Sex-specific machine learning classification models improve outcome prediction for abdominal aortic aneurysms.

Area of Science:

• Cardiovascular Medicine
• Medical Imaging Analysis
• Machine Learning in Healthcare

Background:

• Abdominal aortic aneurysm (AAA) is a life-threatening condition with high rupture mortality.
• Female patients experience AAA rupture more frequently than males, despite lower prevalence.
• Current diameter-based criteria for surgical intervention have limitations, as some aneurysms rupture below thresholds.

Purpose of the Study:

• To investigate the impact of sex-specific machine learning (ML) models on predicting patient outcomes for abdominal aortic aneurysms (AAA).
• To compare the performance of sex-specific ML models against a general, sex-agnostic model.
• To assess the effect of equalizing sample sizes in ML models on prediction accuracy.

Main Methods:

• Acquired CT image data from 537 patients (159 female, 378 male).
• Utilized clinical, biomechanical, and morphological features for ML model training.
• Trained and tested general, male-specific, and female-specific ML models using a 20% holdout set.

Main Results:

• Both female-specific and male-specific ML models demonstrated superior predictive performance compared to the general model for their respective sexes.
• Equalizing sample sizes between male and female patients improved prediction outcomes for females.
• Performance for male patients was not negatively impacted by equalizing sample sizes.

Conclusions:

• Sex-specific ML models show significant promise for improving AAA patient outcome prediction.
• Addressing underrepresentation of female patients in AAA datasets through sex-specific models and data balancing is crucial.
• Consideration of population representation in datasets is vital for reducing bias in medical ML applications.