The AI-based MPA model had higher AUC (0.757 vs 0.698, p=0.004) and lower false negative rate (0.4% vs 3.0%) for relevant ischemia than the 2024 ESC RF-CL model.
Does an artificial intelligence-based tool (MPA) improve the detection of ischemia compared to the 2024 ESC risk-factor clinical likelihood model in patients with suspected chronic coronary syndrome?
An AI-based memetic pattern-based algorithm significantly improved the discriminatory power for detecting relevant ischemia on PET compared to the newly proposed 2024 ESC risk-factor clinical likelihood model.
Absolute Event Rate: 0% vs 0%
Abstract Introduction Patients are referred to coronary artery disease (CAD) testing based on their pre-test probability (PTP). The novel risk factor-weighted clinical likelihood (RF-CL) proposed by the 2024 ESC guidelines on chronic coronary syndromes (CCS) uses symptoms, age, sex and risk factors. Artificial intelligence-based (AI) tools can incorporate many more variables, and could therefore improve PTP assessment. Purpose Study aim was to compare a novel AI tool with the RF-CL model. Methods RF-CL and a memetic pattern-based algorithm (MPA) were applied to 1657 consecutive patients with suspected CCS referred for 82Rb-Positron Emission Tomography (PET). Distribution across risk groups, receiver operator curve and test characteristics were calculated. Endpoints were defined as small ischemia (Summed Difference Score ≥2) and relevant ischemia (≥10% of myocardium). Results Mean age of the patients was 65±11 years; 43% were female. Median RF-CL was 11% 6, 19, respectively. Typical and atypical angina were present in 20% and 25%, respectively. Small ischemia was observed in 19%, relevant ischemia in 9%. Distribution of patients across risk categories is shown in figure 1. The MPA allocated also patients in the high and very high risk category. For small ischemia, AUC was similar (RF-CL: 0.689 (0.658 – 0.720) vs. MPA 0.706 (0.676 – 0.736), p = 0.227). For relevant ischemia, AUC was significantly higher for the MPA: 0.698 (0.654 – 0.742) vs. 0.757 (0.721 – 0.793), p = 0.004. As shown in table 1, sensitivity and negative predictive value (NPV) were higher for the MPA model. Consequently, the false negative rate was lower with the MPA for small (4.3% vs. 6.7%) and relevant ischemia (0.4% vs. 3.0%). Conclusion The MPA model showed a significantly better discriminatory power for detection of relevant ischemia in patients with suspected CCS compared to the proposed RF-CL model. In addition, relevant ischemia could be excluded with a very high degree of certainty. The MPA allocated patients more evenly across all five risk groups. Tools using artificial intelligence should be used more frequently in daily routine for assessment of PTP.
Frey et al. (Sat,) reported a other. The AI-based MPA model had higher AUC (0.757 vs 0.698, p=0.004) and lower false negative rate (0.4% vs 3.0%) for relevant ischemia than the 2024 ESC RF-CL model.