Key result
A random forest model combining strain elastography and traditional risk factors identified carotid artery plaques with AUC-ROCs between 0.76 and 0.79, outperforming logistic regression (P=0.0001).
Why the study?
There is a need for a specific atherosclerotic risk assessment for people living with HIV.
Does a machine learning model combining ultrasound strain elastography and traditional cardiovascular risk factors improve the identification of carotid artery plaques in people living with HIV?
Cohort (n=186)
Does a machine learning model combining ultrasound strain elastography and traditional cardiovascular risk factors improve the identification of carotid artery plaques in people living with HIV?
Effect estimate: AUC-ROC 0.76-0.79
p-value: p=0.0001
Ultrasound strain elastography combined with traditional cardiovascular risk factors in a machine learning model can effectively identify carotid artery plaques in people living with HIV.
May support ML-ultrasound integration for plaque detection in HIV; leaves open prospective validation before clinical adoption.
BACKGROUND: There is a need for a specific atherosclerotic risk assessment for people living with HIV (PLWH). SETTING: A machine learning classification model was applied to PLWH and control subjects with low-to-intermediate cardiovascular risks to identify associative predictors of diagnosed carotid artery plaques. Associations with plaques were made using strain elastography in normal sections of the common carotid artery and traditional cardiovascular risk factors. METHODS: One hundred two PLWH and 84 control subjects were recruited from the prospective Canadian HIV and Aging Cohort Study (57 ± 8 years; 159 men). Plaque presence was based on clinical ultrasound scans of left and right common carotid arteries and internal carotid arteries. A classification task for identifying subjects with plaque was defined using random forest (RF) and logistic regression models. Areas under the receiver operating characteristic curves (AUC-ROCs) were applied to select 5 among 50 combinations of 4 or less features yielding the highest AUC-ROCs. RESULTS: To retrospectively classify individuals with and without plaques, the 5 most discriminant combinations of features had AUC-ROCs between 0.76 and 0.79. AUC-ROCs from RF were statistically significantly higher than those obtained with logistic regressions ( P = 0.0001). The most discriminant features of RF classifications in PLWH were age, smoking status, maximum axial strain and pulse pressure (equal weights), and sex and antiretroviral therapy exposure (equal weights). When considering the whole population, the HIV status was identified as a cofactor associated with carotid artery plaques. CONCLUSIONS: Strain elastography adds to traditional cardiovascular risk factors for identifying individuals with carotid artery plaques.
No takes yet. Share an insight, caveat, or question.
Cardinal et al. (2022) conducted a cohort in Carotid artery plaques in people living with HIV (n=186). Strain elastography and traditional cardiovascular risk factors vs. Logistic regression models / traditional risk factors was evaluated on Classification of individuals with and without carotid artery plaques (AUC-ROC 0.76-0.79, p=0.0001). A random forest model combining strain elastography and traditional risk factors identified carotid artery plaques with AUC-ROCs between 0.76 and 0.79, outperforming logistic regression (P=0.0001).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: