External validation of models for estimating pretest probability of coronary artery disease among individuals undergoing myocardial perfusion imaging

BackgroundClinical decisions regarding the appropriateness of noninvasive cardiac imaging among individuals with suspected coronary artery disease (CAD) rely heavily on the pretest probability of coronary artery disease (pCAD), often estimated from clinical prediction models. These models have not been validated among individuals undergoing noninvasive myocardial perfusion imaging (MPI) for suspected CAD. Thus, the objective of this study was to compare the extent of agreement and predictive performance between four published pCAD models among individuals undergoing positron emission tomography (PET MPI).MethodsThis cross-sectional study performed at a cardiac referral center included 2383 patients with stable symptoms undergoing PET MPI for the evaluation of suspected CAD. pCAD was estimated on a per-patient basis using four distinct pCAD estimation models. All pCAD estimates were calibrated to a common standard to allow fair comparisons of agreement and predictive performance. Pairwise pCAD model disagreement was defined as percent discordance in classifying patients as low versus intermediate pCAD (< 10% vs. ≥ 10%). Predictive performance was quantified by c-statistics with abnormal myocardial perfusion as a binary outcome.ResultsPairwise pCAD estimates demonstrated non-negligible disagreement with percent discordance between models ranging from 11% to 23%. Agreement worsened when higher thresholds for distinguishing low-intermediate pCAD were employed. All pCAD models demonstrated poor predictive performance for identifying abnormal stress perfusion with c-statistics ranging from 0.554 to 0.616.ConclusionspCAD estimation models showed suboptimal agreement and poor predictive performance in patients undergoing PET MPI. The transportability of pCAD models to MPI patients should be questioned and further evaluated in future studies.