JMIR Publications

ABSTRACT

Background:

Cancer is a major public health problem and poses a health threat to the population, and Peripherally Inserted Central Catheters-Unplanned Extubation (PICC-UE) is considered the most adverse event for patient safety. Identifying independent risk factors for PICC-UE, applying high-quality assessment tool early in high-risk population, and adopting precise prevention and treatment can effectively reduce the occurrence of PICC-UE.

Objective:

The objective is to identify the independent risk factors for PICC-UE in cancer patients, and to develop a predictive model for PICC-UE for cancer patients, providing a theoretical basis for the prevention and prediction of PICC-UE in cancer patients.

Methods:

Prospective data were collected from January to December 2022 from cancer patients with PICC at Xiangya Hospital, Central South University, and each patient was followed up until the catheter removal. The patients were divided into UE group (n=3107) and non-UE group (n=284), and independent risk factors were determined by univariate, LASSO algorithm and multivariate analysis. The 3391 patients were then divided into a trainset and a testset according to the ratio of 7:3. The screened predictors were used to build three predictive models using Logistic Regression, Support Vector Machine and Random Forest algorithms, and the optimal model was screened by ROC curve and TOPSIS synthesis analysis. We collected prospective data of 600 cancer patients with PICC from June to December 2022 at the Affiliated Hospital of Qinghai University and Hainan Provincial People's Hospital for external validation. The Area Under the Curve (AUC) of the ROC was used to test the differentiation of the model, and the Calibration Curve to assess the calibration capability and Decision Curve Analysis (DCA) to evaluate the clinical applicability of the model.

Results:

Independent risk factors for PICC-UE in cancer patients included impaired physical mobility (OR=2.775), diabetes (OR=1.754), surgical history (OR=1.734), elevated D-dimer concentration (OR=2.376), targeted therapy (OR=1.441), surgical protocol (OR=1.543), and more than one catheter puncture (OR=1.715); protective factors included normal BMI (OR=0.449), polyurethane catheter material (OR=0.305), and valved catheter (OR=0.639). The results of the TOPSIS synthesis analysis: the Ci values were 0.00, 0.82 and 0.85 in the trainset, and the Ci values were 0.00, 1.00 and 0.81 in the testset for the Logistic, Support Vector Machine and Random Forest models, respectively. The optimal model constructed based on Support Vector Machine was obtained and validated externally, the ROC curve, Calibration curve and DCA curve showed that the model had excellent accuracy, stability, generalizability and clinical applicability.

Conclusions:

Ten independent predictors of PICC-UE in cancer patients were obtained in this study. The predictive model was constructed based on Support Vector Machine, which has clinical application value through external validation, and provides significant support for the early prediction of PICC-UE in cancer patients.