Machine learning approaches for risk prediction in aortic dissection: a systematic review and meta-analysis

Area of Science:

• Cardiovascular Medicine
• Medical Informatics
• Biostatistics

Background:

• Aortic dissection (AD) is a critical condition with high mortality.
• Traditional models struggle with AD's complex pathophysiology.
• Machine learning (ML) offers potential for improved AD risk prediction.

Purpose of the Study:

• To systematically review and meta-analyze ML models for AD outcome prediction.
• To evaluate the performance, quality, and clinical applicability of these ML models.
• To identify limitations and guide future research in ML for AD.

Main Methods:

• Systematic review and meta-analysis following PRISMA, CHARMS, TRIPOD guidelines.
• Searched six databases for studies on ML models predicting AD adverse outcomes.
• Assessed risk of bias using PROBAST and synthesized C-statistics (AUC).

Main Results:

• Forty studies included; ML models showed strong predictive performance for mortality and AKI.
• Pooled AUCs for early mortality (0.891) and long-term mortality (0.847) were high.
• Significant heterogeneity and high risk of bias (100% of studies) were noted, with limited external validation and reporting.

Conclusions:

• ML models show potential for AD risk stratification.
• Current evidence is insufficient for routine clinical implementation due to methodological weaknesses.
• Future research must focus on rigorous validation and transparent reporting for clinical utility.