Independent and incremental prognostic value of tests performed in hierarchical order to evaluate patients with suspected coronary artery disease. Validation of models based on these tests.

Abstract

BACKGROUND The additive prognostic value of tests done in a hierarchical order for the detection of coronary artery disease (CAD) is not always known. The principal goal of this study, therefore, was to assess the incremental prognostic value of data obtained in succession (clinical, exercise stress testing, 201Tl imaging, and coronary angiography) in patients with suspected CAD. A second goal was to develop models for determining prognosis based on results of these tests and to test the clinical validity of these models in unrelated patients. METHODS AND RESULTS Data from two groups of patients who had undergone such evaluation and had been followed for a mean of 4.4 years were analyzed. There were 204 patients from Massachusetts General Hospital (MGH) and 299 from the University of Virginia (UVA). There were 20 deaths and 21 nonfatal infarctions in the MGH group and 41 deaths and nine infarctions in the UVA group. Both univariate and multivariate Cox regression analyses were performed to assess the individual and incremental prognostic value of these tests. In both groups, 201Tl imaging provided significant additional prognostic information compared with clinical and exercise stress test data (p less than 0.05). At MGH, where the lung/heart 201Tl ratio had been analyzed, coronary angiography did not provide additional prognostic information. In this group of patients, the combination of clinical and exercise 201Tl variables provided greater prognostic information than the combination of clinical and angiographic data (p less than 0.001). In the UVA cohort, in which the lung/heart ratio had not been analyzed, coronary angiography provided incremental prognostic information compared with clinical and exercise 201Tl data alone (p less than 0.05). When models developed using data from either sample were applied to the other unrelated sample, there was often close agreement between the overall observed rates and those predicted by the models. This was also true for the low-risk and high-risk subgroups. Some models, however, did not perform as well as other models, which suggests that models that do well in one sample may not always be generalized to other groups. CONCLUSIONS Tests performed in hierarchical order for the evaluation of suspected CAD provide additional prognostic information. Models developed using clinically relevant combinations of test results obtained from different patient populations are frequently able to predict absolute levels of survival in unrelated but similar samples.