Back

Prediction and Screening for Asymptomatic Carotid Artery Stenosis in Post-Radiotherapy Nasopharyngeal Carcinoma Patients

Lin, C.-Y.; Kuo, P.-H.; Wu, S.-Y.; Chen, C.-N.; Ko, J.-Y.

2025-08-24 otolaryngology
10.1101/2025.08.19.25334024 medRxiv
Show abstract

ObjectiveHead and neck radiotherapy (RT) is associated with an increased risk of carotid artery stenosis; yet standardized surveillance guidelines for nasopharyngeal carcinoma (NPC) survivors remain lacking. This study aimed to develop a risk prediction model to identify individuals at heightened risk. MethodsWe conducted this retrospective cohort study using claims data from the National Taiwan University Hospital-integrative Medical Database (NTUH-iMD). A Cox-based prediction model using stepwise variable selection was developed and model performance was evaluated using the area under receiver operating characteristic (ROC) curre (AUC) and integrated Brier score (IBS). To identify the optimal timing for initiating carotid ultrasound screening in post-RT NPC patients, a multivariable Generalized Linear Mixed Model (GLMM) was used to determine the post-treatment year most strongly associated with the development of moderate or greater internal carotid artery (ICA) stenosis. ResultsPatients in the high-risk group (risk score[≥]5) have significantly increased risk of moderate or greater ICA stenosis, which may begin to manifest as early as the fourth year after receiving radiation therapy. Additionally, we found that, compared to the average risk of the entire cohort, the risk of developing moderate-to-severe ICA stenosis began to increase significantly and persistently around the seventh year after radiotherapy. Moreover, the model achieved an AUC of 0.623 and an IBS of 0.084, indicating moderate discriminatory ability and good overall predictive accuracy. ConclusionAccording to our proposed risk prediction model, we recommend that carotid ultrasound screening begin in the seventh year following radiation therapy, while high-risk patients should start screening earlier, beginning in the fourth year.

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.