Development and validation of a predictive model for peripherally inserted central catheter-related thrombosis in...

Peripherally inserted central catheters have been extensively applied in clinical practices. However, they are associated with an increased risk of thrombosis. To improve patient care, it is critical to timely identify patients at risk of developing peripherally inserted central catheter-related thrombosis. Artificial neural networks have been successfully used in many areas of clinical events prediction and affected clinical decisions and practice.

To develop and validate a novel clinical model based on artificial neural network for predicting peripherally inserted central catheter-related thrombosis in breast cancer patients who underwent chemotherapy and determine whether it may improve the prediction performance compared with the logistic regression model.

A prospective cohort study.

A large general hospital in Fujian Province, China.

One thousand eight hundred and forty-four breast cancer patients with peripherally inserted central catheters placement for chemotherapy were eligible for the study.

The dataset was divided into a training set (N = 1497) and an independent validation set (N = 347). The synthetic minority oversampling technique (SMOTE) was used to handle the effect of imbalance class. Both the artificial neural network and logistic regression models were then developed on the training set with and without SMOTE, respectively. The performance of each model was evaluated on the validation set using accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUC).

Of the 1844 enrolled patients, 256 (13.9%) were diagnosed with peripherally inserted central catheter-related thrombosis. Predictive models were constructed in the training set and assessed in the validation set. Eight factors were selected as input variables to develop the artificial neural network model. Without SMOTE, the artificial neural network model (AUC = 0.725) outperformed the logistic regression model (AUC = 0.670, p = 0.039). SMOTE improved the performance of both two models based on AUC. With the SMOTE sampling, the artificial neural network model performed the best across all evaluated models, the AUC value remained statistically better than that of the logistic regression model (0.742 vs. 0.675, p = 0.004).

Artificial neural network model can effectively predict peripherally inserted central catheter-related thrombosis in breast cancer patients receiving chemotherapy. Identifying high-risk groups with peripherally inserted central catheter-related thrombosis can provide close monitoring and an opportune time for intervention.