, which enrolled patients presenting with acute large vessel occlusion and received standard treatment within 24 h of their last known well state in China (URL: http://www.chictr.org.cn. Uniform identifier: ChiCTR2100051664). Ethical approval for this study was granted by the institutional review board at the Second Affiliated Hospital of Army Medical University, and additional authorization was obtained from all participating centers’ ethics committees. From the originally published cohort (10), we conducted a subanalysis focusing exclusively on patients who met two additional criteria: (1) large ischemic core on NCCT (defined as an ASPECTS of 3–5); (2) eTICI ≥2b after EVT plus standard medical treatment (SMT).

Baseline characteristics included: (1) demographic data; (2) stroke risk factors; (3) laboratory results; (4) stroke severity assessed by NIHSS (11); (5) collateral status evaluated using the American Society of Interventional and Therapeutic Neuroradiology/Society of Interventional Radiology collateral grading system (ASITN/SIR) (12); (6) stroke etiology classified according to the Trial of ORG10172 in Acute Stroke Treatment (TOAST) classification (13); (7) workflow durations; (8) EVT methodology; (9) recanalization grade (eTICI); (10) location of occlusion; (11) baseline core infarct volume determined by the NCCT-based ASPECTS; (12) Procedure-related complications.

An independent imaging core lab was blinded to treatment allocation and clinical results. Two trained neuroradiologists analyzed all imaging data; they independently evaluated the baseline NCCT-based ASPECTS and assessed occlusion sites using supplemental angiographic imaging. In cases of discrepancies, a third senior neuroradiologist adjudicated any discrepancies. Successful recanalization was defined as eTICI ≥2b (blood flow to greater than 50%). In our study, 90-day functional outcomes were assessed using the mRS through follow-up visits or telephone interviews conducted by trained and experienced local physicians. According to the 90-day mRS, patients were categorized into two groups: FR group (90-day mRS 4–6) and meaningful recanalization (MR) group (90-day mRS 0–3). We present details of the data elements in Table 1.

We performed statistical analyses using SPSS 29.0, employing Mann–Whitney U tests for continuous variables and χ²/Fisher’s exact tests for categorical variables. Significant univariate predictors (p < 0.05) and variables with established clinical relevance as key determinants of infarct progression and recanalization outcomes in ACLVO infarction were included in our multivariable logistic regression model (3, 14). The adjusted covariates comprised age, sex, history of atrial fibrillation, baseline NIHSS score, blood glucose (random admission glucose), ASITN/SIR score, and eTICI grade. The results of multivariable logistic regression were expressed as adjusted odds ratios (aOR) and 95% confidence intervals (CI). Missing data were minimal (0.6–2.9% per variable, see Table 1 footnotes) and handled via complete-case analysis, which is consistent with recommendations for negligible missingness (15). To evaluate the predictive performance of continuous variables in our multivariable logistic regression model, we generated individual ROC curves for each univariate predictor (age, glucose, NIHSS score) and multivariate combinations (age + baseline NIHSS, glucose + baseline NIHSS, glucose + age, age + glucose + NIHSS) using GraphPad Prism 10. The area under the curve (AUC) was used to measure both independent contributions and synergistic predictive effects. To explore potential synergistic effects, we generated distribution surface plots using SigmaPlot 15 to visualize predicted outcome probability variations. For subgroup analyses, we similarly incorporated variables meeting both statistical significance (p < 0.05) and clinical relevance into logistic regression models to explore the occurrence of FR.

The MAGIC registry collected data from 38 comprehensive stroke centers throughout China, a total of 745 eligible patients (November 2021 and February 2023). We excluded 255 patients who received only SMT, 135 patients with NCCT-based ASPECTS scores of 0–2, and 42 patients who did not achieve successful recanalization (eTICI ≥ 2b). Ultimately, 313 patients were included in the final analysis. Among these, 142 patients (45.4%) were categorized into the meaningful recanalization (MR) group, while 171 patients (54.6%) were classified into the FR group. The study flowchart is presented in Figure 1.

Compared with the MR group, patients in the FR group were significantly older, had higher blood glucose levels, and higher baseline NIHSS scores. Additionally, the FR group showed a greater proportion of male patients, a higher percentage of atrial fibrillation, and a worse distribution of ASITN/SIR grades and eTICI grades (Table 1).

A multivariate logistic regression model was developed to identify independent predictors of FR in patients who achieved successful recanalization after EVT. The analysis revealed that higher baseline NIHSS scores, older age, elevated blood glucose, poor collateral circulation, and incomplete recanalization were significantly associated with FR (Table 2).

ROC analysis confirmed the predictive utility of these three independent predictors, with combined models showing improved discrimination (AUC = 0.7546, 95% CI, 0.7015–0.8078, p < 0.001) (Figure 2). To enhance clinical applicability, we derived optimal glucose cut-off values (7.5 mmol/L, which could effectively distinguish between MR and FR) using the maximum Youden’s index for subsequent subgroup analyses.

All variables in Figure 3 are continuous. The figure demonstrates the relationships between various predictors and the probability of FR. Figure 3A reveals a linear association between increasing blood glucose levels and higher FR probability when NIHSS scores are held constant. Notably, patients with higher NIHSS scores exhibit a steeper glucose-FR probability slope. Figure 3B demonstrates that concurrent elevations in both blood glucose and age are associated with increased probability of FR.

Establishing population-specific treatment thresholds for blood glucose is critical. Hyperglycemia in hospitalized patients was defined as any glucose level ≥140 mg/dL (7.8 mmol/L) (16). We adopted the optimal predictive cutoff value (7.5 mmol/L, closely aligned to 7.8 mmol/L) identified in this analysis as the dichotomization threshold to explore the relationship and mechanisms between glucose levels and FR in our study.

After excluding 6 cases with missing data, the final cohort comprised 307 patients. We used the cutoff value (7.5 mmol/L) to stratify patients into two groups: low glucose (LG) group (glucose <7.5 mmol/L, n = 170) and high glucose (HG) group (glucose ≥7.5 mmol/L, n = 137). Univariate analysis results are presented in Supplementary Table 1.

In the multivariate logistic regression analysis, age was identified as an independent predictor in both groups. Additionally, in the LG group, higher baseline NIHSS scores were associated with FR. Interestingly, in the HG group, baseline NIHSS scores were not a predictor of FR. We also found that FR was significantly associated with ASITN/SIR grade in the HG group, whereas in the LG group, FR showed a correlation with the eTICI grade (Supplementary Table 2). We assessed collinearity between age, high glucose levels (glucose ≥ 7.5 mmoL/L), and collateral status (ASTIN/SIR grade). All VIF values were < 2 (age: 1.042, glucose: 1.000, collateral status: 1.042).

There were 94 (68.6%) patients with FR in the HG group, and 72 (42.4%) patients with FR in the LG group. We also found that among 260 (84.7%) non-diabetic patients and 47 (15.3%) diabetic patients in the cohort, the LG group, 7 (4.1%) patients had diabetes, with 5 (5.1%) cases in the MR group and 2 (2.8%) cases in the FR group. In contrast, the HG group contained 40 (29.2%) diabetic patients, comprising 13 (30.2%) in the MR group and 27 (28.7%) in the FR group. Notably, diabetes prevalence was 7-fold higher in the HG stratum (29.2% vs. 4.1%), yet diabetic status itself did not modify FR risk within either group (HG: 28.7% FR in diabetics vs. 30.2% MR; LG: 2.8% vs. 5.1%) (Supplementary Table 3). These findings suggest that acute hyperglycemia, rather than chronic diabetic status, drives FR.

As with all multicenter prospective clinical studies, center-specific effects are inevitable. Due to the limited sample size in our study, we were unable to further validate the robustness of our findings using mixed-effects models, and the subanalysis’s results should also be interpreted with caution. Additionally, the lack of documentation regarding whether random admission blood glucose levels were measured under fasting conditions introduces a potential limitation to our subgroup analyses based on dichotomized cut-off values.