Meta-Analysis of 40 Machine Learning Models for Aortic Dissection Risk

40 studies from a systematic review up to September 2025 synthesize performance metrics for machine learning models predicting aortic dissection outcomes. The dataset includes pooled C-statistics for outcomes like early mortality, long-term mortality, and acute kidney injury, with performance metrics and bias assessments. It was compiled by author Yijun Mao following PRISMA and CHARMS guidelines.

Use Cases

• Compare pooled C-statistics (AUC) for early mortality versus long-term mortality across the 40 included studies.
• Analyze the distribution of machine learning algorithms (e.g., random forest, XGBoost) used for predicting specific complications like acute kidney injury.
• Assess the correlation between PROBAST risk-of-bias ratings (high/unclear) and reported model performance metrics.
• Examine reporting adherence rates (e.g., 78.7% average TRIPOD compliance) across different outcome categories.

Strengths

• Includes data from 40 distinct studies identified through a systematic review of six major databases.
• Provides quantitative performance synthesis, such as a pooled C-statistic of 0.891 for early mortality prediction.
• Documents methodological assessments, including that 100% of studies were at high or unclear risk of bias per PROBAST.

Limitations

• Significant heterogeneity reported across analyses (I² = 61.3–78.8%), indicating high variability in model performance.
• Critical reporting shortcomings are present, with 82.5% of studies lacking sample size justification and 72.5% omitting full model specifications.
• Limited external validation, with only 5 out of the 40 studies employing robust external validation methods.

Provenance

Source

figshare, authored by Yijun Mao.

Collection Method

Systematic review and meta-analysis following PRISMA, CHARMS, and TRIPOD guidelines, registered with PROSPERO.

Time Range

Literature search from database inception to September 30, 2025.

Freshness

Last updated March 2026, with literature search covering studies up to September 2025.

Geography

null