Predictive Model for the Detection of Subclinical Atherosclerosis in HIV Patients on Antiretroviral Treatment.
Authors
Affiliations (6)
Affiliations (6)
- Research Area, Consorci Sanitari Alt Penedés-Garraf, Barcelona, Spain.
- Department of Hospital Pharmacy, Hospital Universitari de Bellvitge, Barcelona, Spain.
- Department of Infectious Diseases, Hospital San Pedro, Logroño, Spain.
- Department of Infectious Diseases, Hospital Clínico Universitario Lozano Blesa, Zaragoza, Spain.
- Department of Genetics, Microbiology and Statistics, Faculty of Biology, University of Barcelona, Barcelona, Spain.
- Department of Internal Medicine, Consorci Sanitari Alt Penedés-Garraf, Barcelona, Spain.
Abstract
Patients living with HIV (PLHIV) have a higher cardiovascular risk than others, which is why the early detection of atherosclerosis in this population is important. The present study reports predictive models of subclinical atherosclerosis for this population of patients, made up of variables that are easily collected in the clinic. The study design is a cross-sectional observational study. PLHIV without established cardiovascular disease were recruited for this study. Predictive models of subclinical atherosclerosis (Doppler ultrasound) were developed by testing sociodemographic variables, pathological history, data related to HIV infection, laboratory parameters, and capillaroscopy as potential predictors. Logistic regression with internal validation (bootstrapping) and machine learning techniques were used to develop the models. Data from 96 HIV patients were analysed, 19 (19.8%) of whom had subclinical atherosclerosis. The predictors that went into both machine learning models and the regression model were hypertension, dyslipidaemia, protease inhibitors, triglycerides, fibrinogen, and alkaline phosphatase. Age and C-reactive protein were also part of the machine learning models. The logistic regression model had an area under the receiver operating characteristic curve (AUC) of 0.91 (95% CI: 0.84-0.99), which became 0.80 after internal validation by bootstrapping. The ma-chine learning techniques produced models with AUCs ranging from 0.73 to 0.86. We report predictive models for subclinical atherosclerosis in PLHIV, demonstrating relevant predictive performance based on easily accessible parameters, making them potentially useful as a screening tool. However, given the study's limitations-primarily the sample size-external validation in larger cohorts is warranted.