Machine-learning-based online prediction models for new-onset atrial fibrillation in ICU patients with severe sepsis: development and validation in a retrospective cohort

Machine-learning-based online prediction models for new-onset atrial fibrillation in ICU patients with severe sepsis: development and validation in a retrospective cohort | 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 Machine-learning-based online prediction models for new-onset atrial fibrillation in ICU patients with severe sepsis: development and validation in a retrospective cohort Fengying Sun, Jian Ouyang, Minmin Xiao, Jiajia Zhou, Jian Zhang, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9208556/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 New-onset atrial fibrillation (NOAF) is a common complication in patients with severe sepsis admitted to the intensive care unit (ICU), and is associated with increased in-hospital mortality. Therefore, early identification and prediction of patients at high risk for NOAF are of great significance. This study aims to establish and validate an interpretable machine learning (ML) model for the prediction of NOAF in critically ill patients with severe sepsis. Methods Data from patients with severe sepsis were extracted from three datasets: the Medical Information Mart for Intensive Care (MIMIC)-IV database, the MIMIC-III database, and the eICU Collaborative Research Database (eICU). The MIMIC-IV cohort was randomly split into a training set (70%) and an internal validation set (30%). The MIMIC-III and eICU databases served as two external validation cohorts. Feature selection was performed using a combination of the LASSO regression and the Boruta algorithm. Subsequently, ten ML algorithms were used to construct prediction models. The SHapley Additive exPlanations (SHAP) method was applied to interpret the model outputs visually. Results A total of 7,168 critically ill patients with severe sepsis were included in this study, with 3,993 from the MIMIC-IV database, 1,604 from the MIMIC-III database, and 1,571 from the eICU database. 19 features were selected from 88 variables for model construction. The Random Forest (RF) method demonstrated the most optimal predictive performance in terms of discriminaiton, calibration, and clinical utility among the ten models, achieving an AUC of 0.936 in the internal validation set, 0.779 in the MIMIC-III cohort, and 0.724 in the eICU cohort. SHAP analysis revealed that age, urine output, and respiratory failure were the most important contributors to the most. The simplified model incorporating 10 features demonstrated comparable efficacy, with an AUC of 0.907 in the internal validation set, 0.769 in the MIMIC-III cohort, and 0.707 in the eICU cohort. The final optimized model has been translated into a user-friendly interface for clinical use. The web application is accessible online at http://119.3.41.228/mm/index.php . Conclusions This study developed an interpretable RF model to accurately predict the NOAF risk in ICU patients with severe sepsis. Clinical trial number Not applicable. Severe sepsis New-onset atrial fibrillation intensive care unit Risk prediction MIMIC database Machine learning models Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Introduction Sepsis is defined by life-threatening organ dysfunction during infection and is the leading cause of death in hospitals[ 1 ]. New-onset atrial fibrillation (NOAF), defined as atrial fibrillation (AF) occurring in patients without a prior AF history, is one of the most common complications in patients with sepsis. Sepsis-induced systemic inflammation, myocardial stress, electrolyte imbalances, and autonomic dysfunction make septic patients highly susceptible to AF[ 2 – 5 ]. The incidence of NOAF in patients with sepsis ranges from 5% to 25%[ 6 – 8 ]. Notably, the incidence of NOAF among patients with severe sepsis is remarkably high, reaching 22% in severe sepsis and up to 40% in septic shock[ 6 ]. Patients with severe sepsis are defined as a subset of patients with sepsis, with hospital mortality rates ranging from 18% to 50%. Multiple studies have demonstrated a strong association between NOAF during severe sepsis and prolonged hospitalization, elevated risk of in-hospital death and intensive care unit (ICU) mortality, and a higher incidence of ischemic stroke[ 6 , 9 – 11 ]. Consequently, early identification of and intervention for AF in sepsis patients are crucial for minimizing adverse hospital outcomes. Machine learning (ML), as an emerging technological paradigm, has been extensively used in predictive models of various diseases and exhibits good clinical application value[ 12 – 15 ]. In recent years, several ML-based clinical prediction models have been developed for predicting the risk of AF at the individual level. For example, the XGBoost model developed by Guan and colleagues predicts the risk of NOAF in critically ill patients based on variables such as age, weight, blood urea nitrogen, percutaneous arterial oxygen saturation, urine output, sepsis, mechanical ventilation, and continuous renal replacement therapy[ 16 ]. Hou et al. built a LightGBM model to evaluate the risk of NOAF in coronary heart disease by incorporating patient-level clinical factors[ 17 ]. These ML models have exhibited robust predictive capabilities within their target populations. It is noteworthy that there is currently no ML-based AF risk prediction model specifically designed for the population of severe sepsis patients in the ICU. Despite the high accuracy achieved by ML models, they are often criticized for their “black-box” nature, as the influence of individual variables on these models often remains largely uninterpretable, which limits their clinical application[ 18 ]. The SHapley Additive exPlanations (SHAP) algorithm assigns an SHAP value to each feature in model predictions. The absolute value of SHAP measures the significance of each feature in the model’s decision-making process, thereby enhancing the model’s interpretability. In this study, we aim to construct an efficient ML model to predict the NOAF risk in ICU patients with severe sepsis based on three datasets, and to visually interpret the model using SHAP methods. This model aims to enable the early identification of high-risk groups, assisting clinicians in making appropriate decisions and implementing strategies in a timely manner, thereby improving patient outcomes. Methods Data source This study is a retrospective cohort analysis based on three datasets from the MIMIC-IV database, MIMIC-III database, and eICU database. The MIMIC is a large, publicly available critical care database that contains de-identified electronic health records from Beth Israel Deaconess Medical Center (BIDMC). The MIMIC-III (V.1.4) contains 58,976 admissions for 46,520 patients collected from June 2001 to October 2012, while MIMIC-IV (V.3.1) includes 546,028 admissions for 364,627 patients between 2008 and 2022. The eICU database (version 2.0) is constructed from a multi-center cohort encompassing over 200,000 ICU patients from 208 hospitals throughout the United States during the 2014–2015 period. These databases provide comprehensive clinical data such as demographics, vital signs, laboratory results, medications, surgical procedures, disease diagnoses, survival outcomes, and more. Data extraction was conducted through Structured Query Language (SQL) queries following authorization granted upon completion of the CITI Program certification (Record ID: 67165439). Since all patient data were anonymized, the study was exempt from ethical approval and individual consent requirements. Study population Patients with a diagnosis of severe sepsis according to the International Classification of Diseases, 9th Revision (ICD-9) code ‘99592’ or the International Classification of Diseases, 10th Revision (ICD-10) code ‘R6520, R6521’ who were admitted to the ICU were recruited. For patients with multiple ICU admissions, only the first admission records to the ICU were included. Patients were excluded if they met any of the following criteria: (1) patients under the age of 18 years; (2) patients with ICU length of stay less than 48 hours; (3) patients with a history of AF; (4) AF as an admission diagnosis; (5) diagnosis of AF within 24 hours of ICU admission; and (6) patients with more than 20% missing data. The patient screening process is shown in Fig. 1 . Data collection This study extracted the following information: (1) demographic features, including admission age, sex, admission weight, length of hospital stay, and ICU stay; (2) laboratory indicators, including hematocrit (HCT), hemoglobin (HGB), platelet count (PLT), white blood cell count (WBC), anion gap, bicarbonate, blood urea nitrogen (BUN), calcium, chloride, creatinine, glucose, sodium, potassium, international normalized ratio (INR), prothrombin time (PT), partial thromboplastin time (PTT), alanine aminotransferase (ALT), alkaline phosphatase (ALP), aspartate aminotransferase (AST), total bilirubin (TBIL), red blood cell count (RBC), mean corpuscular volume (MCV), mean corpuscular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), and red cell distribution width (RDW); (3) vital signs, including heart rate (HR), systolic blood pressure (SBP), diastolic blood pressure (DBP), mean blood pressure (MBP), respiratory rate (RR), temperature, and oxygen saturation (SpO2); (4) severity scores, including Glasgow Coma Scale (GCS) and Sequential Organ Failure Assessment (SOFA); (5) therapeutics, including mechanical ventilation, continuous renal replacement therapy (CRRT), and vasopressors and sedatives use; (6) comorbidities, including hypertension, diabetes, respiratory failure, heart failure, kidney failure, chronic kidney disease (CKD), and peripheral vascular disease (PVD). For vital signs, we collected the maximum, minimum, and average values. For other variables measured repeatedly during hospitalization, the maximum and minimum values were extracted. All data were extracted within the first 24 hours of the patient’s admission to the ICU, with a total of 132 variables collected. Variables with missing data are a common occurrence in the MIMIC and eICU databases. Therefore, to minimize the impact of missing data on model construction, patients or variables with over 20% missing values were excluded, and the remaining missing values were imputated using the K-nearest neighbors method. Supplementary Figure S1 illustrates the missing data proportion for each variable. After preprocessing, data from 3,993 patients with 88 variables were extracted from MIMIC-IV for model development. Statistical analysis As all continuous variables were non-normally distributed, they were described using the median and IQR (25th percentile, 75th percentile). Categorical variables were presented as frequencies (n) and percentages (%). The Kruskal-Wallis H test was used to analyse differences in non-normally distributed continuous variables among the three groups, while the chi-square test was employed to compare categorical variables. All statistical tests were two-tailed, and P-values < 0.05 were considered statistically significant. R software (version 4.3.3) was used for all statistical analyses. Cohort division and feature selection To address class imbalance between the negative and positive groups, undersampling and the synthetic minority oversampling technique (SMOTE) were employed to resample the data to achieve balance. The balanced dataset was divided into training and internal validation sets at a 7:3 ratio. Feature selection was a key step in the model building process. In this study, LASSO regression and the Boruta algorithm were used to identify the most important features based on the training set. The Boruta method, based on random forest classification, selects the most important features by comparing the Z-score of each feature against that of its “shadow features”. LASSO regression performs feature selection and dimensionality reduction through L1 regularization, which shrinks the coefficients of less relevant variables to zero, retaining only features with significant contributions. The top 10 features identified by each method, along with the common features identified by both methods, were selected as the final feature set for model construction. Model construction and evaluation A total of ten ML algorithms were used to construct the prediction model, namely logistic regression (LR), support vector machine (SVM), gradient boosting machine (GBM), neural network (NN), random forest (RF), extreme gradient boosting (XGBoost), k-nearest neighbors (KNN), adaptive boosting (AdaBoost), light gradient boosting machine (LightGBM), and category boosting (CatBoost). The optimal parameters of each algorithm were obtained by grid search. The training set was used for model building, and the best model was validated in the internal validation set and the external validation set. Model performance was evaluated through multiple metrics such as area under the curve (AUC) of receiver operating characteristic (ROC), accuracy, sensitivity, specificity, precision, and F1 score. Additionally, we assessed the consistency between the prediction probabilities and the sample probabilities by the calibration curve, and compared the clinical benefit at different threshold probabilities of each model via decision curve analysis (DCA). The model with the best performance was selected as the final prediction model. The modeling process was performed in R software (version 4.3.3). SHapley Additive exPlanations analysis The SHapley Additive exPlanations (SHAP) method, based on the concept of Shapley values from cooperative game theory, enhanced the interpretability of the final model by quantifying the contribution of each feature to the model prediction. A SHAP value of a feature greater than 0 indicates that the feature has a positive impact on the prediction result; conversely, it has a negative impact. The global importance of each feature on the model’s predictions was visually demonstrated through SHAP visualization tools such as histograms and bee swarm plots. The importance of each feature was determined by the average absolute value of its SHAP value, where a larger absolute value of SHAP indicates a stronger impact of the feature on the prediction outcome. SHAP can also generate force plots and waterfall plots for individual samples to provide personalized interpretation of prediction results, thereby guiding clinical decision-making. This method enables clear identification of key predictive factors and their directional effects, making it particularly useful for explaining complex ML models like RF. Results Baseline characteristics Following the application of predefined inclusion and exclusion criteria, a total of 3,993 patients with severe sepsis were identified from the MIMIC-IV database for analysis. Among these, 716 patients (17.93%) developed NOAF after the first 24 hours of ICU admission. For external validation cohorts, the incidence of AF was 18.64% (299/1,604) in the MIMIC-III database and 16.49% (259/1,571) in the eICU database. Supplementary Table S1 summarizes the differences in baseline characteristics among patients in the derivation cohort (MIMIC-IV) and the two external validation cohorts (MIMIC-III and eICU). Table 1 summarizes the comparison of baseline demographic and clinical characteristics between the non-AF group and the NOAF group in the derivation cohort. Male patients were more likely to develop AF than female patients during ICU admission. Patients who suffered from AF tended to be older and heavier. In addition, these patients had longer hospital and ICU stays, higher in-hospital mortality rates and SOFA score, higher rate of mechanical ventilation, vasopressor, and sedatives use, and higher respiratory failure and kidney failure risks compared with those without AF. Laboratory and vital signs assessments revealed that the maximum values of HCT, HGB, anion gap, BUN, calcium, creatinine, glucose, potassium, INR, PT, PTT, ALT, AST, RBC, and MCV; the minimum values of HCT, HGB, anion gap, BUN, creatinine, potassium, ALT, and AST; and the mean values of RR were much higher in severe sepsis with AF compared with those without AF (P < 0.05). However, the maximum values of PLT, WBC, chloride, ALP, and HR; the minimum values of PLT, WBC, ALP, MCHC, HR, SBP, DBP, temperature, and SpO2; the mean values of HR, DBP, temperature, and SpO2, and urine output were lower than those in the non-AF group (P < 0.05). Table 1 Comparison of baseline characteristics of patients with severe sepsis with and without NOAF. Characteristic Non-NOAF (n = 3277) NOAF (n = 716) P Admission age, M (Q1, Q3) 63.24 (51.99, 73.65) 70.40 (62.15, 79.87) < 0.001 Admission weight, M (Q1, Q3) 77.90 (64.25, 94.20) 80.90 (68.13, 98.30) < 0.001 Gender, n (%) < 0.001 Female 1532 (46.75) 266 (37.15) Male 1745 (53.25) 450 (62.85) Los hospital, M (Q1, Q3) 13.53 (7.75, 24.00) 16.59 (9.51, 28.25) < 0.001 Los icu, M (Q1, Q3) 5.13 (3.01, 10.01) 9.79 (5.67, 17.75) < 0.001 Hospital expire flag, n (%) < 0.001 No 2387 (72.84) 373 (52.09) Yes 890 (27.16) 343 (47.91) RBC_min, M (Q1, Q3) 3.13 (2.66, 3.67) 3.20 (2.71, 3.74) 0.085 RBC_max, M (Q1, Q3) 3.46 (2.99, 3.99) 3.58 (3.07, 4.11) 0.001 HGB_min, M (Q1, Q3) 9.20 (7.80, 10.80) 9.40 (8.00, 11.00) 0.009 HGB_max, M (Q1, Q3) 10.60 (9.20, 12.40) 11.00 (9.50, 12.70) 0.004 HCT_min, M (Q1, Q3) 28.40 (24.00, 33.10) 28.95 (24.23, 34.18) 0.012 HCT_max, M (Q1, Q3) 32.90 (28.55, 38.00) 34.00 (29.20, 39.38) 0.002 MCV_min, M (Q1, Q3) 91.00 (87.00, 96.00) 92.00 (87.00, 96.00) 0.064 MCV_max, M (Q1, Q3) 93.00 (88.00, 98.00) 94.00 (89.00, 98.00) 0.003 MCH_min, M (Q1, Q3) 29.70 (28.10, 31.30) 30.00 (28.30, 31.50) 0.156 MCH_max, M (Q1, Q3) 30.30 (28.50, 32.00) 30.50 (28.90, 32.20) 0.072 MCHC_min, M (Q1, Q3) 32.20 (31.10, 33.20) 32.00 (31.00, 33.10) 0.026 MCHC_max, M (Q1, Q3) 33.00 (31.80, 34.10) 32.80 (31.80, 34.00) 0.293 RDW_min, M (Q1, Q3) 15.20 (14.00, 17.20) 15.00 (13.90, 17.00) 0.281 RDW_max, M (Q1, Q3) 15.60 (14.30, 17.60) 15.40 (14.20, 17.50) 0.696 WBC_min, M (Q1, Q3) 11.00 (6.60, 16.10) 10.50 (6.13, 15.68) 0.018 WBC_max, M (Q1, Q3) 16.20 (10.70, 23.20) 15.40 (9.80, 21.40) 0.011 PLT_min, M (Q1, Q3) 162.00 (97.00, 245.00) 149.00 (90.25, 212.00) < 0.001 PLT_max, M (Q1, Q3) 209.00 (132.00, 307.00) 194.00 (128.00, 268.00) 0.001 Anion gap_min, M (Q1, Q3) 13.00 (11.00, 16.00) 14.00 (11.00, 17.00) < 0.001 Anion gap_max, M (Q1, Q3) 17.00 (14.00, 21.00) 18.00 (15.00, 22.00) < 0.001 Bicarbonate_min, M (Q1, Q3) 19.00 (16.00, 22.00) 18.00 (15.00, 22.00) 0.069 Bicarbonate_max, M (Q1, Q3) 22.00 (19.00, 25.00) 22.00 (19.00, 25.00) 0.326 BUN_min, M (Q1, Q3) 24.00 (14.00, 40.00) 31.00 (20.00, 50.00) < 0.001 BUN_max, M (Q1, Q3) 30.00 (18.00, 49.00) 38.00 (25.00, 60.00) < 0.001 Calcium_min, M (Q1, Q3) 7.60 (7.10, 8.20) 7.70 (7.10, 8.30) 0.094 Calcium_max, M (Q1, Q3) 8.30 (7.80, 8.90) 8.40 (7.90, 9.00) 0.015 Chloride_min, M (Q1, Q3) 101.00 (97.00, 106.00) 101.00 (97.00, 105.00) 0.069 Chloride_max, M (Q1, Q3) 106.00 (101.00, 111.00) 105.00 (101.00, 110.00) 0.026 Creatinine_min, M (Q1, Q3) 1.10 (0.70, 1.90) 1.40 (0.90, 2.40) < 0.001 Creatinine_max, M (Q1, Q3) 1.50 (0.90, 2.50) 1.90 (1.20, 3.10) < 0.001 Glucose_min, M (Q1, Q3) 109.00 (89.00, 135.00) 110.50 (89.00, 142.00) 0.223 Glucose_max, M (Q1, Q3) 158.00 (123.50, 217.00) 165.00 (129.00, 226.75) 0.015 Sodium_min, M (Q1, Q3) 136.00 (132.00, 139.00) 136.00 (132.00, 139.00) 0.510 Sodium_max, M (Q1, Q3) 139.00 (136.00, 143.00) 140.00 (136.00, 143.00) 0.927 Potassium_min, M (Q1, Q3) 3.80 (3.40, 4.20) 4.00 (3.50, 4.40) < 0.001 Potassium_max, M (Q1, Q3) 4.50 (4.10, 5.10) 4.70 (4.30, 5.30) < 0.001 ALT_min, M (Q1, Q3) 29.00 (17.00, 53.00) 31.00 (19.00, 61.00) < 0.001 ALT_max, M (Q1, Q3) 30.00 (19.00, 66.00) 38.00 (22.00, 87.75) < 0.001 ALP_min, M (Q1, Q3) 89.00 (65.00, 135.00) 78.00 (58.00, 121.75) < 0.001 ALP_max, M (Q1, Q3) 98.00 (71.00, 156.00) 87.50 (63.00, 138.75) < 0.001 AST_min, M (Q1, Q3) 40.00 (24.00, 83.00) 51.00 (29.00, 96.00) < 0.001 AST_max, M (Q1, Q3) 50.00 (28.00, 110.50) 62.00 (35.00, 136.75) < 0.001 TBIL_min, M (Q1, Q3) 0.70 (0.40, 1.65) 0.70 (0.40, 1.50) 0.238 TBIL_max, M (Q1, Q3) 0.80 (0.50, 2.20) 0.90 (0.50, 2.00) 0.449 INR_min, M (Q1, Q3) 1.30 (1.20, 1.60) 1.30 (1.20, 1.60) 0.488 INR_max, M (Q1, Q3) 1.50 (1.20, 1.90) 1.50 (1.30, 2.00) 0.017 PT_min, M (Q1, Q3) 14.60 (12.80, 17.50) 14.40 (12.80, 17.08) 0.237 PT_max, M (Q1, Q3) 15.90 (13.60, 20.10) 16.10 (13.80, 21.78) 0.026 PTT_min, M (Q1, Q3) 30.60 (27.10, 36.40) 31.10 (27.30, 36.80) 0.384 PTT_max, M (Q1, Q3) 35.10 (29.70, 48.75) 39.80 (30.63, 59.98) < 0.001 Urine output, M (Q1, Q3) 1265.00 (696.00, 2120.00) 928.50 (454.25, 1642.50) < 0.001 HR_min, M (Q1, Q3) 76.00 (65.00, 88.00) 74.00 (62.00, 85.00) 0.001 HR_max, M (Q1, Q3) 111.00 (97.00, 124.00) 106.00 (92.00, 123.00) < 0.001 HR_mean, M (Q1, Q3) 91.07 (79.61, 103.70) 88.89 (76.17, 101.17) 0.001 RR_min, M (Q1, Q3) 13.00 (11.00, 16.00) 13.00 (10.00, 16.00) 0.629 RR_max, M (Q1, Q3) 30.00 (25.50, 34.00) 30.00 (26.00, 34.00) 0.993 RR_mean, M (Q1, Q3) 20.95 (18.00, 24.18) 21.52 (18.46, 24.55) 0.009 SBP_min, M (Q1, Q3) 82.00 (75.00, 90.00) 82.00 (73.00, 90.00) 0.036 SBP_max, M (Q1, Q3) 140.00 (127.00, 155.00) 142.00 (129.00, 156.75) 0.095 SBP_mean, M (Q1, Q3) 107.90 (101.30, 116.06) 107.83 (101.36, 115.74) 0.850 DBP_min, M (Q1, Q3) 43.00 (37.00, 49.00) 42.00 (36.00, 47.88) 0.002 DBP_max, M (Q1, Q3) 83.00 (73.00, 96.00) 83.00 (71.00, 95.00) 0.098 DBP_mean, M (Q1, Q3) 59.08 (53.90, 64.91) 57.80 (52.19, 63.68) < 0.001 MBP_min, M (Q1, Q3) 55.00 (48.00, 61.00) 55.00 (47.00, 60.00) 0.076 MBP_max, M (Q1, Q3) 99.00 (88.00, 113.00) 98.00 (88.00, 113.00) 0.965 MBP_mean, M (Q1, Q3) 72.86 (68.42, 78.56) 72.51 (67.91, 77.83) 0.278 Temperature_min, M (Q1, Q3) 36.50 (36.17, 36.72) 36.44 (36.00, 36.72) 0.008 Temperature_max, M (Q1, Q3) 37.50 (37.06, 38.29) 37.50 (37.06, 38.17) 0.303 Temperature_mean, M (Q1, Q3) 36.95 (36.65, 37.37) 36.92 (36.60, 37.31) 0.040 SpO2_min, M (Q1, Q3) 92.00 (89.00, 94.00) 91.00 (88.00, 94.00) < 0.001 SpO2_max, M (Q1, Q3) 100.00 (100.00, 100.00) 100.00 (99.00, 100.00) 0.079 SpO2_mean, M (Q1, Q3) 96.93 (95.51, 98.33) 96.71 (95.03, 98.20) 0.004 SOFA, M (Q1, Q3) 8.00 (5.00, 11.00) 9.00 (6.00, 11.00) < 0.001 GCS, M (Q1, Q3) 15.00 (13.00, 15.00) 15.00 (14.00, 15.00) 0.090 MV, n (%) < 0.001 No 1712 (52.24%) 274 (38.27) Yes 1565 (47.76) 442 (61.73) CRRT, n (%) 0.172 No 3104 (94.72) 669 (93.44) Yes 173 (5.28) 47 (6.56) Vasopressors, n (%) 0.023 No 1289 (39.33) 249 (34.78) Yes 1988 (60.67) 467 (65.22) sedatives, n(%) < 0.001 No 1518 (46.32) 252 (35.20) Yes 1759 (53.68) 464 (64.80) Hypertension, n(%) 0.156 No 1830 (55.84) 379 (52.93) Yes 1447 (44.16) 337 (47.07) Diabetes 0.003 No 2191 (66.86) 437 (61.03) Yes 1086 (33.14) 279 (38.97) Respiratory failure, n(%) < 0.001 No 1392 (42.48) 197 (27.51) Yes 1885 (57.52) 519 (72.49) Heart failure, n(%) < 0.001 No 2542 (77.57) 442 (61.73) Yes 735 (22.43) 274 (38.27) Kidney failure, n(%) < 0.001 No 1086 (33.14) 152 (21.23) Yes 2191 (66.86) 564 (78.77) CKD < 0.001 No 2596 (79.22) 493 (68.85) Yes 681 (20.78) 223 (31.15) PVD, n (%) 0.257 No 3180 (97.04) 689 (96.23) Yes 97 (2.96) 27 (3.77) M: Median, Q1: 1st Quartile, Q3: 3st Quartile NOAF New-onset atrial fibrillation; Los: length of stay; RBC: red blood cell; HGB: hemoglobin; HCT: hematocrit; MCV: mean corpuscular volume; MCH: mean corpuscular hemoglobin; MCHC: mean corpuscular hemoglobin concentration; RDW: red cell distribution width; WBC: white blood cell; PLT: platelet; BUN: blood urea nitrogen; AST: aspartate aminotransferase; ALP: alkaline phosphatase; ALT: alanine aminotransferase; TBIL: total bilirubin; INR: international normalized ratio; PT: prothrombin time; PTT: partial thromboplastin time; HR: heart rate; RR: respiratory rate; SBP: systolic blood pressure; DBP: diastolic blood pressure; MBP: mean blood pressure; SpO2: percutaneous arterial oxygen saturation; GCS: Glasgow Coma Scale; SOFA: Sequential Organ Failure Assessment; CRRT: continuous renal replacement therapy; MV: mechanical ventilation; CKD: chronic kidney disease; PVD: peripheral vascular disease. Feature selection Two methods were used for feature screening. The results of LASSO regression and Boruta algorithm analysis are presented in Fig. 2 . Thirty-two out of 88 variables were identified to be closely associated with NOAF using the Boruta algorithm (Fig. 2 A). Based on variable importance, the top ten features were as follows: age, AST_max, AST_min, ALP_max, PT_max, PTT_max, ALP_min, INR_max, ALT_max, and ALT_min. In the LASSO regression, 16 variables were selected as potential predictors. The coefficient profile and cross-validated error plot of the LASSO model are shown in Fig. 2 B and C, respectively, with the full list of coefficients provided in Supplementary Table S2. Based on the ranking of variable importance according to the absolute values of LASSO regression coefficients, the ten variables most closely associated with NOAF were identified: heart failure, respiratory failure, mechanical ventilation, sedatives, gender, age, SpO2_mean, potassium_min, PTT_max, and BUN_min. According to the feature selection results, we constructed the final feature set by merging the features common to both methods and the top ten features from each method. After removing duplicates, we obtained 19 features for model development (Supplementary Figure S2). These selected features were mechanical ventilation, heart failure, age, BUN_min, PT_max, PTT_max, urine output, respiratory failure, sedatives, gender, SpO2_mean, potassium_min, AST_max, AST_min, ALP_max, ALP_min, INR_max, ALT_max, and ALT_min. Model development and evaluation Using LASSO regression and Boruta algorithm, we selected 19 key features from 88 clinical variables associated with the occurrence of AF. Based on these features, we developed ten ML models to predict the risk of NOAF in severe sepsis patients in the ICU using the training data set. The performance of these models was evaluated on the internal validation set. The ROCs of the ten ML models are presented in Fig. 3 . Among the ten models, the RF model demonstrated the highest predictive performance (AUC = 0.936, 95% confidence interval (CI): 0.921–0.951). The AdaBoost model exhibited comparable efficacy to the RF model, with an AUC of 0.920 (95% CI: 0.902–0.937), followed by SVM (AUC = 0.905, 95% CI: 0.886–0.924), GBM (AUC = 0.893, 95% CI: 0.870–0.915), and XGBoost (AUC = 0.861, 95% CI: 0.836–0.886) models. The remaining models showed relatively lower generalization ability, ranked in descending order of performance as follows: KNN (AUC = 0.814, 95% CI: 0.788–0.839), LightGBM (AUC = 0.797, 95% CI: 0.767–0.827), CatBoost (AUC = 0.793, 95% CI: 0.763–0.823), LR (AUC = 0.748, 95% CI: 0.716–0.780) and NN (AUC = 0.740, 95% CI: 0.707–0.773) models. A set of detailed performance metrics for the ten ML models is shown in Table 2 , among which the RF model exhibited superior overall performance with the highest accuracy (0.851), sensitivity (0.902), F1 score (0.858), and the third highest precision (0.818) and the fourth highest specificity (0.800). The calibration curves revealed that the RF model had excellent agreement between predicted probabilities and actual probabilities (Fig. 4 A). According to the DCA curves (Fig. 4 B), the RF model obtained a higher net benefit across a wider range of threshold probabilities than other models, indicating its superior clinical utility. Therefore, the RF model was ultimately selected as the optimal model for predicting NOAF. Table 2 Performances of the ten ML models for predicting NOAF in internal validation set. Model Threshold AUC (95% CI) Accuracy Sensitivity Specificity Precision F1 score LR 0.496 AUC = 0.748 (0.716–0.780) 0.671 0.697 0.646 0.663 0.680 SVM 0.542 AUC = 0.905 (0.886–0.924) 0.781 0.695 0.867 0.839 0.760 GBM 0.500 AUC = 0.893 (0.870–0.915) 0.825 0.844 0.807 0.813 0.828 NN 0.517 AUC = 0.740 (0.707–0.773) 0.698 0.772 0.625 0.673 0.719 RF 0.500 AUC = 0.936 (0.921–0.951) 0.851 0.902 0.800 0.818 0.858 XGBoost 0.514 AUC = 0.861 (0.836–0.886) 0.782 0.795 0.769 0.775 0.785 KNN 0.500 AUC = 0.814 (0.788–0.839) 0.814 0.862 0.765 0.786 0.822 AdaBoost 0.500 AUC = 0.920 (0.902–0.937) 0.843 0.879 0.807 0.820 0.848 LightGBM 0.543 AUC = 0.797 (0.767–0.827) 0.749 0.713 0.786 0.769 0.740 CatBoost 0.629 AUC = 0.793 (0.763–0.823) 0.721 0.730 0.713 0.718 0.724 NOAF: new-onset atrial fibrillation, ML: machine learning; AUC: area under the curve; CI: confidence interval; LR: logistic regression; SVM: support vector machine; GBM: gradient boosting machine; NN: neural network; RF: random forest; XGBoost: extreme gradient boosting; KNN: k-nearest neighbors; AdaBoost: adaptive boosting; LightGBM: light gradient boosting machine; CatBoost: category boosting. Model interpretability The SHAP algorithm was used to visually exhibit the contributions of each feature in the RF model for predicting the risk of NOAF in severe sepsis patients in the ICU. Figure 5 A shows the feature importance plot, which revealed that age, urine output, respiratory failure, BUN_min, mechanical ventilation, PTT_max, heart failure, SpO2_mean, potassium_min, and sedatives were the ten most important variables in predicting NOAF. Figure 5 B presents a comprehensive beeswarm plot illustrating the features in the RF model. Each data point represented a sample, with a yellow-to-purple color gradient indicating the magnitude of the feature value, where yellow denoted a higher value and purple denoted a lower value. Baseline variables with higher SHAP values contributed to a higher risk of developing NOAF during ICU hospitalization. According to the SHAP beeswarm plots, older patients with respiratory failure and heart failure, those receiving sedatives and mechanical ventilation, and those who presented with low urine output, low SpO2_mean, high PTT_max, high BUN_min, and high potassium_min were at significantly increased risk of developing NOAF during their ICU hospitalization. To facilitate a granular understanding of the model’s decision logic at the individual level, we performed an interpretability analysis on two representative cases, as illustrated in Fig. 5 C and D. The SHAP values were used to show the magnitude and direction of each feature’s contribution to the prediction. Features with positive values increase the AF risk, while those with negative values decrease the AF risk. Each feature contributed to varying degrees to the predicted probability of the outcome, and their combined effect determined the final result. The value f(x) represents the model’s predicted probability, whereas E[f(x)] is the expected model output without input features. For a patient who did not develop AF, our RF model predicted an f(x) close to 0 (Fig. 5 C), correctly indicating a low risk of the event. In contrast, for a patient who developed AF, the model predicted a high risk, with an f(x) value of 1 (Fig. 5 D). External validation To further verify the predictive accuracy of the selected RF model, two external validation cohorts were additionally incorporated: 1,604 patients with severe sepsis from MIMIC-III (Cohort 1) and 1,571 from the eICU database (Cohort 2). Their baseline characteristics are provided in Supplementary Tables S3 and S4, respectively. Notwithstanding the inherent differences in baseline characteristics among the three cohorts (Supplementary Table S1 ), the externally validated ROC curve achieved an AUC of 0.779 (95% CI: 0.749–0.808) in the MIMIC-III cohort and 0.724 (0.692–0.756) in the eICU cohort (Supplementary Figure S3), demonstrating the strong generalization capability of our model. Application of the model To improve the clinical applicability of our model and facilitate rapid clinical decision-making, we employed a recursive feature elimination (RFE) approach to refine the predictor set (Supplementary Figure S4). This method facilitated the identification of a minimal yet highly informative subset of variables, balancing interpretability and predictive power. Through this process, the model was optimized to include only 10 clinically relevant features: age, BUN_min, urine output, AST_min, SpO2_mean, AST_max, ALP_max, PTT_max, ALP_min, and ALT_min. The simplified model demonstrated robust discriminatory performance in both internal and external validation, with AUCs of 0.907 (95% CI: 0.887–0.926) in the internal validation set, 0.769 (95% CI: 0.739–0.799) in the MIMIC-III cohort, and 0.707 (95% CI: 0.676–0.739) (Fig. 6 ). The final simplified model was deployed as a web application designed to facilitate its use in clinical settings (Fig. 7 ). Users can automatically obtain prediction outcomes of NOAF risk by entering the actual values of the ten features required for the model into the designated text fields on the webpage. Furthermore, the platform will also generate a bar chart for the individual patient to display the important features that contribute to the decision of NOAF. Positive or negative values in bar chart indicate positive or negative contributions to the prediction. The web application can be freely accessed at http://119.3.41.228/mm/index.php . Discussion Despite considerable advances in medicine, sepsis is a leading cause of death in patients in the ICU. Patients with sepsis who develop AF often experience a worsening of their condition and have a poor prognosis. Studies have shown that sepsis patients with NOAF have a higher risk of mortality compared to those without AF, with mortality rates of 45.7% and 26.8%, respectively[ 19 ]. Therefore, it is of paramount importance to establish predictive tools to identify the risk of NOAF for improving patients’ prognosis. In this study, we constructed and validated an ML model for predicting the risk of AF in severe sepsis patients admitted to the ICU using three large-scale datasets. Based on 88 clinical variables extracted within the first 24 hours of ICU admission, we selected 19 key features through the use of LASSO regression combined with the Boruta algorithm and developed ten ML models using these 19 features. Our results showed that the RF model exhibited the best predictive performance among all ML models. Moreover, the RF model demonstrated good predictive accuracy in terms of discrimination and calibration, and provided a substantial net benefit in clinical practice. The results from two external validation cohorts provided additional confirmation of the stability and accuracy of the RF model for predicting the risk of AF in severe sepsis. Additionally, the SHAP method was employed to interpret and visualize the RF model, further improving its applicability in the clinical setting. According to SHAP histograms and bee swarm plots, the study identified ten key factors most closely associated with the risk of NOAF in patients with severe sepsis, including age, urine output, respiratory failure, BUN_min, mechanical ventilation, PTT_max, heart failure, SpO2_mean, potassium_min, and sedatives. We employed SHAP force plots to visualize individualized AF risk predictions, offering an intuitive interpretation of the model’s decision-making process in distinguishing high-risk cases. Previous studies have attempted to construct predictive models for AF risk in sepsis patients, and a retrospective study extracted electrocardiogram (ECG) data of 198 subjects from MIMIC-III database and modelled them using KNN algorithms, achieving sensitivity, specificity, and accuracy of 98.40%, 99.80%, and 99.32%, respectively[ 20 ]. Syed et al. utilized ECG data from 100 patients sourced from the AFPDB and MIMIC-III databases to develop a risk prediction model for NOAF in sepsis patients, with the SVM model reaching 80% sensitivity, 100% specificity, and 90% accuracy. Although both aforementioned predictive models demonstrated high accuracy, both studies were performed on small sample sizes and lacked external validation and interpretation of the models[ 21 ]. Jiming et al. used a multicenter dataset of 2,492 patients with sepsis from two hospitals in China to build a nomogram model for the prediction of NOAF based on clinical risk factors, achieving an AUC of 0.861 and 0.845 in the internal and external validation, respectively, but did not perform any model interpretation[ 22 ]. Although previous studies have attempted to construct predictive models for AF risk in patients with sepsis, most have not specifically targeted severe sepsis, and none have yet undergone clinical transformation, limiting their clinical utility. In contrast, our study is the first to build an interpretable RF model for specifically targeting the prediction of NOAF in patients with severe sepsis across three datasets. The model achieved an AUC of 0.936 (95% CI: 0.921–0.951) in the internal validation set, 0.779 (95% CI: 0.749–0.808) in the MIMIC-III cohort, and 0.724 (95% CI: 0.692–0.756) in the eICU cohort, demonstrating robust generalizability. Currently, the MIMIC database has been widely applied in the prediction of various diseases such as acute kidney injury, rhabdomyolysis, and hepatorenal syndrome[ 23 – 25 ]. Notably, compared to single-center datasets, the MIMIC databases provide large-scale, diverse patient data that enhance the robustness and generalizability of the predictive model, thereby improving the applicability of the model across varied healthcare settings. A key advantage of our study is the simplification of the model structure through RFE. The simplified version of the model required fewer input variables while maintaining similar accuracy, thereby improving its practicality and user-friendliness. Moreover, all the selected variables were readily available in most hospitals, enhancing our model’s potential for widespread application in medical institutions of different levels. We had also deployed the simplified model on a dedicated website to further facilitate its use in clinical settings. The user-friendly platform will assist clinicians in rapidly and accurately identifying patients at high risk of AF and optimizing treatment strategies based on predicted outcomes, thereby improving patient outcomes. SHAP values demonstrate that age is the most influential feature in the development of AF, consistent with previous studies[ 26 , 27 ]. The aging left atrium undergoes structural remodeling due to fatty infiltration, fibrosis, and loss of cardiomyocytes, leading to conduction block and multiple wavelet reentry. Concurrently, calcium homeostasis disorders such as cytoplasmic calcium overload and spontaneous calcium release can shorten action potentials and predispose to AF. Furthermore, abnormalities at the cellular and molecular levels, including ion channel dysfunction, autonomic imbalance, and oxidative stress, further increase atrial spontaneous excitability and predisposition to arrhythmias[ 28 ]. Among the comorbidities, respiratory failure, manifested by hypoxemia (SpO2 reduction) and an increased requirement for mechanical ventilation, had the strongest association with NOAF. In patients with type 2 respiratory failure, the incidence of AF is relatively high, approximately 50%[ 29 ]. Previous studies have shown that hypoxemia and hypercapnia caused by respiratory failure are significant factors contributing to the occurrence of AF. The mechanism is that hypoxemia and hypercapnia can lead to right ventricular overload and right atrial dilation by increasing pulmonary arterial pressure. These changes disrupt the hemodynamic balance of the heart, thereby contributing to the onset of AF[ 30 , 31 ]. Renal failure, characterized by elevated BUN and potassium levels and a reduction in urine output, is also an important predictor of NOAF. Patients with renal failure are prone to AF, potentially due to electrolyte imbalances, neurohormonal activation, and systemic inflammation/oxidative stress[ 32 ]. In addition, we noted that the AF prevalence in renal failure patients increased from 45.80% in 2008 to 65.51% in 2014, which may be attributed to improved survival of critically ill patients, enhanced detection methods, an aging population, and increased comorbidities[ 33 ]. We observed that AF was more frequently present in patients with severe sepsis who were administered sedative drugs than in those who were not. The potential mechanism by which sedative agents trigger the onset of AF could be as follows: (1) direct blockade of cardiac ion channels, leading to atrial conduction abnormalities; (2) drug-induced reflex sympathetic activation or a reflex increase in sympathetic tone, which promotes ectopic firing; and (3) anticholinergic effects through M₂ receptor antagonism, resulting in shortening of the atrial effective refractory period and facilitation of re-entrant circuits[ 34 ]. Our finding was consistent with previous studies showing that prolonged PTT was a risk factor for NOAF[ 27 ]. The underlying logic was that prolonged PTT indicated that the patient population was undergoing heparin treatment due to a high risk of thrombosis, and the underlying conditions necessitating anticoagulation (such as post-cardiac surgery, heart failure, sepsis, etc.) are precisely the classic triggers for AF[ 9 , 35 , 36 ]. Therefore, prolonged PTT was a consequence of anticoagulation therapy, not a cause of AF; it acted as an effective surrogate marker, identifying those high-risk populations with primary diseases that are highly prone to developing AF. Several limitations must be acknowledged in this study. First, the sample size of patients with AF in this large cohort of patients with severe sepsis was relatively modest, and the incidence of events was relatively low. These class-unbalanced data posed a challenge for ML. Although combined undersampling and oversampling techniques addressed class imbalance problem, the characteristics of the cohort may limit the predictive performance. Second, despite the use of data from MIMIC, their retrospective design potentially introduces selection bias arising from demographic variations and patient inclusion criteria. Furthermore, while the prediction model showed robust performance in two external validation cohorts (MIMIC-III and eICU), all data originated from the United States, which potentially limits the generalizability of our findings to other populations. Therefore, future external validation across multiple centers will be essential to further evaluate the model’s generalizability and clinical applicability. Finally, due to the limitations inherent in the MIMIC and eICU databases, the study lacked features related to cardiac function, such as brain natriuretic peptide and N-terminal pro-brain natriuretic peptide. The absence of these key variables may affect the comprehensiveness of our predictive model. Conclusion We developed an RF model to predict the risk of NOAF in patients with severe sepsis in the ICU, which may assist clinicians in tailoring management and implementing early interventions for these patients at risk of AF to improve outcomes. SHAP enhances the interpretability of the model, providing insights into the contribution of each variable to the model’s predictions. Additionally, the optimized compact model and a web-based application further improve the clinical applicability of our model. Abbreviations New-onset atrial fibrillation (NOAF), atrial fibrillation (AF), intensive care unit (ICU), machine learning (ML), Medical Information Mart for Intensive Care (MIMIC), SHapley Additive exPlanations (SHAP), receiver operating characteristic (ROC), area under the curve (AUC), random forest (RF), logistic regression (LR), support vector machine (SVM), gradient boosting machine (GBM), neural network (NN), extreme gradient boosting (XGBoost), k-nearest neighbors (KNN), adaptive boosting (AdaBoost), light gradient boosting machine (LightGBM), category boosting (CatBoost), decision curve analysis (DCA), hematocrit (HCT), hemoglobin (HGB), platelet count (PLT), white blood cell count (WBC), blood urea nitrogen (BUN), international normalized ratio (INR), prothrombin time (PT), partial thromboplastin time (PTT), alanine aminotransferase (ALT), alkaline phosphatase (ALP), aspartate aminotransferase (AST), total bilirubin (TBIL), red blood cell count (RBC), mean corpuscular volume (MCV), mean corpuscular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), red cell distribution width (RDW), heart rate (HR), systolic blood pressure (SBP), diastolic blood pressure (DBP), mean blood pressure (MBP), respiratory rate (RR), oxygen saturation (SpO2), Glasgow Coma Scale (GCS), Sequential Organ Failure Assessment (SOFA), continuous renal replacement therapy (CRRT), chronic kidney disease (CKD), and peripheral vascular disease (PVD). Declarations Acknowledgments The authors would like to thank the developers of the MIMIC and eICU databases for sharing their data resources. Additionally, we also acknowledge all individuals who contributed to this work and the financial support received from funding bodies. Author contributions FYS, MMX, and TLS conceived and designed the study. FYS, JZ, and JJZ conducted the data collection, data analysis, and data interpretation. FYS, JJZ, and JOY established the ML models and website. FYS, JZ, and TLS drafted the initial manuscript. TLS, MMX, and JZ revised the manuscript and performed some of the revised analyses. All authors read and approved the final manuscript. Funding This study was funded by the National Key Research and Development Program of China (No. 2023YFC2706503), the Shanghai Municipal Science and Technology Major Project (No. 2017SHZDZX01), the Cooperative Research Fund of the Affiliated Wuhu Hospital of East China Normal University (No. 40500-20104-222400), Beihang University & Capital Medical University Plan (No. BHME-201904), the Special Fund of the Pediatric Medical Coordinated Development Center of Beijing Hospitals Authority (No. XTCX201809), and the Open Research Fund of Key Laboratory of Advanced Theory and Application in Statistics and Data Science-MOE, ECNU and Key Laboratory of MEA, Ministry of Education, ECNU. Data availability The MIMIC-IV (V.3.1) and MIMIC-III (V.1.4) datasets used in this study are publicly available via the PhysioNet platform (https://physionet.org/) following authorization granted upon completion of the CITI Program certification. The data analyzed and the codes used during the current study are available from the corresponding author on reasonable request. Ethics approval and consent to participate Our research was based on the MIMIC database, a de-identified public resource. The creation of MIMIC database was approved by the institutional review boards of Beth Israel Deaconess Medical Center and the Massachusetts Institute of Technology, with a waiver of informed consent. The author (Fengying Sun) passed the Collaborative Institutional Training Initiative (CITI) program exam and obtained permission to access the database (Record ID: 67165439). Consequently, our analysis, which involved secondary use of this data, was exempt from further ethical approval and individual consent requirements. Clinical trial number Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests. References Singer M, Deutschman CS, Seymour CW, Shankar-Hari M, Annane D, Bauer M, et al. The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3). JAMA. 2016;315:801–10. Bai W, Liu ZQ, He PY, Muhuyati. The role of IL-6, IL-10, TNF-α and PD-1 expression on CD4 T cells in atrial fibrillation. Heliyon. 2023;9:e18818. Pinho-Gomes AC, Reilly S, Brandes RP, Casadei B. Targeting inflammation and oxidative stress in atrial fibrillation: role of 3-hydroxy-3-methylglutaryl-coenzyme a reductase inhibition with statins. Antioxid Redox Signal. 2014;20:1268–85. Rafaqat S, Rafaqat S, Khurshid H, Rafaqat S. Electrolyte's imbalance role in atrial fibrillation: Pharmacological management. Int J Arrhythmia. 2022;23:1–10. Carrara M, Ferrario M, Bollen Pinto B, Herpain A. The autonomic nervous system in septic shock and its role as a future therapeutic target: a narrative review. Ann Intensive Care. 2021;11:80. Klein Klouwenberg PM, Frencken JF, Kuipers S, Ong DS, Peelen LM, van Vught LA, et al. Incidence, Predictors, and Outcomes of New-Onset Atrial Fibrillation in Critically Ill Patients with Sepsis. A Cohort Study. Am J Respir Crit Care Med. 2017;195:205–11. Wetterslev M, Haase N, Hassager C, Belley-Cote EP, McIntyre WF, An Y, et al. New-onset atrial fibrillation in adult critically ill patients: a scoping review. Intensive Care Med. 2019;45:928–38. Fernando SM, Mathew R, Hibbert B, Rochwerg B, Munshi L, Walkey AJ, et al. New-onset atrial fibrillation and associated outcomes and resource use among critically ill adults-a multicenter retrospective cohort study. Crit Care. 2020;24:15. Corica B, Romiti GF, Basili S, Proietti M. Prevalence of New-Onset Atrial Fibrillation and Associated Outcomes in Patients with Sepsis: A Systematic Review and Meta-Analysis. J Pers Med. 2022;12. Liu YW, Wang YF, Chen Y, Dong R, Li S, Peng JM, et al. A nationwide study on new onset atrial fibrillation risk factors and its association with hospital mortality in sepsis patients. Sci Rep. 2024;14:12206. Huo Y, Yoshimura H, Gonzalez-Izquierdo A, Lip GYH, Schmidt F, Providencia R. Risk Factors and Prognosis of New-Onset Atrial Fibrillation in Sepsis: A Nationwide Electronic Health Record Study. JACC Adv. 2025;4:101681. Deng B, Zhao Z, Ruan T, Zhou R, Liu C, Li Q, et al. Development and external validation of a machine learning model for brain injury in pediatric patients on extracorporeal membrane oxygenation. Crit Care. 2025;29:17. Kwiendacz H, Huang B, Chen Y, Janota O, Irlik K, Liu Y, et al. Predicting major adverse cardiac events in diabetes and chronic kidney disease: a machine learning study from the Silesia Diabetes-Heart Project. Cardiovasc Diabetol. 2025;24:76. Lei J, Zhai J, Zhang Y, Qi J, Sun C. Supervised Machine Learning Models for Predicting Sepsis-Associated Liver Injury in Patients With Sepsis: Development and Validation Study Based on a Multicenter Cohort Study. J Med Internet Res. 2025;27:e66733. Chen H, Song H, Huang H, Fang X, Chen H, Yang Q, et al. Machine learning prediction and interpretability analysis of high-risk chest pain: a study from the MIMIC-IV database. Front Physiol. 2025;16:1594277. Guan C, Gong A, Zhao Y, Yin C, Geng L, Liu L, et al. Interpretable machine learning model for new-onset atrial fibrillation prediction in critically ill patients: a multi-center study. Crit Care. 2024;28:349. Hou L, Su K, Zhao J, He T, Li Y. Application of Multi-Inflammatory Index to Predict Atrial Fibrillation Risk in Patients with Coronary Heart Disease: A Retrospective Machine Learning Study. Risk Manag Healthc Policy. 2024;17:2907–15. Wiggerthale J, Reich C. Explainable Machine Learning in Critical Decision Systems: Ensuring Safe Application and Correctness. 2024. Meng H, Guo L, Pan Y, Kong B, Shuai W, Huang H. Machine learning based clinical prediction model for 1-year mortality in Sepsis patients with atrial fibrillation. Heliyon. 2024;10:e38730. Bashar SK, Hossain MB, Ding E, Walkey AJ, McManus DD, Chon KH. Atrial Fibrillation Detection During Sepsis: Study on MIMIC III ICU Data. IEEE J Biomed Health Inf. 2020;24:3124–35. Bashar SK, Ding EY, Walkey AJ, McManus DD, Chon KH. Atrial Fibrillation Prediction from Critically Ill Sepsis Patients. Biosens (Basel). 2021;11. Li Z, Pang M, Li Y, Yu Y, Peng T, Hu Z, et al. Development and validation of a predictive model for new-onset atrial fibrillation in sepsis based on clinical risk factors. Front Cardiovasc Med. 2022;9:968615. Liu C, Shi J, Wang F, Li D, Luo Y, Yang B, et al. Development and validation of an interpretable multi-task model to predict outcomes in patients with rhabdomyolysis: a multicenter retrospective cohort study. EClinicalMedicine. 2025;87:103438. Shi B, Ye J, Chen W, Liao B, Gu W, Yin H, et al. Prognosis of critically ill patients with early and late sepsis-associated acute kidney injury: an observational study based on the MIMIC-IV. Ren Fail. 2025;47:2441393. Yao F, Luo J, Zhou Q, Wang L, He Z. Development and validation of a machine learning-based prediction model for hepatorenal syndrome in liver cirrhosis patients using MIMIC-IV and eICU databases. Sci Rep. 2025;15:2743. Ding M, Murata S, Louro J, Hammar N, Modig K. Machine-learning approach to atrial fibrillation prediction among individuals without prior cardiovascular diseases. Open Heart. 2025;12. Alomari L, Jarrar Y, Al-Fakhouri Z, Otabor E, Lam J, Alomari J. A machine learning-based risk prediction model for atrial fibrillation in critically ill patients. Heart Rhythm O2. 2025;6:652–60. Hakim FA, Shen WK. Atrial fibrillation in the elderly: a review. Future Cardiol. 2014;10:745–58. Mentes O, Celik D, Yıldız M, Özdemir T, Ari M, Aksoy Güney EN et al. Atrial Fibrillation Among ICU Patients with Type 2 Respiratory Failure: Who Is at Risk and What Are the Outcomes? Diagnostics. (Basel). 2025;15. Bemis CE, Serur JR, Borkenhagen D, Sonnenblick EH, Urschel CW. Influence of right ventricular filling pressure on left ventricular pressure and dimension. Circ Res. 1974;34:498–504. Terzano C, Conti V, Di Stefano F, Petroianni A, Ceccarelli D, Graziani E, et al. Comorbidity, hospitalization, and mortality in COPD: results from a longitudinal study. Lung. 2010;188:321–9. Navarro-Garcia JA, Keefe JA, Song J, Li N, Wehrens XHT. Mechanisms underlying atrial fibrillation in chronic kidney disease. J Mol Cell Cardiol. 2025;205:37–51. Lin Z, Han H, Guo W, Wei X, Guo Z, Zhai S, et al. Atrial fibrillation in critically ill patients who received prolonged mechanical ventilation: a nationwide inpatient report. Korean J Intern Med. 2021;36:1389–401. Tamargo J, Villacastín J, Caballero R, Delpón E. Drug-induced atrial fibrillation. A narrative review of a forgotten adverse effect. Pharmacol Res. 2024;200:107077. Bessissow A, Khan J, Devereaux PJ, Alvarez-Garcia J, Alonso-Coello P. Postoperative atrial fibrillation in non-cardiac and cardiac surgery: an overview. J Thromb Haemost. 2015;13(Suppl 1):S304–312. Karnik AA, Gopal DM, Ko D, Benjamin EJ, Helm RH. Epidemiology of Atrial Fibrillation and Heart Failure: A Growing and Important Problem. Cardiol Clin. 2019;37:119–29. Additional Declarations No competing interests reported. Supplementary Files Additionalfile1.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 28 Apr, 2026 Reviewers agreed at journal 28 Apr, 2026 Reviewers invited by journal 21 Apr, 2026 Editor invited by journal 26 Mar, 2026 Editor assigned by journal 25 Mar, 2026 Submission checks completed at journal 25 Mar, 2026 First submitted to journal 24 Mar, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9208556","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":631382342,"identity":"d247d0b3-d320-4a72-93c8-c743aca9f580","order_by":0,"name":"Fengying Sun","email":"","orcid":"","institution":"The Second People’s Hospital of Wuhu City","correspondingAuthor":false,"prefix":"","firstName":"Fengying","middleName":"","lastName":"Sun","suffix":""},{"id":631382343,"identity":"8ddc07fe-9e8c-4815-8c21-7a76b565932a","order_by":1,"name":"Jian Ouyang","email":"","orcid":"","institution":"Zhengzhou University","correspondingAuthor":false,"prefix":"","firstName":"Jian","middleName":"","lastName":"Ouyang","suffix":""},{"id":631382344,"identity":"132df5f1-66dc-44e1-8a70-0a1c0ada0b14","order_by":2,"name":"Minmin Xiao","email":"","orcid":"","institution":"The Second People’s Hospital of Wuhu City","correspondingAuthor":false,"prefix":"","firstName":"Minmin","middleName":"","lastName":"Xiao","suffix":""},{"id":631382345,"identity":"675ba247-8858-47a0-99fd-e7a6d6ea3c00","order_by":3,"name":"Jiajia Zhou","email":"","orcid":"","institution":"Suzhou Municipal Hospital, Affiliated to Nanjing Medical University","correspondingAuthor":false,"prefix":"","firstName":"Jiajia","middleName":"","lastName":"Zhou","suffix":""},{"id":631382346,"identity":"732c1e28-9d32-4b91-9e25-0300b27be2c3","order_by":4,"name":"Jian Zhang","email":"","orcid":"","institution":"The Second People’s Hospital of Wuhu City","correspondingAuthor":false,"prefix":"","firstName":"Jian","middleName":"","lastName":"Zhang","suffix":""},{"id":631382347,"identity":"82ce82c0-259d-4e9d-bbcf-226c5953e61a","order_by":5,"name":"Tieliu Shi","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA20lEQVRIiWNgGAWjYDACdgY2hoSKAzwIkQMEdPAwg7ScIVkLYxuyMkJa7JnZnz14OO+ODD8D88HPhW0Mcnw3Ehg/F+C1hcfcIHHbMx7JBrZk6ZltDMaSNxKYpWfg18ImkbjtMI/BAR4zZt42hsQNNxLYmHnwamF/JpE4B6SF/xtISz0RWhjMJBIbwLawgbQkGBDUcpjHTCLhGNAvzWzG0jznJAxnnnnYLI1PC3t7+zPJHzV37PnZmx9+5imzkec7nnzwMz4tCMAMJiWAmLGBKA2jYBSMglEwCnADABD+P9r83+KDAAAAAElFTkSuQmCC","orcid":"","institution":"The Second People’s Hospital of Wuhu City","correspondingAuthor":true,"prefix":"","firstName":"Tieliu","middleName":"","lastName":"Shi","suffix":""}],"badges":[],"createdAt":"2026-03-24 07:55:31","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9208556/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9208556/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":108392546,"identity":"ed3b2976-e929-42b2-ae8f-bca28f236fdf","added_by":"auto","created_at":"2026-05-04 07:16:05","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":99986,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eStudy flowchart.\u003c/strong\u003e AF: atrial fibrillation; NOAF: new-onset atrial fibrillation; ML: machine learning; RFE: recursive feature elimination.\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-9208556/v1/f8244b353a3204b89c46cd71.png"},{"id":108392547,"identity":"fc5f2a50-58d3-4840-a835-81b934222606","added_by":"auto","created_at":"2026-05-04 07:16:05","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":61625,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFeature selection.\u003c/strong\u003e \u003cstrong\u003e(A)\u003c/strong\u003e Feature selection based on Boruta algorithm. The horizontal axis denotes the name of each variable, and the vertical axis denotes the Z-value of each variable. In the plot, the colors correspond to the variable importance assigned by the algorithm: green (important), yellow (tentative), and red (unimportant). The blue boxes represent max, average, and minimum shadow score. \u003cstrong\u003e(B)\u003c/strong\u003e The correlation between Lambda (regularization parameter) and different coefficients in LASSO regression. (\u003cstrong\u003eC)\u003c/strong\u003e LASSO regression cross-validation plot. Vertical dashed line on the left side represents the best lambda value for the evaluation metrics (lambda.min) while the right one indicates one standard error away from the lambda.min (lambda.1se). The λ value of 0.02098573, with log(λ), -3.863912 was chosen (lambda.1se) according to 10-fold cross-validation. RBC: red blood cell; HGB: hemoglobin; HCT: hematocrit; MCV: mean corpuscular volume; MCH: mean corpuscular hemoglobin; MCHC: mean corpuscular hemoglobin concentration; RDW: red cell distribution width; WBC: white blood cell; PLT: platelet; BUN: blood urea nitrogen; AST: aspartate aminotransferase; ALP: alkaline phosphatase; ALT: alanine aminotransferase; TBIL: total bilirubin; INR: international normalized ratio; PT: prothrombin time; PTT: partial thromboplastin time; HR: heart rate; RR: respiratory rate; SBP: systolic blood pressure; DBP: diastolic blood pressure; MBP: mean blood pressure; SpO2: percutaneous arterial oxygen saturation; GCS: Glasgow Coma Scale; SOFA: Sequential Organ Failure Assessment; CRRT: continuous renal replacement therapy; MV: mechanical ventilation; CKD: chronic kidney disease; PVD: peripheral vascular disease.\u003c/p\u003e","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-9208556/v1/d693edb7194ed24a1d76031d.png"},{"id":108392549,"identity":"6d3f0870-3e1a-4802-bd0f-93c7b6ac6681","added_by":"auto","created_at":"2026-05-04 07:16:05","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":29708,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eROC curve of ten ML models in internal validation set.\u003c/strong\u003e ROC: receiver operating characteristic; AUC: area under curve; CI: confidence interval; ML: machine learning; LR: logistic regression; SVM: support vector machine; GBM: gradient boosting machine; NN: neural network; RF: random forest; XGBoost: extreme gradient boosting; KNN: k-nearest neighbors; AdaBoost: adaptive boosting; LightGBM: light gradient boosting machine; CatBoost: category boosting.\u003c/p\u003e","description":"","filename":"Figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-9208556/v1/3f524d8c77c725c8d7b9d6df.png"},{"id":108392553,"identity":"d2c841e6-f849-407c-99eb-2d1b50fa729c","added_by":"auto","created_at":"2026-05-04 07:16:06","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":50699,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eEvaluation of validity and reliability of ten ML models in internal validation set.\u003c/strong\u003e \u003cstrong\u003e(A)\u003c/strong\u003e Calibration curve analysis. \u003cstrong\u003e(B)\u003c/strong\u003e Decision curve analysis. ML: machine learning; LR: logistic regression; SVM: support vector machine; GBM: gradient boosting machine; NN: neural network; RF: random forest; XGBoost: extreme gradient boosting; KNN: k-nearest neighbors; AdaBoost: adaptive boosting; LightGBM: light gradient boosting machine; CatBoost: category boosting.\u003c/p\u003e","description":"","filename":"Figure4.png","url":"https://assets-eu.researchsquare.com/files/rs-9208556/v1/04cfed5325680d962fc591ea.png"},{"id":108392550,"identity":"788f84d8-ee51-4473-8555-c4e80b8c1c86","added_by":"auto","created_at":"2026-05-04 07:16:05","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":68418,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eVisual model explanation by the SHAP method.\u003c/strong\u003e \u003cstrong\u003e(A)\u003c/strong\u003e SHAP histogram. \u003cstrong\u003e(B)\u003c/strong\u003e SHAP swarm plot. \u003cstrong\u003e(C and D)\u003c/strong\u003e SHAP force plots for two representative cases. SHAP: Shapley Additive Explanations; SpO2: percutaneous arterial oxygen saturation; PT: prothrombin time; PTT: partial thromboplastin time; BUN: blood urea nitrogen; AST: aspartate aminotransferase; ALP: alkaline phosphatase; ALT: alanine aminotransferase; INR: international normalized ratio; MV: mechanical ventilation.\u003c/p\u003e","description":"","filename":"Figure5.png","url":"https://assets-eu.researchsquare.com/files/rs-9208556/v1/626327886b80e7b99a7b2131.png"},{"id":108492387,"identity":"e5f6c136-ef59-4554-812a-8ed78d1b7d9c","added_by":"auto","created_at":"2026-05-05 09:57:39","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":58225,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe diagnostic performance evaluation of the simplified RF prediction model.\u003c/strong\u003e \u003cstrong\u003e(A)\u003c/strong\u003e ROC curve of the simplified RF model in the internal validation set. \u003cstrong\u003e(B)\u003c/strong\u003e ROC curve of the simplified RF model in the MIMIC-III cohort. \u003cstrong\u003e(C)\u003c/strong\u003e ROC curve of the simplified RF model in the eICU cohort. ROC: receiver operating characteristic; AUC: area under curve; CI: confidence interval; RF: random forest.\u003c/p\u003e","description":"","filename":"Figure6.png","url":"https://assets-eu.researchsquare.com/files/rs-9208556/v1/bb6158db761ca7ef472f02f0.png"},{"id":108392552,"identity":"020b3512-03f1-4005-b892-36de02c93183","added_by":"auto","created_at":"2026-05-04 07:16:05","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":335184,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eWeb interface for clinical utility.\u003c/strong\u003e An example of the web interface in use. The prediction of NOAF risk was conducted by inputting ten clinical variables. When entering actual values of the ten variables, this application automatically predicts the probability of 77.0%. The patient is at high risk for NOAF, with a predicted outcome of Yes. In addition, the bar chart for a single patient with severe sepsis shows how each feature influences the “NOAF” prediction. Features on the right (red) increase the NOAF risk prediction value, while those on the left (blue) decrease the risk prediction value.\u003c/p\u003e","description":"","filename":"Figure7.png","url":"https://assets-eu.researchsquare.com/files/rs-9208556/v1/5a11cda269aa7f15fd8a882a.png"},{"id":108495112,"identity":"a8d26e79-f528-4137-9c9a-4780f805c5b6","added_by":"auto","created_at":"2026-05-05 10:08:51","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1287993,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9208556/v1/72f96c61-2ac0-4943-988d-7040c51b70bb.pdf"},{"id":108492577,"identity":"a86913ee-fbc9-412b-9f02-381ed02ade4f","added_by":"auto","created_at":"2026-05-05 09:58:05","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":1068269,"visible":true,"origin":"","legend":"","description":"","filename":"Additionalfile1.docx","url":"https://assets-eu.researchsquare.com/files/rs-9208556/v1/0da04cf43506735ccb1ec5a8.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Machine-learning-based online prediction models for new-onset atrial fibrillation in ICU patients with severe sepsis: development and validation in a retrospective cohort","fulltext":[{"header":"Introduction","content":"\u003cp\u003eSepsis is defined by life-threatening organ dysfunction during infection and is the leading cause of death in hospitals[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. New-onset atrial fibrillation (NOAF), defined as atrial fibrillation (AF) occurring in patients without a prior AF history, is one of the most common complications in patients with sepsis. Sepsis-induced systemic inflammation, myocardial stress, electrolyte imbalances, and autonomic dysfunction make septic patients highly susceptible to AF[\u003cspan additionalcitationids=\"CR3 CR4\" citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. The incidence of NOAF in patients with sepsis ranges from 5% to 25%[\u003cspan additionalcitationids=\"CR7\" citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Notably, the incidence of NOAF among patients with severe sepsis is remarkably high, reaching 22% in severe sepsis and up to 40% in septic shock[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Patients with severe sepsis are defined as a subset of patients with sepsis, with hospital mortality rates ranging from 18% to 50%. Multiple studies have demonstrated a strong association between NOAF during severe sepsis and prolonged hospitalization, elevated risk of in-hospital death and intensive care unit (ICU) mortality, and a higher incidence of ischemic stroke[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan additionalcitationids=\"CR10\" citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Consequently, early identification of and intervention for AF in sepsis patients are crucial for minimizing adverse hospital outcomes.\u003c/p\u003e \u003cp\u003eMachine learning (ML), as an emerging technological paradigm, has been extensively used in predictive models of various diseases and exhibits good clinical application value[\u003cspan additionalcitationids=\"CR13 CR14\" citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. In recent years, several ML-based clinical prediction models have been developed for predicting the risk of AF at the individual level. For example, the XGBoost model developed by Guan and colleagues predicts the risk of NOAF in critically ill patients based on variables such as age, weight, blood urea nitrogen, percutaneous arterial oxygen saturation, urine output, sepsis, mechanical ventilation, and continuous renal replacement therapy[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. Hou et al. built a LightGBM model to evaluate the risk of NOAF in coronary heart disease by incorporating patient-level clinical factors[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. These ML models have exhibited robust predictive capabilities within their target populations. It is noteworthy that there is currently no ML-based AF risk prediction model specifically designed for the population of severe sepsis patients in the ICU.\u003c/p\u003e \u003cp\u003eDespite the high accuracy achieved by ML models, they are often criticized for their \u0026ldquo;black-box\u0026rdquo; nature, as the influence of individual variables on these models often remains largely uninterpretable, which limits their clinical application[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. The SHapley Additive exPlanations (SHAP) algorithm assigns an SHAP value to each feature in model predictions. The absolute value of SHAP measures the significance of each feature in the model\u0026rsquo;s decision-making process, thereby enhancing the model\u0026rsquo;s interpretability.\u003c/p\u003e \u003cp\u003eIn this study, we aim to construct an efficient ML model to predict the NOAF risk in ICU patients with severe sepsis based on three datasets, and to visually interpret the model using SHAP methods. This model aims to enable the early identification of high-risk groups, assisting clinicians in making appropriate decisions and implementing strategies in a timely manner, thereby improving patient outcomes.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eData source\u003c/h2\u003e \u003cp\u003eThis study is a retrospective cohort analysis based on three datasets from the MIMIC-IV database, MIMIC-III database, and eICU database. The MIMIC is a large, publicly available critical care database that contains de-identified electronic health records from Beth Israel Deaconess Medical Center (BIDMC). The MIMIC-III (V.1.4) contains 58,976 admissions for 46,520 patients collected from June 2001 to October 2012, while MIMIC-IV (V.3.1) includes 546,028 admissions for 364,627 patients between 2008 and 2022. The eICU database (version 2.0) is constructed from a multi-center cohort encompassing over 200,000 ICU patients from 208 hospitals throughout the United States during the 2014\u0026ndash;2015 period. These databases provide comprehensive clinical data such as demographics, vital signs, laboratory results, medications, surgical procedures, disease diagnoses, survival outcomes, and more. Data extraction was conducted through Structured Query Language (SQL) queries following authorization granted upon completion of the CITI Program certification (Record ID: 67165439). Since all patient data were anonymized, the study was exempt from ethical approval and individual consent requirements.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eStudy population\u003c/h3\u003e\n\u003cp\u003ePatients with a diagnosis of severe sepsis according to the International Classification of Diseases, 9th Revision (ICD-9) code \u0026lsquo;99592\u0026rsquo; or the International Classification of Diseases, 10th Revision (ICD-10) code \u0026lsquo;R6520, R6521\u0026rsquo; who were admitted to the ICU were recruited. For patients with multiple ICU admissions, only the first admission records to the ICU were included. Patients were excluded if they met any of the following criteria: (1) patients under the age of 18 years; (2) patients with ICU length of stay less than 48 hours; (3) patients with a history of AF; (4) AF as an admission diagnosis; (5) diagnosis of AF within 24 hours of ICU admission; and (6) patients with more than 20% missing data. The patient screening process is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eData collection\u003c/h3\u003e\n\u003cp\u003eThis study extracted the following information: (1) demographic features, including admission age, sex, admission weight, length of hospital stay, and ICU stay; (2) laboratory indicators, including hematocrit (HCT), hemoglobin (HGB), platelet count (PLT), white blood cell count (WBC), anion gap, bicarbonate, blood urea nitrogen (BUN), calcium, chloride, creatinine, glucose, sodium, potassium, international normalized ratio (INR), prothrombin time (PT), partial thromboplastin time (PTT), alanine aminotransferase (ALT), alkaline phosphatase (ALP), aspartate aminotransferase (AST), total bilirubin (TBIL), red blood cell count (RBC), mean corpuscular volume (MCV), mean corpuscular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), and red cell distribution width (RDW); (3) vital signs, including heart rate (HR), systolic blood pressure (SBP), diastolic blood pressure (DBP), mean blood pressure (MBP), respiratory rate (RR), temperature, and oxygen saturation (SpO2); (4) severity scores, including Glasgow Coma Scale (GCS) and Sequential Organ Failure Assessment (SOFA); (5) therapeutics, including mechanical ventilation, continuous renal replacement therapy (CRRT), and vasopressors and sedatives use; (6) comorbidities, including hypertension, diabetes, respiratory failure, heart failure, kidney failure, chronic kidney disease (CKD), and peripheral vascular disease (PVD). For vital signs, we collected the maximum, minimum, and average values. For other variables measured repeatedly during hospitalization, the maximum and minimum values were extracted. All data were extracted within the first 24 hours of the patient\u0026rsquo;s admission to the ICU, with a total of 132 variables collected.\u003c/p\u003e \u003cp\u003eVariables with missing data are a common occurrence in the MIMIC and eICU databases. Therefore, to minimize the impact of missing data on model construction, patients or variables with over 20% missing values were excluded, and the remaining missing values were imputated using the K-nearest neighbors method. Supplementary Figure \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e illustrates the missing data proportion for each variable. After preprocessing, data from 3,993 patients with 88 variables were extracted from MIMIC-IV for model development.\u003c/p\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eStatistical analysis\u003c/h2\u003e \u003cp\u003eAs all continuous variables were non-normally distributed, they were described using the median and IQR (25th percentile, 75th percentile). Categorical variables were presented as frequencies (n) and percentages (%). The Kruskal-Wallis H test was used to analyse differences in non-normally distributed continuous variables among the three groups, while the chi-square test was employed to compare categorical variables. All statistical tests were two-tailed, and P-values\u0026thinsp;\u0026lt;\u0026thinsp;0.05 were considered statistically significant. R software (version 4.3.3) was used for all statistical analyses.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eCohort division and feature selection\u003c/h3\u003e\n\u003cp\u003eTo address class imbalance between the negative and positive groups, undersampling and the synthetic minority oversampling technique (SMOTE) were employed to resample the data to achieve balance. The balanced dataset was divided into training and internal validation sets at a 7:3 ratio. Feature selection was a key step in the model building process. In this study, LASSO regression and the Boruta algorithm were used to identify the most important features based on the training set. The Boruta method, based on random forest classification, selects the most important features by comparing the Z-score of each feature against that of its \u0026ldquo;shadow features\u0026rdquo;. LASSO regression performs feature selection and dimensionality reduction through L1 regularization, which shrinks the coefficients of less relevant variables to zero, retaining only features with significant contributions. The top 10 features identified by each method, along with the common features identified by both methods, were selected as the final feature set for model construction.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eModel construction and evaluation\u003c/h2\u003e \u003cp\u003eA total of ten ML algorithms were used to construct the prediction model, namely logistic regression (LR), support vector machine (SVM), gradient boosting machine (GBM), neural network (NN), random forest (RF), extreme gradient boosting (XGBoost), k-nearest neighbors (KNN), adaptive boosting (AdaBoost), light gradient boosting machine (LightGBM), and category boosting (CatBoost). The optimal parameters of each algorithm were obtained by grid search. The training set was used for model building, and the best model was validated in the internal validation set and the external validation set. Model performance was evaluated through multiple metrics such as area under the curve (AUC) of receiver operating characteristic (ROC), accuracy, sensitivity, specificity, precision, and F1 score. Additionally, we assessed the consistency between the prediction probabilities and the sample probabilities by the calibration curve, and compared the clinical benefit at different threshold probabilities of each model via decision curve analysis (DCA). The model with the best performance was selected as the final prediction model. The modeling process was performed in R software (version 4.3.3).\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eSHapley Additive exPlanations analysis\u003c/h3\u003e\n\u003cp\u003eThe SHapley Additive exPlanations (SHAP) method, based on the concept of Shapley values from cooperative game theory, enhanced the interpretability of the final model by quantifying the contribution of each feature to the model prediction. A SHAP value of a feature greater than 0 indicates that the feature has a positive impact on the prediction result; conversely, it has a negative impact. The global importance of each feature on the model\u0026rsquo;s predictions was visually demonstrated through SHAP visualization tools such as histograms and bee swarm plots. The importance of each feature was determined by the average absolute value of its SHAP value, where a larger absolute value of SHAP indicates a stronger impact of the feature on the prediction outcome. SHAP can also generate force plots and waterfall plots for individual samples to provide personalized interpretation of prediction results, thereby guiding clinical decision-making. This method enables clear identification of key predictive factors and their directional effects, making it particularly useful for explaining complex ML models like RF.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eBaseline characteristics\u003c/h2\u003e \u003cp\u003eFollowing the application of predefined inclusion and exclusion criteria, a total of 3,993 patients with severe sepsis were identified from the MIMIC-IV database for analysis. Among these, 716 patients (17.93%) developed NOAF after the first 24 hours of ICU admission. For external validation cohorts, the incidence of AF was 18.64% (299/1,604) in the MIMIC-III database and 16.49% (259/1,571) in the eICU database. Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e summarizes the differences in baseline characteristics among patients in the derivation cohort (MIMIC-IV) and the two external validation cohorts (MIMIC-III and eICU).\u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e summarizes the comparison of baseline demographic and clinical characteristics between the non-AF group and the NOAF group in the derivation cohort. Male patients were more likely to develop AF than female patients during ICU admission. Patients who suffered from AF tended to be older and heavier. In addition, these patients had longer hospital and ICU stays, higher in-hospital mortality rates and SOFA score, higher rate of mechanical ventilation, vasopressor, and sedatives use, and higher respiratory failure and kidney failure risks compared with those without AF. Laboratory and vital signs assessments revealed that the maximum values of HCT, HGB, anion gap, BUN, calcium, creatinine, glucose, potassium, INR, PT, PTT, ALT, AST, RBC, and MCV; the minimum values of HCT, HGB, anion gap, BUN, creatinine, potassium, ALT, and AST; and the mean values of RR were much higher in severe sepsis with AF compared with those without AF (P\u0026thinsp;\u0026lt;\u0026thinsp;0.05). However, the maximum values of PLT, WBC, chloride, ALP, and HR; the minimum values of PLT, WBC, ALP, MCHC, HR, SBP, DBP, temperature, and SpO2; the mean values of HR, DBP, temperature, and SpO2, and urine output were lower than those in the non-AF group (P\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\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\u003eComparison of baseline characteristics of patients with severe sepsis with and without NOAF.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\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\u003eNon-NOAF (n\u0026thinsp;=\u0026thinsp;3277)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNOAF (n\u0026thinsp;=\u0026thinsp;716)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eP\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAdmission age, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e63.24 (51.99, 73.65)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e70.40 (62.15, 79.87)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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 weight, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e77.90 (64.25, 94.20)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e80.90 (68.13, 98.30)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c4\"\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\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1532 (46.75)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e266 (37.15)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1745 (53.25)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e450 (62.85)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLos hospital, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e13.53 (7.75, 24.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e16.59 (9.51, 28.25)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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\u003eLos icu, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e5.13 (3.01, 10.01)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e9.79 (5.67, 17.75)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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\u003eHospital expire flag, 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=\"char\" char=\".\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2387 (72.84)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e373 (52.09)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e890 (27.16)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e343 (47.91)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRBC_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.13 (2.66, 3.67)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.20 (2.71, 3.74)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.085\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRBC_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.46 (2.99, 3.99)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.58 (3.07, 4.11)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHGB_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e9.20 (7.80, 10.80)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e9.40 (8.00, 11.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.009\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHGB_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e10.60 (9.20, 12.40)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e11.00 (9.50, 12.70)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.004\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHCT_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e28.40 (24.00, 33.10)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e28.95 (24.23, 34.18)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.012\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHCT_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e32.90 (28.55, 38.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e34.00 (29.20, 39.38)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.002\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMCV_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e91.00 (87.00, 96.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e92.00 (87.00, 96.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.064\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMCV_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e93.00 (88.00, 98.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e94.00 (89.00, 98.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.003\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMCH_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e29.70 (28.10, 31.30)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e30.00 (28.30, 31.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.156\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMCH_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e30.30 (28.50, 32.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e30.50 (28.90, 32.20)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.072\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMCHC_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e32.20 (31.10, 33.20)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e32.00 (31.00, 33.10)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.026\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMCHC_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e33.00 (31.80, 34.10)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e32.80 (31.80, 34.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.293\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRDW_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e15.20 (14.00, 17.20)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e15.00 (13.90, 17.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.281\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRDW_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e15.60 (14.30, 17.60)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e15.40 (14.20, 17.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.696\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWBC_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e11.00 (6.60, 16.10)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e10.50 (6.13, 15.68)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.018\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWBC_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e16.20 (10.70, 23.20)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e15.40 (9.80, 21.40)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.011\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePLT_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e162.00 (97.00, 245.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e149.00 (90.25, 212.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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\u003ePLT_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e209.00 (132.00, 307.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e194.00 (128.00, 268.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAnion gap_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e13.00 (11.00, 16.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e14.00 (11.00, 17.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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\u003eAnion gap_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e17.00 (14.00, 21.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e18.00 (15.00, 22.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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\u003eBicarbonate_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e19.00 (16.00, 22.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e18.00 (15.00, 22.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.069\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBicarbonate_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e22.00 (19.00, 25.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e22.00 (19.00, 25.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.326\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBUN_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e24.00 (14.00, 40.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e31.00 (20.00, 50.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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\u003eBUN_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e30.00 (18.00, 49.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e38.00 (25.00, 60.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e7.60 (7.10, 8.20)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e7.70 (7.10, 8.30)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.094\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCalcium_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e8.30 (7.80, 8.90)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e8.40 (7.90, 9.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.015\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChloride_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e101.00 (97.00, 106.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e101.00 (97.00, 105.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.069\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChloride_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e106.00 (101.00, 111.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e105.00 (101.00, 110.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.026\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCreatinine_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.10 (0.70, 1.90)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.40 (0.90, 2.40)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.50 (0.90, 2.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.90 (1.20, 3.10)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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\u003eGlucose_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e109.00 (89.00, 135.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e110.50 (89.00, 142.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.223\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGlucose_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e158.00 (123.50, 217.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e165.00 (129.00, 226.75)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.015\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSodium_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e136.00 (132.00, 139.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e136.00 (132.00, 139.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.510\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSodium_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e139.00 (136.00, 143.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e140.00 (136.00, 143.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.927\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePotassium_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.80 (3.40, 4.20)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4.00 (3.50, 4.40)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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\u003ePotassium_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4.50 (4.10, 5.10)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4.70 (4.30, 5.30)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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\u003eALT_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e29.00 (17.00, 53.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e31.00 (19.00, 61.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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\u003eALT_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e30.00 (19.00, 66.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e38.00 (22.00, 87.75)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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\u003eALP_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e89.00 (65.00, 135.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e78.00 (58.00, 121.75)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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\u003eALP_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e98.00 (71.00, 156.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e87.50 (63.00, 138.75)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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\u003eAST_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e40.00 (24.00, 83.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e51.00 (29.00, 96.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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\u003eAST_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e50.00 (28.00, 110.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e62.00 (35.00, 136.75)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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\u003eTBIL_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.70 (0.40, 1.65)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.70 (0.40, 1.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.238\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTBIL_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.80 (0.50, 2.20)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.90 (0.50, 2.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.449\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eINR_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.30 (1.20, 1.60)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.30 (1.20, 1.60)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.488\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eINR_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.50 (1.20, 1.90)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.50 (1.30, 2.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.017\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePT_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e14.60 (12.80, 17.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e14.40 (12.80, 17.08)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.237\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePT_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e15.90 (13.60, 20.10)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e16.10 (13.80, 21.78)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.026\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePTT_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e30.60 (27.10, 36.40)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e31.10 (27.30, 36.80)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.384\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePTT_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e35.10 (29.70, 48.75)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e39.80 (30.63, 59.98)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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\u003eUrine output, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1265.00 (696.00, 2120.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e928.50 (454.25, 1642.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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\u003eHR_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e76.00 (65.00, 88.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e74.00 (62.00, 85.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHR_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e111.00 (97.00, 124.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e106.00 (92.00, 123.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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\u003eHR_mean, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e91.07 (79.61, 103.70)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e88.89 (76.17, 101.17)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRR_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e13.00 (11.00, 16.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e13.00 (10.00, 16.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.629\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRR_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e30.00 (25.50, 34.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e30.00 (26.00, 34.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.993\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRR_mean, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e20.95 (18.00, 24.18)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e21.52 (18.46, 24.55)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.009\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSBP_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e82.00 (75.00, 90.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e82.00 (73.00, 90.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.036\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSBP_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e140.00 (127.00, 155.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e142.00 (129.00, 156.75)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.095\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSBP_mean, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e107.90 (101.30, 116.06)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e107.83 (101.36, 115.74)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.850\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDBP_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e43.00 (37.00, 49.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e42.00 (36.00, 47.88)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.002\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDBP_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e83.00 (73.00, 96.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e83.00 (71.00, 95.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.098\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDBP_mean, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e59.08 (53.90, 64.91)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e57.80 (52.19, 63.68)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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\u003eMBP_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e55.00 (48.00, 61.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e55.00 (47.00, 60.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.076\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMBP_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e99.00 (88.00, 113.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e98.00 (88.00, 113.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.965\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMBP_mean, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e72.86 (68.42, 78.56)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e72.51 (67.91, 77.83)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.278\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTemperature_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e36.50 (36.17, 36.72)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e36.44 (36.00, 36.72)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.008\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTemperature_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e37.50 (37.06, 38.29)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e37.50 (37.06, 38.17)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.303\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTemperature_mean, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e36.95 (36.65, 37.37)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e36.92 (36.60, 37.31)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.040\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSpO2_min, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e92.00 (89.00, 94.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e91.00 (88.00, 94.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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\u003eSpO2_max, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e100.00 (100.00, 100.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e100.00 (99.00, 100.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.079\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSpO2_mean, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e96.93 (95.51, 98.33)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e96.71 (95.03, 98.20)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.004\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSOFA, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e8.00 (5.00, 11.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e9.00 (6.00, 11.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\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\u003eGCS, M (Q1, Q3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e15.00 (13.00, 15.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e15.00 (14.00, 15.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.090\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMV, 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=\"char\" char=\".\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1712 (52.24%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e274 (38.27)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1565 (47.76)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e442 (61.73)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\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=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.172\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3104 (94.72)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e669 (93.44)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e173 (5.28)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e47 (6.56)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVasopressors, 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=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.023\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1289 (39.33)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e249 (34.78)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1988 (60.67)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e467 (65.22)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003esedatives, 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=\"char\" char=\".\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1518 (46.32)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e252 (35.20)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1759 (53.68)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e464 (64.80)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\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=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.156\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1830 (55.84)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e379 (52.93)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1447 (44.16)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e337 (47.07)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDiabetes\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=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.003\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2191 (66.86)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e437 (61.03)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1086 (33.14)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e279 (38.97)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1392 (42.48)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e197 (27.51)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1885 (57.52)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e519 (72.49)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHeart 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=\"char\" char=\".\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2542 (77.57)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e442 (61.73)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e735 (22.43)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e274 (38.27)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKidney 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=\"char\" char=\".\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1086 (33.14)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e152 (21.23)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2191 (66.86)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e564 (78.77)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCKD\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=\"char\" char=\".\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2596 (79.22)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e493 (68.85)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e681 (20.78)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e223 (31.15)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePVD, 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=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.257\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3180 (97.04)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e689 (96.23)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e97 (2.96)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e27 (3.77)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"4\"\u003eM: Median, Q1: 1st Quartile, Q3: 3st Quartile\u003c/td\u003e\u003c/tr\u003e \u003ctr\u003e\u003ctd colspan=\"4\"\u003eNOAF New-onset atrial fibrillation; Los: length of stay; RBC: red blood cell; HGB: hemoglobin; HCT: hematocrit; MCV: mean corpuscular volume; MCH: mean corpuscular hemoglobin; MCHC: mean corpuscular hemoglobin concentration; RDW: red cell distribution width; WBC: white blood cell; PLT: platelet; BUN: blood urea nitrogen; AST: aspartate aminotransferase; ALP: alkaline phosphatase; ALT: alanine aminotransferase; TBIL: total bilirubin; INR: international normalized ratio; PT: prothrombin time; PTT: partial thromboplastin time; HR: heart rate; RR: respiratory rate; SBP: systolic blood pressure; DBP: diastolic blood pressure; MBP: mean blood pressure; SpO2: percutaneous arterial oxygen saturation; GCS: Glasgow Coma Scale; SOFA: Sequential Organ Failure Assessment; CRRT: continuous renal replacement therapy; MV: mechanical ventilation; CKD: chronic kidney disease; PVD: peripheral vascular disease.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eFeature selection\u003c/h2\u003e \u003cp\u003eTwo methods were used for feature screening. The results of LASSO regression and Boruta algorithm analysis are presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. Thirty-two out of 88 variables were identified to be closely associated with NOAF using the Boruta algorithm (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA). Based on variable importance, the top ten features were as follows: age, AST_max, AST_min, ALP_max, PT_max, PTT_max, ALP_min, INR_max, ALT_max, and ALT_min. In the LASSO regression, 16 variables were selected as potential predictors. The coefficient profile and cross-validated error plot of the LASSO model are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB and C, respectively, with the full list of coefficients provided in Supplementary Table S2. Based on the ranking of variable importance according to the absolute values of LASSO regression coefficients, the ten variables most closely associated with NOAF were identified: heart failure, respiratory failure, mechanical ventilation, sedatives, gender, age, SpO2_mean, potassium_min, PTT_max, and BUN_min. According to the feature selection results, we constructed the final feature set by merging the features common to both methods and the top ten features from each method. After removing duplicates, we obtained 19 features for model development (Supplementary Figure S2). These selected features were mechanical ventilation, heart failure, age, BUN_min, PT_max, PTT_max, urine output, respiratory failure, sedatives, gender, SpO2_mean, potassium_min, AST_max, AST_min, ALP_max, ALP_min, INR_max, ALT_max, and ALT_min.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eModel development and evaluation\u003c/h2\u003e \u003cp\u003eUsing LASSO regression and Boruta algorithm, we selected 19 key features from 88 clinical variables associated with the occurrence of AF. Based on these features, we developed ten ML models to predict the risk of NOAF in severe sepsis patients in the ICU using the training data set. The performance of these models was evaluated on the internal validation set. The ROCs of the ten ML models are presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. Among the ten models, the RF model demonstrated the highest predictive performance (AUC\u0026thinsp;=\u0026thinsp;0.936, 95% confidence interval (CI): 0.921\u0026ndash;0.951). The AdaBoost model exhibited comparable efficacy to the RF model, with an AUC of 0.920 (95% CI: 0.902\u0026ndash;0.937), followed by SVM (AUC\u0026thinsp;=\u0026thinsp;0.905, 95% CI: 0.886\u0026ndash;0.924), GBM (AUC\u0026thinsp;=\u0026thinsp;0.893, 95% CI: 0.870\u0026ndash;0.915), and XGBoost (AUC\u0026thinsp;=\u0026thinsp;0.861, 95% CI: 0.836\u0026ndash;0.886) models. The remaining models showed relatively lower generalization ability, ranked in descending order of performance as follows: KNN (AUC\u0026thinsp;=\u0026thinsp;0.814, 95% CI: 0.788\u0026ndash;0.839), LightGBM (AUC\u0026thinsp;=\u0026thinsp;0.797, 95% CI: 0.767\u0026ndash;0.827), CatBoost (AUC\u0026thinsp;=\u0026thinsp;0.793, 95% CI: 0.763\u0026ndash;0.823), LR (AUC\u0026thinsp;=\u0026thinsp;0.748, 95% CI: 0.716\u0026ndash;0.780) and NN (AUC\u0026thinsp;=\u0026thinsp;0.740, 95% CI: 0.707\u0026ndash;0.773) models.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eA set of detailed performance metrics for the ten ML models is shown in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, among which the RF model exhibited superior overall performance with the highest accuracy (0.851), sensitivity (0.902), F1 score (0.858), and the third highest precision (0.818) and the fourth highest specificity (0.800). The calibration curves revealed that the RF model had excellent agreement between predicted probabilities and actual probabilities (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA). According to the DCA curves (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eB), the RF model obtained a higher net benefit across a wider range of threshold probabilities than other models, indicating its superior clinical utility. Therefore, the RF model was ultimately selected as the optimal model for predicting NOAF.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformances of the ten ML models for predicting NOAF in internal validation set.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" 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=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThreshold\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAUC (95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSensitivity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSpecificity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eF1 score\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.496\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAUC\u0026thinsp;=\u0026thinsp;0.748 (0.716\u0026ndash;0.780)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.671\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.697\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.646\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.663\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.680\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSVM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.542\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAUC\u0026thinsp;=\u0026thinsp;0.905 (0.886\u0026ndash;0.924)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.781\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.695\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.867\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.839\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.760\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGBM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAUC\u0026thinsp;=\u0026thinsp;0.893 (0.870\u0026ndash;0.915)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.825\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.844\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.807\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.813\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.828\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.517\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAUC\u0026thinsp;=\u0026thinsp;0.740 (0.707\u0026ndash;0.773)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.698\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.772\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.625\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.673\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.719\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAUC\u0026thinsp;=\u0026thinsp;0.936 (0.921\u0026ndash;0.951)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.851\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.902\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.800\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.818\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.858\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eXGBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.514\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAUC\u0026thinsp;=\u0026thinsp;0.861 (0.836\u0026ndash;0.886)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.782\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.795\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.769\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.775\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.785\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAUC\u0026thinsp;=\u0026thinsp;0.814 (0.788\u0026ndash;0.839)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.814\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.862\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.765\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.786\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.822\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAdaBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAUC\u0026thinsp;=\u0026thinsp;0.920 (0.902\u0026ndash;0.937)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.843\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.879\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.807\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.820\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.848\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLightGBM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.543\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAUC\u0026thinsp;=\u0026thinsp;0.797 (0.767\u0026ndash;0.827)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.749\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.713\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.786\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.769\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.740\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCatBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.629\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAUC\u0026thinsp;=\u0026thinsp;0.793 (0.763\u0026ndash;0.823)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.721\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.730\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.713\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.718\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.724\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"8\"\u003eNOAF: new-onset atrial fibrillation, ML: machine learning; AUC: area under the curve; CI: confidence interval; LR: logistic regression; SVM: support vector machine; GBM: gradient boosting machine; NN: neural network; RF: random forest; XGBoost: extreme gradient boosting; KNN: k-nearest neighbors; AdaBoost: adaptive boosting; LightGBM: light gradient boosting machine; CatBoost: category boosting.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eModel interpretability\u003c/h2\u003e \u003cp\u003eThe SHAP algorithm was used to visually exhibit the contributions of each feature in the RF model for predicting the risk of NOAF in severe sepsis patients in the ICU. Figure\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eA shows the feature importance plot, which revealed that age, urine output, respiratory failure, BUN_min, mechanical ventilation, PTT_max, heart failure, SpO2_mean, potassium_min, and sedatives were the ten most important variables in predicting NOAF. Figure\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eB presents a comprehensive beeswarm plot illustrating the features in the RF model. Each data point represented a sample, with a yellow-to-purple color gradient indicating the magnitude of the feature value, where yellow denoted a higher value and purple denoted a lower value. Baseline variables with higher SHAP values contributed to a higher risk of developing NOAF during ICU hospitalization. According to the SHAP beeswarm plots, older patients with respiratory failure and heart failure, those receiving sedatives and mechanical ventilation, and those who presented with low urine output, low SpO2_mean, high PTT_max, high BUN_min, and high potassium_min were at significantly increased risk of developing NOAF during their ICU hospitalization.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTo facilitate a granular understanding of the model\u0026rsquo;s decision logic at the individual level, we performed an interpretability analysis on two representative cases, as illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eC and D. The SHAP values were used to show the magnitude and direction of each feature\u0026rsquo;s contribution to the prediction. Features with positive values increase the AF risk, while those with negative values decrease the AF risk. Each feature contributed to varying degrees to the predicted probability of the outcome, and their combined effect determined the final result. The value f(x) represents the model\u0026rsquo;s predicted probability, whereas E[f(x)] is the expected model output without input features. For a patient who did not develop AF, our RF model predicted an f(x) close to 0 (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eC), correctly indicating a low risk of the event. In contrast, for a patient who developed AF, the model predicted a high risk, with an f(x) value of 1 (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eD).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eExternal validation\u003c/h2\u003e \u003cp\u003eTo further verify the predictive accuracy of the selected RF model, two external validation cohorts were additionally incorporated: 1,604 patients with severe sepsis from MIMIC-III (Cohort 1) and 1,571 from the eICU database (Cohort 2). Their baseline characteristics are provided in Supplementary Tables S3 and S4, respectively. Notwithstanding the inherent differences in baseline characteristics among the three cohorts (Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e), the externally validated ROC curve achieved an AUC of 0.779 (95% CI: 0.749\u0026ndash;0.808) in the MIMIC-III cohort and 0.724 (0.692\u0026ndash;0.756) in the eICU cohort (Supplementary Figure S3), demonstrating the strong generalization capability of our model.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eApplication of the model\u003c/h2\u003e \u003cp\u003eTo improve the clinical applicability of our model and facilitate rapid clinical decision-making, we employed a recursive feature elimination (RFE) approach to refine the predictor set (Supplementary Figure S4). This method facilitated the identification of a minimal yet highly informative subset of variables, balancing interpretability and predictive power. Through this process, the model was optimized to include only 10 clinically relevant features: age, BUN_min, urine output, AST_min, SpO2_mean, AST_max, ALP_max, PTT_max, ALP_min, and ALT_min. The simplified model demonstrated robust discriminatory performance in both internal and external validation, with AUCs of 0.907 (95% CI: 0.887\u0026ndash;0.926) in the internal validation set, 0.769 (95% CI: 0.739\u0026ndash;0.799) in the MIMIC-III cohort, and 0.707 (95% CI: 0.676\u0026ndash;0.739) (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe final simplified model was deployed as a web application designed to facilitate its use in clinical settings (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e). Users can automatically obtain prediction outcomes of NOAF risk by entering the actual values of the ten features required for the model into the designated text fields on the webpage. Furthermore, the platform will also generate a bar chart for the individual patient to display the important features that contribute to the decision of NOAF. Positive or negative values in bar chart indicate positive or negative contributions to the prediction. The web application can be freely accessed at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://119.3.41.228/mm/index.php\u003c/span\u003e\u003cspan address=\"http://119.3.41.228/mm/index.php\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eDespite considerable advances in medicine, sepsis is a leading cause of death in patients in the ICU. Patients with sepsis who develop AF often experience a worsening of their condition and have a poor prognosis. Studies have shown that sepsis patients with NOAF have a higher risk of mortality compared to those without AF, with mortality rates of 45.7% and 26.8%, respectively[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. Therefore, it is of paramount importance to establish predictive tools to identify the risk of NOAF for improving patients\u0026rsquo; prognosis.\u003c/p\u003e \u003cp\u003eIn this study, we constructed and validated an ML model for predicting the risk of AF in severe sepsis patients admitted to the ICU using three large-scale datasets. Based on 88 clinical variables extracted within the first 24 hours of ICU admission, we selected 19 key features through the use of LASSO regression combined with the Boruta algorithm and developed ten ML models using these 19 features. Our results showed that the RF model exhibited the best predictive performance among all ML models. Moreover, the RF model demonstrated good predictive accuracy in terms of discrimination and calibration, and provided a substantial net benefit in clinical practice. The results from two external validation cohorts provided additional confirmation of the stability and accuracy of the RF model for predicting the risk of AF in severe sepsis. Additionally, the SHAP method was employed to interpret and visualize the RF model, further improving its applicability in the clinical setting. According to SHAP histograms and bee swarm plots, the study identified ten key factors most closely associated with the risk of NOAF in patients with severe sepsis, including age, urine output, respiratory failure, BUN_min, mechanical ventilation, PTT_max, heart failure, SpO2_mean, potassium_min, and sedatives. We employed SHAP force plots to visualize individualized AF risk predictions, offering an intuitive interpretation of the model\u0026rsquo;s decision-making process in distinguishing high-risk cases.\u003c/p\u003e \u003cp\u003ePrevious studies have attempted to construct predictive models for AF risk in sepsis patients, and a retrospective study extracted electrocardiogram (ECG) data of 198 subjects from MIMIC-III database and modelled them using KNN algorithms, achieving sensitivity, specificity, and accuracy of 98.40%, 99.80%, and 99.32%, respectively[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Syed et al. utilized ECG data from 100 patients sourced from the AFPDB and MIMIC-III databases to develop a risk prediction model for NOAF in sepsis patients, with the SVM model reaching 80% sensitivity, 100% specificity, and 90% accuracy. Although both aforementioned predictive models demonstrated high accuracy, both studies were performed on small sample sizes and lacked external validation and interpretation of the models[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. Jiming et al. used a multicenter dataset of 2,492 patients with sepsis from two hospitals in China to build a nomogram model for the prediction of NOAF based on clinical risk factors, achieving an AUC of 0.861 and 0.845 in the internal and external validation, respectively, but did not perform any model interpretation[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Although previous studies have attempted to construct predictive models for AF risk in patients with sepsis, most have not specifically targeted severe sepsis, and none have yet undergone clinical transformation, limiting their clinical utility.\u003c/p\u003e \u003cp\u003eIn contrast, our study is the first to build an interpretable RF model for specifically targeting the prediction of NOAF in patients with severe sepsis across three datasets. The model achieved an AUC of 0.936 (95% CI: 0.921\u0026ndash;0.951) in the internal validation set, 0.779 (95% CI: 0.749\u0026ndash;0.808) in the MIMIC-III cohort, and 0.724 (95% CI: 0.692\u0026ndash;0.756) in the eICU cohort, demonstrating robust generalizability. Currently, the MIMIC database has been widely applied in the prediction of various diseases such as acute kidney injury, rhabdomyolysis, and hepatorenal syndrome[\u003cspan additionalcitationids=\"CR24\" citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Notably, compared to single-center datasets, the MIMIC databases provide large-scale, diverse patient data that enhance the robustness and generalizability of the predictive model, thereby improving the applicability of the model across varied healthcare settings.\u003c/p\u003e \u003cp\u003eA key advantage of our study is the simplification of the model structure through RFE. The simplified version of the model required fewer input variables while maintaining similar accuracy, thereby improving its practicality and user-friendliness. Moreover, all the selected variables were readily available in most hospitals, enhancing our model\u0026rsquo;s potential for widespread application in medical institutions of different levels. We had also deployed the simplified model on a dedicated website to further facilitate its use in clinical settings. The user-friendly platform will assist clinicians in rapidly and accurately identifying patients at high risk of AF and optimizing treatment strategies based on predicted outcomes, thereby improving patient outcomes.\u003c/p\u003e \u003cp\u003eSHAP values demonstrate that age is the most influential feature in the development of AF, consistent with previous studies[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. The aging left atrium undergoes structural remodeling due to fatty infiltration, fibrosis, and loss of cardiomyocytes, leading to conduction block and multiple wavelet reentry. Concurrently, calcium homeostasis disorders such as cytoplasmic calcium overload and spontaneous calcium release can shorten action potentials and predispose to AF. Furthermore, abnormalities at the cellular and molecular levels, including ion channel dysfunction, autonomic imbalance, and oxidative stress, further increase atrial spontaneous excitability and predisposition to arrhythmias[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. Among the comorbidities, respiratory failure, manifested by hypoxemia (SpO2 reduction) and an increased requirement for mechanical ventilation, had the strongest association with NOAF. In patients with type 2 respiratory failure, the incidence of AF is relatively high, approximately 50%[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. Previous studies have shown that hypoxemia and hypercapnia caused by respiratory failure are significant factors contributing to the occurrence of AF. The mechanism is that hypoxemia and hypercapnia can lead to right ventricular overload and right atrial dilation by increasing pulmonary arterial pressure. These changes disrupt the hemodynamic balance of the heart, thereby contributing to the onset of AF[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. Renal failure, characterized by elevated BUN and potassium levels and a reduction in urine output, is also an important predictor of NOAF. Patients with renal failure are prone to AF, potentially due to electrolyte imbalances, neurohormonal activation, and systemic inflammation/oxidative stress[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. In addition, we noted that the AF prevalence in renal failure patients increased from 45.80% in 2008 to 65.51% in 2014, which may be attributed to improved survival of critically ill patients, enhanced detection methods, an aging population, and increased comorbidities[\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. We observed that AF was more frequently present in patients with severe sepsis who were administered sedative drugs than in those who were not. The potential mechanism by which sedative agents trigger the onset of AF could be as follows: (1) direct blockade of cardiac ion channels, leading to atrial conduction abnormalities; (2) drug-induced reflex sympathetic activation or a reflex increase in sympathetic tone, which promotes ectopic firing; and (3) anticholinergic effects through M₂ receptor antagonism, resulting in shortening of the atrial effective refractory period and facilitation of re-entrant circuits[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. Our finding was consistent with previous studies showing that prolonged PTT was a risk factor for NOAF[\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. The underlying logic was that prolonged PTT indicated that the patient population was undergoing heparin treatment due to a high risk of thrombosis, and the underlying conditions necessitating anticoagulation (such as post-cardiac surgery, heart failure, sepsis, etc.) are precisely the classic triggers for AF[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. Therefore, prolonged PTT was a consequence of anticoagulation therapy, not a cause of AF; it acted as an effective surrogate marker, identifying those high-risk populations with primary diseases that are highly prone to developing AF.\u003c/p\u003e \u003cp\u003eSeveral limitations must be acknowledged in this study. First, the sample size of patients with AF in this large cohort of patients with severe sepsis was relatively modest, and the incidence of events was relatively low. These class-unbalanced data posed a challenge for ML. Although combined undersampling and oversampling techniques addressed class imbalance problem, the characteristics of the cohort may limit the predictive performance. Second, despite the use of data from MIMIC, their retrospective design potentially introduces selection bias arising from demographic variations and patient inclusion criteria. Furthermore, while the prediction model showed robust performance in two external validation cohorts (MIMIC-III and eICU), all data originated from the United States, which potentially limits the generalizability of our findings to other populations. Therefore, future external validation across multiple centers will be essential to further evaluate the model\u0026rsquo;s generalizability and clinical applicability. Finally, due to the limitations inherent in the MIMIC and eICU databases, the study lacked features related to cardiac function, such as brain natriuretic peptide and N-terminal pro-brain natriuretic peptide. The absence of these key variables may affect the comprehensiveness of our predictive model.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eWe developed an RF model to predict the risk of NOAF in patients with severe sepsis in the ICU, which may assist clinicians in tailoring management and implementing early interventions for these patients at risk of AF to improve outcomes. SHAP enhances the interpretability of the model, providing insights into the contribution of each variable to the model\u0026rsquo;s predictions. Additionally, the optimized compact model and a web-based application further improve the clinical applicability of our model.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eNew-onset atrial fibrillation (NOAF), atrial fibrillation (AF), intensive care unit (ICU), machine learning (ML), Medical Information Mart for Intensive Care (MIMIC), SHapley Additive exPlanations (SHAP), receiver operating characteristic (ROC), area under the curve (AUC), random forest (RF), logistic regression (LR), support vector machine (SVM), gradient boosting machine (GBM), neural network (NN), extreme gradient boosting (XGBoost), k-nearest neighbors (KNN), adaptive boosting (AdaBoost), light gradient boosting machine (LightGBM), category boosting (CatBoost), decision curve analysis (DCA), hematocrit (HCT), hemoglobin (HGB), platelet count (PLT), white blood cell count (WBC), blood urea nitrogen (BUN), international normalized ratio (INR), prothrombin time (PT), partial thromboplastin time (PTT), alanine aminotransferase (ALT), alkaline phosphatase (ALP), aspartate aminotransferase (AST), total bilirubin (TBIL), red blood cell count (RBC), mean corpuscular volume (MCV), mean corpuscular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), red cell distribution width (RDW), heart rate (HR), systolic blood pressure (SBP), diastolic blood pressure (DBP), mean blood pressure (MBP), respiratory rate (RR), oxygen saturation (SpO2), Glasgow Coma Scale (GCS), Sequential Organ Failure Assessment (SOFA), continuous renal replacement therapy (CRRT), chronic kidney disease (CKD), and peripheral vascular disease (PVD).\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors would like to thank the developers of the MIMIC and eICU databases for sharing their data resources. Additionally, we also acknowledge all individuals who contributed to this work and the financial support received from funding bodies.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFYS, MMX, and TLS conceived and designed the study. FYS, JZ, and JJZ conducted the data collection, data analysis, and data interpretation. FYS, JJZ, and JOY established the ML models and website. FYS, JZ, and TLS drafted the initial manuscript. TLS, MMX, and JZ revised the manuscript and performed some of the revised analyses. All authors read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was funded by the National Key Research and Development Program of China (No. 2023YFC2706503), the Shanghai Municipal Science and Technology Major Project (No. 2017SHZDZX01), the Cooperative Research Fund of the Affiliated Wuhu Hospital of East China Normal University (No. 40500-20104-222400), Beihang University \u0026amp; Capital Medical University Plan (No. BHME-201904), the Special Fund of the Pediatric Medical Coordinated Development Center of Beijing Hospitals Authority (No. XTCX201809), and the Open Research Fund of Key Laboratory of Advanced Theory and Application in Statistics and Data Science-MOE, ECNU and Key Laboratory of MEA, Ministry of Education, ECNU.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe MIMIC-IV (V.3.1) and MIMIC-III (V.1.4) datasets used in this study are publicly available via the PhysioNet platform (https://physionet.org/) following authorization granted upon completion of the CITI Program certification. The data analyzed and the codes used during the current study are available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eOur research was based on the MIMIC database, a de-identified public resource. The creation of MIMIC database was approved by the institutional review boards of Beth Israel Deaconess Medical Center and the Massachusetts Institute of Technology, with a waiver of informed consent. The author (Fengying Sun) passed the Collaborative Institutional Training Initiative (CITI) program exam and obtained permission to access the database (Record ID: 67165439). Consequently, our analysis, which involved secondary use of this data, was exempt from further ethical approval and individual consent requirements.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClinical trial number\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eSinger M, Deutschman CS, Seymour CW, Shankar-Hari M, Annane D, Bauer M, et al. The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3). JAMA. 2016;315:801\u0026ndash;10.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBai W, Liu ZQ, He PY, Muhuyati. The role of IL-6, IL-10, TNF-α and PD-1 expression on CD4 T cells in atrial fibrillation. Heliyon. 2023;9:e18818.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePinho-Gomes AC, Reilly S, Brandes RP, Casadei B. Targeting inflammation and oxidative stress in atrial fibrillation: role of 3-hydroxy-3-methylglutaryl-coenzyme a reductase inhibition with statins. Antioxid Redox Signal. 2014;20:1268\u0026ndash;85.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRafaqat S, Rafaqat S, Khurshid H, Rafaqat S. Electrolyte's imbalance role in atrial fibrillation: Pharmacological management. Int J Arrhythmia. 2022;23:1\u0026ndash;10.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCarrara M, Ferrario M, Bollen Pinto B, Herpain A. The autonomic nervous system in septic shock and its role as a future therapeutic target: a narrative review. Ann Intensive Care. 2021;11:80.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKlein Klouwenberg PM, Frencken JF, Kuipers S, Ong DS, Peelen LM, van Vught LA, et al. Incidence, Predictors, and Outcomes of New-Onset Atrial Fibrillation in Critically Ill Patients with Sepsis. A Cohort Study. Am J Respir Crit Care Med. 2017;195:205\u0026ndash;11.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWetterslev M, Haase N, Hassager C, Belley-Cote EP, McIntyre WF, An Y, et al. New-onset atrial fibrillation in adult critically ill patients: a scoping review. Intensive Care Med. 2019;45:928\u0026ndash;38.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFernando SM, Mathew R, Hibbert B, Rochwerg B, Munshi L, Walkey AJ, et al. New-onset atrial fibrillation and associated outcomes and resource use among critically ill adults-a multicenter retrospective cohort study. Crit Care. 2020;24:15.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCorica B, Romiti GF, Basili S, Proietti M. Prevalence of New-Onset Atrial Fibrillation and Associated Outcomes in Patients with Sepsis: A Systematic Review and Meta-Analysis. J Pers Med. 2022;12.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu YW, Wang YF, Chen Y, Dong R, Li S, Peng JM, et al. A nationwide study on new onset atrial fibrillation risk factors and its association with hospital mortality in sepsis patients. Sci Rep. 2024;14:12206.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuo Y, Yoshimura H, Gonzalez-Izquierdo A, Lip GYH, Schmidt F, Providencia R. Risk Factors and Prognosis of New-Onset Atrial Fibrillation in Sepsis: A Nationwide Electronic Health Record Study. JACC Adv. 2025;4:101681.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeng B, Zhao Z, Ruan T, Zhou R, Liu C, Li Q, et al. Development and external validation of a machine learning model for brain injury in pediatric patients on extracorporeal membrane oxygenation. Crit Care. 2025;29:17.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKwiendacz H, Huang B, Chen Y, Janota O, Irlik K, Liu Y, et al. Predicting major adverse cardiac events in diabetes and chronic kidney disease: a machine learning study from the Silesia Diabetes-Heart Project. Cardiovasc Diabetol. 2025;24:76.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLei J, Zhai J, Zhang Y, Qi J, Sun C. Supervised Machine Learning Models for Predicting Sepsis-Associated Liver Injury in Patients With Sepsis: Development and Validation Study Based on a Multicenter Cohort Study. J Med Internet Res. 2025;27:e66733.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen H, Song H, Huang H, Fang X, Chen H, Yang Q, et al. Machine learning prediction and interpretability analysis of high-risk chest pain: a study from the MIMIC-IV database. Front Physiol. 2025;16:1594277.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuan C, Gong A, Zhao Y, Yin C, Geng L, Liu L, et al. Interpretable machine learning model for new-onset atrial fibrillation prediction in critically ill patients: a multi-center study. Crit Care. 2024;28:349.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHou L, Su K, Zhao J, He T, Li Y. Application of Multi-Inflammatory Index to Predict Atrial Fibrillation Risk in Patients with Coronary Heart Disease: A Retrospective Machine Learning Study. Risk Manag Healthc Policy. 2024;17:2907\u0026ndash;15.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWiggerthale J, Reich C. Explainable Machine Learning in Critical Decision Systems: Ensuring Safe Application and Correctness. 2024.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMeng H, Guo L, Pan Y, Kong B, Shuai W, Huang H. Machine learning based clinical prediction model for 1-year mortality in Sepsis patients with atrial fibrillation. Heliyon. 2024;10:e38730.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBashar SK, Hossain MB, Ding E, Walkey AJ, McManus DD, Chon KH. Atrial Fibrillation Detection During Sepsis: Study on MIMIC III ICU Data. IEEE J Biomed Health Inf. 2020;24:3124\u0026ndash;35.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBashar SK, Ding EY, Walkey AJ, McManus DD, Chon KH. Atrial Fibrillation Prediction from Critically Ill Sepsis Patients. Biosens (Basel). 2021;11.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi Z, Pang M, Li Y, Yu Y, Peng T, Hu Z, et al. Development and validation of a predictive model for new-onset atrial fibrillation in sepsis based on clinical risk factors. Front Cardiovasc Med. 2022;9:968615.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu C, Shi J, Wang F, Li D, Luo Y, Yang B, et al. Development and validation of an interpretable multi-task model to predict outcomes in patients with rhabdomyolysis: a multicenter retrospective cohort study. EClinicalMedicine. 2025;87:103438.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShi B, Ye J, Chen W, Liao B, Gu W, Yin H, et al. Prognosis of critically ill patients with early and late sepsis-associated acute kidney injury: an observational study based on the MIMIC-IV. Ren Fail. 2025;47:2441393.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYao F, Luo J, Zhou Q, Wang L, He Z. Development and validation of a machine learning-based prediction model for hepatorenal syndrome in liver cirrhosis patients using MIMIC-IV and eICU databases. Sci Rep. 2025;15:2743.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDing M, Murata S, Louro J, Hammar N, Modig K. Machine-learning approach to atrial fibrillation prediction among individuals without prior cardiovascular diseases. Open Heart. 2025;12.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlomari L, Jarrar Y, Al-Fakhouri Z, Otabor E, Lam J, Alomari J. A machine learning-based risk prediction model for atrial fibrillation in critically ill patients. Heart Rhythm O2. 2025;6:652\u0026ndash;60.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHakim FA, Shen WK. Atrial fibrillation in the elderly: a review. Future Cardiol. 2014;10:745\u0026ndash;58.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMentes O, Celik D, Yıldız M, \u0026Ouml;zdemir T, Ari M, Aksoy G\u0026uuml;ney EN et al. Atrial Fibrillation Among ICU Patients with Type 2 Respiratory Failure: Who Is at Risk and What Are the Outcomes? Diagnostics. (Basel). 2025;15.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBemis CE, Serur JR, Borkenhagen D, Sonnenblick EH, Urschel CW. Influence of right ventricular filling pressure on left ventricular pressure and dimension. Circ Res. 1974;34:498\u0026ndash;504.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTerzano C, Conti V, Di Stefano F, Petroianni A, Ceccarelli D, Graziani E, et al. Comorbidity, hospitalization, and mortality in COPD: results from a longitudinal study. Lung. 2010;188:321\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNavarro-Garcia JA, Keefe JA, Song J, Li N, Wehrens XHT. Mechanisms underlying atrial fibrillation in chronic kidney disease. J Mol Cell Cardiol. 2025;205:37\u0026ndash;51.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLin Z, Han H, Guo W, Wei X, Guo Z, Zhai S, et al. Atrial fibrillation in critically ill patients who received prolonged mechanical ventilation: a nationwide inpatient report. Korean J Intern Med. 2021;36:1389\u0026ndash;401.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTamargo J, Villacast\u0026iacute;n J, Caballero R, Delp\u0026oacute;n E. Drug-induced atrial fibrillation. A narrative review of a forgotten adverse effect. Pharmacol Res. 2024;200:107077.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBessissow A, Khan J, Devereaux PJ, Alvarez-Garcia J, Alonso-Coello P. Postoperative atrial fibrillation in non-cardiac and cardiac surgery: an overview. J Thromb Haemost. 2015;13(Suppl 1):S304\u0026ndash;312.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKarnik AA, Gopal DM, Ko D, Benjamin EJ, Helm RH. Epidemiology of Atrial Fibrillation and Heart Failure: A Growing and Important Problem. Cardiol Clin. 2019;37:119\u0026ndash;29.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-health-services-research","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bhsr","sideBox":"Learn more about [BMC Health Services Research](http://bmchealthservres.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/BHSR/default.aspx","title":"BMC Health Services Research","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Severe sepsis, New-onset atrial fibrillation, intensive care unit, Risk prediction, MIMIC database, Machine learning models","lastPublishedDoi":"10.21203/rs.3.rs-9208556/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9208556/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eNew-onset atrial fibrillation (NOAF) is a common complication in patients with severe sepsis admitted to the intensive care unit (ICU), and is associated with increased in-hospital mortality. Therefore, early identification and prediction of patients at high risk for NOAF are of great significance. This study aims to establish and validate an interpretable machine learning (ML) model for the prediction of NOAF in critically ill patients with severe sepsis.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eData from patients with severe sepsis were extracted from three datasets: the Medical Information Mart for Intensive Care (MIMIC)-IV database, the MIMIC-III database, and the eICU Collaborative Research Database (eICU). The MIMIC-IV cohort was randomly split into a training set (70%) and an internal validation set (30%). The MIMIC-III and eICU databases served as two external validation cohorts. Feature selection was performed using a combination of the LASSO regression and the Boruta algorithm. Subsequently, ten ML algorithms were used to construct prediction models. The SHapley Additive exPlanations (SHAP) method was applied to interpret the model outputs visually.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eA total of 7,168 critically ill patients with severe sepsis were included in this study, with 3,993 from the MIMIC-IV database, 1,604 from the MIMIC-III database, and 1,571 from the eICU database. 19 features were selected from 88 variables for model construction. The Random Forest (RF) method demonstrated the most optimal predictive performance in terms of discriminaiton, calibration, and clinical utility among the ten models, achieving an AUC of 0.936 in the internal validation set, 0.779 in the MIMIC-III cohort, and 0.724 in the eICU cohort. SHAP analysis revealed that age, urine output, and respiratory failure were the most important contributors to the most. The simplified model incorporating 10 features demonstrated comparable efficacy, with an AUC of 0.907 in the internal validation set, 0.769 in the MIMIC-III cohort, and 0.707 in the eICU cohort. The final optimized model has been translated into a user-friendly interface for clinical use. The web application is accessible online at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://119.3.41.228/mm/index.php\u003c/span\u003e\u003cspan address=\"http://119.3.41.228/mm/index.php\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eThis study developed an interpretable RF model to accurately predict the NOAF risk in ICU patients with severe sepsis.\u003c/p\u003e\u003ch2\u003eClinical trial number\u003c/h2\u003e \u003cp\u003eNot applicable.\u003c/p\u003e","manuscriptTitle":"Machine-learning-based online prediction models for new-onset atrial fibrillation in ICU patients with severe sepsis: development and validation in a retrospective cohort","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-05-04 07:16:00","doi":"10.21203/rs.3.rs-9208556/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-04-29T00:59:42+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"106446367228224090849783919637530139872","date":"2026-04-29T00:07:57+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-04-21T16:16:51+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-03-26T13:24:50+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-25T09:43:42+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-25T09:43:26+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Health Services Research","date":"2026-03-24T07:43:23+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-health-services-research","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bhsr","sideBox":"Learn more about [BMC Health Services Research](http://bmchealthservres.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/BHSR/default.aspx","title":"BMC Health Services Research","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"0f17df8e-d88b-440f-b24b-d64ec3b8b032","owner":[],"postedDate":"May 4th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-04T07:16:01+00:00","versionOfRecord":[],"versionCreatedAt":"2026-05-04 07:16:00","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9208556","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9208556","identity":"rs-9208556","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}