Abstract 177: Predictors Of All-cause Mortality In Young Onset Stroke: An Artificial Neural Network Analysis Using A Nationwide Cohort

Abstract

Background: It is crucial for providers to screen high risk patients with limited data being available on young onset stroke (YOS). We aimed to determine predictors of all-cause mortality in this population using the Artificial Neural Network (ANN) model in a national cohort. Methods: We identified young adult (18-44 yrs) YOS hospitalizations from the National Inpatient Sample (2018). ANN’s predictive factors were selected for all-cause mortality. YOS admissions were randomly split between training (70%) & testing datasets (30%). Training data was used to calibrate ANN while testing data was used to evaluate accuracy of the algorithm. We compared the frequency of incorrect prediction between training and testing data and measured area under the Receiver Operating Curve (AUC) to determine ANN’s efficacy in predicting in-hospital mortality in YOS. Results: The 2018 YOS cohort consisted of 39,040 admissions with a mean age of 36 ± 6 years (50.1% male, 51.1% white, 26.2% black, 15.5% Hispanic, 3.2% Asian or Pacific islanders) patients). The all-cause-in-hospital mortality was 5.3%. Training data showed an improved lower predictions in testing model vs testing (5.0% vs 5.2% error rate) , thereby depicting better accuracy. Normalized predictors are displayed in Figure 1a. The AUC was 0.82 (Fig 1b) which shows an excellent ANN model for inpatient mortality in YOS patients. Conclusion: The ANN model successfully revealed the order of prevalent predictors for all-cause mortality that can eventually be utilized to improve survival in high-risk patients.