Background In gastrointestinal endoscopy (GIE), the growing use of sedation and anesthesia often leads to intraoperative hypoxemia, escalating the risk of respiratory, renal, and cardiovascular complications. Providing anesthesiologists with a machine learning (ML) model tailored for GIE patients is essential for mitigating these risks. Methods A cohort of patients who underwent painless GIE were retrospective collected in the Nanjing First Hospital. We developed five ML models separately and selected ones with best performance as the final preoperative and holistic model. To verify the generalization ability of the final models, an independent cohort was collected from the Fourth Affiliated Hospital of Nanjing Medical University to conduct external validation. In addition, the Shapley additive explanations method was applied to interpret the optimal models. Results According to the inclusion and exclusion standard, we included a total of 10170 patients in our study. In both preoperative and holistic models, the XGBoost models were selected as the final models because of excellent performance (AUROC was 0.875 in preoperative model and 0.941 in holistic model). In external validation including 995 patients, the model’s the model showed consistent performance, supporting its robustness and generalization. Conclusions Our project proposed preoperative and holistic ML models for predicting hypoxemia during outpatient GIE operations, ensuring patient safety during surgery, and providing them with safe and effective care. Factors such as hemoglobin levels, diabetes presence, nasopharyngeal ventilation, and high-flow oxygen use significantly influence hypoxemia in these patients.

Background:Hypoxemia is a common and serious complication during sedated gastrointestinal endoscopy for out- and in-patients. Though diagnostic and severity scoring systems of obstructive sleep apnea (OSA) and difficult airway assessment (DAA) are widely used to assess hypoxemia risk, there is no exclusively designed prediction model and convenient tool in real-world practice. We aimed to develop and validate a robust and accurate hypoxemia risk prediction model for pre-operative use in this context. Methods:Using data from out-patients undergoing gastrointestinal endoscopy between May 2020 and November 2023 across seven hospitals in China with diverse regional and ethnic backgrounds, we developed and independently validated a hypoxemia risk prediction model for sedated gastrointestinal endoscopy (HAPPY-12K). The model was developed to pre-operatively predict occurrence of hypoxemia during sedated gastrointestinal endoscopy, defined as SpO2 falling below 95% for a duration exceeding 10 s. HAPPY-12K was a logistic regression model incorporating eight predictors: body mass index, Mallampati grade, limited jaw protrusion, short thyromental distance, large tongue, history of snoring, short neck with large circumference and pre-operative mean arterial pressure. The model was constructed by a well-established 3-D modeling strategy composed of Double types of effects, Double steps of screening, and Double steps of modeling. The discriminative ability was evaluated using the area under the receiver operating characteristic curve (AUC). The model calibration was examined through calibration slope, expected-to-observed (E:O) ratio and Brier score. For clinical utility, decision curve analysis was performed to assess net benefit (NB) and net reduction (NR). Furthermore, we systematically compared HAPPY-12K with other newly developed models using scores or raw variables from questionaries of OSA and DAA using DeLong's test. This study is registered in the Chinese Clinical Trial Registry (ChiCTR2300074128). Findings:We included 11,957 patients, divided into a Training Set (n = 2,518, hypoxemia rate 10.37%), and five validation sets (n = 9,439, hypoxemia rate raining from 8.40% to 28.45%). HAPPY-12K was developed in a Han Chinese population and exhibited satisfactory discrimination ability with AUCs ranging from 0.818 to 0.895 in external populations of the same ethnicity, and an acceptable AUC of 0.771 in a Uygur Chinese population. Although its Brier scores were satisfactory across all ethnic populations, HAPPY-12K displayed acceptable calibration (calibration slope < 1.2) in external Han Chinese populations, and good calibration (E:O ratio = 0.962) in independent homogenous populations comparable to the training set. The average NB and NR were 45.2‰ and 59.5%, respectively. It was estimated that HAPPY-12K would identify over half a million patients with truly developing hypoxemia during sedated gastrointestinal endoscopy and could help avoid over six million unnecessary interventions annually in China. Meanwhile, a head-to-head comparison revealed that HAPPY-12K outperformed other models. HAPPY-12K has been implemented as an interactive online tool available at http://bigdata.njmu.edu.cn/HAPPY-12K/. Interpretation:HAPPY-12K could enable efficient and precise hypoxemia risk assessment before sedated gastrointestinal endoscopy, providing timely alerts for high-risk outpatients. Funding:National Natural Science Foundation of China; Noncommunicable Chronic Diseases-National Science and Technology Major Project; Science and Technology Project of Jiangsu Disease Control and Prevention Administration; Science and Technology Development Project of Nanjing Medical University; Priority Academic Program Development of Jiangsu Higher Education Institutions; and Outstanding Young Level Academic Leadership Training Program of Nanjing Medical University.

BACKGROUND:Acute kidney injury is a common complication after orthotopic heart transplantation. Previous models have failed to consider the impact of multiple hemodynamic parameters on prediction performance. The objective of this study was aimed to develop machine-learning models incorporating multiple hemodynamic parameters to predict the risk of developing acute kidney injury after orthotopic heart transplantation. METHODS:We retrospectively analyzed 114 recipients of orthotopic heart transplantation, with postoperative stage 2-3 acute kidney injury as the prediction outcome. Preoperative characteristics, laboratory parameters, and intraoperative hemodynamics were evaluated. Potential predictors were first screened by univariate analysis, followed by least absolute shrinkage and selection operator regression for feature selection. Five machine-learning models were developed and evaluated via 5-fold cross-validation using multiple performance metrics. RESULTS:Postoperative stage 2-3 acute kidney injury incidence was 21.9% (25/114). Intraoperative hypotension and venous congestion were significantly associated with acute kidney injury. A total of 8 factors were ultimately filtered for constructing multiple machine-learning models, with the light gradient boosting machine model demonstrating the optimal performance (area under the receiver operating characteristic curve, 0.898; area under the precision-recall curve, 0.802; Brier score: 0.106). CONCLUSION:The machine-learning model based on incorporating perioperative hemodynamics (pulmonary artery systolic pressure, mean arterial pressure, central venous pressure) accurately predict stage 2-3 acute kidney injury postorthotopic heart transplantation, thereby optimizing hemodynamic management and improving outcomes.

Background:Most existing models for predicting postoperative pulmonary complications (PPCs) rely solely on preoperative variables and lack integration of intraoperative data or clinical tools for application. Methods:This study developed and externally validated logistic regression and machine learning (ML) models for predicting PPCs in non-cardiothoracic surgical patients undergoing general anesthesia. A cohort of 997 patients was randomly divided into training and test sets in a 7:3 ratio. The stepwise regression method was adopted to conduct feature screening on the preoperative dataset and the combined preoperative and intraoperative dataset, and a variety of ML algorithm models were developed. Model performance was evaluated using area under the receiver operating characteristic (AUROC), area under the precision-recall curve (AUPRC), brier score, sensitivity, specificity, precision, and F1-score. External validation was performed on an independent cohort of 286 patients from a different institution. The features of the optimized model were interpreted by introducing SHapley Additive exPlanations (SHAP) method. Results:The AUROC value of the logistic regression model using only preoperative variables in the test set was 0.828. After including intraoperative variables, the CatBoost model demonstrated superior performance, with an AUROC value of 0.927. In the preoperative model, C-reactive protein (CRP), American Society of Anesthesiologists (ASA) physical status classification, age, and smoking status were the most significant predictors. However, operative duration, CRP, age, and intraoperative blood loss were the dominant predictors of postoperative outcomes. External validation confirmed general applicability, with preoperative logistic regression and the Assess Respiratory Risk in Surgical Patients in Catalonia Tool (ARISCAT) yielding AUROC values of 0.827 and 0.796, respectively. The postoperative CatBoost model achieved an AUROC of 0.865, AUPRC of 0.809, sensitivity of 0.787, specificity of 0.921, and the lowest Brier score (0.132). Conclusions:The preoperative model enabled early risk stratification, while the postoperative model provided improved accuracy. A publicly accessible web-based calculator was developed to support clinical implementation and facilitate prospective validation.

Background:N-methyl-D-aspartate receptors (NMDARs) are known to contribute to neurological and neurodegenerative diseases. Meanwhile, glutamate ionotropic receptor N-methyl-D-aspartate-type subunit 2D (GRIN2D), which encodes NMDAR subunit 2D, may play a role in colorectal cancer. The study examined whether GRIN2D is involved in lung cancer. Methods:Through the use of quantitative reverse-transcription polymerase chain reaction (qRT-PCR), GRIN2D expression was detected in tissue samples and cell lines. EdU staining assay was used to determine the cell proliferation ability, while TUNEL staining assay was used to evaluate apoptosis. The phosphorylation levels of PI3K, AKT, and mTOR were measured via Western blotting; cell viability was evaluated via Cell Counting Kit-8 (CCK-8) assay; and colony formation ability was examined via colony formation assay. Results:In this study, we demonstrated that lung adenocarcinoma and related cancer cell lines had significantly high levels of GRIN2D expression. GRIN2D promoted cancer cell proliferation while inhibiting apoptosis via the PI3K/mTOR signaling pathway. Esketamine, a GRIN2D inhibitor, and LY294002, a PI3K inhibitor, either alone or in combination, could suppress the tumor growth induced by high GRIN2D levels both in vitro and in vivo. Conclusions:This study is the first to identify the involvement of GRIN2D in lung cancer and to clarify the underlying mechanism of its effect; the findings further suggest that ketamine in cancer treatment may extend beyond relieving pain and depression.

​Abstract Background: Magnesium deficiency is common in critically ill patients, and serum-based thresholds may underestimate clinically relevant deficiency within the normal range. The association between magnesium levels and mortality across the full range of values remains incompletely characterized. Objective: To determine the magnesium level associated with the lowest mortality and to assess whether supplementation and early target attainment are associated with improved survival in patients with low–normal levels. Methods: We conducted a retrospective cohort study of 42,419 ICU adults from the MIMIC-IV database. Restricted cubic splines were used to characterize the nonlinear association between admission magnesium and 28-day mortality. Among patients with baseline magnesium 1.7–2.1 mg/dL, representing the low–normal range above the conventional hypomagnesemia threshold, intravenous magnesium supplementation and early target attainment (≥2.1 mg/dL within 24 h) were evaluated using multivariable Cox models and propensity score–matched analyses. Results: Across the full magnesium range, mortality was lowest at approximately 2.1 mg/dL. In patients with magnesium 1.7–2.1 mg/dL, supplementation was associated with lower 28-day mortality (HR 0.85, 95% CI 0.74–0.97; P=0.010) after matching. Early target attainment was also associated with reduced mortality (HR 0.81, 95% CI 0.65–0.99; P=0.041), with consistent findings in time-varying models and stratified models. Conclusion: Low–normal magnesium levels may represent a clinically relevant risk state in critically ill patients. In this subgroup, magnesium supplementation and early target attainment were associated with lower mortality, supporting a target-based approach that should be evaluated prospectively.

STUDY OBJECTIVE:Quality of bowel preparation correlates with the time interval between its completion and colonoscopy in patients receiving colonoscopy. Previous studies focused primarily on awake patients, neglecting the risk of regurgitation and aspiration in those receiving sedation. We investigated the relationship between gastric emptying time, colonoscopy waiting time, and bowel preparation quality in patients undergoing colonoscopy with sedation. DESIGN:A prospective, randomized trial. SETTING:Endoscopy center. PATIENTS:134 patients receiving colonoscopy with sedation. INTERVENTIONS:Gastric ultrasound assessment was initiated 2 or 4 h after completing the 3 L split-dose oral sulfate solution (OSS), followed by colonoscopy within 2-4 h (Group A) or 4-6 h (Group B) after confirming gastric emptying. MEASUREMENTS:The primary outcomes were full stomach rate at various timepoints and bowel preparation quality. Secondary outcomes included colonoscopy waiting time, operation time, terminal ileal intubation rate, polyp and adenoma detection rate, and overall satisfaction. MAIN RESULTS:Gastric emptying was confirmed at 2 or 4 h after bowel preparation in all patients. The bowel preparation score was significantly higher in Group A than in Group B (7.0 [6.0-8.0] vs 6.0 [5.0-7.0], P < 0.001), and patient satisfaction was also higher (10.0 [8.0-10.0] vs 8.0 [7.0-9.0], P < 0.001). CONCLUSIONS:Performing colonoscopy 2 h after 3 L split-dose OSS intake does not increase the risk of intraoperative regurgitation or aspiration as assessed by ultrasound in this selected, low-risk population. Colonoscopy performed within 2-4 h after bowel preparation completion yields superior bowel preparation quality compared to 4-6 h.

Background: A full stomach during elective sedated upper endoscopy (UE) is highly associated with aspiration. Aspiration is related to increased morbidity and mortality. We aimed to develop a nomogram model to predict the risk of having a full stomach in adult outpatients undergoing elective sedated UE. Methods: Data of 1171 adult outpatients undergoing elective sedated UE were collected retrospectively between July 2021 and February 2022. Univariable analysis, multivariable logistic regression, and Boruta were used to identify independent risk factors associated with full stomach. Using the identified risk factors, a nomogram was developed. The area under the receiver operating characteristic (AUROC), decision curve analysis (DCA), and calibration curve (CC) were used to evaluate the model’s performance. Results: After exclusion, 990 patients were enrolled in this study. In the training set, 305 patients (38.5%) had a full stomach, and 71 patients (35.9%) had a full stomach for the testing set. The AUROC (95% Cl) was 0.819 (0.788-0.85) and 0.823(0.759-0.887) for the training and testing set, respectively. The variables incorporated in the development of our STAGED model were Sex = female (OR: 2.147), fasting Time (OR: 0.839), Age (OR: 0.971), GERD (OR: 15.61), diabEtes (OR: 8.614) and Diet (OR: 3.691). DCA and CIC analyses showed that our model had a significant net benefit compared to the "treat all or "treat none" strategies and had high predictive value, making it clinically useful. Conclusion: The “STAGED” nomogram model was successfully developed and made easily accessible online via a visual web-based calculator. Summary of the patient’s inclusion along with the study design, the methodology used for model development, the implementation of the model, and model performance.

Sepsis-associated encephalopathy (SAE), one of the common complications of sepsis, is associated with higher ICU mortality, prolonged hospitalization, and long-term cognitive decline. Sepsis can induce neuroinflammation, which negatively affects hippocampal neurogenesis. Dexmedetomidine has been shown to protect against SAE. However, the potential mechanism remains unclear. In this study, we added lipopolysaccharide (LPS)-stimulated astrocytes-conditioned media (LPS-CM) to neural stem cells (NSCs) culture, which were pretreated with dexmedetomidine in the presence or absence of the α2-adrenoceptor antagonist yohimbine or the α2A-adrenoceptor antagonist BRL-44408. LPS-CM impaired the neurogenesis of NSCs, characterized by decreased proliferation, enhanced gliogenesis, and declined viability. Dexmedetomidine alleviated LPS-CM-induced impairment of neurogenesis in a dose-dependent manner. Yohimbine, as well as BRL-44408, reversed the effects of dexmedetomidine. We established a mouse model of SAE via cecal ligation and perforation (CLP). CLP-induced astrocyte-related neuroinflammation and hippocampal neurogenesis deficits, accompanied by learning and memory decline, which were reversed by dexmedetomidine. The effect of dexmedetomidine was blocked by BRL-44408. Collectively, our findings support the conclusion that dexmedetomidine can protect against SAE, likely mediated by the combination of inhibiting neuroinflammation via the astrocytic α2A-adrenoceptor with attenuating neuroinflammation-induced hippocampal neurogenesis deficits via NSCs α2A-adrenoceptor.

BACKGROUND:Interleukin-9 (IL-9) is an emerging pro-inflammatory cytokine that promotes intestinal barrier injury (IBI) in sepsis. The specific mechanisms of IL-9-induced IBI still need to be clarified. As a newly discovered form of programmed cell death, ferroptosis was demonstrated to be involved in sepsis-related organ dysfunction, yet its role in IL-9-induced IBI in sepsis remains unexplored. METHODS:Serum levels of IL-9, D-lactate, intestinal fatty acid binding protein (iFABP), glutathione (GSH), and glutathione peroxidase 4 (GPX4) were tested in septic patients and control subjects. Biomarkers reflecting barrier function in serum and intestinal tissue were tested in treated rats. Rats underwent sepsis induction, IL-9, IL-9 inhibition (IL-9i), and ferroptosis inhibition (Fei) treatment were selected to examine the severity of IBI and survival rates. RESULTS:Significantly elevated levels of IL-9, D-lactate, iFABP, GSH, and GPX4 were observed in septic patients and rats. IL-9 levels showed a negative correlation with GSH and GPX4 levels, while GSH or GPX4 levels showed an inverse correlation with D-lactate and iFABP levels. Serum GSH and GPX4 levels demonstrated strong predictive value for acute gastrointestinal injury of grade II and above in septic patients. IL-9 administration increased levels of transferrin receptor, Fe 2+ , and iFABP in serum and intestinal tissue of septic rats, while decreasing GSH, GPX4, and zonula occludens 1 levels. Inhibition of ferroptosis reversed these biomarkers alterations. Intestinal permeability, transmission electron microscopy, histopathology, and apoptosis assays confirmed exacerbated IBI following IL-9 upregulation and its attenuation upon ferroptosis inhibition. CONCLUSION:Ferroptosis was implicated in the IL-9-induced intestinal barrier injury in sepsis.

Sepsis-associated encephalopathy is a debilitating complication of systemic infection, marked by acute cognitive impairment and long-term neurological deficits in the absence of direct central nervous system (CNS) infection. Its pathogenesis involves a multifactorial interplay of neuroinflammation (e.g., cytokine storms), immune dysregulation, blood-brain barrier (BBB) disruption, metabolic derangements, and impaired neuronal repair. These mechanisms synergistically contribute to neuronal injury and persistent cognitive dysfunction. Emerging therapeutic strategies-such as targeted immunomodulators, BBB-stabilizing agents, and novel CNS-targeted drug delivery-aim to interrupt this cascade and improve outcomes. Concurrently, precision medicine approaches leverage molecular profiling to tailor interventions. However, current clinical management remains supportive, hindered by incomplete mechanistic understanding and a paucity of disease-modifying therapies. This review synthesizes recent advances in the pathophysiology of sepsis-associated encephalopathy, critically evaluates these mechanism-based therapeutic approaches, and highlights translational roadblocks in biomarker development and preclinical-to-clinical bridging. We also propose future directions to accelerate the development of targeted pharmacotherapies and personalized treatment paradigms for sepsis-associated encephalopathy.

BACKGROUND:Prior research on postoperative pneumonia (POP) risk models focused on preoperative factors but overlooked intraoperative variables vital for precision. These models also neglected the higher-risk elderly population. This study seeks to develop and evaluate preoperative and combined models to predict POP risk in elderly patients undergoing non-cardiothoracic surgery. METHODS:A retrospective cohort of 444 patients who underwent non-cardiothoracic surgery at Nanjing First Hospital from March 2021 to April 2022 was included. Univariate analysis and Least Absolute Shrinkage and Selection Operator (LASSO) regression were employed to develop preoperative and combined logistic regression models. RESULTS:The area under the receiver operating characteristic curve for both models exceeded 0.80, indicating excellent discriminatory ability. Furthermore, the combined model demonstrated superior predictive accuracy compared to the preoperative model. CONCLUSION:This study developed preoperative and combined nomograms that offer practical and innovative tools for clinicians to predict POP risk and improve patient care.

Postoperative respiratory failure following cardiac surgery (CS-PRF) remains a critical complication with substantial morbidity and mortality. Current risk prediction models are limited by static assessments and suboptimal accuracy. This study aimed to develop and validate a dynamic, machine learning–based model to enhance perioperative risk stratification for CS-PRF. We retrospectively analyzed 1,016 adult patients who underwent cardiac surgery. Feature selection was conducted via the Least Absolute Shrinkage and Selection Operator (LASSO) and Boruta algorithms. Five machine learning models, including logistic regression, multilayer perceptron, extreme gradient boosting, categorical boosting, and deep neural network (DNN), were trained using preoperative and intraoperative variables. Model performance was evaluated by the area under the receiver operating characteristic curve (AUROC), area under the precision–recall curve (AUPRC), and calibration metrics. Model interpretability was evaluated via SHapley additive exPlanation (SHAP), and restricted cubic spline (RCS) analyses were used to explore nonlinear associations. The incidence of CS-PRF was 16.3

Background:The impact of adjusting intraoperative mean arterial pressure (MAP) management strategies on early detection and prevention of acute kidney injury (AKI) after Type A acute aortic dissection (TA-AAD) repair has not been elucidated. This study sought to investigate the association between different degrees of hypotension exposure and stage 3 AKI, and examine how intraoperative time-series dynamic variables influence the performance of risk stratification models. Methods:We analyzed intraoperative data and divided 336 adult patients into groups based on MAP below different thresholds (< 65, 60, 55, 50 mmHg). Logistic regression algorithms were used to identify and screen for predictors other than blood pressure (BP) features and develop initial model. Hypotensive exposure indicators, including cumulative time and area under the curve below a certain MAP threshold, were considered for confounding correction. Subsequently all independent predictors were incorporated to develop upgraded models. Predictive performance was assessed by area under the receiver operating characteristic curve (AUROC) and calibration curves. Results:227 patients (67.6 %) developed postoperative AKI, including 114 (33.9 %) stage 1, 54 (16.1 %) stage 2, and 59 (17.6 %) stage 3. Multivariate logistic regression analysis identified preoperative serum creatinine (odds ratio [OR] = 1.007 [95 % CI 1.002-1.015], P = 0.047), operation duration (OR = 1.007 [95 % CI, 1.002-1.012], P = 0.008) and intraoperative urine output (OR = 0.576 [95 % CI, 0.417-0.768], P < 0.001) as independent predictors of stage 3 AKI. After confounding correction, hypotensive exposure indicators were significant at all four thresholds, and ORs increased with decreasing thresholds. Integrating BP features yielded eight upgraded models with AUROC ranging from 0.797 to 0.805. Conclusions:With a worsening degree of intraoperative hypotension, i.e., lower absolute MAP thresholds and longer exposure times, the odds of stage 3 AKI risk after TA-AAD repair increased. Incorporating BP time-series variables into models could improve the accuracy of early prediction, while the two presentations of hypotension features yield scarcely any difference in predictive outcomes.

OBJECTIVE:Sepsis-induced encephalopathy is a critical determinant of mortality, driven by microglial activation and excessive autophagy. However, the underlying mechanisms remain unclear. METHODS:Sepsis was induced in wild-type and nuclear factor of activated T cells (NFAT) 1-deficient mice via cecal ligation and puncture. Hippocampal morphology, microglial autophagy, polarization, and inflammatory cytokine expressions were analyzed. Cognitive function was evaluated using the Morris water maze and fear conditioning tests. In vitro, BV2 microglia were stimulated with lipopolysaccharide (LPS), followed by genetic manipulation of calcineurin, NFAT1 or Smad2 to investigate underlying mechanisms. RESULTS:In wild-type mice, sepsis induced microglial autophagy, M1 polarization and neuroinflammation, resulting in cognitive impairment. These changes were accompanied by upregulated NFAT1 and elevated phosphorylation of Smad2 in hippocampal microglia. Notably, the sepsis-induced effects were attenuated by either pharmacological inhibition of autophagy (using 3-methyladenine) or genetic NFAT1 deficiency. Smad2 overexpression in NFAT1-deficient mice reversed sepsis-induced pathological phenotype, suggesting a functional dependency on Smad2 downstream of NFAT1. Corroborating the in vivo findings, in vitro experiments demonstrated that knockdown of calcineurin, NFAT1 or Smad2 suppressed LPS-induced autophagy and inflammatory responses in microglial cells. Furthermore, Smad2 overexpression rescued the effects of NFAT1 knockdown on LPS-exposure cells. CONCLUSION:The calcineurin/NFAT1 pathway may interact with Smad2 signaling promotes microglial autophagy during sepsis, exacerbating neuroinflammation and cognitive impairment. These findings support targeting this pathway for treating or even preventing sepsis-induced encephalopathy.

Background:Temperature variations during cardiopulmonary bypass (CPB) may significantly contribute to the development and progression of acute kidney injury (AKI) after cardiac surgery. We tested the hypothesis that temperature time-series variables during CPB are associated with AKI after cardiac surgery. Methods:We conducted a retrospective analysis of the data from 2041 patients. The primary outcome of interest in this study was cardiac surgery-associated AKI. By analyzing time-series nasopharyngeal temperature (Tnp) monitored intraoperatively with multiple categories, we obtained indicators reflecting the duration and depth of hypothermia. We used univariate and multivariate logistic regression analyses to identify perioperative factors associated with the outcome and construct a prediction model, which was evaluated in terms of discrimination and calibration. Results:Mild hypothermia (32-34 °C) was independently associated with a reduced risk of AKI compared to other temperature categories. The proportion of the duration with temperature at 32-34 °C during CPB (CPB_32-34 °C_DurProp) demonstrated the greatest predictive value for AKI compared to other temperature-related characteristics. We found age, preoperative creatinine, history of hypertension, intraoperative blood loss, intraoperative transfusion of allogeneic blood, and the use of left ventricular assist device (LVAD) were risk factors for the development of AKI. In contrast, preoperative hemoglobin, intraoperative urine output, and CPB_32-34 °C_DurProp were protective factors. Conclusions:The proportion of duration of mild hypothermia (32-34 °C) during CPB was identified as an independent risk factor for AKI after cardiac surgery. A prediction model incorporating this factor demonstrated good predictive performance. This emphasizes the importance of maintaining mild hypothermia during CPB in reducing the risk of AKI in postoperative cardiac surgery patients.

Postoperative delirium (POD) is a common and severe complication following cardiac surgery, with an incidence of about 65 https://baozexiang.shinyapps.io/dynnomapp/ ) was developed to enhance clinical risk assessment and patient outcomes. Predictive model for postoperative delirium (POD) in cardiac surgery patients, demonstrating strong accuracy and clinical utility, with a web-based tool for individualized risk assessment. Figure created with https://BioRender.com .

Postoperative Delirium (POD) has an incidence of up to 65 https://xxh152.shinyapps.io/Pre-POD/ and https://xxh152.shinyapps.io/Post-POD/ . We established two nomogram models based on the preoperative and postoperative time points to predict POD risk and guide the flexible implementation of possible interventions at different time points.

OBJECTIVE:Postoperative delirium (POD) is strongly associated with poor early and long-term prognosis in cardiac surgery patients with cardiopulmonary bypass (CPB). This study aimed to develop dynamic prediction models for POD after cardiac surgery under CPB using machine learning (ML) algorithms.METHODS:From July 2021 to June 2022, clinical data were collected from patients undergoing cardiac surgery under CPB at Nanjing First Hospital. A dataset from the same center (October 2022 to November 2022) was also used for temporal external validation. We used ML and deep learning to build models in the training set, optimized parameters in the test set, and finally validated the best model in the validation set. The SHapley Additive exPlanations (SHAP) method was introduced to explain the best models.RESULTS:Of the 885 patients enrolled, 221 (25.0%) developed POD. 22 (22.0%) of 100 validation cohort patients developed POD. The preoperative and postoperative artificial neural network (ANN) models exhibited optimal performance. The validation results demonstrated satisfactory predictive performance of the ANN model, with area under the receiver operator characteristic curve (AUROC) values of 0.776 and 0.684 for the preoperative and postoperative models, respectively. Based on the ANN algorithm, we constructed dynamic, highly accurate, and interpretable web risk calculators for POD.CONCLUSIONS:We successfully developed online interpretable dynamic ANN models as clinical decision aids to identify patients at high risk of POD before and after cardiac surgery to facilitate early intervention or care.

BACKGROUND:Burst suppression (BS) is a specific electroencephalogram (EEG) pattern that may contribute to postoperative delirium and negative outcomes. Few prediction models of BS are available and some factors such as frailty and intraoperative hypotension (IOH) which have been reported to promote the occurrence of BS were not included. Therefore, we look forward to creating a straightforward, precise, and clinically useful prediction model by incorporating new factors, such as frailty and IOH. MATERIALS AND METHODS:We retrospectively collected 540 patients and analyzed the data from 418 patients. Univariate analysis and backward stepwise logistic regression were used to select risk factors to develop a dynamic nomogram model, and then we developed a web calculator to visualize the process of prediction. The performance of the nomogram was evaluated in terms of discrimination, calibration, and clinical utility. RESULTS:According to the receiver operating characteristic (ROC) analysis, the nomogram showed good discriminative ability (AUC = 0.933) and the Hosmer-Lemeshow goodness-of-fit test demonstrated the nomogram had good calibration (p = 0.0718). Age, Clinical Frailty Scale (CFS) score, midazolam dose, propofol induction dose, total area under the hypotensive threshold of mean arterial pressure (MAP_AUT), and cerebrovascular diseases were the independent risk predictors of BS and used to construct nomogram. The web-based dynamic nomogram calculator was accessible by clicking on the URL: https://eegbsnomogram.shinyapps.io/dynnomapp/ or scanning a converted Quick Response (QR) code. CONCLUSIONS:Incorporating two distinctive new risk factors, frailty and IOH, we firstly developed a visualized nomogram for accurately predicting BS in non-cardiac surgery patients. The model is expected to guide clinical decision-making and optimize anesthesia management.