Machine learning risk prediction model for bloodstream infections related to totally implantable venous access ports in patients with cancer

Abstract

Objective: This study aimed to develop and validate a machine learning-based risk prediction model for catheter-related bloodstream infection (CRBSI) following implantation of totally implantable venous access ports (TIVAPs) in patients.

Methods: A retrospective cohort study design was employed, utilizing the R software package mlr3. Various algorithms including logistic regression, naive Bayes, K nearest neighbor, classification tree, and random forest were applied. Addressing class imbalance, benchmarks were used, and model performance was assessed using the area under the curve (AUC). The final model, chosen for its superior performance, was interpreted using variable importance scores. Additionally, a nomogram was developed to calculate individualized risk probabilities, enhancing clinical utility.

Results: The study involved 755 patients across both development and validation cohorts, with a TIVAP-CRBSI rate of 14.17%. The random forest model demonstrated the highest discrimination ability, achieving a validated AUC of 0.94, which was consistent in the validation cohort.

Conclusions: This study successfully developed a robust predictive model for TIVAP-CRBSI risk post-implantation. Implementation of this model may aid healthcare providers in making informed decisions, thereby potentially improving patient outcomes.

Keywords: Bloodstream infection; Cancer; Machine learning; Prediction model; Risk factor; Totally implantable venous access port.

Figure 1

Flow diagram of the model-building approach. AUC, area under the curve; FNR, false negative rate; FPR, false positive rate; ROC, receiver operating characteristic; DCA, curve of nomogram.

Figure 2

LASSO regression analysis result chart. A: LASSO regression cross validation plot; B: LASSO regression coefficient path diagram. LASSO, least absolute shrinkage and selection operator.

Figure 3

ROC curves and precision-recall (PRC) curves for the five different models. A: ROC curves of different machine learning models; B: PRC curves of different machine learning models. ROC, receiver operating characteristic.

Figure 4

Risk prediction nomogram for TIVAP-CRBSI. PCT: procalcitonin; catheter month: catheter usage time, in months; TPN: receiving parenteral nutrition; NRS-2002: nutritional risk screening assessment score; NEUT: neutrophil count, normal value is 1.8–6.3 × 10⁹/L. NEUT, neutrophil count levels; NRS, nutritional risk screening; PCT, procalcitonin level; TIVAP-CRBSI, totally implantable venous access port-catheter-related bloodstream infection; TPN, total parenteral nutrition.

Figure 5

Predictive performance of the nomogram. A: ROC curve of nomogram; B: calibration curve of nomogram; C: DCA curve of nomogram; D: nomogram clinical impact curve. ROC, receiver operating characteristic.