Abstract
Commonly used tools to assess the probability of obstructive-coronary artery disease (CAD) were derived based on Caucasian cohorts, with their performance in China is still unknown. Furthermore, most were established based on non-laboratory variables, contributing to the limited predictive ability to some extent. Thus, we developed and internally validated a laboratory-based model with data from a Chinese cohort of 8963 inpatients, with suspected stable chest pain, referred to catheter-based coronary angiography (CAG) from September 2007 to April 2019, and then compared the present model's performance with the four most commonly used prediction tools, Coronary Artery Disease Consortium 1/2 Score (CAD1/2), Duke clinical score (DCS), and Diamond-Forrester score (DF). The final model was developed by random forest method, including 8 predictors derived from 70 variables. Five-fold cross-validation was performed to evaluate the model's prediction accuracy. In the external validation set, the present model showed a superior area under the receiver-operating curve (0.816), followed by DCS (0.66), CAD2 (0.61), CAD1 (0.59) and at last DF (0.58), respectively. Furthermore, the present model correctly classified 74.4% of obstructive-CAD patients as high-risk, and correctly classified more than one third of non-obstructive-CAD patients as low-risk. The present model's net reclassification improvement (NRI) showed a significant positive reclassification over CAD1 (NRI = 0.60, P < 0.001), DF (NRI = 0.59, P < 0.001), CAD2 (NRI = 0.57, P < 0.001), and DCS (NRI = 0.43, P < 0.001). Decision curve analysis demonstrated that the present model provided a larger net benefit compared with CAD1/2, DCS, and DF. In conclusion, the novel model, using 8 laboratory and non-laboratory variables, performed well in risk stratifying patients with suspected chest pain regarding the presence of obstructive-CAD in the present Chinese cohort.
