Abstract 485: On the Relative Effectiveness of Machine Learning and Statistical Methods in Predicting Abdominal Aortic Aneurysm Rupture in the Asian Population

Abstract

Background: Abdominal Aortic Aneurysms (AAA) have a high risk of mortality if they rupture, making it one of the leading causes of death in elderly men. Hence, a better prediction method is necessary to identify high rupture risk patients. Methods: Retrospective data (clinical and AAA geometric parameters) were collected of 114 Asian AAA patients from an IRB approved database. We applied a novel machine learning algorithm (MLA) to predict AAA rupture by generating a decision tree of rupture risk predictors. For comparative analysis, a binary logistic regression (backward stepwise Wald) was also performed. The rupture status was the dependent variable for both methods; each method provided the highest weighted risk factors and effectiveness of classification as assessed by Receiver Operating Characteristic (ROC) curves. Results: For the MLA, the most significant risk factors were DAAA (maximum diameter) (> 57 mm), presence of hypertension, D4 (diameter at the iliac bifurcation) (> 23.7 mm), age (> 79 years), and D3 (right common iliac artery diameter) (> 14 mm), as seen in Fig. 1a. Similarly, the backward stepwise logistic regression analysis showed a strong association for rupture with DAAA (O.R. 2.635), L1 (distance between the lowest renal artery and the iliac bifurcation) (O.R. 1.658), D3 (O.R. 1.876), NL (AAA neck length) (O.R. 0.294), and age (O.R. 0.681). The MLA exhibited a lower classification accuracy (80.7% vs. 99.2%) and area under the curve (ROC analysis) (56.3% vs. 98.2%) when compared to the regression analysis, as seen in Fig. 1b. Conclusions: This is the first known report of predictive rupture risk in the Asian AAA patient population using MLA and statistical methods. Although the MLA did not provide improved classification accuracy compared to a standard statistical method, the risk factors obtained from both methods were similar. Future studies will aim at improving the prediction ability of MLA in a racially diverse AAA patient population.