A nomogram for predicting early bacterial infection after liver transplantation: a retrospective study

Bacterial infection is a common complication of liver transplantation and is associated with high mortality rates. However, multifactor-based early-prediction tools are currently lacking. Therefore, this study investigated the risk factors of early bacterial infections after liver transplantation and used them to establish a nomogram.

We retrospectively collected the clinical data of 232 patients who underwent liver transplantation. We excluded 15 patients aged less than 18 years, 7 patients with infection before transplantation, and 3 patients with incomplete laboratory test results based on the sample exclusion criteria, and finally included 207 liver transplant patients. The patients were divided into the bacterial infection group (75 cases) and non-infected group (132 cases) according to whether bacterial infection had occurred within 30 days after surgery. The associated risk factors were determined using stepwise regression, and a nomogram was established based on the results of the multifactorial analysis. The predictive performance of the model was compared by assessing the area under the receiver operating characteristic curve (AUC-ROC), decision curve analysis (DCA), and the calibration curve, which was validated using cross-validation and repeated sampling.

Preoperative systemic immune inflammation index (SII) (OR = 1.003, p = 0.001), duration of surgery (OR = 1.008, p = 0.005), duration of postoperative ventilator use (OR = 1.013, p = 0.025), neutrophil to lymphocyte ratio (NLR) (OR = 1.017, p = 0.024), ICU stay time (OR = 1.125, p = 0.015) were independent risk factors for early bacterial infection after liver transplantation. The nomogram was constructed based on the above factors, achieving an AUC of 0.863 (95%CI: 0.808, 0.918), which showed that the mean absolute error between the predicted risk and the actual risk of the model was 0.044. The decision curve analysis showed that it was located above both extreme curves in a range of more than the 14% threshold, which indicated that there was a good clinical benefit in this range. Internal validation using 10-fold cross validation and bootstrap replicate sampling yielded areas under the corrected ROC curves of 0.842 and 0.854, respectively. These results indicate that the developed model exhibits good predictive performance and a moderate error in training and validation.

The nomogram constructed in this study showed good differentiation, calibration, and clinical applicability. It can effectively identify the high-risk group for bacterial infection in the early postoperative period after liver transplantation, while simultaneously helping the transplant team dynamically monitor the key indicators and optimize perioperative management.