Construction of a nomogram-based risk prediction model for midline catheter-related thrombosis in people living with HIV

Midline catheters (MCs) are increasingly used for venous access in people living with HIV (PLHIV) who require prolonged intravenous therapy for management of opportunistic infections and other HIV-associated complications. Catheter-related thrombosis (CRT) represents a clinically significant complication in this population, driven by HIV-associated coagulopathy, chronic immune activation, and endothelial dysfunction. This study aimed to identify independent risk factors for CRT in PLHIV undergoing MC placement and to develop a validated nomogram-based risk prediction model. This retrospective cohort study enrolled 1216 HIV-infected patients who underwent midline catheter placement at Changzhou Third People's Hospital between January 2018 and December 2024. Patients were divided chronologically into a discovery cohort (2018–2022) and a temporal validation cohort (2023–2024); the discovery cohort was randomly split 7:3 into training and internal test sets, while the validation cohort was withheld from all model development. Candidate predictors were screened by univariate logistic regression in the training set and entered multivariable backward selection. A nomogram was constructed, compared with random forest, XGBoost, and support-vector-machine models, and evaluated by receiver operating characteristic (ROC) analysis, calibration, decision curve analysis (DCA), and SHapley Additive exPlanations (SHAP). CRT occurred in 216 of 1,216 patients (17.8%). Multivariable analysis identified 11 independent risk factors: catheter-to-vein ratio (CVR; OR=2.79 per 0.1-unit), previous thrombosis history (OR=2.48), fibrinogen (OR=2.17), LDL-C (OR=1.69), insertion attempts (OR=1.51), BMI (OR=1.30), APTT (OR=1.14), platelet count (OR=1.11 per 10×10⁹/L), age (OR=1.07 per year), D-dimer (OR=1.07), and hemoglobin (OR=0.98). The nomogram showed strong discrimination (training AUC=0.922, 95% CI 0.892–0.953; internal test AUC=0.917; temporal validation AUC=0.910, 95% CI 0.876–0.945) with good calibration, and its discrimination did not differ significantly from the machine-learning models (DeLong all P>0.4). At the Youden cut-off of 0.23, validation accuracy was 84.0% and negative predictive value (NPV) 94.6%. Performance was robust across age and sex strata. This study presents the first temporally validated nomogram for predicting MC-related CRT specifically in PLHIV. The model demonstrates good discrimination and clinical utility, with a high negative predictive value particularly suited to safely identifying low-risk patients and supporting decisions to avoid unnecessary prophylactic anticoagulation in this high-risk population.