Development and validation of a machine-learning prediction model to improve abdominal aortic aneurysm screening

Area of Science:

• Cardiovascular Medicine
• Artificial Intelligence in Healthcare
• Preventive Medicine

Background:

• Screening for abdominal aortic aneurysms (AAAs) is recommended but adoption remains low due to low detection rates in unscreened populations.
• Identifying individuals at high risk for AAAs is crucial to improve screening effectiveness and patient outcomes.

Purpose of the Study:

• To develop and validate a machine learning (ML) model to identify individuals at highest risk for AAAs.
• To increase the detection rate of AAA screenings by targeting high-risk individuals.

Main Methods:

• A retrospective cohort study utilized longitudinal medical records from an institutional database to train an ML model.
• Patients aged 65-75 years with current or past smoking history were stratified by sex and smoking status.
• The model was adjusted for fairness between sexes and validated using six-fold cross-validation.

Main Results:

• The ML algorithm identified 41 factors associated with AAAs, including novel factors.
• Validation on 18,660 patients over 2 years identified 314 AAAs.
• The algorithm demonstrated a 200% lift in AAA detection compared to guideline-based screening, with statistically significant increases across all cutoff points.

Conclusions:

• A novel ML-based algorithm effectively identifies high-risk individuals for AAAs.
• Targeted screening using this algorithm significantly increases AAA detection rates by 200% compared to standard guidelines.
• The automated process, integrated into workflows, improves screening rates and yield for high-risk individuals.