[Paper Review] Assessment of Aortic Aneurysm Rupture Risk

Rafael Izbicki, Ann B. Lee+1 more|arXiv (Cornell University)|Jun 3, 2011

Aortic aneurysm repair treatments6 references3 citations

TL;DR

This study develops a risk prediction model for abdominal aortic aneurysm (AAA) rupture using 144 CT-derived geometric and hemodynamic features from patients in elective and emergent repair groups. By adapting logistic regression with customized loss functions and prior probability adjustments to address case-control sampling bias, the model identifies wall thickness, aneurysm size, and tortuosity as key predictors, improving risk stratification beyond diameter alone.

ABSTRACT

The rupture of an abdominal aortic aneurysm (AAA) is associated with a high mortality. When an AAA ruptures, 50% of the patients die before reaching the hospital. Of the patients that are able to reach the operating room, only 50% have it successfully repaired (Fillinger et al, 2003). Therefore, it is important to find good predictors for immediate risk of rupture. Clinically, the size of the aneurysm is the variable vascular surgeons usually use to evaluate this risk. Patients with large aneurysms are often sent to surgery. However, many studies have shown that even small aneurysms can rupture and deserve attention as well. It is important to find good predictors of rupture that also avoid unnecessary surgery as all surgeries are associated with possible complications. Here, we use data obtained from 144 computed tomographies of patients from the Western Pennsylvania Allegheny Health System to predict the high risk of rupture of an aneurysm and also to examine which features are important for this goal.

Motivation & Objective

• To develop a predictive model for abdominal aortic aneurysm (AAA) rupture risk that improves upon diameter-based clinical criteria.
• To identify the most informative geometric and morphological features from 3D CT reconstructions for rupture prediction.
• To address the limitations of standard machine learning methods in a case-control study design with imbalanced groups and asymmetric misclassification costs.
• To enable clinically interpretable risk assessment that reduces unnecessary surgeries while identifying high-risk patients.
• To guide future research toward hemodynamic features and improved patient stratification.

Proposed method

• Adapted logistic regression with a modified risk function that accounts for asymmetric misclassification costs (higher penalty for misclassifying emergent patients).
• Incorporated prior probabilities from the target population (p0 = 0.886, p1 = 0.114) to correct for case-control sampling bias in model training.
• Used 144 patients with 28 geometric and morphological features extracted from 3D segmented CT images, including diameters, volume, surface area, wall thickness statistics, and tortuosity indices.
• Applied cross-validation with a modified score to estimate model performance, adjusting for unequal group proportions and loss asymmetry.
• Evaluated variable importance using univariate Mann-Whitney tests, standardized logistic regression coefficients, and Random Forest Gini importance.
• Treated the logistic regression cutoff as a tuning parameter to optimize risk prediction under clinical cost constraints.

Experimental results

Research questions

• RQ1 Which geometric and morphological features from 3D CT reconstructions are most predictive of AAA rupture?
• RQ2 How can traditional logistic regression be adapted to handle case-control sampling and asymmetric misclassification costs in medical risk prediction?
• RQ3 Can a model based on morphological features alone outperform diameter-based clinical thresholds in identifying high-risk AAA patients?
• RQ4 What characteristics do patients who are consistently misclassified across models share, and can these inform new predictors?
• RQ5 How do the estimated risks and confidence intervals of the model compare across different feature selection and model fitting approaches?

Key findings

• Wall thickness, particularly minimum and maximum values, emerged as highly significant predictors of rupture risk, even when univariate tests showed otherwise.
• Aneurysm size metrics—length (L), height (H), and volume (Vcm3)—were consistently important predictors across all models.
• Tortuosity, a measure of aortic shape complexity not directly tied to size, was identified as a strong predictor, suggesting its role in hemodynamic stress.
• The model achieved fair classification performance, but confidence intervals for risk estimates were wide, indicating limited improvement with more complex models given current sample size.
• Patients misclassified across multiple models tended to have predicted probabilities near the decision threshold, suggesting uncertainty in risk estimation for borderline cases.
• Inclusion of interaction terms did not improve model performance, supporting the use of simpler, interpretable models.