Objective To conduct a multi-center external validation of a visual assessment model for thrombosis risk related to peripherally inserted central catheter (PICC) based on Caprini score and coagulation indicators, with a focus on evaluating the model's generalization and clinical applicability, in order to deeply explore its practical application value in clinical practice.

Methods A total of 671 inpatients with PICC from three tertiary hospitals (Hospital 1, Hospital 2 and Hospital 3) in Hefei City, Anhui Province were retrospectively analyzed, and consecutively included as the research subjects. The patient inclusion periods for each hospital were as follows: for Hospital 1 and Hospital 2, it was from January 1, 2024 to October 31, 2024; for Hospital 3, it was from October 1, 2023 to October 31, 2024. Electronic medical records were used to collect information such as the age of the research subjects, previous medical history, chemotherapy, Caprini score, D-dimer, BMI, limb selection and catheter specifications. To evaluate the discriminative ability of model, the area (AUC) under the receiver operating characteristic curve (ROC) was quantitatively analyzed. Meanwhile, the calibration of model was comprehensively evaluated through the Hosmer-Lemeshow test to systematically examine the consistency between the actual prediction probability of prediction model and observed results, ensuring the reliability of the model in terms of calibration consistency.

Results A total of 671 PICC catheterization cases were ultimately selected to form the external validation dataset, among which 62 cases (9.2%) were in the thrombosis group and 609 cases (90.8%) were in the non-thrombosis group. Comparison of the area under the ROC curve (AUC): the Data showed that the Caprini thrombosis Risk Assessment Scale (AUC = 0.636, 95%CI: 0.598–0.672) and the D-dimer index (AUC = 0.673, 95%CI: 0.636–0.708) had limited predictive efficacy when used alone, while the prediction model jointly constructed by the two had better predictive efficacy (AUC = 0.824, 95%CI: 0.784–0.864). The results of H-L test (χ² = 0.4029, P ＞ 0.05) showed that the fit degree of this model was good.

Conclusions The visual prediction model based on the Caprini risk assessment model combined with D-dimer coagulation index shows good predictive efficacy. It can provide an effective assessment tool for the clinical identification of patients at high risk of PICC-related thrombosis, and a reference basis for medical staff to formulate individualized intervention strategies for the prevention of PICC-related thrombosis.