Abstract 14740: Machine Learning Regression for Prediction of Abdominal Aortic Aneurysm Growth
Machine-Learning-Regression zur Vorhersage des Wachstums eines abdominalen Aortenaneurysmas
Circulation · 02.11.2026 · 154 (Suppl_1)
DOI: 10.1161/circ.154.suppl_1.14740 ↗Auf einen Blick
Bei 4.399 Patient:innen sagte ein ML-Modell die künftige AAA-Größe etwas besser voraus als lineare Regression. Der Zugewinn war gering; die maximale Aneurysmadurchmesser war der wichtigste Prädiktor.
Fragestellung
Verbessern Machine-Learning-Modelle die Vorhersage der künftigen AAA-Größe gegenüber linearer Regression, und bringt Vorwachstum zusätzlichen Nutzen?
Methode
Retrospektive Kohortenstudie mit 4.399 Patient:innen; AAA-Messungen aus CT, MRT und Ultraschall von 1993 bis 2025; Vergleich von LR und ML; interne Validierung und SHAP-Analysen.
Ergebnisse
16.101 Bildgebungsstudien. Erstes Modell: SVR MAE 0,38 cm, RMSE 0,58 cm, R 2 0,62; LR 0,42 cm, 0,60 cm, 0,60. Zweites Modell: SVR MAE 0,37 cm, RMSE 0,57 cm, R 2 0,64; LR 0,42 cm, 0,60 cm, 0,60.
Schlussfolgerung der Autor:innen
ML-Regression sagte die künftige AAA-Größe innerhalb von 0,38 cm vorher, mit nur geringem Vorteil gegenüber LR. Frühere Wachstumsverläufe lieferten zusätzlich nur wenig Vorhersagewert.
Kernaussagen
- SVR besser als LR, aber nur gering
- Maximaler AAA-Durchmesser war der Hauptprädiktor
- Historische Wachstumsdaten brachten wenig Zusatznutzen
Deutsche Zusammenfassung: KI-gestützt aus dem Original-Abstract erstellt; alle Zahlen wurden automatisch mit dem Original abgeglichen. Maßgeblich ist der Original-Abstract.
Kurz gefasst
Fachartikel in der Zeitschrift Circulation, erschienen am 02.11.2026. Verfasst von Hala Deeb und 8 weiteren Autor:innen.
Abstract (Originalsprache)
Introduction
Abdominal aortic aneurysm (AAA) surveillance guidelines are primarily based on maximum aneurysm diameter despite substantial variability in individual growth patterns. We evaluated whether machine learning (ML) models improve prediction of future AAA size compared with linear regression (LR), and whether incorporation of prior aneurysm growth information provides additional predictive value beyond diameter alone.
Methods
This retrospective cohort study identified patients with AAA across Mayo Clinic Enterprise using ICD-9 and ICD-10 codes. Longitudinal AAA measurements were extracted from CT, MRI, and ultrasound reports from 1993 to 2025. Patients with thoracoabdominal aortic aneurysms, genetic syndromes, prior aortic surgery, initial AAA diameter > 5.5 cm, or fewer than three pre-intervention imaging studies were excluded. Two modeling approaches were evaluated: (1) using current AAA diameter and time to subsequent imaging (days), and (2) the same variables plus overall and most recent growth rates (cm/year). LR and ML models were compared. Model performance was assessed using mean absolute error (MAE), root mean squared error (RMSE), and coefficient of determination (R 2 ). Grouped cross-validation, hyperparameter tuning, bootstrap internal validation, and SHapley Additive exPlanations (SHAP) analyses were performed.
Results
The cohort consisted of 4,399 patients with 16,101 imaging studies. The median initial AAA size was 3.6 cm (IQR 3.2-4.3), the median final AAA size was 4.8 cm (IQR 3.9-5.4), and the median follow-up was 955 days. In the first modeling approach, which incorporated AAA diameter and time to subsequent imaging, the best-performing ML model (SVR) achieved an MAE of 0.38 cm, RMSE of 0.58 cm, and R 2 of 0.62, compared with 0.42 cm, 0.60 cm, and 0.60, respectively, for LR. In the second modeling approach, which also incorporated overall and most recent AAA growth rates, the best-performing ML model (SVR) achieved an MAE of 0.37 cm, RMSE of 0.57 cm, and R 2 of 0.64, compared with 0.42 cm, 0.60 cm, and 0.60, respectively, for LR (figure 1). SHAP analysis identified maximal AAA diameter as the dominant predictor, while prior diameters contributed comparatively little additional information. Bootstrap validation demonstrated stable ML model performance (mean R 2 0.638 ± 0.055; RMSE 0.563 ± 0.056 cm; MAE 0.375 ± 0.020 cm).
Conclusion
ML regression predicted future AAA size within 0.38 cm, with only modest improvement over LR. Notably, historical AAA growth trends added minimal predictive value beyond current maximum diameter alone. These findings suggest that current AAA diameter captures much of the information needed to predict future aneurysm size, while historical growth information provides limited additional predictive value.
Zitieren
APA
Deeb, H., Brandt, M., Barzani, I. K., Casanegra, A. I., Henkin, S., McBane, R. D., Wennberg, P. W., Kuczmik, W., & Houghton, D. (2026). Abstract 14740: Machine Learning Regression for Prediction of Abdominal Aortic Aneurysm Growth. Circulation, 154(Suppl_1). https://doi.org/10.1161/circ.154.suppl_1.14740
Vancouver
Deeb H, Brandt M, Barzani IK, Casanegra AI, Henkin S, McBane RD, et al. Abstract 14740: Machine Learning Regression for Prediction of Abdominal Aortic Aneurysm Growth. Circulation. 2026;154(Suppl_1). doi:10.1161/circ.154.suppl_1.14740
BibTeX
@article{Deeb2026,
author = {Deeb, Hala and Brandt, Michael and Barzani, Izad Khawajah and Casanegra, Ana I and Henkin, Stanislav and McBane, Robert D and Wennberg, Paul W and Kuczmik, Wiktoria and Houghton, Damon},
title = {Abstract 14740: Machine Learning Regression for Prediction of Abdominal Aortic Aneurysm Growth},
journal = {Circulation},
year = {2026},
volume = {154},
number = {Suppl_1},
doi = {10.1161/circ.154.suppl_1.14740}
}
EUVIMED