Participants between February 2023 and April 2023, we conducted a prospective observational study at the Department of Cardio-Thoracic Surgery. The study was ethically approved by the institutional review board was registered on the Chinese Clinical Trial Registry (ChiCTR2000038762).

Between February 1, 2023, and April 14, 2023, this study recruited patients who underwent cardiac surgery under general anesthesia for cardiopulmonary bypass with the inclusion criteria of being admitted to the Cardio-Thoracic Surgery ICU for more than 24 h and providing written informed consent after receiving information sheets and potential risk disclosures. Exclusion criteria included patients under the age of 18, those diagnosed with delirium and stroke during admission, those with a Glasgow Coma Scale score of ≤8 points who required intubation and mechanical ventilation, those who were deeply sedated (as determined by Richmond Agitation Sedation Scale scores of −4 and −5), and those with alcohol withdrawal reactions (15), patients with infective endocarditis and preoperative fever. A total of 232 eligible patients were enrolled (Figure 1), and measures were taken to ensure the validity and robustness of the research findings.

Delirium is a challenging condition to diagnose in the intensive care unit (ICU), and many patients may go unrecognized. Despite the use of many delirium assessment instruments in published studies, the most widely used instrument is the Confusion Assessment Method (CAM) (16). CAM has a sensitivity of 94% and specificity of 89% compared to the gold standard diagnosis by psychiatrists (17). CAM-ICU was developed to accurately diagnose delirium in ICU patients who are often unable to speak due to mechanical ventilation. CAM-ICU has a sensitivity of 95% and specificity of 89% (16, 18).

In surgical patients who do not have permanent neurological impairment, postoperative delirium (POD) is identified using the Confusion Assessment Method for the ICU (CAM-ICU) (18) and evaluated twice daily for seven consecutive days. This method allows for consistent monitoring and diagnosis of POD, which is a common complication after surgery and can have significant negative effects on patient outcomes (19).

We define high fever as bladder temperature greater than 39°C and continuously monitor body temperature for 24 h postoperatively. This definition of high fever allows for consistent and accurate measurement of postoperative fever, which can indicate a pathological state and may require further evaluation and treatment.

In this study, we conducted a comprehensive analysis of the data using a range of statistical methods. Continuous variables were assessed for normality using the Kolmogorov–Smirnov test and reported either as mean ± standard deviation or median with interquartile ranges (Q1–Q3), depending on their distribution. Student's t-test and Mann–Whitney U-test were used to analyze normally and non-normally distributed continuous variables, respectively. Categorical variables were presented as frequencies and percentages and analyzed using either chi-squared test or Fisher's exact test. All statistical analyses were two-tailed, and a P-value less than 0.05 was considered statistically significant. Additionally, we employed single-variable binary logistic regression analysis to assess the relationship between various variables and the outcome, calculating odds ratios and 95% confidence intervals.

After conducting univariate analysis, variables that demonstrated statistical significance with a P-value less than 0.05 were included in a stepwise (backward: conditional) multivariate logistic regression analysis model to establish a predictive model. Furthermore, we constructed a nomogram using the variables with a P-value less than 0.05 in the multivariate analysis to facilitate clinical decision-making. This approach allowed us to identify the most significant predictors of the outcome of interest and develop a useful tool to aid in clinical management.

To assess the performance of the developed nomogram for predicting the probability of postoperative neurological complications in aortic surgery, we conducted internal validation using the bootstrap method with 1,000 resamples, evaluating both discrimination and calibration. The discrimination ability was assessed using the C-statistic, equivalent to the area under the receiver operating characteristic (ROC) curve (20). Calibration was assessed by plotting calibration curves and calculating the Brier score, which is the squared difference between observed and predicted probabilities (21). Furthermore, we conducted a decision curve analysis (DCA) to assess the clinical usefulness of the nomogram across various threshold probabilities. The statistical analysis was performed using IBM SPSS Statistics 26 and R.4.2.2, and significance was determined at P < 0.05. This rigorous approach enabled us to gain valuable insights into the data and identify potential predictors of the outcome of interest.

During the research period from February 1, 2023, to April 14, 2023, our analysis included a total of 232 patients, as shown in Figure 1. Preoperative risk factors, such as age, gender, BMI, hypertension, NYHA class, as well as important intraoperative risk factors, including surgical category and cardiopulmonary bypass time (CPB), were identified, as presented in Table 1. The overall incidence of POD was found to be 12.1%, as shown in Table 1, and was significantly associated with CPB and postoperative lactate levels. Moreover, there were significant variations in Tmax between the POD and Non-POD groups. POD was also found to be related to ICU length of stay.

The results of multivariate analysis for POD are listed in Table 2, For this phase of the analysis, three variables were determined to be statistically significant. Multivariate analysis identified that lactate [Odds ratio (OR) = 1.787, 95% CI, 1.192–2.785], Tmax (OR = 11.290, 95% CI, 4.369–32.129), and CPB time (OR = 1.015, 95% CI, 1.004–1.026) were independent predictors for POD (Table 2).

The prediction model exhibited a C-statistic value of 0.852 (95% CI, 0.763–0.941), demonstrating excellent discriminatory power. Sensitivity and specificity, based on the area under the receiver operating characteristic (AUROC) curve, were 91.2% and 67.9%, respectively (Figure 2). The apparent calibration curve closely approximated the ideal 45° line, indicating consistent agreement between observed and predicted probabilities within the development cohort (Figure 3). To mitigate any potential overoptimism in the model, internal validation using the 1,000 bootstrap approach was performed, which confirmed its robust discrimination ability with a Brier score of 0.0686 (Figure 3). Utilizing these three candidate variables, we constructed a nomogram for predicting the probability of postoperative delirium (Figure 4). Furthermore, decision curve analysis demonstrated a favorable net clinical benefit (Figure 5).