A Discrete Event Simulation To Assess The Impact Of Ems Routing Algorithms On The Number Needed To Route And Transport Time

Introduction: Proposed EMS routing algorithms permit additional transport time to an endovascular center (EC) beyond the closest non-EC for patients with suspected large vessel occlusion acute ischemic stroke (LVO). The effectiveness of these algorithms depends on screening tools and patient location relative to EC and non-ECs. We implemented routing algorithms in a discrete event simulation to examine their impact on one region. Methods: We simulated stroke and stroke mimic patients screened by EMS over a year using hospital locations and demographics of Mecklenburg County, NC. We used an 8% LVO prevalence among those screened and geographically distributed patients using published stroke incidence rates and census tract population estimates, stratified by age, sex, and race. We estimated distance from census tract centroids to the nearest EC and non-EC using real road travel times. Last known well (LKW) was probabilistically assigned using county data. A patient was EC-routed if they screened positive, had LKW ≤6 hours and were within an allowable additional transport time. We simulated policies that varied by stroke severity screen (LAMS ≥ 4, RACE ≥ 5, C-STAT ≥ 2) and allowable additional transport time (10, 20, and 30 minutes). We define Number Needed to Route (NNR) as the number of patients enduring additional transport time to route one LVO patient to an EC. Results: Over 100 replications, EMS screened an average of 3102 patients annually; 249 were LVOs. NNRs were 2.6 (LAMS ≥ 4), 5.3 (RACE ≥ 5), and 9.3 (C-STAT ≥ 2). The number of EC-routed non-LVOs ranged from 87 (LAMS ≥ 4, 10 minutes) to 859 (C-STAT ≥ 2, 30 minutes). The proportion of LVOs within 10 and 20 minutes of added transport time to an EC was 67% and 99.6% respectively. EC-routing added a mean of 5.5 and 9.5 minutes to transport time for 10 and 20 minute policies respectively. A 20 minute policy EC-routed 1.8 times more patients than a 10 minute policy (e.g. C-STAT: 957 vs. 535). Increasing from a 20 to 30 minute policy routed only 4 more patients, thus these policies had similar results. Conclusions: We designed and tested a simulation tool to evaluate LVO routing policies. It is easily modifiable to aid in tailoring routing policies to a specific region. We propose using NNR as an intuitive metric of non-LVO overtriage.