Abstract TMP41: Machine Learning-Enabled Automated Large Vessel Occlusion Detection Improves Treatment Times: A Multi-Center Cluster Randomized Clinical Trial

Introduction: Automated LVO detection from CT angiography may improve treatment times in LVO acute ischemic stroke (AIS), as suggested by prior single center non-randomized studies. Here, we perform a cluster randomized trial to evaluate this question with improved rigor. Methods: We performed a cluster randomized stepped wedge clinical trial across four comprehensive stroke centers January 2021 to February 2022. An automated LVO detection software (Viz.AI) was implemented at the 4 hospitals in a stepped fashion, with a randomly determined order (Figure 1). We included patients who underwent EVT. Patients presenting within the 2-week software transition period were excluded. The primary outcome was door-to-groin puncture time (DTG) adjusted for age, sex and NIHSS and was determined using a mixed-effect linear regression, with a random effect for cluster (hospital site) and fixed effect for exposure status. Results: Among 249 patients that met inclusion criteria, median age was 67 [IQR 57-79], NIHSS 17 [IQR 11-22] and 50.6% were female. Median time from last known well to arrival was 222 minutes and 90% had ASPECTS ≥ 8. A total of 199 patients were treated in the unexposed period and 50 in the exposed. In univariable analysis, median DTG improved in the unexposed vs. exposed periods (91 min [IQR 63 - 108 min] vs. 73 [IQR 32-102 min], p<0.05). In multivariable adjusted mixed-effects model, DTG improved by approximately 15 minutes [4 - 26 min, 95% CI, p<0.01]. Discussion: In a multi-center cluster randomized trial, implementation of an automated LVO detection algorithm improved treatment times in patients with LVO AIS.