Neurologic outcomes of carotid and other emergent interventions for ischemic stroke over 6 years with dataset enhanced by machine learning

Abstract

Background: The current mainstays of ischemic stroke treatment include the use of thrombolysis (tissue plasminogen activator [tPA]), urgent carotid endarterectomy (uCEA) or urgent carotid artery stenting (uCAS), and mechanical endovascular reperfusion/thrombectomy (MER). Scarce data describe the presenting stroke severity and neurologic outcomes for these acute ischemic stroke interventions, alone or in combination. The authors hypothesize that patients undergoing carotid interventions experience better functional neurologic outcomes than other stroke interventions.

Methods: A comprehensive stroke center dataset was combined with data for stroke-related procedures, comorbidities, complications, and physician documentation collected from electronic medical record data. A total of 10,975 patient encounter records from January 1, 2015, through July 31, 2021, were retrieved. The presenting stroke severity was determined by vascular/stroke neurologists using the National Institutes of Health Stroke Scale (NIHSS). Functional neurologic outcomes were reported using the modified Rankin scale (mRS) score, which quantifies the degree of neurologic disability. Because mRS values were only available for 3627 encounters in the original dataset, the authors developed a machine learning algorithm to analyze physician documentation and assign an mRS value. After the exclusion and machine learning analysis, a total of 5170 patient encounters were included for statistical analysis. Statistical analyses included the χ² test, one-way analysis of variance and logistic regression on 30-day complications, stroke severity, and neurologic outcomes.

Results: Patients were divided into five cohorts: (1) uCEA or uCAS (n = 189), (2) tPA alone (n = 1053), (3) MER alone (n = 418), (4) tPA + MER (n = 199), and (5) no intervention (n = 3311). Patients undergoing uCEA/uCAS were significantly more likely to be male, smokers, and have a history of peripheral arterial disease compared with other stroke cohorts. The length of stay was shortest for patients who only received tPA or no intervention (6 days), followed by uCEA/uCAS (7.2 days), MER (10.2 days), and tPA + MER (8.8 days) cohorts (P < .001). The 30-day mortality was highest in the MER cohort (12.2%) and lowest in the uCEA/uCAS cohort (2.6%). The uCEA/uCAS cohort compared with other cohorts had the lowest presenting stroke severity (NIHSS 4.9 vs NIHSS 6.9-16.0), and best neurologic outcomes (mRS 1.7 vs mRS 1.8-2.6).

Conclusions: After an ischemic stroke, patients undergoing urgent carotid interventions had the lowest presenting stroke severity (NIHSS) and highest rate of independent neurologic outcomes (mRS) compared with other stroke interventions. Incoming stroke severity correlates with functional neurologic outcomes, and patients who present with an NIHSS of 10 or less who undergo uCEA/uCAS have a high likelihood of independent neurologic functional outcome (mRS of ≤2).

Keywords: Ischemic stroke; Machine learning; Neurologic outcomes; Stroke severity; Urgent carotid.