Procedure time and clinical outcomes after thrombectomy: a pooled cohort analysis of three randomised trials

Abstract

Introduction

Procedure time (PT) during endovascular thrombectomy (EVT) may be associated with clinical outcomes in acute ischaemic stroke (AIS). We evaluated the association between PT and clinical outcomes using a large pooled analysis.

Patients and methods

We performed a post hoc analysis of 3 multicentre randomised controlled trials including patients with anterior circulation large-vessel occlusion treated with EVT in China (2018–2023). The primary outcome was a favourable outcome (mRS, 0–2) at 90 days; the secondary outcomes were excellent outcome (mRS 0–1), sICH and all-cause mortality. Procedure time was evaluated as a continuous variable (per 10-min increase) and in categories (≤60, 60–120, > 120 min) using adjusted logistic and mixed-effects models.

Results

Among 2845 eligible patients (median age 68 years, 57.4% were male, median PT 69 min), each 10-min increase in PT was associated with an adjusted risk difference (aRD) of −1.52% (95% CI, −1.85% to −1.20%) in the probability of achieving mRS 0–2 and an aRD of 0.77% (95% CI, 0.53%–1.01%) increase in mortality, corresponding to lower odds of mRS 0–2 (aOR 0.928; 95% CI, 0.912–0.944) and higher odds of sICH (aOR 1.028; 95% CI, 1.007–1.049) and mortality (aOR 1.055; 95% CI, 1.037–1.073). Compared with PT ≤ 60 min, PT 60–120 and > 120 min were associated with reduced odds of mRS 0–2 (aOR 0.658 and 0.355, respectively) and increased mortality (aOR 1.648 and 2.422, respectively). These associations were consistent across sensitivity analyses. In exploratory mediation analyses, total passes and the first pass effect (FPE) explained only 12.9% and 11.4% of the association between PT and favourable outcome, respectively.

Discussion and Conclusion

Prolonged PT during EVT was associated with a reduced rate of favourable outcomes and increased sICH and mortality. These findings support continued efforts to minimise avoidable intraprocedural delays during EVT.

Graphical Abstract

Graphical Abstract.

Introduction

Endovascular thrombectomy (EVT) has been established as the standard of care for patients with acute ischaemic stroke (AIS) due to LVO,^1–3 and recent evidence has extended the therapeutic time window to 24 h after the last known well.^4,5 Nevertheless, despite successful recanalisation, nearly 50% of patients treated with EVT remain functionally dependent at 90 days.^6,7

The benefit of EVT is highly time-dependent, and most prior studies have emphasised the onset-to-reperfusion time as a critical determinant of outcome.^8,9 Procedure time (PT), defined as the interval from arterial puncture to reperfusion or the last failed attempt, is another key time metric.¹⁰ Procedure time not only reflects thrombus characteristics and procedural complexity but may also exert an effect on clinical outcomes. Although recent studies have highlighted the prognostic significance of PT,^11–15 available evidence remains limited and inconsistent. Moreover, whether the impact of PT differs across clinically important subgroups, such as onset-to-puncture time (OPT), anaesthesia type, baseline infarct core size and occlusion site, has not been systematically examined.

To further investigate this question, we conducted a post hoc, hypothesis-generating pooled analysis of individual participant data from 3 multicentre randomised controlled trials (RCTs). This study aimed to investigate the association between PT and clinical outcomes in AIS patients undergoing EVT. In particular, we sought to determine whether prolonged PT was independently associated with worse prognosis after accounting for established pre-procedural time metrics (such as OPT).

Methods

Study design and population

This post hoc cohort analysis pooled data from 3 multicentre RCTs: RESCUE BT (Endovascular Treatment With versus Without Tirofiban for Patients with Large Vessel Occlusion Stroke; ChiCTR-IOR-17014167), DEVT (Direct Endovascular Thrombectomy vs Combined Intravenous Thrombolysis and Endovascular Thrombectomy for Patients With Acute Large Vessel Occlusion in the Anterior Circulation; ChiCTR-IOR-17013568) and MARVEL (Methylprednisolone as Adjunctive to Endovascular Treatment for Acute Large Vessel Occlusion; ChiCTR2100051729), which have been described in detail previously.^16–18 In brief, all 3 trials prospectively enrolled patients with AIS due to anterior circulation LVO who were eligible for EVT across multiple comprehensive stroke centres in China between 2018 and 2023. Investigators from each trial agreed to participate in the present study, and anonymised individual participant data were pooled. The key characteristics of each trial are summarised in Table S1.

For the current analysis, patients were included if they: (i) were aged ≥18 years, (ii) had anterior circulation LVO confirmed by CTA, MRA or digital subtraction angiography (DSA) and (iii) had available data on PT and 90-day mRS. Patients were excluded if they (i) withdrew informed consent, or (ii) had a pre-stroke mRS score > 2. After all exclusions, 2845 patients were included in the final analysis set (Figure 1).

Figure 1.

Flowchart of patient selection.

Each trial was approved by the institutional review board or ethics committee of the participating centres, and written informed consent was obtained from all patients or their legally authorised representatives. This pooled analysis was conducted in accordance with the Declaration of Helsinki.

Data collection

Demographic characteristics, vascular risk factors (hypertension, diabetes mellitus, hyperlipidaemia, etc.), stroke severity (assessed by the NIHSS score¹⁹), Alberta Stroke Program Early CT Score (ASPECTS),²⁰ occlusion site, stroke aetiology, time metrics (OPT, PT, etc.) and clinical outcomes were collected. If the time of onset was unknown, the last known well time was used as the onset time. Procedure time was defined as the interval from arterial puncture to successful recanalisation (expanded thrombolysis in cerebral infarction [eTICI] score ≥ 2b)²¹ or the last failed attempt, measured in minutes. Anaesthesia was categorised as general anaesthesia vs local anaesthesia. In the pooled datasets, the local-anaesthesia category referred to EVT performed without general anaesthesia, typically under local anaesthesia with or without conscious sedation.

Outcomes measurement

The primary outcome was a favourable functional outcome, defined as an mRS score of 0–2 at 90 days. The mRS ranges from 0 (no symptoms) to 6 (death). Clinical follow-up was conducted at 90 ± 7 days by trained neurologists at each site who were masked to treatment allocation. Assessments were performed in person or, when not feasible, via telephone or video. Subsequently, all outcomes were centrally adjudicated by investigators at the coordinating centre, who were also blinded to the treatment assignments. Secondary outcomes included excellent outcome (mRS score of 0–1), sICH within 48 h confirmed by CT or MRI and all-cause mortality at 90 days. sICH was defined according to the Heidelberg Bleeding Classification²² and was assessed in all 3 parent trials using the same pre-specified definition.

Statistical analysis

Individual participant data from the 3 RCTs were pooled and harmonised before analysis, following a predefined harmonisation workflow to ensure data consistency and minimise manual errors (eMethod 1). Although the parent trials differed in adjunctive treatments and some eligibility criteria, between-trial heterogeneity was addressed through harmonised variable definitions, pre-specified covariate adjustment and mixed-effects modelling. All analyses were performed using IBM SPSS Statistics (Version 25.0, IBM Software, Chicago, Illinois, USA), R version 4.5.0 (R Foundation for Statistical Computing, Vienna, Austria) and GraphPad Prism 8 (GraphPad, La Jolla, CA).

The pre-specified statistical analysis plan is described in eMethod 2 of the Online Supplementary Material. Univariate analyses were performed according to variable type: Student’s t-test or analysis of variance was applied for parametric data, the Mann–Whitney U test or Kruskal–Wallis test for non-parametric data and the χ² test for categorical variables. Potential confounders were identified a priori using directed acyclic graphs (DAGs) (eMethod 2). Because the proportion of missing data for potential confounders was low (Table 1), the primary analyses were conducted using a complete-case approach.

Table 1.

Baseline patient characteristics stratified by procedure time.

	All patients   (n = 2845)	≤60 Min   (n = 1260)	60–120 Min   (n = 978)	>120 Min   (n = 607)	P value
Age, y, median (IQR)	68.0 [58.0, 76.0]	69.0 [59.0, 76.0]	68.0 [58.0, 75.0]	68.0 [58.0, 75.0]	.087
Sex, male, n (%)	1633 (57.4)	678 (53.8)	575 (58.8)	380 (62.6)	.001
Medical history, n (%)
Hypertension	1683 (59.2)	743 (59.0)	558 (57.1)	382 (62.9)	.068
Hyperlipidaemia	679 (23.9)	296 (23.5)	235 (24.0)	148 (24.4)	.905
Diabetes mellitus	566 (19.9)	254 (20.2)	201 (20.6)	111 (18.3)	.521
Atrial fibrillation	1128 (39.6)	561 (44.5)	369 (37.7)	198 (32.6)	<.001
Coronary heart disease	521 (18.3)	215 (17.1)	184 (18.8)	122 (20.1)	.250
TIA	20 (0.7)	7 (0.6)	9 (0.9)	4 (0.7)	.586
Ischaemic stroke	441 (15.5)	187 (14.8)	156 (16.0)	98 (16.1)	.683
Intracranial haemorrhage	35 (1.2)	12 (1.0)	16 (1.6)	7 (1.2)	.340
Smoking	753 (26.5)	295 (23.4)	275 (28.1)	183 (30.1)	.003
Pre-stroke mRS, n (%)					.205
0	2693 (94.7)	1188 (94.3)	928 (94.9)	577 (95.1)
1	117 (4.1)	58 (4.6)	33 (3.4)	26 (4.3)
2	35 (1.2)	14 (1.1)	17 (1.7)	4 (0.7)
Glucose, mmol/L, median (IQR)a	7.0 [6.0, 8.7]	7.0 [5.9, 8.6]	7.0 [6.0, 8.7]	7.0 [6.0, 8.8]	.566
SBP, mmHg, median (IQR)	145.0 [128.0, 160.0]	143.5 [127.0, 160.0]	145.0 [127.2, 160.8]	145.0 [130.0, 164.0]	.120
DBP, mmHg, median (IQR)	84.0 [75.0, 94.0]	83.0 [74.0, 92.2]	83.0 [75.0, 93.0]	85.0 [76.0, 98.0]	.002
Baseline ASPECTS, median (IQR)b	7.0 [5.0, 8.0]	7.0 [5.0, 8.0]	7.0 [5.0, 8.0]	7.0 [5.0, 8.0]	.944
Baseline NIHSS score, median (IQR)	18.0 [15.0, 21.0]	18.0 [15.0, 21.0]	18.0 [15.0, 20.0]	18.0 [14.5, 21.0]	.454
IVT, n (%)	764 (26.9)	362 (28.7)	245 (25.1)	157 (25.9)	.124
Occlusion site, n (%)					<.001
ICA	811 (28.5)	286 (22.7)	285 (29.1)	240 (39.5)
MCA M1	1667 (58.6)	798 (63.3)	564 (57.7)	305 (50.2)
MCA M2	367 (12.9)	176 (14.0)	129 (13.2)	62 (10.2)
ASITN/SIR, n (%)c					.149
0–1	994 (35.1)	458 (36.5)	327 (33.8)	209 (34.6)
2	1061 (37.5)	463 (36.9)	387 (40.0)	211 (34.9)
3–4	773 (27.3)	335 (26.7)	254 (26.2)	184 (30.5)
Tandem occlusion, n (%)d	218 (7.7)	65 (5.2)	71 (7.3)	82 (13.5)	<.001
TOAST aetiology, n (%)					<.001
LAA	1129 (39.7)	385 (30.6)	417 (42.6)	327 (53.9)
ce	1344 (47.2)	701 (55.6)	446 (45.6)	197 (32.5)
Others and unknown	372 (13.1)	174 (13.8)	115 (11.8)	83 (13.7)
Local anaesthesia, n (%)	2001 (70.4)	940 (74.6)	689 (70.4)	372 (61.4)	<.001
First choice of EVT, n (%)e					.002
Stent thrombectomy	664 (23.3)	274 (21.7)	244 (24.9)	146 (24.1)
Aspiration	1241 (43.6)	608 (48.3)	394 (40.3)	239 (39.4)
Aspiration and stent retriever	626 (22.0)	254 (20.2)	228 (23.3)	144 (23.8)
Others	313 (11.0)	124 (9.8)	112 (11.5)	77 (12.7)
Total passes, median (IQR)	1.0 [1.0, 2.0]	1.0 [1.0, 2.0]	2.0 [1.0, 3.0]	2.0 [1.0, 3.0]	<.001
FPE, n (%)	1122 (39.4)	679 (53.9)	312 (31.9)	131 (21.6)	<.001
Onset to puncture time, min, median (IQR)	341.0 [226.0, 584.0]	330.0 [224.0, 555.0]	340.0 [226.2, 590.0]	380.0 [235.0, 627.5]	.052
Reperfusion, n (%)f	2559 (90.1)	1197 (95.1)	877 (89.8)	485 (80.3)	<.001
Clinical outcomes
mRS 0–2, n (%)	1275 (44.8)	662 (52.5)	425 (43.5)	188 (31.0)	<.001
mRS 0–1, n (%)	835 (29.3)	449 (35.6)	270 (27.6)	116 (19.1)	<.001
sICH, n (%)g	258 (9.2)	87 (7.0)	95 (9.8)	76 (12.9)	<.001
Mortality, n (%)	638 (22.4)	212 (16.8)	235 (24.0)	191 (31.5)	<.001

Abbreviations: ASITN/SIR, American Society of Interventional and Therapeutic Neuroradiology/Society of Interventional Radiology collateral score; ASPECTS, Alberta Stroke Program Early CT Score; ce, cardioembolism; DBP, diastolic blood pressure; FPE, first pass effect; ICA, internal carotid artery; IVT, intravenous thrombolysis; LAA, large-artery atherosclerosis; MCA, middle cerebral artery; SBP, systolic blood pressure; TOAST, Trial of ORG 10172 in Acute Stroke Treatment.

^aMissing data for glucose were 278 (9.8% overall): 131(10.4%) in the PT ≤ 60 -min group, 93 (9.5%) in the PT 60- to 120-min group and 54 (8.9%) in the PT > 120-min group.

^bMissing data for baseline ASPECTS were 6 (0.2% overall): 3 (0.2) in the PT ≤ 60-min group, 1 (0.1%) in the PT 60- to 120-min group and 2 (0.3%) in the PT > 120-min group.

^cMissing data for ASITN/SIR were 17 (0.6% overall): 4 (0.3%) in the PT ≤ 60-min group, 10 (0.6%) in the PT 60- to 120-min group and 3 (0.5%) in the PT > 120-min group.

^dMissing data for tandem occlusion were 1 (<0.1% overall): 0 (0.0%) in the PT ≤ 60-min group, 0 (0.0%) in the PT 60- to 120-min group and 1 (0.2%) in the PT > 120-min group.

^eMissing data for first choice of EVT were 1 (<0.1% overall): 0 (0.0%) in the PT ≤ 60-min group, 0 (0.0%) in the PT 60- to 120-min group and 1 (0.2%) in the PT > 120-min group.

^fMissing data for reperfusion were 5 (0.2% overall): 1 (0.1%) in the PT ≤ 60-min group, 1 (0.1%) in the PT 60- to 120-min group and 3 (0.5%) in the PT > 120-min group.

^gMissing data for sICH were 32 (1.1% overall): 9 (0.7%) in the PT ≤ 60-min group, 6 (0.7%) in the PT 60- to 120-min group and 17 (2.8%) in the PT > 120-min group.

The association between PT and clinical outcomes was evaluated using logistic regression, treating PT either as a continuous variable (per 10-min increase) or as pre-specified categories (≤60, 60–120 and > 120 min). Covariates for adjustment were pre-specified using DAGs and clinical judgement (eMethod 2). For each endpoint, 3 hierarchical models were constructed: model 1 adjusted for core confounders defined by the DAGs (intravenous thrombolysis [IVT], occlusion site, tandem occlusion, Trial of ORG 10172 in Acute Stroke Treatment (TOAST) aetiology and anaesthesia type); model 2 additionally adjusted for age, diabetes mellitus, pre-stroke mRS, diastolic blood pressure, baseline ASPECTS, baseline NIHSS and OPT and model 3 was a mixed-effects logistic regression model with the same fixed effects as model 2 and a random intercept for stroke centre. Estimated effects are reported as odds ratios (ORs) and absolute risk differences (RDs), with corresponding 95% CIs.

Cumulative distribution curves and fitted curves of favourable outcome rates with PT were generated using GraphPad Prism 8. Both linear and non-linear best-fit models were examined, and R² values were used to assess the correlation and model fit. Restricted cubic spline (RCS) analysis with 3 knots at the 10th, 50th and 90th percentiles of PT was used to explore the association between continuous PT (minutes) and clinical outcomes. The Wald χ² test was used to assess the non-linearity.

Subgroup analyses

Subgroup analyses were pre-planned to explore potential effect modification of PT on the primary endpoint (mRS 0–2) and to better characterise patient groups most vulnerable to intra-procedural delays. These analyses are exploratory and will not be used to redefine the primary conclusions (detailed in eMethod2).

Sensitivity analysis

To assess the robustness of the findings, we conducted 4 sensitivity analyses. First, stabilised inverse probability of treatment weights (IPTW) based on the covariates in model 2 were applied, and a doubly robust estimation was performed by combining these weights with outcome regression models (eMethod 3). Second, the analyses were repeated in the subset of patients with successful reperfusion (eTICI ≥ 2b). Third, extreme PT values (>240 min) were excluded to reduce the influence of the outliers. Fourth, multiple imputation by chained equations (MICE) (10 imputed datasets) was used for variables with missing data; logistic regression models were fitted in each imputed dataset and estimates were combined using Rubin’s rules (eMethod 4). In addition, an exploratory descriptive learning-curve analysis of PT was performed for visualisation purposes.

Causal mediation analysis

Finally, an exploratory causal mediation analysis was performed to decompose the total effect of PT on favourable outcome into direct and indirect effects mediated by procedural complexity (eg, total passes and the first pass effect [FPE]), using the mediation package in R with 1000 bootstrap resamples (eMethod 2).

Results

Patient characteristics

Among the 2872 screened patients, 2845 with anterior circulation LVO who were treated with EVT were included in the present analysis (Figure 1). The median age was 68 years (IQR, 58–76), 57.4% were male and the baseline NIHSS score was 18 (IQR, 15–21). The median PT was 69.0 min (IQR, 40.0–110.0). In an exploratory descriptive analysis, PT appeared relatively stable across cumulative case experience, with no significant difference between the early and late trial periods (P = .251; Figure S1).

Compared with patients whose PT was ≤ 60 min, those with PT of 60–120 min or > 120 min were more likely to be male (P = .001), more frequently had internal carotid artery (ICA) occlusion (P < .001) and tandem occlusions (P < .001). Regarding stroke aetiology, large-artery atherosclerosis (LAA) was more common in patients with prolonged PT, whereas cardioembolism predominated in the ≤ 60-min group (P < .001). Conversely, the use of local anaesthesia decreased with increasing PT (P < .001). Procedural characteristics also differed across groups: the > 120-min group required more passes, had lower rates of FPE (P < .001), and achieved reperfusion less frequently (P < .001). Clinically, these patients experienced worse outcomes, including lower rates of mRS 0–2 (P < .001) and mRS 0–1 (P < .001) at 90 days, as well as higher rates of sICH (P < .001) and mortality (P < .001). The detailed baseline and procedural characteristics stratified by PT are presented in Table 1.

Association between continuous PT and outcomes

When modelled as a continuous variable, longer PT was consistently associated with worse clinical outcomes (Table 2). For a favourable outcome (mRS 0–2), each 10-min increase in PT was associated with approximately 7% lower odds [adjusted OR (aOR) 0.928; 95% CI, 0.912–0.944], corresponding to an adjusted RD (aRD) of −1.52% (95% CI, −1.85% to −1.2%). Similar associations were observed for excellent outcome (mRS 0–1; aOR 0.924; 95% CI, 0.906–0.942; aRD −1.37%; 95% CI, −1.69% to −1.05%). In contrast, PT was positively associated with sICH (aOR 1.028; 95% CI, 1.007–1.049; aRD +0.22%; 95% CI, 0.05%–0.38%) and mortality, with each 10-min increment associated with higher odds of death (aOR 1.055; 95% CI, 1.037–1.073; aRD +0.77%; 95% CI, 0.53%–1.01%). These associations remained similar in the mixed-effects models accounting for centre-level clustering (model 3).

Table 2.

Association of continuous procedure time (per 10 min) with outcomes.

Outcome	Metric	Unadjusted (95% CI)	Model 1 a  (95% CI)	Model 2 b  (95% CI)	Model 3 c  (95% CI)	P value (Model 3)
mRS 0–2	OR	0.938 (0.924–0.952)	0.939 (0.925–0.953)	0.928 (0.912–0.944)	0.930 (0.914–0.947)	<.001
	RD	−1.54% (−1.880%–−1.21%)	−1.50% (−1.84%–−1.15%)	−1.52% (−1.85%–−1.20%)	NA	NA
mRS 0–1	OR	0.936 (0.920–0.952)	0.936 (0.920–0.953)	0.924 (0.906–0.942)	0.926 (0.909–0.945)	<.001
	RD	−1.35% (−1.68%–−1.01%)	−1.32% (−1.67%–−0.98%)	−1.37% (−1.69%–−1.05%)	NA	NA
sICH	OR	1.032 (1.012–1.053)	1.030 (1.009–1.051)	1.028 (1.007–1.049)	1.028 (1.007–1.049)	.010
	RD	0.26% (0.10%–0.42%)	0.24% (0.07%–0.41%)	0.22% (0.05%–0.38%)	NA	NA
Mortality	OR	1.050 (1.034–1.065)	1.049 (1.033–1.065)	1.055 (1.037–1.073)	1.054 (1.036–1.073)	<.001
	RD	0.81% (0.56%–1.05%)	0.78% (0.53%–1.03%)	0.77% (0.53%–1.01%)	NA	NA

Abbreviations: ASPECTS, Acute Stroke Program Early CT Score; DBP, diastolic blood pressure; IVT, intravenous thrombolysis; OR, odds ratio; RD, risk difference; TOAST, Trial of ORG 10172 in Acute Stroke Treatment.

^aModel 1 was adjusted for IVT, occlusion site, tandem occlusion, TOAST aetiology and local anaesthesia.

^bModel 2 adjusted for confounders in model 1 plus clinical important variables including age, diabetes mellitus, pre-stroke mRS, DBP, baseline ASPECTS, baseline NIHSS score and onset to puncture time.

^cModel 3 was a mixed-effect model adjusting for same covariates in model 2 and including stroke centre as random effect.

Second, RCS analyses (Figure S2) showed no evidence of significant non-linearity for either favourable outcome (P = .886) or mortality (P = .095). As shown in Figure 2A, the proportion of patients with a favourable outcome declined progressively with increasing PT, whereas the proportion with an unfavourable outcome increased, with unfavourable outcomes predominating once PT exceeded approximately 60 min. The cumulative distribution of favourable outcome and mortality by PT are presented in Figure 2B. The cumulative rate of favourable outcomes rose steeply within the first 120 min and then plateaued, ultimately reaching 44.8%. In contrast, cumulative mortality increased more gradually, eventually reaching 22.4%. Linear regression models confirmed strong correlations between PT and mortality (R² = 0.86, P < .001, Figure 2C).

Figure 2.

Association between PT and clinical outcomes. (A) Proportions of patients with favourable outcome and unfavourable outcome based on PT 20-min increments. Longer PT was associated with a progressive decline in favourable outcome and a corresponding increase in unfavourable outcome. (B) Cumulative percentage curves of favourable outcome and mortality according to PT. (C) Linear regression demonstrated a positive correlation between PT and mortality (R² = 0.86, P < .001). Abbreviation: PT = procedure time.

Association of PT categories with outcomes

When PT was analysed categorically, the rate of mRS 0–2 declined stepwise with longer PT, from 52.5% in the ≤ 60-min group to 43.5% in the 60- to 120-min group and 31.0% in the > 120-min group (P < 0.001; Table 3). In fully adjusted analyses (model 2), patients with a PT of 60-120 min and > 120 min had lower odds of mRS 0–2 compared with those with PT ≤ 60 min (aOR = 0.658; 95% CI, 0.546–0.792 and aOR = 0.355; 95% CI, 0.281–0.448, respectively). These associations remained consistent in the mixed-effects model accounting for centre (model 3, Table 3). A similar gradient was observed for mRS 0–1, with aOR of 0.645 (95% CI, 0.528–0.789) for the 60- to 120-min group and 0.365 (95% CI, 0.280–0.476) for the > 120 min group (Table 3). The stacked distribution of mRS scores (Figure 3) further illustrated these trends: patients with shorter PT were more likely to achieve favourable outcome, whereas those with longer PT had a higher likelihood of disability and death.

Table 3.

Association of categorical procedure time with outcomes.

Outcome	Procedure time	Frequencies	χ 2	P value	Estimate (c) OR (95% CI)
					Model 1 a	Model 2 b	Model 3 c
mRS 0–2			78.164	<.001
	≤60 min	662/1260 (52.5%)			Reference	Reference	Reference
	60–120 min	425/978 (43.5%)			0.701 (0.591–0.832)	0.658 (0.546–0.792)	0.680 (0.562–0.823)
	>120 min	188/607 (31.0%)			0.420 (0.340–0.520)	0.355 (0.281–0.448)	0.368 (0.289–0.469)
mRS 0–1			56.128	<.001
	≤60 min	449/1260 (35.6%)			Reference	Reference	Reference
	60–120 min	270/978 (27.6%)			0.693 (0.576–0.833)	0.645 (0.528–0.789)	0.666 (0.543–0.816)
	>120 min	116/607 (19.1%)			0.438 (0.344–0.558)	0.365 (0.280–0.476)	0.380 (0.290–0.497)
sICH			17.552	<.001
	≤60 min	87/1251 (7.0%)			Reference	Reference	Reference
	60–120 min	95/972 (9.8%)			1.445 (1.062–1.965)	1.433 (1.048–1.958)	1.433 (1.048–1.958)
	>120 min	76/590 (12.9%)			1.894 (1.347–2.663)	1.837 (1.299–2.597)	1.837 (1.299–2.597)
Mortality			52.679	<.001
	≤60 min	212/1260 (16.8%)			Reference	Reference	Reference
	60–120 min	235/978 (24.0%)			1.562 (1.263–1.931)	1.648 (1.315–2.067)	1.635 (1.301–2.056)
	>120 min	191/607 (31.5%)			2.193 (1.728–2.784)	2.422 (1.873–3.132)	2.398 (1.846–3.115)

Abbreviations: ASPECTS, Acute Stroke Program Early CT Score; DBP, diastolic blood pressure; IVT, intravenous thrombolysis; OR, odds ratio; TOAST, Trial of ORG 10172 in Acute Stroke Treatment.

^aModel 1 was adjusted for IVT, occlusion site, tandem occlusion, TOAST aetiology and local anaesthesia.

^bModel 2 adjusted for confounders in model 1 plus clinical important variables including age, diabetes mellitus, pre-stroke mRS, DBP, baseline ASPECTS, baseline NIHSS score and onset to puncture time.

^cModel 3 was a mixed-effect model adjusting for same covariates in model 2 and including stroke centre as random effect.

Figure 3.

Distribution of 90-day mRS scores by PT. Stacked bar plots show the proportion of patients in each mRS category (0–6) at 90 days across 3 PT groups (≤ 60 min, 60–120 min and > 120 min). Longer PT was associated with a stepwise shift towards higher disability and death, with fewer patients achieving favourable outcome (mRS 0–2; P < .001). Abbreviation: PT = procedure time.

The rate of sICH increased with longer PT, from 7.0% in the ≤ 60-min group to 9.8% in the 60- to 120-min group and 12.9% in the > 120-min group (P < .001). After adjustment, both intermediate and prolonged PT were associated with higher risks of sICH (model 2: aOR 1.433; 95% CI, 1.048–1.958, and 1.837, 95% CI, 1.299–2.597, respectively), with virtually identical estimates in model 3. Mortality also increased progressively across PT categories (16.8%, 24.0% and 31.5%; P < .001). In fully adjusted analyses, patients with PT of 60–120 min and > 120 min had substantially higher odds of death compared with those with a PT ≤ 60 min (model 2: aOR 1.648; 95% CI, 1.315–2.067 and 2.422; 95% CI 1.873–3.132 and model 3: aOR 1.635; 95% CI, 1.301–2.056 and 2.398; 95% CI, 1.846–3.115). The multivariable model for mRS 0–2 showed good overall calibration and maintained stable performance across PT strata (Figure S3).

Causal mediation analysis

In exploratory causal mediation analyses, procedural complexity explained only a modest proportion of the association between PT and favourable outcome (Table S2). For each 10-min increase in PT, the average direct effect on mRS 0–2 remained significant, whereas the indirect effects mediated by total passes and FPE accounted for approximately 12.9% and 11.4% of the total effect, respectively. This suggests that most of the detrimental impact of prolonged PT is not fully captured by these measures of procedural effort alone.

Subgroup analysis

In pre-specified subgroup analyses, the detrimental association between longer PT and lower odds of favourable outcome was directionally consistent across all subgroups (Figure 4). There was no evidence of heterogeneity by age, baseline NIHSS, baseline ASPECTS, IVT or occlusion site (all P for interaction > .05). However, significant interactions were observed for OPT and type of anaesthesia (P  _{for interaction} = .009 and .003, respectively).

Figure 4.

Subgroup analysis of the association between procedure time (per 10-min) and favourable outcome at 90 days. Abbreviations: aOR = adjusted odds ratio; ASPECTS = Alberta Stroke Program Early CT Score; IVT = intravenous thrombolysis; ICA = internal carotid artery; MCA = middle cerebral artery; OPT = onset-to-puncture time.

When stratified by OPT, the adverse association between longer PT and favourable outcome was stronger in earlier presenters (Table S3; Figure S4). For patients with OPT ≤ 6 h, both intermediate and prolonged PT were linked to substantially lower odds of mRS 0–2 compared with PT ≤ 60 min (aOR 0.567 and 0.302, respectively), with a clear stepwise decline in favourable outcomes as PT increased. For OPT > 6 h, the effect of PT was attenuated: PT 60–120 min was not clearly different from ≤ 60 min (aOR 0.771; 95% CI, 0.587–1.013), whereas PT > 120 min remained associated with reduced odds of independence (aOR 0.422; 95% CI, 0.304–0.585). Cumulative curves (Figure S4C) showed consistently higher favourable-outcome rates across the PT spectrum in the OPT ≤ 6-h group.

When stratified by anaesthesia type, prolonged PT was associated with worse outcomes under both general and local anaesthesia, but with different patterns (Table S4; Figure S5). Under general anaesthesia, even moderate prolongation (60–120 min) was linked to a lower likelihood of achieving mRS 0–2 compared with PT ≤ 60 min (aOR 0.552; 95% CI, 0.443–0.688), with a further reduction when PT exceeded 120 min (aOR 0.287; 95% CI, 0.214–0.385). In contrast, among patients treated under local anaesthesia, PT 60–120 min was not clearly different from ≤ 60 min (aOR 1.028; 95% CI, 0.721–1.466), whereas PT > 120 min was associated with reduced odds of favourable outcome (aOR 0.572; 95% CI, 0.384–0.851). Cumulative curves (Figure S5C) showed slightly higher cumulative rates of favourable outcome with local compared with general anaesthesia across the PT range.

When stratified by baseline ASPECTS, prolonged PT predicted worse favourable outcomes in both strata but with a steeper gradient in patients with ASPECTS ≤ 5, although the formal interaction test was not significant (Table S5; Figure S6). In the low-ASPECTS group, even PT of 60–120 min was associated with reduced odds of mRS 0–2. In patients with ASPECTS > 5, the association was attenuated but remained evident, with both PT 60–120 min and PT > 120 min associated with lower odds of mRS 0–2; the magnitude was greater when PT exceeded 120 min. Cumulative curves showed consistently lower favourable-outcome rates across the PT range in patients with ASPECTS ≤ 5.

When stratified by occlusion site, the harmful association between longer PT and favourable outcome was evident for both ICA and M1 occlusions, with a clear stepwise decrease in the odds of mRS 0–2 as PT increased (Table S6). For M2 occlusions, the point estimates similarly favoured shorter PT, although categorical comparisons did not reach statistical significance. Stacked mRS distributions (Figure S7) showed a consistent shift towards higher disability and death with prolonged PT across all occlusion sites.

Sensitivity analysis

In 4 pre-specified sensitivity analyses, the associations between longer PT and clinical outcomes remained largely unchanged (Table S7). For favourable outcome, the aOR per 10-min delay ranged from 0.926 to 0.944 across the IPTW, eTICI ≥ 2b, outlier-excluded and MICE models, all closely matching the main analysis. Similar robustness was observed for excellent outcome (mRS 0–1), sICH and mortality, supporting the stability of our findings.

Exploratory factors associated with longer PT

In exploratory mixed-effects analyses of factors associated with longer PT, ICA occlusion, tandem occlusion, LAA, general anaesthesia and stent-based first-choice EVT strategies were independently associated with longer PT, whereas IVT was associated with shorter PT (Table S8). Compared with M2 occlusion, ICA occlusion was associated with 28.2% longer PT (95% CI, 18.2%–39.0%), and LAA was associated with 34.6% longer PT (95% CI, 27.1%–42.5%) compared with ce.

Discussion

In this post hoc pooled individual-patient analysis of 3 RCTs, longer PT was consistently associated with worse functional and safety outcomes after EVT for anterior-circulation LVO stroke. Each incremental increase in PT was associated with lower odds of functional independence and higher risks of sICH and death.

While the overall association between longer PT and worse outcomes has been reported previously, the present study extends this literature by using harmonised individual-patient data from RCTs and by further exploring the extent to which this association is explained by conventional procedural complexity metrics. Previous studies have consistently shown that prolonged PT is associated with poor outcomes after EVT.¹⁵ Some studies have suggested a plateau of benefits beyond 60 min, whereas Huang et al.²³ reported a 43% higher risk of poor outcome and a 39% higher risk of sICH for every additional 30 min, and Hassan et al.²⁴ found that each 10-min delay reduced the odds of favourable outcome by ~ 11% even after adjustment. More recently, a large multicentre registry of nearly 9000 patients confirmed the “60-min golden hour,” showing a doubling of poor outcomes once PT exceeded 1 h, with ~ 10% higher odds of disability per 10-min delay.²⁵ In contrast, our multicentre randomised dataset did not show strong evidence of a clear threshold effect, and the observed association was broadly compatible with an approximately linear pattern over the range studied. These associations were robust across multiple adjustment strategies, including mixed-effects modelling, inverse probability of treatment weighting, restriction to patients with successful reperfusion, exclusion of extreme values and multiple imputation.

Our exploratory causal mediation analysis adds nuance to those findings. Only about 11%–13% of the association between PT and favourable outcome was explained by procedural complexity, as captured by total passes and FPE. This suggests that prolonged PT may reflect not only more difficult thrombectomy but also other factors—such as longer exposure to hypoperfusion that may independently worsen outcomes.²⁶ This interpretation is further supported by our exploratory analysis of factors associated with longer PT. Internal carotid artery occlusion, tandem occlusion, large-artery atherosclerosis, general anaesthesia and stent-based first-choice EVT strategies were independently associated with longer PT, whereas IVT was associated with shorter PT. Although these exploratory findings do not establish causality, they suggest that PT reflects a combination of occlusion anatomy, stroke aetiology, anaesthesia strategy and procedural approach, rather than a single workflow metric.

The detrimental impact of procedural delay was most evident in patients treated within 6 h, where longer PT was more strongly associated with a low likelihood of good outcome, whereas the association was weaker but still present in late presenters. Consistent with the findings of Sadeler et al.,¹³ who analysed a pooled, predominantly late-window (6–24 h) cohort and similarly showed that longer intra-hospital delays, including PT, remained harmful even when patients were selected with advanced imaging. In contrast, Song et al.¹¹ reported no such association beyond 6 h, which may be attributable to their relatively small sample size. Early-presenting patients likely have substantial salvageable brain tissue, making them highly sensitive to intra-procedural delays. In contrast, late-window patients are often selected for favourable collaterals and slower infarct growth, which may grant a degree of resilience to delays.^27,28

We observed significant heterogeneity by anaesthesia type, but the pattern differed across anaesthesia strata. Under general anaesthesia, both intermediate and prolonged PT were associated with worse outcomes, whereas under local anaesthesia the adverse association was mainly evident when PT exceeded 120 min. These patterns may reflect differences in case selection, haemodynamic stability or workflow efficiency between anaesthesia strategies.^29,30 However, our study was not designed to determine the optimal anaesthetic approach, and minimising PT is likely important regardless of anaesthesia type.

Recent trials have shown that EVT can still benefit patients with large ischaemic cores (ASPECTS ≤ 5).^31–35 In our cohort, however, outcomes in this group appeared to decline more steeply with increasing PT than in patients with ASPECTS > 5, consistent with the findings of Sadeler et al.¹³ This pattern likely reflects greater baseline injury and poorer collateral circulation in large-core strokes, where earlier presenters may still have salvageable but highly vulnerable tissue, making these patients particularly vulnerable to additional intra-procedural delay. The occlusion site also appeared to influence the effect of PT. In our cohort, the association between longer PT and poor outcome appeared stronger for proximal ICA/M1 occlusions than for M2 occlusions, although formal interaction by occlusion site was not observed. Nevertheless, even distal occlusions showed a trend towards higher disability with longer PT, highlighting that rapid reperfusion is advantageous across all occlusion sites. This pattern is consistent with the findings of Ash et al.,²⁵ which included nearly 9000 real-world patients with both anterior and posterior circulation strokes and demonstrated a “60-min golden hour” across proximal and distal occlusions as well as small and large cores. However, based on our data, these subgroup findings should be interpreted with caution because of the relatively small sample sizes within some strata.

Taken together, our findings support PT as a clinically relevant and potentially modifiable process metric after EVT, but also suggest that PT represents more than procedure speed alone. Efforts to reduce PT may need to address not only workflow efficiency but also the anatomical, aetiological and procedural features that predispose to prolonged thrombectomy. Future studies with more granular imaging and device-level data will be needed to clarify which components of this broader PT-related burden are most amenable to intervention.

Limitations

This study has several limitations. First, it is a post hoc analysis of randomised trial data and is, therefore, observational in nature, with the attendant risk of residual confounding despite multivariable adjustment. In addition, because the parent trials differed in adjunctive treatments and some eligibility criteria, residual confounding related to between-trial heterogeneity cannot be fully excluded. Second, PT is influenced by multiple factors—including clot characteristics, cervical tortuosity, extracranial carotid disease, the use of stenting/angioplasty, blood pressure management, intraprocedural hypotension, centre volume and operator experience—several of which were not fully captured in our dataset and may have affected the observed associations. In addition, the exploratory analyses of factors associated with longer PT were limited to variables uniformly available across the pooled datasets and could not account for more granular determinants such as clot length, vessel tortuosity or device-level procedural details. Third, all trials were conducted at high-volume stroke centres in China, where practice patterns, patient characteristics and health-system structures may differ from those in other regions; therefore, the generalisability of our findings to Western populations, lower-volume centres or other healthcare systems may be limited.

Conclusion

In this pooled analysis, shorter PT during EVT for anterior circulation LVO stroke was associated with higher rates of favourable outcome and lower risks of sICH and mortality. These findings support PT as a clinically relevant process metric after EVT and highlight the need to better understand and address the anatomical, etiologic and procedural factors associated with prolonged thrombectomy.

Author contributions

Xuanyu Chen (Data curation, Formal analysis, Writing—original draft [equal]), Xu Xu (Data curation, Formal analysis, Writing—original draft [equal]), Jinfu Ma (Data curation, Formal analysis, Writing—original draft [equal]), Haoxuan Zhu (Data curation, Formal analysis, Writing—original draft [equal]), Xiaolei Shi (Investigation [equal]), Shihai Yang (Investigation [equal]), Zhixi Wang (Investigation [equal]), Mingyang Chen (Investigation [equal]), Yihui Yang (Investigation [equal]), Yuhan Fan (Investigation [equal]), Binghan Wang (Investigation [equal]), Guojian Liu (Investigation [equal]), Linyu Li (Investigation [equal]), Zhenxuan Tian (Investigation [equal]), Boyu Chen (Investigation [equal]), Chawen Ding (Investigation [equal]), Dahong Yang (Investigation [equal]), Wenzhe Sun (Investigation [equal]), Gaoming Li (Investigation [equal]), Lilan Wang (Investigation [equal]), Shitao Fan (Investigation [equal]), Chengsong Yue (Investigation [equal]), Nizhen Yu (Investigation [equal]), Jie Yang (Investigation [equal]), Zhuang Li (Investigation [equal]), Lingyu Zhang (Investigation [equal]), Wenjie Zi (Conceptualisation, Funding acquisition, Writing—review & editing [equal]), Kunxin Lin (Conceptualisation, Funding acquisition, Writing—review & editing [equal])

Conflicts of interest

The authors declare that they have no competing interests.

Funding

This study was supported by the National Science Fund for Distinguished Young Scholars of China (No. 82425021), the Noncommunicable Chronic Diseases-National Science and Technology Major Project (No. 2024ZD0527905), the National Natural Science Foundation of China (No. 82271349) and the Key Laboratory of Geriatric Cardiovascular and Cerebrovascular Disease Research, Ministry of Education of China (KFKT202506).

Data availability

The data supporting the findings of this study are available from the corresponding author upon reasonable request.

Ethical approval

The trial protocol was approved by medical ethics committee of the Second Affiliated Hospital of the Army Medical University and all participating centres.

Informed consent

All enrolled patients or their legally authorised representatives provided written informed consent before enrolment.

Trial registration

Chinese Clinical Trial Registry Identifier: ChiCTR-IOR-17014167, ChiCTR-IOR-17013568 and ChiCTR2100051729.