Artificial Intelligence-based 5-lead 3D-vectorcardiography (5L3DVCG-AI) offers additional information over standard 12-lead electrocardiography (ECG) in the detection of coronary vascular disease (CVD) and identified those with significant stenoses in need for coronary intervention. 5L3DVCG-AI is under investigation as a new screening tool for CVD in prospective multinational trials.

We tested the hypothesis of 5L3DVCG-AI being able to detect patients with mild to overt signs and / or history of CVD, as diagnosed according to current guidelines.

In this monocentric exploratory retrospective register study, all consecutive raw data of 331 patients with 5L3DVCG-AI were included in the analysis. The Perfusion (P)-Factor for cardiac ischaemic pathologies, based on the P-Index including 731 parameters, e.g. QRS-T angle and in-house features calculated in time and frequency domains (e.g. beat moments), classified patients as high, medium or low risk for CVD. Diagnosis of CVD was based upon current guidelines by 2 independent cardiologists blinded to the 5L3DVCG-AI and categorised as exclusion of CVD (control), mild signs or overt CVD. Diagnostic accuracy was calculated validating the P-Index against clinical CVD and clinical course. Cardiovascular risk factor (CVRF) score was quantified with the modified PROCAM score.

In this sample of 331 patients (m:w 60:40%, age: 50.0 ± 19.8 years) of mixed ethnicity (caucasian, turkish, arabic, african, far-east) and moderate CVRF (2.1 ± 1.2), 71% were controls, 20% had mild signs of CVD and 9% overt CVD. Follow-up period was 16.2 ± 7.5 months. CVRF was significantly higher in CVD compared to controls (2.7 ± 0.5 vs. 2.0 ± 1.2, p<0.05), and P-Factor correlated with number of CVRF (p<0.01), with significantly higher CVRF in higher P-factor (KW p<0.001). P-factor indicated significantly more often higher risk in CVD compared to controls (0.78 ± 0.4 vs. 0.34 ± 0.48, p<0.01). P-Index from 5L3DVCG-AI at rest differentiated between CVD and controls (Chi2= 6.7, p=0.01). ROC curve showed correlation between P-Index and presence of CVD (R2=0.72, p<0.05). Expectedly, ECG at rest was not able to differentiate between CVD and controls.

These data extend the previous findings of 5L3DVCG-AI identifying CVD patients with cardiac ischaemia from those without to now differentiating healthy controls from CVD and those with higher risk for CVD. 5L3DVCG-AI may thus be a further scalable screening method to identify patients at risk for CVD in need for risk modification or further diagnostic procedures. The ongoing prospective large-scale performance clinical trials will have to confirm these preliminary data to verify the diagnostic accuracy.