Prediction of post-insertion infections related to totally implantable subcutaneous venous access ports in tumor patients using a nomogram

Abstract

Totally implantable subcutaneous venous access ports (TISVAPs) are essential for long-term central venous chemotherapy, delivering medication directly into the central veins of patients. While they play a critical role in reducing patient discomfort, TISVAPs pose a notable risk of post-insertion infections-particularly concerning for oncology patients with compromised immune systems due to aggressive treatment regimens. Our research addresses this issue by developing a predictive nomogram to estimate the risk of TISVAP-associated infections. The model is based on independent risk factors identified in our study: a history of diabetes, the type of chemotherapy, peripheral blood leukocyte count (WBC), and serum albumin levels. Using retrospective clinical data from 309 oncology patients who underwent TISVAP implantation at a tertiary A-grade comprehensive hospital, we divided the dataset into training (n = 246) and validation (n = 63) subsets. Through logistic and Lasso regression analyses, we identified the independent risk factors associated with infections. The resulting interactive nomogram demonstrated strong accuracy and reliability, with C-indexes of 0.82 and 0.835 for the training and validation sets, respectively. This tool equips healthcare providers to proactively identify high-risk patients and tailor preventive strategies accordingly. Ultimately, our research aims to enhance patient outcomes and improve the quality of life for those undergoing long-term venous chemotherapy.

Figure 1.

Flowchart depicting the enrollment process for the training set and validation set in model construction. TISVAP: Totally implantable subcutaneous venous access port.

Figure 2.

Feature selection by LASSO regression model in training sets. (A) The coefficients change of different genes with different lambda; (B) By verifying the optimal parameter (lambda) in the LASSO model, the partial likelihood deviance (binomial deviance) curve was plotted vs log (lambda). Dotted vertical lines were drawn based on 1 SE of the minimum criteria (the 1-SE criteria). Four features with non-zero coefficients were selected by optimal lambda.

Figure 3.

The nomogram used to predict the probability of infection in patients with TISVAPs surgery. TISVAP: Totally implantable subcutaneous venous access port.

Figure 4.

Assessment and validation of the interactive nomogram model. The model demonstrates high discriminatory power as shown by the AUC for both the training group, with an AUC value of 0.820 (A), and the validation group, with an AUC value of 0.835 (B). Calibration curves indicate a strong fit between predicted and observed outcomes in both the training (D) and validation cohorts (E). Decision curves further reveal the model’s clinical relevance within both cohorts, as displayed in training (C) and validation sets (F). AUC: Area under the curve.