Abstract 14214: Comparative Analysis of Artificial Intelligence Techniques for ST Segment Elevation Myocardial Infarction and Their Risk of Mortality: A Nationwide Analysis

Introduction: ST segment elevation myocardial infarction (STEMI) significantly contributes to patient morbidity and mortality nationwide. Determining which patient characteristics lead to an increased mortality rate is vital, and the use of artificial intelligence may highlight overlooked characteristics. Purpose: To investigate risk factor discordance between Multivariate Logistic Regression (MLR) and Multilayer Perceptron (MLP) Neural Networks in order to determine patient characteristics that lead to increased mortality.

Methods: Patient data was collected from 2011-2018 from the National inpatient database (NIS) using International Classification of Disease Codes (ICD-9 and ICD-10). We selected patients hospitalized with STEMI. MLR and MLP models were used to tabulate adjusted odds ratios and Normalized Importance Scores respectively. Area under the curve (AUC) was calculated to determine if select comorbid conditions were associated with increased risk of mortality.

Results: A total of 925,161 patients were included in our analysis. Our patient population was predominantly Caucasian, with males constituting 61% of our patient population. Notable contributors to mortality via MLR analysis include cardiogenic shock, AKI, ESRD, cirrhosis and pay status. Notable contributors to mortality via MLP analysis include cardiogenic shock, AKI, ESRD, hyperlipidemia and race. Overall MLP model classification accuracy is rated at 95.2%. Receiver operator AUC is calculated at 0.85.

Conclusion: We found that patients STEMI had similar risk factors associated with mortality between MLR and MLP, however there was a discordance. This was predominantly in weighing certain socioeconomic risk factors as well as chronic comorbid conditions. This begets further analysis to optimize predictive capabilities.