Machine Learning Models in the Prediction of Adverse Outcomes in Peripheral Arterial Disease: Meta-Analysis

Area of Science:

• Cardiovascular research
• Medical informatics
• Machine learning applications in healthcare

Background:

• Peripheral arterial disease (PAD) poses significant risks for adverse cardiovascular and limb events.
• Accurate prediction of these outcomes is crucial for patient management.
• Existing statistical methods may have limitations in capturing complex predictive patterns.

Purpose of the Study:

• To systematically review and compare the performance of machine learning (ML) models against traditional statistical approaches for predicting adverse outcomes in PAD patients.
• To assess the current limitations and challenges in ML model development and validation for PAD.
• To provide insights into the future of predictive modeling in PAD management.

Main Methods:

• A comprehensive systematic search of major medical and trial databases (2000-2025) for studies using ML in PAD outcome prediction.
• Independent screening, data extraction, and quality/bias assessment using PROBAST and an AI/ML-specific tool.
• Modified Hierarchical Summary Receiver Operating Characteristic (HSROC) analysis and diagnostic odds ratio (DOR) calculation to compare ML and regression models for MACE and MALE prediction.

Main Results:

• Thirteen studies met inclusion criteria, evaluating various ML models and traditional regression.
• Gradient boosted models showed the highest predictive performance for major adverse limb events (MALE) with a DOR of 36.593.
• All evaluated ML models demonstrated moderate to high predictive ability, consistently outperforming traditional regression models (lowest DOR 3.326).
• Methodological quality varied, with 54% of studies rated as low risk of bias.

Conclusions:

• Machine learning techniques offer superior predictive power for adverse outcomes in PAD patients compared to traditional regression methods.
• ML models, particularly gradient boosted approaches, show significant promise for improving risk stratification in PAD.
• Further research focusing on robust ML model development and validation is warranted to optimize clinical application.