Predicting outcome in coronary disease. Statistical models versus expert clinicians

Area of Science:

• Cardiology
• Medical Statistics
• Health Informatics

Background:

• Accurate long-term prognosis is crucial for managing coronary artery disease (CAD).
• Clinical predictions of patient outcomes can be subjective and variable.
• Data-driven statistical models offer a potential alternative for objective prognostic assessment.

Purpose of the Study:

• To compare the accuracy of prognostic predictions for CAD patients between a multivariable statistical model and senior cardiologists.
• To assess the reliability of one- and three-year survival and infarct-free survival predictions.
• To evaluate interphysician variability in clinical prognostic assessments.

Main Methods:

• A Cox regression model was developed using a dataset of medically treated CAD patients.
• Five senior cardiologists provided one- and three-year survival and infarct-free survival predictions for 100 patients each, based on detailed case summaries.
• Model predictions were compared against actual patient outcomes and contrasted with physician predictions.

Main Results:

• The statistical model's predictions showed better correlation with actual patient outcomes than physician predictions.
• For three-year survival, model correlations (0.61) surpassed physician correlations (0.49).
• For three-year infarct-free survival, model correlations (0.48) were superior to physician correlations (0.29), with substantial interphysician variability observed.

Conclusions:

• Data-based statistical models provide more accurate prognostic predictions in coronary artery disease than experienced clinicians.
• Cox regression models offer significant prognostic information, outperforming individual physician predictions.
• The study highlights the potential of statistical modeling to enhance prognostic accuracy and reduce variability in CAD patient care.