JAHA Publication

“Derivation and External Validation of a High-Sensitivity Cardiac Troponin-Based Proteomic Model to Predict the Presence of Obstructive Coronary Artery Disease”

Journal of the American Heart Association (JAHA)

August 6, 2020 (Print Issue)

Highlights

  • The HART CADhs panel consisted of clinical variables (male sex, age, prior percutaneous coronary intervention [PCI]) and three biomarkers (hsTroponin, adiponectin, kidney injury molecule-1). ​

  • The HART CADhs panel generated had an AUC of 0.85 for coronary stenosis ≥ 70% in the internal validation cohort and had an AUC of 0.86 in the external validation cohort. 

  • Results from dividing the score into 5 categories of predicted risk demonstrated an 89% NPV for a score of 1 and a 97% PPV for a score of 5 in the external validation cohort.

  • In patients who could not be ruled in nor ruled out for myocardial infarction via hs-cTnI (“indeterminate zone”), the HART CADhs panel had an area under the receiver operating characteristic curve of 0.88 (P<0.001).​

  • In patients with renal injury (creatinine ≥1.5 mg/dL) which confounds hs-cTnI, the AUC was 0.79 with a NPV 56% and PPV 87%.

  • As compared to cardiac stress testing, the CADhs panel was substantially more accurate for predicting severe CAD.  CADhs had an AUC of 0.85 compared to an AUC of 0.52 for stress testing in the internal validation cohort.

Study Overview

APPROACH

Investigators used machine learning to train and perform an internal validation in patients who were referred to the heart catheterization laboratory for coronary angiography and/or peripheral angiography.  An external validation was performed on patients presenting to the ED with suspected acute myocardial infarction.  A novel scoring system (HART CADhs panel) was developed to predict the presence of significant Coronary Artery Disease (CAD) stenosis.

GOAL

Identify clinical and biomarker predictors of clinically significant CAD in an at-risk population of subjects enrolled in the Catheter Sampled Blood Archive in Cardiovascular Diseases Study (CASABLANCA; Clinical Trials.Gov NCT00842868). This cohort was undergoing coronary angiography for numerous indications and was used with machine learning to train a model and then internally validate the model.

 

External validation of the model was conducted on patients presenting to an ED with chest pain and suspected acute myocardial infarction enrolled in the BACC (Biomarkers in Acute Cardiac Care) study (ClinicalTrials.Gov NCT02355457).

HYPOTHESIS

The addition of plasma biomarkers to known clinical risk factors may increase the accuracy of predicting clinically significant CAD.

METHODS​

The Massachusetts General Hospital CASABLANCA patients selected for this analysis consisted of the chronologically initial 911 patients who received a coronary angiogram. This cohort was randomly divided after into a training (derivation) cohort of 636 patients (70%) and a separate internal validation cohort of 275 patients (30%). 

 

A convenience sample of 241 patients presenting to the ED with suspected acute myocardial infarction were enrolled at University Medical Center Hamburg-Eppendorf in Germany and were used for external validation of the model. 

MACHINE LEARNING & BIG DATA
Candidate panels of biomarkers and clinical variables were generated and evaluated with machine learning statistical techniques, a subset of Artificial Intelligence (AI). The ultimate result of this process was a final test panel and scoring system predicting the outcome of the presence of significant obstruction in at least one coronary artery. The same panel (model) and cutoffs were externally validation on a separate cohort.

Multivariable logistic regression evaluated the performance of the model in the training set. Operating characteristics of the score were calculated, with sensitivity, specificity, positive and negative predictive value (PPV, NPV) generated. The range of the diagnostic model was the partitioned into five different risk levels, corresponding to the multiple levels of risk for CADhs.

Key Findings

SCORING SYSTEM
The HART CAD panel consisted of three clinical variables (age, male sex, prior percutaneous coronary intervention) and three biomarkers (hs-cTnI, adiponectin, and kidney injury molecule-1).  The HART CADhs panel generated had an AUC of 0.85 for coronary stenosis ≥ 70% in the internal validation cohort and had an AUC of 0.86 in the external validation cohort. 

Results from dividing the score into 5 categories of predicted risk demonstrated an 89% NPV for a score of 1 and a 97% PPV for a score of 5 in the external validation cohort. 

In patients who could not be ruled in nor ruled out for myocardial infarction via hs-cTnI (“indeterminate zone”), the HART CADhs panel had an area under the receiver operating characteristic curve of 0.88 (P<0.001).​

 

In patients with renal injury (creatinine ≥1.5 mg/dL) which confounds hs-cTnI, the AUC was 0.79 with a NPV 56% and PPV 87%.

 

As compared to cardiac stress testing, the CADhs panel was substantially more accurate for predicting severe CAD.  CADhs had an AUC of 0.85 compared to an AUC of 0.52 for stress testing in the internal validation cohort.

 

COMPARISON TO STRESS TESTING

Among the patients undergoing cardiac stress testing per standard of care, the CADhs score was substantially more accurate for predicting angiographically severe CAD (0.85 vs. 0.52; <0.001 for difference in AUC).

​ 

OTHER FINDINGS

Importantly, The HART CADhs panel performed particularly well in women, and while one element of the score was prior PCI, the score performance was similar in subjects without a history of CAD. It was also accurate for predicting those in the “indeterminate zone” in patients who had their myocardial infarction neither ruled in nor ruled out via hs-cTnI testing.

Conclusions

A clinical and biomarker scoring strategy, incorporating hs c-TnI, with high accuracy to reliably diagnose presence of severe epicardial CAD has been developed.  Advantages of a reliable clinical and biomarker score for diagnosing CAD presence include the fact such a technology can be widely disseminated in a cost-effective manner, easily interpreted, and may assist in the evaluation of outpatients presenting with symptoms of stable angina.  Additionally, the test panel may be useful for identifying patients presenting with chest pain with accurate triage decisions in the hospital setting and may be associated with a well-defined therapeutic strategy to reduce the risk for CAD-related events.

Scroll Down for Presentation for HART CADhs

HART CADhs JAHA Publication