Impact of Severe Prestroke Disability on Outcomes After Mechanical Thrombectomy: A Multicenter Analysis

Abstract

BACKGROUND:

Patients with severe prestroke disability (PSD) remain underrepresented in mechanical thrombectomy studies, despite their growing relevance in aging populations. This study used data from the German Stroke Registry-Endovascular Treatment to evaluate functional recovery, mortality, and poststroke care outcomes in this high-risk population.

METHODS:

We analyzed 9456 mechanical thrombectomy–treated patients with stroke from the German Stroke Registry-Endovascular Treatment (2015–2021), categorized by premorbid modified Rankin Scale (mRS): no PSD (mRS score, 0–1), moderate PSD (mPSD; mRS score, 2–3), and severe PSD (sPSD; mRS score, 4–5). Favorable outcomes were defined as an mRS score of 0 to 2 or return to baseline. Logistic regression adjusted for age, National Institutes of Health Stroke Scale, intravenous thrombolysis, reperfusion success, and sex was used to predict outcomes. A neural network subsequently explored feature importance.

RESULTS:

Among 9456 patients, 7387 had no PSD, 1648 mPSD, and 421 sPSD. Unadjusted 90-day outcomes showed increasing mortality with PSD severity and fewer favorable outcomes in both PSD groups. At 90 days, favorable outcomes occurred in 3020 patients without PSD (40.9%), 232 with mPSD (14.1%), and 85 with sPSD (20.2%). After adjustment, only mPSD was associated with lower odds of favorable outcomes, while both mPSD and sPSD remained independent predictors of higher mortality. Complication rates were similar across groups, except for higher vasospasm in patients without PSD. Including rebalanced sPSD samples in predictive models resulted in minor performance improvements but notable shifts in feature importance, with age and Alberta Stroke Program Early Computed Tomography Score emerging as key predictors, National Institutes of Health Stroke Scale decreasing in relevance, and factors such as local anesthesia and occlusion location becoming more prominent.

CONCLUSIONS:

Despite higher mortality, approximately a fifth of patients with PSD achieved favorable outcomes, suggesting that this group should not be routinely excluded from mechanical thrombectomy. Further studies should refine patient selection criteria and outcome definitions for this vulnerable population.

REGISTRATION:

URL: https://www.clinicaltrials.gov; Unique identifier: NCT03356392.

CLINICAL PERSPECTIVE.

What Is New?

• This large multicenter registry analysis is among the first to comprehensively evaluate outcomes after mechanical thrombectomy in patients with prestroke disability, particularly those with severe disability (modified Rankin Scale score, 4–5), a group largely excluded from previous trials.

What Are the Clinical Implications?

• Despite higher mortality, approximately one-fifth of patients with prestroke disability achieved favorable outcomes, and complication rates were not increased, indicating that mechanical thrombectomy can be safe and effective in selected individuals.

• These findings highlight the need for individualized, patient-centered treatment decisions rather than categorical exclusion based on prestroke modified Rankin Scale and call for refined outcome definitions and selection criteria for this vulnerable population.

Stroke is a leading cause of disability and mortality worldwide, imposing a significant burden on health care systems and societies. With an aging population, there is a growing proportion of patients with stroke presenting with preexisting functional disabilities (PSD), further complicating the management of acute stroke.^1,2

Functional disability after stroke is commonly assessed using the modified Rankin Scale (mRS), a clinician-reported, 7-level classification of global functional impairment.³ The mRS has been widely utilized in clinical trials not only to define primary outcomes after stroke but also to assess prestroke functional disability, which has often served as an exclusion criterion for many of these trials.

Mechanical thrombectomy (MT), in conjunction with intravenous thrombolysis (IVT), has revolutionized the treatment of acute ischemic stroke, offering significant improvements in functional outcomes for patients with large vessel occlusion. MT is now established as the standard of care.^4–6 Approximately 30% of patients with stroke have a preexisting disability,^7,8 yet most early trials, apart from the DEFUSE 3 trial (Endovascular Therapy Following Imaging Evaluation for Ischemic Stroke 3), excluded those with a prestroke mRS score ≥2, resulting in limited evidence regarding the effectiveness of MT in this subgroup.^9–13

This gap was recently underscored by the American Heart Association and American Stroke Association Scientific Statement, which highlights the routine exclusion of patients with premorbid disability or dementia from reperfusion trials, despite observational evidence suggesting comparable safety and potential to retain prestroke function. The statement calls for pragmatic, patient-centered decision-making and greater inclusion of this population in future research.¹⁴

Most observational studies evaluating MT in patients with PSD have focused on those with mild impairments or involved small sample sizes. Notably, only a few studies have included patients with more severe PSD (sPSD; mRS score, 4–5), leaving a significant gap in understanding how this group responds to MT. Prestroke functional status plays a critical role in shaping stroke outcomes. Individuals with evidence of PSD, as assessed by the mRS, are known to have significantly worse poststroke outcomes. These include higher risks of adverse events, longer stays in stroke units, and the need for elevated levels of care upon discharge.^15,16 Patients with PSD face a higher risk of mortality after MT, and positive poststroke recovery outcomes are achieved less frequently in those with premorbid mRS scores of 3 or 4 compared with patients with scores of 0 to 2.^17–19 Observational studies suggest that patients with mild to moderate PSD (mPSD) can achieve favorable functional outcomes, with no clear association between prestroke mRS and accumulated disability. However, some evidence indicates a tendency toward higher mortality in patients with preexisting disability, while data on other outcomes remain limited and inconclusive.^18,20–22

Despite growing evidence on the outcomes of MT in patients with PSD, significant gaps remain, particularly for those with sPSD (mRS score, 4–5). This study aims to address these gaps by evaluating the impact of sPSD on post-MT outcomes through a multicenter analysis, with a focus on functional recovery, mortality, and the need for poststroke care.

Methods

The study was conducted in accordance with the Declaration of Helsinki and was centrally approved by the institutional review board of the Ludwig Maximilian University of Munich and institutional review boards according to local regulations.

Patient or proxy consent for registry participation was obtained, as previously described.³⁴

The data set utilized in this study was derived from the German Stroke Registry-Endovascular Treatment (NCT03356392), which is a prospective, multicenter registry that encompasses patients diagnosed with acute ischemic stroke who have large vessel occlusions and were treated via MT. Patient enrollment occurred from 2015 to 2021 across 27 hospitals throughout Germany, with details of the registry having been described in prior publications.^23,24 The research adhered to the principles outlined in the Declaration of Helsinki and received overarching ethical approval from the institutional review board of the Ludwig Maximilian University of Munich (689-15), alongside approvals from local institutional review boards in compliance with regional guidelines.

Cases were excluded from the analysis due to missing baseline or 90-day mRS scores or due to implausible mRS changes (meaning an improvement of >1 point at 90 days compared with baseline or discharge scores that were better than baseline). Patients were categorized into 3 groups based on the severity of PSD, as determined by their premorbid mRS scores. The first group, representing no PSD (nPSD), included patients with premorbid mRS scores of 0 or 1, indicating no symptoms or no significant disability. The second group, mPSD, consisted of patients with premorbid mRS scores of 2 or 3, reflecting moderate disability but preserved independent ambulation. The third group, sPSD, included patients with premorbid mRS scores of 4 or 5, corresponding to severe disability requiring assistance with daily activities. The patient inclusion process is summarized in Figure 1. Group comparisons were conducted to evaluate differences in baseline characteristics, received treatments, in-hospital complications, and clinical outcomes, with a focus on functional status at 90-day follow-up and potential factors associated with poorer outcomes. Symptomatic intracranial hemorrhage was evaluated retrospectively according to the ECASS (European Cooperative Acute Stroke Study) II definition.²⁵ The distribution of the mRS groups across different centers was largely balanced, as shown in Figure S1. At 90-day follow-up, outcomes were categorized as good outcomes, worsened outcomes, or death. A good outcome was defined as having an mRS score of 0 to 2 or no change in the mRS score compared with baseline. A worsened outcome referred to any increase in the mRS score after 90 days, while death indicated patients who had died within this period.

Figure 1.

Patient inclusion flowchart. GSR-ET indicates German Stroke Registry-Endovascular Treatment; mPSD, moderate prestroke disability; mRS, modified Rankin Scale; nPSD, no prestroke disability; PSD, prestroke disability; and sPSD, severe prestroke disability.

All analyses were performed in Python. A detailed list of all packages and their respective version numbers used for the analysis is provided in the Supplemental Material. Initial Shapiro-Wilk testing indicated that the data were not normally distributed; accordingly, nonparametric methods were applied in subsequent analyses. Group comparisons were performed using the χ² test for categorical variables and the Kruskal-Wallis test for continuous numerical variables. Post hoc tests were conducted where appropriate. For categorical variables, χ² tests on contingency tables were used for larger data sets, and the Fisher exact test was applied for smaller data sets. For numerical variables, the Dunn test was used for post hoc analysis. At each stage of analysis, Bonferroni correction was applied to adjust for multiple comparisons. A P value of <0.05 was considered statistically significant. Missing values were handled on a per-analysis basis. For unadjusted group comparisons, pairwise deletion was applied, that is, only nonmissing values were used for each respective variable. The number of missing values and the proportion of complete data used are summarized in Table S1. Logistic regression was performed to evaluate the relationship between PSD groups and poststroke outcomes. The regression model was built using the Logit function from Statsmodels, with odds ratios (OR) and 95% CIs calculated for each predictor. McFadden pseudo-R² was used to assess the model’s fit, and variance inflation factors were computed to rule out higher-order multicollinearity. Age was analyzed with an OR per 10-year increase. In additional analyses, multinomial logistic regression was performed using the MNLogit function from Statsmodels, with relative risk (RR) ratios and 95% CIs calculated for each predictor. We used machine learning to compare 2 models: one trained without patients with sPSD (scenario 1) and another with a rebalanced group also including patients with sPSD (scenario 2). We applied a backward elimination approach for feature selection, alongside data preprocessing steps such as filtering, feature engineering, and one-hot encoding. Synthetic Minority Oversampling Technique Combined With Tomek Links was applied to rebalance the sPSD group by synthetically oversampling its underrepresented examples and then undersampling Tomek link pairs, especially those dominated by mortality cases, thereby preventing the model from defaulting to death prediction and enhancing its discrimination between good, worsened, and fatal outcomes. To ensure robust statistical reliability, model training was repeated over 1000 random seed iterations with 5-fold stratified cross-validation per run, and performance was evaluated using the area under the receiver operating characteristic curve and the average precision score metrics with permutation importance used to assess feature relevance. Comprehensive details of the neural network are provided in the Supplemental Material. For both the logistic regression and machine learning models, we used only complete-case data sets, excluding any rows with missing values, except for age, which was imputed using the median. This approach was chosen to ensure the robustness of the multivariable models and to avoid bias that might arise from imputing missing values. We used GPT-4 (OpenAI) as an assisting tool for coding and language editing. The study was reported in accordance with the recommendations of the Strengthening the Reporting of Observational Studies in Epidemiology. The data supporting the results of this study are available from the corresponding author upon reasonable request and pending authorization of the study sites and ethics committees.

Results

PSD Groups and Baseline Characteristics

After exclusions, 9456 patients remained from the initial cohort of 13 082 and were categorized into nPSD (n=7387), mPSD (n=1648), and sPSD (n=421) groups.

Prestroke mRS score (0–5) proportions are shown in Table S1.

Baseline data are reported in Table 1. Patients in the mPSD and sPSD groups were significantly older (median age: 82 versus 75 years) and had a higher proportion of female patients compared with the nPSD group (62% and 64% versus 48%).

Table 1.

Baseline and Treatment Characteristicshttps://www.ahajournals.org/doi/suppl/10.1161/SVIN.125.002055

Stroke severity, measured by the admission National Institutes of Health Stroke Scale (NIHSS), was higher in the mPSD and sPSD groups (median: 16 and 17 versus 14). Most cardiovascular risk factors, including hypertension, diabetes, and atrial fibrillation, as well as previous stroke, were more prevalent in patients with PSD. The history of smoking was more common in the nPSD group. Treatment differences were also noted, with IVT being less frequently administered in the mPSD group compared with the nPSD group (36% versus 51%). Successful reperfusion (modified Thrombolysis in Cerebral Infarction score ≥2b) was lower in both the mPSD and sPSD groups compared with nPSD. Cardioembolism and large-artery atherosclerosis emerged as the predominant stroke etiologies, with cardioembolism being more frequent in patients with higher PSD, while large-artery atherosclerosis was more common in the nPSD group. No significant difference was observed in the length of the primary hospital stay between the 3 groups. Tables S2 and S3 contain information on data availability proportions and detailed results from post hoc analyses.

Poststroke Complications and Discharge Modalities

Complications

Complications during the thrombectomy procedure and hospital stay were largely similar across the PSD groups (Figure 2A). No significant differences were observed for most complications, including symptomatic intracranial hemorrhage, dissection and perforation, clot migration, groin hematoma, pseudoaneurysm, malignant middle cerebral artery infarction, and myocardial infarction.

Figure 2.

Complications and outcomes by prestroke disability (PSD) group. A, Peri-interventional and postinterventional complication rates across PSD groups. B, Discharge modalities across PSD groups. C, Unadjusted outcomes at 90-day follow-up across the 3 PSD groups. D, Mortality at each time point shown relative to all patients within each PSD group. E, Proportional mortality distribution across time points normalized to 100% for each group. F, Living status at baseline and after 90 days, stratified by PSD groups. MCA indicates middle cerebral artery; mPSD, moderate prestroke disability; nPSD, no prestroke disability; sICH, symptomatic intracranial hemorrhage; and sPSD, severe prestroke disability.

However, a significant difference was found in the incidence of vasospasm (P<0.01). Patients in the nPSD group experienced vasospasm more frequently (5.1%) compared with the mPSD group (2.6%), while post hoc analysis showed no significant difference between the nPSD and sPSD groups (2.6%).

Discharge Modalities

Discharge modalities also varied significantly across the PSD groups (Figure 2B). Home discharge was most frequent in the nPSD group (27.6%) compared with both mPSD (13.1%) and sPSD (11.3%), with no significant difference between the latter 2 groups. Transfer to another hospital was more common in mPSD (19.4%) and sPSD (21.3%) than in nPSD. Neurorehabilitation was less frequent in the sPSD group (40.9%) compared with both nPSD (57%) and mPSD (55.2%), while nursing home discharge was highest in the sPSD group (26.5%) compared with both nPSD (1.3%) and mPSD (12.2%). Tables S4 and S5 show the detailed distribution of complications and discharge modalities across all groups and the detailed post hoc results. In multinomial logistic regression, mPSD and sPSD were associated with higher RRs of discharge to a nursing home compared with neurorehabilitation (mPSD: RR, 8.22 [95% CI, 5.69–11.88]; P<0.001; sPSD: RR, 13.38 [95% CI, 7.57–23.63]; P<0.001). Older age and higher NIHSS increased the likelihood of nonrehabilitation destinations, while IVT and successful reperfusion were associated with a lower probability of hospital transfer. The full set of estimates is provided in Table S7.

Functional Recovery and Mortality at 90-Day Follow-Up

Unadjusted Outcomes

Unadjusted outcomes at 90-day follow-up showed clear differences across the PSD groups (Figure 2C). In the nPSD group, the largest proportion of patients (40.9%) achieved a good functional outcome, while 33.9% had worsened outcomes, and 25.2% died. Restricting good outcomes in nPSD to return to baseline yielded a rate of 13.2%. In the mPSD group, most patients (52.9%) died, with 32.9% experiencing worsened outcomes, and only 14.1% achieving a good outcome. For the sPSD group, mortality was highest, with 63.3% of patients dying, 16.4% experiencing worsened outcomes, and 20.2% achieving a good outcome. Overall, outcome distributions differed significantly between groups. Especially, good outcomes and mortality varied significantly across all pairwise comparisons, while worsened outcomes differed significantly except between nPSD and mPSD. The full post hoc results are shown in Table S6. Exploratory sensitivity analysis (prestroke mRS score, 0–2 versus ≥3) confirmed the pattern: mRS score ≥3 had 58.1% mortality and 18.7% good outcomes versus 27.9% and 37.7% in mRS score of 0 to 2 (Figure S2).

Adjusted Outcomes

Logistic regression was applied to assess the impact of the PSD group on good outcomes and 90-day mortality, independent of age, IVT, NIHSS, reperfusion success, and sex. Regarding good outcomes, patients in the mPSD group had a significantly lower likelihood of achieving a good outcome compared with the nPSD group (OR, 0.41 [95% CI, 0.34–0.49]; P<0.001). sPSD, compared with nPSD, was not associated with a worse outcome (ie, no worsening from baseline; OR, 1.37 [95% CI, 1.01–1.86]; P=0.044). Variables associated with a favorable outcome in the adjusted model were younger age, IVT administration, lower NIHSS score, successful reperfusion, and male sex (Table 2). To account for secular trends, we included the admission year (centered at the cohort median) as a linear covariate. The year term was not statistically significant and did not relevantly change other coefficients (Table S8). Regarding mortality, both the mPSD (OR, 2.13 [95% CI, 1.86–2.45]; P<0.001) and sPSD (OR, 2.75 [95% CI, 2.15–3.51]; P<0.001) groups had significantly higher odds of death compared with the nPSD group. Older age, higher NIHSS score, and male sex were associated with a higher risk of mortality, while IVT and successful reperfusion significantly reduced the risk (Table 3).

Table 2.

Multivariable Logistic Regression Model for Predicting a Good Outcome Defined as Having an mRS Score of 0 to 2 or No Change in the mRS Score Compared With Baseline

Table 3.

Multivariable Logistic Regression Model for Predicting Mortality

Mortality Patterns Across PSD Groups

The proportional distribution of mortality at different time points, relative to all patients within each PSD group, revealed distinct patterns (Figure 2D). After 24 hours, mortality was low across all groups, with 1.1% in the nPSD group, 1.8% in the mPSD group, and 1.3% in the sPSD group, with no significant differences between the groups at this time point. In-hospital mortality increased significantly, with 16.1% of patients dying in the nPSD group, 31.0% in the mPSD group, and 41.1% in the sPSD group. By 90 days, additional mortality was 8.3% in the nPSD group, 20.7% in the mPSD group, and 20.6% in the sPSD group. However, there was no significant difference in mortality between the mPSD and sPSD groups between discharge and 90-day follow-up. When comparing the relative distribution of mortality time points, differences were minor (Figure 2E). Expressed as a proportion of total 90-day mortality, early deaths within 24 hours accounted for 4.2% in nPSD, 3.4% in mPSD, and 2.0% in sPSD; by discharge, 57.9% to 65.3% of these deaths had occurred, and the remaining 32.5% to 38.8% occurred between discharge and day 90. The detailed mortality rates across all groups and time points, along with the corresponding post hoc results, are provided in Table S9.

Living Status at Baseline and 90-Day Follow-Up

To evaluate functional outcomes, living status at baseline and 90-day follow-up was compared across the PSD groups (Figure 2F), revealing significant differences (P<0.001). At baseline, most patients in the nPSD group lived at home (97.3%), while this was less common in the mPSD (56.8%) and sPSD (31.3%) groups. Nearly half of the patients in the sPSD group were already residing in nursing homes (47.0%) at baseline. Nursing care at home was more frequent in the mPSD and sPSD groups, with no significant difference between them. This pattern persisted at 90 days, with home residence observed in 46.4%, 12.9%, and 3.0% of patients in the nPSD, mPSD, and sPSD groups, respectively; group comparisons remained statistically significant across all PSD categories (all P<0.001). Higher nursing home rates in mPSD (15.7%) and sPSD (20.7%) compared with nPSD (7.8%) remained significant, while the difference between mPSD and sPSD was not. Table S10 provides a detailed percentage breakdown and the results of the post hoc tests.

In the adjusted multinomial model with nursing care as the reference, mPSD was associated with a lower likelihood of home residence (RR, 0.30 [95% CI, 0.19–0.46]; P<0.001), rehabilitation (RR, 0.27 [95% CI, 0.22–0.33]; P<0.001), and hospital stay (RR, 0.63 [95% CI, 0.46–0.85]; P=0.003) compared with nPSD. sPSD showed an even stronger shift, with markedly reduced RRs for home (RR, 0.11 [95% CI, 0.04–0.30]; P<0.001) and rehabilitation (RR, 0.05 [95% CI, 0.03–0.10]; P<0.001). Increasing age and higher NIHSS were associated with lower probabilities of home residence, while IVT was associated with a higher probability. Full estimates are reported in Table S11.

Outcome Prediction for sPSD Using a Neural Network Based on Preinterventional Factors

Model Performance

To predict outcomes for patients in the sPSD group based on preinterventional and intervention-related procedural data, a neural network model was tested, with and without including and rebalancing the sPSD group. Overall model performance was similar in both cases. Nevertheless, when looking specifically at the sPSD subgroup, including and rebalancing their data led to small but statistically significant improvements. While overall model performance was strong in both scenarios (area under the receiver operating characteristic curve ≈0.79), it was minimally, but statistically significantly, higher when the rebalanced sPSD cohort was included (0.7914±0.0128 versus 0.7913±0.0125; P<0.05). However, in the sPSD subgroup itself, including and rebalancing their data yielded small yet statistically significant improvements (area under the receiver operating characteristic curve, 0.718±0.091 versus 0.711±0.092; P<0.001). Table 4 shows the exact model metrics and the differences in area under the receiver operating characteristic curve and average precision score.

Table 4.

Comparison of Model Performance Metrics With and Without Inclusion and Rebalancing of the sPSD Group

Feature Importance

We conducted a permutation importance analysis to identify key predictive features of a good outcome for patients with sPSD. Figure S3 shows the absolute feature importance in an average model of the 2 training scenarios. Feature importance values indicate how strongly each input feature contributes to the model’s predictions, with higher values reflecting greater influence. Age and NIHSS emerged as the dominant predictors (importance values between 0.12 and 0.14), followed by a second tier of moderate importance features including IVT, diabetes, Alberta Stroke Program Early Computed Tomography Score, and time to flow restoration with values in the 0.02 to 0.04 range. All other features, including various vessel occlusion patterns, showed substantially lower absolute importance values (<0.01). When comparing models trained with and without sPSD rebalancing, we observed significant shifts in feature relevance. Figure 3A illustrates the absolute differences in feature importance between these scenarios. Features showing the greatest positive shifts after rebalancing included age and Alberta Stroke Program Early Computed Tomography Score, while admission NIHSS demonstrated decreased importance. Intervention parameters and vessel occlusion characteristics showed mixed patterns: local anesthesia and proximal first segment of the middle cerebral artery occlusion gained importance, while time to flow restoration, posterior cerebral artery occlusion, and second segment of the middle cerebral artery occlusion showed diminished relevance. Figure 3B presents these differences normalized by average absolute feature importance, highlighting the relative magnitude of these shifts. Local anesthesia showed a substantial relative increase, reinforcing its enhanced relevance in the rebalanced model. Similarly, the proximal first segment of the middle cerebral artery occlusion demonstrated a notable relative increase, contrasting with the significant decreases seen for the posterior cerebral artery and the second segment of the middle cerebral artery occlusions.

Figure 3.

Impact of severe prestroke disability (sPSD) rebalancing on permutation feature importance. A, Absolute differences in permutation feature importance for the prediction of a good outcome, comparing models trained without sPSD and with rebalanced sPSD. Positive values indicate a feature gained importance in the rebalanced model (ie, became more predictive for patients with sPSD), while negative values indicate reduced importance. B, Relative differences in feature importance, calculated as percentage change relative to the average feature importance across both models. This normalization highlights the proportional magnitude of shifts, allowing better comparison across features with different baseline values. The x axis lists predictors; the y axis indicates the change in feature importance. ACA indicates anterior cerebral artery; ADAPT, A Direct Aspiration First Pass Technique; Adjunctive EVT, any adjunctive or alternative endovascular technique (eg, ADAPT/aspiration-only, SOLUMBRA, mSAVE, intra-arterial drug application, and ADAPT+SAVE rescue); APT, antiplatelet therapy; ASPECTS, Alberta Stroke Program Early Computed Tomography Score; BA, basilar artery; Direct Admission, direct admission to the intervening hospital; DM, diabetes; EVT, endovascular treatment; Female, female sex (vs male sex); ICA, internal carotid artery (intracranial); IQR, interquartile range; IVT, intravenous thrombolysis; Local Anesthesia, anesthesia type: local (vs anesthesia type: general); M1, first segment of the middle cerebral artery; M2, second segment of the middle cerebral artery; mSAVE, modified SAVE; mTICI, modified Thrombolysis in Cerebral Infarction; NIHSS, National Institutes of Health Stroke Scale; OAC, oral anticoagulation; PCA, posterior cerebral artery; SAVE, Stent Retriever Assisted Vacuum-Locked Extraction; SOLUMBRA, combined stent retriever and aspiration thrombectomy technique; and Time to FLR, time from admission to flow restoration of cerebral blood flow.

Discussion

In this study, we analyzed a large, multicenter cohort from the German Stroke Registry-Endovascular Treatment to assess outcomes of patients with PSD (mRS score, ≥2), especially in those with severe disability (mRS score, 4–5), following MT. This large study addresses a critical evidence gap as previous research often excluded or insufficiently represented patients with sPSD. Our findings indicate that sPSD is associated with worse outcomes after MT, particularly regarding mortality: Patients with sPSD had the highest 90-day mortality rate, followed by those with mPSD and nPSD. Functional recovery also varied significantly across disability groups: 41% of the nPSD group achieved a good outcome (mRS score, 0–2 or no change from baseline) compared with only 14% of patients with mPSD and 20% with sPSD.

After adjustment for age, stroke severity (NIHSS after 24 hours), IVT, successful reperfusion (modified Thrombolysis in Cerebral Infarction score ≥2b), and sex, patients with sPSD showed no significant disadvantage in achieving a good outcome compared with those without PSD, suggesting a ceiling effect, where profound baseline deficits limit both recovery potential and scope for further decline. Interestingly, patients with mPSD showed the lowest proportion of good outcomes after adjustment. This may reflect the greater clinical heterogeneity and functional vulnerability within this group, which allows for both noticeable decline and limited recovery, unlike in sPSD, where further deterioration is less pronounced, and recovery potential is already constrained. In contrast, both mPSD and sPSD independently conferred substantially higher odds of 90-day mortality. These findings align with other studies linking PSD to increased mortality.^{17,20,21,26,27} Nevertheless, prior investigations also highlight that PSD does not always preclude favorable outcomes,^{22,26,28–30} and even patients with sPSD may regain their baseline status.^17,31

Data from an Italian registry, for instance, showed that patients with prestroke mRS score of 3 to 4 were older, more often female, and, despite a similar proportion returning to their premorbid status, had approximately double the mortality and lower recanalization rates compared with those with mRS score of 0 to 2.³² They also reported shorter intraprocedural times, possibly due to fewer thrombectomy passes, a nonsignificant trend that we similarly observed. A meta-analysis further quantified that only 20% to 31% of patients with a prestroke mRS score of 2 to 4 regain baseline function, but that achieving Thrombolysis in Cerebral Infarction 2b/3 reperfusion can double this likelihood while reducing mortality, underscoring the important role of technical success.³³ A Czech registry study additionally showed that patients with PSD were less often selected for thrombectomy, experienced longer door-to-puncture times (median 75 versus 54 minutes), and had higher mortality; yet, about one-third returned to their premorbid status, and the relative benefit over medical therapy was preserved.³⁴ Collectively, these data support our observation that PSD is associated with older age, slower workflows, lower recanalization rates, and higher mortality, yet a substantial minority, particularly those achieving successful reperfusion, can regain their baseline function.

In contrast, one study reported markedly poorer 1-year outcomes after thrombectomy among patients aged ≥80 years, with more than half deceased and fewer than 20% functionally independent.³⁵ Prognosis was particularly unfavorable in the presence of comorbidities and prolonged mechanical ventilation (>48 hours). This highlights the need for extended follow-up and underscores that, especially in patients with severe baseline impairment, the likelihood of regaining meaningful independence remains low. Treatment decisions in this vulnerable population should, therefore, be guided by realistic expectations of long-term outcomes. Furthermore, MT decisions in severely disabled patients should respect the presumed will of the patient and be guided by shared decision-making with family members.

Interestingly, our results show that patients with PSD did not experience higher rates of complications, such as symptomatic intracranial hemorrhage, consistent with previous studies.^{17,21,26,28–30,32,36}

In contrast, a Swedish registry study by Larsson et al³¹ likewise found similar symptomatic intracranial hemorrhage rates across both groups, yet observed higher overall in-hospital complications in the PSD cohort. However, our findings suggest that PSD, while associated with poorer functional outcomes and higher mortality, does not inherently increase the procedural risks of MT. The length of hospital stay did not differ significantly between groups though higher in-hospital mortality among patients with PSD may have influenced the similar median duration. Nonetheless, this challenges the assumption that preexisting disability increases hospital resource use. Compared with patients without prior disability, those with mPSD or sPSD were transferred to neurorehabilitation less often and discharged to nursing homes more often, likely because the potential for functional recovery was considered limited by clinicians when marked functional deficits already exist and because insurance coverage for intensive rehabilitation may be less often approved for the same reasons.

Our study is among the first to apply machine learning to a large cohort including patients with sPSD. By leveraging baseline characteristics and preinterventional information, our model detected subtle but robust differences relevant for patients with sPSD. Permutation importance analysis showed that age and Alberta Stroke Program Early Computed Tomography Score remain fundamental predictors, while admission NIHSS importance decreased, likely because baseline neurological deficits of patients with sPSD make admission NIHSS less discriminative of new deficits compared with patients without PSD, where NIHSS clearly reflects acute changes. Moreover, as the premorbid mRS in patients with sPSD often reflects not only focal deficits but also broader functional impairments, NIHSS may carry less prognostic weight. Vessel occlusion territories showed heterogeneous changes, with the proximal first segment of the middle cerebral artery gaining importance, while the second segment of the middle cerebral artery and posterior circulation decreased in relevance, suggesting different prognostic implications in patients with preexisting disabilities. The increased relevance of local anesthesia in the rebalanced model suggests that the choice between local anesthesia and general anesthesia gains prognostic importance in patients with sPSD, possibly reflecting the specific risks and needs of this vulnerable population. These findings highlight the need for tailored methodological approaches when developing predictive models for specific patient subgroups. However, the small representation of patients with sPSD and synthetic oversampling warrants caution. Although such models should not currently be used as a basis for therapy decisions, they provide exploratory insights. Therefore, future studies should validate these findings in larger cohorts and incorporate explainable artificial intelligence to better elucidate the interplay of clinical and imaging predictors in this important subgroup.

A limitation of our study is its observational design, which may be influenced by unmeasured confounders and center-specific practices. However, its nationwide, multicenter registry base confers broad external validity to comparable MT settings. The exclusion of patients with incomplete follow-up could introduce bias, and reliance on the mRS as the primary outcome measure may overlook patient-centered aspects of recovery, such as quality of life. Future studies should, therefore, incorporate patient-reported outcomes and quality-of-life measures to provide a more comprehensive assessment of recovery, particularly in patients with sPSD. In addition, while the 90-day follow-up period is standard and sufficient to capture most of the mortality, which occurred predominantly during hospitalization, it may not adequately reflect long-term functional outcomes.³¹ Another limitation is the lack of an untreated control group, which limits causal inference regarding treatment benefit in patients with PSD. The inclusion of only MT-treated patients might introduce selection bias, as treatment decisions could be influenced by unmeasured factors such as younger age, preserved cognitive function, perceived quality of life, level of social support, or institutional care status. Patients with sPSD might represent a highly selected subgroup based on clinical characteristics (eg, type of preexisting disability) and radiological findings (eg, occlusion site, favorable collateral status, or perfusion mismatch). This selective treatment approach might also have influenced the observed outcome distribution, potentially contributing to the comparable outcomes between patients with sPSD and mPSD. Evidence for MT in PSD largely derives from observational studies prone to selection bias, while randomized data exist only from a HERMES (Highly Effective Reperfusion Evaluated in Multiple Endovascular Stroke Trials) substudy including mildly disabled patients (mRS score, 1–2). Therefore, the benefit of MT in patients with more substantial PSD remains uncertain.^37–40

In conclusion, despite the increased mortality, a notable proportion of patients with PSD, including those with severe impairment, achieved favorable outcomes, supporting the consideration of MT on an individual basis rather than excluding this group a priori. Future studies should prioritize randomized trials and extended follow-up to refine treatment strategies. Redefining success to cover outcomes such as maintaining baseline function or reducing additional care needs will be essential for optimizing care and supporting personalized treatment approaches for this vulnerable population.

Conclusions

sPSD is associated with higher mortality after MT, but a notable proportion of patients maintains baseline function, highlighting the potential benefits of MT in this population. PSD was not linked to increased periprocedural or postprocedural complications or prolonged hospital stays. Patients with sPSD were less frequently discharged to rehabilitation, often requiring continued hospitalization or nursing care instead. Machine learning models further revealed shifts in predictive factors when sPSD was rebalanced. These findings support the need for individualized treatment decisions and refined outcome assessments for patients with preexisting disability.

ARTICLE INFORMATION

Acknowledgments

The authors acknowledge the contribution of all study nurses and extend their gratitude to the patients and their relatives for participating in the study.

Author Contributions

Drs Asperger and Stösser conceptualized and designed the study. Drs Shirvani, Bode, Nitsch, Stösser, Ebrahimi, von Danwitz, Asperger, Layer, Meißner, Dorn, Petzold, Thielscher, and Weller acquired data. Drs Asperger and Stösser analyzed data. Dr Asperger wrote the original draft. Dr Stösser provided critical feedback on the manuscript. All authors approved the final version of the manuscript.

Sources of Funding

This study was conducted without specific funding from public, commercial, or nonprofit organizations.

Disclosures

Dr Dorn is a consultant for Microvention, Balt, and Johnson & Johnson; received scientific grants from Johnson & Johnson; and received speakers honoraria from Johnson & Johnson, Q’Apel, Tonbridge, Asahi, Acandis, Stryker, Microvention, Medtronic, and Penumbra. Dr Bode received speakers honoraria from AstraZeneca. Dr von Danwitz received travel grants from Sanofi and Viatris. The other authors report no conflicts.

Supplemental Material

Supplemental Materials and Methods

Tables S1–S11

Figures S1–S3

APPENDIX

GSR-ET Collaborators: M. Abruscato, A. Alegiani, J. Berrouschot, T. Boeck-Behrens, A. Bormann, B. Eckert, U. Ernemann, J. Fiehler, F. Flottmann, K. Gröschel, F. Keil, L. Kellert, O. Nikoubashman, C.H. Nolte, S. Poli, G.C. Petzold, A. Reich, J. Röther, J.H. Schäfer, M. Schell, E. Siebert, G. Thomalla, S. Thonke, S. Tiedt, T. Uphaus, S. Wunderlich, H. Zimmermann.