Differential risk reclassification improvement by exercise testing and myocardial perfusion imaging in patients with suspected and known coronary artery disease.

OBJECTIVE

The objective of this study is to compare the incremental prognostic and net risk reclassification value of exercise testing alone vs exercise myocardial perfusion imaging (MPI) for estimating the risk of death in patients with suspected and known coronary artery disease (CAD).

METHODS

6702 patients with suspected CAD and 2008 with known CAD had treadmill exercise MPI and were followed for 2.5 ± 0.9 years for the occurrence of all-cause death. The estimation of risk of death and net reclassification improvement (NRI) were examined in three models. Model 1: clinical variables; Model 2: model 1+Duke Treadmill Score; and Model 3: model 2+ MPI variables. Risk estimates were categorized as <1%, 1-3%, and >3% risk of death per year.

RESULTS

In patients with suspected CAD, the global Chi-square for predicting risk of death increased significantly for Model 2 compared to Model 1 (74.78 vs 63.86 to (P = .001). However, adding MPI variables in Model 3 did not further improve predictive value (Chi-square 79.38, P = .10). In patients with suspected CAD risk, reclassification improved significantly in Model 2 over Model 1 (NRI = 0.12, 95% CI 0.02 to 0.22, P = .019), but not in Model 3 (NRI = 0.0009, 95% CI -0.072 to 0.070; P = .98). In contrast, in patients with known CAD Model 2 did not yield significant improvements for predicting risk and risk reclassification compared to Model 1. However, global Chi-square of Model 3 was significantly higher than that of Model 2 (30.03 vs 6.56, P < .0001) with associated significant reclassification improvement (NRI = 0.26 95% CI 0.067 to 0.46. P = .0084).

CONCLUSION

Risk reclassification by diagnostic testing is importantly influenced by baseline characteristics of patient cohorts. In patients with suspected CAD, NRI is predominately achieved by exercise variables, whereas in patients with known CAD, greatest NRI is obtained by MPI variables.