Large Vessel Occlusion Stroke Triage Strategies - Project Summary / Abstract . This is a K23 resubmission for Dr. Kevin Keenan, a vascular neurologist and early-career clinical researcher at the University of California, Davis focused on prehospital diagnostic testing for large vessel occlusion (LVO) stroke triage. It will provide Dr. Keenan with the mentored training he requires to become an independent stroke emergency medical services (EMS) researcher, including large EMS dataset management and analysis, machine learning for classification, prospective study design, emergency research ethics, and R01 grant writing. Primary Co-Mentor Frank Sharp, MD, is a vascular neurologist with expertise in the development of stroke diagnostic tests using machine learning and over 35 years of R01-level funding. Primary Co-Mentor Nishijima, MD, MAS, is an emergency medicine physician with expertise in prehospital neurological emergency trials and decision analysis. Mentors Xin Liu, PhD, Sandra Taylor, PhD, and Kwan Ng, MD, PhD will provide further machine learning, biostatistical, and stroke expertise to ensure Dr. Keenan’s success. Based on recent evidence, Dr. Keenan’s central hypothesis is that combining clinical variables predictive of LVO stroke such as age, sex, time of onset, vitals, atrial fibrillation, the Glasgow Coma Scale, and glucose with the Cincinnati Prehospital Stroke Scale (CPSS) neurological examination will improve diagnostic and decision curve analysis performance compared to existing LVO stroke tests that exclusively rely on the neurological examination. Using machine learning, he will combine these clinical variables and CPSS into a new test called CLEAR-LVO (CLinical and Exam bAsed Recognition – LVO). CLEAR-LVO will be compared to CPSS alone, since CPSS is a widely used stroke screening test that can also identify LVO, and multiple recommended but not widely adopted LVO tests, such as the Los Angeles Motor Scale (LAMS). This research is significant because of the urgent need to improve prehospital LVO stroke testing so the correct patients can be triaged to endovascular stroke centers for time-sensitive treatment. It is innovative because it will use machine learning methods, large EMS datasets, and a decision curve analysis to guide prehospital stroke triage policies. Specific Aim 1 will test the hypotheses that CLEAR-LVO will diagnose LVO stroke with a statistically and clinically significantly higher Youden’s Index (sensitivity + specificity - 1) than CPSS and other LVO scales in a large EMS dataset (n = 86,396). External validation via the Aim 2 dataset will enhance the rigor of our findings. Specific Aim 2 will test the hypothesis that CLEAR-LVO will provide higher decision curve analysis net benefit than prospectively documented LAMS and CPSS scores in a separate Yolo County dataset (n = 2,310). The urgent need to determine whether EMS agencies should use CPSS or LAMS now, while awaiting better tests, will ensure that Aim 2 and Dr. Keenan’s R01 plans are not dependent on CLEAR-LVO. This K23 will lead to a multi-center R01 prospectively comparing CPSS, LAMS, and CLEAR-LVO in ambulances, prospective re- derivation of CLEAR-LVO if indicated, and subsequent triage studies aimed at improving stroke outcomes.
