External validation in a Chinese cohort of a nomogram developed using MIMIC-IV for predicting sepsis-associated delirium in elderly ICU patients

External validation in a Chinese cohort of a nomogram developed using MIMIC-IV for predicting sepsis-associated delirium in elderly ICU patients | 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 External validation in a Chinese cohort of a nomogram developed using MIMIC-IV for predicting sepsis-associated delirium in elderly ICU patients Tengfei Zhou, Xinming Tian, Xuyang Han, Wei Wang, Yanhui Feng, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7380480/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 7 You are reading this latest preprint version Abstract Background : Sepsis-associated delirium (SAD) is a common acute brain dysfunction in elderly patients in the intensive care unit (ICU), which significantly increases the length of hospital stay, medical costs, and the risk of death. Despite the availability of multiple delirium prediction tools, there are few models specific to the elderly septic population, and most lack external validation. Methods : This retrospective cohort study enrolled 5034 elderly ICU patients with sepsis. A prediction model was constructed based on the MIMIC-IV database and externally validated using 281 patients admitted to the First Affiliated Hospital of Jilin University between January 2019 and November 2024. A workflow was developed using R software. Candidate predictors were first identified using the LASSO regression method, then incorporated into a multivariate logistic regression model and visualized as a nomogram. Patients were randomly divided into a training set and an internal validation set in a 6:4 ratio. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), the Hosmer-Lemeshow test, calibration plots, Brier scores, and decision curve analysis (DCA). Results : The overall incidence of delirium was 46.44%. Seven variables were ultimately included in the final model: body temperature, SOFA score, hemoglobin level, serum sodium concentration, history of neurological disease, mechanical ventilation, and midazolam use. The model demonstrated good discriminatory performance, with area under the receiver operating characteristic curve (AUC) values of 0.794 (95% CI: 0.777–0.810) in the training set, 0.784 (95% CI: 0.763–0.804) in the internal validation set, and 0.815 (95% CI: 0.765–0.864) in the external validation set. The Hosmer–Lemeshow goodness-of-fit test showed no significant deviation between predicted and observed outcomes (P > 0.05), indicating good calibration. The Brier scores were 0.185, 0.188, and 0.176 for the training, internal validation, and external validation sets, respectively. Decision curve analysis (DCA) further confirmed the model’s potential clinical utility. Conclusion: The SAD risk prediction model developed in this study features a simple structure and relies on readily available clinical variables. It demonstrated favorable discrimination and calibration in the external validation cohort, suggesting its potential utility as a practical tool for early detection and targeted intervention of delirium in elderly ICU patients with sepsis. Sepsis-associated delirium elderly patients intensive care unit nomogram predictive model Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1 Introduction Sepsis is a clinical syndrome of immune dysregulation caused by infection, which can lead to serious consequences such as septic shock and multiple organ failure (MODS) [ 1 ]. Annually, approximately 48.9 million new sepsis cases occur worldwide, representing about 0.7% of the global population and resulting in an estimated 11 million deaths, accounting for nearly 19.7% of all global deaths during that timeframe [ 2 , 3 ].Delirium is an acute neurocognitive disorder that is costly, serious, and life-threatening, affecting approximately 50% of hospitalized elderly patients [ 4 , 5 ]. Sepsis-associated delirium (SAD) is a common neuropsychiatric complication in elderly ICU patients, who usually have a poor prognosis [ 6 – 8 ]. The reported prevalence of SAD varies greatly across studies, generally ranging from 9–76.9%, with an average of approximately 42.9%. SAD is significantly associated with several risk factors, including elevated Sequential Organ Failure Assessment (SOFA) scores, increased lactate levels, hyperglycemia, and frailty [ 9 – 11 ]. Studies have shown that the risk of in-hospital mortality in patients with SAD is approximately twice that of sepsis patients without delirium, and the duration of mechanical ventilation and hospital stay is significantly prolonged, leading to a 1.5- to 2-fold increase in healthcare costs [ 12 , 13 ]. The complex interaction between systemic organ failure and sepsis makes the clinical management of this population very complicated [ 14 , 15 ]. Given that SAD is challenging to reverse, medical professionals must prevent and address it in time. As the global population ages, the proportion of elderly patients in the ICU continues to rise, and sepsis management in this population has gradually become a clinical priority [ 16 ]. Immune senescence and impaired brain regulatory function render older individuals with sepsis more susceptible to delirium. Among the widely used delirium screening tools, the Confusion Assessment Method for the ICU (CAM-ICU) is the most common [ 17 ]. Meta-analysis shows that CAM-ICU has a high recognition efficiency in general hospitalized patients and has the highest sensitivity and specificity among various screening tools [ 18 ]. However, due to the complex clinical manifestations and multiple comorbidities of elderly patients with sepsis in the ICU, there is a certain risk of missed diagnosis when using CAM-ICU. Current research has found that the sensitivity of CAM-ICU for detecting delirium in elderly hospitalized patients is only 53%, and the sensitivity for patients with moderate delirium is even lower, at 30%, which seriously restricts the early identification of SAD in clinical practice [ 19 ]. Consequently, it is imperative to thoroughly examine the clinical characteristics of the elderly sepsis cohort to facilitate early identification and targeted care for high-risk patients. With the development of precision medicine, clinical prediction models related to delirium are becoming increasingly important in risk assessment, decision making, and personalized care [ 20 ]. Several delirium prediction models have been proposed, among which PRE-DELIRIC and E-PRE-DELIRIC are representative models. The PRE-DELIRIC model is based on 10 variables collected within 24 hours after admission to a multicenter ICU in the Netherlands, including age, APACHE-II score, coma, sedative or morphine use, and emergency admission [ 21 ]. The model's AUC reached 0.87 (95% CI: 0.85–0.89), which was significantly better than the judgment of doctors and nurses (0.59; 95% CI: 0.49–0.70). The model was first prospectively validated in the time dimension, with an AUC of 0.89 (95% CI: 0.86–0.92), a calibration slope of 1.2, and an intercept of 0.22, showing good fit. Subsequently, an international multicenter external validation was conducted in eight ICUs in six European countries. The combined AUC was 0.77 (95% CI: 0.74–0.79), the Hosmer-Lemeshow test P = 0.045, and the calibration was good [ 22 ]. However, the original authors pointed out that the model should be used with caution in populations with a high incidence of delirium. If it is not recalibrated, there may be a bias of overestimating the risk [ 21 ]. The E-PRE-DELIRIC model integrates data from thirteen ICUs in seven countries, taking into account both the generality and stability of the model. Its development set AUC was 0.76 (95% CI: 0.73–0.77) and the validation set AUC was 0.75 (95% CI: 0.71–0.79), showing good calibration ability [ 23 ]. In addition, the model can also predict the risk of delirium on the second and sixth days of ICU hospitalization (AUCs of 0.70 and 0.81, respectively), showing certain advantages in temporal dynamic prediction. The researchers also pointed out that the E-PRE-DELIRIC model can show good fit in different regions without recalibration. As our understanding of delirium continues to deepen, there are many other models that deserve our attention. Shi et al. [ 24 ] explored the predictive value of biomarkers based on the lymphocyte-to-monocyte ratio (LMR) and found that LMR was significantly negatively correlated with SAD. The model AUC was significantly improved after the introduction of this indicator (P = 0.035). After the introduction of LMR, the AUC value of the model increased steadily, and the difference was statistically significant (P = 0.035), suggesting that it has certain predictive potential. Gu et al. [ 25 ] constructed a regression model based on the MIMIC-III database, including SOFA score, mechanical ventilation, phosphate, and lactate. The training set AUC was 0.742, the validation set AUC was 0.713, and the Hosmer-Lemeshow test P = 0.471. The calibration was good, and the decision curve analysis also suggested that it has practical value. Machine learning algorithms have developed rapidly in recent years and have unique advantages in processing complex data. Zhang et al. [ 26 ] and Yu et al. [ 27 ] constructed seven machine learning models based on the MIMIC-IV database and performed external validation in the eICU-CRD database. The AUCs of the XGBoost model in the modeling set were 0.793 and 0.775, and those in the validation set were 0.701 and 0.687. The calibration curve and decision curve were drawn for each model. The results showed that XGBoost performed best and had certain generalization ability and potential application value. In addition, the variable importance ranking results also showed that "mechanical ventilation" was the factor most strongly associated with delirium (SHAP values were 0.593 and 0.663, respectively). Fu et al. [ 28 ] used a machine learning algorithm to construct a nomogram containing four risk factors and also found that the XGBoost model had the best predictive performance. The AUC of XGBoost was 0.79 (95% CI: 0.75–0.83), but the model has not been externally validated, and its generalization ability remains to be examined. It is worth noting that the above models are all constructed and verified based on clinical databases of European and American populations, especially MIMIC-IV and eICU-CRD, which are both American hospital data, and their population characteristics may be different from those in China or Asia [ 29 , 30 ]. Some studies have attempted to introduce these models into the Asian context. Kim et al. [ 31 ] collected data from patients in the cardiac intensive care unit (CICU) of Samsung Medical Center in South Korea and verified the applicability of the PRE-DELIRIC and E-PRE-DELIRIC models in CICU patients. The results showed that the PRE-DELIRIC model had better prediction results, with an AUC of 0.84 (95% CI: 0.82–0.86), and showed good generalization ability in Asian populations. Hsiao et al. [ 32 ] conducted an intervention study based on the PRE-DELIRIC model in a tertiary hospital in Taipei, China. The model was evaluated in combination with SMART bundle care. The incidence of delirium in the intervention group was significantly lower than that in the control group (22.3% vs. 47.7%), proving that the model is feasible and effective in clinical practice in Asia. These studies provide preliminary evidence for the cross-regional applicability of delirium prediction models in Asian populations. However, whether it can be extended to a wider range of high-risk populations still needs further exploration and verification in combination with regional data. China, one of the most populous countries in the world, is facing an increasingly severe challenge of an aging population. A nationwide population-based study estimated that there were approximately 4.8 million to 6.1 million hospitalized sepsis cases in China each year from 2017 to 2019, 57.5% of which occurred in people aged 65 years and above [ 33 ]. Considering that the elderly are a high-risk group for SAD, this study, based on the construction of a prediction model using the MIMIC-IV database, further introduced independent clinical data from an Asian population (from the ICU of a large tertiary hospital in China) as an external validation cohort. By integrating international and local data, the model aims to complement existing tools and enhance early identification and personalized management of high-risk patients. 2 Methods 2.1 Research methodology and data sources This study employed a retrospective cohort design. Model development was based on data from the Medical Information Mart for Intensive Care IV (MIMIC-IV, version 3.1), a publicly available database that includes detailed clinical information on ICU patients admitted to large medical centers in the United States between 2008 and 2022 [ 30 , 34 ]. The principal investigator (ZTF) has completed the CITI ethics examination program and has a valid certification (No: 63365666). Elderly ICU patients aged ≥ 65 years with a diagnosis of sepsis were identified from the MIMIC-IV database. Eligible patients were randomly assigned to a training set (60%) and an internal validation set (40%) using a random number generator based on terminal digits (0–9). The external validation cohort was derived from the central intensive care unit of a tertiary hospital in China, comprising patients admitted between January 2019 and November 2024. Ethical approval for the use of clinical data was obtained from the Institutional Review Board of the First Affiliated Hospital of Jilin University (Approval No: 2024 − 1161). This study complied with the ethical principles of the Declaration of Helsinki and followed the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) guideline for predictive modeling [ 35 ]. 2.2 Patient selection Patients admitted to the ICU for the first time with a confirmed diagnosis of sepsis were selected from the MIMIC-IV database. All enrolled patients had a documented delirium assessment. The inclusion and exclusion criteria for the external validation cohort were consistent with those of the MIMIC-IV dataset. Patients were excluded based on the following criteria: (1) Age < 65 years; (2) ICU stay < 48 hours; (3) Absence of CAM-ICU evaluation or lack of documentation of delirium; (4) Presence of conditions that may confound the diagnosis of delirium, including dementia, epilepsy, schizophrenia, traumatic brain injury, meningitis, brain tumor, alcohol abuse, and drug poisoning. 2.3 Outcome definition The primary outcome was the occurrence of delirium during the ICU stay of the patients involved. (1) Sepsis-3.0: Clinically confirmed life-threatening organ dysfunction caused by suspected or confirmed infection and SOFA score ≥ 2 points [ 1 ]. (2) Delirium: We used the CAM-ICU score [ 17 ] or the medical staff's delirium diagnosis record to determine whether the patient had delirium. The ‘chartevents’ table in the MIMIC-IV database was combined to identify the occurrence of delirium in ICU patients. For CAM-ICU score assessment, we extracted ‘itemid’ including 229326, 228337, 229325, 228334, 228336, 228303, 229324, 228300, 228335, 228302, and 228301. The result in the value field was used for judgment. A record of ‘Yes’ was defined as positive for delirium, a record of ‘No’ was defined as negative, and a record of ‘Unable to assess’ was considered as missing. The CAM-ICU comprises four components: ① abrupt onset of cognitive alterations or a fluctuating trajectory; ② inattention; ③ disorganized thought functions; ④ altered awareness level. A positive diagnosis of CAM-ICU was confirmed if the criteria ① + ② + ③ or ① + ② + ④ were satisfied. For the medical staff assessment records, the extracted ‘itemid’ is 228332, where the value of ‘Positive’ is considered positive, ‘Negative’ is considered negative, and ‘UTA’ is a missing outcome. Patients were excluded if both sources (CAM-ICU and staff notes) were missing or if delirium occurred prior to ICU admission or within the first 24 hours of ICU stay. 2.4 Variables and data extraction The prediction model cohort was derived by linking the MIMIC-IV (v3.1) database in PostgreSQL utilizing Navicat Premium 15.0. We obtained the same variable characteristics as the modeling cohort from the clinical electronic medical records of the central ICU of our hospital. For continuous variables (including vital signs and laboratory parameters) recorded within 24 hours after ICU admission, we extracted statistical features based on clinical significance and data characteristics. Specifically, the minimum, maximum, or mean value was extracted for each variable, with the choice depending on clinical interpretability and the proportion of missing data. Categorical variables (e.g., sex, comorbidities, medications, and interventions) were coded as binary indicators. The following data were extracted: (1) Demographic variables: age, gender, weight, admission type; (2) Vital signs within 24 hours of admission: temperature mean (T), heart rate mean (HR), respiratory rate mean (RR), SBP mean, DBP mean, MBP mean, partial pressure of oxygen (pO 2 ) max, partial pressure of carbon dioxide (pCO 2 ) max; (3) Comorbidity and score: hypertension, diabetes, cancer, respiratory failure, renal failure, neurological disease, sequential organ failure assessment score (SOFA); (4) Biochemistry within 24 hours of admission: hemoglobin min, blood urea nitrogen (BUN) min, calcium min, creatinine min, potassium min, international normalized ratio (INR) min, prothrombin time (PT) min, partial thromboplastin time (PTT) min, pH min, lactate min, platelets max, WBC max, sodium max; (5) Whether treatment was received within 24 hours of admission: continuous renal replacement therapy (CRRT) and the application of mechanical ventilation; (6) Whether to receive medication within 24 hours after admission: vasopressin, midazolam, fentanyl. 2.5 Statistical methods and model construction All statistical analyses were performed using R software (version 4.3.3). Multiple imputation was performed using the “mice” package in R for variables with a missing proportion of less than 20%, and variables with a missing proportion of more than 20% were deleted (see Supplementary Table S1 for details). Descriptive statistical analysis was performed before modeling. We used the Shapiro-Wilk test to see if all continuous variables were normally distributed. The results showed that all were non-normally distributed (P < 0.05). Therefore, continuous variables were summarized as medians and interquartile ranges (IQRs), and intergroup differences were assessed using the Kruskal-Wallis test. Categorical data were reported as frequencies and proportions (n, %), with group comparisons conducted using either Pearson’s chi-square test or Fisher’s exact test, depending on data distribution. All statistical evaluations were performed as two-tailed tests, with a significance level set at P < 0.05. For variable selection, we used least absolute shrinkage and selection operator (LASSO) regression, implemented in the glmnet package in R. LASSO regression was applied to the training cohort, penalizing noninformative variables to shrink coefficients and select the most relevant predictors. Ten-fold cross-validation was used to determine the optimal regularization parameter (λ). The predictor set identified by LASSO was ultimately entered into a multivariable logistic regression model to estimate effect sizes and construct a predictive model. Nomograms were constructed using the rms package to visualize the model. Model validation was performed in an internal validation cohort (MIMIC data with a 60/40 randomization split) and an external validation cohort from the ICU of a tertiary hospital in China. The area under the receiver operating characteristic curve (AUC) was calculated using the pROC package to assess the discriminative ability of the model. Calibration was assessed using 1,000 bootstrap resamplings with the rms package, and nonparametric LOESS smoothed calibration curves were plotted to compare predicted and observed probabilities. Model accuracy was quantified using the Brier score, and its 95% confidence interval was estimated using percentile bootstrapping (n = 1,000). Furthermore, the Hosmer-Lemeshow goodness-of-fit test was performed to assess the consistency between predicted and observed results. Finally, decision curve analysis (DCA) was performed to evaluate the potential clinical utility of the model [ 36 ]. This study utilized the following R packages: mice [ 37 ], rms [ 38 ], pROC [ 39 ], Hmisc [ 40 ], and corrplot [ 41 ]. 3 Results 3.1 Patient selection clinical characteristics Patients in this study were recruited from two cohorts: the MIMIC-IV database and an intensive care unit (ICU) at a Chinese hospital. The complete patient screening process is shown in Fig. 1 . After rigorous screening, a total of 5,034 elderly ICU patients with sepsis were enrolled. Of these, 2,852 patients were assigned to the training cohort, 1,901 to the internal validation cohort, and 281 to the external validation cohort (Table 1 ). The overall incidence of sepsis-associated delirium (SAD) was 46.44% (2,338/5,034). The incidence in the training cohort was 46.91% (1,338/2,852), the internal validation cohort was 46.29% (880/1,901), and the external validation cohort was 42.70% (120/281). Detailed baseline characteristics and between-group differences are provided in Tables S2-S4. Demographic characteristics differed among the three cohorts, with the median age of the external validation cohort being slightly lower than that of the training and internal validation cohorts (P < 0.001). Similarly, weight was significantly lower in the external validation cohort than in the training and internal validation cohorts (P < 0.001). However, no significant differences were observed in gender distribution or hospital admission type. Regarding vital signs, heart rate was higher in the external validation group than in the training group (84 beats/min vs. 80 beats/min, P < 0.001), while no significant differences were observed in other vital sign parameters. The external validation group had a higher prevalence of hypertension and diabetes (44.5% and 27.4%, respectively; both P < 0.001). Regarding treatment interventions, the external validation cohort had a lower proportion of patients receiving mechanical ventilation and vasopressors (32.7% and 24.6%, respectively, both P < 0.001), while there was a higher prevalence of fentanyl use (66.5%, P < 0.001). No significant differences were observed in other treatment interventions. Table 1 Baseline characteristics of the study population in the training, internal validation, and external validation sets. Characteristic Training set (N = 2,852) Internal validation set (N = 1,901) External validation set (N = 281) Statistic 1 p-value 1 Demographic variables Age, Median (Q1, Q3) 77 (71, 83) 77 (71, 84) 74 (70, 80) 23.12 < 0.001 Gender, n (%) 0.33 0.849 Female 1,289 (45.2%) 874 (46.0%) 126 (44.8%) Male 1,563 (54.8%) 1,027 (54.0%) 155 (55.2%) Weight, Median (Q1, Q3) 79 (65, 92) 79 (66, 92) 71 (62, 84) 28.84 < 0.001 Admission type, n (%) 1.71 0.425 Non-Emergency 885 (31.0%) 620 (32.6%) 84 (29.9%) Emergency 1,967 (69.0%) 1,281 (67.4%) 197 (70.1%) Vital signs Temperature mean, Median (Q1, Q3) 36.90 (36.60, 37.20) 36.80 (36.60, 37.20) 36.90 (36.60, 37.30) 1.33 0.514 Heart rate mean, Median (Q1, Q3) 84 (74, 95) 83 (74, 95) 88 (76, 104) 18.15 < 0.001 Resp rate mean, Median (Q1, Q3) 19.0 (17.0, 22.0) 19.0 (17.0, 22.0) 20.0 (17.0, 23.0) 5.35 0.069 SBP mean, Median (Q1, Q3) 113 (105, 125) 113 (105, 125) 112 (102, 122) 6.19 0.045 DBP mean, Median (Q1, Q3) 58 (53, 65) 58 (53, 65) 59 (52, 68) 1.81 0.404 MBP mean, Median (Q1, Q3) 74 (69, 81) 74 (69, 81) 74 (69, 81) 0.16 0.925 pO2 max, Median (Q1, Q3) 153 (82, 312) 150 (85, 324) 128 (81, 254) 5.89 0.052 pCO2 max, Median (Q1, Q3) 45 (40, 53) 45 (39, 53) 42 (35, 50) 27.36 < 0.001 Comorbidity variables Hypertension, n (%) 20.29 < 0.001 No 1,956 (68.6%) 1,296 (68.2%) 156 (55.5%) Yes 896 (31.4%) 605 (31.8%) 125 (44.5%) Diabetes, n (%) 13.97 < 0.001 No 2,332 (81.8%) 1,532 (80.6%) 204 (72.6%) Yes 520 (18.2%) 369 (19.4%) 77 (27.4%) Cancer, n (%) 2.09 0.351 No 2,242 (78.6%) 1,506 (79.2%) 212 (75.4%) Yes 610 (21.4%) 395 (20.8%) 69 (24.6%) Respiratory failure, n (%) 6.77 0.034 No 2,254 (79.0%) 1,522 (80.1%) 206 (73.3%) Yes 598 (21.0%) 379 (19.9%) 75 (26.7%) Renal failure, n (%) 7.25 0.027 No 2,089 (73.2%) 1,388 (73.0%) 226 (80.4%) Yes 763 (26.8%) 513 (27.0%) 55 (19.6%) Neurological disease, n (%) 0.59 0.744 No 2,174 (76.2%) 1,432 (75.3%) 211 (75.1%) Yes 678 (23.8%) 469 (24.7%) 70 (24.9%) SOFA score, Median (Q1, Q3) 5.0 (4.0, 8.0) 6.0 (4.0, 8.0) 6.0 (4.0, 9.0) 2.48 0.290 Biochemistry variables Hemoglobin min, Median (Q1, Q3) 9.10 (7.90, 10.70) 9.10 (7.90, 10.80) 9.60 (8.40, 11.00) 10.86 0.004 Platelets max, Median (Q1, Q3) 212 (153, 285) 202 (150, 274) 197 (134, 279) 10.57 0.005 WBC max, Median (Q1, Q3) 13 (10, 19) 14 (10, 19) 13 (8, 19) 3.51 0.173 BUN min, Median (Q1, Q3) 23 (16, 38) 22 (15, 35) 25 (23, 29) 25.88 < 0.001 Calcium min, Median (Q1, Q3) 8.10 (7.60, 8.60) 8.10 (7.60, 8.60) 8.00 (7.20, 8.10) 50.62 < 0.001 Creatinine min, Median (Q1, Q3) 1.10 (0.80, 1.60) 1.00 (0.80, 1.60) 1.10 (0.80, 1.70) 3.04 0.219 Sodium max, Median (Q1, Q3) 141.0 (138.0, 144.0) 141.0 (138.0, 144.0) 141.0 (138.0, 144.0) 0.04 0.981 Potassium min, Median (Q1, Q3) 3.90 (3.60, 4.30) 3.90 (3.50, 4.30) 3.90 (3.50, 4.30) 1.86 0.395 INR min, Median (Q1, Q3) 1.20 (1.10, 1.40) 1.20 (1.10, 1.40) 1.20 (1.10, 1.40) 1.08 0.582 PT min, Median (Q1, Q3) 13.6 (12.2, 15.6) 13.5 (12.4, 15.7) 13.5 (12.3, 15.2) 0.82 0.665 PTT min, Median (Q1, Q3) 29 (26, 34) 29 (26, 34) 29 (27, 34) 1.56 0.458 Lactate min, Median (Q1, Q3) 1.40 (1.00, 1.90) 1.40 (1.00, 1.90) 1.40 (1.00, 2.20) 2.70 0.260 PH min, Median (Q1, Q3) 7.33 (7.27, 7.39) 7.33 (7.27, 7.39) 7.35 (7.27, 7.40) 2.74 0.255 Treatment variables CRRT, n (%) 10.78 0.005 No 2,666 (93.5%) 1,761 (92.6%) 248 (88.3%) Yes 186 (6.5%) 140 (7.4%) 33 (11.7%) Mechanical ventilation, n (%) 90.34 < 0.001 No 1,113 (39.0%) 726 (38.2%) 189 (67.3%) Yes 1,739 (61.0%) 1,175 (61.8%) 92 (32.7%) Vasopressin use, n (%) 47.20 < 0.001 No 2,537 (89.0%) 1,694 (89.1%) 212 (75.4%) Yes 315 (11.0%) 207 (10.9%) 69 (24.6%) Midazolam use, n (%) 1.66 0.436 No 2,082 (73.0%) 1,419 (74.6%) 209 (74.4%) Yes 770 (27.0%) 482 (25.4%) 72 (25.6%) Fentanyl use, n (%) 30.58 < 0.001 No 1,431 (50.2%) 964 (50.7%) 94 (33.5%) Yes 1,421 (49.8%) 937 (49.3%) 187 (66.5%) Note: SOFA score, Sequential Organ Failure Assessment Score; WBC, white blood cell; SBP, systolic blood pressure; DBP, diastolic blood pressure; MBP, Mean arterial blood pressure; pO2, partial pressure of Oxygen; pCO2, partial pressure of carbon dioxide; WBC, white blood cell; BUN, blood urea nitrogen; INR, International normalized ratio; PT, prothrombin time; PTT, partial thromboplastin time; PH, potential of hydrogen; CRRT, continuous renal replacement therapy. 1 Pearson's Chi-squared test; Kruskal-Wallis rank sum test. --------- Table 1 is about here--------------- 3.2 Predictors of SAD To identify the most relevant predictors of sepsis-associated delirium (SAD) in elderly ICU patients, we used the least absolute shrinkage and selection operator (LASSO) logistic regression for variable selection. Figure 2 A shows the results of 10-fold cross-validation to determine the optimal λ value, and Fig. 2 B shows the coefficient paths for all candidate variables. Based on the minimum binomial deviance criterion, eight variables were selected for further modeling. These predictors were then included in a multivariate logistic regression analysis to estimate effect sizes. The results (Fig. 3 ) showed that the following factors were significantly associated with the development of SAD: neurological disease (OR = 1.660, 95% CI: 1.352–2.033, P < 0.001); mechanical ventilation (OR = 1.883, 95% CI: 1.561–2.264, P < 0.001); midazolam use (OR = 3.931, 95% CI: 3.121–4.958, P < 0.001); fentanyl use (OR = 0.833, 95% CI: 0.672–1.015, P = 0.064); SOFA score (OR = 1.031, 95% CI: 1.010–1.062, P = 0.036); temperature mean (OR = 2.014, 95% CI: 1.719–2.353, P < 0.001); hemoglobin min (OR = 0.843, 95% CI: 0.807–0.881, P < 0.001); sodium max (OR = 1.098, 95% CI: 1.081–1.116, P 0.05 after multivariate analysis ( Table S5 ). 3.3 Development of the predictive model In this study, we constructed a prediction model covering seven risk factors with clinical significance. We also designed a nomogram. In clinical applications, the patient's variable values are sequentially mapped to the scoring scale at the top of the nomogram, yielding a matching score for each variable (0-100), which is subsequently aggregated to give the total score. The overall score, combined with the risk axis beneath the nomogram, allows for the prediction of the patient's chance of SAD (Fig. 4 ). 3.4 Internal and external validation of the predictive model The predictive performance of the model, as measured by the AUC, was evaluated in three cohorts: the training set (AUC = 0.794, 95%CI: 0.777–0.810), the internal validation set (AUC = 0.784, 95%CI: 0.763–0.804), and the external validation set (AUC = 0.815, 95%CI: 0.765–0.864) ( Fig. 5 ) . These results suggest that the model has a certain discriminative ability in predicting SAD in elderly ICU patients. Calibration curves were drawn for the training set, internal validation set, and external validation set, and the Brier score was calculated ( Fig. 6 A–C ) . The calibration curves showed that the predicted probabilities of all cohorts were basically consistent with the ideal diagonal line. The Brier score for the training set was 0.185 (95% CI: 0.177–0.192), the Brier score for the internal validation set was 0.188 (0.179–0.197), and the Brier score for the external validation set was 0.176 (0.154–0.198). These values indicate acceptable calibration performance and suggest that the model has a certain generalization ability. The results of the H–L test showed that none of the training set (χ² = 12.918, P = 0.115), internal validation set (χ² = 4.385, P = 0.821), and external validation set (χ² = 8.954, P = 0.346) reached a significant level, indicating that the model fit was good (Table S6) . Moreover, decision curve analysis (DCA) showed that the model had potential clinical applicability across different decision threshold ranges in all three cohorts ( Fig. 7 A–C ). 4 Discussion This study used the MIMIC-IV database to develop a novel clinical prediction model for sepsis-associated delirium (SAD) in elderly ICU patients and externally validated its performance in an independent cohort at a tertiary hospital in China. The MIMIC database has been widely used in precision medicine research due to its comprehensive and high-quality ICU data. However, previous MIMIC-based delirium prediction models have primarily focused on Western populations and often lack external validation in diverse ethnic groups. This may limit their global applicability due to potential cultural, clinical, and systemic differences. To address this issue, we incorporated an independent external validation cohort consisting of elderly ICU patients with sepsis from a tertiary hospital in China. This approach enabled us to assess the model's applicability across diverse healthcare systems and ethnic groups, thereby enhancing its clinical relevance. Our model incorporated seven conventional variables: SOFA score, neurologic disease, mechanical ventilation, hemoglobin level, temperature, midazolam use, and serum sodium concentration. Both internal and external validation demonstrated good discrimination and calibration, indicating the model's potential clinical utility. The overall incidence of SAD among the elderly in the ICU in this study was 46.44%, aligning closely with the incidence reported in previous studies (9–76.9%) [ 7 ]. The external validation cohort originated from an ICU in a Chinese hospital, where the incidence of SAD was marginally lower (42.70%), potentially attributable to the data being sourced from a single site and the relatively small case volume. It is worth noting that the standardized diagnosis of SAD in the elderly still faces many difficulties, and there is still a lack of accurate identification tools for sepsis-induced neurological dysfunction in clinical practice [ 42 ]. The current diagnostic criteria and screening instruments are primarily based on research with adult ICU patients and may lack enough sensitivity for the older demographic. Although the CAM-ICU score is a commonly used tool in clinical practice, its positive diagnosis requires multiple evaluations by professional nurses and is greatly affected by factors such as the patient's state of consciousness, use of sedatives, endotracheal intubation, and language barriers, which can easily lead to missed diagnoses or biased results [ 26 ]. During the external validation of this study, some patients were left out because their medical records did not have valid CAM-ICU score information, which resulted in some patients being lost in the validation sample and indirectly highlighted the issue of potentially missing diagnoses with the CAM-ICU score for identifying SAD in older adults. Several delirium prediction models have been developed and extensively validated. For example, the PRE-DELIRIC model includes 10 common ICU variables and demonstrates high accuracy (AUC up to 0.85, 95% CI: 0.84–0.87). Although our model had a slightly lower AUC in the training set (0.794, 95% CI: 0.777–0.810), it showed high external validity in the Asian cohort (AUC: 0.815, 95% CI: 0.765–0.864). Researchers van den et al. [ 21 ] believed that the PRE-DELIRIC model should be used with caution to predict delirium in high-risk patients. In contrast, we are more concerned about the high-risk subgroup of elderly sepsis patients. Moreover, Gu et al. [ 25 ] constructed a multifactorial prediction model for delirium using MIMIC, with an AUC of 0.743, but no external validation was performed. Our model is superior to their model (AUC = 0.794) and has additional external validation. Machine learning–based approaches have also been proposed. Fu et al. constructed a SAD prediction model using four algorithms, among which the best algorithm (XGBoost) achieved an AUC of 0.767 and had good internal calibration [ 28 ]. However, the algorithm has not been externally validated, which limits its applicability in practical applications. Zhang et al. [ 26 ] and Yu et al. [ 27 ] applied machine learning to the MIMIC-IV data and validated their models on the eICU-CRD database. While the internal AUC was satisfactory (Zhang: 0.793; Yu: 0.775), performance decreased significantly during external validation (0.701 and 0.687, respectively), demonstrating the performance degradation of the models when transferring across populations. Among these models, XGBoost is considered to be the best algorithm, but the model has many variables and may require more explanation to use. In contrast, the model we constructed added validation of the Asian population, and the seven variables in the model are simple and easy to obtain. The external validation AUC was 0.815, which is relatively stable and has potential applicability. The current study found that midazolam use, body temperature, neurological diseases and mechanical ventilation were the most important risk factors for SAD in elderly patients. Among them, midazolam (OR = 4.131, 95%CI: 3.271–5.218) had the greatest impact on the outcome. Midazolam is a short-acting benzodiazepine sedative. In routine ICU management, patients with severe sepsis are usually treated with mechanical ventilation and assisted with sedatives and analgesics [ 43 ]. Elderly patients are more susceptible to cognitive disorders and changes in consciousness due to their decreased ability to metabolize drugs and increased permeability of the blood-brain barrier. A prospective cohort study found that among 950 ICU patients receiving midazolam after continuous sedation, the incidence of delirium was 73% [ 44 ]. However, a randomized controlled trial by Hughes et al. [ 45 ] demonstrated no significant association in delirium incidence and mortality in patients with severe sepsis sedated with dexmedetomidine, midazolam, or propofol, suggesting that the choice of sedative may not be the sole determining factor. Current studies have not achieved consensus on the effects of sedatives in SAD. The impact of sedatives on SAD in the elderly may be influenced by various parameters, including dosage, timing of administration, and individual patient characteristics [ 46 ]. Our study found that the early use of midazolam in elderly patients with sepsis admitted to the ICU was significantly associated with delirium. Careful sedation in clinical practice may help reduce the risk of SAD in this vulnerable population, but a large number of prospective high-quality studies are still needed to verify this. Mechanical ventilation has been identified as a high-risk factor by multiple delirium prediction models, especially in patients with severe sepsis [ 47 ], which is consistent with the results of this study. Critically ill patients receiving mechanical ventilation usually require continuous sedation, which can easily lead to circadian rhythm disorders, sensory deprivation, reduced cognitive stimulation, and adverse events such as delirium. Studies show that prompt initiation of mechanical ventilation (< 24 hours) and shortened duration of ventilation may alleviate problems and shorten hospital stay [ 48 ]. Mechanical ventilation is one of the important predictors in the model, suggesting that we may need to pay more attention to elderly patients who require early mechanical ventilation. Increased body temperature is usually part of the response to systemic inflammation (such as sepsis), when a large number of proinflammatory cytokines (such as IL-1β and TNFα) can penetrate the blood-brain barrier, interfere with neurotransmitter metabolism, and may induce a series of abnormal neurological reactions [ 6 ]. Multivariate regression analysis showed that the risk of delirium in patients with abnormal body temperature fluctuations was approximately twice that of patients with normal body temperature (OR = 2.180, 95%CI: 1.856–2.543). Tan et al. [ 49 ] found in a large sample study based on the MIMIC and eICU databases that body temperature deviation from the normal range (such as below 36°C or above 38°C) was significantly associated with ICU patient mortality. Irregular changes in patient body temperature may be associated with brain injury, and the possible impact of these temperature fluctuations on neurologic status warrants careful consideration; however, robust prospective studies are needed to confirm this claim. In addition, the history of neurological diseases is also one of the important predictive factors of the model. The patient's brain structure and functional reserve may be damaged. When faced with the pressure of acute inflammation, metabolic disorders or drug stimulation caused by sepsis, it is difficult to withstand the further decline of neurological function, which then evolves into SAD [ 50 , 51 ]. If we pay attention to this kind of medical history in patients, we may be able to prevent it in advance. Overall, the prediction model constructed in this study showed good performance in multiple cohorts and showed certain potential for clinical application. However, the following limitations still need to be considered when interpreting the results and evaluating the practicality. First, although we built a prediction model based on the MIMIC-IV database and conducted external validation in ICU patients of a large tertiary hospital in Northeast China and preliminarily explored the applicability of the model under different ethnic groups and medical systems, the source of the external validation population was relatively single and the sample size was relatively limited, which may limit the generalization ability of the model. Further validation should be carried out in larger, multi-center, and multi-ethnic populations in the future. Second, both the development set and validation set of this study were from retrospective cohorts. Although the efficiency of data acquisition was improved, there were inherent selection biases and confounding factors, which may affect the accuracy of causal inference. In addition, some variables in the MIMIC-IV database were missing. Although we adopted a reasonable missing value filling strategy, the preprocessing process before the variables entered the model may introduce potential bias, thereby affecting the stability and reliability of the model. Third, the current model mainly incorporates seven routine clinical variables and does not include some immune/inflammatory biomarkers (such as C-reactive protein, IL-6, LMR, etc.) that have been confirmed to be closely related to SAD in previous studies. Although this is due to the consideration of data availability and clinical promotion of the model, it may also lead to insufficient identification of certain high-risk factors and limit the further improvement of model performance. In the future, such biomarkers should be considered for inclusion in model construction and optimization. Fourth, although this study has made indirect comparisons with existing models such as PRE-DELIRIC and E-PRE-DELIRIC in the discussion, it has not yet achieved strict horizontal reproduction; that is, these models have not been replicated on the same data set, and performance comparisons have been made. Therefore, our quantitative evaluation of the advantages and disadvantages of different models is still insufficient. Copying the methods of other models (such as score calculation, variable construction logic, verification strategy, etc.) and conducting systematic comparisons on a unified platform is a direction worthy of strengthening in future research. Finally, this study constructed a static prediction model, which has not been prospectively deployed or evaluated for intervention effects in actual clinical workflows, and it is impossible to confirm whether the model has an actual improvement effect on clinical diagnosis and treatment behavior or patient prognosis in real scenarios. Subsequent studies should combine dynamic data update mechanisms to conduct prospective intervention studies to further test the practicality and clinical value of the model. 5 Conclusion This study established and validated a new prediction model for delirium in elderly ICU patients with sepsis. The model showed good performance in both internal and external validation sets, demonstrating its potential for application. Although the overall performance of the model in this study may not necessarily surpass all models, its practicality, interpretability, and cross-regional external validation in a specific subgroup (elderly patients with sepsis) have important complementary value. Declarations Conflict of Interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest . Author Contributions Tengfei Zhou and Xinming Tian conceptualized and designed the study. Tengfei Zhou performed the data analysis and drafted the initial manuscript. Xinming Tian and Wei Wang contributed to data analysis and interpretation. Xuyang Han was responsible for the acquisition and organization of clinical data. Yanhui Feng contributed to data collection. Wenyan Xiong , Chunyan Zhang , and Hongquan Fan provided methodological support and critical feedback during manuscript preparation. Zhe Chu supervised the study, provided critical revisions for statistical analysis and interpretation, and contributed substantially to manuscript revision. All authors reviewed and approved the final manuscript and agree to be accountable for all aspects of the work. Funding No specific grant from any funding agency. Acknowledgments Thanks to all the authors of this article for their contributions. Human Ethics and Consent to Participate declarations This study was conducted in two parts. Modeling was based on the publicly available de-identified database MIMIC-IV v3.1, the use of which was approved by the institutional review boards of Beth Israel Deaconess Medical Center (Boston, MA, USA) and the Massachusetts Institute of Technology (Cambridge, MA, USA). De-identification was completed during the data sharing process, thus eliminating the need for individual informed consent. External validation was based on clinical data from the First Hospital of Jilin University, which was approved by the Institutional Review Committee of the First Hospital of Jilin University (approval number: 2024-1161). Given the retrospective nature of this study and the fact that all patient data were de-identified, the Institutional Review Committee waived written informed consent from patients. Clinical trial number: not applicable. Supplementary Material Supplementary Material should be uploaded separately on submission. Data Availability Statement The mimic-iv database used in this study is available on the official website (https://physionet.org/content/mimiciv/3.1/), and the external validation dataset is available after contacting the corresponding author for consent. The data extraction code and model construction R language code for this study have been uploaded to the website (https://github.com/PAIDAXING-HUP/Frontiers-code). References Singer M, Deutschman CS, Seymour CW, Shankar-hari M, Annane D, Bauer M, Bellomo R, Bernard GR, Chiche J, Coopersmith CM, et al. The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3). JAMA (2016) 315:801–810. doi: 10.1001/jama.2016.0287 Kristina E. R, Sarah Charlotte J, Kareha M. A, Katya Anne S, Derrick T, Daniel Rhodes K, Danny V. C, Kevin S. I, Niranjan K, Simon F, et al. Global, regional, and national sepsis incidence and mortality, 1990-2017: analysis for the Global Burden of Disease Study. Lancet (2020) 395:200–211. doi: 10.1016/S0140-6736(19)32989-7 La Via L, Giuseppe S, Stefania S, Andrea M, Giuseppe N, Salvatore C, Ignazio LM, Bruno C, Stefano S, Serena S, et al. The Global Burden of Sepsis and Septic Shock. Epidemiologia (Basel) (2024) 5:456–478. doi: 10.3390/epidemiologia5030032 Cavallazzi R, Saad M, Marik PE. Delirium in the ICU: an overview. Annals of Intensive Care (2012) 2:49. doi: 10.1186/2110-5820-2-49 Inouye SK, Westendorp RGJ, Saczynski JS. Delirium in elderly people. Lancet (2014) 383:911–922. doi: 10.1016/S0140-6736(13)60688-1 Maldonado JR. Delirium pathophysiology: An updated hypothesis of the etiology of acute brain failure. International Journal of Geriatric Psychiatry (2018) 33:1428–1457. doi: 10.1002/gps.4823 Tsuruta R, Oda Y. A clinical perspective of sepsis-associated delirium. Journal of Intensive Care (2016) 4:18. doi: 10.1186/s40560-016-0145-4 Kim Y, Jin Y, Jin T, Lee S-M. Risk factors and outcomes of sepsis-associated delirium in intensive care unit patients: A secondary data analysis. Intensive and Critical Care Nursing (2020) 59:102844. doi: 10.1016/j.iccn.2020.102844 Brück E, Schandl A, Bottai M, Sackey P. The impact of sepsis, delirium, and psychological distress on self-rated cognitive function in ICU survivors—a prospective cohort study. Journal of Intensive Care (2018) 6:2. doi: 10.1186/s40560-017-0272-6 Td G, Jl T, Pp P, Ne B, Jc J, Mb P, Cg H, R C, Bt P, Lm B, et al. Clinical phenotypes of delirium during critical illness and severity of subsequent long-term cognitive impairment: a prospective cohort study. The Lancet Respiratory medicine (2018) 6: doi: 10.1016/S2213-2600(18)30062-6 Alıcı Ş, Öztürk Birge A. The frequency of sepsis‐associated delirium in intensive care unit and its effect on nurse workload. J Clin Nurs (2025) 34:1383–1397. doi: 10.1111/jocn.17298 McPherson JA, Wagner CE, Boehm LM, Hall JD, Johnson DC, Miller LR, Burns KM, Thompson JL, Shintani AK, Ely EW, et al. Delirium in the cardiovascular ICU: exploring modifiable risk factors. Crit Care Med (2013) 41:405–413. doi: 10.1097/CCM.0b013e31826ab49b Dziegielewski C, Skead C, Canturk T, Webber C, Fernando SM, Thompson LH, Foster M, Ristovic V, Lawlor PG, Chaudhuri D, et al. Delirium and Associated Length of Stay and Costs in Critically Ill Patients. Critical Care Research and Practice (2021) 2021:6612187. doi: 10.1155/2021/6612187 Ahmed S, Leurent B, Sampson EL. Risk factors for incident delirium among older people in acute hospital medical units: a systematic review and meta-analysis. Age and Ageing (2014) 43:326–333. doi: 10.1093/ageing/afu022 Zheng G, Yan J, Li W, Chen Z. Frailty as an independent risk factor for sepsis-associated delirium: a cohort study of 11,740 older adult ICU patients. Aging Clin Exp Res (2025) 37:52. doi: 10.1007/s40520-025-02956-2 Bruno RR, Wernly B, Mamandipoor B, Rezar R, Binnebössel S, Baldia PH, Wolff G, Kelm M, Guidet B, De Lange DW, et al. ICU-Mortality in Old and Very Old Patients Suffering From Sepsis and Septic Shock. Front Med (2021) 8: doi: 10.3389/fmed.2021.697884 Ely EW, Margolin R, Francis J, May L, Truman B, Dittus R, Speroff T, Gautam S, Bernard GR, Inouye SK. Evaluation of delirium in critically ill patients: validation of the Confusion Assessment Method for the Intensive Care Unit (CAM-ICU). Crit Care Med (2001) 29:1370–1379. doi: 10.1097/00003246-200107000-00012 Chen T-J, Chung Y-W, Chang H-C (Rita), Chen P-Y, Wu C-R, Hsieh S-H, Chiu H-Y. Diagnostic accuracy of the CAM-ICU and ICDSC in detecting intensive care unit delirium: A bivariate meta-analysis. International Journal of Nursing Studies (2021) 113:103782. doi: 10.1016/j.ijnurstu.2020.103782 Kuczmarska A, Ngo LH, Guess J, O’Connor MA, Branford-White L, Palihnich K, Gallagher J, Marcantonio ER. Detection of Delirium in Hospitalized Older General Medicine Patients: A Comparison of the 3D-CAM and CAM-ICU. J Gen Intern Med (2016) 31:297–303. doi: 10.1007/s11606-015-3514-0 Wojtara M, Rana E, Rahman T, Khanna P, Singh H. Artificial intelligence in rare disease diagnosis and treatment. Clin Transl Sci (2023) 16:2106–2111. doi: 10.1111/cts.13619 van den Boogaard M, Pickkers P, Slooter AJC, Kuiper MA, Spronk PE, van der Voort PHJ, van der Hoeven JG, Donders R, van Achterberg T, Schoonhoven L. Development and validation of PRE-DELIRIC (PREdiction of DELIRium in ICu patients) delirium prediction model for intensive care patients: observational multicentre study. BMJ (2012) 344:e420. doi: 10.1136/bmj.e420 van den Boogaard M, Schoonhoven L, Maseda E, Plowright C, Jones C, Luetz A, Sackey PV, Jorens PG, Aitken LM, van Haren FMP, et al. Recalibration of the delirium prediction model for ICU patients (PRE-DELIRIC): a multinational observational study. Intensive Care Med (2014) 40:361–369. doi: 10.1007/s00134-013-3202-7 Wassenaar A, van den Boogaard M, van Achterberg T, Slooter AJC, Kuiper MA, Hoogendoorn ME, Simons KS, Maseda E, Pinto N, Jones C, et al. Multinational development and validation of an early prediction model for delirium in ICU patients. Intensive Care Med (2015) 41:1048–1056. doi: 10.1007/s00134-015-3777-2 Shi X, Yang L, Bai W, Jing L, Qin L. Evaluating early lymphocyte-to-monocyte ratio as a predictive biomarker for delirium in older adult patients with sepsis: insights from a retrospective cohort analysis. Front Med (2024) 11:1342568. doi: 10.3389/fmed.2024.1342568 Gu Q, Yang S, Fei D, Lu Y, Yu H. A nomogram for predicting sepsis-associated delirium: a retrospective study in MIMIC III. BMC Medical Informatics and Decision Making (2023) 23:184. doi: 10.1186/s12911-023-02282-5 Zhang Y, Hu J, Hua T, Zhang J, Zhang Z, Yang M. Development of a machine learning-based prediction model for sepsis-associated delirium in the intensive care unit. Sci Rep (2023) 13:12697. doi: 10.1038/s41598-023-38650-4 Yu S, Pan X, Zhang M, Zhang J, Chen D. A machine learning-based prediction model for sepsis-associated delirium in intensive care unit patients with sepsis-associated acute kidney injury. Ren Fail 47:2514186. doi: 10.1080/0886022X.2025.2514186 Fu J, He A, Wang L, Li X, Yu J, Zheng R. Interpretable machine learning model for predicting delirium in patients with sepsis: a study based on the MIMIC data. BMC Infect Dis (2025) 25:1–13. doi: 10.1186/s12879-025-10982-8 Pollard TJ, Johnson AEW, Raffa JD, Celi LA, Mark RG, Badawi O. The eICU Collaborative Research Database, a freely available multi-center database for critical care research. Sci Data (2018) 5:1–13. doi: 10.1038/sdata.2018.178 Alistair Johnson, Tom Pollard, a-chahin, Jim Blundell, Brian Gow, erinhong, Michael Schubert, Kien Dang, Nicolas Paris, shu98, et al. MIT-LCP/mimic-code: v3.0.1. (2025) doi: 10.5281/ZENODO.821871 Kim SE, Ko R-E, Na SJ, Chung CR, Choi KH, Kim D, Park TK, Lee JM, Song YB, Choi J-O, et al. External validation and comparison of two delirium prediction models in patients admitted to the cardiac intensive care unit. Front Cardiovasc Med (2022) 9:947149. doi: 10.3389/fcvm.2022.947149 Hsiao W-L, Hsieh P-Y, Hsu L-F, Lai Y-H, Wang M-L, Yeh Y-C. Effectiveness of a delirium risk assessment and multidisciplinary care approach in reducing delirium incidence among surgical intensive care unit patients: A retrospective pre-post intervention study. Journal of Clinical Anesthesia (2024) 98:111566. doi: 10.1016/j.jclinane.2024.111566 Weng L, Xu Y, Yin P, Wang Y, Chen Y, Liu W, Li S, Peng J, Dong R, Hu X, et al. National incidence and mortality of hospitalized sepsis in China. Crit Care (2023) 27:84. doi: 10.1186/s13054-023-04385-x Johnson AEW, Bulgarelli L, Shen L, Gayles A, Shammout A, Horng S, Pollard TJ, Hao S, Moody B, Gow B, et al. MIMIC-IV, a freely accessible electronic health record dataset. Sci Data (2023) 10: doi: 10.1038/s41597-022-01899-x Collins GS, Reitsma JB, Altman DG, Moons KGM. Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): the TRIPOD statement. Ann Intern Med (2015) 162:55–63. doi: 10.7326/M14-0697 Zhang Z, Rousson V, Lee W-C, Ferdynus C, Chen M, Qian X, Guo Y. Decision curve analysis: a technical note. Ann Transl Med (2018) 6:308. doi: 10.21037/atm.2018.07.02 Buuren S van, Groothuis-Oudshoorn K. mice: Multivariate Imputation by Chained Equations in R. Journal of Statistical Software (2011) 45:1–67. doi: 10.18637/jss.v045.i03 Harrell, FE. Regression Modeling Strategies: With Applications to Linear Models, Logistic and Ordinal Regression, and Survival Analysis . Cham: Springer International Publishing. (2015). doi: 10.1007/978-3-319-19425-7 Robin X, Turck N, Hainard A, Tiberti N, Lisacek F, Sanchez J-C, Müller M. pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics (2011) 12:77. doi: 10.1186/1471-2105-12-77 Lalloué B, Deguen S, Monnez J-M, Padilla C, Kihal W, Zmirou-Navier D, Le Meur N. SesIndexCreatoR: An R Package for Socioeconomic Indices Computation and Visualization. OJS (2015) 05:291–302. doi: 10.4236/ojs.2015.54031 Corrplot: Visualization of a Correlation Matrix. https://taiyun.r-universe.dev/corrplot Evans L, Rhodes A, Alhazzani W, Antonelli M, Coopersmith CM, French C, Machado FR, Mcintyre L, Ostermann M, Prescott HC, et al. Surviving sepsis campaign: international guidelines for management of sepsis and septic shock 2021. Intensive Care Med (2021) 47:1181–1247. doi: 10.1007/s00134-021-06506-y Pham T, Heunks L, Bellani G, Madotto F, Aragao I, Beduneau G, Goligher EC, Grasselli G, Laake JH, Mancebo J, et al. Weaning from mechanical ventilation in intensive care units across 50 countries (WEAN SAFE): a multicentre, prospective, observational cohort study. The Lancet Respiratory Medicine (2023) 11:465–476. doi: 10.1016/S2213-2600(22)00449-0 van Gelder TG, van Diem-Zaal IJ, Dijkstra-Kersten SMA, de Mul N, Lalmohamed A, Slooter AJC. The risk of delirium after sedation with propofol or midazolam in intensive care unit patients. Br J Clin Pharmacol (2024) 90:1471–1479. doi: 10.1111/bcp.16031 Hughes CG, Mailloux PT, Devlin JW, Swan JT, Sanders RD, Anzueto A, Jackson JC, Hoskins AS, Pun BT, Orun OM, et al. Dexmedetomidine or Propofol for Sedation in Mechanically Ventilated Adults with Sepsis. N Engl J Med (2021) 384:1424–1436. doi: 10.1056/NEJMoa2024922 Kawazoe Y, Miyamoto K, Morimoto T, Yamamoto T, Fuke A, Hashimoto A, Koami H, Beppu S, Katayama Y, Itoh M, et al. Effect of Dexmedetomidine on Mortality and Ventilator-Free Days in Patients Requiring Mechanical Ventilation With Sepsis: A Randomized Clinical Trial. JAMA (2017) 317:1321–1328. doi: 10.1001/jama.2017.2088 Yamamoto T, Mizobata Y, Kawazoe Y, Miyamoto K, Ohta Y, Morimoto T, Yamamura H. Incidence, risk factors, and outcomes for sepsis-associated delirium in patients with mechanical ventilation: A sub-analysis of a multicenter randomized controlled trial. J Crit Care (2020) 56:140–144. doi: 10.1016/j.jcrc.2019.12.018 Kim G, Oh DK, Lee SY, Park MH, Lim C-M, the Korean Sepsis Alliance (KSA) investigators. Impact of the timing of invasive mechanical ventilation in patients with sepsis: a multicenter cohort study. Critical Care (2024) 28:297. doi: 10.1186/s13054-024-05064-1 Tan DJ, Chen J, Zhou Y, Ong JSQ, Sin RJX, Bui TV, Mehta AA, Feng M, See KC. Association of body temperature and mortality in critically ill patients: an observational study using two large databases. European Journal of Medical Research (2024) 29:33. doi: 10.1186/s40001-023-01616-3 Mitasova A, Kostalova M, Bednarik J, Michalcakova R, Kasparek T, Balabanova P, Dusek L, Vohanka S, Ely EW. Poststroke delirium incidence and outcomes: validation of the Confusion Assessment Method for the Intensive Care Unit (CAM-ICU). Crit Care Med (2012) 40:484–490. doi: 10.1097/CCM.0b013e318232da12 Frei AI, Wagner AS, Baumann SM, Grzonka P, Berger S, Hunziker S, Rüegg S, Marsch S, Sutter R. Concurrence of seizures and peri-ictal delirium in the critically ill - its frequency, associated characteristics, and outcomes. J Neurol (2024) 271:231–240. doi: 10.1007/s00415-023-11944-3 Additional Declarations No competing interests reported. Supplementary Files Appendixtable2.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 07 Jan, 2026 Reviewers agreed at journal 28 Dec, 2025 Reviewers invited by journal 02 Oct, 2025 Editor invited by journal 09 Sep, 2025 Editor assigned by journal 23 Aug, 2025 Submission checks completed at journal 23 Aug, 2025 First submitted to journal 15 Aug, 2025 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-7380480","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":528977143,"identity":"ce201753-4421-4c25-a361-ba66cceaf344","order_by":0,"name":"Tengfei Zhou","email":"","orcid":"","institution":"The First Hospital of Jilin University","correspondingAuthor":false,"prefix":"","firstName":"Tengfei","middleName":"","lastName":"Zhou","suffix":""},{"id":528977144,"identity":"e25871d1-84d1-4bea-a63f-2bfdf22be20a","order_by":1,"name":"Xinming Tian","email":"","orcid":"","institution":"Jilin University","correspondingAuthor":false,"prefix":"","firstName":"Xinming","middleName":"","lastName":"Tian","suffix":""},{"id":528977145,"identity":"29512be2-2406-4631-929a-472ced61a483","order_by":2,"name":"Xuyang Han","email":"","orcid":"","institution":"The First Hospital of Jilin University","correspondingAuthor":false,"prefix":"","firstName":"Xuyang","middleName":"","lastName":"Han","suffix":""},{"id":528977146,"identity":"eb0e8054-f5a5-4cec-92b9-4663d4ab69df","order_by":3,"name":"Wei Wang","email":"","orcid":"","institution":"First Hospital of Jilin University","correspondingAuthor":false,"prefix":"","firstName":"Wei","middleName":"","lastName":"Wang","suffix":""},{"id":528977147,"identity":"3a24215a-80e3-470b-8a19-7966ad557c55","order_by":4,"name":"Yanhui Feng","email":"","orcid":"","institution":"First Hospital of Jilin University","correspondingAuthor":false,"prefix":"","firstName":"Yanhui","middleName":"","lastName":"Feng","suffix":""},{"id":528977148,"identity":"8e045b5a-aa33-4351-9ebb-36b315396444","order_by":5,"name":"Wenyan Xiong","email":"","orcid":"","institution":"First Hospital of Jilin University","correspondingAuthor":false,"prefix":"","firstName":"Wenyan","middleName":"","lastName":"Xiong","suffix":""},{"id":528977149,"identity":"e1362d01-b823-4fdf-b269-e3eed9d63a30","order_by":6,"name":"Chunyan Zhang","email":"","orcid":"","institution":"First Hospital of Jilin University","correspondingAuthor":false,"prefix":"","firstName":"Chunyan","middleName":"","lastName":"Zhang","suffix":""},{"id":528977150,"identity":"4ae1f97d-2fb2-43dd-99f0-8de759838f4b","order_by":7,"name":"Hongquan Fan","email":"","orcid":"","institution":"First Hospital of Jilin University","correspondingAuthor":false,"prefix":"","firstName":"Hongquan","middleName":"","lastName":"Fan","suffix":""},{"id":528977151,"identity":"1db7f098-7fd9-4fe7-a697-2a80633f3318","order_by":8,"name":"Zhe Chu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA5UlEQVRIiWNgGAWjYDACCQYGZgYGGyDJ3MAgUQASSiBKSxqQZARqMSBey2EgC6iFgRgt/LN7jD8XVJyP5m8HarEwOMzAz55jwPBzBx5L7pwxMJ5x5nbujMNghx1mkOx5Y8DYewa3FgOJHINk3rbbuQ0wLQY3cgyYGdvwaznM++9c7nyYFnsitBg28zYcyN0At0WCgBaJG2nFzDzHknM3ArUckDBI55E486zgYC8eLfwzkjd/5qmxy513/vDBxxIV1nL87ckbH/zEowUFHAZGEg+IcYBIDcDI/EC00lEwCkbBKBhJAACyIEw553/4BwAAAABJRU5ErkJggg==","orcid":"","institution":"First Hospital of Jilin University","correspondingAuthor":true,"prefix":"","firstName":"Zhe","middleName":"","lastName":"Chu","suffix":""}],"badges":[],"createdAt":"2025-08-15 10:08:20","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7380480/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7380480/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":93616985,"identity":"d658fd4a-33ca-4d0f-a807-d10d256e4e09","added_by":"auto","created_at":"2025-10-15 17:02:58","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4907198,"visible":true,"origin":"","legend":"","description":"","filename":"BMCgeriatricmanuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/f0c3001a9942577db7a4cfc1.docx"},{"id":93616960,"identity":"2d7be01e-274b-484b-89a2-7593a6e193ee","added_by":"auto","created_at":"2025-10-15 17:02:58","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":10931,"visible":true,"origin":"","legend":"","description":"","filename":"d6b56d926dc4428b97e90698c412407f.json","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/ca6109e870f600046d28856a.json"},{"id":93616961,"identity":"56b7edbf-b624-4620-8a1a-2461f10fba54","added_by":"auto","created_at":"2025-10-15 17:02:58","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":84664,"visible":true,"origin":"","legend":"","description":"","filename":"Appendixtable2.docx","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/e5f083761a8be9b3b31feecd.docx"},{"id":93617713,"identity":"34cd45a6-c0f5-48d2-82c7-55e0096255d1","added_by":"auto","created_at":"2025-10-15 17:10:58","extension":"xml","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":179479,"visible":true,"origin":"","legend":"","description":"","filename":"d6b56d926dc4428b97e90698c412407f1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/db9fb342a11f7d55a5c54253.xml"},{"id":93618922,"identity":"24653723-a870-4fb1-badc-5bc32bdc36a2","added_by":"auto","created_at":"2025-10-15 17:26:58","extension":"jpeg","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":539204,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/1c708b6695293fe6e7f20ee2.jpeg"},{"id":93616970,"identity":"a211eb6c-56be-4ecb-a9d1-faaff9a95a47","added_by":"auto","created_at":"2025-10-15 17:02:58","extension":"jpeg","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":522280,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/6b34ce134802fc6588c7cd0d.jpeg"},{"id":93617716,"identity":"ee739943-c14f-4a8f-9858-d32baa92de35","added_by":"auto","created_at":"2025-10-15 17:10:58","extension":"jpeg","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":496888,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/d16eb40b9cc661bbe0c1f002.jpeg"},{"id":93617722,"identity":"e18362ee-b28f-4c30-a663-d5bc0c08c97a","added_by":"auto","created_at":"2025-10-15 17:10:58","extension":"jpeg","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":511428,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/aef93b3c44c36567fc719f03.jpeg"},{"id":93616972,"identity":"7b200dbb-a2ee-4372-8983-2d855928c8f7","added_by":"auto","created_at":"2025-10-15 17:02:58","extension":"jpeg","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":643716,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/6b1c9d2bf1e448c0334d735d.jpeg"},{"id":93617720,"identity":"6acbb033-2f23-405c-9a90-82edfd7282ff","added_by":"auto","created_at":"2025-10-15 17:10:58","extension":"jpeg","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1347348,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/c6b323010c7b75b2d85506a7.jpeg"},{"id":93616974,"identity":"4519451a-805d-4621-ad22-a09cae39bff4","added_by":"auto","created_at":"2025-10-15 17:02:58","extension":"jpeg","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":682908,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/d368bce1f7e707b403fc39ab.jpeg"},{"id":93617723,"identity":"3173168f-0223-489a-a1b0-cdc1c2e3bc71","added_by":"auto","created_at":"2025-10-15 17:10:58","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":110877,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/b19266e32e2ef41816031304.png"},{"id":93617725,"identity":"3490d7cd-aba2-49f3-b51e-9013b16ab04b","added_by":"auto","created_at":"2025-10-15 17:10:58","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":43451,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/5842bfb9769d5143f62822ce.png"},{"id":93618156,"identity":"c2373284-2ddf-4bd8-8a1c-debb10fae97f","added_by":"auto","created_at":"2025-10-15 17:18:58","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":72955,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/813e1bd418acf1c8953ab105.png"},{"id":93617719,"identity":"2aabc99f-d42f-44c3-a779-866b97e1697b","added_by":"auto","created_at":"2025-10-15 17:10:58","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":46440,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/bd902ec8c2e123cdf249700b.png"},{"id":93618155,"identity":"ac23cb10-2eaf-46a9-b913-9ea80d8b57c4","added_by":"auto","created_at":"2025-10-15 17:18:58","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":90083,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/7aaa0ae388e12f9b8f4f26b5.png"},{"id":93616973,"identity":"f29b4831-46fa-4ed8-b3f1-f67dd2091c2c","added_by":"auto","created_at":"2025-10-15 17:02:58","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":155540,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/5b61534d5b85c89295134047.png"},{"id":93616978,"identity":"50cd94ee-0d5c-45db-8818-5231b9fccd69","added_by":"auto","created_at":"2025-10-15 17:02:58","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":72356,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/23881aa18ea65bdb56e58a20.png"},{"id":93616983,"identity":"32d4a06b-219d-4b29-846a-ac826e514e86","added_by":"auto","created_at":"2025-10-15 17:02:58","extension":"xml","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":180426,"visible":true,"origin":"","legend":"","description":"","filename":"d6b56d926dc4428b97e90698c412407f1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/4aacbdec13a4fdd4fc86f74a.xml"},{"id":93616984,"identity":"d5142a10-c315-4bd2-9cfa-f91d3002693e","added_by":"auto","created_at":"2025-10-15 17:02:58","extension":"html","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":190568,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/089339adcb235754acd62bec.html"},{"id":93618151,"identity":"954bab8b-ad76-4678-825f-d532c514b3f0","added_by":"auto","created_at":"2025-10-15 17:18:58","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":97633,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of patient selection for the study.\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/c8db03ea8dcf1d84ff6a58c0.jpg"},{"id":93616958,"identity":"a5f2d975-0839-4721-9b14-f492c5b2bbf1","added_by":"auto","created_at":"2025-10-15 17:02:58","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":78797,"visible":true,"origin":"","legend":"\u003cp\u003eLASSO Regression for Variable Selection in SAD Prediction. (A) Coefficient trajectories of variables as the penalty parameter (λ) changes. (B) Ten-fold cross-validation plot showing binomial deviance for different λ values.\u003c/p\u003e","description":"","filename":"2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/7b98c6b3684b8fb5f8db7bb7.jpg"},{"id":93616968,"identity":"9d82c46e-ad4b-4c3e-8d24-5db3b897d192","added_by":"auto","created_at":"2025-10-15 17:02:58","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":73631,"visible":true,"origin":"","legend":"\u003cp\u003eMultivariable Logistic Regression Results of LASSO-Selected Predictors for Sepsis-Associated Delirium\u003c/p\u003e","description":"","filename":"3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/8acd5064f6c6177e8b60cdc7.jpg"},{"id":93617711,"identity":"747a9645-4601-47bc-aa0d-1d74e310eabf","added_by":"auto","created_at":"2025-10-15 17:10:58","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":65088,"visible":true,"origin":"","legend":"\u003cp\u003eNomogram for predicting SAD risk in elderly patients in the ICU.\u003c/p\u003e","description":"","filename":"4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/685311263156bfa951940d2b.jpg"},{"id":93617712,"identity":"614689d8-92b6-4694-9b00-5293ddd037e6","added_by":"auto","created_at":"2025-10-15 17:10:58","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":52078,"visible":true,"origin":"","legend":"\u003cp\u003eAUC values in the training, internal, and external validation cohorts.\u003c/p\u003e","description":"","filename":"5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/eb0d9d00bc0d59008525d336.jpg"},{"id":93618920,"identity":"6df64dfb-a59a-4f82-bc86-dc9bfa8b8eb3","added_by":"auto","created_at":"2025-10-15 17:26:58","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":63994,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of calibration curves of the three sets. (A) Training set, (B) Internal validation set, and (C) External validation set.\u003c/p\u003e\n\u003cp\u003eNote: \u003cstrong\u003eDxy\u003c/strong\u003e: Somers’ D rank correlation; \u003cstrong\u003eR²\u003c/strong\u003e: Pseudo-R²; \u003cstrong\u003eD\u003c/strong\u003e: Discrimination Index; \u003cstrong\u003eU:\u003c/strong\u003e Unreliability Index; \u003cstrong\u003eQ\u003c/strong\u003e: Quality Index; \u003cstrong\u003eBrier\u003c/strong\u003e: Brier score; \u003cstrong\u003eIntercept\u003c/strong\u003e: Calibration-in-the-large; \u003cstrong\u003eSlope\u003c/strong\u003e: Calibration slope; \u003cstrong\u003eEmax\u003c/strong\u003e: Maximum absolute difference between predicted and observed probabilities; \u003cstrong\u003eE90\u003c/strong\u003e: 90th percentile of the absolute difference between predicted and observed probabilities; \u003cstrong\u003eEavg\u003c/strong\u003e: Average absolute difference between predicted and observed probabilities; \u003cstrong\u003eS:z\u003c/strong\u003e: Standardized z-score for Spiegelhalter's z-test; \u003cstrong\u003eS:p\u003c/strong\u003e: P-value for Spiegelhalter’s test.\u003c/p\u003e","description":"","filename":"6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/1d9ae3d0a215ad13c4e0df9a.jpg"},{"id":93619191,"identity":"f3bc6ef2-b605-432f-9091-07b1dbfaa03a","added_by":"auto","created_at":"2025-10-15 17:34:58","extension":"jpg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":59689,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of decision curves analysis (DCA) for the three sets. (A) Training set, (B) Internal validation set, and (C) External validation set.\u003c/p\u003e","description":"","filename":"7.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/77bcf35a71f10917d6bb961b.jpg"},{"id":93681928,"identity":"158a1d3b-e571-4e4f-931c-7b36cbc07d55","added_by":"auto","created_at":"2025-10-16 12:29:13","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1869399,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/3e042c9c-b59a-4fb8-b6e4-de8c6c602b66.pdf"},{"id":93616957,"identity":"df4d3474-5470-4321-a146-9268466ab993","added_by":"auto","created_at":"2025-10-15 17:02:58","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":84664,"visible":true,"origin":"","legend":"","description":"","filename":"Appendixtable2.docx","url":"https://assets-eu.researchsquare.com/files/rs-7380480/v1/702dff12ad9ad7d0d8d7f5c4.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"External validation in a Chinese cohort of a nomogram developed using MIMIC-IV for predicting sepsis-associated delirium in elderly ICU patients","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eSepsis is a clinical syndrome of immune dysregulation caused by infection, which can lead to serious consequences such as septic shock and multiple organ failure (MODS) [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Annually, approximately 48.9\u0026nbsp;million new sepsis cases occur worldwide, representing about 0.7% of the global population and resulting in an estimated 11\u0026nbsp;million deaths, accounting for nearly 19.7% of all global deaths during that timeframe [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e].Delirium is an acute neurocognitive disorder that is costly, serious, and life-threatening, affecting approximately 50% of hospitalized elderly patients [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Sepsis-associated delirium (SAD) is a common neuropsychiatric complication in elderly ICU patients, who usually have a poor prognosis [\u003cspan additionalcitationids=\"CR7\" citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. The reported prevalence of SAD varies greatly across studies, generally ranging from 9\u0026ndash;76.9%, with an average of approximately 42.9%. SAD is significantly associated with several risk factors, including elevated Sequential Organ Failure Assessment (SOFA) scores, increased lactate levels, hyperglycemia, and frailty [\u003cspan additionalcitationids=\"CR10\" citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Studies have shown that the risk of in-hospital mortality in patients with SAD is approximately twice that of sepsis patients without delirium, and the duration of mechanical ventilation and hospital stay is significantly prolonged, leading to a 1.5- to 2-fold increase in healthcare costs [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. The complex interaction between systemic organ failure and sepsis makes the clinical management of this population very complicated [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Given that SAD is challenging to reverse, medical professionals must prevent and address it in time.\u003c/p\u003e\u003cp\u003eAs the global population ages, the proportion of elderly patients in the ICU continues to rise, and sepsis management in this population has gradually become a clinical priority [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. Immune senescence and impaired brain regulatory function render older individuals with sepsis more susceptible to delirium. Among the widely used delirium screening tools, the Confusion Assessment Method for the ICU (CAM-ICU) is the most common [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Meta-analysis shows that CAM-ICU has a high recognition efficiency in general hospitalized patients and has the highest sensitivity and specificity among various screening tools [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. However, due to the complex clinical manifestations and multiple comorbidities of elderly patients with sepsis in the ICU, there is a certain risk of missed diagnosis when using CAM-ICU. Current research has found that the sensitivity of CAM-ICU for detecting delirium in elderly hospitalized patients is only 53%, and the sensitivity for patients with moderate delirium is even lower, at 30%, which seriously restricts the early identification of SAD in clinical practice [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. Consequently, it is imperative to thoroughly examine the clinical characteristics of the elderly sepsis cohort to facilitate early identification and targeted care for high-risk patients.\u003c/p\u003e\u003cp\u003eWith the development of precision medicine, clinical prediction models related to delirium are becoming increasingly important in risk assessment, decision making, and personalized care [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Several delirium prediction models have been proposed, among which PRE-DELIRIC and E-PRE-DELIRIC are representative models. The PRE-DELIRIC model is based on 10 variables collected within 24 hours after admission to a multicenter ICU in the Netherlands, including age, APACHE-II score, coma, sedative or morphine use, and emergency admission [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. The model's AUC reached 0.87 (95% CI: 0.85\u0026ndash;0.89), which was significantly better than the judgment of doctors and nurses (0.59; 95% CI: 0.49\u0026ndash;0.70). The model was first prospectively validated in the time dimension, with an AUC of 0.89 (95% CI: 0.86\u0026ndash;0.92), a calibration slope of 1.2, and an intercept of 0.22, showing good fit. Subsequently, an international multicenter external validation was conducted in eight ICUs in six European countries. The combined AUC was 0.77 (95% CI: 0.74\u0026ndash;0.79), the Hosmer-Lemeshow test P\u0026thinsp;=\u0026thinsp;0.045, and the calibration was good [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. However, the original authors pointed out that the model should be used with caution in populations with a high incidence of delirium. If it is not recalibrated, there may be a bias of overestimating the risk [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. The E-PRE-DELIRIC model integrates data from thirteen ICUs in seven countries, taking into account both the generality and stability of the model. Its development set AUC was 0.76 (95% CI: 0.73\u0026ndash;0.77) and the validation set AUC was 0.75 (95% CI: 0.71\u0026ndash;0.79), showing good calibration ability [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. In addition, the model can also predict the risk of delirium on the second and sixth days of ICU hospitalization (AUCs of 0.70 and 0.81, respectively), showing certain advantages in temporal dynamic prediction. The researchers also pointed out that the E-PRE-DELIRIC model can show good fit in different regions without recalibration. As our understanding of delirium continues to deepen, there are many other models that deserve our attention. Shi et al. [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e] explored the predictive value of biomarkers based on the lymphocyte-to-monocyte ratio (LMR) and found that LMR was significantly negatively correlated with SAD. The model AUC was significantly improved after the introduction of this indicator (P\u0026thinsp;=\u0026thinsp;0.035). After the introduction of LMR, the AUC value of the model increased steadily, and the difference was statistically significant (P\u0026thinsp;=\u0026thinsp;0.035), suggesting that it has certain predictive potential. Gu et al. [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] constructed a regression model based on the MIMIC-III database, including SOFA score, mechanical ventilation, phosphate, and lactate. The training set AUC was 0.742, the validation set AUC was 0.713, and the Hosmer-Lemeshow test P\u0026thinsp;=\u0026thinsp;0.471. The calibration was good, and the decision curve analysis also suggested that it has practical value.\u003c/p\u003e\u003cp\u003eMachine learning algorithms have developed rapidly in recent years and have unique advantages in processing complex data. Zhang et al. [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] and Yu et al. [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e] constructed seven machine learning models based on the MIMIC-IV database and performed external validation in the eICU-CRD database. The AUCs of the XGBoost model in the modeling set were 0.793 and 0.775, and those in the validation set were 0.701 and 0.687. The calibration curve and decision curve were drawn for each model. The results showed that XGBoost performed best and had certain generalization ability and potential application value. In addition, the variable importance ranking results also showed that \"mechanical ventilation\" was the factor most strongly associated with delirium (SHAP values were 0.593 and 0.663, respectively). Fu et al. [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e] used a machine learning algorithm to construct a nomogram containing four risk factors and also found that the XGBoost model had the best predictive performance. The AUC of XGBoost was 0.79 (95% CI: 0.75\u0026ndash;0.83), but the model has not been externally validated, and its generalization ability remains to be examined.\u003c/p\u003e\u003cp\u003eIt is worth noting that the above models are all constructed and verified based on clinical databases of European and American populations, especially MIMIC-IV and eICU-CRD, which are both American hospital data, and their population characteristics may be different from those in China or Asia [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Some studies have attempted to introduce these models into the Asian context. Kim et al. [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e] collected data from patients in the cardiac intensive care unit (CICU) of Samsung Medical Center in South Korea and verified the applicability of the PRE-DELIRIC and E-PRE-DELIRIC models in CICU patients. The results showed that the PRE-DELIRIC model had better prediction results, with an AUC of 0.84 (95% CI: 0.82\u0026ndash;0.86), and showed good generalization ability in Asian populations. Hsiao et al. [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e] conducted an intervention study based on the PRE-DELIRIC model in a tertiary hospital in Taipei, China. The model was evaluated in combination with SMART bundle care. The incidence of delirium in the intervention group was significantly lower than that in the control group (22.3% vs. 47.7%), proving that the model is feasible and effective in clinical practice in Asia. These studies provide preliminary evidence for the cross-regional applicability of delirium prediction models in Asian populations. However, whether it can be extended to a wider range of high-risk populations still needs further exploration and verification in combination with regional data.\u003c/p\u003e\u003cp\u003eChina, one of the most populous countries in the world, is facing an increasingly severe challenge of an aging population. A nationwide population-based study estimated that there were approximately 4.8\u0026nbsp;million to 6.1\u0026nbsp;million hospitalized sepsis cases in China each year from 2017 to 2019, 57.5% of which occurred in people aged 65 years and above [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. Considering that the elderly are a high-risk group for SAD, this study, based on the construction of a prediction model using the MIMIC-IV database, further introduced independent clinical data from an Asian population (from the ICU of a large tertiary hospital in China) as an external validation cohort. By integrating international and local data, the model aims to complement existing tools and enhance early identification and personalized management of high-risk patients.\u003c/p\u003e"},{"header":"2 Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.1 Research methodology and data sources\u003c/h2\u003e\u003cp\u003eThis study employed a retrospective cohort design. Model development was based on data from the Medical Information Mart for Intensive Care IV (MIMIC-IV, version 3.1), a publicly available database that includes detailed clinical information on ICU patients admitted to large medical centers in the United States between 2008 and 2022 [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. The principal investigator (ZTF) has completed the CITI ethics examination program and has a valid certification (No: 63365666). Elderly ICU patients aged\u0026thinsp;\u0026ge;\u0026thinsp;65 years with a diagnosis of sepsis were identified from the MIMIC-IV database. Eligible patients were randomly assigned to a training set (60%) and an internal validation set (40%) using a random number generator based on terminal digits (0\u0026ndash;9). The external validation cohort was derived from the central intensive care unit of a tertiary hospital in China, comprising patients admitted between January 2019 and November 2024. Ethical approval for the use of clinical data was obtained from the Institutional Review Board of the First Affiliated Hospital of Jilin University (Approval No: 2024\u0026thinsp;\u0026minus;\u0026thinsp;1161). This study complied with the ethical principles of the Declaration of Helsinki and followed the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) guideline for predictive modeling [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e].\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e2.2 Patient selection\u003c/h2\u003e\u003cp\u003ePatients admitted to the ICU for the first time with a confirmed diagnosis of sepsis were selected from the MIMIC-IV database. All enrolled patients had a documented delirium assessment. The inclusion and exclusion criteria for the external validation cohort were consistent with those of the MIMIC-IV dataset. Patients were excluded based on the following criteria: (1) Age\u0026thinsp;\u0026lt;\u0026thinsp;65 years; (2) ICU stay\u0026thinsp;\u0026lt;\u0026thinsp;48 hours; (3) Absence of CAM-ICU evaluation or lack of documentation of delirium; (4) Presence of conditions that may confound the diagnosis of delirium, including dementia, epilepsy, schizophrenia, traumatic brain injury, meningitis, brain tumor, alcohol abuse, and drug poisoning.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e\u003cem\u003e2.3 Outcome definition\u003c/em\u003e\u003c/h2\u003e\u003cp\u003eThe primary outcome was the occurrence of delirium during the ICU stay of the patients involved.\u003c/p\u003e\u003cp\u003e(1) Sepsis-3.0: Clinically confirmed life-threatening organ dysfunction caused by suspected or confirmed infection and SOFA score\u0026thinsp;\u0026ge;\u0026thinsp;2 points [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e].\u003c/p\u003e\u003cp\u003e(2) Delirium: We used the CAM-ICU score [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e] or the medical staff's delirium diagnosis record to determine whether the patient had delirium. The \u0026lsquo;chartevents\u0026rsquo; table in the MIMIC-IV database was combined to identify the occurrence of delirium in ICU patients. For CAM-ICU score assessment, we extracted \u0026lsquo;itemid\u0026rsquo; including 229326, 228337, 229325, 228334, 228336, 228303, 229324, 228300, 228335, 228302, and 228301. The result in the value field was used for judgment. A record of \u0026lsquo;Yes\u0026rsquo; was defined as positive for delirium, a record of \u0026lsquo;No\u0026rsquo; was defined as negative, and a record of \u0026lsquo;Unable to assess\u0026rsquo; was considered as missing. The CAM-ICU comprises four components: ① abrupt onset of cognitive alterations or a fluctuating trajectory; ② inattention; ③ disorganized thought functions; ④ altered awareness level. A positive diagnosis of CAM-ICU was confirmed if the criteria ① + ② + ③ or ① + ② + ④ were satisfied. For the medical staff assessment records, the extracted \u0026lsquo;itemid\u0026rsquo; is 228332, where the value of \u0026lsquo;Positive\u0026rsquo; is considered positive, \u0026lsquo;Negative\u0026rsquo; is considered negative, and \u0026lsquo;UTA\u0026rsquo; is a missing outcome. Patients were excluded if both sources (CAM-ICU and staff notes) were missing or if delirium occurred prior to ICU admission or within the first 24 hours of ICU stay.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\u003ch2\u003e2.4 Variables and data extraction\u003c/h2\u003e\u003cp\u003eThe prediction model cohort was derived by linking the MIMIC-IV (v3.1) database in PostgreSQL utilizing Navicat Premium 15.0. We obtained the same variable characteristics as the modeling cohort from the clinical electronic medical records of the central ICU of our hospital. For continuous variables (including vital signs and laboratory parameters) recorded within 24 hours after ICU admission, we extracted statistical features based on clinical significance and data characteristics. Specifically, the minimum, maximum, or mean value was extracted for each variable, with the choice depending on clinical interpretability and the proportion of missing data. Categorical variables (e.g., sex, comorbidities, medications, and interventions) were coded as binary indicators. The following data were extracted: (1) Demographic variables: age, gender, weight, admission type; (2) Vital signs within 24 hours of admission: temperature mean (T), heart rate mean (HR), respiratory rate mean (RR), SBP mean, DBP mean, MBP mean, partial pressure of oxygen (pO\u003csub\u003e2\u003c/sub\u003e) max, partial pressure of carbon dioxide (pCO\u003csub\u003e2\u003c/sub\u003e) max; (3) Comorbidity and score: hypertension, diabetes, cancer, respiratory failure, renal failure, neurological disease, sequential organ failure assessment score (SOFA); (4) Biochemistry within 24 hours of admission: hemoglobin min, blood urea nitrogen (BUN) min, calcium min, creatinine min, potassium min, international normalized ratio (INR) min, prothrombin time (PT) min, partial thromboplastin time (PTT) min, pH min, lactate min, platelets max, WBC max, sodium max; (5) Whether treatment was received within 24 hours of admission: continuous renal replacement therapy (CRRT) and the application of mechanical ventilation; (6) Whether to receive medication within 24 hours after admission: vasopressin, midazolam, fentanyl.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003e2.5 Statistical methods and model construction\u003c/h2\u003e\u003cp\u003eAll statistical analyses were performed using R software (version 4.3.3). Multiple imputation was performed using the \u0026ldquo;mice\u0026rdquo; package in R for variables with a missing proportion of less than 20%, and variables with a missing proportion of more than 20% were deleted (see \u003cb\u003eSupplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003c/b\u003e for details). Descriptive statistical analysis was performed before modeling. We used the Shapiro-Wilk test to see if all continuous variables were normally distributed. The results showed that all were non-normally distributed (P\u0026thinsp;\u0026lt;\u0026thinsp;0.05). Therefore, continuous variables were summarized as medians and interquartile ranges (IQRs), and intergroup differences were assessed using the Kruskal-Wallis test. Categorical data were reported as frequencies and proportions (n, %), with group comparisons conducted using either Pearson\u0026rsquo;s chi-square test or Fisher\u0026rsquo;s exact test, depending on data distribution. All statistical evaluations were performed as two-tailed tests, with a significance level set at P\u0026thinsp;\u0026lt;\u0026thinsp;0.05.\u003c/p\u003e\u003cp\u003eFor variable selection, we used least absolute shrinkage and selection operator (LASSO) regression, implemented in the glmnet package in R. LASSO regression was applied to the training cohort, penalizing noninformative variables to shrink coefficients and select the most relevant predictors. Ten-fold cross-validation was used to determine the optimal regularization parameter (λ). The predictor set identified by LASSO was ultimately entered into a multivariable logistic regression model to estimate effect sizes and construct a predictive model. Nomograms were constructed using the rms package to visualize the model. Model validation was performed in an internal validation cohort (MIMIC data with a 60/40 randomization split) and an external validation cohort from the ICU of a tertiary hospital in China. The area under the receiver operating characteristic curve (AUC) was calculated using the pROC package to assess the discriminative ability of the model. Calibration was assessed using 1,000 bootstrap resamplings with the rms package, and nonparametric LOESS smoothed calibration curves were plotted to compare predicted and observed probabilities. Model accuracy was quantified using the Brier score, and its 95% confidence interval was estimated using percentile bootstrapping (n\u0026thinsp;=\u0026thinsp;1,000). Furthermore, the Hosmer-Lemeshow goodness-of-fit test was performed to assess the consistency between predicted and observed results. Finally, decision curve analysis (DCA) was performed to evaluate the potential clinical utility of the model [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. This study utilized the following R packages: mice [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e], rms [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e], pROC [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e], Hmisc [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e], and corrplot [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e].\u003c/p\u003e\u003c/div\u003e"},{"header":"3 Results","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\u003ch2\u003e3.1 Patient selection clinical characteristics\u003c/h2\u003e\u003cp\u003ePatients in this study were recruited from two cohorts: the MIMIC-IV database and an intensive care unit (ICU) at a Chinese hospital. The complete patient screening process is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. After rigorous screening, a total of 5,034 elderly ICU patients with sepsis were enrolled. Of these, 2,852 patients were assigned to the training cohort, 1,901 to the internal validation cohort, and 281 to the external validation cohort (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The overall incidence of sepsis-associated delirium (SAD) was 46.44% (2,338/5,034). The incidence in the training cohort was 46.91% (1,338/2,852), the internal validation cohort was 46.29% (880/1,901), and the external validation cohort was 42.70% (120/281). Detailed baseline characteristics and between-group differences are provided in Tables S2-S4. Demographic characteristics differed among the three cohorts, with the median age of the external validation cohort being slightly lower than that of the training and internal validation cohorts (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Similarly, weight was significantly lower in the external validation cohort than in the training and internal validation cohorts (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001). However, no significant differences were observed in gender distribution or hospital admission type. Regarding vital signs, heart rate was higher in the external validation group than in the training group (84 beats/min vs. 80 beats/min, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001), while no significant differences were observed in other vital sign parameters. The external validation group had a higher prevalence of hypertension and diabetes (44.5% and 27.4%, respectively; both P\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Regarding treatment interventions, the external validation cohort had a lower proportion of patients receiving mechanical ventilation and vasopressors (32.7% and 24.6%, respectively, both P\u0026thinsp;\u0026lt;\u0026thinsp;0.001), while there was a higher prevalence of fentanyl use (66.5%, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001). No significant differences were observed in other treatment interventions.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eBaseline characteristics of the study population in the training, internal validation, and external validation sets.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCharacteristic\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTraining set\u003c/p\u003e\u003cp\u003e(N\u0026thinsp;=\u0026thinsp;2,852)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eInternal validation set\u003c/p\u003e\u003cp\u003e(N\u0026thinsp;=\u0026thinsp;1,901)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eExternal validation set\u003c/p\u003e\u003cp\u003e(N\u0026thinsp;=\u0026thinsp;281)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eStatistic\u003csup\u003e1\u003c/sup\u003e\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003ep-value\u003csup\u003e1\u003c/sup\u003e\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colspan=\"6\" nameend=\"c6\" namest=\"c1\"\u003e\u003cp\u003eDemographic variables\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAge, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e77 (71, 83)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e77 (71, 84)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e74 (70, 80)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e23.12\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGender, n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.33\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.849\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFemale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1,289 (45.2%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e874 (46.0%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e126 (44.8%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1,563 (54.8%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1,027 (54.0%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e155 (55.2%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWeight, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e79 (65, 92)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e79 (66, 92)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e71 (62, 84)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e28.84\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAdmission type, n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.71\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.425\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNon-Emergency\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e885 (31.0%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e620 (32.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e84 (29.9%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eEmergency\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1,967 (69.0%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1,281 (67.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e197 (70.1%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"6\" nameend=\"c6\" namest=\"c1\"\u003e\u003cp\u003e\u003cb\u003eVital signs\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTemperature mean, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e36.90 (36.60, 37.20)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e36.80 (36.60, 37.20)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e36.90 (36.60, 37.30)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.33\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.514\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHeart rate mean, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e84 (74, 95)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e83 (74, 95)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e88 (76, 104)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e18.15\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eResp rate mean, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e19.0 (17.0, 22.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e19.0 (17.0, 22.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e20.0 (17.0, 23.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e5.35\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.069\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSBP mean, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e113 (105, 125)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e113 (105, 125)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e112 (102, 122)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e6.19\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.045\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDBP mean, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e58 (53, 65)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e58 (53, 65)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e59 (52, 68)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.81\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.404\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMBP mean, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e74 (69, 81)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e74 (69, 81)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e74 (69, 81)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.16\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.925\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003epO2 max, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e153 (82, 312)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e150 (85, 324)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e128 (81, 254)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e5.89\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.052\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003epCO2 max, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e45 (40, 53)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e45 (39, 53)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e42 (35, 50)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e27.36\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"6\" nameend=\"c6\" namest=\"c1\"\u003e\u003cp\u003e\u003cb\u003eComorbidity variables\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHypertension, n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e20.29\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1,956 (68.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1,296 (68.2%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e156 (55.5%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e896 (31.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e605 (31.8%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e125 (44.5%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDiabetes, n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e13.97\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2,332 (81.8%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1,532 (80.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e204 (72.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e520 (18.2%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e369 (19.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e77 (27.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCancer, n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e2.09\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.351\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2,242 (78.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1,506 (79.2%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e212 (75.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e610 (21.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e395 (20.8%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e69 (24.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRespiratory failure, n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e6.77\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.034\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2,254 (79.0%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1,522 (80.1%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e206 (73.3%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e598 (21.0%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e379 (19.9%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e75 (26.7%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRenal failure, n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e7.25\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.027\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2,089 (73.2%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1,388 (73.0%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e226 (80.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e763 (26.8%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e513 (27.0%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e55 (19.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNeurological disease, n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.59\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.744\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2,174 (76.2%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1,432 (75.3%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e211 (75.1%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e678 (23.8%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e469 (24.7%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e70 (24.9%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSOFA score, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e5.0 (4.0, 8.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e6.0 (4.0, 8.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e6.0 (4.0, 9.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e2.48\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.290\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"6\" nameend=\"c6\" namest=\"c1\"\u003e\u003cp\u003e\u003cb\u003eBiochemistry variables\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHemoglobin min, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e9.10 (7.90, 10.70)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e9.10 (7.90, 10.80)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e9.60 (8.40, 11.00)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e10.86\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.004\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePlatelets max, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e212 (153, 285)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e202 (150, 274)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e197 (134, 279)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e10.57\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.005\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWBC max, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e13 (10, 19)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e14 (10, 19)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e13 (8, 19)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e3.51\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.173\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBUN min, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e23 (16, 38)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e22 (15, 35)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e25 (23, 29)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e25.88\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCalcium min, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e8.10 (7.60, 8.60)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e8.10 (7.60, 8.60)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e8.00 (7.20, 8.10)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e50.62\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCreatinine min, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1.10 (0.80, 1.60)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1.00 (0.80, 1.60)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1.10 (0.80, 1.70)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e3.04\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.219\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSodium max, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e141.0 (138.0, 144.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e141.0 (138.0, 144.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e141.0 (138.0, 144.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.04\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.981\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePotassium min, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.90 (3.60, 4.30)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3.90 (3.50, 4.30)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e3.90 (3.50, 4.30)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.86\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.395\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eINR min, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1.20 (1.10, 1.40)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1.20 (1.10, 1.40)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1.20 (1.10, 1.40)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.08\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.582\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePT min, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e13.6 (12.2, 15.6)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e13.5 (12.4, 15.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e13.5 (12.3, 15.2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.82\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.665\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePTT min, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e29 (26, 34)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e29 (26, 34)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e29 (27, 34)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.56\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.458\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLactate min, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1.40 (1.00, 1.90)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1.40 (1.00, 1.90)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1.40 (1.00, 2.20)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e2.70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.260\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePH min, Median (Q1, Q3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e7.33 (7.27, 7.39)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e7.33 (7.27, 7.39)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e7.35 (7.27, 7.40)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e2.74\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.255\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"6\" nameend=\"c6\" namest=\"c1\"\u003e\u003cp\u003e\u003cb\u003eTreatment variables\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCRRT, n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e10.78\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.005\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2,666 (93.5%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1,761 (92.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e248 (88.3%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e186 (6.5%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e140 (7.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e33 (11.7%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMechanical ventilation, n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e90.34\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1,113 (39.0%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e726 (38.2%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e189 (67.3%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1,739 (61.0%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1,175 (61.8%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e92 (32.7%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eVasopressin use, n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e47.20\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2,537 (89.0%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1,694 (89.1%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e212 (75.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e315 (11.0%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e207 (10.9%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e69 (24.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMidazolam use, n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.66\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.436\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2,082 (73.0%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1,419 (74.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e209 (74.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e770 (27.0%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e482 (25.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e72 (25.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFentanyl use, n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e30.58\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1,431 (50.2%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e964 (50.7%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e94 (33.5%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1,421 (49.8%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e937 (49.3%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e187 (66.5%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"6\"\u003eNote: SOFA score, Sequential Organ Failure Assessment Score; WBC, white blood cell; SBP, systolic blood pressure; DBP, diastolic blood pressure; MBP, Mean arterial blood pressure; pO2, partial pressure of Oxygen; pCO2, partial pressure of carbon dioxide; WBC, white blood cell; BUN, blood urea nitrogen; INR, International normalized ratio; PT, prothrombin time; PTT, partial thromboplastin time; PH, potential of hydrogen; CRRT, continuous renal replacement therapy.\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd colspan=\"6\"\u003e\u003csup\u003e1\u003c/sup\u003ePearson's Chi-squared test; Kruskal-Wallis rank sum test.\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cem\u003e---------\u003c/em\u003e\u003cb\u003eTable\u0026nbsp;1\u003c/b\u003e \u003cem\u003eis about here---------------\u003c/em\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\u003ch2\u003e3.2 Predictors of SAD\u003c/h2\u003e\u003cp\u003eTo identify the most relevant predictors of sepsis-associated delirium (SAD) in elderly ICU patients, we used the least absolute shrinkage and selection operator (LASSO) logistic regression for variable selection. Figure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA shows the results of 10-fold cross-validation to determine the optimal λ value, and Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB shows the coefficient paths for all candidate variables. Based on the minimum binomial deviance criterion, eight variables were selected for further modeling. These predictors were then included in a multivariate logistic regression analysis to estimate effect sizes. The results (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e) showed that the following factors were significantly associated with the development of SAD: neurological disease (OR\u0026thinsp;=\u0026thinsp;1.660, 95% CI: 1.352\u0026ndash;2.033, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001); mechanical ventilation (OR\u0026thinsp;=\u0026thinsp;1.883, 95% CI: 1.561\u0026ndash;2.264, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001); midazolam use (OR\u0026thinsp;=\u0026thinsp;3.931, 95% CI: 3.121\u0026ndash;4.958, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001); fentanyl use (OR\u0026thinsp;=\u0026thinsp;0.833, 95% CI: 0.672\u0026ndash;1.015, P\u0026thinsp;=\u0026thinsp;0.064); SOFA score (OR\u0026thinsp;=\u0026thinsp;1.031, 95% CI: 1.010\u0026ndash;1.062, P\u0026thinsp;=\u0026thinsp;0.036); temperature mean (OR\u0026thinsp;=\u0026thinsp;2.014, 95% CI: 1.719\u0026ndash;2.353, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001); hemoglobin min (OR\u0026thinsp;=\u0026thinsp;0.843, 95% CI: 0.807\u0026ndash;0.881, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001); sodium max (OR\u0026thinsp;=\u0026thinsp;1.098, 95% CI: 1.081\u0026ndash;1.116, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001). The predictor fentanyl use was removed from the final model because P\u0026thinsp;\u0026gt;\u0026thinsp;0.05 after multivariate analysis (\u003cb\u003eTable S5\u003c/b\u003e).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003e3.3 Development of the predictive model\u003c/h2\u003e\u003cp\u003eIn this study, we constructed a prediction model covering seven risk factors with clinical significance. We also designed a nomogram. In clinical applications, the patient's variable values are sequentially mapped to the scoring scale at the top of the nomogram, yielding a matching score for each variable (0-100), which is subsequently aggregated to give the total score. The overall score, combined with the risk axis beneath the nomogram, allows for the prediction of the patient's chance of SAD (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003e3.4 Internal and external validation of the predictive model\u003c/h2\u003e\u003cp\u003eThe predictive performance of the model, as measured by the AUC, was evaluated in three cohorts: the training set (AUC\u0026thinsp;=\u0026thinsp;0.794, 95%CI: 0.777\u0026ndash;0.810), the internal validation set (AUC\u0026thinsp;=\u0026thinsp;0.784, 95%CI: 0.763\u0026ndash;0.804), and the external validation set (AUC\u0026thinsp;=\u0026thinsp;0.815, 95%CI: 0.765\u0026ndash;0.864) \u003cb\u003e(\u003c/b\u003eFig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e. These results suggest that the model has a certain discriminative ability in predicting SAD in elderly ICU patients. Calibration curves were drawn for the training set, internal validation set, and external validation set, and the Brier score was calculated \u003cb\u003e(\u003c/b\u003eFig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eA\u0026ndash;C\u003cb\u003e)\u003c/b\u003e. The calibration curves showed that the predicted probabilities of all cohorts were basically consistent with the ideal diagonal line. The Brier score for the training set was 0.185 (95% CI: 0.177\u0026ndash;0.192), the Brier score for the internal validation set was 0.188 (0.179\u0026ndash;0.197), and the Brier score for the external validation set was 0.176 (0.154\u0026ndash;0.198). These values indicate acceptable calibration performance and suggest that the model has a certain generalization ability. The results of the H\u0026ndash;L test showed that none of the training set (χ\u0026sup2; = 12.918, P\u0026thinsp;=\u0026thinsp;0.115), internal validation set (χ\u0026sup2; = 4.385, P\u0026thinsp;=\u0026thinsp;0.821), and external validation set (χ\u0026sup2; = 8.954, P\u0026thinsp;=\u0026thinsp;0.346) reached a significant level, indicating that the model fit was good \u003cb\u003e(Table S6)\u003c/b\u003e. Moreover, decision curve analysis (DCA) showed that the model had potential clinical applicability across different decision threshold ranges in all three cohorts \u003cb\u003e(\u003c/b\u003eFig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eA\u0026ndash;C\u003cb\u003e).\u003c/b\u003e\u003c/p\u003e"},{"header":"4 Discussion","content":"\u003cp\u003eThis study used the MIMIC-IV database to develop a novel clinical prediction model for sepsis-associated delirium (SAD) in elderly ICU patients and externally validated its performance in an independent cohort at a tertiary hospital in China. The MIMIC database has been widely used in precision medicine research due to its comprehensive and high-quality ICU data. However, previous MIMIC-based delirium prediction models have primarily focused on Western populations and often lack external validation in diverse ethnic groups. This may limit their global applicability due to potential cultural, clinical, and systemic differences. To address this issue, we incorporated an independent external validation cohort consisting of elderly ICU patients with sepsis from a tertiary hospital in China. This approach enabled us to assess the model's applicability across diverse healthcare systems and ethnic groups, thereby enhancing its clinical relevance. Our model incorporated seven conventional variables: SOFA score, neurologic disease, mechanical ventilation, hemoglobin level, temperature, midazolam use, and serum sodium concentration. Both internal and external validation demonstrated good discrimination and calibration, indicating the model's potential clinical utility.\u003c/p\u003e\u003cp\u003eThe overall incidence of SAD among the elderly in the ICU in this study was 46.44%, aligning closely with the incidence reported in previous studies (9\u0026ndash;76.9%) [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. The external validation cohort originated from an ICU in a Chinese hospital, where the incidence of SAD was marginally lower (42.70%), potentially attributable to the data being sourced from a single site and the relatively small case volume. It is worth noting that the standardized diagnosis of SAD in the elderly still faces many difficulties, and there is still a lack of accurate identification tools for sepsis-induced neurological dysfunction in clinical practice [\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]. The current diagnostic criteria and screening instruments are primarily based on research with adult ICU patients and may lack enough sensitivity for the older demographic. Although the CAM-ICU score is a commonly used tool in clinical practice, its positive diagnosis requires multiple evaluations by professional nurses and is greatly affected by factors such as the patient's state of consciousness, use of sedatives, endotracheal intubation, and language barriers, which can easily lead to missed diagnoses or biased results [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. During the external validation of this study, some patients were left out because their medical records did not have valid CAM-ICU score information, which resulted in some patients being lost in the validation sample and indirectly highlighted the issue of potentially missing diagnoses with the CAM-ICU score for identifying SAD in older adults.\u003c/p\u003e\u003cp\u003eSeveral delirium prediction models have been developed and extensively validated. For example, the PRE-DELIRIC model includes 10 common ICU variables and demonstrates high accuracy (AUC up to 0.85, 95% CI: 0.84\u0026ndash;0.87). Although our model had a slightly lower AUC in the training set (0.794, 95% CI: 0.777\u0026ndash;0.810), it showed high external validity in the Asian cohort (AUC: 0.815, 95% CI: 0.765\u0026ndash;0.864). Researchers van den et al. [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e] believed that the PRE-DELIRIC model should be used with caution to predict delirium in high-risk patients. In contrast, we are more concerned about the high-risk subgroup of elderly sepsis patients. Moreover, Gu et al. [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] constructed a multifactorial prediction model for delirium using MIMIC, with an AUC of 0.743, but no external validation was performed. Our model is superior to their model (AUC\u0026thinsp;=\u0026thinsp;0.794) and has additional external validation. Machine learning\u0026ndash;based approaches have also been proposed. Fu et al. constructed a SAD prediction model using four algorithms, among which the best algorithm (XGBoost) achieved an AUC of 0.767 and had good internal calibration [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. However, the algorithm has not been externally validated, which limits its applicability in practical applications. Zhang et al. [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] and Yu et al. [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e] applied machine learning to the MIMIC-IV data and validated their models on the eICU-CRD database. While the internal AUC was satisfactory (Zhang: 0.793; Yu: 0.775), performance decreased significantly during external validation (0.701 and 0.687, respectively), demonstrating the performance degradation of the models when transferring across populations. Among these models, XGBoost is considered to be the best algorithm, but the model has many variables and may require more explanation to use. In contrast, the model we constructed added validation of the Asian population, and the seven variables in the model are simple and easy to obtain. The external validation AUC was 0.815, which is relatively stable and has potential applicability.\u003c/p\u003e\u003cp\u003eThe current study found that midazolam use, body temperature, neurological diseases and mechanical ventilation were the most important risk factors for SAD in elderly patients. Among them, midazolam (OR\u0026thinsp;=\u0026thinsp;4.131, 95%CI: 3.271\u0026ndash;5.218) had the greatest impact on the outcome. Midazolam is a short-acting benzodiazepine sedative. In routine ICU management, patients with severe sepsis are usually treated with mechanical ventilation and assisted with sedatives and analgesics [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]. Elderly patients are more susceptible to cognitive disorders and changes in consciousness due to their decreased ability to metabolize drugs and increased permeability of the blood-brain barrier. A prospective cohort study found that among 950 ICU patients receiving midazolam after continuous sedation, the incidence of delirium was 73% [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]. However, a randomized controlled trial by Hughes et al. [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e] demonstrated no significant association in delirium incidence and mortality in patients with severe sepsis sedated with dexmedetomidine, midazolam, or propofol, suggesting that the choice of sedative may not be the sole determining factor. Current studies have not achieved consensus on the effects of sedatives in SAD. The impact of sedatives on SAD in the elderly may be influenced by various parameters, including dosage, timing of administration, and individual patient characteristics [\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e]. Our study found that the early use of midazolam in elderly patients with sepsis admitted to the ICU was significantly associated with delirium. Careful sedation in clinical practice may help reduce the risk of SAD in this vulnerable population, but a large number of prospective high-quality studies are still needed to verify this. Mechanical ventilation has been identified as a high-risk factor by multiple delirium prediction models, especially in patients with severe sepsis [\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e], which is consistent with the results of this study. Critically ill patients receiving mechanical ventilation usually require continuous sedation, which can easily lead to circadian rhythm disorders, sensory deprivation, reduced cognitive stimulation, and adverse events such as delirium. Studies show that prompt initiation of mechanical ventilation (\u0026lt;\u0026thinsp;24 hours) and shortened duration of ventilation may alleviate problems and shorten hospital stay [\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e]. Mechanical ventilation is one of the important predictors in the model, suggesting that we may need to pay more attention to elderly patients who require early mechanical ventilation. Increased body temperature is usually part of the response to systemic inflammation (such as sepsis), when a large number of proinflammatory cytokines (such as IL-1β and TNFα) can penetrate the blood-brain barrier, interfere with neurotransmitter metabolism, and may induce a series of abnormal neurological reactions [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Multivariate regression analysis showed that the risk of delirium in patients with abnormal body temperature fluctuations was approximately twice that of patients with normal body temperature (OR\u0026thinsp;=\u0026thinsp;2.180, 95%CI: 1.856\u0026ndash;2.543). Tan et al. [\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e] found in a large sample study based on the MIMIC and eICU databases that body temperature deviation from the normal range (such as below 36\u0026deg;C or above 38\u0026deg;C) was significantly associated with ICU patient mortality. Irregular changes in patient body temperature may be associated with brain injury, and the possible impact of these temperature fluctuations on neurologic status warrants careful consideration; however, robust prospective studies are needed to confirm this claim. In addition, the history of neurological diseases is also one of the important predictive factors of the model. The patient's brain structure and functional reserve may be damaged. When faced with the pressure of acute inflammation, metabolic disorders or drug stimulation caused by sepsis, it is difficult to withstand the further decline of neurological function, which then evolves into SAD [\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e, \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e]. If we pay attention to this kind of medical history in patients, we may be able to prevent it in advance.\u003c/p\u003e\u003cp\u003eOverall, the prediction model constructed in this study showed good performance in multiple cohorts and showed certain potential for clinical application. However, the following limitations still need to be considered when interpreting the results and evaluating the practicality. First, although we built a prediction model based on the MIMIC-IV database and conducted external validation in ICU patients of a large tertiary hospital in Northeast China and preliminarily explored the applicability of the model under different ethnic groups and medical systems, the source of the external validation population was relatively single and the sample size was relatively limited, which may limit the generalization ability of the model. Further validation should be carried out in larger, multi-center, and multi-ethnic populations in the future. Second, both the development set and validation set of this study were from retrospective cohorts. Although the efficiency of data acquisition was improved, there were inherent selection biases and confounding factors, which may affect the accuracy of causal inference. In addition, some variables in the MIMIC-IV database were missing. Although we adopted a reasonable missing value filling strategy, the preprocessing process before the variables entered the model may introduce potential bias, thereby affecting the stability and reliability of the model. Third, the current model mainly incorporates seven routine clinical variables and does not include some immune/inflammatory biomarkers (such as C-reactive protein, IL-6, LMR, etc.) that have been confirmed to be closely related to SAD in previous studies. Although this is due to the consideration of data availability and clinical promotion of the model, it may also lead to insufficient identification of certain high-risk factors and limit the further improvement of model performance. In the future, such biomarkers should be considered for inclusion in model construction and optimization. Fourth, although this study has made indirect comparisons with existing models such as PRE-DELIRIC and E-PRE-DELIRIC in the discussion, it has not yet achieved strict horizontal reproduction; that is, these models have not been replicated on the same data set, and performance comparisons have been made. Therefore, our quantitative evaluation of the advantages and disadvantages of different models is still insufficient. Copying the methods of other models (such as score calculation, variable construction logic, verification strategy, etc.) and conducting systematic comparisons on a unified platform is a direction worthy of strengthening in future research. Finally, this study constructed a static prediction model, which has not been prospectively deployed or evaluated for intervention effects in actual clinical workflows, and it is impossible to confirm whether the model has an actual improvement effect on clinical diagnosis and treatment behavior or patient prognosis in real scenarios. Subsequent studies should combine dynamic data update mechanisms to conduct prospective intervention studies to further test the practicality and clinical value of the model.\u003c/p\u003e"},{"header":"5 Conclusion","content":"\u003cp\u003eThis study established and validated a new prediction model for delirium in elderly ICU patients with sepsis. The model showed good performance in both internal and external validation sets, demonstrating its potential for application. Although the overall performance of the model in this study may not necessarily surpass all models, its practicality, interpretability, and cross-regional external validation in a specific subgroup (elderly patients with sepsis) have important complementary value.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eConflict of Interest\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eThe authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest\u003c/em\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTengfei Zhou\u003c/strong\u003e and \u003cstrong\u003eXinming Tian\u003c/strong\u003e conceptualized and designed the study. \u003cstrong\u003eTengfei Zhou\u003c/strong\u003e performed the data analysis and drafted the initial manuscript. \u003cstrong\u003eXinming Tian\u003c/strong\u003e and \u003cstrong\u003eWei Wang\u003c/strong\u003e contributed to data analysis and interpretation. \u003cstrong\u003eXuyang Han\u003c/strong\u003e was responsible for the acquisition and organization of clinical data. \u003cstrong\u003eYanhui Feng\u003c/strong\u003e contributed to data collection. \u003cstrong\u003eWenyan Xiong\u003c/strong\u003e, \u003cstrong\u003eChunyan Zhang\u003c/strong\u003e, and \u003cstrong\u003eHongquan Fan\u003c/strong\u003e provided methodological support and critical feedback during manuscript preparation. \u003cstrong\u003eZhe Chu\u003c/strong\u003e supervised the study, provided critical revisions for statistical analysis and interpretation, and contributed substantially to manuscript revision. All authors reviewed and approved the final manuscript and agree to be accountable for all aspects of the work.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNo\u0026nbsp;specific grant from any funding agency.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThanks to all the authors of this article for their contributions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;Human Ethics and Consent to Participate declarations\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was conducted in two parts. Modeling was based on the publicly available de-identified database MIMIC-IV v3.1, the use of which was approved by the institutional review boards of Beth Israel Deaconess Medical Center (Boston, MA, USA) and the Massachusetts Institute of Technology (Cambridge, MA, USA). De-identification was completed during the data sharing process, thus eliminating the need for individual informed consent. External validation was based on clinical data from the First Hospital of Jilin University, which was approved by the Institutional Review Committee of the First Hospital of Jilin University (approval number: 2024-1161). Given the retrospective nature of this study and the fact that all patient data were de-identified, the Institutional Review Committee waived written informed consent from patients. Clinical trial number: not applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSupplementary Material\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSupplementary Material should be uploaded separately on submission.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe mimic-iv database used in this study is available on the official website (https://physionet.org/content/mimiciv/3.1/), and the external validation dataset is available after contacting the corresponding author for consent. The data extraction code and model construction R language code for this study have been uploaded to the website (https://github.com/PAIDAXING-HUP/Frontiers-code).\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eSinger M, Deutschman CS, Seymour CW, Shankar-hari M, Annane D, Bauer M, Bellomo R, Bernard GR, Chiche J, Coopersmith CM, et al. The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3). \u003cem\u003eJAMA\u003c/em\u003e (2016) 315:801\u0026ndash;810. doi: 10.1001/jama.2016.0287\u003c/li\u003e\n\u003cli\u003eKristina E. R, Sarah Charlotte J, Kareha M. A, Katya Anne S, Derrick T, Daniel Rhodes K, Danny V. C, Kevin S. I, Niranjan K, Simon F, et al. Global, regional, and national sepsis incidence and mortality, 1990-2017: analysis for the Global Burden of Disease Study. \u003cem\u003eLancet\u003c/em\u003e (2020) 395:200\u0026ndash;211. doi: 10.1016/S0140-6736(19)32989-7\u003c/li\u003e\n\u003cli\u003eLa Via L, Giuseppe S, Stefania S, Andrea M, Giuseppe N, Salvatore C, Ignazio LM, Bruno C, Stefano S, Serena S, et al. The Global Burden of Sepsis and Septic Shock. \u003cem\u003eEpidemiologia (Basel)\u003c/em\u003e (2024) 5:456\u0026ndash;478. doi: 10.3390/epidemiologia5030032\u003c/li\u003e\n\u003cli\u003eCavallazzi R, Saad M, Marik PE. Delirium in the ICU: an overview. \u003cem\u003eAnnals of Intensive Care\u003c/em\u003e (2012) 2:49. doi: 10.1186/2110-5820-2-49\u003c/li\u003e\n\u003cli\u003eInouye SK, Westendorp RGJ, Saczynski JS. Delirium in elderly people. \u003cem\u003eLancet\u003c/em\u003e (2014) 383:911\u0026ndash;922. doi: 10.1016/S0140-6736(13)60688-1\u003c/li\u003e\n\u003cli\u003eMaldonado JR. Delirium pathophysiology: An updated hypothesis of the etiology of acute brain failure. \u003cem\u003eInternational Journal of Geriatric Psychiatry\u003c/em\u003e (2018) 33:1428\u0026ndash;1457. doi: 10.1002/gps.4823\u003c/li\u003e\n\u003cli\u003eTsuruta R, Oda Y. A clinical perspective of sepsis-associated delirium. \u003cem\u003eJournal of Intensive Care\u003c/em\u003e (2016) 4:18. doi: 10.1186/s40560-016-0145-4\u003c/li\u003e\n\u003cli\u003eKim Y, Jin Y, Jin T, Lee S-M. Risk factors and outcomes of sepsis-associated delirium in intensive care unit patients: A secondary data analysis. \u003cem\u003eIntensive and Critical Care Nursing\u003c/em\u003e (2020) 59:102844. doi: 10.1016/j.iccn.2020.102844\u003c/li\u003e\n\u003cli\u003eBr\u0026uuml;ck E, Schandl A, Bottai M, Sackey P. The impact of sepsis, delirium, and psychological distress on self-rated cognitive function in ICU survivors\u0026mdash;a prospective cohort study. \u003cem\u003eJournal of Intensive Care\u003c/em\u003e (2018) 6:2. doi: 10.1186/s40560-017-0272-6\u003c/li\u003e\n\u003cli\u003eTd G, Jl T, Pp P, Ne B, Jc J, Mb P, Cg H, R C, Bt P, Lm B, et al. Clinical phenotypes of delirium during critical illness and severity of subsequent long-term cognitive impairment: a prospective cohort study. \u003cem\u003eThe Lancet Respiratory medicine\u003c/em\u003e (2018) 6: doi: 10.1016/S2213-2600(18)30062-6\u003c/li\u003e\n\u003cli\u003eAlıcı Ş, \u0026Ouml;zt\u0026uuml;rk Birge A. The frequency of sepsis‐associated delirium in intensive care unit and its effect on nurse workload. \u003cem\u003eJ Clin Nurs\u003c/em\u003e (2025) 34:1383\u0026ndash;1397. doi: 10.1111/jocn.17298\u003c/li\u003e\n\u003cli\u003eMcPherson JA, Wagner CE, Boehm LM, Hall JD, Johnson DC, Miller LR, Burns KM, Thompson JL, Shintani AK, Ely EW, et al. Delirium in the cardiovascular ICU: exploring modifiable risk factors. \u003cem\u003eCrit Care Med\u003c/em\u003e (2013) 41:405\u0026ndash;413. doi: 10.1097/CCM.0b013e31826ab49b\u003c/li\u003e\n\u003cli\u003eDziegielewski C, Skead C, Canturk T, Webber C, Fernando SM, Thompson LH, Foster M, Ristovic V, Lawlor PG, Chaudhuri D, et al. Delirium and Associated Length of Stay and Costs in Critically Ill Patients. \u003cem\u003eCritical Care Research and Practice\u003c/em\u003e (2021) 2021:6612187. doi: 10.1155/2021/6612187\u003c/li\u003e\n\u003cli\u003eAhmed S, Leurent B, Sampson EL. Risk factors for incident delirium among older people in acute hospital medical units: a systematic review and meta-analysis. \u003cem\u003eAge and Ageing\u003c/em\u003e (2014) 43:326\u0026ndash;333. doi: 10.1093/ageing/afu022\u003c/li\u003e\n\u003cli\u003eZheng G, Yan J, Li W, Chen Z. Frailty as an independent risk factor for sepsis-associated delirium: a cohort study of 11,740 older adult ICU patients. \u003cem\u003eAging Clin Exp Res\u003c/em\u003e (2025) 37:52. doi: 10.1007/s40520-025-02956-2\u003c/li\u003e\n\u003cli\u003eBruno RR, Wernly B, Mamandipoor B, Rezar R, Binneb\u0026ouml;ssel S, Baldia PH, Wolff G, Kelm M, Guidet B, De Lange DW, et al. ICU-Mortality in Old and Very Old Patients Suffering From Sepsis and Septic Shock. \u003cem\u003eFront Med\u003c/em\u003e (2021) 8: doi: 10.3389/fmed.2021.697884\u003c/li\u003e\n\u003cli\u003eEly EW, Margolin R, Francis J, May L, Truman B, Dittus R, Speroff T, Gautam S, Bernard GR, Inouye SK. Evaluation of delirium in critically ill patients: validation of the Confusion Assessment Method for the Intensive Care Unit (CAM-ICU). \u003cem\u003eCrit Care Med\u003c/em\u003e (2001) 29:1370\u0026ndash;1379. doi: 10.1097/00003246-200107000-00012\u003c/li\u003e\n\u003cli\u003eChen T-J, Chung Y-W, Chang H-C (Rita), Chen P-Y, Wu C-R, Hsieh S-H, Chiu H-Y. Diagnostic accuracy of the CAM-ICU and ICDSC in detecting intensive care unit delirium: A bivariate meta-analysis. \u003cem\u003eInternational Journal of Nursing Studies\u003c/em\u003e (2021) 113:103782. doi: 10.1016/j.ijnurstu.2020.103782\u003c/li\u003e\n\u003cli\u003eKuczmarska A, Ngo LH, Guess J, O\u0026rsquo;Connor MA, Branford-White L, Palihnich K, Gallagher J, Marcantonio ER. Detection of Delirium in Hospitalized Older General Medicine Patients: A Comparison of the 3D-CAM and CAM-ICU. \u003cem\u003eJ Gen Intern Med\u003c/em\u003e (2016) 31:297\u0026ndash;303. doi: 10.1007/s11606-015-3514-0\u003c/li\u003e\n\u003cli\u003eWojtara M, Rana E, Rahman T, Khanna P, Singh H. Artificial intelligence in rare disease diagnosis and treatment. \u003cem\u003eClin Transl Sci\u003c/em\u003e (2023) 16:2106\u0026ndash;2111. doi: 10.1111/cts.13619\u003c/li\u003e\n\u003cli\u003evan den Boogaard M, Pickkers P, Slooter AJC, Kuiper MA, Spronk PE, van der Voort PHJ, van der Hoeven JG, Donders R, van Achterberg T, Schoonhoven L. Development and validation of PRE-DELIRIC (PREdiction of DELIRium in ICu patients) delirium prediction model for intensive care patients: observational multicentre study. \u003cem\u003eBMJ\u003c/em\u003e (2012) 344:e420. doi: 10.1136/bmj.e420\u003c/li\u003e\n\u003cli\u003evan den Boogaard M, Schoonhoven L, Maseda E, Plowright C, Jones C, Luetz A, Sackey PV, Jorens PG, Aitken LM, van Haren FMP, et al. Recalibration of the delirium prediction model for ICU patients (PRE-DELIRIC): a multinational observational study. \u003cem\u003eIntensive Care Med\u003c/em\u003e (2014) 40:361\u0026ndash;369. doi: 10.1007/s00134-013-3202-7\u003c/li\u003e\n\u003cli\u003eWassenaar A, van den Boogaard M, van Achterberg T, Slooter AJC, Kuiper MA, Hoogendoorn ME, Simons KS, Maseda E, Pinto N, Jones C, et al. Multinational development and validation of an early prediction model for delirium in ICU patients. \u003cem\u003eIntensive Care Med\u003c/em\u003e (2015) 41:1048\u0026ndash;1056. doi: 10.1007/s00134-015-3777-2\u003c/li\u003e\n\u003cli\u003eShi X, Yang L, Bai W, Jing L, Qin L. Evaluating early lymphocyte-to-monocyte ratio as a predictive biomarker for delirium in older adult patients with sepsis: insights from a retrospective cohort analysis. \u003cem\u003eFront Med\u003c/em\u003e (2024) 11:1342568. doi: 10.3389/fmed.2024.1342568\u003c/li\u003e\n\u003cli\u003eGu Q, Yang S, Fei D, Lu Y, Yu H. A nomogram for predicting sepsis-associated delirium: a retrospective study in MIMIC III. \u003cem\u003eBMC Medical Informatics and Decision Making\u003c/em\u003e (2023) 23:184. doi: 10.1186/s12911-023-02282-5\u003c/li\u003e\n\u003cli\u003eZhang Y, Hu J, Hua T, Zhang J, Zhang Z, Yang M. Development of a machine learning-based prediction model for sepsis-associated delirium in the intensive care unit. \u003cem\u003eSci Rep\u003c/em\u003e (2023) 13:12697. doi: 10.1038/s41598-023-38650-4\u003c/li\u003e\n\u003cli\u003eYu S, Pan X, Zhang M, Zhang J, Chen D. A machine learning-based prediction model for sepsis-associated delirium in intensive care unit patients with sepsis-associated acute kidney injury. \u003cem\u003eRen Fail\u003c/em\u003e 47:2514186. doi: 10.1080/0886022X.2025.2514186\u003c/li\u003e\n\u003cli\u003eFu J, He A, Wang L, Li X, Yu J, Zheng R. Interpretable machine learning model for predicting delirium in patients with sepsis: a study based on the MIMIC data. \u003cem\u003eBMC Infect Dis\u003c/em\u003e (2025) 25:1\u0026ndash;13. doi: 10.1186/s12879-025-10982-8\u003c/li\u003e\n\u003cli\u003ePollard TJ, Johnson AEW, Raffa JD, Celi LA, Mark RG, Badawi O. The eICU Collaborative Research Database, a freely available multi-center database for critical care research. \u003cem\u003eSci Data\u003c/em\u003e (2018) 5:1\u0026ndash;13. doi: 10.1038/sdata.2018.178\u003c/li\u003e\n\u003cli\u003eAlistair Johnson, Tom Pollard, a-chahin, Jim Blundell, Brian Gow, erinhong, Michael Schubert, Kien Dang, Nicolas Paris, shu98, et al. MIT-LCP/mimic-code: v3.0.1. (2025) doi: 10.5281/ZENODO.821871\u003c/li\u003e\n\u003cli\u003eKim SE, Ko R-E, Na SJ, Chung CR, Choi KH, Kim D, Park TK, Lee JM, Song YB, Choi J-O, et al. External validation and comparison of two delirium prediction models in patients admitted to the cardiac intensive care unit. \u003cem\u003eFront Cardiovasc Med\u003c/em\u003e (2022) 9:947149. doi: 10.3389/fcvm.2022.947149\u003c/li\u003e\n\u003cli\u003eHsiao W-L, Hsieh P-Y, Hsu L-F, Lai Y-H, Wang M-L, Yeh Y-C. Effectiveness of a delirium risk assessment and multidisciplinary care approach in reducing delirium incidence among surgical intensive care unit patients: A retrospective pre-post intervention study. \u003cem\u003eJournal of Clinical Anesthesia\u003c/em\u003e (2024) 98:111566. doi: 10.1016/j.jclinane.2024.111566\u003c/li\u003e\n\u003cli\u003eWeng L, Xu Y, Yin P, Wang Y, Chen Y, Liu W, Li S, Peng J, Dong R, Hu X, et al. National incidence and mortality of hospitalized sepsis in China. \u003cem\u003eCrit Care\u003c/em\u003e (2023) 27:84. doi: 10.1186/s13054-023-04385-x\u003c/li\u003e\n\u003cli\u003eJohnson AEW, Bulgarelli L, Shen L, Gayles A, Shammout A, Horng S, Pollard TJ, Hao S, Moody B, Gow B, et al. MIMIC-IV, a freely accessible electronic health record dataset. \u003cem\u003eSci Data\u003c/em\u003e (2023) 10: doi: 10.1038/s41597-022-01899-x\u003c/li\u003e\n\u003cli\u003eCollins GS, Reitsma JB, Altman DG, Moons KGM. Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): the TRIPOD statement. \u003cem\u003eAnn Intern Med\u003c/em\u003e (2015) 162:55\u0026ndash;63. doi: 10.7326/M14-0697\u003c/li\u003e\n\u003cli\u003eZhang Z, Rousson V, Lee W-C, Ferdynus C, Chen M, Qian X, Guo Y. Decision curve analysis: a technical note. \u003cem\u003eAnn Transl Med\u003c/em\u003e (2018) 6:308. doi: 10.21037/atm.2018.07.02\u003c/li\u003e\n\u003cli\u003eBuuren S van, Groothuis-Oudshoorn K. mice: Multivariate Imputation by Chained Equations in R. \u003cem\u003eJournal of Statistical Software\u003c/em\u003e (2011) 45:1\u0026ndash;67. doi: 10.18637/jss.v045.i03\u003c/li\u003e\n\u003cli\u003eHarrell, FE. \u003cem\u003eRegression Modeling Strategies: With Applications to Linear Models, Logistic and Ordinal Regression, and Survival Analysis\u003c/em\u003e. Cham: Springer International Publishing. (2015). doi: 10.1007/978-3-319-19425-7\u003c/li\u003e\n\u003cli\u003eRobin X, Turck N, Hainard A, Tiberti N, Lisacek F, Sanchez J-C, M\u0026uuml;ller M. pROC: an open-source package for R and S+ to analyze and compare ROC curves. \u003cem\u003eBMC Bioinformatics\u003c/em\u003e (2011) 12:77. doi: 10.1186/1471-2105-12-77\u003c/li\u003e\n\u003cli\u003eLallou\u0026eacute; B, Deguen S, Monnez J-M, Padilla C, Kihal W, Zmirou-Navier D, Le Meur N. SesIndexCreatoR: An R Package for Socioeconomic Indices Computation and Visualization. \u003cem\u003eOJS\u003c/em\u003e (2015) 05:291\u0026ndash;302. doi: 10.4236/ojs.2015.54031\u003c/li\u003e\n\u003cli\u003eCorrplot: Visualization of a Correlation Matrix. https://taiyun.r-universe.dev/corrplot\u003c/li\u003e\n\u003cli\u003eEvans L, Rhodes A, Alhazzani W, Antonelli M, Coopersmith CM, French C, Machado FR, Mcintyre L, Ostermann M, Prescott HC, et al. Surviving sepsis campaign: international guidelines for management of sepsis and septic shock 2021. \u003cem\u003eIntensive Care Med\u003c/em\u003e (2021) 47:1181\u0026ndash;1247. doi: 10.1007/s00134-021-06506-y\u003c/li\u003e\n\u003cli\u003ePham T, Heunks L, Bellani G, Madotto F, Aragao I, Beduneau G, Goligher EC, Grasselli G, Laake JH, Mancebo J, et al. Weaning from mechanical ventilation in intensive care units across 50 countries (WEAN SAFE): a multicentre, prospective, observational cohort study. \u003cem\u003eThe Lancet Respiratory Medicine\u003c/em\u003e (2023) 11:465\u0026ndash;476. doi: 10.1016/S2213-2600(22)00449-0\u003c/li\u003e\n\u003cli\u003evan Gelder TG, van Diem-Zaal IJ, Dijkstra-Kersten SMA, de Mul N, Lalmohamed A, Slooter AJC. The risk of delirium after sedation with propofol or midazolam in intensive care unit patients. \u003cem\u003eBr J Clin Pharmacol\u003c/em\u003e (2024) 90:1471\u0026ndash;1479. doi: 10.1111/bcp.16031\u003c/li\u003e\n\u003cli\u003eHughes CG, Mailloux PT, Devlin JW, Swan JT, Sanders RD, Anzueto A, Jackson JC, Hoskins AS, Pun BT, Orun OM, et al. Dexmedetomidine or Propofol for Sedation in Mechanically Ventilated Adults with Sepsis. \u003cem\u003eN Engl J Med\u003c/em\u003e (2021) 384:1424\u0026ndash;1436. doi: 10.1056/NEJMoa2024922\u003c/li\u003e\n\u003cli\u003eKawazoe Y, Miyamoto K, Morimoto T, Yamamoto T, Fuke A, Hashimoto A, Koami H, Beppu S, Katayama Y, Itoh M, et al. Effect of Dexmedetomidine on Mortality and Ventilator-Free Days in Patients Requiring Mechanical Ventilation With Sepsis: A Randomized Clinical Trial. \u003cem\u003eJAMA\u003c/em\u003e (2017) 317:1321\u0026ndash;1328. doi: 10.1001/jama.2017.2088\u003c/li\u003e\n\u003cli\u003eYamamoto T, Mizobata Y, Kawazoe Y, Miyamoto K, Ohta Y, Morimoto T, Yamamura H. Incidence, risk factors, and outcomes for sepsis-associated delirium in patients with mechanical ventilation: A sub-analysis of a multicenter randomized controlled trial. \u003cem\u003eJ Crit Care\u003c/em\u003e (2020) 56:140\u0026ndash;144. doi: 10.1016/j.jcrc.2019.12.018\u003c/li\u003e\n\u003cli\u003eKim G, Oh DK, Lee SY, Park MH, Lim C-M, the Korean Sepsis Alliance (KSA) investigators. Impact of the timing of invasive mechanical ventilation in patients with sepsis: a multicenter cohort study. \u003cem\u003eCritical Care\u003c/em\u003e (2024) 28:297. doi: 10.1186/s13054-024-05064-1\u003c/li\u003e\n\u003cli\u003eTan DJ, Chen J, Zhou Y, Ong JSQ, Sin RJX, Bui TV, Mehta AA, Feng M, See KC. Association of body temperature and mortality in critically ill patients: an observational study using two large databases. \u003cem\u003eEuropean Journal of Medical Research\u003c/em\u003e (2024) 29:33. doi: 10.1186/s40001-023-01616-3\u003c/li\u003e\n\u003cli\u003eMitasova A, Kostalova M, Bednarik J, Michalcakova R, Kasparek T, Balabanova P, Dusek L, Vohanka S, Ely EW. Poststroke delirium incidence and outcomes: validation of the Confusion Assessment Method for the Intensive Care Unit (CAM-ICU). \u003cem\u003eCrit Care Med\u003c/em\u003e (2012) 40:484\u0026ndash;490. doi: 10.1097/CCM.0b013e318232da12\u003c/li\u003e\n\u003cli\u003eFrei AI, Wagner AS, Baumann SM, Grzonka P, Berger S, Hunziker S, R\u0026uuml;egg S, Marsch S, Sutter R. Concurrence of seizures and peri-ictal delirium in the critically ill - its frequency, associated characteristics, and outcomes. \u003cem\u003eJ Neurol\u003c/em\u003e (2024) 271:231\u0026ndash;240. doi: 10.1007/s00415-023-11944-3\u003c/li\u003e\n\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":"bmc-geriatrics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bgtc","sideBox":"Learn more about [BMC Geriatrics](http://bmcgeriatr.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bgtc/default.aspx","title":"BMC Geriatrics","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Sepsis-associated delirium, elderly patients, intensive care unit, nomogram, predictive model","lastPublishedDoi":"10.21203/rs.3.rs-7380480/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7380480/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cem\u003e\u003cstrong\u003eBackground\u003c/strong\u003e\u003c/em\u003e\u003cstrong\u003e:\u003c/strong\u003e\u003cem\u003e\u003cstrong\u003e \u003c/strong\u003e\u003c/em\u003eSepsis-associated delirium (SAD) is a common acute brain dysfunction in elderly patients in the intensive care unit (ICU), which significantly increases the length of hospital stay, medical costs, and the risk of death. Despite the availability of multiple delirium prediction tools, there are few models specific to the elderly septic population, and most lack external validation.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u003cstrong\u003eMethods\u003c/strong\u003e\u003c/em\u003e\u003cstrong\u003e:\u003c/strong\u003e\u003cem\u003e\u003cstrong\u003e \u003c/strong\u003e\u003c/em\u003eThis retrospective cohort study enrolled 5034 elderly ICU patients with sepsis. A prediction model was constructed based on the MIMIC-IV database and externally validated using 281 patients admitted to the First Affiliated Hospital of Jilin University between January 2019 and November 2024. A workflow was developed using R software. Candidate predictors were first identified using the LASSO regression method, then incorporated into a multivariate logistic regression model and visualized as a nomogram. Patients were randomly divided into a training set and an internal validation set in a 6:4 ratio. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), the Hosmer-Lemeshow test, calibration plots, Brier scores, and decision curve analysis (DCA).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u003cstrong\u003eResults\u003c/strong\u003e\u003c/em\u003e\u003cstrong\u003e: \u003c/strong\u003eThe overall incidence of delirium was 46.44%. Seven variables were ultimately included in the final model: body temperature, SOFA score, hemoglobin level, serum sodium concentration, history of neurological disease, mechanical ventilation, and midazolam use. The model demonstrated good discriminatory performance, with area under the receiver operating characteristic curve (AUC) values of 0.794 (95% CI: 0.777–0.810) in the training set, 0.784 (95% CI: 0.763–0.804) in the internal validation set, and 0.815 (95% CI: 0.765–0.864) in the external validation set. The Hosmer–Lemeshow goodness-of-fit test showed no significant deviation between predicted and observed outcomes (P \u0026gt; 0.05), indicating good calibration. The Brier scores were 0.185, 0.188, and 0.176 for the training, internal validation, and external validation sets, respectively. Decision curve analysis (DCA) further confirmed the model’s potential clinical utility.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion:\u003c/strong\u003e The SAD risk prediction model developed in this study features a simple structure and relies on readily available clinical variables. It demonstrated favorable discrimination and calibration in the external validation cohort, suggesting its potential utility as a practical tool for early detection and targeted intervention of delirium in elderly ICU patients with sepsis.\u003c/p\u003e","manuscriptTitle":"External validation in a Chinese cohort of a nomogram developed using MIMIC-IV for predicting sepsis-associated delirium in elderly ICU patients","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-15 17:02:53","doi":"10.21203/rs.3.rs-7380480/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-01-07T07:07:35+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"34930043284434131102559218490893987159","date":"2025-12-28T23:48:40+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-10-02T20:00:05+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-09-09T11:38:35+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-08-23T09:43:15+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-08-23T09:41:52+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Geriatrics","date":"2025-08-15T10:05:21+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-geriatrics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bgtc","sideBox":"Learn more about [BMC Geriatrics](http://bmcgeriatr.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bgtc/default.aspx","title":"BMC Geriatrics","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"24e7b0a8-3403-45b1-be5a-e0af2f520248","owner":[],"postedDate":"October 15th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-10-15T17:02:53+00:00","versionOfRecord":[],"versionCreatedAt":"2025-10-15 17:02:53","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7380480","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7380480","identity":"rs-7380480","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}