Machine Learning Predictive Models for Prognosis in Patients Undergoing Endovascular Abdominal Aortic Aneurysm Repair.

BACKGROUND

An increasing number of patients with abdominal aortic aneurysms (AAAs) are opting for endovascular aneurysm repair (EVAR), and predicting postoperative survival is important for patient management. The development of a postoperative prognostic model using machine learning (ML) can be effective in predicting postoperative survival, and research on this issue needs to be further enhanced. This study aims to establish predictive models for prognosis in AAA patients after EVAR.

METHODS

Perioperative and follow-up information of 163 AAA patients were collected in Beijing Hospital who underwent EVAR from January 2016 to April 2023. The patients were divided into a training set and a test set in a ratio of 7:3. ML methods such as least absolute shrinkage and selection operator (LASSO) regression, random forests, linear discriminant analysis, naive Bayes, K-nearest neighbor algorithm, support vector machines, and decision trees were selected to build prediction models, and these models were evaluated by the receiver operating characteristic curves (ROC).

RESULTS

The study cohort comprised 163 patients with a mean age of 72 ± 8.4 years, 33 (20.2%) patients died during the follow-up period, and the majority were male (88.3%). Patients were categorized into survivors (n = 130) and nonsurvivors (n = 33). LASSO regression selected chronic obstructive pulmonary disease (COPD), maximum diameter of aneurysm to body mass index ratio (DBR), and stroke history. The area under the ROC (AUC) for the training and test sets of nomogram was 0.77 and 0.75, respectively. K-Nearest Neighbors (KNN) was the most effective ML algorithm, with the AUC of 0.85 and 0.81 superlatively for the training and test sets.

CONCLUSION

Stroke, DBR, and COPD can predict prognosis after EVAR in AAA patients. KNN is better than other algorithms in the study.