Nomogram predicts catheter-related thrombosis risk in people living with HIV

People living with HIV who need weeks of intravenous therapy increasingly receive midline catheters, thin tubes threaded into a peripheral vein that can stay in place for prolonged treatment of opportunistic infections and other HIV-associated complications. A clinically significant complication of such catheters is catheter-related thrombosis, the formation of blood clots around or downstream of the device. In this population, the risk is shaped by HIV-associated coagulopathy, chronic immune activation, and endothelial dysfunction, which together make clotting harder to predict from general ward experience alone.

A new retrospective cohort study set out to identify which factors independently drive this risk and to convert them into a bedside scoring tool. The researchers enrolled 1,216 HIV-infected patients who underwent midline catheter placement at Changzhou Third People’s Hospital between January 2018 and December 2024. Rather than testing their model on the same patients used to build it, they divided the records chronologically: data from 2018 to 2022 formed a discovery cohort, randomly split seven-to-three into a training set and an internal test set, while records from 2023 and 2024 were held back entirely as a temporal validation cohort, untouched during all model development.

Catheter-related thrombosis occurred in 216 of the 1,216 patients, an overall rate of 17.8 percent. Candidate predictors were first screened with univariate logistic regression in the training set and then entered into multivariable backward selection. Eleven independent risk factors survived the analysis: catheter-to-vein ratio, with an odds ratio of 2.79 per 0.1-unit increase; previous thrombosis history, odds ratio 2.48; fibrinogen, 2.17; LDL cholesterol, 1.69; number of insertion attempts, 1.51; body mass index, 1.30; activated partial thromboplastin time, 1.14; platelet count, 1.11 per 10×10⁹/L; age, 1.07 per year; D-dimer, 1.07; and hemoglobin, which showed a small inverse association at 0.98.

The selected factors were assembled into a nomogram, a point-based chart that lets clinicians total a patient’s individual values into an estimated probability of thrombosis. The authors also built three machine-learning comparators using random forest, XGBoost, and support-vector-machine algorithms, and evaluated everything with receiver operating characteristic analysis, calibration plots, decision curve analysis, and SHapley Additive exPlanations, a method for quantifying each variable’s contribution to individual predictions.

Discrimination, the model’s ability to separate patients who develop clots from those who do not, was strong and, crucially, held up over time. The area under the ROC curve reached 0.922 in the training set (95 percent CI 0.892–0.953), 0.917 in the internal test set, and 0.910 in the temporal validation cohort (95 percent CI 0.876–0.945). Calibration was good, and performance remained robust across age and sex strata. Notably, the nomogram’s discrimination did not differ significantly from the machine-learning models, with DeLong tests all yielding P values above 0.4, meaning the transparent point-based chart matched the more opaque algorithms on this dataset.

At the Youden-index cut-off of 0.23, the threshold balancing sensitivity and specificity, the model achieved 84.0 percent accuracy in the validation cohort and a negative predictive value of 94.6 percent. That last figure is the study’s most clinically pointed result: a low-risk score reliably identified patients who did not go on to develop catheter-related thrombosis. The authors argue this makes the tool particularly suited to safely identifying low-risk patients and supporting decisions to avoid unnecessary prophylactic anticoagulation in this population, where anticoagulant use carries its own bleeding considerations.

The study’s conclusion positions the work as the first temporally validated nomogram for predicting midline catheter-related thrombosis specifically in people living with HIV. Because the validation cohort came from later calendar years than the development data, the reported performance reflects how the model behaves on patients treated after the predictors were fixed, a stricter test than a random internal split. The reported utility, however, is a model property measured in this cohort, not a demonstrated outcome of deploying the tool in clinical practice; whether scoring patients actually reduces thrombosis or changes anticoagulation decisions would require prospective evaluation that this retrospective design cannot provide.

Several scope limits follow directly from the design. All patients came from a single hospital in China, so the predictor list and the 17.8 percent event rate may not transfer to settings with different catheter practices, patient mixes, or HIV care pathways. As a retrospective cohort, the study depends on variables already recorded at insertion, and the odds ratios describe associations within this population rather than experimentally established causal effects. The article is being shared early as peer-reviewed, accepted research carrying a permanent DOI, subject to further edits before the final version of record replaces it. Even so, the combination of a large single-center cohort, temporal validation, and a simple chart that performs on par with machine-learning models offers a concrete starting point for clinicians weighing who among their HIV-positive patients with midline catheters genuinely needs closer clot surveillance.

Subject of Research: A risk prediction model for midline catheter-related thrombosis in people living with HIV

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

Article References: Yue, Y., Qi, S., Dong, Z., Wang, K., & Luo, Y. (2026). Construction of a nomogram-based risk prediction model for midline catheter-related thrombosis in people living with HIV. AIDS Research and Therapy. https://doi.org/10.1186/s12981-026-00957-3

Image Credits: AI Generated

DOI: 10.1186/s12981-026-00957-3

Keywords: HIV, midline catheter, catheter-related thrombosis, nomogram, risk prediction model, retrospective cohort study, temporal validation, machine learning comparison, fibrinogen, D-dimer, anticoagulation, predictive modeling