A validated simple model to predict coexistent coronary disease in patients undergoing mitral valve surgery

The primary limitation of the American Heart Association/American College of Cardiology guidelines is specificity. To improve the selection process, we proposed a simple additive model including age (1 point for every 5 years above 50), male sex (2 points), hypercholesterolemia (2 points), angina (3 points), and electrocardiographic evidence of ischemia (3 points). We recommend screening angiography at 3 or more points. This model was previously derived from 359 patients at Papworth Hospital.

The validation cohort was a consecutive series of patients who underwent mitral valve surgery at the Royal Brompton Hospital. Preoperative coronary angiography reports were obtained, and coronary disease was defined as luminal narrowing of 50% in 2 or more views. Sensitivities and specificities were calculated for the American Heart Association/American College of Cardiology criteria, the simple additive model, and a logistic regression model. Receiver operating characteristic curves were used to validate accuracy and compare discrimination with logistic regression.

From 1998 through 2003, angiographic details were available for 342 (86%) of 396 patients who underwent mitral valve surgery. The sensitivity and specificity of the American Heart Association/American College of Cardiology guidelines were 100% and 5%, respectively; those of the simple additive model were 91% and 44%, respectively; and those of logistic regression were 93% and 41%, respectively. The receiver operating characteristic areas for the simple additive and logistic regression model were 0.78 (95% confidence interval, 0.73-0.84) and 0.80 (95% confidence interval, 0.74-0.85), respectively.

This is the third independent cohort to highlight the poor specificity of the American Heart Association/American College of Cardiology guidelines. Although high sensitivity is achieved, the cost is the majority of patients requiring screening angiography. Our validated simple model improved the specificity and selection; however, this was achieved at the expense of decreased sensitivity.