Construction and Validation of a Nomogram Prediction Model for Catheter Related Thrombosis Risk of Central Venous Access Devices in Cancer Patients -- Prospective, Machine Learning Study (Preprint)

Abstract

Central venous access devices (CVADs) play a crucial role in providing treatment and supportive care for cancer patients. However, catheter-related thrombosis (CRT) poses a significant risk to patient safety., which will interrupt patient treatment, delay the patient's therapeutic period, prolong hospitalization, and increase the patient’s physical, mental, and economic burden. Identifying independent risk factors for CRT in cancer patients, proactively utilizing high-quality risk assessment tools in high-risk groups, and implementing precise prevention and treatment can effectively reduce the occurrence of CRT.

Construction and validation of a prediction model for predicting the risk of catheter-related thrombosis (CRT) in cancer patients’ central venous access devices (CVADs).

Using a prospective study design, cancer patients with CVADs in Xiangya Hospital of Central South University were followed up from January 2021 to December 2022 until catheter removal and 539 cases of CVADs-CRT occurred. Five hundred patients who met the inclusion and exclusion criteria were taken as the case group. Two cases of cancer patients without CRT were taken according to the number of CRTs/number of non-CRTs (1:2) in the same month in which a case of cancer patient with CRT was diagnosed by using the random number table method, for a total of 1,000 cases of cancer patients without CRT as the control group. Patient data were randomly divided into a training group (n=1050) and a testing group (n=450) according to the ratio of 7:3. Univariate and Least Absolute Shrinkage and Selection Operator (LASSO) algorithm was used to determine the risk factors for CRT formation. Risk prediction models were constructed based on Logistic Regression, Random Forest, and Support Vector Machine and evaluated by area under the curve (AUC). Data from patients with CVADs placement in Qinghai University Affiliated Hospital and Hainan Provincial People's Hospital (January 2023 to June 2023) were applied to externally validate the optimal model's differentiation, calibration, and clinical applicability.

The incidence of CVADs-CRT in cancer patients was 5.02%. There was a difference in the occurrence of CRT in patients with different cancers (P<.05), and the top three highest risks of occurrence were in patients with head and neck tumors (9.66%), haematological tumors (6.97%) and respiratory tumors (6.58%). There was a difference in the occurrence of CRT in patients with different catheters placed (P<.05), with the top three highest risks occurring in hemodialysis catheters (13.91%), CVC (8.39%), and PICC (4.68%). Eleven independent risk factors were screened for age, catheter method, catheter valve, catheter material, infection, catheter history, D-Dimer, operation history, anemia, diabetes, and targeted drugs. The Logistic prediction model had the best discriminative ability among the three machine learning-constructed models, with AUCs of 0.868 (0.846-0.890) for the training group. The externally validated AUC was 0.708 (0.618-0.797), the Nomogram model calibration curve was consistent with the ideal curve, and the Hosmer-Lemeshow test showed a good fit (P > .05) and a high net benefit value for the clinical decision curve.

The Nomogram constructed in this study can be personalised to predict the risk of developing CVADs-CRT in cancer patients, which can help in the early identification and screening of patients at high risk of cancer CVADs-CRT.