The derivation cohorts in this study were enrolled in the Jrecan trial between 1 March 2018 and 30 June 2019. The Jrecan trial, a prospective, multicenter, randomized, non-inferior clinical trial, was designed to verify the safety and efficacy of the Jrecan Flow Reconstruction Device manufactured by Hunan Ruikang Technologies Co., Ltd. for endovascular treatment of acute ischemic stroke, and registered with Chinese Clinical Trial Registry (NO. ChiCTR-TOC-17013822, http://www.chictr.org.cn/showproj.aspx?proj=23396). The trial involved 16 high-capacity stroke centers in China and was approved by PLA Rocket Force General Hospital Ethics Committee (NO. X2017008). The validation cohorts in this study were from Baotou central hospital (March 2019–December 2021) and Ningbo No.2 Hospital (February 2019–December 2021), with 76 and 94 cases, respectively.

The studies were included if they met the following criteria: (1) Aged ≥18 years old; (2) AIS: within 8 h; (3) baseline admission National Institute of Health Stroke Scale (NIHSS) score ≥ 6；(4) modified Rankin Scale (mRS) score before onset ≤ 2; (5) angiography confirmed intracranial internal carotid artery, middle cerebral artery (MCA) trunk acute obstruction in patients; (6) The subject (or his guardian) agreed to participate in the study and signed an informed consent form.

The exclusion criteria were as follows: (1) CT or MR showed evidence of intracranial hemorrhage, or is known to have bleeding tendency; (2) CT showed low-density lesions more than 1/3 of the middle cerebral artery area, or ASPECTS < 7; (3) DSA showed carotid artery dissection, complete occlusion of cervical carotid artery or vasculitis; (4) DSA prompted serious tortuous path blood vessels, blood flow reconstruction device is difficult to reach the target location; (5) DSA angiography prompted an acute blockage of 2 vessels; (6) had a history of stroke within 3 months; (7) had heart, lung, liver and kidney failure or other serious diseases (such as brain tumors, cerebral arteriovenous malformations, systemic infections, active disseminated intravascular coagulation, myocardial infarction within 12 months before enrollment, A serious history of mental illness); (8) Pregnant or lactating women; (9) was known to be severely allergic to contrast media; (10) were participating in other drug or device research; (11) Life expectancy was less than 90 days.

The epidemiological information and clinical data of the admitted patients were the same as in our previous studies (7). In this study, clinical data of patients were collected as follows: (1) Demographic information: age and sex; (2) Previous medical history: hypertension, diabetes, coronary heart disease, atrial fibrillation, previous stroke history; (3) Baseline data: preoperative mRS Score, blood glucose, blood pressure, preoperative NIHSS score, ASPECT score, whether intravenous thrombolysis was performed; (4) Intraoperative and postoperative: first pass effect, anesthesia method, lesion site, residual severe stenosis degree, remedial treatment, anticoagulation therapy, recanalization time; (5) Complications: 24 h ICH, 24 h sICH, survival status. See Table 1 for details.

Effective recanalization (ER) was defined as modified thrombolysis in cerebral infarction (mTICI) scores of 2b or higher and associated rates of functional independence or good functional outcomes, defined as mRS ≤ 2 at 90 days. FR was defined as the level of modified TICI (mTICI) 2b to 3, despite successful recanalization, modified Rankin Scale (mRS) scores ≥3 still appeared after 90d (4, 8). Symptomatic intracranial hemorrhage (sICH) was defined as an increase of ≥4 in the NIHSS score within 24 h or death. Early neurological deterioration (END) was defined as increase in ≥4 points of the NIHSS between admission and 24 h.

All data analyses were performed using SPSS software (version 22.0, IBM Corporation, Armonk, New York, United States) and R version 4.2.1 software (http://www.R-project.org, foundation for statistical computing, Vienna, Austria). The normally distributed continuous data were expressed as mean ± standard deviation ( X ¯ ± s), and the T-test was used for comparison between groups. Non-normally distributed continuous variables were expressed as median (quartile) [M (Q1, Q3)], and Mann Whitney U test was used for comparison between groups. Count data were expressed in cases (percentages) and comparisons between groups were performed using either the x² test or Fisher’s exact probability test. Multivariate logistic regression analysis was used to determine significant predictors of FR, and a nomogram was established. The performance of the nomogram was verified in the modeling group and the verification group by C-index, calibration chart, Hosmer Lemeshow test and decision curve analysis (DCA), respectively. C-index were used to measure discrimination; The calibration plot described the degree of fit between the occurrence of the actual invalid reconnection and the nomogram predicted invalid reconnection; DCA to present the net return at various threshold probabilities.

The derivation set was based on RCT data from Jrecan and initially included 193 patients. 2 cases did not use any instrument, 4 cases were seriously deviated from experimental protocol, 1 case was lost for follow-up, and 24 cases had a mTICI of 2a or 0 after opening. A total of 162 eligible patients were finally included in the derivation set (Figure 1). The median age of the patients was 66 years old (56–74), and 58.9% were male. Other clinical information is shown in Supplementary Table 1. In the derivation set, there were 88 cases of FR and 74 cases of ER, with FR accounting for 54.32%.

The validation set initially included 238 people, 38 cases did not use any instrument, 15 cases were seriously deviated from experimental protocol, 1 case was lost for follow-up, and 14 cases had mTICI of 2a or 0 after MT. Therefore, a total of 170 people were eventually included in the validation set (Figure 1). The median age of patients was 67 years (59–75), the proportion of males was 52.9%, and other clinical information was shown in Supplementary Table 1. In the validation set, 82 cases were FR and 88 cases were ER, with FR accounting for 48.24%.

There was no significant statistical difference in FR rates between the derivation set and validation set [88/162 (54.32%) and 82/170 (48.23%), p = 0.267] (Supplementary Table 1). The 90d mRS Scores of the patients in the derivation set were shown in Figure 2, including ER group/FR group, ICH group/No-ICH group, sICH group/No-sICH group, and survival group/death group, with significant statistical differences in each group.

The baseline characteristics of ER and FR in the derivation set and the results of univariate analysis are listed in Table 1. To prevent missing meaningful variables, factors with p < 0.2 were shown as potential predictors of FR in the derivation set. The predictors included in multivariate analysis were Age (p = 0.004), Hypertension (p = 0.084), Glucose (p = 0.012), Systolic pressure (p = 0.014), and Initial NIHSS (p = 0.105), ASPECT (p = 0.068), Anesthesia methods (p = 0.090), Location of lesions (p = 0.080), Residual severe stenosis (p = 0.118), OTR (p = 0.056), 24 h ICH (p = 0.001).

The risk factors of FR in the derivation set based on the results of multivariate Logistic analysis are shown in Table 2. The results showed that Age ≥ 65 years old (OR = 2.096, 95%CI 1.024–4.289, p = 0.043), SBP ≥ 180 mmHg (OR = 5.624, 95%CI 1.141–27.717, p = 0.034), OTR ≥ 453 min (OR = 2.759, 95%CI 1.323~5.754, p = 0.007), 24 h ICH (OR = 4.029, 95%CI 1.844~8.803, p < 0.001) was an independent risk factor for FR.

The risk prediction nomogram of FR in AIS based on multivariate Logistic analysis is shown in Figure 3A, and each independent risk factor was assigned with a score ranging from 0 to 100 points. The four independent predictors obtained by multivariate analysis were scored according to the weight: when age ≥ 65, 43 points; When SBP ≥ 180 mmHg, 100 points; OTR ≥ 453 min, 59 points; When 24 h ICH occurs, 81 points, the final total score is 260 points. Thus, it can be concluded that there is a 50% probability of FR occurring when the total score is 73 (Figure 3A). For the model, after the internal verification of 1,000 bootstrap samples, the calibration diagrams of the derivation set (Figure 3B) and the validation set revealed that the nomogram model had good prediction accuracy (Hosmer-Lemeshow test: the derivation set X² = 3.348, p = 0.764; the validation set X² = 4.892, p = 0.429). The C-index of FR in the derivation set and the validation set were 0.739 (95%CI 0.662~0.816, st = 0.039; Figure 3B) and 0.703 (95%CI 0.621~0.785, st = 0.042; Figure 3C), respectively, suggesting a remarkable sensitivity and specificity of the nomogram in the clinical context. In addition, to evaluate the net benefit of the nomogram model in predicting a FR in AIS after MT, the DCA was performed. The results shown that when threshold probabilities are ranged between 25% and 93% in the derivation set (Figure 4A) and between 28% and 74% in the validation set (Figure 4B), using a nomogram to predict FR may have a greater net benefit than a “treat all” or “treat none” strategy, suggesting that the nomogram model has good clinical utility.

According to the points of each variable contributed to the nomogram, we translated the predictive model into practice. The total scores of each patient was calculated as follows: total scores = (0 for age < 65 or 1for age ≥ 65) × 43 + (0 for SBP < 180 mmHg or 1 for SBP ≥ 180 mmHg) × 100+ (0 for OTR < 453 min or 1for OTR ≥ 453 min) × 59 + (0 for 24 h no-ICH or 1for 24 h ICH occurs) × 81. The ROC curve and the scatter plot of the total scores of each patient in the derivation and the validation sets are shown in Figures 5A,B.

There are still some limitations in this study. First, although the data in our study came from multiple centers, it only included Asian populations and may lack generality. Secondly, the sample size of this study is limited, and high-quality large-scale RCT studies are still needed for further verification in the future. Third, missing samples in data collection may cause the accuracy of the model to be affected. Fourth, this study only included patients with internal carotid artery and middle cerebral artery occlusion, which may not be applicable to patients with posterior circulation occlusion. Fifth, Because the majority of enrolled patients were within the thrombectomy time window, the study design at that time did not use advanced imaging assessment to assess the ischemic penumbra, which was one of the limitations of this study.