Predicting Long-Term Survival after Endovascular Aneurysm Repair Using Machine Learning-Based Decision Tree Analysis

Abstract

Objective

Endovascular aneurysm repair (EVAR) has become a preferred method for treating abdominal aortic aneurysms (AAA) due to its minimally invasive approach. However, identifying factors that influence long-term patient outcomes is crucial for improving prognosis. This study investigates whether machine learning (ML)-based decision tree analysis (DTA) can predict long-term survival (over 5 years postoperatively) by uncovering complex patterns in patient data.

Methods

We retrospectively analyzed data from 142 patients who underwent elective EVAR for AAA at Tokyo Medical University Hospital between October 2013 and July 2018. The dataset comprised 24 variables, including age, gender, nutritional status, comorbidities, and surgical details. The decision tree classifier was developed and validated using Python 3.7 and the scikit-learn toolkit.

Results

DTA identified poor nutritional status as the most significant predictor, followed by compromised immunity, active cancer, octogenarians, chronic kidney disease, and chronic obstructive pulmonary disease. The decision tree identified 9 terminal nodes with probabilities of long-term survival. Four of these terminal nodes represented groups of patients with a high probability of long-term survival: 100%, 84%, 77%, and 60%, whereas the other 5 terminal nodes represented groups of patients with a low probability of long-term survival: 17%, 25%, 30%, 45%, and 47%. The model achieved a moderately high accuracy of 76.1%, specificity of 72.4%, sensitivity of 81.8%, precision of 65.2%, and area under the receiver operating characteristic curve of 0.84.

Conclusion

ML-based DTA effectively predicts long-term survival after EVAR, highlighting the importance of comprehensive preoperative assessments and personalized management strategies to improve patient outcomes.