Diagnostic performance of artificial intelligence-based myocardial ischemia determination compared to FFR-CT across key clinical cohorts: a pooled analysis of four cohorts

European Heart Journal, vol. 46

Abstract

Abstract Introduction Artificial intelligence-guided quantitative coronary computed tomography angiography (AI-QCTIschemia), is a non-invasive tool to evaluate myocardial ischemia associated with coronary artery disease on a per vessel basis. Prior individual cohorts have demonstrated a strong diagnostic performance of AI-QCTIschemi. as compared to invasive fractional flow reserve (FFR). Purpose To evaluate the diagnostic performance of AI-QCTIschemia versus fractional flow reserve computed tomography (FFR-CT), using invasive FFR as the reference standard. Patient-level data from four cohort studies were pooled to enable assessment of clinically important subgroups: men, women, patients ≤60 or >60 years, and those with or without diabetes. Methods This analysis pooled patient and vessel level data from 592 patients and 1575 vessels from four cohorts undergoing CCTA with comparison of AI-QCTIschemia and FFR-CT. FFR-CT and invasive FFR <0.8 were considered positive for ischemia. Sensitivity, specificity, and AUC were compared between subgroups using Generalized Estimating Equations models and DeLong’s test. Uninterpretable test results were considered positive. Results Mean age was 62 years, and 32% were females. A total of 356 (60.1%) of vessels had >50% stenosis, with 230 (38.9%) involving the proximal LAD. At the vessel level, 494 (28.6%) were characterized as mild stenosis (25-49%), 270 (16%) moderate (50-69%), and 173 (10%) severe (70-99%). These results were then subclassified based on gender, age (>60 or ≤60), and those with or without diabetes. Across all three key clinical cohorts, AI-QCTIschemia demonstrated higher diagnostic accuracy for detection of vessel level ischemia. Conclusion In this pooled analysis, we demonstrated that AI-QCTIschemia a model which emphasizes vessel characterization of coronary artery disease to predict myocardial ischemia showed higher diagnostic accuracy than a computational fluid dynamics-based prediction model to predict vessel level ischemia across clinically relevant subgroups: women, older adults, and diabetics.

Authors 11

1. T Beckmann

   University of Tennessee Medical Center

   Affiliation as printed

   University of Tennessee Medical Center, Cardiology , Knoxville ,

2. Tyler Coombes

   University of Tennessee Medical Center

   Affiliation as printed

   University of Tennessee Medical Center, Cardiology , Knoxville ,

3. R Dieter Iv

   University of Tennessee Medical Center

   Affiliation as printed

   University of Tennessee Medical Center, Cardiology , Knoxville ,

4. A R Van Rosendael Aachen

   Leiden University Medical Center

   Affiliation as printed

   Leiden University Medical Center, Cardiology , Leiden ,

5. R. Karlsberg

   Cedars-Sinai Smidt Heart Institute

   Affiliation as printed

   Cedars-Sinai Smidt Heart Institute, Cardiovascular Research Foundation of Southern California , Los Angeles ,

6. J Park
   Affiliation as printed

   Cleerly Inc., Biostatistics , Denver ,

7. Ibrahim Danad

   Amsterdam University Medical Centers

   Affiliation as printed

   Amsterdam UMC, Cardiology , Amsterdam ,

8. Paul Knaapen

   Amsterdam University Medical Centers

   Affiliation as printed

   Amsterdam UMC, Cardiology , Amsterdam ,

9. Asim Rizvi

   Cornell University · Weill Cornell Medicine · Weill Cornell Medical Center

   Affiliation as printed

   New York Presbyterian Weill Cornell, Cardiology , New York ,

10. George E Wesbey

    Scripps Clinic

    Affiliation as printed

    Scripps Clinic, Department of Cardiovascular Computed Tomography , La Jolla ,

11. Kashif Abbas Shaikh

    University of Tennessee Medical Center

    Affiliation as printed

    University of Tennessee Medical Center, Cardiology , Knoxville ,