Abstract Body: Introduction
Determining ischemic stroke etiology using transthoracic echocardiography (TTE) is often considered to be a standard part of routine stroke care and secondary prevention of stroke. However, TTE’s are not always readily available, can delay hospital discharge, and increase the cost of each hospitalization. Identifying which patients specifically benefit from TTE’s could optimize resource allocation. The aim of the study is to develop a model using machine learning methods to confidently identify patients unlikely to have an actionable finding with enough statistical certainty to defer an inpatient TTE.

Methods
The model was developed using 874 patients admitted for stroke/TIA from 2017-2018 and was validated on another data set of 200 patients randomly selected from 2017-2018 from a different institution. 23 variables were considered as potential inputs for the model, which was developed using logistic regression. All variables except six were truncated from the model based on coefficient magnitude and adjusted to be the same low-term improper fraction for simplicity. This model was internally validated using a 5-fold cross-validation and was then tested on an external validation data set with performance being compared to other standard machine learning models.

Results
The training/test data consisted of 874 patients (52.9% male; median age 64 years). Validation data set consisted of 200 patients (53.5% male; median age, 63 years). For the final model, termed ALO₂HA, mean AUC on the training/test data across five-fold cross validation was 0.79 (95% CI, 0.74 - 0.84). The model consisted of six variables, and one point was awarded for each: atrial fibrillation, large artery atherosclerosis, large vessel occlusion, obesity, prior anti-hypertensive medication usage and if patient’s age was 18-39 or > 69. Risk of positive findings was 6.2% for score of 0, 26.1% for score of 1, 65.0% for score of 2, 90.7% for score of 3, 98% for score of 4, and 99.7% for score of 5 or greater. When tested on the external validation data set, AUC was determined to not be significantly different than the AUC for the training/test data.

Conclusion
The ALO₂HA model is a clinical tool which can stratify which patients admitted for acute ischemic stroke/TIA are more likely to benefit from inpatient TTE's.