Guidelines recommend performing a non-invasive testing for ischemia to diagnose coronary artery disease (CAD). However, these tests are frequently inconclusive (20–30%) and can lead to uncertainty about the likelihood of CAD. This population has been poorly studied because it is a heterogeneous population difficult to characterise.

To characterise a population of patients with inconclusive test using an unsupervised classification algorithm, then to compare the prognosis of each clusters identified.

Between 2008 and 2018, consecutive patients with inconclusive stress test to detect CAD prospectively referred for a second stress test were followed for major adverse cardiovascular events (MACE) defined as cardiac death or non-fatal myocardial infarction (MI). Inconclusive stress test was defined by stress echocardiography or nuclear stress testing with uncertain conclusion. Data were analysed by a hierarchical clustering on principal components (HCPC) to perform this unsupervised classification algorithm. Among baseline, clinical and CMR characteristics, 18 variables were used for the clustering of patients.

Of 1502 patients with inconclusive stress test (61.9±11.8 years, 59% men), 1397 (93%) completed the follow-up (median 5.5±2.3 years). An unsupervised clustering analysis of those patients identified 3 clusters of phenotypes: cluster 1 (n=524, 35%) had the highest prevalence of previous PCI (59%), the highest presence of a myocardial scar defined by CMR (58%), the lowest LVEF (35±7%) and the highest degree of LV dilatation (LVEDVi=124±25mL/m²). Cluster 2 (n=406, 27%) had the highest prevalence of previous CABG (82%), preserved LVEF (54±10%), absence of LV dilatation (LVEDVi=82±20mL/m²), and presence of myocardial scar defined by CMR (52%). This cluster comprised predominantly male patients (89%), with the highest rate of dyslipidemia (81%) or hypertension (71%). Cluster 3 (n=572, 38%) had the lowest rate of previous CABG or PCI (9%) and the lowest rate of myocardial scar in CMR (6%). This cluster gathered the oldest patients (73±11 years) and was predominantly female (60%) with the highest rate of atrial fibrillation (51%) or body mass index (31±7.8kg/m²). Survival analysis found significant differences across clusters for the occurrence of MACE (P=0.02) et cardiovascular mortality (P<0.0001) (Fig. 1). Using Cox proportional hazards modelling, clusters 1 (hazard ratio: 0.72; 95% confidence interval: 0.50 to 1.04) and 3 (hazard ratio: 0.62; 95% confidence interval: 0.30 to 1.25) had a similar prognostic, which was better than cluster 2 (HR: 1.00 [reference]).

Cluster analysis of clinical and CMR variables identified 3 different phenotypes of patients with inconclusive stress test to detect CAD that were associated with distinct clinical and prognostic profiles.