This study received approval from the hospital ethics committee (PJ202139) and encompassed a total of 831 patients who underwent craniotomy at the hospital between January 1, 2022, and June 31, 2022. The inclusion criteria consisted of adult patients who developed new-onset pneumonia within 30 days after craniotomy. Exclusion criteria encompassed patients with pre-existing pneumonia or those diagnosed with pneumonia during surgery, patients who abandoned the surgical procedure, patients who experienced intraoperative cardiac arrest or required rescue measures, and patients with severely incomplete medical records. The diagnostic criteria for pneumonia established by the US Centers for Disease Control were employed in this study (Abbott et al., 2018). These criteria, which rely on chest X-ray findings and clinical symptoms, are primarily designed for adult patients and offer comprehensive and objective assessments. Adhering strictly to this definition, we identified patients with postoperative pneumonia and assigned them to the POP group. Subsequently, controls were matched by surgical specialty and randomly selected at 1:1 from the remaining surgical patients without pneumonia and assigned to the non-POP group.

Patients’ medical records were meticulously collected and documented through the hospital’s digital system. The collected data encompassed demographic factors (such as age, gender, height, weight, alcohol consumption history, and smoking history), general condition assessments (such as the Glasgow Coma Scale [GCS] and American Society of Anesthesiologists [ASA] classification), pre-existing medical conditions (such as hypertension, diabetes, chronic obstructive pulmonary disease), auxiliary examination indicators (such as chest X-ray findings, serum albumin levels, hemoglobin levels), operative factors (such as surgical duration, emergency surgery, unplanned re-operation, surgical incision), anesthetic factors (such as anesthesia method, duration of mechanical ventilation, blood transfusion, duration of postoperative bed rest), and invasive procedures (such as deep vein catheterization, radial artery cannulation, and urethral catheterization). In addition, the study also documented the causative pathogens and the presence of multi-drug resistant bacteria (MDR), which was defined as non-susceptibility to at least one agent in three or more antimicrobial categories.

Statistical analysis was performed using SPSS 25.0 and Rstudio software. For normally distributed and homogenous metric data, the mean ± standard deviation (x ± s) was used to represent the data, and the independent sample T-test was employed to compare between groups. For metric data that did not meet the normal distribution and homogeneity of variance, the median (interquartile range) was used, and the rank-sum test was utilized for group comparisons. Frequency was used to express all count data, and the X² test was employed for group comparisons. All factors that showed statistical significance in the univariate analysis were included in the multivariate regression analysis using a binary logistic regression model. A risk prediction model and a nomogram model were established using Rstudio software. The fitting degree of the model was evaluated using the Hosmer-Lemeshow test, and the predictive performance was assessed using the receiver operating characteristic curve (ROC). The bootstrap method was used to repeat the sampling 1000 times for internal validation and use the C-index for evaluation. Furthermore, the prediction model was evaluated using calibration curves and DCA. A significance level of P<0.05 was considered statistically significant.

Within the scope of this study, a total of 831 patients who underwent craniotomy was examined. Among them, 150 patients were diagnosed with pneumonia, while 47 patients were excluded (46 had pre-existing pneumonia, and 1 patient discontinued surgery midway). Ultimately, 103 patients were diagnosed with postoperative pneumonia, resulting in an overall incidence rate of 12.39% (103/831). The flowchart of patient selection is shown in Figure 1 . Co-culture analysis of sputum specimens from these patients revealed the isolation of 68 pathogenic strains. Among them, 51 strains were gram-negative bacteria, predominantly Klebsiella pneumoniae (26.47%), while 13 strains were gram-positive bacteria, mainly Staphylococcus epidermidis (13.24%). Additionally, 4 strains of fungi were identified, with Candida albicans (4.41%) being the most prevalent. Furthermore, 19 strains (27.94%) exhibited multidrug resistance, including 9 strains (13.24%) of carbapenem-resistant Enterobacteriaceae, 8 strains (11.76%) of carbapenem-resistant Acinetobacter baumannii, and 2 strains (2.94%) of carbapenem-resistant Pseudomonas aeruginosa. The distribution of pathogens is detailed in Table 1 .

Table 2 presents the significant risk factors associated with postoperative pneumonia as identified through univariate analysis. Notably, 15 perioperative factors demonstrated a significant correlation with postoperative pneumonia (P<0.05). These factors include age, smoking, hypertension, diabetes, coma (GCS<8), ASA physical status, emergency surgery, postoperative hemoglobin, postoperative albumin, duration of ventilation, blood transfusion, surgical duration, unplanned re-operation, surgical incision, and deep vein catheterization. However, factors such as gender, body mass index, alcohol consumption, chronic obstructive pulmonary disease, preoperative hemoglobin levels, preoperative serum albumin levels, anesthesia method, radial artery cannulation, and urethral catheterization did not demonstrate a significant correlation with postoperative pneumonia (P>0.05).

For multivariate logistic regression analysis, postoperative pneumonia was defined as the dependent variable, while risk factors were considered as independent variables. The variables were assigned values with a significance level of α=0.05. The analysis revealed five independent risk factors for postoperative pneumonia after craniotomy ( Table 3 ), as follows: smoking history (OR: 4.123 [1.619, 10.499], P=0.003), surgical duration (OR: 3.447[1.332, 8.919], P=0.011), postoperative albumin (OR: 3.525 [1.380, 9.002], P=0.008), unplanned re-operation (OR: 28.223[5.840, 136.406], P<0.001), and deep vein catheterization (OR: 5.766 [1.956, 16.996], P=0.001).

Based on the results of the multivariate logistic regression analysis, a risk prediction model using logistic regression was developed to predict the likelihood of postoperative pneumonia after craniotomy. The model’s predicted probability was calculated using the equation: logit (P) = 1/[1+e ^{- 4.372 + 1.417× smoking + 1.260 × postoperative albumin + 1.237× surgical duration +1.752 × deep vein catheterization +3.340 × unplanned re-operation}]. The Hosmer-Lemeshow test confirmed that this model exhibited a good fit with the observed values (X² = 3.871, P=0.869). A nomogram model, incorporating all independent risk factors, is presented in Figure 2 for the prediction of postoperative pneumonia in patients undergoing craniotomy.

The C-index of the model was 0.898(95%CI, 0.853~0.941). The greater the C-index was, the better the differentiation of the model was, indicating that the accuracy of nomogram prediction was great. The receiver operator characteristic (ROC) analysis revealed a remarkable area under the curve (AUC) of 0.898 (95% CI: 0.856-0.940), with a sensitivity of 79.6% and a specificity of 85.4%, indicating a commendable predictive performance, as shown in Figure 3 . The calibration plots of the model showed that the nomogram had a good fit with the reference line, as shown in Figure 4 , with p-values of 0.797, respectively (p > 0.05). The Brier scores were 0.127, respectively (close to 0), indicating that the prediction of the probability of POP after craniotomy by the nomogram model was consistent with the actual infection percentage of the observed population.

The DCA determined the value of the clinical application of the nomogram model by calculating the net benefit under the POP risk threshold probability of each patient after craniotomy. The horizontal coordinate of the DCA was the threshold probability of high risk, and the vertical coordinate was the net benefit (NB). The high-risk threshold was set as (0, 1), and the net benefit rate and the range of effective pretest probability were assessed by subtracting the false-positive population that was incorrectly judged by the model. The threshold probability of the model set was between 0.01~1, all of which had a net benefit rate >0 and had clinical practical value, which suggested that the model had good clinical value in predicting POP after craniotomy, as shown in Figure 5 .

The retrospective analysis of outcomes of patients undergoing craniotomy with postoperative pneumonia is presented in Table 4 . The findings demonstrated a significantly prolonged hospitalization duration in the POP group (28.09 ± 13.97) compared to the non-POP group (18.29 ± 7.13). Moreover, the proportion of neurosurgical patients requiring hospitalization for more than 14 or 30 days was significantly higher in the POP group than in the non-POP group (P<0.05). Additionally, the rate of ICU admission in the POP group was significantly higher than that in the non-POP group (P<0.05). Although there was no significant difference in the postoperative mortality rate between the two groups (P>0.05), it is worth noting that all three postoperative deaths occurred in the POP group.