A novel Artificial Intelligence Model for Predicting Totally Implantable Venous Access Device (TIVAD)-Related Infections.

PURPOSE

To develop and validate a model for predicting totally implantable venous access device (TIVAD)-related infections.

MATERIALS & METHODS: Patients who underwent placement of TIVADs between 2014 and 2023 were included and divided into a training set (n = 5465) and a validation set (n = 2341). The clinical characteristics of 7806 patients with TIVADs were analysed. Model 1 was developed using both least absolute shrinkage and selection operator (LASSO) regression and bidirectional stepwise logistic regression, whereas Model 2 was developed using only bidirectional stepwise logistic regression. Two models were compared using the Akaike information criterion (AIC), decision curve analysis (DCA), clinical impact curves (CICs), area under the curve (AUC) of receiver operating characteristic curves (ROCs) and calibration curves, and the better model was chosen as the final model.

RESULTS

Model 1 included body mass index (BMI), sex, outpatient or inpatient status, tunnel length, catheter-related thrombosis (CRT), primary catheter dislodgement and age. Model 2 included sex, catheter diameter, CRT, age, outpatient or inpatient status, catheter length, primary catheter dislodgement, BMI and tumour type. Both models demonstrated clinical usefulness according to DCA and CICs. Model 2 was chosen as the final model because of its superior AIC value and calibration performance.

CONCLUSIONS

We developed and validated a model for predicting TIVAD-related infections. We recommend a 5F catheter and prophylactic antibiotics for patients with a Model 2 infection risk score above 0.026. We also suggest appropriately increasing the tunnel length, using a smaller catheter, and strictly adhering to heparin flush protocols to prevent thrombosis.