In order to bring you the best possible user experience, this site uses Javascript. If you are seeing this message, it is likely that the Javascript option in your browser is disabled. For optimal viewing of this site, please ensure that Javascript is enabled for your browser.

Abstract of the day - Can AI improve risk stratification in symptomatic patients with suspected CAD?

26 Aug 2023
Abstract of the Day

Although coronary computed tomography angiography (CCTA) is highly useful in non-invasive coronary artery disease (CAD) detection in patients with low-to-intermediate pre-test probability of CAD, it cannot directly assess the presence of ischaemia, which can lead to further downstream testing.

Today, Doctor Sarah Bär (Turku University Hospital - Turku, Finland) presents results from a novel AI-based algorithm that leverages atherosclerosis and vascular morphological features to determine coronary ischaemia from CCTA images.

The AI-based ischaemia algorithm was calculated by blinded analysts using data from 2,214 patients with suspected CAD who underwent clinically indicated CCTA at Turku University Hospital from 2007 to 2016. The median follow-up duration was 6.9 years.

Around one-fifth (22.2%) of patients had positive AI-based ischaemia findings and these patients had a significantly higher crude rate of the primary endpoint (death, myocardial infarction [MI] and unstable angina pectoris [uAP]) compared with patients who had negative AI-based ischaemia findings (hazard ratio [HR] 2.80; 95% CI 2.18–3.61; p<0.001). Indeed, higher rates of all the primary endpoint components – death (HR 1.84; 95% CI 1.34–2.52; p<0.001), MI (HR 5.95; 95% CI 3.60–9.83; p<0.001) and uAP (HR 9.10; 95% CI 4.23–19.59; p<0.001) – were observed in patients with positive AI-based ischaemia findings. Furthermore, results for the primary outcome were consistent and significant after adjusting for confounders, including age, sex, diabetes, smoking, hypertension, dyslipidaemia, body mass index >25 kg/m2 and family history of CAD (adjusted HR 1.96; 95% CI 1.50–2.57; p<0.001). Of interest, positive AI-based ischaemia findings were associated with a significantly higher rate of the primary endpoint among 1,641 patients with no/non-obstructive CAD in CCTA (50% stenosis; HR 2.78; 95% CI 1.72–4.48; p<0.001), but not among 517 patients with anatomically obstructive CAD in CCTA (>50% stenosis; HR 1.31; 95% CI 0.85–2.00; p=0.218).

Taken together, these results suggest that the AI-based ischaemia algorithm may be helpful to improve risk stratification, especially among patients with normal/non-­obstructive CCTA, and further studies on its agreement with other methods to determine myocardial ischaemia and its prognostic power in different patient populations are warranted.