Can machine learning unravel unsuspected, clinically important factors predictive of long-term mortality in complex...

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Risk stratification is crucial for treatment decisions in complex coronary artery disease (CAD).
• Accurate prediction of long-term mortality is essential for guiding interventions like percutaneous coronary intervention (PCI) or coronary artery bypass grafting (CABG).

Purpose of the Study:

• To evaluate the effectiveness of machine learning (ML) algorithms in predicting long-term mortality in patients with complex CAD.
• To identify novel prognostic factors for long-term mortality using ML.

Main Methods:

• ML algorithms were applied to the SYNTAXES database, utilizing 75 pre-procedural variables.
• A 10-fold cross-validation approach was used to assess model discrimination and feature importance.
• The study included demographic, clinical, laboratory, imaging, and patient-reported outcome data.

Main Results:

• The ML model demonstrated acceptable discrimination for predicting 10-year mortality, with an area under the curve of 0.76.
• Important predictors identified include C-reactive protein, pre-procedural mental status, gamma-glutamyl transferase, and HbA1c.
• These factors were found to be potentially important, even if previously unsuspected.

Conclusions:

• ML algorithms can uncover significant, previously unrecognized prognostic factors for long-term mortality in CAD patients.
• Further validation using large-scale "big data" analyses is recommended to confirm these findings.
• ML holds promise for refining risk prediction in complex cardiovascular disease.