Development and external validation of a machine learning model for predicting 28-day mortality in ICU patients with acute coronary syndrome complicated by heart failure

Development and external validation of a machine learning model for predicting 28-day mortality in ICU patients with acute coronary syndrome complicated by heart failure | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Development and external validation of a machine learning model for predicting 28-day mortality in ICU patients with acute coronary syndrome complicated by heart failure wencai Jiang, Longzhou Chen, Yucai Tang, Xuejun Deng This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9218650/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 5 You are reading this latest preprint version Abstract Background Acute coronary syndrome complicated by heart failure is a common condition in intensive care units (ICUs) and is associated with a poor prognosis and a markedly increased short-term mortality risk. Most existing prediction models are based on traditional regression approaches and lack external validation, and their applicability in critically ill populations remains uncertain. This study aimed to develop and externally validate a machine learning–based model to predict 28-day mortality in this population. Methods Patients with acute coronary syndrome complicated by heart failure were identified from the MIMIC-IV database (v3.1) as the training cohort (n = 3,410) and from the eICU-CRD database (v2.0) as the external validation cohort (n = 984). Feature selection was performed using the Boruta algorithm and LASSO regression, resulting in 20 predictors. Ten machine learning algorithms were evaluated, including logistic regression, decision tree, random forest, gradient boosting, AdaBoost, XGBoost, CatBoost, LightGBM, support vector machine, and k-nearest neighbors. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value, and negative predictive value. SHAP analysis was used to interpret the model, and a web-based calculator was developed based on the optimal model. Results Among the evaluated models, CatBoost demonstrated the best performance in the internal validation cohort, with an AUC of 0.855, accuracy of 0.862, specificity of 0.976, and negative predictive value of 0.872. In the external validation cohort, the model achieved an AUC of 0.676, with specificity of 0.960 and negative predictive value of 0.832, although sensitivity remained relatively low (0.189). SHAP analysis identified SAPS II, red cell distribution width, blood urea nitrogen, arterial partial pressure of oxygen, and vasopressor use as the most influential predictors. Conclusion The CatBoost model demonstrated high specificity and negative predictive value for predicting 28-day mortality in ICU patients with acute coronary syndrome complicated by heart failure, suggesting potential utility in identifying low-risk patients. A web-based calculator was developed to facilitate individualized and interpretable risk assessment. Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction Acute coronary syndrome (ACS) remains a major contributor to morbidity and mortality worldwide, particularly among critically ill patients requiring intensive care. According to the Global Burden of Disease (GBD) 2019 study, ischemic heart disease is still the leading cause of death globally, accounting for approximately 9.14 million deaths each year( 1 – 3 ). Heart failure (HF), a frequent complication of ACS, has been reported in approximately 15% to 30% of hospitalized patients and is associated with a markedly increased risk of short-term mortality( 4 – 6 ). In the ICU, patients with ACS complicated by HF often present with hemodynamic instability, multi-organ dysfunction, and multiple comorbidities, all of which contribute to poor clinical outcomes( 7 – 9 ). The interplay between ACS and HF involves a complex and dynamic pathophysiological process. Myocardial ischemia can impair cardiac function and trigger ventricular remodeling, ultimately leading to the development or worsening of HF( 10 – 12 ). Conversely, HF further reduces systemic perfusion, promoting renal dysfunction, metabolic abnormalities, and systemic inflammation, thereby creating a vicious cycle that worsens prognosis, especially in critically ill patients( 13 – 15 ). Most existing prognostic models for cardiovascular disease or ICU mortality are developed using conventional statistical approaches and are typically derived from general populations( 16 , 17 ). As a result, they may not adequately capture nonlinear relationships or complex interactions among variables, limiting their applicability in specific conditions such as ACS complicated by HF. Therefore, more accurate and individualized prediction tools are still needed for this high-risk population( 18 , 19 ). In recent years, machine learning (ML) techniques have shown strong potential in analyzing high-dimensional clinical data and identifying complex patterns( 20 , 21 ). In addition, the use of explainable methods, such as Shapley additive explanations (SHAP), allows for better interpretation of model predictions and may enhance clinical acceptance. In this study, we developed and externally validated an interpretable ML model to predict 28-day mortality in ICU patients with ACS complicated by HF, and further implemented a web-based calculator to facilitate individualized risk assessment in clinical practice( 22 , 23 ). Methods Data sources In this retrospective study, the training cohort was obtained from the Medical Information Mart for Intensive Care IV (MIMIC-IV, version 3.1), a large, publicly accessible database containing detailed clinical data of patients admitted to intensive care units at the Beth Israel Deaconess Medical Center (BIDMC), USA, between 2008 and 2022. The database was developed with approval from the Institutional Review Boards of BIDMC and the Massachusetts Institute of Technology. The author, Longzhou Chen, completed the Collaborative Institutional Training Initiative (CITI) program and obtained certification to access the database (certification number: 74101068). The external validation cohort was derived from the eICU Collaborative Research Database (eICU-CRD, version 2.0), which includes deidentified, multicenter clinical data from more than 200,000 ICU admissions across 208 hospitals in the United States between 2014 and 2015. As both databases are publicly available and contain fully anonymized patient information, this study was exempt from the requirement for informed consent and additional ethical approval. Study population This retrospective study was based on two large critical care databases. The training cohort was obtained from the MIMIC-IV database (version 3.1), including ICU admissions from 2008 to 2022, and the external validation cohort was derived from the eICU Collaborative Research Database (eICU-CRD, version 2.0), which contains multicenter ICU data collected between 2014 and 2015. Patients diagnosed with both ACS and HF were included in the study. The diagnoses were identified according to standard clinical criteria and corresponding ICD codes. The inclusion criteria were: ( 1 ) age ≥ 18 years; ( 2 ) first ICU admission; and ( 3 ) concurrent diagnosis of ACS and HF. The patient selection flowchart is presented in Fig. 1 . Patients with an ICU stay of less than 24 hours were excluded. For the external validation cohort, the variables were selected according to the features identified in the training cohort using Boruta and LASSO. Patients with substantial missing data in these variables were excluded to ensure data completeness. The primary outcome was 28-day all-cause mortality after ICU admission. Data extraction and feature engineering Data extraction was performed using Navicat Premium (version 16.2.11; PremiumSoft CyberTech Ltd., Singapore) and pgAdmin 4 (version 8.13). Patients meeting the diagnostic criteria for ACS complicated by HF were identified, and the relevant clinical data were retrieved from the databases. Data preprocessing and subsequent analyses were conducted using Python (version 3.11; Python Software Foundation, Wilmington, DE, USA). Clinical variables within the first 24 hours after ICU admission were included. An initial set of 50 candidate variables was collected based on clinical relevance and data availability. Variables with more than 30% missing values were excluded. For the remaining variables, missing data were handled using multiple imputation with predictive mean matching (PMM). Feature selection was performed in several steps. First, the Boruta algorithm was applied to screen variables with potential predictive value. Then, multicollinearity was assessed, and highly correlated variables were removed. Finally, least absolute shrinkage and selection operator (LASSO) regression with cross-validation was used to further reduce the number of variables and select the most informative predictors(Figure S1 -S2). To further reduce redundancy among variables, multicollinearity was assessed and highly correlated variables were removed, resulting in a more stable feature set for model development. After this process, 20 variables were retained and used for model development(Figure S3). Statistical analysis All statistical analyses were performed using Python (version 3.11.9). Continuous variables were assessed for normality using the Shapiro–Wilk test. Variables following a normal distribution are presented as mean ± standard deviation (SD) and were compared using the Student's t-test. Non-normally distributed variables are reported as median (interquartile range [IQR]) and were compared using the Mann–Whitney U test. Categorical variables are expressed as counts (percentages) and were analyzed using the chi-square test or Fisher's exact test, as appropriate. The MIMIC-IV database (version 3.1) was used as the training cohort for model development, while the eICU Collaborative Research Database (eICU-CRD, version 2.0) served as an independent external validation cohort to evaluate model generalizability. Missing data were addressed using multiple imputation with predictive mean matching prior to analysis. A total of ten machine learning algorithms were developed and compared for predicting 28-day mortality in ICU patients with acute coronary syndrome complicated by heart failure. These included logistic regression, decision tree, random forest, gradient boosting, AdaBoost, extreme gradient boosting (XGBoost), CatBoost, LightGBM, support vector machine (SVM), and k-nearest neighbors (KNN). Model training was conducted exclusively on the training cohort, and no data from the external validation cohort were used during model development to prevent data leakage. Model performance was evaluated using accuracy, area under the receiver operating characteristic curve (AUC), F1 score, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). AUC was used as the primary measure of discriminative ability. The F1 score was included to reflect the balance between sensitivity and precision. PPV and NPV indicate the proportions of true positive and true negative predictions among all positive and negative predictions, respectively. To improve model interpretability, SHapley Additive exPlanations (SHAP) were applied. SHAP values, derived from cooperative game theory, were used to quantify the contribution of each feature to individual predictions. A SHAP summary plot was generated to present global feature importance, and SHAP dependence plots were used to examine relationships between feature values and model outputs. In addition, SHAP force plots and waterfall plots were used to visualize feature contributions at the individual level. Partial dependence plots (PDPs) were further constructed to evaluate the marginal effects of key variables on predicted risk. All statistical tests were two-sided, and a p-value < 0.05 was considered statistically significant. Results Baseline characteristics Based on the predefined inclusion and exclusion criteria, a total of 3,410 ICU patients diagnosed with ACS complicated by HF were identified from the MIMIC-IV database (version 3.1) and included as the training cohort. Among them, 2,825 patients survived within 28 days, while 585 patients died. For external validation, 984 eligible patients were extracted from the eICU Collaborative Research Database (eICU-CRD, version 2.0), including 794 survivors and 190 non-survivors at 28 days. The baseline characteristics of patients in both cohorts are presented in Table 1 . Table 1 Patient demographics and baseline characteristics. Variables MIMIC-IV (v3.1) eICU-CRD (v2.0) Total (n = 3410) Survivors (n = 2,825) Non-survivors (n = 585) p-value Total (n = 984) Survivors (n = 794) Non-survivors (n = 190) p-value Age (years) 75.00 (66.00–84.00) 75.00 (66.00–84.00) 75.00 (66.00–84.00) 0.802 70.00 (60.00–79.00) 69.00 (59.00–78.00) 73.50 (63.00–81.00) 0.009 Pco2 (mmHg) 41.00 (35.00–47.00) 41.00 (36.00–46.00) 41.00 (35.00–49.00) 0.05 45.00 (38.80–55.00) 45.00 (39.00–54.50) 46.00 (38.00–56.65) 0.522 Po2 (mmHg) 96.00 (50.00–258.00) 104.00 (53.00–287.00) 76.00 (42.00–135.00) < 0.001 145.00 (100.00–229.80) 144.00 (100.00–220.00) 156.90 (93.00–246.00) 0.793 PH 7.37 (7.31–7.42) 7.38 (7.32–7.43) 7.33 (7.25–7.40) < 0.001 7.44 (7.39–7.49) 7.44 (7.39–7.49) 7.43 (7.35–7.51) 0.013 Na+(mmol/L) 138.00 (135.00–141.00) 138.00 (135.00–141.00) 138.00 (134.00–142.00) 0.476 143.00 (140.00–147.00) 143.00 (140.00–146.00) 143.00 (139.00–148.00) 0.342 CL-(mmol/L) 103.00 (99.00–107.00) 104.00 (100.00–107.00) 101.00 (97.00–106.00) < 0.001 109.00 (105.00–113.00) 109.00 (105.00–113.00) 110.00 (105.00–114.00) 0.141 Mg2+(mg/dL) 2.10 (1.90–2.40) 2.10 (1.90–2.40) 2.10 (1.90–2.40) 0.92 2.30 (2.07–2.60) 2.30 (2.01–2.60) 2.40 (2.20–2.80) 0.028 AG (mEq/L) 15.00 (12.00–18.00) 14.00 (12.00–17.00) 17.00 (14.00–21.00) < 0.001 13.00 (10.00–16.30) 12.00 (10.00–15.10) 16.00 (13.00–20.00) < 0.001 RBC (m/uL) 3.47 (2.92–4.06) 3.47 (2.92–4.07) 3.48 (2.94–4.03) 0.925 3.92 (3.49–4.44) 3.92 (3.49–4.44) 3.88 (3.50–4.46) 0.336 WBC (K/uL) 11.70 (8.60–16.00) 11.50 (8.40–15.60) 13.00 (9.40–18.20) < 0.001 16.00 (11.60–21.35) 15.57 (11.30–20.40) 18.00 (13.70–25.10) < 0.001 PLT (K/uL) 195.00 (145.00–256.00) 196.00 (146.00–256.00) 191.00 (139.00–259.00) 0.929 241.00 (178.00–326.00) 251.50 (186.50–334.00) 213.00 (156.00–263.00) < 0.001 RDW(%) 14.60 (13.60–16.00) 14.50 (13.50–15.80) 15.40 (14.10–17.30) < 0.001 15.60 (14.50–17.30) 15.50 (14.40–17.00) 16.30 (14.70–18.30) 0.002 Glu (mg/dL) 139.00 (113.00–190.00) 136.00 (112.00–185.00) 160.00 (118.00–213.00) < 0.001 221.50 (167.00–295.50) 221.00 (165.00–296.00) 224.50 (180.00–293.00) 0.51 BUN (mg/dL) 28.00 (18.00–46.50) 26.00 (18.00–43.00) 39.50 (25.00–62.00) < 0.001 41.00 (25.00–62.00) 38.00 (24.00–62.00) 49.00 (31.00–62.00) 0.004 Creatinine (mg/dL) 1.30 (0.90–2.10) 1.20 (0.90–2.00) 1.70 (1.20–2.80) < 0.001 1.72 (1.12–3.11) 1.57 (1.08–2.76) 2.44 (1.57–3.72) 0.001 ALT (IU/L) 32.00 (18.00–77.00) 30.00 (17.00–66.00) 44.00 (21.00–126.00) < 0.001 48.00 (27.00–137.00) 42.00 (25.00–100.50) 132.00 (38.00–632.00) < 0.001 SOFA 5.00 (3.00–8.00) 5.00 (3.00–7.00) 7.00 (5.00–10.00) < 0.001 6.00 (4.00–9.00) 6.00 (4.00–9.00) 9.00 (5.00–13.00) < 0.001 SAPSII 39.00 (32.00–49.00) 38.00 (31.00–46.00) 50.00 (40.00–62.00) < 0.001 42.00 (32.00–55.00) 40.00 (31.00–51.00) 55.50 (42.00–69.00) < 0.001 Diuretic use 1,308.00 (38.36%) 1,225.00 (43.36%) 83.00 (14.19%) < 0.001 11.00 (1.12%) 10.00 (1.26%) 1.00 (0.53%) 0.388 Inotrope use 684.00 (20.06%) 473.00 (16.74%) 211.00 (36.07%) < 0.001 454.00 (46.14%) 380.00 (47.86%) 74.00 (38.95%) 0.027 Vasopressor use 2,014.00 (59.06%) 1,552.00 (54.94%) 462.00 (78.97%) < 0.001 260.00 (26.42%) 177.00 (22.29%) 83.00 (43.68%) < 0.001 Data are presented as median (interquartile range, 25th–75th percentile) or n (%). AG, anion gap; ALT, alanine aminotransferase; BUN, blood urea nitrogen; Cl⁻, chloride; Cr, creatinine; Glu, glucose; Mg²⁺, magnesium; Na⁺, sodium; PaCO₂, partial pressure of carbon dioxide; PaO₂, partial pressure of oxygen; pH, potential of hydrogen; PLT, platelet count; RBC, red blood cell count; RDW, red cell distribution width; WBC, white blood cell count; SOFA, Sequential Organ Failure Assessment; SAPS II, Simplified Acute Physiology Score II; loop diuretic use, use of loop diuretics; inotrope use, use of inotropic agents; vasopressor use, use of vasopressors. Comparison of machine learning models Among all the models evaluated, CatBoost showed the best overall performance. It achieved the highest AUC (0.855) and accuracy (0.862), together with a relatively high F1 score (0.438) and PPV (0.733). In addition, it maintained a good balance between sensitivity (0.312) and specificity (0.976), with a high NPV (0.872) (Table 2 ). Gradient Boosting and XGBoost also demonstrated good discrimination, with AUC values of 0.839 and 0.830, respectively, although their overall performance was slightly lower than that of CatBoost. Logistic regression showed high specificity (0.967), but its sensitivity was limited (0.335), suggesting reduced ability to identify high-risk patients. Similar patterns were observed in Random Forest and SVM, which tended to have high specificity but relatively low sensitivity. Taken together, CatBoost provided a more balanced performance across different evaluation metrics. Therefore, it was selected as the final model for subsequent analyses (Fig. 2 A). Table 2 Prediction performance of each model in internal validation sets. Model Accuracy AUC F1 Sensitivity Specificity PPV NPV Logistic Regression 0.858 0.824 0.449 0.335 0.967 0.678 0.875 Decision Tree 0.785 0.618 0.349 0.335 0.878 0.364 0.864 Random Forest 0.85 0.831 0.326 0.21 0.983 0.725 0.857 Gradient Boosting 0.86 0.839 0.435 0.312 0.974 0.714 0.872 AdaBoost 0.856 0.821 0.405 0.284 0.975 0.704 0.868 XGBoost 0.857 0.83 0.437 0.324 0.967 0.671 0.873 CatBoost 0.862 0.855 0.438 0.312 0.976 0.733 0.872 LightGBM 0.849 0.828 0.412 0.307 0.962 0.628 0.87 SVM 0.852 0.799 0.335 0.216 0.985 0.745 0.858 KNN 0.847 0.775 0.379 0.273 0.966 0.623 0.865 AUC: Area under the receiver operating characteristic curve; NPV: Negative predictive value; PPV: Positive predictive value. (A) ROC curves of ten machine learning models in the training cohort (MIMIC-IV). Among them, CatBoost achieved the highest discriminative performance (AUC = 0.857). (B) ROC curves of the CatBoost model in the internal and external validation cohorts. The model maintained good discrimination in the external validation cohort (eICU-CRD), although a slight decrease in AUC was observed compared to the training cohort. AUC: Area under the curve; eICU-CRD: eICU collaborative research database; ICU: Intensive care unit; MIMIC-IV: Medical information mart for intensive care IV; ROC: Receiver operating characteristic. For external validation, the CatBoost model was further evaluated using the eICU-CRD (v2.0) cohort. The receiver operating characteristic curve for the external cohort is presented in Fig. 2 B. In the external cohort, the model achieved an AUC of 0.676 with an accuracy of 0.811 and an F1-score of 0.279. Sensitivity was relatively low (0.189), whereas specificity remained high (0.960). The PPV and NPV were 0.529 and 0.832, respectively. As shown in Fig. 1 B, the ROC curve of the external cohort was consistently lower than that of the internal cohort (AUC: 0.803 vs. 0.676), indicating a reduction in discriminative performance across datasets. Despite this decline, the model maintained a relatively high specificity and NPV, suggesting that it remains useful for identifying patients at low risk of 28-day mortality. However, the low sensitivity indicates that a proportion of high-risk patients may not be captured in the external population. Overall, these findings suggest that although the model performance decreased in the external cohort, it retained moderate discrimination and may still provide supportive information for clinical risk stratification. Explanation of risk factors To further interpret the CatBoost model and identify the main contributors, SHAP analysis was performed. Figure 3 A shows the feature importance ranked by mean absolute SHAP values. Among all variables, diuretic use had the greatest impact on model output, followed by SAPSII score, vasopressor use, and Po2. Other important features included RDW, BUN, chloride (Cl−), and creatinine, suggesting that both clinical interventions and laboratory indicators contributed to mortality prediction. Figure 3 B presents the distribution of SHAP values for each feature, illustrating how feature values influence the predicted risk. Higher SAPSII scores, RDW, BUN, and the use of vasopressors were generally associated with increased mortality risk. In contrast, higher Po2 levels and chloride tended to be associated with lower predicted risk. The effect of diuretic use appeared complex, with both positive and negative contributions depending on the clinical context. AG, anion gap; ALT, alanine aminotransferase; BUN, blood urea nitrogen; Cl⁻, chloride; Cr, creatinine; Glu, glucose; Mg²⁺, magnesium; Na⁺, sodium; PaCO₂, partial pressure of carbon dioxide; PaO₂, partial pressure of oxygen; pH, potential of hydrogen; PLT, platelet count; RBC, red blood cell count; RDW, red cell distribution width; WBC, white blood cell count; SOFA, Sequential Organ Failure Assessment; SAPS II, Simplified Acute Physiology Score II; loop diuretic use, use of loop diuretics; inotrope use, use of inotropic agents; vasopressor use, use of vasopressors. Interpretation of individual prediction To illustrate how individual features contribute to model predictions, two representative cases were selected (Figure S). In the SHAP waterfall plots, red bars represent features associated with increased predicted risk, whereas blue bars indicate features that reduce the predicted risk. In Figure S5A, a patient with a low predicted risk (0.01) is shown. Several variables contributed to a reduction in the model output. In particular, the presence of diuretic use and a relatively low SAPSII score had the largest negative impact. Lower BUN levels, higher pH, and lower RDW were also associated with decreased predicted risk. Although vasopressor use and AG contributed in the opposite direction, their effects were modest. Overall, the combined effect of these features resulted in a low predicted probability, consistent with the observed outcome. Figure S5B shows a patient with a higher predicted risk (0.21). In this case, the absence of diuretic use, a higher SAPSII score, and vasopressor use contributed to an increase in the model output. Elevated BUN, lower Po2, and higher WBC were also associated with higher predicted risk. In contrast, lower RDW and higher chloride levels contributed to a reduction in risk. The net effect of these factors led to a higher predicted probability compared with the low-risk case. These examples highlight how different combinations of clinical variables can influence individual predictions, reflecting the model's ability to capture complex patterns in the data. Implementation of web-based calculator Based on the final CatBoost model, an interactive web-based calculator was developed to facilitate clinical application (Fig. 4 ). The interface allows users to input routinely available clinical variables, after which the model provides an estimated probability of 28-day mortality in real time. As shown in Fig. 4 , the tool also incorporates SHAP-based visualization to enhance interpretability. A force plot is used to display the overall direction and magnitude of feature contributions, while a waterfall plot illustrates how individual variables cumulatively influence the final prediction for a given patient. In addition, the predicted probability is presented in a dedicated results panel, allowing for straightforward interpretation. The calculator is publicly available at https://repository-name-mlapp-ujfpaagtkdpqmzgklgjjbf.streamlit.app and is also available via QR code, enabling convenient use on mobile devices. This tool may assist clinicians in individualized risk assessment and support clinical decision-making in patients with ACS complicated by heart HF. AG, anion gap; ALT, alanine aminotransferase; BUN, blood urea nitrogen; Cl⁻, chloride; Cr, creatinine; Glu, glucose; Mg²⁺, magnesium; Na⁺, sodium; PaCO₂, partial pressure of carbon dioxide; PaO₂, partial pressure of oxygen; pH, potential of hydrogen; PLT, platelet count; RBC, red blood cell count; RDW, red cell distribution width; WBC, white blood cell count; SOFA, Sequential Organ Failure Assessment; SAPS II, Simplified Acute Physiology Score II; loop diuretic use, use of loop diuretics; inotrope use, use of inotropic agents; vasopressor use, use of vasopressors. Discussion In this study, we developed and externally validated a machine learning model based on the MIMIC-IV database to predict 28-day mortality in ICU patients with acute coronary syndrome complicated by heart failure. Among the candidate algorithms, CatBoost showed the most consistent performance. Although the model demonstrated satisfactory discrimination in the internal cohort, a decline in performance was observed in the external validation dataset, suggesting that its generalizability across different clinical settings remains limited. Notably, the model achieved high specificity and negative predictive value, indicating potential utility in identifying patients at relatively low risk. To facilitate clinical implementation, we further developed a web-based calculator integrating SHAP-based visualization. While the application of machine learning in critical care prognostication has increased substantially, external validation is still lacking in many studies, despite its importance in assessing real-world applicability ( 24 , 25 ). Traditional ICU prognostic tools, such as APACHE and SAPS scores, are widely used but are generally derived from regression-based approaches that assume linear relationships between predictors and outcomes( 26 , 27 ). In practice, however, clinical data often exhibit nonlinear patterns and complex interactions. A systematic review including 30 studies reported that machine learning approaches achieved comparable or even improved discrimination compared with conventional scoring systems in predicting ICU mortality( 28 ). In addition, widely used models for acute coronary syndrome or heart failure risk assessment, including GRACE, MAGGIC, and the Seattle Heart Failure model, were primarily developed in non-ICU populations, which may limit their applicability in critically ill patients( 29 – 31 ). Although machine learning methods have been increasingly introduced into this field, most studies rely on single datasets and lack external validation( 32 ). By incorporating both MIMIC-IV and eICU-CRD for model development and validation, the present study attempts to improve the robustness of model evaluation( 33 ). The relatively favorable performance of CatBoost in this study may be related to its ability to handle categorical variables efficiently and to mitigate overfitting( 34 , 35 ). SHAP analysis provided further insight into the model's decision process. SAPS II, as a comprehensive measure of disease severity, contributed most strongly to the prediction, consistent with its established role in ICU prognostication( 27 , 36 ). In addition, RDW and BUN were identified as important contributors, suggesting that inflammatory status and renal function play key roles in short-term outcomes in this population. Previous studies have shown that elevated RDW is associated with impaired cardiac function and adverse cardiovascular events( 37 ), while BUN reflects both renal dysfunction and neurohormonal activation, which may influence circulatory stability( 38 ). Among physiological variables, higher PaO₂ was associated with a lower predicted risk, whereas the use of vasopressors, as an indicator of hemodynamic instability, was associated with worse outcomes( 39 ). Notably, the contribution of diuretic use appeared bidirectional, which may reflect heterogeneity in clinical indications( 40 ). The anion gap also contributed to the prediction, suggesting a potential role of metabolic disturbances in patient outcomes( 41 ). Overall, short-term mortality in this population is likely driven by the combined effects of disease severity, inflammatory response, renal dysfunction, and hemodynamic instability. The decline in model performance observed in the external validation cohort is likely multifactorial. Differences in patient characteristics, disease severity, and treatment strategies between datasets may result in distributional shifts that affect model predictions. In addition, variations in data collection practices and missing data patterns across institutions may further influence performance. Previous studies have reported that performance degradation across populations is a common phenomenon in predictive modeling, underscoring the importance of external validation in evaluating clinical applicability( 42 ). In addition, the relatively low sensitivity observed in this study suggests that the model may have limited ability to identify high-risk patients. This may be partly attributable to class imbalance, with fewer mortality events in the dataset, leading the model to prioritize reducing false positives and thereby increasing specificity at the expense of sensitivity( 43 ). In clinical practice, this trade-off should be interpreted with caution, particularly in settings where early identification of high-risk patients is critical. The high specificity (0.976) and negative predictive value (0.872) indicate that the model may be particularly useful for identifying low-risk patients. In the ICU setting, accurate identification of patients with a relatively favorable prognosis may help optimize resource allocation and improve management efficiency. The web-based calculator developed in this study integrates SHAP visualization, enhancing the transparency of model predictions and facilitating clinician understanding. As machine learning continues to evolve in healthcare, such models are more likely to serve as complementary tools rather than replacements for traditional clinical assessment, supporting decision-making processes. Several limitations should be acknowledged. First, both MIMIC-IV and eICU-CRD are derived from healthcare systems in the United States, and the generalizability of the model to other populations remains uncertain. Second, the retrospective design limited the inclusion of certain clinically relevant variables, such as echocardiographic parameters and NT-proBNP. Third, model performance declined in the external cohort (AUC decreased from 0.855 to 0.676), and sensitivity remained relatively low (0.312 internally and 0.189 externally), indicating suboptimal identification of high-risk patients. Therefore, model predictions should be interpreted cautiously in clinical practice. Future studies should focus on prospective validation in more diverse populations and explore the integration of temporal data to further improve predictive performance. Conclusion In this study, we developed and externally validated a CatBoost model for predicting 28-day mortality in ICU patients with acute coronary syndrome complicated by heart failure. The model demonstrated high specificity and negative predictive value, suggesting potential utility in identifying patients at lower risk. A web-based calculator incorporating SHAP visualization was also implemented to support individualized risk assessment with improved interpretability. Further prospective validation in more diverse populations is warranted to better assess the model's generalizability. Abbreviations Abbreviation Full Form GBD Global Burden of Disease eICU-CRD eICU Collaborative Research Database MIMIC-IV Medical Information Mart for Intensive Care IV CITI Collaborative Institutional Training Initiative ICD International Classification of Diseases IQR Interquartile Range LASSO Least Absolute Shrinkage and Selection Operator Boruta Boruta feature selection algorithm ML Machine Learning BIDMC Beth Israel Deaconess Medical Center KNN K-Nearest Neighbors LightGBM Light Gradient Boosting Machine SVM Support Vector Machine CatBoost Categorical Boosting XGBoost Extreme Gradient Boosting PMM Predictive Mean Matching PPV Positive Predictive Value NPV Negative Predictive Value ROC Receiver Operating Characteristic AUC Area Under the Receiver Operating Characteristic Curve SHAP Shapley Additive Explanations ACS Acute Coronary Syndrome HF Heart Failure AG Anion Gap ALT Alanine Aminotransferase BUN Blood Urea Nitrogen Cr Creatinine Glu Glucose Cl- Chloride Mg2+ Magnesium Na+ Sodium PaCO2 Partial Pressure of Carbon Dioxide PaO2 Partial Pressure of Oxygen pH Potential of Hydrogen PLT Platelet Count RBC Red Blood Cell Count WBC White Blood Cell Count RDW Red Cell Distribution Width SAPS II Simplified Acute Physiology Score I SOFA Sequential Organ Failure Assessment Declarations Ethics approval and consent to participate This study was a retrospective analysis of publicly available critical care databases (MIMIC-IV and eICU-CRD). Both databases received institutional review board approval and contain deidentified patient data. The requirement for informed consent was waived. Therefore, additional ethical approval was not required. Consent for publication Not applicable. This study does not contain any individual person’s data in any form Clinical trial number not applicable. Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request. The data are derived from the MIMIC-IV and eICU-CRD database, which is publicly available but requires credentialed access through the official PhysioNet platform. Competing interests The authors declare no competing interests. Funding This work was supported by the Sichuan Science and Technology Program (Grant No. SCJJ25RKX011, "Research on the Precision Prevention and Control System for Common High-Incidence Chronic Major Diseases"). Author contributions Wencai Jiang and Longzhou Chen contributed equally to this work. Wencai Jiang: Conceptualization, Data curation, Formal analysis, Methodology, Software, Visualization, Writing – original draft. Longzhou Chen: Conceptualization, Data curation, Formal analysis, Methodology, Software, Validation, Writing – review & editing. Yucai Tang: Investigation, Resources, Validation, Visualization, Writing – review & editing. Xuejun Deng: Investigation, Resources, Writing – review & editing. All authors approved the final manuscript. Acknowledgements The authors thank the contributors to the MIMIC-IV and eICU-CRD databases for making the data publicly available. We also appreciate the support from the research team at Suining Central Hospital. References Tsao CW, Aday AW, Almarzooq ZI, Anderson CAM, Arora P, Avery CL, et al. Heart disease and stroke statistics-2023 update: a report from the american heart association. Circulation. 2023;147(8):e93–621. 10.1161/CIR.0000000000001123 . Roth GA, Mensah GA, Johnson CO, Addolorato G, Ammirati E, Baddour LM, et al. Global burden of cardiovascular diseases and risk factors, 1990–2019: update from the gbd 2019 study. J Am Coll Cardiol. 2020;76(25):2982–3021. 10.1016/j.jacc.2020.11.010 . Global burden of 369 diseases and injuries in 204 countries and territories, 1990–2019: a systematic analysis for the global burden of disease study 2019. Lancet (London, England). (2020) 396(10258):1204-22. 10.1016/S0140-6736(20)30925-9 Yunoki K, Naruko T, Komatsu R, Ehara S, Shirai N, Sugioka K, et al. Enhanced expression of haemoglobin scavenger receptor in accumulated macrophages of culprit lesions in acute coronary syndromes. Eur Heart J. 2009;30(15):1844–52. 10.1093/eurheartj/ehp257 . De Filippo O, D'Ascenzo F, Wańha W, Leonardi S, Raposeiras Roubin S, Fabris E, et al. Incidence and predictors of heart failure after acute coronary syndrome: the coralys registry. Int J Cardiol. 2023;370:35–42. https://doi.org/10.1016/j.ijcard.2022.10.146 . Song L, Li Y, Nie S, Feng Z, Liu Y, Ding F, et al. Using machine learning to predict adverse events in acute coronary syndrome: a retrospective study. Clin Cardiol. 2023;46(12):1594–602. https://doi.org/10.1002/clc.24127 . Li J, Liu S, Hu Y, Zhu L, Mao Y, Liu J. Predicting mortality in intensive care unit patients with heart failure using an interpretable machine learning model: retrospective cohort study. J Med Internet Res. 2022;24(8):e38082. 10.2196/38082 . Tu G, Cai Z, Wu L, Yu H, Jiang H, Luo H. Predicting in-hospital mortality in icu patients with coronary heart disease and diabetes mellitus using machine learning models. PLoS ONE. 2025;20(8):e330381. 10.1371/journal.pone.0330381 . Lee H, Yang H, Ryu HG, Jung C, Cho YJ, Yoon SB, et al. Real-time machine learning model to predict in-hospital cardiac arrest using heart rate variability in icu. Npj Digit Med. 2023;6(1):215. 10.1038/s41746-023-00960-2 . Frangogiannis NG. The inflammatory response in myocardial injury, repair, and remodelling. Nat Rev Cardiol. 2014;11(5):255–65. 10.1038/nrcardio.2014.28 . Heusch G. Myocardial ischemia/reperfusion: translational pathophysiology of ischemic heart disease. Med (New York N Y). 2024;5(1):10–31. 10.1016/j.medj.2023.12.007 . Buja LM. Pathobiology of myocardial ischemia and reperfusion injury: models, modes, molecular mechanisms, modulation, and clinical applications. Cardiol Rev. 2023;31(5):252–64. 10.1097/CRD.0000000000000440 . Rangaswami J, Bhalla V, Blair JEA, Chang TI, Costa S, Lentine KL, et al. Cardiorenal syndrome: classification, pathophysiology, diagnosis, and treatment strategies: a scientific statement from the american heart association. Circulation. 2019;139(16):e840–78. 10.1161/CIR.0000000000000664 . Dick SA, Epelman S. Chronic heart failure and inflammation: what do we really know? Circ Res. 2016;119(1):159–76. 10.1161/CIRCRESAHA.116.308030 . Kawakami R, Sakamoto A, Kawai K, Gianatti A, Pellegrini D, Nasr A, et al. Pathological evidence for SARS-cov-2 as a cause of myocarditis: jacc review topic of the week. J Am Coll Cardiol. 2021;77(3):314–25. 10.1016/j.jacc.2020.11.031 . Johnson AEW, Bulgarelli L, Shen L, Gayles A, Shammout A, Horng S, et al. Mimic-iv, a freely accessible electronic health record dataset. Sci Data. 2023;10(1):1. 10.1038/s41597-022-01899-x . Shillan D, Sterne JAC, Champneys A, Gibbison B. Use of machine learning to analyse routinely collected intensive care unit data: a systematic review. Crit Care (London England). 2019;23(1):284. 10.1186/s13054-019-2564-9 . Goldstein BA, Navar AM, Pencina MJ, Ioannidis JPA. Opportunities and challenges in developing risk prediction models with electronic health records data: a systematic review. J Am Med Inf Association: Jamia. 2017;24(1):198–208. 10.1093/jamia/ocw042 . Beam AL, Kohane IS. Big data and machine learning in health care. JAMA. 2018;319(13):1317–18. 10.1001/jama.2017.18391 . Ren H, Sun Y, Xu C, Fang M, Xu Z, Jing F, et al. Predicting acute onset of heart failure complicating acute coronary syndrome: an explainable machine learning approach. Curr Probl Cardiol. 2023;48(2):101480. https://doi.org/10.1016/j.cpcardiol.2022.101480 . Gui L, Hu Y, Ouyang H, Zhuang H, Peng Y, Yang J, et al. Predicting the rapid progression of coronary artery lesions in patients with acute coronary syndrome based on machine learning. Front Cardiovasc Med. 2025;12:1535406. 10.3389/fcvm.2025.1535406 . Lundberg SM, Erion G, Chen H, DeGrave A, Prutkin JM, Nair B, et al. From local explanations to global understanding with explainable ai for trees. Nat Mach Intell. 2020;2(1):56–67. 10.1038/s42256-019-0138-9 . Sun H, Kang M, Zhang H, Jia J, Wang Q. Machine learning for predicting mortality in intensive care unit patients: a prognostic performance systematic review and meta-analysis. Nurs Crit Care. 2025;30(6):e70206. 10.1111/nicc.70206 . Arshi B, Cowley LE, Rijnhart E, Reeve K, Smits LJ, Wynants L. External validation, impact assessment and clinical utilization of clinical prediction models: a prospective cohort study. J Clin Epidemiol. 2025;186:111902. 10.1016/j.jclinepi.2025.111902 . Gallitto G, Englert R, Kincses B, Kotikalapudi R, Li J, Hoffschlag K, et al. External validation of machine learning models-registered models and adaptive sample splitting. Gigascience. 2025;14. 10.1093/gigascience/giaf036 . Kahraman F, Yılmaz AS, Demir M, Beşiroğlu F. Apache ii score predicts in-hospital mortality more accurately than inflammatory indices in patients with acute coronary syndrome. Kardiologiia. 2022;62(9):54–9. 10.18087/cardio.2022.9.n1979 . Le Gall JR, Lemeshow S, Saulnier F. A new simplified acute physiology score (saps ii) based on a european/north american multicenter study. JAMA. 1993;270(24):2957–63. Pocock SJ, Ariti CA, McMurray JJV, Maggioni A, Køber L, Squire IB, et al. Predicting survival in heart failure: a risk score based on 39 372 patients from 30 studies. Eur Heart J. 2013;34(19):1404–13. 10.1093/eurheartj/ehs337 . Fox KAA, Dabbous OH, Goldberg RJ, Pieper KS, Eagle KA, Van de Werf F, et al. Prediction of risk of death and myocardial infarction in the six months after presentation with acute coronary syndrome: prospective multinational observational study (grace). Bmj (Clinical Res Ed). 2006;333(7578):1091. Li S, Marcus P, Núñez J, Núñez E, Sanchis J, Levy WC. Validity of the seattle heart failure model after heart failure hospitalization. Esc Heart Fail. 2019;6(3):509–15. 10.1002/ehf2.12427 . Wenzl FA, Kraler S, Ambler G, Weston C, Herzog SA, Räber L, et al. Sex-specific evaluation and redevelopment of the grace score in non-st-segment elevation acute coronary syndromes in populations from the uk and switzerland: a multinational analysis with external cohort validation. Lancet (London England). 2022;400(10354):744–56. 10.1016/S0140-6736(22)01483-0 . Kelly CJ, Karthikesalingam A, Suleyman M, Corrado G, King D. Key challenges for delivering clinical impact with artificial intelligence. Bmc Med. 2019;17(1):195. 10.1186/s12916-019-1426-2 . Pollard TJ, Johnson AEW, Raffa JD, Celi LA, Mark RG, Badawi O. The eicu collaborative research database, a freely available multi-center database for critical care research. Sci Data. 2018;5:180178. 10.1038/sdata.2018.178 . Hancock JT, Khoshgoftaar TM. Catboost for big data: an interdisciplinary review. J Big Data. 2020;7(1):94. 10.1186/s40537-020-00369-8 . Safaei N, Safaei B, Seyedekrami S, Talafidaryani M, Masoud A, Wang S, et al. E-catboost: an efficient machine learning framework for predicting icu mortality using the eicu collaborative research database. PLoS ONE. 2022;17(5):e262895. 10.1371/journal.pone.0262895 . Minne L, Abu-Hanna A, de Jonge E. Evaluation of sofa-based models for predicting mortality in the icu: a systematic review. Crit Care (London England). 2008;12(6):R161. 10.1186/cc7160 . Li S, Zhang W, Liang X. Red blood cell distribution width and mortality risk in critically ill cardiovascular patients. Heliyon. 2023;9(11):e22225. 10.1016/j.heliyon.2023.e22225 . Aronson D, Hammerman H, Beyar R, Yalonetsky S, Kapeliovich M, Markiewicz W, et al. Serum blood urea nitrogen and long-term mortality in acute st-elevation myocardial infarction. Int J Cardiol. 2008;127(3):380–85. Jentzer JC, Coons JC, Link CB, Schmidhofer M. Pharmacotherapy update on the use of vasopressors and inotropes in the intensive care unit. J Cardiovasc Pharmacol Ther. 2015;20(3):249–60. 10.1177/1074248414559838 . Testani JM, Hanberg JS, Cheng S, Rao V, Onyebeke C, Laur O, et al. Rapid and highly accurate prediction of poor loop diuretic natriuretic response in patients with heart failure. Circ Heart Fail. 2016;9(1):e2370. 10.1161/CIRCHEARTFAILURE.115.002370 . Li R, Jin X, Ren J, Deng G, Li J, Gao Y, et al. Relationship of admission serum anion gap and prognosis of critically ill patients: a large multicenter cohort study. Dis Markers. 2022;2022:5926049. 10.1155/2022/5926049 . Siontis GCM, Tzoulaki I, Castaldi PJ, Ioannidis JPA. External validation of new risk prediction models is infrequent and reveals worse prognostic discrimination. J Clin Epidemiol. 2015;68(1):25–34. 10.1016/j.jclinepi.2014.09.007 . Saito T, Rehmsmeier M. The precision-recall plot is more informative than the roc plot when evaluating binary classifiers on imbalanced datasets. PLoS ONE. 2015;10(3):e118432. 10.1371/journal.pone.0118432 . Additional Declarations No competing interests reported. Supplementary Files Supplementarymaterial1.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviewers invited by journal 21 Apr, 2026 Editor invited by journal 30 Mar, 2026 Editor assigned by journal 27 Mar, 2026 Submission checks completed at journal 27 Mar, 2026 First submitted to journal 25 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-9218650","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":631268680,"identity":"74e8eff5-aa65-4b07-afc7-4234f6cc601a","order_by":0,"name":"wencai Jiang","email":"","orcid":"","institution":"Suining Central Hospital","correspondingAuthor":false,"prefix":"","firstName":"wencai","middleName":"","lastName":"Jiang","suffix":""},{"id":631268681,"identity":"f47287dc-c825-411c-a6ae-7a448d30ac93","order_by":1,"name":"Longzhou Chen","email":"","orcid":"","institution":"Suining Central Hospital","correspondingAuthor":false,"prefix":"","firstName":"Longzhou","middleName":"","lastName":"Chen","suffix":""},{"id":631268682,"identity":"f87ebd4f-29a3-47ed-95ee-596bffdcfa29","order_by":2,"name":"Yucai Tang","email":"","orcid":"","institution":"Suining Central Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yucai","middleName":"","lastName":"Tang","suffix":""},{"id":631268684,"identity":"bbaa1d62-6a2a-47ab-954d-b0801cc7cce3","order_by":3,"name":"Xuejun Deng","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAx0lEQVRIiWNgGAWjYLACHjDJfIAhgWjVQEKCgYEtgWQtPAbEucme/4zxhzcVdnX80mc+f3i4w46Bv70bv2U8EjlmknPOJEtI9uVuk0g8k8wgcebsBgJaeMyYedsOSBic4d3GkNjGzGAgkUtAC9BhnyFaeB5/SGyrJ0ILQ46BNFQLg0Ri22EitNxIKwP5RXJmD5sZUMtxHoJ+Ye8/vBkUYvz8PMyPP/5sq5bjb+/Fr4WBgQM1OngIKAfb84AIRaNgFIyCUTCiAQDmKkAYYmyWuAAAAABJRU5ErkJggg==","orcid":"","institution":"Suining Central Hospital","correspondingAuthor":true,"prefix":"","firstName":"Xuejun","middleName":"","lastName":"Deng","suffix":""}],"badges":[],"createdAt":"2026-03-25 05:55:08","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9218650/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9218650/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":108492368,"identity":"03b73086-c2cf-42dd-b4db-c48fc1aded07","added_by":"auto","created_at":"2026-05-05 09:57:36","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":49281,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of patient selection. eICU-CRD: eICU Collaborative Research Database; ICU: Intensive care unit; MIMIC-IV: Medical Information Mart for Intensive Care IV.\u003c/p\u003e","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-9218650/v1/ab38fe0f9df6c997ca2753b2.png"},{"id":108389699,"identity":"0f68f782-057f-4810-9127-f5749cced8d6","added_by":"auto","created_at":"2026-05-04 06:50:57","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":39972,"visible":true,"origin":"","legend":"\u003cp\u003eReceiver operating characteristic (ROC) curves of machine learning models for predicting 28-day mortality.\u003c/p\u003e\n\u003cp\u003e(A) ROC curves of ten machine learning models in the training cohort (MIMIC-IV). Among them, CatBoost achieved the highest discriminative performance (AUC = 0.857). (B) ROC curves of the CatBoost model in the internal and external validation cohorts. The model maintained good discrimination in the external validation cohort (eICU-CRD), although a slight decrease in AUC was observed compared to the training cohort.\u003c/p\u003e\n\u003cp\u003eAUC: Area under the curve; eICU-CRD: eICU collaborative research database; ICU: Intensive care unit; MIMIC-IV: Medical information mart for intensive care IV; ROC: Receiver operating characteristic.\u003c/p\u003e","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-9218650/v1/1e8c0fa7d37ab81d126945bd.png"},{"id":108389698,"identity":"e8147497-8442-4604-838d-5b1e4230c2e3","added_by":"auto","created_at":"2026-05-04 06:50:57","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":36678,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP value analysis (color gradient: blue = lower values; red = higher values). A: Top 20 important features. B: The impact of the top 20 features on model output.\u003c/p\u003e\n\u003cp\u003eAG, anion gap; ALT, alanine aminotransferase; BUN, blood urea nitrogen; Cl⁻, chloride; Cr, creatinine; Glu, glucose; Mg²⁺, magnesium; Na⁺, sodium; PaCO₂, partial pressure of carbon dioxide; PaO₂, partial pressure of oxygen; pH, potential of hydrogen; PLT, platelet count; RBC, red blood cell count; RDW, red cell distribution width; WBC, white blood cell count; SOFA, Sequential Organ Failure Assessment; SAPS II, Simplified Acute Physiology Score II; loop diuretic use, use of loop diuretics; inotrope use, use of inotropic agents; vasopressor use, use of vasopressors.\u003c/p\u003e","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-9218650/v1/00d788c2db36ce29468182e8.png"},{"id":108492882,"identity":"2b866259-f654-4b13-bf37-8ada91cd9f6c","added_by":"auto","created_at":"2026-05-05 09:58:53","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":41777,"visible":true,"origin":"","legend":"\u003cp\u003eA web-based calculator for predicting 28-day mortality in patients with acute coronary syndrome complicated by heart failure.\u003c/p\u003e\n\u003cp\u003eAG, anion gap; ALT, alanine aminotransferase; BUN, blood urea nitrogen; Cl⁻, chloride; Cr, creatinine; Glu, glucose; Mg²⁺, magnesium; Na⁺, sodium; PaCO₂, partial pressure of carbon dioxide; PaO₂, partial pressure of oxygen; pH, potential of hydrogen; PLT, platelet count; RBC, red blood cell count; RDW, red cell distribution width; WBC, white blood cell count; SOFA, Sequential Organ Failure Assessment; SAPS II, Simplified Acute Physiology Score II; loop diuretic use, use of loop diuretics; inotrope use, use of inotropic agents; vasopressor use, use of vasopressors.\u003c/p\u003e","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-9218650/v1/a35402a0bf5482d1d16ddd01.png"},{"id":108804501,"identity":"3a645de8-83ad-435c-8886-5a99e89df746","added_by":"auto","created_at":"2026-05-08 15:20:55","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":681973,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9218650/v1/26b14d3b-2fff-479d-bc27-ee6d18f164dc.pdf"},{"id":108389696,"identity":"3d71649b-7d66-4505-823f-f0a3b4ce295c","added_by":"auto","created_at":"2026-05-04 06:50:57","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":771978,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarymaterial1.docx","url":"https://assets-eu.researchsquare.com/files/rs-9218650/v1/9df06032f5d5ed192011973b.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Development and external validation of a machine learning model for predicting 28-day mortality in ICU patients with acute coronary syndrome complicated by heart failure","fulltext":[{"header":"Introduction","content":"\u003cp\u003eAcute coronary syndrome (ACS) remains a major contributor to morbidity and mortality worldwide, particularly among critically ill patients requiring intensive care. According to the Global Burden of Disease (GBD) 2019 study, ischemic heart disease is still the leading cause of death globally, accounting for approximately 9.14\u0026nbsp;million deaths each year(\u003cspan additionalcitationids=\"CR2\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e). Heart failure (HF), a frequent complication of ACS, has been reported in approximately 15% to 30% of hospitalized patients and is associated with a markedly increased risk of short-term mortality(\u003cspan additionalcitationids=\"CR5\" citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e). In the ICU, patients with ACS complicated by HF often present with hemodynamic instability, multi-organ dysfunction, and multiple comorbidities, all of which contribute to poor clinical outcomes(\u003cspan additionalcitationids=\"CR8\" citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe interplay between ACS and HF involves a complex and dynamic pathophysiological process. Myocardial ischemia can impair cardiac function and trigger ventricular remodeling, ultimately leading to the development or worsening of HF(\u003cspan additionalcitationids=\"CR11\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). Conversely, HF further reduces systemic perfusion, promoting renal dysfunction, metabolic abnormalities, and systemic inflammation, thereby creating a vicious cycle that worsens prognosis, especially in critically ill patients(\u003cspan additionalcitationids=\"CR14\" citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eMost existing prognostic models for cardiovascular disease or ICU mortality are developed using conventional statistical approaches and are typically derived from general populations(\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e). As a result, they may not adequately capture nonlinear relationships or complex interactions among variables, limiting their applicability in specific conditions such as ACS complicated by HF. Therefore, more accurate and individualized prediction tools are still needed for this high-risk population(\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn recent years, machine learning (ML) techniques have shown strong potential in analyzing high-dimensional clinical data and identifying complex patterns(\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e). In addition, the use of explainable methods, such as Shapley additive explanations (SHAP), allows for better interpretation of model predictions and may enhance clinical acceptance. In this study, we developed and externally validated an interpretable ML model to predict 28-day mortality in ICU patients with ACS complicated by HF, and further implemented a web-based calculator to facilitate individualized risk assessment in clinical practice(\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e).\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eData sources\u003c/h2\u003e \u003cp\u003eIn this retrospective study, the training cohort was obtained from the Medical Information Mart for Intensive Care IV (MIMIC-IV, version 3.1), a large, publicly accessible database containing detailed clinical data of patients admitted to intensive care units at the Beth Israel Deaconess Medical Center (BIDMC), USA, between 2008 and 2022. The database was developed with approval from the Institutional Review Boards of BIDMC and the Massachusetts Institute of Technology. The author, Longzhou Chen, completed the Collaborative Institutional Training Initiative (CITI) program and obtained certification to access the database (certification number: 74101068).\u003c/p\u003e \u003cp\u003eThe external validation cohort was derived from the eICU Collaborative Research Database (eICU-CRD, version 2.0), which includes deidentified, multicenter clinical data from more than 200,000 ICU admissions across 208 hospitals in the United States between 2014 and 2015. As both databases are publicly available and contain fully anonymized patient information, this study was exempt from the requirement for informed consent and additional ethical approval.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eStudy population\u003c/h3\u003e\n\u003cp\u003eThis retrospective study was based on two large critical care databases. The training cohort was obtained from the MIMIC-IV database (version 3.1), including ICU admissions from 2008 to 2022, and the external validation cohort was derived from the eICU Collaborative Research Database (eICU-CRD, version 2.0), which contains multicenter ICU data collected between 2014 and 2015. Patients diagnosed with both ACS and HF were included in the study. The diagnoses were identified according to standard clinical criteria and corresponding ICD codes. The inclusion criteria were: (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e) age\u0026thinsp;\u0026ge;\u0026thinsp;18 years; (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e) first ICU admission; and (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e) concurrent diagnosis of ACS and HF. The patient selection flowchart is presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Patients with an ICU stay of less than 24 hours were excluded. For the external validation cohort, the variables were selected according to the features identified in the training cohort using Boruta and LASSO. Patients with substantial missing data in these variables were excluded to ensure data completeness. The primary outcome was 28-day all-cause mortality after ICU admission.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eData extraction and feature engineering\u003c/h3\u003e\n\u003cp\u003eData extraction was performed using Navicat Premium (version 16.2.11; PremiumSoft CyberTech Ltd., Singapore) and pgAdmin 4 (version 8.13). Patients meeting the diagnostic criteria for ACS complicated by HF were identified, and the relevant clinical data were retrieved from the databases. Data preprocessing and subsequent analyses were conducted using Python (version 3.11; Python Software Foundation, Wilmington, DE, USA). Clinical variables within the first 24 hours after ICU admission were included. An initial set of 50 candidate variables was collected based on clinical relevance and data availability. Variables with more than 30% missing values were excluded. For the remaining variables, missing data were handled using multiple imputation with predictive mean matching (PMM). Feature selection was performed in several steps. First, the Boruta algorithm was applied to screen variables with potential predictive value. Then, multicollinearity was assessed, and highly correlated variables were removed. Finally, least absolute shrinkage and selection operator (LASSO) regression with cross-validation was used to further reduce the number of variables and select the most informative predictors(Figure \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e-S2). To further reduce redundancy among variables, multicollinearity was assessed and highly correlated variables were removed, resulting in a more stable feature set for model development. After this process, 20 variables were retained and used for model development(Figure S3).\u003c/p\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eStatistical analysis\u003c/h2\u003e \u003cp\u003eAll statistical analyses were performed using Python (version 3.11.9). Continuous variables were assessed for normality using the Shapiro\u0026ndash;Wilk test. Variables following a normal distribution are presented as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation (SD) and were compared using the Student's t-test. Non-normally distributed variables are reported as median (interquartile range [IQR]) and were compared using the Mann\u0026ndash;Whitney U test. Categorical variables are expressed as counts (percentages) and were analyzed using the chi-square test or Fisher's exact test, as appropriate.\u003c/p\u003e \u003cp\u003eThe MIMIC-IV database (version 3.1) was used as the training cohort for model development, while the eICU Collaborative Research Database (eICU-CRD, version 2.0) served as an independent external validation cohort to evaluate model generalizability. Missing data were addressed using multiple imputation with predictive mean matching prior to analysis.\u003c/p\u003e \u003cp\u003eA total of ten machine learning algorithms were developed and compared for predicting 28-day mortality in ICU patients with acute coronary syndrome complicated by heart failure. These included logistic regression, decision tree, random forest, gradient boosting, AdaBoost, extreme gradient boosting (XGBoost), CatBoost, LightGBM, support vector machine (SVM), and k-nearest neighbors (KNN). Model training was conducted exclusively on the training cohort, and no data from the external validation cohort were used during model development to prevent data leakage.\u003c/p\u003e \u003cp\u003eModel performance was evaluated using accuracy, area under the receiver operating characteristic curve (AUC), F1 score, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). AUC was used as the primary measure of discriminative ability. The F1 score was included to reflect the balance between sensitivity and precision. PPV and NPV indicate the proportions of true positive and true negative predictions among all positive and negative predictions, respectively.\u003c/p\u003e \u003cp\u003eTo improve model interpretability, SHapley Additive exPlanations (SHAP) were applied. SHAP values, derived from cooperative game theory, were used to quantify the contribution of each feature to individual predictions. A SHAP summary plot was generated to present global feature importance, and SHAP dependence plots were used to examine relationships between feature values and model outputs. In addition, SHAP force plots and waterfall plots were used to visualize feature contributions at the individual level. Partial dependence plots (PDPs) were further constructed to evaluate the marginal effects of key variables on predicted risk. All statistical tests were two-sided, and a p-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was considered statistically significant.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eBaseline characteristics\u003c/h2\u003e \u003cp\u003eBased on the predefined inclusion and exclusion criteria, a total of 3,410 ICU patients diagnosed with ACS complicated by HF were identified from the MIMIC-IV database (version 3.1) and included as the training cohort. Among them, 2,825 patients survived within 28 days, while 585 patients died. For external validation, 984 eligible patients were extracted from the eICU Collaborative Research Database (eICU-CRD, version 2.0), including 794 survivors and 190 non-survivors at 28 days. The baseline characteristics of patients in both cohorts are presented in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePatient demographics and baseline characteristics.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"9\"\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 \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 \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eVariables\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"4\" nameend=\"c5\" namest=\"c2\"\u003e \u003cp\u003eMIMIC-IV (v3.1)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"4\" nameend=\"c9\" namest=\"c6\"\u003e \u003cp\u003eeICU-CRD (v2.0)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTotal (n\u0026thinsp;=\u0026thinsp;3410)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSurvivors (n\u0026thinsp;=\u0026thinsp;2,825)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNon-survivors (n\u0026thinsp;=\u0026thinsp;585)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ep-value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eTotal (n\u0026thinsp;=\u0026thinsp;984)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eSurvivors (n\u0026thinsp;=\u0026thinsp;794)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eNon-survivors (n\u0026thinsp;=\u0026thinsp;190)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003ep-value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge (years)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e75.00 (66.00\u0026ndash;84.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e75.00 (66.00\u0026ndash;84.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e75.00 (66.00\u0026ndash;84.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.802\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e70.00 (60.00\u0026ndash;79.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e69.00 (59.00\u0026ndash;78.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e73.50 (63.00\u0026ndash;81.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.009\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePco2 (mmHg)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e41.00 (35.00\u0026ndash;47.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e41.00 (36.00\u0026ndash;46.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e41.00 (35.00\u0026ndash;49.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e45.00 (38.80\u0026ndash;55.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e45.00 (39.00\u0026ndash;54.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e46.00 (38.00\u0026ndash;56.65)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.522\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePo2 (mmHg)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e96.00 (50.00\u0026ndash;258.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e104.00 (53.00\u0026ndash;287.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e76.00 (42.00\u0026ndash;135.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e145.00 (100.00\u0026ndash;229.80)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e144.00 (100.00\u0026ndash;220.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e156.90 (93.00\u0026ndash;246.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.793\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePH\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e7.37 (7.31\u0026ndash;7.42)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e7.38 (7.32\u0026ndash;7.43)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e7.33 (7.25\u0026ndash;7.40)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e7.44 (7.39\u0026ndash;7.49)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e7.44 (7.39\u0026ndash;7.49)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e7.43 (7.35\u0026ndash;7.51)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.013\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNa+(mmol/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e138.00 (135.00\u0026ndash;141.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e138.00 (135.00\u0026ndash;141.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e138.00 (134.00\u0026ndash;142.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.476\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e143.00 (140.00\u0026ndash;147.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e143.00 (140.00\u0026ndash;146.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e143.00 (139.00\u0026ndash;148.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.342\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCL-(mmol/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e103.00 (99.00\u0026ndash;107.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e104.00 (100.00\u0026ndash;107.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e101.00 (97.00\u0026ndash;106.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e109.00 (105.00\u0026ndash;113.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e109.00 (105.00\u0026ndash;113.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e110.00 (105.00\u0026ndash;114.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.141\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMg2+(mg/dL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.10 (1.90\u0026ndash;2.40)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2.10 (1.90\u0026ndash;2.40)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2.10 (1.90\u0026ndash;2.40)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e2.30 (2.07\u0026ndash;2.60)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e2.30 (2.01\u0026ndash;2.60)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e2.40 (2.20\u0026ndash;2.80)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.028\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAG (mEq/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e15.00 (12.00\u0026ndash;18.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e14.00 (12.00\u0026ndash;17.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e17.00 (14.00\u0026ndash;21.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e13.00 (10.00\u0026ndash;16.30)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e12.00 (10.00\u0026ndash;15.10)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e16.00 (13.00\u0026ndash;20.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\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\u003eRBC (m/uL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.47 (2.92\u0026ndash;4.06)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.47 (2.92\u0026ndash;4.07)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.48 (2.94\u0026ndash;4.03)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.925\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3.92 (3.49\u0026ndash;4.44)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e3.92 (3.49\u0026ndash;4.44)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e3.88 (3.50\u0026ndash;4.46)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.336\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWBC (K/uL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e11.70 (8.60\u0026ndash;16.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e11.50 (8.40\u0026ndash;15.60)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e13.00 (9.40\u0026ndash;18.20)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e16.00 (11.60\u0026ndash;21.35)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e15.57 (11.30\u0026ndash;20.40)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e18.00 (13.70\u0026ndash;25.10)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\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 (K/uL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e195.00 (145.00\u0026ndash;256.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e196.00 (146.00\u0026ndash;256.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e191.00 (139.00\u0026ndash;259.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.929\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e241.00 (178.00\u0026ndash;326.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e251.50 (186.50\u0026ndash;334.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e213.00 (156.00\u0026ndash;263.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\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\u003eRDW(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e14.60 (13.60\u0026ndash;16.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e14.50 (13.50\u0026ndash;15.80)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e15.40 (14.10\u0026ndash;17.30)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e15.60 (14.50\u0026ndash;17.30)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e15.50 (14.40\u0026ndash;17.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e16.30 (14.70\u0026ndash;18.30)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.002\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGlu (mg/dL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e139.00 (113.00\u0026ndash;190.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e136.00 (112.00\u0026ndash;185.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e160.00 (118.00\u0026ndash;213.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e221.50 (167.00\u0026ndash;295.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e221.00 (165.00\u0026ndash;296.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e224.50 (180.00\u0026ndash;293.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.51\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBUN (mg/dL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e28.00 (18.00\u0026ndash;46.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e26.00 (18.00\u0026ndash;43.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e39.50 (25.00\u0026ndash;62.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e41.00 (25.00\u0026ndash;62.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e38.00 (24.00\u0026ndash;62.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e49.00 (31.00\u0026ndash;62.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.004\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCreatinine (mg/dL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.30 (0.90\u0026ndash;2.10)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.20 (0.90\u0026ndash;2.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.70 (1.20\u0026ndash;2.80)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.72 (1.12\u0026ndash;3.11)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e1.57 (1.08\u0026ndash;2.76)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e2.44 (1.57\u0026ndash;3.72)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eALT (IU/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e32.00 (18.00\u0026ndash;77.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e30.00 (17.00\u0026ndash;66.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e44.00 (21.00\u0026ndash;126.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e48.00 (27.00\u0026ndash;137.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e42.00 (25.00\u0026ndash;100.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e132.00 (38.00\u0026ndash;632.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\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\u003eSOFA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e5.00 (3.00\u0026ndash;8.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e5.00 (3.00\u0026ndash;7.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e7.00 (5.00\u0026ndash;10.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e6.00 (4.00\u0026ndash;9.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e6.00 (4.00\u0026ndash;9.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e9.00 (5.00\u0026ndash;13.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\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\u003eSAPSII\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e39.00 (32.00\u0026ndash;49.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e38.00 (31.00\u0026ndash;46.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e50.00 (40.00\u0026ndash;62.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e42.00 (32.00\u0026ndash;55.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e40.00 (31.00\u0026ndash;51.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e55.50 (42.00\u0026ndash;69.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\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\u003eDiuretic use\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1,308.00 (38.36%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1,225.00 (43.36%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e83.00 (14.19%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e11.00 (1.12%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e10.00 (1.26%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.00 (0.53%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.388\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInotrope use\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e684.00 (20.06%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e473.00 (16.74%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e211.00 (36.07%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e454.00 (46.14%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e380.00 (47.86%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e74.00 (38.95%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.027\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVasopressor use\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2,014.00 (59.06%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1,552.00 (54.94%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e462.00 (78.97%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e260.00 (26.42%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e177.00 (22.29%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e83.00 (43.68%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eData are presented as median (interquartile range, 25th\u0026ndash;75th percentile) or n (%).\u003c/p\u003e \u003cp\u003eAG, anion gap; ALT, alanine aminotransferase; BUN, blood urea nitrogen; Cl⁻, chloride; Cr, creatinine; Glu, glucose; Mg\u0026sup2;⁺, magnesium; Na⁺, sodium; PaCO₂, partial pressure of carbon dioxide; PaO₂, partial pressure of oxygen; pH, potential of hydrogen; PLT, platelet count; RBC, red blood cell count; RDW, red cell distribution width; WBC, white blood cell count; SOFA, Sequential Organ Failure Assessment; SAPS II, Simplified Acute Physiology Score II; loop diuretic use, use of loop diuretics; inotrope use, use of inotropic agents; vasopressor use, use of vasopressors.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eComparison of machine learning models\u003c/h3\u003e\n\u003cp\u003eAmong all the models evaluated, CatBoost showed the best overall performance. It achieved the highest AUC (0.855) and accuracy (0.862), together with a relatively high F1 score (0.438) and PPV (0.733). In addition, it maintained a good balance between sensitivity (0.312) and specificity (0.976), with a high NPV (0.872) (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eGradient Boosting and XGBoost also demonstrated good discrimination, with AUC values of 0.839 and 0.830, respectively, although their overall performance was slightly lower than that of CatBoost. Logistic regression showed high specificity (0.967), but its sensitivity was limited (0.335), suggesting reduced ability to identify high-risk patients. Similar patterns were observed in Random Forest and SVM, which tended to have high specificity but relatively low sensitivity.\u003c/p\u003e \u003cp\u003e Taken together, CatBoost provided a more balanced performance across different evaluation metrics. Therefore, it was selected as the final model for subsequent analyses (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA).\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\u003ePrediction performance of each model in internal validation sets.\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=\"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 \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\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eF1\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\u003ePPV\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eNPV\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLogistic Regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.858\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.824\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.449\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.335\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.967\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.678\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.875\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDecision Tree\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.785\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.618\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.349\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.335\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.878\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.364\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.864\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.831\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.326\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.21\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.983\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.725\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.857\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGradient Boosting\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.839\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.435\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.312\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.974\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.714\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.872\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\u003e0.856\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.821\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.405\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.284\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.975\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.704\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.868\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\u003e0.857\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.437\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.324\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.967\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.671\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.873\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\u003e0.862\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.855\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.438\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.312\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.976\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.733\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.872\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\u003e0.849\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.828\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.412\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.307\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.962\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.628\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.87\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\u003e0.852\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.799\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.335\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.216\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.985\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.745\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\u003eKNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.847\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.775\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.379\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.273\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.966\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.623\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.865\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAUC: Area under the receiver operating characteristic curve; NPV: Negative predictive value; PPV: Positive predictive value.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e(A) ROC curves of ten machine learning models in the training cohort (MIMIC-IV). Among them, CatBoost achieved the highest discriminative performance (AUC\u0026thinsp;=\u0026thinsp;0.857). (B) ROC curves of the CatBoost model in the internal and external validation cohorts. The model maintained good discrimination in the external validation cohort (eICU-CRD), although a slight decrease in AUC was observed compared to the training cohort.\u003c/p\u003e \u003cp\u003eAUC: Area under the curve; eICU-CRD: eICU collaborative research database; ICU: Intensive care unit; MIMIC-IV: Medical information mart for intensive care IV; ROC: Receiver operating characteristic.\u003c/p\u003e \u003cp\u003eFor external validation, the CatBoost model was further evaluated using the eICU-CRD (v2.0) cohort. The receiver operating characteristic curve for the external cohort is presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB. In the external cohort, the model achieved an AUC of 0.676 with an accuracy of 0.811 and an F1-score of 0.279. Sensitivity was relatively low (0.189), whereas specificity remained high (0.960). The PPV and NPV were 0.529 and 0.832, respectively. As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eB, the ROC curve of the external cohort was consistently lower than that of the internal cohort (AUC: 0.803 vs. 0.676), indicating a reduction in discriminative performance across datasets. Despite this decline, the model maintained a relatively high specificity and NPV, suggesting that it remains useful for identifying patients at low risk of 28-day mortality. However, the low sensitivity indicates that a proportion of high-risk patients may not be captured in the external population. Overall, these findings suggest that although the model performance decreased in the external cohort, it retained moderate discrimination and may still provide supportive information for clinical risk stratification.\u003c/p\u003e\n\u003ch3\u003eExplanation of risk factors\u003c/h3\u003e\n\u003cp\u003eTo further interpret the CatBoost model and identify the main contributors, SHAP analysis was performed. Figure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA shows the feature importance ranked by mean absolute SHAP values. Among all variables, diuretic use had the greatest impact on model output, followed by SAPSII score, vasopressor use, and Po2. Other important features included RDW, BUN, chloride (Cl\u0026minus;), and creatinine, suggesting that both clinical interventions and laboratory indicators contributed to mortality prediction. Figure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB presents the distribution of SHAP values for each feature, illustrating how feature values influence the predicted risk. Higher SAPSII scores, RDW, BUN, and the use of vasopressors were generally associated with increased mortality risk. In contrast, higher Po2 levels and chloride tended to be associated with lower predicted risk. The effect of diuretic use appeared complex, with both positive and negative contributions depending on the clinical context.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAG, anion gap; ALT, alanine aminotransferase; BUN, blood urea nitrogen; Cl⁻, chloride; Cr, creatinine; Glu, glucose; Mg\u0026sup2;⁺, magnesium; Na⁺, sodium; PaCO₂, partial pressure of carbon dioxide; PaO₂, partial pressure of oxygen; pH, potential of hydrogen; PLT, platelet count; RBC, red blood cell count; RDW, red cell distribution width; WBC, white blood cell count; SOFA, Sequential Organ Failure Assessment; SAPS II, Simplified Acute Physiology Score II; loop diuretic use, use of loop diuretics; inotrope use, use of inotropic agents; vasopressor use, use of vasopressors.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eInterpretation of individual prediction\u003c/h2\u003e \u003cp\u003eTo illustrate how individual features contribute to model predictions, two representative cases were selected (Figure S). In the SHAP waterfall plots, red bars represent features associated with increased predicted risk, whereas blue bars indicate features that reduce the predicted risk. In Figure S5A, a patient with a low predicted risk (0.01) is shown. Several variables contributed to a reduction in the model output. In particular, the presence of diuretic use and a relatively low SAPSII score had the largest negative impact. Lower BUN levels, higher pH, and lower RDW were also associated with decreased predicted risk. Although vasopressor use and AG contributed in the opposite direction, their effects were modest. Overall, the combined effect of these features resulted in a low predicted probability, consistent with the observed outcome. Figure S5B shows a patient with a higher predicted risk (0.21). In this case, the absence of diuretic use, a higher SAPSII score, and vasopressor use contributed to an increase in the model output. Elevated BUN, lower Po2, and higher WBC were also associated with higher predicted risk. In contrast, lower RDW and higher chloride levels contributed to a reduction in risk. The net effect of these factors led to a higher predicted probability compared with the low-risk case. These examples highlight how different combinations of clinical variables can influence individual predictions, reflecting the model's ability to capture complex patterns in the data.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eImplementation of web-based calculator\u003c/h2\u003e \u003cp\u003eBased on the final CatBoost model, an interactive web-based calculator was developed to facilitate clinical application (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). The interface allows users to input routinely available clinical variables, after which the model provides an estimated probability of 28-day mortality in real time.\u003c/p\u003e \u003cp\u003eAs shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, the tool also incorporates SHAP-based visualization to enhance interpretability. A force plot is used to display the overall direction and magnitude of feature contributions, while a waterfall plot illustrates how individual variables cumulatively influence the final prediction for a given patient. In addition, the predicted probability is presented in a dedicated results panel, allowing for straightforward interpretation. The calculator is publicly available at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://repository-name-mlapp-ujfpaagtkdpqmzgklgjjbf.streamlit.app\u003c/span\u003e\u003cspan address=\"https://repository-name-mlapp-ujfpaagtkdpqmzgklgjjbf.streamlit.app\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e and is also available via QR code, enabling convenient use on mobile devices. This tool may assist clinicians in individualized risk assessment and support clinical decision-making in patients with ACS complicated by heart HF.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAG, anion gap; ALT, alanine aminotransferase; BUN, blood urea nitrogen; Cl⁻, chloride; Cr, creatinine; Glu, glucose; Mg\u0026sup2;⁺, magnesium; Na⁺, sodium; PaCO₂, partial pressure of carbon dioxide; PaO₂, partial pressure of oxygen; pH, potential of hydrogen; PLT, platelet count; RBC, red blood cell count; RDW, red cell distribution width; WBC, white blood cell count; SOFA, Sequential Organ Failure Assessment; SAPS II, Simplified Acute Physiology Score II; loop diuretic use, use of loop diuretics; inotrope use, use of inotropic agents; vasopressor use, use of vasopressors.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn this study, we developed and externally validated a machine learning model based on the MIMIC-IV database to predict 28-day mortality in ICU patients with acute coronary syndrome complicated by heart failure. Among the candidate algorithms, CatBoost showed the most consistent performance. Although the model demonstrated satisfactory discrimination in the internal cohort, a decline in performance was observed in the external validation dataset, suggesting that its generalizability across different clinical settings remains limited. Notably, the model achieved high specificity and negative predictive value, indicating potential utility in identifying patients at relatively low risk. To facilitate clinical implementation, we further developed a web-based calculator integrating SHAP-based visualization. While the application of machine learning in critical care prognostication has increased substantially, external validation is still lacking in many studies, despite its importance in assessing real-world applicability (\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eTraditional ICU prognostic tools, such as APACHE and SAPS scores, are widely used but are generally derived from regression-based approaches that assume linear relationships between predictors and outcomes(\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e). In practice, however, clinical data often exhibit nonlinear patterns and complex interactions. A systematic review including 30 studies reported that machine learning approaches achieved comparable or even improved discrimination compared with conventional scoring systems in predicting ICU mortality(\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e). In addition, widely used models for acute coronary syndrome or heart failure risk assessment, including GRACE, MAGGIC, and the Seattle Heart Failure model, were primarily developed in non-ICU populations, which may limit their applicability in critically ill patients(\u003cspan additionalcitationids=\"CR30\" citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e). Although machine learning methods have been increasingly introduced into this field, most studies rely on single datasets and lack external validation(\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e). By incorporating both MIMIC-IV and eICU-CRD for model development and validation, the present study attempts to improve the robustness of model evaluation(\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e). The relatively favorable performance of CatBoost in this study may be related to its ability to handle categorical variables efficiently and to mitigate overfitting(\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSHAP analysis provided further insight into the model's decision process. SAPS II, as a comprehensive measure of disease severity, contributed most strongly to the prediction, consistent with its established role in ICU prognostication(\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e). In addition, RDW and BUN were identified as important contributors, suggesting that inflammatory status and renal function play key roles in short-term outcomes in this population. Previous studies have shown that elevated RDW is associated with impaired cardiac function and adverse cardiovascular events(\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e), while BUN reflects both renal dysfunction and neurohormonal activation, which may influence circulatory stability(\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e). Among physiological variables, higher PaO₂ was associated with a lower predicted risk, whereas the use of vasopressors, as an indicator of hemodynamic instability, was associated with worse outcomes(\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e). Notably, the contribution of diuretic use appeared bidirectional, which may reflect heterogeneity in clinical indications(\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e). The anion gap also contributed to the prediction, suggesting a potential role of metabolic disturbances in patient outcomes(\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e). Overall, short-term mortality in this population is likely driven by the combined effects of disease severity, inflammatory response, renal dysfunction, and hemodynamic instability.\u003c/p\u003e \u003cp\u003eThe decline in model performance observed in the external validation cohort is likely multifactorial. Differences in patient characteristics, disease severity, and treatment strategies between datasets may result in distributional shifts that affect model predictions. In addition, variations in data collection practices and missing data patterns across institutions may further influence performance. Previous studies have reported that performance degradation across populations is a common phenomenon in predictive modeling, underscoring the importance of external validation in evaluating clinical applicability(\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn addition, the relatively low sensitivity observed in this study suggests that the model may have limited ability to identify high-risk patients. This may be partly attributable to class imbalance, with fewer mortality events in the dataset, leading the model to prioritize reducing false positives and thereby increasing specificity at the expense of sensitivity(\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e). In clinical practice, this trade-off should be interpreted with caution, particularly in settings where early identification of high-risk patients is critical. The high specificity (0.976) and negative predictive value (0.872) indicate that the model may be particularly useful for identifying low-risk patients. In the ICU setting, accurate identification of patients with a relatively favorable prognosis may help optimize resource allocation and improve management efficiency. The web-based calculator developed in this study integrates SHAP visualization, enhancing the transparency of model predictions and facilitating clinician understanding. As machine learning continues to evolve in healthcare, such models are more likely to serve as complementary tools rather than replacements for traditional clinical assessment, supporting decision-making processes.\u003c/p\u003e \u003cp\u003eSeveral limitations should be acknowledged. First, both MIMIC-IV and eICU-CRD are derived from healthcare systems in the United States, and the generalizability of the model to other populations remains uncertain. Second, the retrospective design limited the inclusion of certain clinically relevant variables, such as echocardiographic parameters and NT-proBNP. Third, model performance declined in the external cohort (AUC decreased from 0.855 to 0.676), and sensitivity remained relatively low (0.312 internally and 0.189 externally), indicating suboptimal identification of high-risk patients. Therefore, model predictions should be interpreted cautiously in clinical practice. Future studies should focus on prospective validation in more diverse populations and explore the integration of temporal data to further improve predictive performance.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn this study, we developed and externally validated a CatBoost model for predicting 28-day mortality in ICU patients with acute coronary syndrome complicated by heart failure. The model demonstrated high specificity and negative predictive value, suggesting potential utility in identifying patients at lower risk. A web-based calculator incorporating SHAP visualization was also implemented to support individualized risk assessment with improved interpretability. Further prospective validation in more diverse populations is warranted to better assess the model's generalizability.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAbbreviation\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eFull Form\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eGBD\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eGlobal Burden of Disease\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eeICU-CRD\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eeICU Collaborative Research Database\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMIMIC-IV\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMedical Information Mart for Intensive Care IV\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCITI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eCollaborative Institutional Training Initiative\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eICD\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eInternational Classification of Diseases\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eIQR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eInterquartile Range\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eLASSO\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eLeast Absolute Shrinkage and Selection Operator\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eBoruta\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eBoruta feature selection algorithm\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eML\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMachine Learning\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eBIDMC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eBeth Israel Deaconess Medical Center\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eKNN\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eK-Nearest Neighbors\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eLightGBM\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eLight Gradient Boosting Machine\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eSVM\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eSupport Vector Machine\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCatBoost\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eCategorical Boosting\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eXGBoost\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eExtreme Gradient Boosting\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePMM\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePredictive Mean Matching\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePPV\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePositive Predictive Value\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eNPV\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eNegative Predictive Value\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eROC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eReceiver Operating Characteristic\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAUC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eArea Under the Receiver Operating Characteristic Curve\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eSHAP\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eShapley Additive Explanations\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eACS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eAcute Coronary Syndrome\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eHF\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eHeart Failure\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAG\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eAnion Gap\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eALT\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eAlanine Aminotransferase\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eBUN\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eBlood Urea Nitrogen\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCr\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eCreatinine\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eGlu\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eGlucose\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCl-\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eChloride\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMg2+\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMagnesium\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eNa+\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eSodium\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePaCO2\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePartial Pressure of Carbon Dioxide\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePaO2\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePartial Pressure of Oxygen\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003epH\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePotential of Hydrogen\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePLT\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePlatelet Count\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eRBC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eRed Blood Cell Count\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eWBC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eWhite Blood Cell Count\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eRDW\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eRed Cell Distribution Width\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eSAPS II\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eSimplified Acute Physiology Score I\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eSOFA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eSequential Organ Failure Assessment\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was a retrospective analysis of publicly available critical care databases (MIMIC-IV and eICU-CRD). Both databases received institutional review board approval and contain deidentified patient data. The requirement for informed consent was waived. Therefore, additional ethical approval was not required.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable. This study does not contain any individual person’s data in any form\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\u003eData Availability Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eThe data that support the findings of this study are available from the corresponding author upon reasonable request. The data are derived from the MIMIC-IV\u0026nbsp;\u003c/strong\u003eand eICU-CRD\u003cstrong\u003e\u0026nbsp;database, which is publicly available but requires credentialed access through the official PhysioNet platform.\u003c/strong\u003e\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\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003cbr\u003e\u0026nbsp;This work was supported by the Sichuan Science and Technology Program (Grant No. SCJJ25RKX011, \"Research on the Precision Prevention and Control System for Common High-Incidence Chronic Major Diseases\").\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eWencai Jiang and Longzhou Chen contributed equally to this work. Wencai Jiang: Conceptualization, Data curation, Formal analysis, Methodology, Software, Visualization, Writing – original draft. Longzhou Chen: Conceptualization, Data curation, Formal analysis, Methodology, Software, Validation, Writing – review \u0026amp; editing. Yucai Tang: Investigation, Resources, Validation, Visualization, Writing – review \u0026amp; editing. Xuejun Deng: Investigation, Resources, Writing – review \u0026amp; editing. All authors approved the final manuscript.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors thank the contributors to the MIMIC-IV and eICU-CRD databases for making the data publicly available. We also appreciate the support from the research team at Suining Central Hospital.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eTsao CW, Aday AW, Almarzooq ZI, Anderson CAM, Arora P, Avery CL, et al. Heart disease and stroke statistics-2023 update: a report from the american heart association. Circulation. 2023;147(8):e93\u0026ndash;621. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1161/CIR.0000000000001123\u003c/span\u003e\u003cspan address=\"10.1161/CIR.0000000000001123\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRoth GA, Mensah GA, Johnson CO, Addolorato G, Ammirati E, Baddour LM, et al. Global burden of cardiovascular diseases and risk factors, 1990\u0026ndash;2019: update from the gbd 2019 study. J Am Coll Cardiol. 2020;76(25):2982\u0026ndash;3021. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jacc.2020.11.010\u003c/span\u003e\u003cspan address=\"10.1016/j.jacc.2020.11.010\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGlobal burden of 369 diseases and injuries in 204 countries and territories, 1990\u0026ndash;2019: a systematic analysis for the global burden of disease study 2019. Lancet (London, England). (2020) 396(10258):1204-22. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/S0140-6736(20)30925-9\u003c/span\u003e\u003cspan address=\"10.1016/S0140-6736(20)30925-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYunoki K, Naruko T, Komatsu R, Ehara S, Shirai N, Sugioka K, et al. Enhanced expression of haemoglobin scavenger receptor in accumulated macrophages of culprit lesions in acute coronary syndromes. Eur Heart J. 2009;30(15):1844\u0026ndash;52. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/eurheartj/ehp257\u003c/span\u003e\u003cspan address=\"10.1093/eurheartj/ehp257\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDe Filippo O, D'Ascenzo F, Wańha W, Leonardi S, Raposeiras Roubin S, Fabris E, et al. Incidence and predictors of heart failure after acute coronary syndrome: the coralys registry. Int J Cardiol. 2023;370:35\u0026ndash;42. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ijcard.2022.10.146\u003c/span\u003e\u003cspan address=\"10.1016/j.ijcard.2022.10.146\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSong L, Li Y, Nie S, Feng Z, Liu Y, Ding F, et al. Using machine learning to predict adverse events in acute coronary syndrome: a retrospective study. Clin Cardiol. 2023;46(12):1594\u0026ndash;602. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/clc.24127\u003c/span\u003e\u003cspan address=\"10.1002/clc.24127\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi J, Liu S, Hu Y, Zhu L, Mao Y, Liu J. Predicting mortality in intensive care unit patients with heart failure using an interpretable machine learning model: retrospective cohort study. J Med Internet Res. 2022;24(8):e38082. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.2196/38082\u003c/span\u003e\u003cspan address=\"10.2196/38082\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTu G, Cai Z, Wu L, Yu H, Jiang H, Luo H. Predicting in-hospital mortality in icu patients with coronary heart disease and diabetes mellitus using machine learning models. PLoS ONE. 2025;20(8):e330381. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1371/journal.pone.0330381\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0330381\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee H, Yang H, Ryu HG, Jung C, Cho YJ, Yoon SB, et al. Real-time machine learning model to predict in-hospital cardiac arrest using heart rate variability in icu. Npj Digit Med. 2023;6(1):215. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41746-023-00960-2\u003c/span\u003e\u003cspan address=\"10.1038/s41746-023-00960-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFrangogiannis NG. The inflammatory response in myocardial injury, repair, and remodelling. Nat Rev Cardiol. 2014;11(5):255\u0026ndash;65. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/nrcardio.2014.28\u003c/span\u003e\u003cspan address=\"10.1038/nrcardio.2014.28\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHeusch G. Myocardial ischemia/reperfusion: translational pathophysiology of ischemic heart disease. Med (New York N Y). 2024;5(1):10\u0026ndash;31. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.medj.2023.12.007\u003c/span\u003e\u003cspan address=\"10.1016/j.medj.2023.12.007\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBuja LM. Pathobiology of myocardial ischemia and reperfusion injury: models, modes, molecular mechanisms, modulation, and clinical applications. Cardiol Rev. 2023;31(5):252\u0026ndash;64. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/CRD.0000000000000440\u003c/span\u003e\u003cspan address=\"10.1097/CRD.0000000000000440\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRangaswami J, Bhalla V, Blair JEA, Chang TI, Costa S, Lentine KL, et al. Cardiorenal syndrome: classification, pathophysiology, diagnosis, and treatment strategies: a scientific statement from the american heart association. Circulation. 2019;139(16):e840\u0026ndash;78. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1161/CIR.0000000000000664\u003c/span\u003e\u003cspan address=\"10.1161/CIR.0000000000000664\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDick SA, Epelman S. Chronic heart failure and inflammation: what do we really know? Circ Res. 2016;119(1):159\u0026ndash;76. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1161/CIRCRESAHA.116.308030\u003c/span\u003e\u003cspan address=\"10.1161/CIRCRESAHA.116.308030\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKawakami R, Sakamoto A, Kawai K, Gianatti A, Pellegrini D, Nasr A, et al. Pathological evidence for SARS-cov-2 as a cause of myocarditis: jacc review topic of the week. J Am Coll Cardiol. 2021;77(3):314\u0026ndash;25. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jacc.2020.11.031\u003c/span\u003e\u003cspan address=\"10.1016/j.jacc.2020.11.031\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJohnson AEW, Bulgarelli L, Shen L, Gayles A, Shammout A, Horng S, et al. Mimic-iv, a freely accessible electronic health record dataset. Sci Data. 2023;10(1):1. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41597-022-01899-x\u003c/span\u003e\u003cspan address=\"10.1038/s41597-022-01899-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShillan D, Sterne JAC, Champneys A, Gibbison B. Use of machine learning to analyse routinely collected intensive care unit data: a systematic review. Crit Care (London England). 2019;23(1):284. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s13054-019-2564-9\u003c/span\u003e\u003cspan address=\"10.1186/s13054-019-2564-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGoldstein BA, Navar AM, Pencina MJ, Ioannidis JPA. Opportunities and challenges in developing risk prediction models with electronic health records data: a systematic review. J Am Med Inf Association: Jamia. 2017;24(1):198\u0026ndash;208. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/jamia/ocw042\u003c/span\u003e\u003cspan address=\"10.1093/jamia/ocw042\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBeam AL, Kohane IS. Big data and machine learning in health care. JAMA. 2018;319(13):1317\u0026ndash;18. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1001/jama.2017.18391\u003c/span\u003e\u003cspan address=\"10.1001/jama.2017.18391\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRen H, Sun Y, Xu C, Fang M, Xu Z, Jing F, et al. Predicting acute onset of heart failure complicating acute coronary syndrome: an explainable machine learning approach. Curr Probl Cardiol. 2023;48(2):101480. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cpcardiol.2022.101480\u003c/span\u003e\u003cspan address=\"10.1016/j.cpcardiol.2022.101480\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGui L, Hu Y, Ouyang H, Zhuang H, Peng Y, Yang J, et al. Predicting the rapid progression of coronary artery lesions in patients with acute coronary syndrome based on machine learning. Front Cardiovasc Med. 2025;12:1535406. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fcvm.2025.1535406\u003c/span\u003e\u003cspan address=\"10.3389/fcvm.2025.1535406\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLundberg SM, Erion G, Chen H, DeGrave A, Prutkin JM, Nair B, et al. From local explanations to global understanding with explainable ai for trees. Nat Mach Intell. 2020;2(1):56\u0026ndash;67. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s42256-019-0138-9\u003c/span\u003e\u003cspan address=\"10.1038/s42256-019-0138-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSun H, Kang M, Zhang H, Jia J, Wang Q. Machine learning for predicting mortality in intensive care unit patients: a prognostic performance systematic review and meta-analysis. Nurs Crit Care. 2025;30(6):e70206. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/nicc.70206\u003c/span\u003e\u003cspan address=\"10.1111/nicc.70206\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArshi B, Cowley LE, Rijnhart E, Reeve K, Smits LJ, Wynants L. External validation, impact assessment and clinical utilization of clinical prediction models: a prospective cohort study. J Clin Epidemiol. 2025;186:111902. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jclinepi.2025.111902\u003c/span\u003e\u003cspan address=\"10.1016/j.jclinepi.2025.111902\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGallitto G, Englert R, Kincses B, Kotikalapudi R, Li J, Hoffschlag K, et al. External validation of machine learning models-registered models and adaptive sample splitting. Gigascience. 2025;14. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/gigascience/giaf036\u003c/span\u003e\u003cspan address=\"10.1093/gigascience/giaf036\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKahraman F, Yılmaz AS, Demir M, Beşiroğlu F. Apache ii score predicts in-hospital mortality more accurately than inflammatory indices in patients with acute coronary syndrome. Kardiologiia. 2022;62(9):54\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.18087/cardio.2022.9.n1979\u003c/span\u003e\u003cspan address=\"10.18087/cardio.2022.9.n1979\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLe Gall JR, Lemeshow S, Saulnier F. A new simplified acute physiology score (saps ii) based on a european/north american multicenter study. JAMA. 1993;270(24):2957\u0026ndash;63.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePocock SJ, Ariti CA, McMurray JJV, Maggioni A, K\u0026oslash;ber L, Squire IB, et al. Predicting survival in heart failure: a risk score based on 39 372 patients from 30 studies. Eur Heart J. 2013;34(19):1404\u0026ndash;13. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/eurheartj/ehs337\u003c/span\u003e\u003cspan address=\"10.1093/eurheartj/ehs337\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFox KAA, Dabbous OH, Goldberg RJ, Pieper KS, Eagle KA, Van de Werf F, et al. Prediction of risk of death and myocardial infarction in the six months after presentation with acute coronary syndrome: prospective multinational observational study (grace). Bmj (Clinical Res Ed). 2006;333(7578):1091.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi S, Marcus P, N\u0026uacute;\u0026ntilde;ez J, N\u0026uacute;\u0026ntilde;ez E, Sanchis J, Levy WC. Validity of the seattle heart failure model after heart failure hospitalization. Esc Heart Fail. 2019;6(3):509\u0026ndash;15. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/ehf2.12427\u003c/span\u003e\u003cspan address=\"10.1002/ehf2.12427\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWenzl FA, Kraler S, Ambler G, Weston C, Herzog SA, R\u0026auml;ber L, et al. Sex-specific evaluation and redevelopment of the grace score in non-st-segment elevation acute coronary syndromes in populations from the uk and switzerland: a multinational analysis with external cohort validation. Lancet (London England). 2022;400(10354):744\u0026ndash;56. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/S0140-6736(22)01483-0\u003c/span\u003e\u003cspan address=\"10.1016/S0140-6736(22)01483-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKelly CJ, Karthikesalingam A, Suleyman M, Corrado G, King D. Key challenges for delivering clinical impact with artificial intelligence. Bmc Med. 2019;17(1):195. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12916-019-1426-2\u003c/span\u003e\u003cspan address=\"10.1186/s12916-019-1426-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePollard TJ, Johnson AEW, Raffa JD, Celi LA, Mark RG, Badawi O. The eicu collaborative research database, a freely available multi-center database for critical care research. Sci Data. 2018;5:180178. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/sdata.2018.178\u003c/span\u003e\u003cspan address=\"10.1038/sdata.2018.178\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHancock JT, Khoshgoftaar TM. Catboost for big data: an interdisciplinary review. J Big Data. 2020;7(1):94. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s40537-020-00369-8\u003c/span\u003e\u003cspan address=\"10.1186/s40537-020-00369-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSafaei N, Safaei B, Seyedekrami S, Talafidaryani M, Masoud A, Wang S, et al. E-catboost: an efficient machine learning framework for predicting icu mortality using the eicu collaborative research database. PLoS ONE. 2022;17(5):e262895. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1371/journal.pone.0262895\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0262895\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMinne L, Abu-Hanna A, de Jonge E. Evaluation of sofa-based models for predicting mortality in the icu: a systematic review. Crit Care (London England). 2008;12(6):R161. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/cc7160\u003c/span\u003e\u003cspan address=\"10.1186/cc7160\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi S, Zhang W, Liang X. Red blood cell distribution width and mortality risk in critically ill cardiovascular patients. Heliyon. 2023;9(11):e22225. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.heliyon.2023.e22225\u003c/span\u003e\u003cspan address=\"10.1016/j.heliyon.2023.e22225\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAronson D, Hammerman H, Beyar R, Yalonetsky S, Kapeliovich M, Markiewicz W, et al. Serum blood urea nitrogen and long-term mortality in acute st-elevation myocardial infarction. Int J Cardiol. 2008;127(3):380\u0026ndash;85.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJentzer JC, Coons JC, Link CB, Schmidhofer M. Pharmacotherapy update on the use of vasopressors and inotropes in the intensive care unit. J Cardiovasc Pharmacol Ther. 2015;20(3):249\u0026ndash;60. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1177/1074248414559838\u003c/span\u003e\u003cspan address=\"10.1177/1074248414559838\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTestani JM, Hanberg JS, Cheng S, Rao V, Onyebeke C, Laur O, et al. Rapid and highly accurate prediction of poor loop diuretic natriuretic response in patients with heart failure. Circ Heart Fail. 2016;9(1):e2370. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1161/CIRCHEARTFAILURE.115.002370\u003c/span\u003e\u003cspan address=\"10.1161/CIRCHEARTFAILURE.115.002370\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi R, Jin X, Ren J, Deng G, Li J, Gao Y, et al. Relationship of admission serum anion gap and prognosis of critically ill patients: a large multicenter cohort study. Dis Markers. 2022;2022:5926049. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1155/2022/5926049\u003c/span\u003e\u003cspan address=\"10.1155/2022/5926049\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSiontis GCM, Tzoulaki I, Castaldi PJ, Ioannidis JPA. External validation of new risk prediction models is infrequent and reveals worse prognostic discrimination. J Clin Epidemiol. 2015;68(1):25\u0026ndash;34. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jclinepi.2014.09.007\u003c/span\u003e\u003cspan address=\"10.1016/j.jclinepi.2014.09.007\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSaito T, Rehmsmeier M. The precision-recall plot is more informative than the roc plot when evaluating binary classifiers on imbalanced datasets. PLoS ONE. 2015;10(3):e118432. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1371/journal.pone.0118432\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0118432\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-cardiovascular-disorders","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bcar","sideBox":"Learn more about [BMC Cardiovascular Disorders](http://bmccardiovascdisord.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bcar/default.aspx","title":"BMC Cardiovascular Disorders","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-9218650/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9218650/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eAcute coronary syndrome complicated by heart failure is a common condition in intensive care units (ICUs) and is associated with a poor prognosis and a markedly increased short-term mortality risk. Most existing prediction models are based on traditional regression approaches and lack external validation, and their applicability in critically ill populations remains uncertain. This study aimed to develop and externally validate a machine learning\u0026ndash;based model to predict 28-day mortality in this population.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003ePatients with acute coronary syndrome complicated by heart failure were identified from the MIMIC-IV database (v3.1) as the training cohort (n\u0026thinsp;=\u0026thinsp;3,410) and from the eICU-CRD database (v2.0) as the external validation cohort (n\u0026thinsp;=\u0026thinsp;984). Feature selection was performed using the Boruta algorithm and LASSO regression, resulting in 20 predictors. Ten machine learning algorithms were evaluated, including logistic regression, decision tree, random forest, gradient boosting, AdaBoost, XGBoost, CatBoost, LightGBM, support vector machine, and k-nearest neighbors. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value, and negative predictive value. SHAP analysis was used to interpret the model, and a web-based calculator was developed based on the optimal model.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eAmong the evaluated models, CatBoost demonstrated the best performance in the internal validation cohort, with an AUC of 0.855, accuracy of 0.862, specificity of 0.976, and negative predictive value of 0.872. In the external validation cohort, the model achieved an AUC of 0.676, with specificity of 0.960 and negative predictive value of 0.832, although sensitivity remained relatively low (0.189). SHAP analysis identified SAPS II, red cell distribution width, blood urea nitrogen, arterial partial pressure of oxygen, and vasopressor use as the most influential predictors.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eThe CatBoost model demonstrated high specificity and negative predictive value for predicting 28-day mortality in ICU patients with acute coronary syndrome complicated by heart failure, suggesting potential utility in identifying low-risk patients. A web-based calculator was developed to facilitate individualized and interpretable risk assessment.\u003c/p\u003e","manuscriptTitle":"Development and external validation of a machine learning model for predicting 28-day mortality in ICU patients with acute coronary syndrome complicated by heart failure","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-05-04 06:50:52","doi":"10.21203/rs.3.rs-9218650/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewersInvited","content":"","date":"2026-04-21T13:52:44+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-03-30T07:23:50+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-27T04:05:44+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-27T04:05:21+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Cardiovascular Disorders","date":"2026-03-25T05:47:39+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-cardiovascular-disorders","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bcar","sideBox":"Learn more about [BMC Cardiovascular Disorders](http://bmccardiovascdisord.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bcar/default.aspx","title":"BMC Cardiovascular Disorders","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"0df0d828-92f2-4736-ad78-d9a849289faf","owner":[],"postedDate":"May 4th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-04T06:50:53+00:00","versionOfRecord":[],"versionCreatedAt":"2026-05-04 06:50:52","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9218650","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9218650","identity":"rs-9218650","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}