An externally validated model for predicting long-term survival after exercise treadmill testing in patients with suspected coronary artery disease and a normal electrocardiogram.

Background: The exercise treadmill test is recommended for risk stratification among patients with intermediate to high pretest probability of coronary artery disease. Posttest risk stratification is based on the Duke treadmill score, which includes only functional capacity and measures of ischemia.

Objective: To develop and externally validate a post-treadmill test, multivariable mortality prediction rule for adults with suspected coronary artery disease and normal electrocardiograms.

Design: Prospective cohort study conducted from September 1990 to May 2004.

Setting: Exercise treadmill laboratories in a major medical center (derivation set) and a separate HMO (validation set).

Patients: 33,268 patients in the derivation set and 5821 in the validation set. All patients had normal electrocardiograms and were referred for evaluation of suspected coronary artery disease.

Measurements: The derivation set patients were followed for a median of 6.2 years. A nomogram-illustrated model was derived on the basis of variables easily obtained in the stress laboratory, including age; sex; history of smoking, hypertension, diabetes, or typical angina; and exercise findings of functional capacity, ST-segment changes, symptoms, heart rate recovery, and frequent ventricular ectopy in recovery.

Results: The derivation data set included 1619 deaths. Although both the Duke treadmill score and our nomogram-illustrated model were significantly associated with death (P < 0.001), the nomogram was better at discrimination (concordance index for right-censored data, 0.83 vs. 0.73) and calibration. We reclassified many patients with intermediate- to high-risk Duke treadmill scores as low risk on the basis of the nomogram. The model also predicted 3-year mortality rates well in the validation set: Based on an optimal cut-point for a negative predictive value of 0.97, derivation and validation rates were, respectively, 1.7% and 2.5% below the cut-point and 25% and 29% above the cut-point.

Limitations: Blood test-based measures or left ventricular ejection fraction were not included. The nomogram can be applied only to patients with a normal electrocardiogram. Clinical utility remains to be tested.

Conclusion: A simple nomogram based on easily obtained pretest and exercise test variables predicted all-cause mortality in adults with suspected coronary artery disease and normal electrocardiograms.