Association between early carotid artery plaque presence, vascular strain imaging features and traditional cardiovascular risk factors in HIV infected individuals

A machine learning classification model was used to discriminate subjects with and without carotid artery plaques in a population of HIV -positive and -negative individuals with low-to-intermediate cardiovascular risks. The aim was to identify associations between the presence of plaques and features consisting of traditional cardiovascular risk factors and strain elastography in normal sections of the common carotid artery. One hundred and two HIV-infected and 84 age-matched control subjects were recruited from a prospective, controlled cohort study (mean age 57 years ± 8 years; 159 men). Plaques, defined as an intima-media thickness (IMT) greater than 1 mm, were identified on ultrasound images in longitudinal views of the common and internal carotid arteries. A classification task of identifying subjects with presence of plaque was defined using a random forests model. Six strain features on normal CCA in addition to 21 clinical biomarkers and clinical ultrasound imaging features* were selected as inputs in the classification model. The 5 most discriminant combinations of features had ROC-AUCs between 0.76 and 0.80.