Meta-Analysis of Machine Learning Models for Aortic Dissection Risk

Forty studies on machine learning for aortic dissection risk prediction are systematically reviewed. The meta-analysis synthesizes performance metrics like pooled C-statistics for outcomes including early mortality, long-term mortality, and acute kidney injury. It was authored by Yijun Mao and follows PRISMA, CHARMS, and TRIPOD guidelines.

Use Cases

• Analyze the pooled C-statistic performance across outcomes like early mortality and acute kidney injury to compare model discriminative ability.
• Assess risk of bias using PROBAST results, where 100% of the 40 studies were at high or unclear risk.
• Evaluate adherence to TRIPOD reporting standards, noting specific shortcomings in predictor definitions and sample size justification.
• Compare the performance of specific ML algorithms like random forest, SVM, and XGBoost for predicting various adverse outcomes.

Strengths

• Includes 40 studies sourced from six major databases, providing a broad evidence base.
• Reports specific pooled C-statistics, such as 0.891 for early mortality prediction.
• Follows established systematic review protocols including PRISMA, CHARMS, and TRIPOD.

Limitations

• Significant heterogeneity was observed in analyses, with I² values ranging from 61.3% to 78.8%.
• All 40 included studies were assessed as having high or unclear risk of bias, primarily in the analysis domain.
• Reporting is suboptimal, with critical details like predictor definitions missing in 62.5% of studies.

Provenance

Source

Systematic review and meta-analysis of studies from PubMed, Web of Science, Cochrane Library, Embase, CNKI, and Wanfang.

Collection Method

Literature search from database inception to September 30, 2025, following PRISMA guidelines.

Time Range

Studies from database inception to September 30, 2025.

Freshness

null

Geography

null