214 participants with ESRD who underwent PD catheterization and received regular PD treatment for over 3 months at Wuxi No. 2 People’s Hospital from June 2017 to June 2024 were parts of this study. The primary endpoint of our study is the occurrence of infections in PD patients during the observation period from June 2017 to June 2024. Based on patients’ infection status after 3 months of PD, they were categorized into an infection group and a non-infection one. Infections primarily included intra-abdominal, skin, and catheter-related infections (tunnel and exit-site infections). The Ethics Committee of Wuxi No. 2 People’s Hospital (Ethics Approval No.2023Y-3) issued strong approval for the ethical conduct of this study. Inclusion criteria: (1) diagnosed with ESRD based on a glomerular filtration rateless than 15 ml/(min.1.73 m²), (2) treated with PD following the “Chinese Guidelines for Peritoneal Dialysis Catheterization” with a treatment duration of ≥ 3 months (9), (3) aged 18–85 years, (4) complete clinical laboratory and imaging examination data available, (5) signed informed consent. Exclusion criteria: (1) lack of regular follow-up at the hospital, (2) severe cardiopulmonary dysfunction, (3) severe intra-abdominal infections (occurring after PD catheter placement surgery but before the initiation of the first PD treatment) or adhesions, (4) pregnancy, and (5) failure to sign informed consent.

At the time of admission, relevant examinations were conducted, and general data were collected. These comprised essential information (age, gender, weight, height, BMI), laboratory results (serum creatinine, blood urea nitrogen, serum sodium, serum potassium, albumin and other blood count results), surgical-related indicators (operative time, intraoperative blood loss, surgical incision length, postoperative hospitalization duration, surgery costs, surgery costs/total costs), and peritoneal dialysis-related complications (catheter flow dysfunction, dialysate leakage, catheter migration, exit-site infections, and abdominal catheter tunnel infection).

We carried out multivariable logistic regression analysis with stepwise selection guided by the Akaike Information Criterion (AIC) to identify independent risk factors for infections in PD patients. Variables demonstrating statistical significance (p < 0.05) were retained as independent predictors. Random forest plotting was subsequently implemented to visualize predictor accuracy and variable importance metrics, which were ultimately integrated into a nomogram for predicting infection risk in PD patients. We plotted the ROC curve and evaluated the model’s discriminative ability by calculating the area under the curve (AUC). Additionally, calibration performance was quantified via 1,000-round bootstrap validation of calibration curves, while decision curve analysis (DCA) was carried out by calculating standardized net benefits across varying threshold probabilities to evaluate the clinical utility of the nomogram (10), which compensates for the limitations of ROC curves (11).

We used SPSS 20.0 statistical to make an analysis of the classified and structured data. Continuous variables were expressed as mean ± standard deviation or median (interquartile range, IQR), depending on normality. We made use of t-test or Mann-Whitney U test to compare the means or medians across different sets. With respect to categorical variables, the data were presented in terms of frequency counts and percentage, at the same time, the implementation of Pearson’s chi-square test facilitated a comprehensive evaluation of the variations identified between the two groups. Odds ratios (ORs), often with 95% confidence intervals (CIs), were used to present the entire set of findings. Univariate analysis was performed using Pearson’s chi-square test or Fisher’s exact test, as appropriate. Variables with statistically significant differences (p < 0.05) were included in a multivariate logistic regression model (backward direction) to identify independent risk factors for long-term infection in PD patients. And validation of the nomogram were executed using R software (x64 for Windows, version 3.6.1). In all analyses, a two-sided p-value < 0.05 was considered statistically significant.

235 individuals with ESRD who underwent PD were initially screened. Following the screening process, 214 participants satisfying the inclusion criteria were finally included. The procedural steps involved in the data selection process are depicted in the flowchart feature in Figure 1.

We collected clinical data from 214 patients with ESRD undergoing regular peritoneal dialysis. Based on the follow-up results after 3 months of PD treatment, all participants were stratified into infection and non-infection groups. Significant intergroup differences were observed in the following aspects: age, serum albumin, PNI, GNRI, surgery-related indicators (surgical duration, intraoperative bleeding volume, surgical incision length), primary disease and peritoneal dialysis-associated complications [exit-site infection (ESI) and abdominal catheter tunnel infection] (all with p < 0.05). Individuals in the infection group were older than those in the non-infection one (57.81 ± 11.95 years vs. 52.25 ± 13.29 years). Moreover, the infection group exhibited significantly longer surgical incision lengths (3.28 ± 0.93 cm vs. 2.83 ± 0.82 cm) and lower PNI values (37.04 ± 6.00 vs. 41.99 ± 4.97) compared to the non-infection group. In both groups, diabetic nephropathy was documented as the second most frequent primary disease, highlighting its substantial impact (12). There were no statistically significant intergroup differences in: hematologic indicators (serum creatinine, blood urea nitrogen, serum sodium, serum calcium, etc.), peritoneal dialysis-related complications (catheter malfunction, dialysate leakage, catheter migration). The clinical features of both groups are shown in Table 1.

Univariate and multivariate logistic regression analyses were executed in both groups (Table 2). Univariate logistic regression analysis indicated that patient age, operative time for PD catheterization, surgical incision length, intraoperative blood loss, PNI, and GPNI emerged as independent potential risk indicators for infection in patients undergoing PD (all p < 0.05). We threw these variables into a multivariate logistic regression analysis to pick out independent factors related to long-term infections in these patients, including age, surgical incision length and PNI. A random forest plot was developed to further assess the precision and significance of these predictive variables. As is shown in Figure 2, increasing patient age (OR = 1.042, 95% CI: 1.013–1.072, p = 0.004), longer surgical incision length (OR = 2.366, 95% CI: 1.161–4.824, p = 0.018), in addition, lower PNI (OR = 0.785, 95% CI: 0.702–0.878, p < 0.001) were elements that increase susceptibility for long-term infection in patients undergoing PD.

As illustrated in Figure 3, based on the three independent predictive factors obtained from multivariable logistic regression, the nomogram model was rigorously formulated to predict infection risk in PD individuals. The results indicated that a PNI score of 20 corresponded to the highest risk score (100), followed by an incision length of 5 cm (43 points). When the risk factors for infections in patients are visualized, individual risks for developing long-term infections can be predicted. First, each independent risk factor contributing to infections in patients undergoing PD was projected onto the first row of the scale to derive scores for each factor, and by adding up the values of all 3 risk factors, total score was calculated. A higher total score indicates a higher risk of infection for these patients. The research sample was used as the training set, with 1,000 repeated samplings performed, and internal validation was conducted using five-fold cross-validation. As shown in Figure 4A, the AUC value was 0.807 (95% CI: 0.746–0.869), indicating that the model demonstrated good discriminative accuracy in predicting the risk of long-term infection in PD patients. The corresponding AUC values for PNI, surgical incision length, and age were 0.743, 0.636, and 0.619. Specifically, PNI exhibited higher predictive accuracy compared to surgical incision length and age (Figure 4B). To evaluate the validity of the modeling results, a calibration plot for the infection risk prediction model in PD individuals was generated. Figure 4C showed that the observed outcomes were in high agreement with the predicted outcomes. DCA was performed to assess its clinical utility. As illustrated in Figure 4D, according to the decision curve, when the threshold probability for a specific patient ranged between 14 and 92%, the net benefit of this predictive model (red curve) was consistently higher than that of the “All” and “None” lines, demonstrating its practical value.