Hidden Blood Signatures Reveal Which Stroke Patients Benefit From Clot Removal

Endovascular thrombectomy has transformed the treatment of acute ischemic stroke caused by large vessel occlusion. By mechanically extracting the clot that blocks a major cerebral artery, the procedure restores blood flow to endangered brain tissue in more than 80 percent of treated patients, and multiple randomized trials have demonstrated substantial gains in functional independence compared with medical therapy alone. Yet a stubborn paradox persists at the bedside: roughly half of patients whose vessels are successfully reopened still fail to achieve functional independence at three months, a phenomenon clinicians call futile recanalization. A new study published in Frontiers in Neurology suggests that the answer to this paradox may lie not in the success of the mechanical procedure itself, but in the distinct biological states that patients bring to, and carry through, the reperfusion period.

The research team, led by investigators at Taizhou Hospital of Zhejiang Province, affiliated with Wenzhou Medical University in China, applied a statistical technique known as latent class analysis to a cohort of 825 consecutive patients who underwent thrombectomy between January 2021 and December 2024. Rather than examining each laboratory marker in isolation, as most prior studies have done, latent class analysis searches for hidden subgroups of patients who share similar patterns across many variables simultaneously. The method has already proven its worth in critical care medicine, where it revealed biologically coherent subphenotypes in sepsis and acute respiratory distress syndrome that respond differently to treatment. Until now, however, no study had used it to integrate inflammatory, metabolic, and coagulatory biomarkers with clinical characteristics in thrombectomy patients.

The researchers fed the algorithm twelve routine post-procedural indicators, including six binary variables such as sex, hypertension, atrial fibrillation, prior stroke, intravenous thrombolysis, and stroke territory, and six continuous measures: age, admission stroke severity on the National Institutes of Health Stroke Scale, the neutrophil-to-lymphocyte ratio, blood glucose, D-dimer, and albumin. The neutrophil-to-lymphocyte ratio captures systemic inflammatory imbalance, blood glucose reflects metabolic derangement, and D-dimer, a fibrin degradation product, signals activation of the coagulation and fibrinolysis systems. Models with two, three, and four latent classes were estimated, and the three-class solution emerged as the clear winner, improving the Bayesian Information Criterion by 349 points over the two-class model while maintaining acceptable classification certainty, with an entropy of 0.745 and a minimum average posterior probability of 0.883.

What emerged were three strikingly different biological profiles. The largest group, accounting for 44.4 percent of patients, was labeled the Low-burden phenotype. These patients had the lowest inflammatory and coagulatory activation across all measured markers, with a median neutrophil-to-lymphocyte ratio of 5.48, median blood glucose of 6.40 mmol/L, and median D-dimer of just 0.70 mg/L. They were also the youngest, had the mildest neurological deficits on admission, and the lowest prevalence of diabetes and atrial fibrillation. Crucially, 70.8 percent of these patients achieved a favorable outcome at three months, defined as a modified Rankin Scale score of 0 to 2, meaning they could walk and attend to their own bodily needs without assistance.

The second group, comprising 42.7 percent of the cohort, was named the Metabolic phenotype. Its defining signature was hyperglycemia, with median blood glucose of 7.96 mmol/L, combined with an elevated neutrophil-to-lymphocyte ratio of 8.68 and the highest prevalence of diabetes mellitus at 28.7 percent. These patients fared considerably worse: only 44.9 percent achieved functional independence, and 7.1 percent suffered a symptomatic intracranial hemorrhage, more than four times the rate seen in the Low-burden group. The biology behind this pattern is well understood. Elevated glucose after reperfusion promotes blood-brain barrier disruption, augments oxidative stress, and sustains thromboinflammation, mechanisms that prior studies, including analyses of the large MR CLEAN Registry, have linked to worse functional outcomes and nearly doubled hemorrhagic risk after endovascular treatment.

The third and smallest group, 13.0 percent of patients, was the Hypercoagulable phenotype, and it carried the most dramatic signature of all. Median D-dimer in this class reached 5.83 mg/L, nearly eight times the level in the Low-burden group, alongside the highest neutrophil-to-lymphocyte ratio at 9.66, the oldest median age of 73 years, and the greatest prevalence of atrial fibrillation at 49.5 percent. A sensitivity analysis showed that this D-dimer elevation was not an artifact of thrombolysis-induced fibrinolysis, since levels were similar whether or not patients had received intravenous clot-dissolving drugs. Only 40.2 percent of these patients achieved favorable outcome, and 8.4 percent experienced symptomatic intracranial hemorrhage, the highest rate of any class.

After statistical adjustment for age, sex, admission stroke severity, intravenous thrombolysis, complete reperfusion, atrial fibrillation, and systolic blood pressure, both the Metabolic and Hypercoagulable phenotypes remained independently associated with unfavorable outcome. Compared with the Low-burden class, the odds of favorable recovery were 49 percent lower in the Metabolic group and 59 percent lower in the Hypercoagulable group, both highly statistically significant. Importantly, the researchers deliberately excluded the biomarkers that defined each phenotype from the regression model to avoid circular reasoning, and a sensitivity analysis dropping the overlapping covariates produced even stronger associations, confirming that the phenotype-outcome link was not an artifact of the modeling strategy.

To test whether these findings were a one-off statistical fluke, the team assembled an independent temporal validation cohort of 211 thrombectomy patients treated between January 2025 and November 2025 at the same institution, using identical criteria and applying the classification algorithm without re-estimation. The class distribution reproduced almost exactly, with proportions stable within three percentage points, and the rank-order of outcomes was preserved: 70.0 percent favorable outcome in the Low-burden class, 29.2 percent in the Metabolic class, and 22.7 percent in the Hypercoagulable class. Although the absolute outcome rates were lower in the validation cohort, a formal test for heterogeneity between cohorts was not statistically significant, and bootstrap resampling with 1,000 replicates confirmed the stability of the estimates despite the modest sample size.

The study also quantified how much the phenotypes add to conventional prediction. A base model containing seven established clinical predictors achieved an area under the receiver operating characteristic curve of 0.727, and adding latent class membership raised this to 0.743, a statistically significant but modest improvement. This nuance matters: the phenotypes are not a replacement for clinical judgment but a complementary lens that captures biological information invisible to standard variables. Notably, the non-monotonic pattern of blood glucose across the classes, rising in the Metabolic group but falling back near baseline in the Hypercoagulable group, argues against interpreting the three phenotypes as points on a single severity spectrum and instead supports two partially distinct disease processes sharing an inflammatory core.

The authors are careful to frame their findings as hypothesis-generating rather than practice-changing. The single-center retrospective design limits generalizability, the timing of biomarker sampling within the 24-hour post-procedural window was not standardized, and conditions such as infection or malignancy that can independently elevate the neutrophil-to-lymphocyte ratio and D-dimer were not systematically captured. Baseline anticoagulant use, particularly relevant for the atrial fibrillation-heavy Hypercoagulable class, was also unavailable. Still, the framework points toward testable therapeutic hypotheses: intensified glycemic control during the peri-reperfusion period for the Metabolic phenotype, and carefully timed anticoagulation strategies for the Hypercoagulable phenotype, both of which would require prospective multicenter validation before informing clinical practice. If those trials succeed, the humble blood draw performed within a day of clot removal may become one of the most consequential moments in stroke care.

Subject of Research: Biological phenotyping of patients undergoing endovascular thrombectomy for acute ischemic stroke using latent class analysis

Article Title: Latent class analysis identifies three biologically distinct phenotypes among patients undergoing endovascular thrombectomy: a derivation and temporal validation study

Article References: Latent class analysis identifies three biologically distinct phenotypes among patients undergoing endovascular thrombectomy: a derivation and temporal validation study. (n.d.). https://doi.org/10.3389/fneur.2026.1855033

Image Credits: AI Generated

DOI: 10.3389/fneur.2026.1855033

Keywords: endovascular thrombectomy, acute ischemic stroke, latent class analysis, patient phenotypes, D-dimer, neutrophil-to-lymphocyte ratio, hyperglycemia, functional outcome, futile recanalization, symptomatic intracranial hemorrhage, stroke prognosis, biomarkers