An XGBoost model using high sensitive troponin T predicted obstructive CAD with 81.9% AUC, 92.7% sensitivity, and correctly reclassified 42% without CAD.
Does a machine learning prediction model accurately identify patients with suspected NSTE-ACS or UAP who have a low probability of obstructive epicardial coronary artery disease?
A prediction model incorporating high-sensitive troponin T, sex, age, LDL, and ECG findings can accurately identify patients with suspected NSTE-ACS who have a low probability of obstructive CAD, potentially reducing unnecessary invasive angiographies.
Absolute Event Rate: 0% vs 0%
Abstract Background A significant proportion of patients presenting with symptoms and signs suggestive of non-ST-elevation myocardial infarction (NSTE-ACS) or unstable angina pectoris (UAP) do not have obstructive epicardial coronary artery disease (CAD) on invasive coronary angiography. An objective clinical tool to identify NSTE-ACS and UAP patients with a low pre-test probability of obstructive epicardial CAD may be useful to guide the down-stream diagnostic work-up. Purpose To develop a prediction model for identification of patients admitted to the cardiac care unit (CCU) with a low probability of obstructive coronary artery disease. Methods We merged and analyzed data from the Swedish Web-system for Enhancement and Development of Evidence-based care in Heart Disease Evaluated According to Recommended Therapies (SWEDEHEART). Patients referred to invasive coronary angiography based on symptoms and signs suggestive of NSTE-ACS or UAP) and where high-sensitive Troponin T was used as biomarker were included for analyses. The outcome was based on the first coronary angiography during index admission and was defined as obstructive CAD defined as at least one stenosis with 50 diameter stenosis and/or revascularization. Machine learning (XGBoost) and logistic regression modelling was applied considering pre-defined variables. Results The final study population included 73.244 patients from 2010 to 2022 with mean age 68±12 of which 25.144 (34%) were female. The prevalence of non-obstructive epicardial CAD was 25%. The gradient boosting model slightly outperformed the logistic regression model with an area under the receiver operating curve (AUC) of 81.9 (95%CI: 81.3-82.6) vs. AUC 80.0 (95% CI: 78.7-81.2). The top five most important factors based on shapley additive explanations (SHAP) values were peak high sensitive troponin T, sex, age, low density lipoprotein and ECG (ST-depression). Using a probability cut-off (0.52) ensuring a negative predictive value of 0.7 in the training cohort, total of 2.294 (42%) without CAD at invasive coronary angiography were correctly re-classified in the validation cohort with a sensitivity of 92.7% (95%CI: 92.3-93.1), specificity 42.3% (95%CI: 41.0-43.6), positive predictive value 83.0 (95%CI: 82.5–83.6) and negative predictive value 65.5% (95%CI: 63.9-67.1). Conclusion We developed a high sensitive troponin T based prediction model that may have potential for planning diagnostic work-up in patients presenting with symptoms suggestive of NSTE-ACS or UAP but with a low probability of obstructive epicardial coronary artery disease. The latter could include a first-line coronary computed tomography angiography strategy to decrease invasive coronary angiography procedures without need for intervention. Further validation, including external validation, is required.Top factors for predicting CAD CAD prevalence across model categories
Westra et al. (Sat,) reported a other. An XGBoost model using high sensitive troponin T predicted obstructive CAD with 81.9% AUC, 92.7% sensitivity, and correctly reclassified 42% without CAD.