Differential Effect of Endovascular Treatment in Medium or Distal Vessel Occlusion: A Distal Secondary Analysis

Why clinicians should know about it

• Picked for Health Informatics (top studies of the week, 1 November 2026): External model validation in randomized trial data informs treatment decisions

Abstract

BACKGROUND: Patients with medium or distal vessel occlusion stroke are heterogeneous. Randomized trials show no net benefit of endovascular treatment (EVT) over best medical treatment (BMT) alone, which may mask heterogeneity of treatment effect. We used an externally developed machine learning model to predict excellent early neurological outcome (EENO) with BMT alone and examined whether the effect of EVT differed across predicted risk. METHODS: A Tabular Prior-Data Fitted Network (TabPFN) model was trained to predict EENO (24-hour National Institutes of Health Stroke Scale [NIHSS] score of 0-2) in patients with medium or distal vessel occlusion treated with BMT alone in a local Bernese Stroke Registry. This model was applied to patients from the DISTAL randomized trial (55 centers and 11 countries), who were randomized to EVT plus BMT or BMT alone. We tested for interaction between predicted EENO and treatment allocation on the primary outcome (90-day modified Rankin Scale shift), repeating this using admission NIHSS alone. RESULTS: In the development cohort (n=257, 42% female, and mean age 70.4±14.8 years), 54% achieved EENO. In the DISTAL cohort (270 BMT alone, 269 EVT plus BMT, 44% female, and mean age 74.9±12.0 years), the model showed moderate external discrimination (area under the receiver operating characteristic curve 0.73). A significant interaction was found between predicted EENO and treatment effect (Pinteraction=0.012): in 152 patients predicted to achieve EENO, EVT was associated with worse 90-day modified Rankin Scale score (common odds ratio, 0.48 [95% CI, 0.27-0.84]), whereas no treatment effect was seen in 387 patients predicted not to achieve EENO (common odds ratio 1.13 [95% CI, 0.79-1.60]). A model using admission NIHSS alone showed comparable discrimination (area under the receiver operating characteristic curve 0.71) and a common odds ratio for EVT of 0.47 (95% CI, 0.25-0.89) in patients with an NIHSS score of ≤4 and a common odds ratio of 1.05 (95% CI, 0.75-1.47) in patients with an NIHSS score of >4 (Pinteraction=0.029). CONCLUSIONS: EVT was harmful in patients with medium or distal vessel occlusion predicted to achieve EENO with BMT alone, and an NIHSS-alone model performed comparably to the multivariable model. REGISTRATION: URL: https://www.clinicaltrials.gov; Unique identifier: NCT05029414.

Abstract as published, via PubMed.