Machine learning approaches for risk prediction after percutaneous coronary intervention: a systematic review and...

Area of Science:

• Cardiology
• Medical Informatics
• Data Science

Background:

• Accurate prediction of clinical outcomes after percutaneous coronary intervention (PCI) is crucial for patient management and risk mitigation.
• Traditional risk models offer modest predictive value, necessitating exploration of advanced methods.
• Machine learning (ML) models present a potential alternative for enhanced risk stratification in PCI patients.

Purpose of the Study:

• To systematically compare the predictive accuracy of ML models against traditional statistical methods for key clinical events post-PCI.
• To evaluate the performance of ML models in stratifying risk for all-cause mortality, major bleeding, and major adverse cardiovascular events (MACE).

Main Methods:

• A systematic review and meta-analysis adhering to PRISMA-P guidelines was conducted.
• Searches of PubMed, EMBASE, Web of Science, and Cochrane databases were performed up to November 1, 2023.
• Comparative discrimination was assessed using C-statistics, comparing 34 ML models with traditional methods across 13 observational studies involving over 4 million patients.

Main Results:

• ML models demonstrated a pooled C-statistic of 0.89 for all-cause mortality versus 0.86 for traditional methods (P=0.54).
• For major bleeding, ML models achieved a pooled C-statistic of 0.80 compared to 0.78 for traditional methods (P=0.02).
• ML models showed superior discrimination for MACE with a C-statistic of 0.83 versus 0.71 for traditional methods (P=0.007).

Conclusions:

• Machine learning models offer marginal improvements in predicting MACE and major bleeding following PCI compared to traditional risk scores.
• The clinical implementation of ML for peri-procedural risk stratification requires further investigation due to current methodological and validation limitations.
• Additional research is essential to address biases and enhance the reliability of ML models in clinical practice.