Endovascular Stroke Thrombectomy for Patients With Concomitant Head Trauma: A Propensity Score–Matched and Machine Learning Study

BACKGROUND: The safety and efficacy of endovascular thrombectomy (EVT) for patients with large vessel occlusion stroke presenting with concomitant head trauma is unclear. METHODS: Using the Nationwide Readmissions Database in the United States (2016–2022), we identified adults with nonelective admissions for large vessel occlusion stroke and concurrent head trauma. Patients who underwent EVT were compared with patients who did not undergo EVT using 1:1 propensity score matching. A causal forest machine learning model was deployed to estimate individual-level conditional average treatment effects for excellent neurological outcomes, with Shapley Additive Explanation values quantifying variable importance. RESULTS: Of 15 527 patients with large vessel occlusion stroke large vessel occlusion stroke with concomitant head trauma (median age, 78 years; 52.6% female; 51.6% with superficial injury), 2750 (17.7%) received EVT; propensity score matching yielded 2616 balanced pairs. After matching, EVT was associated with significantly higher rates of excellent outcomes (19.0% versus 13.7%; P =0.002) with no difference in mortality (14.6% versus 14.7%; P =0.93). In causal forest analysis, younger age, higher National Institutes of Health Stroke Scale score, lack of intravenous thrombolysis, and lack of altered mental status emerged as the strongest drivers of larger treatment benefit, and patients were dichotomized into EVT-favorable (n=8097) and EVT-unfavorable (n=7430) subgroups. In the favorable subgroup, EVT was associated with higher routine discharge (25.7% versus 21.5%; P =0.01); in the unfavorable subgroup, EVT conferred no benefit in routine discharge (10.0% versus 11.7%; P =0.26) but was associated with higher mortality (19.0% versus 14.0%; P =0.004). CONCLUSIONS: EVT is overall associated with higher rates of favorable discharge outcomes; however, treatment effect heterogeneity is substantial.