RESUMO
BACKGROUND: Abdominal aortic aneurysm (AAA) is one of the most common diseases in vascular surgery. Endovascular aneurysm repair (EVAR) can effectively treat AAA. It is essential to accurately classify patients with AAA who need EVAR. METHODS: We enrolled 266 patients with AAA who underwent EVAR. Unsupervised machine learning algorithms (UMLAs) were used to cluster subjects according to similar clinical characteristics. To verify UMLA's accuracy, the operative and postoperative results of the 2 clusters were analyzed. Finally, a prediction model was developed using binary logistic regression analysis. RESULTS: UMLAs could correctly classify patients based on their clinical characteristics. Patients in Cluster 1 were older, had a higher BMI, and were more likely than patients in Cluster 2 to develop pneumonia, chronic obstructive pulmonary disease, and cerebrovascular disease. The aneurysm diameter, neck angulation, diameter and angulation of bilateral common iliac arteries, and incidence of iliac artery aneurysm were significantly higher in cluster 1 patients than in cluster 2. Cluster 1 had a longer operative time, a longer length of stay in the intensive care unit and hospital, a higher medical expense, and a higher incidence of reintervention. A nomogram was established based on the BMI, neck angulation, left common iliac artery (LCIA) diameter and angulation, and right common iliac artery (RCIA) diameter and angulation. The nomogram was evaluated using receiver operating characteristic curve analysis, with an area under the curve of 0.933 (95% confidence interval, 0.902-0.963) and a C-index of 0.927. CONCLUSIONS: Our findings demonstrate that UMLAs can be used to rationally classify a heterogeneous cohort of patients with AAA effectively, and the analysis of postoperative variables also verified the accuracy of UMLAs. We established a prediction model for new subtypes of AAA, which can improve the quality of management of patients with AAA.
