Development and validation of a machine learning-based model for predicting delirium risk in postoperative brain tumor patients in the Intensive Care Unit

Development and validation of a machine learning-based model for predicting delirium risk in postoperative brain tumor patients in the Intensive Care Unit | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Development and validation of a machine learning-based model for predicting delirium risk in postoperative brain tumor patients in the Intensive Care Unit Li Liu, Siyao Huang, Yating Huang, Jianmei Xu This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8893713/v1 This work is licensed under a CC BY 4.0 License Status: Under Revision Version 1 posted 9 You are reading this latest preprint version Abstract Objective This study aimed to clarify the incidence and influencing factors of delirium in ICU patients after brain tumor surgery, construct and validate delirium risk prediction models using multiple machine learning algorithms, identify the optimal model, and develop a personalized risk calculation tool to provide evidence-based support for early precise identification of high-risk patients and targeted preventive interventions. Methods Consecutive convenient sampling was adopted. A total of 600 patients who underwent brain tumor surgery in a Grade A tertiary hospital in Nanchang (July 2021–December 2024) served as the modeling cohort, and 160 similar patients (January–August 2025) as the external validation cohort. LASSO regression screened independent risk factors from 26 candidates. Five models (LR, ANN, RF, DT, SVM) were constructed and evaluated by sensitivity, AUC, DCA, etc. SHAP interpreted the optimal model’s feature importance. Results Five independent risk factors were identified: age, APACHE II score, postoperative GCS score, use of diuretics or dehydrating agents, and mechanical ventilation duration. The SVM model performed best: validation set AUC = 0.857 (95%CI:0.814–0.898), test set AUC = 0.847, external validation accuracy 70.00%. SHAP showed diuretics or dehydrating agents were the most important feature (mean |SHAP|=0.1625). An online risk calculator was developed for convenient personalized assessment. Conclusions The SVM-based predictive model has excellent efficacy and generalizability. The easy-to-operate risk calculator can effectively identify high-risk patients early, providing scientific support for precise delirium prevention and control. Brain tumor Machine learning Postoperative delirium Predictive model Intensive Care Unit (ICU) Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1. Introduction Postoperative delirium (POD) is a common neuropsychiatric complication in ICU patients after brain tumor surgery, characterized by disturbances in consciousness, impaired attention, and altered cognitive function. It is classified into hyperactive, hypoactive, and mixed subtypes [ 1 – 2 ]. POD not only hinders patient recovery, prolongs hospital stays, and increases medical costs but also is closely associated with long-term cognitive impairment and elevated mortality, imposing a substantial burden on patients’ families and the healthcare system [ 3 – 6 ]. Studies have shown that 30%–40% of delirium cases can be prevented through early intervention [ 7 ], and the precise identification of high-risk patients is a prerequisite for effective prevention and control. The reported incidence of POD in ICU patients after brain tumor surgery varies widely, ranging from 28.5% to 37.8% [ 8 ], which is related to differences in study population characteristics, assessment tools, and medical management levels. In recent years, with the in-depth application of artificial intelligence in healthcare, machine learning algorithms have become important tools for constructing disease risk prediction models due to their ability to handle complex nonlinear relationships and mine hidden patterns in large datasets [ 9 ]. Currently, several studies on predictive models for POD have been conducted worldwide, but most focus on other diseases such as hip fracture and cardiac surgery [ 10 – 11 ]. There are few models specifically for ICU patients after brain tumor surgery, and existing studies mostly adopt a single Logistic regression algorithm without fully comparing the performance of multiple machine learning models. Additionally, the lack of in-depth analysis of model interpretability limits their clinical application. Furthermore, although previous studies have made some progress in identifying risk factors for POD after brain tumor surgery, with advanced age, surgical trauma, and medication use being confirmed as associated factors [ 8 , 12 – 13 ], the results of different studies are inconsistent, and a unified risk assessment system has not yet been established. Based on clinical big data, this study integrates multiple machine learning algorithms to construct predictive models, systematically compares their performance, and conducts interpretability analysis. Addressing the limitations of previous studies, this research focuses on the clinical positioning of delirium incidence, the limitations of external model validation, and the advantages over traditional prediction methods, aiming to provide a more precise and practical tool for the prevention and control of POD in ICU patients after brain tumor surgery. 2. Materials and Methods 2.1. Study population Consecutive convenient sampling was used to recruit 600 patients admitted to the Neurosurgery ICU of a Grade A tertiary hospital in Nanchang, Jiangxi Province, after brain tumor surgery from July 2021 to December 2024 as the modeling cohort. The modeling cohort was randomly divided into a training set (n = 420) and a test set (n = 180) at a ratio of 7:3 using a random number table. An additional 160 patients from the same institution from January to August 2025 were enrolled as the external validation cohort. 2.2. Inclusion and exclusion criteria Inclusion criteria: (1) Age ≥ 18 years; (2) Admitted to the Neurosurgery ICU after brain tumor surgery; (3) Complete clinical data for the modeling cohort; (4) Written informed consent obtained from patients or their family members for the external validation cohort. Exclusion criteria: (1) Preoperative cognitive impairment such as mental illness or Alzheimer’s disease; (2) Severe visual or auditory sensory impairment; (3) Persistent coma (GCS score ≤ 8) during the study observation period, making it impossible to complete the assessment; (4) ICU stay < 24 hours after surgery; (5) Withdrawal of treatment or death during the study; (6) Incomplete clinical data or non-compliance with the study protocol. 2.3. Study tools and data collection The Confusion Assessment Method for the Intensive Care Unit (CAM-ICU) was used to assess delirium. This scale evaluates four dimensions: consciousness state, attention, altered level of consciousness, and thinking, with a specificity of 98.6% and sensitivity of 90.2% [14]. A two-step screening process was adopted by combining the CAM-ICU with the Richmond Agitation-Sedation Scale (RASS) [15]. A self-designed data collection form was used to collect three categories of patient information: ① Patient-related factors: age, gender, smoking history, drinking history, history of mental illness, C-reactive protein (CRP) level, diabetes mellitus, hypertension, electrolyte imbalance, chronic lung disease, postoperative pain, sleep disturbance, APACHE II score, and postoperative GCS score; ② Tumor-related factors: bilateral brain tumor occupancy, tumor location, tumor diameter, and World Health Organization (WHO) tumor grade; ③ Medical risk factors: physical restraint, frontal craniotomy, use of benzodiazepines, use of diuretics or dehydrating agents, surgical duration, duration of mechanical ventilation, blood transfusion, and tracheotomy. Data were extracted from the hospital electronic medical record system and nursing documents, and double data entry and verification were performed by a uniformly trained research team to ensure data accuracy. 2.4. Data preprocessing and statistical analysis Data were entered using Excel 16.0. Descriptive statistics and univariate analysis were performed using SPSS 26.0, and model construction and validation were conducted using RStudio 4.5.1. Continuous variables were expressed as median (interquartile range, IQR), and the Mann-Whitney U test was used for intergroup comparisons. Categorical variables were expressed as frequencies (percentages), and the chi-square test was used for intergroup comparisons. Data preprocessing: For variables with missing values < 15%, continuous variables were imputed with the mean, and categorical variables were imputed using multiple imputation. Z-score standardization was performed on continuous variables. Independent risk factors were screened through univariate analysis ( P < 0.05) and LASSO regression. Five predictive models (LR, ANN, RF, DT, SVM) were constructed. Model parameters were optimized using 10-fold cross-validation and Bootstrap resampling. Model performance was evaluated using AUC, sensitivity, specificity, accuracy, precision, F1-score, recall, Youden’s index, calibration curves, Brier score, and DCA curves. The SHAP method was used to interpret the feature importance of the optimal model. A two-sided P < 0.05 was considered statistically significant. 3. Results 3.1. Baseline characteristics of the study population In the modeling cohort (n = 600), patients aged 18–80 years, with 248 males (41.33%) and 352 females (58.67%). Delirium occurred in 165 patients, with an incidence of 27.5%. The training set included 420 patients and the test set included 180 patients. In the external validation cohort (n = 160), there were 76 males (47.50%) and 84 females (52.50%), with 43 cases of delirium (incidence: 26.9%). 3.2. Screening of risk factors for delirium in ICU patients after brain tumor surgery Univariate analysis showed significant differences between POD and non-POD patients in 13 variables (all P < 0.05): age, history of diabetes mellitus, history of hypertension, sleep disturbance, postoperative electrolyte imbalance, APACHE II score, postoperative GCS score, CRP level, tumor location, WHO tumor grade, physical restraint, use of diuretics or dehydrating agents, and duration of mechanical ventilation. Detailed data descriptions and univariate analysis results are presented in Table 1 . Table 1 Univariate analysis of risk factors for delirium in 600 ICU patients following brain tumor surgery Predictor Non-POD ( N = 435 ) POD ( N = 165 ) Test statistic P Value Age[years, M ( P 25, P 75)] 51.00 (42.50, 58.00) 64.00 (54.00, 70.00) Z = 9.214 < 0.001 Sex(n, %) χ2 = 0.270 0.603 Male 177(40.7) 71(43.0) Female 258(59.3) 94(57.0) Tobacco use history(n, %) χ2 = 0.584 0.445 No 422(97.0) 158(95.8) Yes 13(3.0) 7(4.2) Alcohol use history(n, %) χ2 = 0.120 0.729 No 408(93.8) 156(94.5) Yes 27(6.2) 9(5.5) Psychiatric history(n, %) χ2 = 1.539 0.215 No 432(99.3) 162(98.2) Yes 3(0.7) 3(1.8) History of diabetes mellitus(n, %) χ2 = 5.114 0.024 No 418(96.1) 151(91.5) Yes 17(3.9) 14(8.5) History of hypertension(n, %) χ2 = 10.714 0.001 No 363(83.4) 118(71.5) Yes 72(16.6) 47(28.5) Chronic pulmonary disease(n, %) χ2 = 0.395 0.530 No 432(99.3) 163(98.8) Yes 3(0.7) 2(1.2) Postoperative pain(n, %) χ2 = 1.696 0.193 No 8(1.8) 6(3.6) Yes 427(98.2) 159(96.4) Sleep disturbances(n, %) χ2 = 10.100 0.001 No 429(98.6) 155(93.9) Yes 6(1.4) 10(6.1) Electrolyte disturbances(n, %) χ2 = 4.477 0.034 No 189 56 Yes 246 109 APACHE II score [ M ( P 25, P 75)] 7.00(5.00, 9.00) 10.00 (8.00, 13.00) Z = 8.298 < 0.001 GCS score[ M ( P 25, P 75)] 15.00 (14.00, 15.00) 14.00 (14.00, 15.00) Z = -4.457 < 0.001 C-reactive protein concentration [mg/L, M ( P 25, P 75)] 13.10 (5.82, 32.52) 21.60 (7.97, 45.83) Z = 2.945 0.003 Bilateral brain tumor occupancy(n, %) χ2 = 2.517 0.113 No 421(96.8) 155(93.9) Yes 14(3.2) 10(6.1) Tumor site(n, %) χ2 = 22.475 0.002 Frontal lobe 60(13.8) 26(15.8) Temporal lobe 34(7.8) 24(14.5) Parietal lobe 17(3.9) 9(5.5) Occipital lobe 9(2.1) 10(6.1) Pituitarium 38(8.7) 19(11.5) Cerebellopontine angle 136(31.3) 29(17.6) Cerebellum 16(3.7) 4(2.4) Others 125(28.7) 44(26.7) Tumor diameter [cm, M ( P 25, P 75)] 3.50(2.50, 5.25) 3.60(2.50, 6.00) Z = 0.888 0.375 WHO tumor grade(n, %) χ2 = 8.645 0.034 Grade Ⅰ 312(71.7) 103(62.4) Grade Ⅱ 55(12.6) 28(17.0) Grade Ⅲ 24(5.5) 6(3.6) Grade IV 44(10.1) 28(17.0) Physical restraint(n, %) χ2 = 17.787 < 0.001 No 169(38.9) 34(20.6) Yes 266(61.1) 131(79.4) Transfrontal craniotomy approach(n, %) χ2 = 2.671 0.102 No 407(93.6) 160(97.0) Yes 28(6.4) 5(3.0) Administration of benzodiazepines(n, %) χ2 = 2.582 0.108 No 430(98.9) 160(97.0) Yes 5(1.1) 5(3.0) Administration of diuretic/dehydrating agents(n, %) χ2 = 108.164 < 0.001 No 333(65.1) 51(7.9) Yes 102(34.9) 114(92.1) Duration of mechanical ventilation[min, M( P 25, P 75)] 555.00 (450.00, 700.00) 655.00 (500.00, 1070.00) Z = 5.494 < 0.001 Duration of operation[min, M( P 25, P 75)] 345.00 (240.00, 450.00) 345.00 (250.00, 470.00) Z = 1.357 0.175 Blood transfusion(n, %) χ2 = 1.032 0.310 No 265(60.9) 93(56.4) Yes 170(39.1) 72(43.6) Tracheostomy(n, %) χ2 = 0.209 0.648 No 427(98.2) 161(97.6) Yes 8(1.8) 4(2.4) Note: APACHE II Acute physiology and chronic health evaluation Ⅱ;GCS Glasgow Coma Scale score After data standardization, the 13 independent variables were included in LASSO regression analysis using R software. Finally, five risk factors were identified as independent predictors of POD in ICU patients after brain tumor surgery: age, APACHE II score, postoperative GCS score, use of diuretics or dehydrating agents, and duration of mechanical ventilation. The optimal λ value was determined using 10-fold cross-validation (Fig. 1 A). At λ.1se = 0.039, the fluctuation of the model standard deviation within the interval defined by the two dashed lines was small (Fig. 1 A), indicating a relatively simple model structure in this region. Figure 1 B depicts the path of each variable coefficient with the change of the regularization parameter λ. It can be seen that as the λ value increases, the compression effect of the model on variables gradually enhances, thereby achieving effective screening of candidate variables and ultimately obtaining a simplified model with the minimum number of predictive variables. 3.3. Comparison of predictive model performance 3.3.1. Internal validation of predictive models and selection of the optimal model Five POD predictive models were constructed using LR, ANN, RF, DT, and SVM algorithms, respectively, based on the five variables identified by LASSO regression. Delirium occurrence was defined as the dependent variable (1 = delirium, 0 = no delirium), and the independent variables were assigned as follows: age (actual value), APACHE II score (actual value), postoperative GCS score (actual value), use of diuretics or dehydrating agents (1 = yes, 0 = no), and duration of mechanical ventilation (actual value). ROC curves were plotted with sensitivity as the ordinate and 1 - specificity as the abscissa, and the area under the ROC curve (AUROC) was used to evaluate the discrimination ability of each model. The ROC curves of the five predictive models are shown in Fig. 2 . The detailed performance of each predictive model in the training and validation sets is presented in Tables2 and 3. In the training set, the RF model exhibited excellent discrimination ability, with an AUC of 0.944 and an ideal steep upward trend of the curve. In addition, the RF model achieved a specificity of 0.947, accuracy of 0.868, precision of 0.829, and F1-score of 0.737, showing outstanding classification performance. Table 2 Performance comparison of different prediction models for POD development in postoperative ICU patients following brain tumor surgery within the training set Prediction Model AUC(95% CI ) Sensitivity Specificity Accuracy Precision Recall F1-Score Youden's Index LR 0.864(0.851 ~ 0.878) 0.577 0.921 0.825 0.738 0.577 0.648 0.498 ANN 0.865(0.851 ~ 0.878) 0.585 0.920 0.827 0.740 0.585 0.653 0.505 RF 0.944(0.937 ~ 0.951) 0.665 0.947 0.868 0.829 0.665 0.737 0.612 DT 0.876(0.863 ~ 0.889) 0.683 0.930 0.861 0.793 0.683 0.731 0.613 SVM 0.869(0.855 ~ 0.882) 0.631 0.905 0.828 0.719 0.631 0.672 0.535 Note: AUC Area Under the Curve;LR Logistic Regression༛ANN Artificial Neural Network༛RF Random Forest༛DT Decision Tree༛SVM Support Vector Machine Table 3 Performance comparison of different prediction models for POD development in postoperative ICU patients following brain tumor surgery within the validation set Prediction Model AUC(95% CI ) Sensitivity Specificity Accuracy Precision Recall F1-Score Youden's Index LR 0.854(0.809 ~ 0.895) 0.575 0.917 0.821 0.740 0.575 0.642 0.492 ANN 0.852(0.804 ~ 0.895) 0.574 0.917 0.821 0.740 0.574 0.642 0.491 RF 0.859(0.819 ~ 0.897) 0.583 0.904 0.814 0.732 0.583 0.630 0.487 DT 0.799(0.744 ~ 0.848) 0.524 0.894 0.791 0.685 0.524 0.575 0.418 SVM 0.857(0.814 ~ 0.898) 0.651 0.907 0.836 0.744 0.651 0.691 0.558 Abbreviations as per Table 2 . In the validation set, there were differences in the generalizability of the models: the AUC of SVM (0.857) was slightly lower than that of RF (0.859), indicating better generalizability of the SVM model. The SVM model had the highest sensitivity (0.651), accuracy (0.836), precision (0.744), recall (0.651), F1-score (0.691), and Youden’s index (0.558) among all models, while its specificity (0.907) was slightly lower than that of LR and ANN (both 0.917). The performance of the RF model decreased most significantly: the AUC dropped from 0.944 to 0.859, and other indicators also decreased significantly, indicating severe overfitting, with the validation set performance far lower than the training set. All indicators of the DT model decreased, among which sensitivity, accuracy, precision, recall, F1-score, and Youden’s index decreased the most, indicating a serious overfitting problem. Although the specificity of LR and ANN (both 0.917) was higher than that of SVM, their other performance indicators were lower than those of SVM. Comprehensive performance analysis of the validation set showed that the SVM model had the optimal efficacy in predicting POD after brain tumor surgery: its sensitivity, accuracy, precision, recall, F1-score, and Youden’s index all ranked first, and the performance fluctuation from the training set to the validation set was small (AUC only decreased by 0.012), showing the best comprehensive performance. Calibration is a key criterion for evaluating the accuracy of a model in predicting the probability of a specific outcome event in individuals. Calibration curves can intuitively show the consistency between predicted and actual risks. Therefore, calibration curves were used to evaluate the calibration of each model in this study. From the comprehensive performance of calibration curves and Brier scores, LR had the lowest Brier score (0.129), and its calibration curve was the closest to the ideal dashed line throughout the entire range with no significant fluctuations, indicating the optimal calibration performance. The SVM model had a Brier score of 0.139, and its curve was generally close to the ideal dashed line, with only a slight deviation in the medium-to-high probability range, ranking second in performance. The RF (Brier = 0.140) and ANN (Brier = 0.141) models had good curve fitting, with overall calibration performance at a good to moderate level. The DT model had the highest Brier score (0.157), and its calibration curve fluctuated sharply, especially a significant decline in the high predicted probability range (> 0.8), indicating a serious overestimation problem and the worst calibration performance (Fig. 3 A). DCA is a method to evaluate the feasibility of clinical decisions by considering the possible range of patient risks and benefits. DCA curves were used in this study to assess the actual clinical utility of the predictive models. The DCA curves of each delirium risk prediction model for ICU patients after brain tumor surgery are shown in Fig.3B. The results showed that the ANN model had the best DCA curve performance, with an average net benefit of 0.197, followed by LR (average net benefit: 0.194) and SVM (average net benefit: 0.188). Based on the superior performance of the SVM predictive model in various aspects compared with the other four models, the SVM model was finally selected as the optimal model in this study. In the test set, the SVM predictive model achieved an AUC of 0.847, sensitivity of 0.896, specificity of 0.712, accuracy of 0.761, precision of 0.531, recall of 0.896, F1-score of 0.667, and Youden’s index of 0.608. The ROC curve is shown in Fig.4. 3.3.2. External validation of the optimal delirium predictive model Data of 160 patients admitted to the Neurosurgery ICU of the same Grade A tertiary hospital in Nanchang, Jiangxi Province, after brain tumor surgery from January to August 2025 were collected for external validation of the predictive model. The validation cohort included 76 male patients (47.50%) and 84 female patients (52.50%), among whom 43 cases (26.90%) developed POD. In external validation, the discrimination, calibration, and clinical utility of the predictive model are shown in Fig. 5 . The SVM predictive model achieved an accuracy of 70.00%, specificity of 67.52%, and sensitivity of 76.74% in the external validation set. 3.4. Interpretation and application of the delirium risk predictive model 3.4.1. Feature importance The SHAP method was used to measure the importance of each feature. Figure 6 A shows the importance of the five predictive factors for delirium in ICU patients after brain tumor surgery, among which the use of diuretics or dehydrating agents was the most important feature. Based on the SHAP package, Fig. 6 B shows the degree of influence of each feature on delirium in ICU patients after brain tumor surgery. 3.4.2. Introduction to model application Based on the core predictive indicators and the model, a user-friendly POD risk calculator was developed, which can be accessed at https://pod66.shinyapps.io/shiny-svm1/ . The optimal cutoff value of the SVM model calculated based on Youden’s index was 0.6006. That is, when medical staff input the patient’s age, APACHE II score, postoperative GCS score, use of diuretics or dehydrating agents, and duration of mechanical ventilation into the calculator, a result > 60.06% indicates that the patient has a risk of developing POD, and early intervention is required. The page view of the online risk calculator is shown in Fig. 7 . 4. Discussion 4.1. Incidence of delirium in ICU patients after brain tumor surgery The incidence of delirium in ICU patients after brain tumor surgery is generally high. Previous studies have clearly shown that it can range from 28.5% to 37.8% [ 8 ], highlighting the high risk of this complication in neurosurgical critically ill patients. The results of this study showed that the incidence of delirium in the modeling cohort was 27.5%, and that in the external validation cohort was 26.9%. Although this result is slightly lower than the lower limit of the above range, it is still in the clinically high-risk range, and is highly close to the 28.5% reported by Chen et al. [ 8 ], only lower than the 37.8% reported by French et al. [ 16 ]. The discrepancies in incidence rates can be attributed to four key factors: ① Stringency of inclusion and exclusion criteria: Patients with persistent coma (GCS score ≤ 8) were excluded because their severely impaired consciousness made it impossible to complete effective delirium assessment, thus ensuring the reliability of endpoint event determination. Meanwhile, to avoid confusion between pre-existing chronic cognitive impairment and acute postoperative delirium, all patients with pre-existing cognitive impairment were also excluded to ensure that the model predicted newly developed acute brain dysfunction. In addition, patients with an ICU stay of less than 24 hours were excluded to ensure that all enrolled subjects had a unified and sufficiently long observation window, thereby completely capturing delirium events and avoiding missing endpoint data due to early transfer out. ② Accuracy of the assessment system: This study adopted a dual assessment model combining the CAM-ICU scale and consultation with neurologists, with a specificity of 98.6% and a sensitivity of 90.2% [ 14 ], which effectively reduced missed diagnoses and misdiagnoses. In contrast, some studies used the Intensive Care Delirium Screening Checklist (ICDSC) or subjective assessment by a single medical staff member, which may lead to high or low incidence rates due to screening errors. ③ Standardization of medical management: The hospital in this study is a Grade A tertiary hospital, where refined sedation and analgesia management, sleep protection, infection prevention and control, and other interventions have been routinely carried out during the perioperative period. This has reduced the impact of controllable risk factors to a certain extent, making the incidence rate closer to the inherent risk level of the disease. ④ Heterogeneity of population characteristics: There are differences in sample size, patient age structure (the higher the proportion of the elderly, the higher the incidence rate), tumor type (the higher the degree of malignancy, the higher the risk), and surgical complexity among different studies. For example, the proportion of patients with WHO grade IV tumors in this study was 12.0% (72 out of 600 cases), while some studies focusing on malignant brain tumors are more likely to approach the high value of 37.8% due to the concentration of high-risk cases. Although the incidence of 27.5% in this study is slightly lower than the high value range reported in previous studies, it is still at a clinically high-risk level, fully indicating that the prevention and control of delirium in ICU patients after brain tumor surgery still faces severe challenges. This incidence not only objectively reflects the inherent risks of the disease such as surgical trauma, changes in the intracranial environment, and damage to the blood-brain barrier but also reflects the effectiveness of standardized medical management in intervening in controllable factors (such as infection and inappropriate sedation). The results have strong representativeness for Grade A tertiary hospitals and institutions with similar medical levels. It should be noted that in primary hospitals with relatively limited medical resources and inadequate intervention measures, the incidence of delirium may be closer to the high value range of 28.5%–37.8%. Therefore, when popularizing the predictive model of this study, it is necessary to calibrate the model threshold in a targeted manner according to the baseline level of delirium occurrence in local medical institutions to ensure its universality and clinical practicality. 4.2. Clinical implications of delirium risk factors This study identified age, APACHE II score, postoperative GCS score, use of diuretics or dehydrating agents, and duration of mechanical ventilation as independent risk factors for delirium in ICU patients after brain tumor surgery, which is partially consistent with previous research results [ 8 , 12 – 13 ], but also highlights the characteristics of neurosurgery. Age is a well-recognized risk factor for delirium. In this study, the median age of the delirium group was significantly higher than that of the non-delirium group. Physiological changes associated with aging, such as reduced neural reserve, impaired integrity of the blood-brain barrier, and inflammaging, reduce the tolerance of elderly patients to surgical stress and anesthesia. Cholinergic system dysfunction further increases the susceptibility to delirium [ 17 – 20 ]. The APACHE II score reflects the severity of the patient’s condition; a high score indicates multiple organ dysfunction, which indirectly impairs cerebral perfusion through hypotension, hypoxia, etc., leading to neurotransmitter system disorders and inducing delirium [ 21 ]. A decreased postoperative GCS score indicates brain function damage, which may result from edema in the surgical area, increased intracranial pressure, or occult hemorrhage. This damage disrupts cortico-subcortical network connections, impairs attention and executive function, and simultaneously increases the risk of pulmonary infection and hypoxemia, further promoting the occurrence of delirium [ 22 ]. Prolonged duration of mechanical ventilation increases the risk of delirium through multiple pathways such as sleep deprivation, use of sedative drugs, and respiratory complications [ 23 – 24 ]. In this study, the duration of mechanical ventilation in the delirium group (655 min) was significantly longer than that in the non-delirium group (555 min), which is consistent with the research results of Habeeb-Allah et al. [ 23 ]. Notably, the use of diuretics or dehydrating agents was the most important risk factor in this study (SHAP value: 0.1625), with a usage rate of 92.1% in the delirium group, much higher than 34.9% in the non-delirium group. Patients after brain tumor surgery often use dehydrating agents for increased intracranial pressure, but these drugs can cause electrolyte imbalances (especially hyponatremia), leading to cerebral edema, abnormal nerve impulse conduction, and disruption of neurotransmitter balance. Given that the blood-brain barrier is damaged after brain tumor surgery, patients are more sensitive to electrolyte imbalances [ 25 ]. This mechanism is particularly prominent in neurosurgical patients, providing a clear target for clinical precise intervention. 4.3. Performance and advantages of machine learning models This study compared five machine learning algorithms and found that the SVM model had the best comprehensive performance, with good discrimination, calibration, and clinical utility, as well as strong stability and generalizability. The AUC only decreased by 0.012 from the training set to the validation set, and the accuracy in external validation reached 70.00%, which is significantly better than the LR, ANN, RF, and DT models. The RF model performed excellently in the training set (AUC = 0.944) but its performance decreased significantly in the validation set, suggesting overfitting, which is related to the characteristics of the RF algorithm that is susceptible to noise data and has high requirements for sample size [ 26 ]. Compared with previous single Logistic regression models, the SVM model has the following advantages: ① It can effectively handle complex nonlinear relationships and interactions between variables, which is more in line with the complexity of clinical data; ② It improves the robustness of the model by maximizing the classification margin and reduces the risk of overfitting; ③ It has stronger ability to process high-dimensional data and performs more stably in multi-factor prediction [ 27 – 29 ]. In addition, this study used the SHAP method to interpret the SVM model, clarifying the importance and direction of action of each risk factor, solving the black-box problem of traditional machine learning models, and enhancing the trust and acceptance of the model by clinical medical staff [ 30 ]. Compared with other postoperative delirium predictive models, the model in this study has the following characteristics: ① It focuses on the high-risk population of ICU patients after brain tumor surgery, and the screening of risk factors is more specialized, especially highlighting the use of diuretics or dehydrating agents as a characteristic risk factor in neurosurgery; ② It integrates five machine learning algorithms and screens the optimal model after systematically comparing their performance, avoiding the limitations of a single algorithm; ③ An online risk calculator was developed, which is easy to operate and can generate personalized risk assessment results in real time, meeting the needs of rapid clinical decision-making; ④ Internal 10-fold cross-validation and external independent cohort validation were performed to ensure the reliability and generalizability of the model. 4.4. Connection and comparison with previous studies The results of this study form a good connection with previous studies on postoperative delirium. In terms of risk factors, factors such as advanced age, APACHE II score, and duration of mechanical ventilation have been confirmed to be associated with postoperative delirium in multiple studies [ 8 , 12 – 13 ]. This study further verified the importance of these factors in the population of ICU patients after brain tumor surgery, and at the same time added the use of diuretics or dehydrating agents as a neurosurgery-specific risk factor, enriching the risk factor spectrum of postoperative delirium. In terms of predictive models, Davoudi et al [ 31 ] constructed a postoperative delirium predictive model based on seven machine learning algorithms, with the generalized additive model achieving an AUC of 0.860; Corradi et al. [ 32 ] also achieved good predictive results using a model constructed by the random forest algorithm. The AUC of the SVM model in this study is 0.857, which is at the same level as the above studies. However, this study focuses on the specific population of ICU patients after brain tumor surgery, with more targeted risk factors and stronger model interpretability. Compared with traditional non-machine learning studies, the predictive efficacy of the model in this study is better. For example, Li et al. [ 33 ] constructed a predictive model for delirium in ICU patients after brain tumor surgery using Logistic regression, with an AUC of 0.78, which is lower than the 0.857 of the SVM model in this study. This benefit from the advantage of machine learning algorithms in processing complex data. However, it should be objectively recognized that machine learning algorithms are not absolutely superior to traditional methods. The advantages of Logistic regression models are simple principles, easy interpretation of results, and low requirements for data quality, which still have application value in scenarios with small sample sizes and clear variable relationships [ 34 – 35 ]. The advantage of the model in this study is that it can more accurately mine data patterns and improve prediction performance in the case of large samples, multiple factors, and complex relationships. Therefore, in clinical applications, appropriate prediction methods should be selected according to the actual situation, rather than blindly pursuing complex algorithms. 5. Limitation First, this study adopted a single-center retrospective design. Although the sample size meets statistical requirements, there is still selection bias, and the characteristics of the patient population are relatively single, which may affect the generaliz 6. Conclusion The incidence of delirium in ICU patients after brain tumor surgery is 27.5%. Age, APACHE II score, postoperative GCS score, use of diuretics or dehydrating agents, and duration of mechanical ventilation are independent risk factors. The delirium risk pre Declarations Conflicts of interest On behalf of all authors, the corresponding author states that there is no conflict of interest. Ethical standard statement This observational study was conducted in accordance with the ethical principles of the Declaration of Helsinki. The study protocol was reviewed and approved by the Biomedical Research Ethics Committee of the Second Affiliated Hospital of Nanchang University. As this was a low-risk observational study, it underwent an expedited review process according to the committee's guidelines. Therefore, a specific approval number was not issued; instead, the approval is validated by the official notification dated 24/02/2025. Informed consent was obtained from the patient, agreeing to the use of her medical information for article publication. All authors declare that the submitted work has not been published before (neither in English nor in any other language) and that the work is not under consideration for publication elsewhere. Acknowledgments We extend our sincere gratitude to the research colleagues, clinicians, nursing staff, and administrative support team of the Second Affiliated Hospital of Nanchang University for their essential contributions, and to all participants and their caregivers for generously donating their time—their invaluable cooperation formed the cornerstone of this study. Author contributions Conception and design: Li Liu. Acquisition of data: Li Liu, Siyao Huang and Yating Huang. Analysis and interpretation of data: Li Liu, Siyao Huang. Drafting the article: Li Liu. Critically revising the article: Li Liu. Reviewed submitted version of manuscript: Yating Huang and Jianmei Xu. Approved the final version of the manuscript on behalf of all authors: Jianmei Xu. Statistical analysis: Li Liu, Siyao Huang and Yating Huang. Administrative/technical/material support: Jianmei Xu. Study supervision: Jianmei Xu. Data Availability The data supporting the conclusions of this article will be made available by the corresponding author, without undue reservation, to any qualified researcher upon reasonable request. References Eschweiler GW, Czornik M, Herrmann ML et al (2021) Presurgical screening improves risk prediction for delirium in elective surgery of older patients: The PAWEL RISK Study. Front Aging Neurosci 13:679933. https://doi.org/10.3389/fnagi.2021.679933 Katayama ES, Stecko H, Woldesenbet S et al (2024) The role of delirium on short - and long-term postoperative outcomes following major gastrointestinal surgery for cancer. Ann Surg Oncol 31(8):5232–5239. https://doi.org/10.1245/s10434-024-15358-x Huang H, Li H, Zhang X et al (2021) Association of postoperative delirium with cognitive outcomes: A meta-analysis. J Clin Anesth 75:110496. https://doi.org/10.1016/j.jclinane.2021.110496 Xiang PY, Boyle L, Short TG et al (2023) Incidence of postoperative delirium in surgical patients: An observational retrospective cohort study. Anaesth Intensive Care 51(4):260–267. https://doi.org/10.1177/0310057X231156459 Ou-Yang CL, Ma LB, Wu XD et al (2024) Association of sleep quality on the night of operative day with postoperative delirium in elderly patients: A prospective cohort study. Eur J Anaesthesiol 41(3):226–233. https://doi.org/10.1097/EJA.0000000000001952 Leslie DL, Zhang Y, Bogardus ST et al (2005) Consequences of preventing delirium in hospitalized older adults on nursing home costs. J Am Geriatr Soc 53(3):405–409. https://doi.org/10.1111/j.1532-5415.2005.53156.x Moon KJ, Park H (2018) Outcomes of Patients With Delirium in Long-Term Care Facilities: A Prospective Cohort Study. J Gerontol Nurs 44(9):41–50. https://doi.org/10.3928/00989134-20180808-08 Chen H, Jiang H, Chen B et al (2020) The Incidence and predictors of postoperative delirium after brain tumor resection in adults: A Cross-Sectional Survey. World Neurosurg 140:e129–e139. https://doi.org/10.1016/j.wneu.2020.04.195 Chen Y, Fan Z, Luo Z et al (2025) Impacts of Nutlin-3a and exercise on murine double minute 2-enriched glioma treatment. Neural Regen Res 20(4):1135–1152. https://doi.org/10.4103/NRR.NRR-D-23-00875 Kim EM, Li G, Kim M (2020) Development of a risk score to predict postoperative delirium in patients with hip fracture. Anesth Analg 130(1):79–86. https://doi.org/10.1213/ANE.0000000000004386 de la Varga-Martínez O, Gómez-Pesquera E, Muñoz-Moreno MF et al (2021) Development and validation of a delirium risk prediction preoperative model for cardiac surgery patients (DELIPRECAS): An observational multicentre study. J Clin Anesth 69:110158. https://doi.org/10.1016/j.jclinane.2020.110158 Wu J, Yin Y, Jin M et al (2021) The risk factors for postoperative delirium in adult patients after hip fracture surgery: a systematic review and meta-analysis. Int J Geriatr Psychiatry 36(1):3–14. https://doi.org/10.1002/gps.5408 Nyholm L, Zetterling M, Elf K (2023) Sleep in neurointensive care patients, and patients after brain tumor surgery. PLoS ONE 18(6):e0286389. https://doi.org/10.1371/journal.pone.0286389 Ely EW, Margolin R, Francis J et al (2001) Evaluation of delirium in critically ill patients: validation of the Confusion Assessment Method for the Intensive Care Unit (CAM-ICU). Crit Care Med 29(7):1370–1379. https://doi.org/10.1097/00003246-200107000-00012 Zhang WY, Wu WL, Gu JJ et al (2015) Risk factors for postoperative delirium in patients after coronary artery bypass grafting: A prospective cohort study. J Crit Care 30(3):606–612. https://doi.org/10.1016/j.jcrc.2015.02.003 French J, Weber T, Ge B et al (2021) Postoperative delirium in patients after brain tumor surgery. World Neurosurg 155:e472–e479. https://doi.org/10.1016/j.wneu.2021.08.089 Gunther ML, Morandi A, Ely EW (2008) Pathophysiology of delirium in the intensive care unit. Crit Care Clin 24(1):45–viii. https://doi.org/10.1016/j.ccc.2007.10.002 Flanigan PM, Jahangiri A, Weinstein D et al (2018) Postoperative delirium in glioblastoma patients: risk factors and prognostic implications. Neurosurgery 83(6):1161–1172. https://doi.org/10.1093/neuros/nyx606 Zheng YB, Ruan GM, Fu JX et al (2016) Postoperative plasma 8-iso-prostaglandin F2α levels are associated with delirium and cognitive dysfunction in elderly patients after hip fracture surgery. Clin Chim Acta 455:149–153. https://doi.org/10.1016/j.cca.2016.02.007 Chen XW, Shi JW, Yang PS et al (2014) Preoperative plasma leptin levels predict delirium in elderly patients after hip fracture surgery. Peptides 57:31–35. https://doi.org/10.1016/j.peptides.2014.04.016 Nellis ME, Goel R, Feinstein S et al (2018) Association between transfusion of RBCs and subsequent development of delirium in critically ill Children. Pediatr Crit Care Med 19(10):925–929. https://doi.org/10.1097/PCC.0000000000001675 Wang CM, Huang HW, Wang YM et al (2020) Incidence and risk factors of postoperative delirium in patients admitted to the ICU after elective intracranial surgery: A prospective cohort study. Eur J Anaesthesiol 37(1):14–24. https://doi.org/10.1097/EJA.0000000000001074 Habeeb-Allah A, Alshraideh JA (2021) Delirium post-cardiac surgery: Incidence and associated factors. Nurs Crit Care 26(3):150–155. https://doi.org/10.1111/nicc.12492 Kwon DH, Kim BS, Chang H et al (2013) Exercise ameliorates cognition impairment due to restraint stress-induced oxidative insult and reduced BDNF level. Biochem Biophys Res Commun 434(2):245–251. https://doi.org/10.1016/j.bbrc.2013.02.111 Cohen DM (2016) Modeling the neurologic and cognitive effects of hyponatremia. J Am Soc Nephrol 27(3):659–661. https://doi.org/10.1681/ASN.2015060714 Li Y, Li Y, Sen G et al (2026) Deep exploration and precise identification of key risk factors for diabetic peripheral neuropathy using the random forest algorithm. Front Endocrinol 16:1740545. https://doi.org/10.3389/fendo.2025.1740545 El-Bashbishy AE, El-Bakry HM (2024) Pediatric diabetes prediction using deep learning. Sci Rep 14(1):4206. https://doi.org/10.1038/s41598-024-51438-4 Kamal MM, Khan W, Shambour QY et al (2026) A hybrid stacked autoencoder and support vector machines-based expert system for heart failure detection. Sci Rep 16(1):3886. https://doi.org/10.1038/s41598-025-34430-4 Guo YF, Li YZ, Qi Y et al (2026) Diagnosis of cognitive impairment in chronic kidney disease: A radiomics and machine learning approach with quantitative susceptibility mapping. Brain Res Bull 234:111714. https://doi.org/10.1016/j.brainresbull.2025.111714 Zhou Y, Li X, Huang W et al (2026) Predicting activities of daily living at discharge in stroke patients using rehabilitation robot training-induced functional connectivity. Top Stroke Rehabil 1–12. https://doi.org/10.1080/10749357.2026.2612712 Davoudi A, Ebadi A, Rashidi P et al (2017) Delirium Prediction using Machine Learning Models on Preoperative Electronic Health Records Data. Proc IEEE Int Symp Bioinf Bioeng 568–573. https://doi.org/10.1109/BIBE.2017.00014 Corradi JP, Thompson S, Mather JF et al (2018) Prediction of incident delirium using a random forest classifier. J Med Syst 42(12):261. https://doi.org/10.1007/s10916-018-1109-0 Li XM, Wang XQ, Xu L et al (2022) Construction and validation of a risk prediction model for ICU delirium in patients after brain tumor surgery. Chin J Mod Nurs 28(29):3991–3997. https://doi.org/10.3760/cma.j.cn115682-20220119-00316 Ma Y, Ye L, Pan J et al (2026) Lung involvement percentage in patients with COVID-19 during the Omicron wave in China: a SHAP-explained machine learning study from a single center. Front Public Health 13:1728282. https://doi.org/10.3389/fpubh.2025.1728282 Nakajima R, Kinoshita M, Okita H et al (2026) Surgical history and temporal muscle thickness as predictors of successful awake surgery in older patients with glioma. J Neurooncol 176(2):161. https://doi.org/10.1007/s11060-025-05400-7 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Revision Version 1 posted Editorial decision: Revision requested 30 Apr, 2026 Reviews received at journal 30 Apr, 2026 Reviews received at journal 31 Mar, 2026 Reviewers agreed at journal 20 Feb, 2026 Reviewers agreed at journal 19 Feb, 2026 Reviewers invited by journal 18 Feb, 2026 Editor assigned by journal 18 Feb, 2026 Submission checks completed at journal 18 Feb, 2026 First submitted to journal 16 Feb, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8893713","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":594572892,"identity":"e64a4ff2-538b-4b9f-a5a5-4b8b29def03a","order_by":0,"name":"Li Liu","email":"","orcid":"","institution":"Jiangxi Medical College, Nanchang University","correspondingAuthor":false,"prefix":"","firstName":"Li","middleName":"","lastName":"Liu","suffix":""},{"id":594572893,"identity":"119f4ffd-b2e6-4ce7-9134-866411b50740","order_by":1,"name":"Siyao Huang","email":"","orcid":"","institution":"Jiangxi Medical College, Nanchang University","correspondingAuthor":false,"prefix":"","firstName":"Siyao","middleName":"","lastName":"Huang","suffix":""},{"id":594572894,"identity":"07b3fb12-64d0-431a-958c-22e0ed5b9dfb","order_by":2,"name":"Yating Huang","email":"","orcid":"","institution":"Jiangxi Medical College, Nanchang University","correspondingAuthor":false,"prefix":"","firstName":"Yating","middleName":"","lastName":"Huang","suffix":""},{"id":594572895,"identity":"77e6748a-ec19-421c-88bd-31eedd37ea80","order_by":3,"name":"Jianmei Xu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA3klEQVRIiWNgGAWjYHCChAMgko29+eCDhIoaUrTwHEs2eHDmGCmWSeSYST5sYSas0OBGwsMDP3dsk+MDaqlIbGBj4G/vTsCrRXJGQsLB3jO3jdl4npXdSNwhwyBx5uwGvFr4JRISDvC23U5sY0/ediPxDBuDgUQufi1sQC0H/7bdrm9jSDArSGxjJqwFZMthoC0JbBwpZgxEaZHseZBwWLbttmEbMJAlEs4c4yHoF4PjOckf37bdlpdvbz748UdFjRx/ey9+LQwMPAmoXALKQYD9ABGKRsEoGAWjYEQDABSeTn9MiY/DAAAAAElFTkSuQmCC","orcid":"","institution":"The Second Affiliated Hospital, Jiangxi Medical College, Nanchang University","correspondingAuthor":true,"prefix":"","firstName":"Jianmei","middleName":"","lastName":"Xu","suffix":""}],"badges":[],"createdAt":"2026-02-16 13:53:38","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8893713/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8893713/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":103212875,"identity":"f9af700b-c6f8-4e14-839a-af99258b22b0","added_by":"auto","created_at":"2026-02-23 08:55:58","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":58399,"visible":true,"origin":"","legend":"\u003cp\u003eLASSO regression analysis figure. (A)Ten fold cross validation diagram. (B)Convergence path diagram of LASSO regression coefficients.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8893713/v1/dd87f744e11776ed26a7393f.png"},{"id":103212995,"identity":"3f8a4e33-947d-40f6-8ff9-c628f6e31a89","added_by":"auto","created_at":"2026-02-23 08:56:26","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":113551,"visible":true,"origin":"","legend":"\u003cp\u003eComparative receiver operating characteristic (ROC) curves of different POD prediction models for postoperative ICU patients following brain tumor surgery in the training and validation sets. ( A: Training Set; B: Validation Set)\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8893713/v1/b7281c041c15b97f6a0d8f2b.png"},{"id":103213014,"identity":"4728abef-ed87-4db0-ab96-83dc1bfe2c9e","added_by":"auto","created_at":"2026-02-23 08:56:33","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":127772,"visible":true,"origin":"","legend":"\u003cp\u003eValidation set figures. (A)Calibration curves of different POD prediction models for postoperative ICU patients following brain tumor surgery. (B)Decision curve analysis for different\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-8893713/v1/d5a7b725f9a887989447ccf5.png"},{"id":103212390,"identity":"e5999800-25bf-4779-803a-218d7977fb88","added_by":"auto","created_at":"2026-02-23 08:53:17","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":32525,"visible":true,"origin":"","legend":"\u003cp\u003eROC curve of the SVM prediction model for delirium in ICU patients after brain tumor surgery in the test set.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-8893713/v1/fda810f279d71fe61ad7a759.png"},{"id":103212387,"identity":"38f0e6ef-8fb2-42f6-9098-c2d15178f87d","added_by":"auto","created_at":"2026-02-23 08:53:12","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":92078,"visible":true,"origin":"","legend":"\u003cp\u003eExternal validation set figures. (A)ROC curve of the SVM prediction model for delirium in ICU patients after brain tumor surgery. (B)Calibration curve of the SVM prediction model for delirium in ICU patients after brain tumor surgery. (C)Decision curve of the SVM prediction model for delirium in ICU patients after brain tumor surgery.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-8893713/v1/898dd2950abb9bb88db229ba.png"},{"id":103212395,"identity":"631e5b8c-b18e-4ace-a6d5-4babbfca2f1e","added_by":"auto","created_at":"2026-02-23 08:53:24","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":97100,"visible":true,"origin":"","legend":"\u003cp\u003e(A)Feature importance of predictive factors for POD in ICU patients after brain tumor surgery. (B)Summary plot of predictive factors for POD in ICU patients after brain tumor surgery.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-8893713/v1/6054634c190147b720359181.png"},{"id":103212391,"identity":"291dca4e-fab3-4845-8cdd-bcae40fbe046","added_by":"auto","created_at":"2026-02-23 08:53:18","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":105610,"visible":true,"origin":"","legend":"\u003cp\u003eOnline risk calculator for POD in ICU patients after brain tumor surgery.\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-8893713/v1/1bf99eb0ac9703f15e502a2c.png"},{"id":103505492,"identity":"45a70481-d8c2-45f8-852e-062f483ee999","added_by":"auto","created_at":"2026-02-26 13:31:27","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1833746,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8893713/v1/5a3905e2-af26-4dee-bde9-ac46a9224e73.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Development and validation of a machine learning-based model for predicting delirium risk in postoperative brain tumor patients in the Intensive Care Unit","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003ePostoperative delirium (POD) is a common neuropsychiatric complication in ICU patients after brain tumor surgery, characterized by disturbances in consciousness, impaired attention, and altered cognitive function. It is classified into hyperactive, hypoactive, and mixed subtypes [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. POD not only hinders patient recovery, prolongs hospital stays, and increases medical costs but also is closely associated with long-term cognitive impairment and elevated mortality, imposing a substantial burden on patients\u0026rsquo; families and the healthcare system [\u003cspan additionalcitationids=\"CR4 CR5\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Studies have shown that 30%\u0026ndash;40% of delirium cases can be prevented through early intervention [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e], and the precise identification of high-risk patients is a prerequisite for effective prevention and control.\u003c/p\u003e \u003cp\u003eThe reported incidence of POD in ICU patients after brain tumor surgery varies widely, ranging from 28.5% to 37.8% [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e], which is related to differences in study population characteristics, assessment tools, and medical management levels. In recent years, with the in-depth application of artificial intelligence in healthcare, machine learning algorithms have become important tools for constructing disease risk prediction models due to their ability to handle complex nonlinear relationships and mine hidden patterns in large datasets [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Currently, several studies on predictive models for POD have been conducted worldwide, but most focus on other diseases such as hip fracture and cardiac surgery [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. There are few models specifically for ICU patients after brain tumor surgery, and existing studies mostly adopt a single Logistic regression algorithm without fully comparing the performance of multiple machine learning models. Additionally, the lack of in-depth analysis of model interpretability limits their clinical application.\u003c/p\u003e \u003cp\u003eFurthermore, although previous studies have made some progress in identifying risk factors for POD after brain tumor surgery, with advanced age, surgical trauma, and medication use being confirmed as associated factors [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e], the results of different studies are inconsistent, and a unified risk assessment system has not yet been established. Based on clinical big data, this study integrates multiple machine learning algorithms to construct predictive models, systematically compares their performance, and conducts interpretability analysis. Addressing the limitations of previous studies, this research focuses on the clinical positioning of delirium incidence, the limitations of external model validation, and the advantages over traditional prediction methods, aiming to provide a more precise and practical tool for the prevention and control of POD in ICU patients after brain tumor surgery.\u003c/p\u003e"},{"header":"2. Materials and Methods","content":"\u003cdiv\u003e\n\u003ch2\u003e2.1. Study population\u003c/h2\u003e\n\u003c/div\u003e\n\u003cdiv\u003e\n\u003cp\u003eConsecutive convenient sampling was used to recruit 600 patients admitted to the Neurosurgery ICU of a Grade A tertiary hospital in Nanchang, Jiangxi Province, after brain tumor surgery from July 2021 to December 2024 as the modeling cohort. The modeling cohort was randomly divided into a training set (n\u0026thinsp;=\u0026thinsp;420) and a test set (n\u0026thinsp;=\u0026thinsp;180) at a ratio of 7:3 using a random number table. An additional 160 patients from the same institution from January to August 2025 were enrolled as the external validation cohort.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv\u003e\n\u003ch2\u003e2.2. Inclusion and exclusion criteria\u003c/h2\u003e\n\u003c/div\u003e\n\u003cdiv\u003e\n\u003cp\u003eInclusion criteria: (1) Age\u0026thinsp;\u0026ge;\u0026thinsp;18 years; (2) Admitted to the Neurosurgery ICU after brain tumor surgery; (3) Complete clinical data for the modeling cohort; (4) Written informed consent obtained from patients or their family members for the external validation cohort.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv\u003e\n\u003cp\u003eExclusion criteria: (1) Preoperative cognitive impairment such as mental illness or Alzheimer\u0026rsquo;s disease; (2) Severe visual or auditory sensory impairment; (3) Persistent coma (GCS score\u0026thinsp;\u0026le;\u0026thinsp;8) during the study observation period, making it impossible to complete the assessment; (4) ICU stay\u0026thinsp;\u0026lt;\u0026thinsp;24 hours after surgery; (5) Withdrawal of treatment or death during the study; (6) Incomplete clinical data or non-compliance with the study protocol.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv\u003e\n\u003ch2\u003e2.3. Study tools and data collection\u003c/h2\u003e\n\u003c/div\u003e\n\u003cdiv\u003e\n\u003cp\u003eThe Confusion Assessment Method for the Intensive Care Unit (CAM-ICU) was used to assess delirium. This scale evaluates four dimensions: consciousness state, attention, altered level of consciousness, and thinking, with a specificity of 98.6% and sensitivity of 90.2% [14]. A two-step screening process was adopted by combining the CAM-ICU with the Richmond Agitation-Sedation Scale (RASS) [15].\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv\u003e\n\u003cp\u003eA self-designed data collection form was used to collect three categories of patient information: ① Patient-related factors: age, gender, smoking history, drinking history, history of mental illness, C-reactive protein (CRP) level, diabetes mellitus, hypertension, electrolyte imbalance, chronic lung disease, postoperative pain, sleep disturbance, APACHE II score, and postoperative GCS score; ② Tumor-related factors: bilateral brain tumor occupancy, tumor location, tumor diameter, and World Health Organization (WHO) tumor grade; ③ Medical risk factors: physical restraint, frontal craniotomy, use of benzodiazepines, use of diuretics or dehydrating agents, surgical duration, duration of mechanical ventilation, blood transfusion, and tracheotomy. Data were extracted from the hospital electronic medical record system and nursing documents, and double data entry and verification were performed by a uniformly trained research team to ensure data accuracy.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv\u003e\n\u003ch2\u003e2.4. Data preprocessing and statistical analysis\u003c/h2\u003e\n\u003c/div\u003e\n\u003cdiv\u003e\n\u003cp\u003eData were entered using Excel 16.0. Descriptive statistics and univariate analysis were performed using SPSS 26.0, and model construction and validation were conducted using RStudio 4.5.1. Continuous variables were expressed as median (interquartile range, IQR), and the Mann-Whitney U test was used for intergroup comparisons. Categorical variables were expressed as frequencies (percentages), and the chi-square test was used for intergroup comparisons.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv\u003e\n\u003cp\u003eData preprocessing: For variables with missing values\u0026thinsp;\u0026lt;\u0026thinsp;15%, continuous variables were imputed with the mean, and categorical variables were imputed using multiple imputation. Z-score standardization was performed on continuous variables. Independent risk factors were screened through univariate analysis (\u003cem\u003eP\u0026thinsp;\u0026lt;\u003c/em\u003e\u0026thinsp;0.05) and LASSO regression. Five predictive models (LR, ANN, RF, DT, SVM) were constructed. Model parameters were optimized using 10-fold cross-validation and Bootstrap resampling. Model performance was evaluated using AUC, sensitivity, specificity, accuracy, precision, F1-score, recall, Youden\u0026rsquo;s index, calibration curves, Brier score, and DCA curves. The SHAP method was used to interpret the feature importance of the optimal model. A two-sided \u003cem\u003eP\u0026thinsp;\u0026lt;\u003c/em\u003e\u0026thinsp;0.05 was considered statistically significant.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\n\u003ch2\u003e3.1. Baseline characteristics of the study population\u003c/h2\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\n\u003cp\u003eIn the modeling cohort (n\u0026thinsp;=\u0026thinsp;600), patients aged 18\u0026ndash;80 years, with 248 males (41.33%) and 352 females (58.67%). Delirium occurred in 165 patients, with an incidence of 27.5%. The training set included 420 patients and the test set included 180 patients. In the external validation cohort (n\u0026thinsp;=\u0026thinsp;160), there were 76 males (47.50%) and 84 females (52.50%), with 43 cases of delirium (incidence: 26.9%).\u003c/p\u003e\n\u003c/div\u003e\n\u003ch2\u003e3.2. Screening of risk factors for delirium in ICU patients after brain tumor surgery\u003c/h2\u003e\n\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e\n\u003cp\u003eUnivariate analysis showed significant differences between POD and non-POD patients in 13 variables (all \u003cem\u003eP\u0026thinsp;\u0026lt;\u003c/em\u003e\u0026thinsp;0.05): age, history of diabetes mellitus, history of hypertension, sleep disturbance, postoperative electrolyte imbalance, APACHE II score, postoperative GCS score, CRP level, tumor location, WHO tumor grade, physical restraint, use of diuretics or dehydrating agents, and duration of mechanical ventilation. Detailed data descriptions and univariate analysis results are presented in Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n\u003ctable id=\"Tab1\" border=\"1\"\u003e\u003ccaption\u003e\n\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n\u003cdiv class=\"CaptionContent\"\u003e\n\u003cp\u003eUnivariate analysis of risk factors for delirium in 600 ICU patients following brain tumor surgery\u003c/p\u003e\n\u003c/div\u003e\n\u003c/caption\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003ePredictor\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eNon-POD\u003c/p\u003e\n\u003cp\u003e(\u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;435 )\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003ePOD\u003c/p\u003e\n\u003cp\u003e(\u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;165 )\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eTest statistic\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eP Value\u003c/em\u003e\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eAge[years, \u003cem\u003eM\u003c/em\u003e(\u003cem\u003eP\u003c/em\u003e25, \u003cem\u003eP\u003c/em\u003e75)]\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e51.00\u003c/p\u003e\n\u003cp\u003e(42.50, 58.00)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e64.00\u003c/p\u003e\n\u003cp\u003e(54.00, 70.00)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eZ\u003c/em\u003e\u0026thinsp;=\u0026thinsp;9.214\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eSex(n, %)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026chi;2\u0026thinsp;=\u0026thinsp;0.270\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.603\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eMale\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e177(40.7)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e71(43.0)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eFemale\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e258(59.3)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e94(57.0)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eTobacco use history(n, %)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026chi;2\u0026thinsp;=\u0026thinsp;0.584\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.445\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNo\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e422(97.0)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e158(95.8)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eYes\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e13(3.0)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e7(4.2)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eAlcohol use history(n, %)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026chi;2\u0026thinsp;=\u0026thinsp;0.120\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.729\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNo\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e408(93.8)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e156(94.5)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eYes\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e27(6.2)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e9(5.5)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003ePsychiatric history(n, %)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026chi;2\u0026thinsp;=\u0026thinsp;1.539\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.215\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNo\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e432(99.3)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e162(98.2)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eYes\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e3(0.7)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e3(1.8)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eHistory of diabetes mellitus(n, %)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026chi;2\u0026thinsp;=\u0026thinsp;5.114\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.024\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNo\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e418(96.1)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e151(91.5)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eYes\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e17(3.9)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e14(8.5)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eHistory of hypertension(n, %)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026chi;2\u0026thinsp;=\u0026thinsp;10.714\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNo\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e363(83.4)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e118(71.5)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eYes\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e72(16.6)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e47(28.5)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eChronic pulmonary disease(n, %)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026chi;2\u0026thinsp;=\u0026thinsp;0.395\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.530\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNo\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e432(99.3)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e163(98.8)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eYes\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e3(0.7)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2(1.2)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003ePostoperative pain(n, %)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026chi;2\u0026thinsp;=\u0026thinsp;1.696\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.193\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNo\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e8(1.8)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6(3.6)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eYes\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e427(98.2)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e159(96.4)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eSleep disturbances(n, %)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026chi;2\u0026thinsp;=\u0026thinsp;10.100\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNo\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e429(98.6)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e155(93.9)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eYes\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6(1.4)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e10(6.1)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eElectrolyte disturbances(n, %)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026chi;2\u0026thinsp;=\u0026thinsp;4.477\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.034\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNo\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e189\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e56\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eYes\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e246\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e109\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eAPACHE II score\u003c/p\u003e\n\u003cp\u003e[\u003cem\u003eM\u003c/em\u003e(\u003cem\u003eP\u003c/em\u003e25, \u003cem\u003eP\u003c/em\u003e75)]\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e7.00(5.00, 9.00)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e10.00\u003c/p\u003e\n\u003cp\u003e(8.00, 13.00)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eZ\u003c/em\u003e\u0026thinsp;=\u0026thinsp;8.298\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eGCS score[\u003cem\u003eM\u003c/em\u003e(\u003cem\u003eP\u003c/em\u003e25, \u003cem\u003eP\u003c/em\u003e75)]\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e15.00\u003c/p\u003e\n\u003cp\u003e(14.00, 15.00)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e14.00\u003c/p\u003e\n\u003cp\u003e(14.00, 15.00)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eZ\u003c/em\u003e= -4.457\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eC-reactive protein concentration\u003c/p\u003e\n\u003cp\u003e[mg/L, \u003cem\u003eM\u003c/em\u003e(\u003cem\u003eP\u003c/em\u003e25, \u003cem\u003eP\u003c/em\u003e75)]\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e13.10\u003c/p\u003e\n\u003cp\u003e(5.82, 32.52)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e21.60\u003c/p\u003e\n\u003cp\u003e(7.97, 45.83)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eZ\u003c/em\u003e\u0026thinsp;=\u0026thinsp;2.945\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.003\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eBilateral brain tumor occupancy(n, %)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026chi;2\u0026thinsp;=\u0026thinsp;2.517\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.113\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNo\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e421(96.8)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e155(93.9)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eYes\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e14(3.2)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e10(6.1)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eTumor site(n, %)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026chi;2\u0026thinsp;=\u0026thinsp;22.475\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.002\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eFrontal lobe\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e60(13.8)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e26(15.8)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eTemporal lobe\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e34(7.8)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e24(14.5)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eParietal lobe\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e17(3.9)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e9(5.5)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eOccipital lobe\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e9(2.1)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e10(6.1)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003ePituitarium\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e38(8.7)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e19(11.5)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eCerebellopontine angle\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e136(31.3)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e29(17.6)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eCerebellum\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e16(3.7)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4(2.4)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eOthers\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e125(28.7)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e44(26.7)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eTumor diameter\u003c/p\u003e\n\u003cp\u003e[cm, \u003cem\u003eM\u003c/em\u003e(\u003cem\u003eP\u003c/em\u003e25, \u003cem\u003eP\u003c/em\u003e75)]\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e3.50(2.50, 5.25)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e3.60(2.50, 6.00)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eZ\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.888\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.375\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eWHO tumor grade(n, %)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026chi;2\u0026thinsp;=\u0026thinsp;8.645\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.034\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eGrade Ⅰ\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e312(71.7)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e103(62.4)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eGrade Ⅱ\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e55(12.6)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e28(17.0)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eGrade Ⅲ\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e24(5.5)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6(3.6)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eGrade IV\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e44(10.1)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e28(17.0)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003ePhysical restraint(n, %)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026chi;2\u0026thinsp;=\u0026thinsp;17.787\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNo\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e169(38.9)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e34(20.6)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eYes\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e266(61.1)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e131(79.4)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eTransfrontal craniotomy approach(n, %)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026chi;2\u0026thinsp;=\u0026thinsp;2.671\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.102\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNo\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e407(93.6)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e160(97.0)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eYes\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e28(6.4)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e5(3.0)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eAdministration of benzodiazepines(n, %)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026chi;2\u0026thinsp;=\u0026thinsp;2.582\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.108\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNo\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e430(98.9)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e160(97.0)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eYes\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e5(1.1)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e5(3.0)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eAdministration of diuretic/dehydrating agents(n, %)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026chi;2\u0026thinsp;=\u0026thinsp;108.164\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNo\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e333(65.1)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e51(7.9)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eYes\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e102(34.9)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e114(92.1)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eDuration of mechanical ventilation[min, M(\u003cem\u003eP\u003c/em\u003e25, \u003cem\u003eP\u003c/em\u003e75)]\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e555.00\u003c/p\u003e\n\u003cp\u003e(450.00, 700.00)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e655.00\u003c/p\u003e\n\u003cp\u003e(500.00, 1070.00)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eZ\u003c/em\u003e\u0026thinsp;=\u0026thinsp;5.494\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eDuration of operation[min, M(\u003cem\u003eP\u003c/em\u003e25, \u003cem\u003eP\u003c/em\u003e75)]\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e345.00\u003c/p\u003e\n\u003cp\u003e(240.00, 450.00)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e345.00\u003c/p\u003e\n\u003cp\u003e(250.00, 470.00)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eZ\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1.357\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.175\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eBlood transfusion(n, %)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026chi;2\u0026thinsp;=\u0026thinsp;1.032\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.310\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNo\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e265(60.9)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e93(56.4)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eYes\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e170(39.1)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e72(43.6)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eTracheostomy(n, %)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026chi;2\u0026thinsp;=\u0026thinsp;0.209\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.648\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNo\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e427(98.2)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e161(97.6)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eYes\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e8(1.8)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4(2.4)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003ctfoot\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"5\"\u003eNote: APACHE II Acute physiology and chronic health evaluation Ⅱ;GCS Glasgow Coma Scale score\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"5\"\u003eAfter data standardization, the 13 independent variables were included in LASSO regression analysis using R software. Finally, five risk factors were identified as independent predictors of POD in ICU patients after brain tumor surgery: age, APACHE II score, postoperative GCS score, use of diuretics or dehydrating agents, and duration of mechanical ventilation. The optimal \u0026lambda; value was determined using 10-fold cross-validation (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003eA). At \u0026lambda;.1se\u0026thinsp;=\u0026thinsp;0.039, the fluctuation of the model standard deviation within the interval defined by the two dashed lines was small (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003eA), indicating a relatively simple model structure in this region. Figure\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003eB depicts the path of each variable coefficient with the change of the regularization parameter \u0026lambda;. It can be seen that as the \u0026lambda; value increases, the compression effect of the model on variables gradually enhances, thereby achieving effective screening of candidate variables and ultimately obtaining a simplified model with the minimum number of predictive variables.\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tfoot\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec19\" class=\"Section2\"\u003e\n\u003ch2\u003e3.3. Comparison of predictive model performance\u003c/h2\u003e\n\u003cdiv id=\"Sec20\" class=\"Section3\"\u003e\n\u003ch2\u003e3.3.1. Internal validation of predictive models and selection of the optimal model\u003c/h2\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec21\" class=\"Section3\"\u003e\n\u003cp\u003eFive POD predictive models were constructed using LR, ANN, RF, DT, and SVM algorithms, respectively, based on the five variables identified by LASSO regression. Delirium occurrence was defined as the dependent variable (1\u0026thinsp;=\u0026thinsp;delirium, 0\u0026thinsp;=\u0026thinsp;no delirium), and the independent variables were assigned as follows: age (actual value), APACHE II score (actual value), postoperative GCS score (actual value), use of diuretics or dehydrating agents (1\u0026thinsp;=\u0026thinsp;yes, 0\u0026thinsp;=\u0026thinsp;no), and duration of mechanical ventilation (actual value). ROC curves were plotted with sensitivity as the ordinate and 1 - specificity as the abscissa, and the area under the ROC curve (AUROC) was used to evaluate the discrimination ability of each model. The ROC curves of the five predictive models are shown in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec22\" class=\"Section3\"\u003e\n\u003ch2\u003e\u0026nbsp;\u003c/h2\u003e\n\u003cp\u003eThe detailed performance of each predictive model in the training and validation sets is presented in Tables2 and 3. In the training set, the RF model exhibited excellent discrimination ability, with an AUC of 0.944 and an ideal steep upward trend of the curve. In addition, the RF model achieved a specificity of 0.947, accuracy of 0.868, precision of 0.829, and F1-score of 0.737, showing outstanding classification performance.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n\u003cdiv class=\"colspec\" align=\"char\"\u003e\u0026nbsp;\u003c/div\u003e\n\u003ctable id=\"Tab2\" border=\"1\"\u003e\u003ccaption\u003e\n\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n\u003cdiv class=\"CaptionContent\"\u003e\n\u003cp\u003ePerformance comparison of different prediction models for POD development in postoperative ICU patients following brain tumor surgery within the training set\u003c/p\u003e\n\u003c/div\u003e\n\u003c/caption\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003ePrediction Model\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eAUC(95%\u003cem\u003eCI\u003c/em\u003e)\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eSensitivity\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eSpecificity\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eAccuracy\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003ePrecision\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eRecall\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eF1-Score\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eYouden's Index\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eLR\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.864(0.851\u0026thinsp;~\u0026thinsp;0.878)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.577\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.921\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.825\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.738\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.577\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.648\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.498\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eANN\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.865(0.851\u0026thinsp;~\u0026thinsp;0.878)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.585\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.920\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.827\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.740\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.585\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.653\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.505\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eRF\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.944(0.937\u0026thinsp;~\u0026thinsp;0.951)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.665\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.947\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.868\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.829\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.665\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.737\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.612\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eDT\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.876(0.863\u0026thinsp;~\u0026thinsp;0.889)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.683\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.930\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.861\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.793\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.683\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.731\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.613\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eSVM\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.869(0.855\u0026thinsp;~\u0026thinsp;0.882)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.631\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.905\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.828\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.719\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.631\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.672\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.535\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003ctfoot\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"9\"\u003eNote: AUC Area Under the Curve;LR Logistic Regression༛ANN Artificial Neural Network༛RF Random Forest༛DT Decision Tree༛SVM Support Vector Machine\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tfoot\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n\u003cdiv class=\"colspec\" align=\"char\"\u003e\u0026nbsp;\u003c/div\u003e\n\u003cdiv class=\"colspec\" align=\"char\"\u003e\u0026nbsp;\u003c/div\u003e\n\u003ctable id=\"Tab3\" border=\"1\"\u003e\u003ccaption\u003e\n\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\n\u003cdiv class=\"CaptionContent\"\u003e\n\u003cp\u003ePerformance comparison of different prediction models for POD development in postoperative ICU patients following brain tumor surgery within the validation set\u003c/p\u003e\n\u003c/div\u003e\n\u003c/caption\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003ePrediction Model\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eAUC(95%\u003cem\u003eCI\u003c/em\u003e)\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eSensitivity\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eSpecificity\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eAccuracy\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003ePrecision\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eRecall\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eF1-Score\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eYouden's Index\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eLR\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.854(0.809\u0026thinsp;~\u0026thinsp;0.895)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.575\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.917\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.821\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.740\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.575\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.642\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.492\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eANN\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.852(0.804\u0026thinsp;~\u0026thinsp;0.895)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.574\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.917\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.821\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.740\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.574\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.642\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.491\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eRF\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.859(0.819\u0026thinsp;~\u0026thinsp;0.897)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.583\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.904\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.814\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.732\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.583\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.630\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.487\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eDT\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.799(0.744\u0026thinsp;~\u0026thinsp;0.848)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.524\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.894\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.791\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.685\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.524\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.575\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.418\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eSVM\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.857(0.814\u0026thinsp;~\u0026thinsp;0.898)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.651\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.907\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.836\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.744\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.651\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.691\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.558\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eAbbreviations as per Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e\n\u003cp\u003eIn the validation set, there were differences in the generalizability of the models: the AUC of SVM (0.857) was slightly lower than that of RF (0.859), indicating better generalizability of the SVM model. The SVM model had the highest sensitivity (0.651), accuracy (0.836), precision (0.744), recall (0.651), F1-score (0.691), and Youden\u0026rsquo;s index (0.558) among all models, while its specificity (0.907) was slightly lower than that of LR and ANN (both 0.917). The performance of the RF model decreased most significantly: the AUC dropped from 0.944 to 0.859, and other indicators also decreased significantly, indicating severe overfitting, with the validation set performance far lower than the training set. All indicators of the DT model decreased, among which sensitivity, accuracy, precision, recall, F1-score, and Youden\u0026rsquo;s index decreased the most, indicating a serious overfitting problem. Although the specificity of LR and ANN (both 0.917) was higher than that of SVM, their other performance indicators were lower than those of SVM. Comprehensive performance analysis of the validation set showed that the SVM model had the optimal efficacy in predicting POD after brain tumor surgery: its sensitivity, accuracy, precision, recall, F1-score, and Youden\u0026rsquo;s index all ranked first, and the performance fluctuation from the training set to the validation set was small (AUC only decreased by 0.012), showing the best comprehensive performance.\u003c/p\u003e\n\u003cp\u003eCalibration is a key criterion for evaluating the accuracy of a model in predicting the probability of a specific outcome event in individuals. Calibration curves can intuitively show the consistency between predicted and actual risks. Therefore, calibration curves were used to evaluate the calibration of each model in this study. From the comprehensive performance of calibration curves and Brier scores, LR had the lowest Brier score (0.129), and its calibration curve was the closest to the ideal dashed line throughout the entire range with no significant fluctuations, indicating the optimal calibration performance. The SVM model had a Brier score of 0.139, and its curve was generally close to the ideal dashed line, with only a slight deviation in the medium-to-high probability range, ranking second in performance. The RF (Brier\u0026thinsp;=\u0026thinsp;0.140) and ANN (Brier\u0026thinsp;=\u0026thinsp;0.141) models had good curve fitting, with overall calibration performance at a good to moderate level. The DT model had the highest Brier score (0.157), and its calibration curve fluctuated sharply, especially a significant decline in the high predicted probability range (\u0026gt;\u0026thinsp;0.8), indicating a serious overestimation problem and the worst calibration performance (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003eA).\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eDCA is a method to evaluate the feasibility of clinical decisions by considering the possible range of patient risks and benefits. DCA curves were used in this study to assess the actual clinical utility of the predictive models. The DCA curves of each delirium risk prediction model for ICU patients after brain tumor surgery are shown in Fig.3B. The results showed that the ANN model had the best DCA curve performance, with an average net benefit of 0.197, followed by LR (average net benefit: 0.194) and SVM (average net benefit: 0.188).\u003c/p\u003e\n\u003cp\u003eBased on the superior performance of the SVM predictive model in various aspects compared with the other four models, the SVM model was finally selected as the optimal model in this study. In the test set, the SVM predictive model achieved an AUC of 0.847, sensitivity of 0.896, specificity of 0.712, accuracy of 0.761, precision of 0.531, recall of 0.896, F1-score of 0.667, and Youden\u0026rsquo;s index of 0.608. The ROC curve is shown in Fig.4.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec23\" class=\"Section3\"\u003e\n\u003ch2\u003e3.3.2. External validation of the optimal delirium predictive model\u003c/h2\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec24\" class=\"Section3\"\u003e\n\u003cp\u003eData of 160 patients admitted to the Neurosurgery ICU of the same Grade A tertiary hospital in Nanchang, Jiangxi Province, after brain tumor surgery from January to August 2025 were collected for external validation of the predictive model. The validation cohort included 76 male patients (47.50%) and 84 female patients (52.50%), among whom 43 cases (26.90%) developed POD. In external validation, the discrimination, calibration, and clinical utility of the predictive model are shown in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e. The SVM predictive model achieved an accuracy of 70.00%, specificity of 67.52%, and sensitivity of 76.74% in the external validation set.\u0026nbsp;\u003c/p\u003e\n\u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec26\" class=\"Section2\"\u003e\n\u003ch2\u003e3.4. Interpretation and application of the delirium risk predictive model\u003c/h2\u003e\n\u003cdiv id=\"Sec27\" class=\"Section3\"\u003e\n\u003ch2\u003e3.4.1. Feature importance\u003c/h2\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec28\" class=\"Section3\"\u003e\n\u003cp\u003eThe SHAP method was used to measure the importance of each feature. Figure\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003eA shows the importance of the five predictive factors for delirium in ICU patients after brain tumor surgery, among which the use of diuretics or dehydrating agents was the most important feature. Based on the SHAP package, Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003eB shows the degree of influence of each feature on delirium in ICU patients after brain tumor surgery.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec30\" class=\"Section3\"\u003e\n\u003ch2\u003e3.4.2. Introduction to model application\u003c/h2\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec31\" class=\"Section3\"\u003e\n\u003cp\u003eBased on the core predictive indicators and the model, a user-friendly POD risk calculator was developed, which can be accessed at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://pod66.shinyapps.io/shiny-svm1/\u003c/span\u003e\u003c/span\u003e. The optimal cutoff value of the SVM model calculated based on Youden\u0026rsquo;s index was 0.6006. That is, when medical staff input the patient\u0026rsquo;s age, APACHE II score, postoperative GCS score, use of diuretics or dehydrating agents, and duration of mechanical ventilation into the calculator, a result\u0026thinsp;\u0026gt;\u0026thinsp;60.06% indicates that the patient has a risk of developing POD, and early intervention is required. The page view of the online risk calculator is shown in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/div\u003e\n\u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cdiv id=\"Sec33\" class=\"Section2\"\u003e\n\u003ch2\u003e4.1. Incidence of delirium in ICU patients after brain tumor surgery\u003c/h2\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec34\" class=\"Section2\"\u003eThe incidence of delirium in ICU patients after brain tumor surgery is generally high. Previous studies have clearly shown that it can range from 28.5% to 37.8% [\u003cspan class=\"CitationRef\"\u003e8\u003c/span\u003e], highlighting the high risk of this complication in neurosurgical critically ill patients. The results of this study showed that the incidence of delirium in the modeling cohort was 27.5%, and that in the external validation cohort was 26.9%. Although this result is slightly lower than the lower limit of the above range, it is still in the clinically high-risk range, and is highly close to the 28.5% reported by Chen et al. [\u003cspan class=\"CitationRef\"\u003e8\u003c/span\u003e], only lower than the 37.8% reported by French et al. [\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e].\u003c/div\u003e\n\u003cp\u003eThe discrepancies in incidence rates can be attributed to four key factors: ① Stringency of inclusion and exclusion criteria: Patients with persistent coma (GCS score\u0026thinsp;\u0026le;\u0026thinsp;8) were excluded because their severely impaired consciousness made it impossible to complete effective delirium assessment, thus ensuring the reliability of endpoint event determination. Meanwhile, to avoid confusion between pre-existing chronic cognitive impairment and acute postoperative delirium, all patients with pre-existing cognitive impairment were also excluded to ensure that the model predicted newly developed acute brain dysfunction. In addition, patients with an ICU stay of less than 24 hours were excluded to ensure that all enrolled subjects had a unified and sufficiently long observation window, thereby completely capturing delirium events and avoiding missing endpoint data due to early transfer out. ② Accuracy of the assessment system: This study adopted a dual assessment model combining the CAM-ICU scale and consultation with neurologists, with a specificity of 98.6% and a sensitivity of 90.2% [\u003cspan class=\"CitationRef\"\u003e14\u003c/span\u003e], which effectively reduced missed diagnoses and misdiagnoses. In contrast, some studies used the Intensive Care Delirium Screening Checklist (ICDSC) or subjective assessment by a single medical staff member, which may lead to high or low incidence rates due to screening errors. ③ Standardization of medical management: The hospital in this study is a Grade A tertiary hospital, where refined sedation and analgesia management, sleep protection, infection prevention and control, and other interventions have been routinely carried out during the perioperative period. This has reduced the impact of controllable risk factors to a certain extent, making the incidence rate closer to the inherent risk level of the disease. ④ Heterogeneity of population characteristics: There are differences in sample size, patient age structure (the higher the proportion of the elderly, the higher the incidence rate), tumor type (the higher the degree of malignancy, the higher the risk), and surgical complexity among different studies. For example, the proportion of patients with WHO grade IV tumors in this study was 12.0% (72 out of 600 cases), while some studies focusing on malignant brain tumors are more likely to approach the high value of 37.8% due to the concentration of high-risk cases.\u003c/p\u003e\n\u003cdiv id=\"Sec36\" class=\"Section2\"\u003eAlthough the incidence of 27.5% in this study is slightly lower than the high value range reported in previous studies, it is still at a clinically high-risk level, fully indicating that the prevention and control of delirium in ICU patients after brain tumor surgery still faces severe challenges. This incidence not only objectively reflects the inherent risks of the disease such as surgical trauma, changes in the intracranial environment, and damage to the blood-brain barrier but also reflects the effectiveness of standardized medical management in intervening in controllable factors (such as infection and inappropriate sedation). The results have strong representativeness for Grade A tertiary hospitals and institutions with similar medical levels. It should be noted that in primary hospitals with relatively limited medical resources and inadequate intervention measures, the incidence of delirium may be closer to the high value range of 28.5%\u0026ndash;37.8%. Therefore, when popularizing the predictive model of this study, it is necessary to calibrate the model threshold in a targeted manner according to the baseline level of delirium occurrence in local medical institutions to ensure its universality and clinical practicality.\u003c/div\u003e\n\u003cdiv id=\"Sec37\" class=\"Section2\"\u003e\n\u003ch2\u003e4.2. Clinical implications of delirium risk factors\u003c/h2\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec38\" class=\"Section2\"\u003e\n\u003cp\u003eThis study identified age, APACHE II score, postoperative GCS score, use of diuretics or dehydrating agents, and duration of mechanical ventilation as independent risk factors for delirium in ICU patients after brain tumor surgery, which is partially consistent with previous research results [\u003cspan class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e12\u003c/span\u003e\u0026ndash;\u003cspan class=\"CitationRef\"\u003e13\u003c/span\u003e], but also highlights the characteristics of neurosurgery.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec39\" class=\"Section2\"\u003e\n\u003cp\u003eAge is a well-recognized risk factor for delirium. In this study, the median age of the delirium group was significantly higher than that of the non-delirium group. Physiological changes associated with aging, such as reduced neural reserve, impaired integrity of the blood-brain barrier, and inflammaging, reduce the tolerance of elderly patients to surgical stress and anesthesia. Cholinergic system dysfunction further increases the susceptibility to delirium [\u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e\u0026ndash;\u003cspan class=\"CitationRef\"\u003e20\u003c/span\u003e]. The APACHE II score reflects the severity of the patient\u0026rsquo;s condition; a high score indicates multiple organ dysfunction, which indirectly impairs cerebral perfusion through hypotension, hypoxia, etc., leading to neurotransmitter system disorders and inducing delirium [\u003cspan class=\"CitationRef\"\u003e21\u003c/span\u003e].\u003c/p\u003e\n\u003cp\u003eA decreased postoperative GCS score indicates brain function damage, which may result from edema in the surgical area, increased intracranial pressure, or occult hemorrhage. This damage disrupts cortico-subcortical network connections, impairs attention and executive function, and simultaneously increases the risk of pulmonary infection and hypoxemia, further promoting the occurrence of delirium [\u003cspan class=\"CitationRef\"\u003e22\u003c/span\u003e]. Prolonged duration of mechanical ventilation increases the risk of delirium through multiple pathways such as sleep deprivation, use of sedative drugs, and respiratory complications [\u003cspan class=\"CitationRef\"\u003e23\u003c/span\u003e\u0026ndash;\u003cspan class=\"CitationRef\"\u003e24\u003c/span\u003e]. In this study, the duration of mechanical ventilation in the delirium group (655 min) was significantly longer than that in the non-delirium group (555 min), which is consistent with the research results of Habeeb-Allah et al. [\u003cspan class=\"CitationRef\"\u003e23\u003c/span\u003e].\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec41\" class=\"Section2\"\u003e\n\u003cp\u003eNotably, the use of diuretics or dehydrating agents was the most important risk factor in this study (SHAP value: 0.1625), with a usage rate of 92.1% in the delirium group, much higher than 34.9% in the non-delirium group. Patients after brain tumor surgery often use dehydrating agents for increased intracranial pressure, but these drugs can cause electrolyte imbalances (especially hyponatremia), leading to cerebral edema, abnormal nerve impulse conduction, and disruption of neurotransmitter balance. Given that the blood-brain barrier is damaged after brain tumor surgery, patients are more sensitive to electrolyte imbalances [\u003cspan class=\"CitationRef\"\u003e25\u003c/span\u003e]. This mechanism is particularly prominent in neurosurgical patients, providing a clear target for clinical precise intervention.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec42\" class=\"Section2\"\u003e\n\u003ch2\u003e4.3. Performance and advantages of machine learning models\u003c/h2\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec43\" class=\"Section2\"\u003e\n\u003cp\u003eThis study compared five machine learning algorithms and found that the SVM model had the best comprehensive performance, with good discrimination, calibration, and clinical utility, as well as strong stability and generalizability. The AUC only decreased by 0.012 from the training set to the validation set, and the accuracy in external validation reached 70.00%, which is significantly better than the LR, ANN, RF, and DT models. The RF model performed excellently in the training set (AUC\u0026thinsp;=\u0026thinsp;0.944) but its performance decreased significantly in the validation set, suggesting overfitting, which is related to the characteristics of the RF algorithm that is susceptible to noise data and has high requirements for sample size [\u003cspan class=\"CitationRef\"\u003e26\u003c/span\u003e].\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec44\" class=\"Section2\"\u003e\n\u003cp\u003eCompared with previous single Logistic regression models, the SVM model has the following advantages: ① It can effectively handle complex nonlinear relationships and interactions between variables, which is more in line with the complexity of clinical data; ② It improves the robustness of the model by maximizing the classification margin and reduces the risk of overfitting; ③ It has stronger ability to process high-dimensional data and performs more stably in multi-factor prediction [\u003cspan class=\"CitationRef\"\u003e27\u003c/span\u003e\u0026ndash;\u003cspan class=\"CitationRef\"\u003e29\u003c/span\u003e]. In addition, this study used the SHAP method to interpret the SVM model, clarifying the importance and direction of action of each risk factor, solving the black-box problem of traditional machine learning models, and enhancing the trust and acceptance of the model by clinical medical staff [\u003cspan class=\"CitationRef\"\u003e30\u003c/span\u003e].\u003c/p\u003e\n\u003cp\u003eCompared with other postoperative delirium predictive models, the model in this study has the following characteristics: ① It focuses on the high-risk population of ICU patients after brain tumor surgery, and the screening of risk factors is more specialized, especially highlighting the use of diuretics or dehydrating agents as a characteristic risk factor in neurosurgery; ② It integrates five machine learning algorithms and screens the optimal model after systematically comparing their performance, avoiding the limitations of a single algorithm; ③ An online risk calculator was developed, which is easy to operate and can generate personalized risk assessment results in real time, meeting the needs of rapid clinical decision-making; ④ Internal 10-fold cross-validation and external independent cohort validation were performed to ensure the reliability and generalizability of the model.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec46\" class=\"Section2\"\u003e\n\u003ch2\u003e4.4. Connection and comparison with previous studies\u003c/h2\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec47\" class=\"Section2\"\u003e\n\u003cp\u003eThe results of this study form a good connection with previous studies on postoperative delirium. In terms of risk factors, factors such as advanced age, APACHE II score, and duration of mechanical ventilation have been confirmed to be associated with postoperative delirium in multiple studies [\u003cspan class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e12\u003c/span\u003e\u0026ndash;\u003cspan class=\"CitationRef\"\u003e13\u003c/span\u003e]. This study further verified the importance of these factors in the population of ICU patients after brain tumor surgery, and at the same time added the use of diuretics or dehydrating agents as a neurosurgery-specific risk factor, enriching the risk factor spectrum of postoperative delirium.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec48\" class=\"Section2\"\u003e\n\u003cp\u003eIn terms of predictive models, Davoudi et al [\u003cspan class=\"CitationRef\"\u003e31\u003c/span\u003e] constructed a postoperative delirium predictive model based on seven machine learning algorithms, with the generalized additive model achieving an AUC of 0.860; Corradi et al. [\u003cspan class=\"CitationRef\"\u003e32\u003c/span\u003e] also achieved good predictive results using a model constructed by the random forest algorithm. The AUC of the SVM model in this study is 0.857, which is at the same level as the above studies. However, this study focuses on the specific population of ICU patients after brain tumor surgery, with more targeted risk factors and stronger model interpretability. Compared with traditional non-machine learning studies, the predictive efficacy of the model in this study is better. For example, Li et al. [\u003cspan class=\"CitationRef\"\u003e33\u003c/span\u003e] constructed a predictive model for delirium in ICU patients after brain tumor surgery using Logistic regression, with an AUC of 0.78, which is lower than the 0.857 of the SVM model in this study. This benefit from the advantage of machine learning algorithms in processing complex data.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec49\" class=\"Section2\"\u003e\n\u003cp\u003eHowever, it should be objectively recognized that machine learning algorithms are not absolutely superior to traditional methods. The advantages of Logistic regression models are simple principles, easy interpretation of results, and low requirements for data quality, which still have application value in scenarios with small sample sizes and clear variable relationships [\u003cspan class=\"CitationRef\"\u003e34\u003c/span\u003e\u0026ndash;\u003cspan class=\"CitationRef\"\u003e35\u003c/span\u003e]. The advantage of the model in this study is that it can more accurately mine data patterns and improve prediction performance in the case of large samples, multiple factors, and complex relationships. Therefore, in clinical applications, appropriate prediction methods should be selected according to the actual situation, rather than blindly pursuing complex algorithms.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"5. Limitation","content":"\u003cp\u003eFirst, this study adopted a single-center retrospective design. Although the sample size meets statistical requirements, there is still selection bias, and the characteristics of the patient population are relatively single, which may affect the generaliz\u003c/p\u003e\n"},{"header":"6. Conclusion","content":"\n\u003cp\u003eThe incidence of delirium in ICU patients after brain tumor surgery is 27.5%. Age, APACHE II score, postoperative GCS score, use of diuretics or dehydrating agents, and duration of mechanical ventilation are independent risk factors. The delirium risk pre\u003c/p\u003e\n"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eConflicts of interest\u003c/strong\u003e On behalf of all authors, the corresponding author states that there is no conflict of interest. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthical standard statement\u003c/strong\u003e This observational study was conducted in accordance with the ethical principles of the Declaration of Helsinki. The study protocol was reviewed and approved by the Biomedical Research Ethics Committee of the Second Affiliated Hospital of Nanchang University.\u003c/p\u003e\n\u003cp\u003eAs this was a low-risk observational study, it underwent an expedited review process according to the committee's guidelines. Therefore, a specific approval number was not issued; instead, the approval is validated by the official notification dated 24/02/2025. Informed consent was obtained from the patient, agreeing to the use of her medical information for article publication.\u003c/p\u003e\n\u003cp\u003eAll authors declare that the submitted work has not been published before (neither in English nor in any other language) and that the work is not under consideration for publication elsewhere.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u0026nbsp;\u003c/strong\u003eWe extend our sincere gratitude to the research colleagues, clinicians, nursing staff, and administrative support team of the Second Affiliated Hospital of Nanchang University for their essential contributions, and to all participants and their caregivers for generously donating their time—their invaluable cooperation formed the cornerstone of this study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u0026nbsp;\u003c/strong\u003eConception and design: Li Liu. Acquisition of data: Li Liu, Siyao Huang and Yating Huang. Analysis and interpretation of data: Li Liu, Siyao Huang. Drafting the article: Li Liu. Critically revising the article: Li Liu. Reviewed submitted version of manuscript: Yating Huang and Jianmei Xu. Approved the final version of the manuscript on behalf of all authors: Jianmei Xu. Statistical analysis: Li Liu, Siyao Huang and Yating Huang. Administrative/technical/material support: Jianmei Xu. Study supervision: Jianmei Xu.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe data supporting the conclusions of this article will be made available by the corresponding author, without undue reservation, to any qualified researcher upon reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eEschweiler GW, Czornik M, Herrmann ML et al (2021) Presurgical screening improves risk prediction for delirium in elective surgery of older patients: The PAWEL RISK Study. Front Aging Neurosci 13:679933. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fnagi.2021.679933\u003c/span\u003e\u003cspan address=\"10.3389/fnagi.2021.679933\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKatayama ES, Stecko H, Woldesenbet S et al (2024) The role of delirium on short - and long-term postoperative outcomes following major gastrointestinal surgery for cancer. Ann Surg Oncol 31(8):5232\u0026ndash;5239. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1245/s10434-024-15358-x\u003c/span\u003e\u003cspan address=\"10.1245/s10434-024-15358-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang H, Li H, Zhang X et al (2021) Association of postoperative delirium with cognitive outcomes: A meta-analysis. J Clin Anesth 75:110496. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jclinane.2021.110496\u003c/span\u003e\u003cspan address=\"10.1016/j.jclinane.2021.110496\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXiang PY, Boyle L, Short TG et al (2023) Incidence of postoperative delirium in surgical patients: An observational retrospective cohort study. Anaesth Intensive Care 51(4):260\u0026ndash;267. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/0310057X231156459\u003c/span\u003e\u003cspan address=\"10.1177/0310057X231156459\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOu-Yang CL, Ma LB, Wu XD et al (2024) Association of sleep quality on the night of operative day with postoperative delirium in elderly patients: A prospective cohort study. Eur J Anaesthesiol 41(3):226\u0026ndash;233. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1097/EJA.0000000000001952\u003c/span\u003e\u003cspan address=\"10.1097/EJA.0000000000001952\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLeslie DL, Zhang Y, Bogardus ST et al (2005) Consequences of preventing delirium in hospitalized older adults on nursing home costs. J Am Geriatr Soc 53(3):405\u0026ndash;409. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/j.1532-5415.2005.53156.x\u003c/span\u003e\u003cspan address=\"10.1111/j.1532-5415.2005.53156.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoon KJ, Park H (2018) Outcomes of Patients With Delirium in Long-Term Care Facilities: A Prospective Cohort Study. J Gerontol Nurs 44(9):41\u0026ndash;50. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3928/00989134-20180808-08\u003c/span\u003e\u003cspan address=\"10.3928/00989134-20180808-08\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen H, Jiang H, Chen B et al (2020) The Incidence and predictors of postoperative delirium after brain tumor resection in adults: A Cross-Sectional Survey. World Neurosurg 140:e129\u0026ndash;e139. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.wneu.2020.04.195\u003c/span\u003e\u003cspan address=\"10.1016/j.wneu.2020.04.195\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen Y, Fan Z, Luo Z et al (2025) Impacts of Nutlin-3a and exercise on murine double minute 2-enriched glioma treatment. Neural Regen Res 20(4):1135\u0026ndash;1152. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.4103/NRR.NRR-D-23-00875\u003c/span\u003e\u003cspan address=\"10.4103/NRR.NRR-D-23-00875\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim EM, Li G, Kim M (2020) Development of a risk score to predict postoperative delirium in patients with hip fracture. Anesth Analg 130(1):79\u0026ndash;86. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1213/ANE.0000000000004386\u003c/span\u003e\u003cspan address=\"10.1213/ANE.0000000000004386\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ede la Varga-Mart\u0026iacute;nez O, G\u0026oacute;mez-Pesquera E, Mu\u0026ntilde;oz-Moreno MF et al (2021) Development and validation of a delirium risk prediction preoperative model for cardiac surgery patients (DELIPRECAS): An observational multicentre study. J Clin Anesth 69:110158. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jclinane.2020.110158\u003c/span\u003e\u003cspan address=\"10.1016/j.jclinane.2020.110158\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu J, Yin Y, Jin M et al (2021) The risk factors for postoperative delirium in adult patients after hip fracture surgery: a systematic review and meta-analysis. Int J Geriatr Psychiatry 36(1):3\u0026ndash;14. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/gps.5408\u003c/span\u003e\u003cspan address=\"10.1002/gps.5408\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNyholm L, Zetterling M, Elf K (2023) Sleep in neurointensive care patients, and patients after brain tumor surgery. PLoS ONE 18(6):e0286389. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1371/journal.pone.0286389\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0286389\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEly EW, Margolin R, Francis J et al (2001) Evaluation of delirium in critically ill patients: validation of the Confusion Assessment Method for the Intensive Care Unit (CAM-ICU). Crit Care Med 29(7):1370\u0026ndash;1379. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1097/00003246-200107000-00012\u003c/span\u003e\u003cspan address=\"10.1097/00003246-200107000-00012\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang WY, Wu WL, Gu JJ et al (2015) Risk factors for postoperative delirium in patients after coronary artery bypass grafting: A prospective cohort study. J Crit Care 30(3):606\u0026ndash;612. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jcrc.2015.02.003\u003c/span\u003e\u003cspan address=\"10.1016/j.jcrc.2015.02.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFrench J, Weber T, Ge B et al (2021) Postoperative delirium in patients after brain tumor surgery. World Neurosurg 155:e472\u0026ndash;e479. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.wneu.2021.08.089\u003c/span\u003e\u003cspan address=\"10.1016/j.wneu.2021.08.089\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGunther ML, Morandi A, Ely EW (2008) Pathophysiology of delirium in the intensive care unit. Crit Care Clin 24(1):45\u0026ndash;viii. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ccc.2007.10.002\u003c/span\u003e\u003cspan address=\"10.1016/j.ccc.2007.10.002\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFlanigan PM, Jahangiri A, Weinstein D et al (2018) Postoperative delirium in glioblastoma patients: risk factors and prognostic implications. Neurosurgery 83(6):1161\u0026ndash;1172. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/neuros/nyx606\u003c/span\u003e\u003cspan address=\"10.1093/neuros/nyx606\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZheng YB, Ruan GM, Fu JX et al (2016) Postoperative plasma 8-iso-prostaglandin F2α levels are associated with delirium and cognitive dysfunction in elderly patients after hip fracture surgery. Clin Chim Acta 455:149\u0026ndash;153. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cca.2016.02.007\u003c/span\u003e\u003cspan address=\"10.1016/j.cca.2016.02.007\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen XW, Shi JW, Yang PS et al (2014) Preoperative plasma leptin levels predict delirium in elderly patients after hip fracture surgery. Peptides 57:31\u0026ndash;35. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.peptides.2014.04.016\u003c/span\u003e\u003cspan address=\"10.1016/j.peptides.2014.04.016\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNellis ME, Goel R, Feinstein S et al (2018) Association between transfusion of RBCs and subsequent development of delirium in critically ill Children. Pediatr Crit Care Med 19(10):925\u0026ndash;929. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1097/PCC.0000000000001675\u003c/span\u003e\u003cspan address=\"10.1097/PCC.0000000000001675\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang CM, Huang HW, Wang YM et al (2020) Incidence and risk factors of postoperative delirium in patients admitted to the ICU after elective intracranial surgery: A prospective cohort study. Eur J Anaesthesiol 37(1):14\u0026ndash;24. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1097/EJA.0000000000001074\u003c/span\u003e\u003cspan address=\"10.1097/EJA.0000000000001074\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHabeeb-Allah A, Alshraideh JA (2021) Delirium post-cardiac surgery: Incidence and associated factors. Nurs Crit Care 26(3):150\u0026ndash;155. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/nicc.12492\u003c/span\u003e\u003cspan address=\"10.1111/nicc.12492\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKwon DH, Kim BS, Chang H et al (2013) Exercise ameliorates cognition impairment due to restraint stress-induced oxidative insult and reduced BDNF level. Biochem Biophys Res Commun 434(2):245\u0026ndash;251. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.bbrc.2013.02.111\u003c/span\u003e\u003cspan address=\"10.1016/j.bbrc.2013.02.111\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCohen DM (2016) Modeling the neurologic and cognitive effects of hyponatremia. J Am Soc Nephrol 27(3):659\u0026ndash;661. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1681/ASN.2015060714\u003c/span\u003e\u003cspan address=\"10.1681/ASN.2015060714\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi Y, Li Y, Sen G et al (2026) Deep exploration and precise identification of key risk factors for diabetic peripheral neuropathy using the random forest algorithm. Front Endocrinol 16:1740545. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fendo.2025.1740545\u003c/span\u003e\u003cspan address=\"10.3389/fendo.2025.1740545\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEl-Bashbishy AE, El-Bakry HM (2024) Pediatric diabetes prediction using deep learning. Sci Rep 14(1):4206. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-024-51438-4\u003c/span\u003e\u003cspan address=\"10.1038/s41598-024-51438-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKamal MM, Khan W, Shambour QY et al (2026) A hybrid stacked autoencoder and support vector machines-based expert system for heart failure detection. Sci Rep 16(1):3886. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-025-34430-4\u003c/span\u003e\u003cspan address=\"10.1038/s41598-025-34430-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuo YF, Li YZ, Qi Y et al (2026) Diagnosis of cognitive impairment in chronic kidney disease: A radiomics and machine learning approach with quantitative susceptibility mapping. Brain Res Bull 234:111714. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.brainresbull.2025.111714\u003c/span\u003e\u003cspan address=\"10.1016/j.brainresbull.2025.111714\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhou Y, Li X, Huang W et al (2026) Predicting activities of daily living at discharge in stroke patients using rehabilitation robot training-induced functional connectivity. Top Stroke Rehabil 1\u0026ndash;12. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/10749357.2026.2612712\u003c/span\u003e\u003cspan address=\"10.1080/10749357.2026.2612712\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDavoudi A, Ebadi A, Rashidi P et al (2017) Delirium Prediction using Machine Learning Models on Preoperative Electronic Health Records Data. Proc IEEE Int Symp Bioinf Bioeng 568\u0026ndash;573. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/BIBE.2017.00014\u003c/span\u003e\u003cspan address=\"10.1109/BIBE.2017.00014\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCorradi JP, Thompson S, Mather JF et al (2018) Prediction of incident delirium using a random forest classifier. J Med Syst 42(12):261. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10916-018-1109-0\u003c/span\u003e\u003cspan address=\"10.1007/s10916-018-1109-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi XM, Wang XQ, Xu L et al (2022) Construction and validation of a risk prediction model for ICU delirium in patients after brain tumor surgery. Chin J Mod Nurs 28(29):3991\u0026ndash;3997. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3760/cma.j.cn115682-20220119-00316\u003c/span\u003e\u003cspan address=\"10.3760/cma.j.cn115682-20220119-00316\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMa Y, Ye L, Pan J et al (2026) Lung involvement percentage in patients with COVID-19 during the Omicron wave in China: a SHAP-explained machine learning study from a single center. Front Public Health 13:1728282. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fpubh.2025.1728282\u003c/span\u003e\u003cspan address=\"10.3389/fpubh.2025.1728282\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNakajima R, Kinoshita M, Okita H et al (2026) Surgical history and temporal muscle thickness as predictors of successful awake surgery in older patients with glioma. J Neurooncol 176(2):161. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11060-025-05400-7\u003c/span\u003e\u003cspan address=\"10.1007/s11060-025-05400-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"journal-of-neuro-oncology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"neon","sideBox":"Learn more about [Journal of Neuro-Oncology](https://www.springer.com/journal/11060)","snPcode":"11060","submissionUrl":"https://submission.nature.com/new-submission/11060/3","title":"Journal of Neuro-Oncology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Brain tumor, Machine learning, Postoperative delirium, Predictive model, Intensive Care Unit (ICU)","lastPublishedDoi":"10.21203/rs.3.rs-8893713/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8893713/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e \u003cb\u003eObjective\u003c/b\u003e This study aimed to clarify the incidence and influencing factors of delirium in ICU patients after brain tumor surgery, construct and validate delirium risk prediction models using multiple machine learning algorithms, identify the optimal model, and develop a personalized risk calculation tool to provide evidence-based support for early precise identification of high-risk patients and targeted preventive interventions.\u003c/p\u003e \u003cp\u003e \u003cb\u003eMethods\u003c/b\u003e Consecutive convenient sampling was adopted. A total of 600 patients who underwent brain tumor surgery in a Grade A tertiary hospital in Nanchang (July 2021\u0026ndash;December 2024) served as the modeling cohort, and 160 similar patients (January\u0026ndash;August 2025) as the external validation cohort. LASSO regression screened independent risk factors from 26 candidates. Five models (LR, ANN, RF, DT, SVM) were constructed and evaluated by sensitivity, AUC, DCA, etc. SHAP interpreted the optimal model\u0026rsquo;s feature importance.\u003c/p\u003e \u003cp\u003e \u003cb\u003eResults\u003c/b\u003e Five independent risk factors were identified: age, APACHE II score, postoperative GCS score, use of diuretics or dehydrating agents, and mechanical ventilation duration. The SVM model performed best: validation set AUC\u0026thinsp;=\u0026thinsp;0.857 (95%CI:0.814\u0026ndash;0.898), test set AUC\u0026thinsp;=\u0026thinsp;0.847, external validation accuracy 70.00%. SHAP showed diuretics or dehydrating agents were the most important feature (mean |SHAP|=0.1625). An online risk calculator was developed for convenient personalized assessment.\u003c/p\u003e \u003cp\u003e \u003cb\u003eConclusions\u003c/b\u003e The SVM-based predictive model has excellent efficacy and generalizability. The easy-to-operate risk calculator can effectively identify high-risk patients early, providing scientific support for precise delirium prevention and control.\u003c/p\u003e","manuscriptTitle":"Development and validation of a machine learning-based model for predicting delirium risk in postoperative brain tumor patients in the Intensive Care Unit","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-23 08:53:04","doi":"10.21203/rs.3.rs-8893713/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-04-30T19:58:52+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-30T14:00:47+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-31T09:30:13+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"207481713934307604976631431329579120660","date":"2026-02-20T16:36:45+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"184414426618972070699642746102331040270","date":"2026-02-19T08:53:21+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-02-18T18:37:23+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-02-18T18:15:49+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-02-18T18:10:05+00:00","index":"","fulltext":""},{"type":"submitted","content":"Journal of Neuro-Oncology","date":"2026-02-16T13:44:17+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"journal-of-neuro-oncology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"neon","sideBox":"Learn more about [Journal of Neuro-Oncology](https://www.springer.com/journal/11060)","snPcode":"11060","submissionUrl":"https://submission.nature.com/new-submission/11060/3","title":"Journal of Neuro-Oncology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"eb680ec2-fb8f-4305-9375-bf0b6ca3cca1","owner":[],"postedDate":"February 23rd, 2026","published":true,"recentEditorialEvents":[{"type":"decision","content":"Revision requested","date":"2026-04-30T19:58:52+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-30T14:00:47+00:00","index":15,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"in-revision","subjectAreas":[],"tags":[],"updatedAt":"2026-04-30T20:08:52+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-23 08:53:04","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8893713","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8893713","identity":"rs-8893713","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}