Impact of machine learning-based coronary computed tomography angiography fractional flow reserve on treatment decisions and clinical outcomes in patients with suspected coronary artery disease

Objectives This study investigated the impact of machine learning (ML)-based fractional flow reserve derived from computed tomography (FFRCT) compared to invasive coronary angiography (ICA) for therapeutic decision-making and patient outcome in patients with suspected coronary artery disease (CAD). Methods One thousand one hundred twenty-one consecutive patients with stable chest pain who underwent coronary computed tomography angiography (CCTA) followed ICA within 90 days between January 2007 and December 2016 were included in this retrospective study. Medical records were reviewed for the endpoint of major adverse cardiac events (MACEs). FFR(CT)values were calculated using an artificial intelligence (AI) ML platform. Disagreements between hemodynamic significant stenosis via FFR(CT)and severe stenosis on qualitative CCTA and ICA were also evaluated. Results After FFR(CT)results were revealed, a change in the proposed treatment regimen chosen based on ICA results was seen in 167 patients (14.9%). Over a median follow-up time of 26 months (4-48 months), FFRCT <= 0.80 was associated with MACE (HR, 6.84 (95% CI, 3.57 to 13.11);p <0.001), with superior prognostic value compared to severe stenosis on ICA (HR, 1.84 (95% CI, 1.24 to 2.73),p= 0.002) and CCTA (HR, 1.47 (95% CI, 1.01 to 2.14,p= 0.045). Reserving ICA and revascularization for vessels with positive FFR(CT)could have reduced the rate of ICA by 54.5% and lead to 4.4% fewer percutaneous interventions. Conclusions This study indicated ML-based FFR(CT)had superior prognostic value when compared to severe anatomic stenosis on CCTA and adding FFR(CT)may direct therapeutic decision-making with the potential to improve efficiency of ICA.