Development and Validation of a Machine Learning‑Based Predictive Model for Assessing the Risk of Comorbid Depression in Patients With Asthma

BackgroundTo develop and validate a machine learning (ML) model for early neurological deterioration (END) risk prediction in patients with symptomatic intracranial atherosclerotic stenosis (SICAS).MethodsThis retrospective cohort study enrolled 557 patients with SICAS between January 2022 and December 2024. Relevant clinical data were collected. Least Absolute Shrinkage and Selection Operator (LASSO) regression selected predictive features from clinical/imaging variables. Five ML algorithms, including Gaussian Naive Bayes (GNB), Gradient Boosting Decision Trees (GBDT), Light Gradient Boosting Machine (LightGBM), Extreme Gradient Boosting (XGBoost), and Logistic Regression (LR), were trained (70% of the data) and validated (30% of the data) using 10-fold cross-validation. Model performance was assessed using the area under the curve (AUC), calibration, and decision curve analysis (DCA). Shapley additive explanations (SHAP) interpreted the feature contributions.ResultsThe overall incidence rate of END was 18.13%. The XGBoost model outperformed the other models, achieving a validation AUC of 0.874 (95% CI, 0.809–0.939), a sensitivity of 0.749, a specificity of 0.859, and excellent calibration (deviation: 0.116). DCA indicates the clinical utility of the XGBoost model. Key predictors included the NIHSS score (strongest driver), vascular stenosis severity, Triglyceride Glucose (TyG) index, age, initial systolic blood pressure (SBP), and diabetes. SHAP analysis provided interpretability for the machine learning model and revealed essential factors related to the risk of END in SICAS.ConclusionThis study demonstrates the potential of ML in predicting END in SICAS patients. The SHAP method enhances the interpretability of the prediction model, providing a practical and implementable solution for the early identification of high-risk patients.

BackgroundPrediction models with high accuracy rates for nonmetastatic cervical cancer (CC) patients are limited. This study aimed to construct and compare predictive models on the basis of machine learning (ML) algorithms for predicting the 5‐year survival status of CC patients through using the Surveillance, Epidemiology, and End Results public database of the National Cancer Institute.MethodsThe data registered from 2004 to 2016 were extracted and randomly divided into training and validation cohorts (8:2). The least absolute shrinkage and selection operator (LASSO) regression was employed to identify significant factors. Then, four predictive models were constructed, including logistic regression (LR), random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGBoost). The predictive models were evaluated and compared using Receiver‐operating characteristics with areas under the curves (AUCs) and decision curve analysis (DCA), respectively.ResultsA total of 13,802 patients were involved and classified into training (N = 11,041) and validation (N = 2761) cohorts. By using the LASSO regression method, seven factors were identified. In the training cohort, the XGBoost model showed the best performance (AUC = 0.8400) compared to the other three models (all p < 0.05 by Delong's test). In the validation cohort, the XGBoost model also demonstrated a superior prediction ability (AUC = 0.8365) than LR and SVM models (both p < 0.05 by Delong's test), although the difference was not statistically significant between the XGBoost and the RF models (p = 0.4251 by Delong's test). Based on the DCA results, the XGBoost model was also superior, and feature importance analysis indicated that the tumor stage was the most important variable among the seven factors.ConclusionsThe XGBoost model proved to be an effective algorithm with better prediction abilities. This model is proposed to support better decision‐making for nonmetastatic CC patients in the future.

Primary-care-focused interpretable machine learning model for depression screening in geriatrics: A comparative study of multiple algorithms.

Undernutrition among children under the age of five is a major public health concern, especially in developing countries. This study aimed to use machine learning (ML) algorithms to predict undernutrition and identify its associated factors. Secondary data analysis of the 2017 Multiple Indicator Cluster Survey (MICS) was performed using R and Python. The main outcomes of interest were undernutrition (stunting: height-for-age (HAZ) < -2 SD; wasting: weight-for-height (WHZ) < -2 SD; and underweight: weight-for-age (WAZ) < -2 SD). Seven ML algorithms were trained and tested: linear discriminant analysis (LDA), logistic model, support vector machine (SVM), random forest (RF), least absolute shrinkage and selection operator (LASSO), ridge regression, and extreme gradient boosting (XGBoost). The ML models were evaluated using the accuracy, confusion matrix, and area under the curve (AUC) receiver operating characteristics (ROC). In total, 8564 children were included in the final analysis. The average age of the children was 926 days, and the majority were females. The weighted prevalence rates of stunting, wasting, and underweight were 17%, 7%, and 12%, respectively. The accuracies of all the ML models for wasting were (LDA: 84%; Logistic: 95%; SVM: 92%; RF: 94%; LASSO: 96%; Ridge: 84%, XGBoost: 98%), stunting (LDA: 86%; Logistic: 86%; SVM: 98%; RF: 88%; LASSO: 86%; Ridge: 86%, XGBoost: 98%), and for underweight were (LDA: 90%; Logistic: 92%; SVM: 98%; RF: 89%; LASSO: 92%; Ridge: 88%, XGBoost: 98%). The AUC values of the wasting models were (LDA: 99%; Logistic: 100%; SVM: 72%; RF: 94%; LASSO: 99%; Ridge: 59%, XGBoost: 100%), for stunting were (LDA: 89%; Logistic: 90%; SVM: 100%; RF: 92%; LASSO: 90%; Ridge: 89%, XGBoost: 100%), and for underweight were (LDA: 95%; Logistic: 96%; SVM: 100%; RF: 94%; LASSO: 96%; Ridge: 82%, XGBoost: 82%). Age, weight, length/height, sex, region of residence and ethnicity were important predictors of wasting, stunting and underweight. The XGBoost model was the best model for predicting wasting, stunting, and underweight. The findings showed that different ML algorithms could be useful for predicting undernutrition and identifying important predictors for targeted interventions among children under five years in Ghana.

Background: Diet plays an important role in preventing and managing the progression from prediabetes to type 2 diabetes mellitus (T2DM). This study aims to develop prediction models incorporating specific dietary indicators and explore the performance in T2DM patients and non-T2DM patients. Methods: This retrospective study was conducted on 2215 patients from the Henan Rural Cohort. The key variables were selected using univariate analysis and the least absolute shrinkage and selection operator (LASSO). Multiple predictive models were constructed separately based on dietary and clinical factors. The performance of different models was compared and the impact of integrating dietary factors on prediction accuracy was evaluated. Receiver operating characteristic (ROC) curve, calibration curve, and decision curve analysis (DCA) were used to evaluate the predictive performance. Meanwhile, group and spatial validation sets were used to further assess the models. SHapley Additive exPlanations (SHAP) analysis was applied to identify key factors influencing the progression of T2DM. Results: Nine dietary indicators were quantitatively collected through standardized questionnaires to construct dietary models. The extreme gradient boosting (XGBoost) model outperformed the other three models in T2DM prediction. The area under the curve (AUC) and F1 score of the dietary model in the validation cohort were 0.929 [95% confidence interval (CI) 0.916-0.942] and 0.865 (95%CI 0.845-0.884), respectively. Both were higher than the traditional model (AUC and F1 score were 0.854 and 0.779, respectively, p < 0.001). SHAP analysis showed that fasting plasma glucose, eggs, whole grains, income level, red meat, nuts, high-density lipoprotein cholesterol, and age were key predictors of the progression. Additionally, the calibration curves displayed a favorable agreement between the dietary model and actual observations. DCA revealed that employing the XGBoost model to predict the risk of T2DM occurrence would be advantageous if the threshold were beyond 9%. Conclusions: The XGBoost model constructed by dietary indicators has shown good performance in predicting T2DM. Emphasizing the role of diet is crucial in personalized patient care and management.

To develop a machine learning (ML)-based admission screening model for hospital-acquired (HA) influenza using routinely available data to support early clinical intervention. The study focused on hospitalized patients from January 2021 to May 2024. The case group consisted of patients with HA influenza, while the control group comprised non-HA influenza patients admitted to the same ward in the HA influenza unit within 2 weeks. The 953 subjects were divided into the training set and the validation set in a 7:3 ratio. Feature screening was performed using least absolute shrinkage and selection operator (LASSO) and the Boruta algorithm. Subsequently eight ML algorithms were applied to analyze and identify the optimal model using a 5-fold cross-validation methodology. And the area under the curve (AUC), area under the precision-recall curve (AP), F1 score, calibration curve and decision curve analysis (DCA) were applied to comprehensively assess the predictive effectiveness of the selected models. Feature factors were selected and feature importance's were assessed using SHapley's additive interpretation (SHAP). Furthermore, an interactive web-based platform was additionally developed to visualize and demonstrate the predictive model. Age, pneumonia on admission, Chronic renal failure, Malignant tumor, hypoproteinemia, glucocorticoid use, admission to ICU, lymphopenia, BMI were identified as key variables. For the eight ML algorithms, ROC values ranging from 0.548 to 0.812 were observed in the validation set. A comprehensive analysis showed that the XGBoost model predicted the highest accuracy (AUC: 0.812) with an F1 score of 0.590 and the highest A p value (0.655). Evaluating the optimal model, the AUC values were 0.995, 0.826, and 0.781 for the training, validation and test sets. The XGBoost model showed strong robust. SHapley's additive interpretation (SHAP) was utilized to analyze the contribution of explanatory variables to the model and their correlation with HA influenza. In addition, we developed a practical online prediction tool to calculate the risk of HA influenza occurrence. Based on the routine data, the XGBoost model demonstrated excellent calibration among all ML algorithms and accurately predicted the risk of HA influenza, thereby serving as an effective tool for early screening of HA influenza.

BackgroundWhile environmental heavy metal exposure has been linked to various metabolic disorders, its association with overactive bladder (OAB) remains poorly characterized. Emerging evidence suggests body mass index (BMI) may mediate heavy metal-induced metabolic dysregulation, though underlying pathways remain unclear. This study investigates the interplay between heavy metal exposure, BMI, and OAB risk via explainable machine learning (ML) and mediation analysis.MethodsDrawing on data from the National Health and Nutrition Examination Survey (NHANES) [2005–2010], we identified OAB-associated heavy metals via least absolute shrinkage and selection operator (LASSO) regression and the Boruta algorithm, then developed ten ML models. The optimal model, Extreme Gradient Boosting (XGBoost), was selected based on performance metrics and interpreted via Permutation Feature Importance (PFI), Shapley Additive Explanations (SHAP), and Partial Dependence Plots (PDP). Dose-response relationships, mixture effects, and BMI-mediated pathways were validated through logistic regression (LR), restricted cubic splines (RCS), Bayesian kernel machine regression (BKMR), and mediation analysis.ResultsAmong 3,201 eligible participants, blood lead, blood iron, urinary barium, urinary cadmium, urinary thallium, and urinary mercury were identified as OAB-associated metals. The XGBoost model achieved superior predictive performance [area under the curve (AUC): 0.736]. PFI highlighted hypertension, urinary cadmium, and age as key OAB determinants, while SHAP emphasized urinary cadmium and blood iron as primary predictors. PDP revealed a positive cadmium-OAB association and an inverse iron-OAB relationship. LR confirmed blood iron [odds ratio (OR) =0.72, 95% confidence interval (CI): 0.57–0.90] and urinary cadmium (OR =1.23, 95% CI: 1.06–1.42) as independent risk factors. RCS demonstrated linear trends for cadmium/iron and nonlinear trends for lead. BKMR analysis confirmed a positive overall mixture effect (conditional posterior inclusion probabilities =0.9860), with urinary cadmium showing the strongest exposure-response relationship. Mediation analysis indicated BMI mediated 14.80% of iron’s protective effect and partially counteracted cadmium/lead risks (mediation proportions: −17.33%).ConclusionsUrinary cadmium, blood lead, and iron emerge as critical OAB risk modulators, with BMI serving as a partial mediator. Integrating explainable ML with conventional epidemiology elucidates environmental-metabolic interactions in OAB pathogenesis, underscoring the need for heavy metal screening and BMI management in high-risk populations.

Early prediction of heart failure (HF) after acute myocardial infarction (AMI) is essential for personalized treatment. We aimed to use interpretable machine learning (ML) methods to develop a risk prediction model for HF in AMI patients. We retrospectively included patients initially with AMI who received percutaneous coronary intervention (PCI) in our hospital from November 2016 to February 2020. The primary endpoint was the occurrence of HF within 3 years after operation. For developing a predictive model for HF risk in AMI patients, the least absolute shrinkage and selection operator (LASSO) Regression was used to feature selection, and four ML algorithms including Random Forest (RF), Extreme Gradient Boost (XGBoost), Support Vector Machine (SVM), and Logistic Regression (LR) were employed to develop the model on the training set. The performance evaluation of the prediction model was carried out on the training set and the testing set, utilizing metrics including AUC (Area under the receiver operating characteristic curve), calibration plot, and decision curve analysis (DCA). In addition, we used the Shapley Additive Explanations (SHAP) value to determine the importance of the selected features and interpret the optimal model. A total of 1220 AMI patients were included and 244 (20%) patients developed HF during follow-up. Among the four evaluated ML models, the XGBoost model exhibited exceptional accuracy, with an AUC value of 0.922. The SHAP method showed that left ventricular ejection fraction (LVEF), left ventricular end-systolic diameter (LVDs) and lactate dehydrogenase (LDH) were identified as the three most important characteristics to predict HF risk in AMI patients. Individual risk assessment was performed using SHAP plots and waterfall plot analysis. Our research demonstrates the potential of ML methods in the early prediction of HF risk in AMI patients. Furthermore, it enhances the interpretability of the XGBoost model through SHAP analysis to guide clinical decision-making.

Background and objectiveSpread through air spaces (STAS) is an important factor in determining the aggressiveness and recurrence risk of lung cancer, especially in early-stage adenocarcinoma. Preoperative identification of STAS is crucial for optimizing surgical strategies. This study aimed to develop and validate machine learning models to predict the presence of STAS using preoperative clinical, radiological, and pathological data in lung cancer patients.Patients and methodsA retrospective analysis was conducted on 1,290 lung cancer patients from two hospitals: Qilu Hospital of Shandong University and Qianfoshan Hospital. Data from 1,174 patients from Qilu Hospital were used for model training and internal validation, while 116 patients from Qianfoshan Hospital were used for external validation. Thirteen key variables, identified using least absolute shrinkage and selection operator (LASSO) regression, were included in the construction of eight machine learning models: decision tree (DT), random forest (RF), regularized support vector machine (RSVM), logistic regression (LR), extreme gradient boosting (XGBoost), multilayer perceptron (MLP), light gradient boosting machine (LightGBM), and K-nearest neighbors (KNN). Model performance was evaluated using receiver operating characteristic (ROC) curves, area under the curve (AUC), calibration curves, decision curve analysis (DCA), and SHapley additive explanations (SHAP) plots.ResultsThe XGBoost model achieved the best performance with an AUC of 0.931 (95% CI: 0.897–0.964) in the internal validation cohort and 0.904 (95% CI: 0.835–0.973) in the external validation cohort, outperforming other models. DCA demonstrated the clinical utility of XGBoost, LightGBM, and RF models, which provided superior net benefit across various threshold probabilities. SHAP analysis revealed that the most influential factors in predicting STAS were carcinoembryonic antigen (CEA), forced expiratory volume in one second (FEV1), consolidation-to-tumor ratio (CTR), maximal voluntary ventilation (MVV), and CT value.ConclusionThe XGBoost model demonstrated robust predictive performance for preoperative identification of STAS in lung cancer patients, showing high generalizability in external validation. These findings suggest that machine learning-based predictions could guide clinical decision-making and improve surgical outcomes by identifying high-risk patients for more aggressive treatment strategies.

ObjectiveTo investigate the determinants affecting live birth outcomes in fresh embryo transfer among polycystic ovary syndrome (PCOS) patients using various machine learning (ML) algorithms and to construct predictive models, offering novel insights for enhancing live birth rates in this specific group.MethodsA sum of 1,062 fresh embryo transfer cycles involving PCOS patients were analyzed, with 466 resulting in live births. The dataset was split randomly into training and testing subsets at a 7:3 ratio. Least absolute shrinkage and selection operator and recursive feature elimination methods were utilized for feature selection within the training data. A grid search strategy identified the optimal parameters for seven ML models: decision tree (DT), K-nearest neighbors (KNN), light gradient boosting machine (LightGBM), naive Bayes model(NBM), random forest (RF), support vector machine (SVM) and extreme gradient boosting (XGBoost). The evaluation of model effectiveness incorporated diverse metrics, encompassing area under the curve (AUC), accuracy, positive predictive value, negative predictive value, F1 score, and Brier score. Calibration curves and decision curve analysis were employed to ascertain the optimal model. Furthermore, Shapley additive explanations were applied to elucidate the importance of predictor variables in the top-performing model.ResultsThe AUC values of DT, KNN, LightGBM, NBM, RF, SVM and XGBoost models in the training set were 0.813, 1.000, 0.724, 0.791, 1.000, 0.819 and 0.853, respectively. Corresponding values in the testing set were 0.773, 0.719, 0.705, 0.764, 0.794, 0.806 and 0.822. XGBoost emerged as the most effective ML model. SHAP analysis revealed that variables encompassing embryo transfer count, embryo type, maternal age, infertility duration, body mass index, serum testosterone (T) levels, and progesterone (P) levels on the day of human chorionic gonadotropin administration were pivotal predictors of live birth outcomes in individuals with PCOS receiving fresh embryo transfer.ConclusionThis study developed a live birth prediction model tailored for PCOS fresh embryo transfer cycles, leveraging ML algorithms to compare the efficacy of multiple models. The XGBoost model demonstrated superior predictive capacity, enabling prompt and precise identification of critical risk factors influencing live birth outcomes in PCOS patients. These findings offer actionable insights for clinical intervention, guiding strategies to improve pregnancy outcomes in this population.Clinical trial numberNot applicable.

ObjectivesTo develop and validate a CT-based radiomic-clinical-dosimetric model to assess the treatment response of lung metastasis following stereotactic body radiation therapy (SBRT).MethodsEighty lung metastases treated with SBRT curative intent in a single institution were analysed. The treatment responses of lung lesions were categorized as a complete responding (CR) group vs a non-complete responding (NCR) group according to Response Evaluation Criteria in Solid Tumors (RECIST) criteria. For each lesion, 107 features were extracted from the CT planning images. The least absolute shrinkage and selection operator (LASSO) was used for features selection. An eXtreme Gradient Boosting (XGBoost) model was trained and validated. Shapley additive explanations (SHAP) analysis was used to provide insights into the impact of each variable on the model’s predictions.ResultsEight radiomic features, 1 dosimetric variable, and no clinical variables were identified by LASSO and used to build the XGBoost model. The model yielded areas under the curve (AUCs) of 0.897 (95% CI 0.860-0.935) and 0.864 (95% CI 0.803-0.924) in the training cohort and validation cohort, respectively. Skewness, surface-to-volume ratio, sphericity, and biological equivalent dose (BED10) were the most significant variables in predicting CR. The SHAP plots illustrated the feature’s global and local impact to the model, explaining the model output in a clinician-friendly way.ConclusionThe integration of the XGBoost model with the SHAP strategy was able to assess lung lesions CR following SBRT, with the potential to assist clinicians in directing personalized SBRT strategies in an understandable manner.Advances in knowledgeThe explainable radiomics model we propose can better predict the treatment response of lung metastasis after SBRT and provide further guidance for clinical practice.

Background Lung metastasis greatly affects medical therapeutic strategies in osteosarcoma. This study aimed to develop and validate a clinical prediction model to predict the risk of lung metastasis among osteosarcoma patients based on machine learning (ML) algorithms. Methods We retrospectively collected osteosarcoma patients from the Surveillance Epidemiology and End Results (SEER) database and from four hospitals in China. Six ML algorithms, including logistic regression (LR), gradient boosting machine (GBM), extreme gradient boosting (XGBoost), random forest (RF), decision tree (DT), and multilayer perceptron (MLP), were applied to build predictive models for predicting lung metastasis using patient's demographics, clinical characteristics, and therapeutic variables from the SEER database. The model was internally validated using 10-fold cross-validation to calculate the mean area under the curve (AUC) and the model was externally validated using the Chinese multicenter osteosarcoma data. Relative importance ranking of predictors was plotted to understand the importance of each predictor in different ML algorithms. The correlation heat map of predictors was plotted to understand the correlation of each predictor, selecting the 10-fold cross-validation with the highest AUC value in the external validation ROC curve to build a web calculator. Results Of all enrolled patients from the SEER database, 17.73% (194/1094) developed lung metastasis. The multiple logistic regression analysis showed that sex, N stage, T stage, surgery, and bone metastasis were all independent risk factors for lung metastasis. In predicting lung metastasis, the mean AUCs of the six ML algorithms ranged from 0.711 to 0.738 in internal validation and 0.697 to 0.729 in external validation. Among the six ML algorithms, the extreme gradient boosting (XGBoost) model had the highest AUC value with an average internal AUC of 0.738 and an external AUC of 0.729. The best performing ML algorithm model was used to build a web calculator to facilitate clinicians to calculate the risk of lung metastasis for each patient. Conclusions The XGBoost model may have the best prediction effect and the online calculator based on this model can help doctors to determine the lung metastasis risk of osteosarcoma patients and help to make individualized medical strategies.

This study aims to develop and validate a machine learning (ML) predictive model for assessing mortality in patients with malignant tumors and hyperkalemia (MTH). We extracted data on patients with MTH from the Medical Information Mart for Intensive Care-IV, version 2.2 (MIMIC-IV v2.2) database. The dataset was split into a training set (75%) and a validation set (25%). We used the Least Absolute Shrinkage and Selection Operator (LASSO) regression to identify potential predictors, which included clinical laboratory indicators and vital signs. Pearson correlation analysis tested the correlation between predictors. In-hospital death was the prediction target. The Area Under the Curve (AUC) and accuracy of the training and validation sets of 7 ML algorithms were compared, and the optimal 1 was selected to develop the model. The calibration curve was used to evaluate the prediction accuracy of the model further. SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) enhanced model interpretability. 496 patients with MTH in the Intensive Care Unit (ICU) were included. After screening, 17 clinical features were included in the construction of the ML model, and the Pearson correlation coefficient was <0.8, indicating that the correlation between the clinical features was small. eXtreme Gradient Boosting (XGBoost) outperformed other algorithms, achieving perfect scores in the training set (accuracy: 1.000, AUC: 1.000) and high scores in the validation set (accuracy: 0.734, AUC: 0.733). The calibration curves indicated good predictive calibration of the model. SHAP analysis identified the top 8 predictive factors: urine output, mean heart rate, maximum urea nitrogen, minimum oxygen saturation, minimum mean blood pressure, maximum total bilirubin, mean respiratory rate, and minimum pH. In addition, SHAP and LIME performed in-depth individual case analyses. This study demonstrates the effectiveness of ML methods in predicting mortality risk in ICU patients with MTH. It highlights the importance of predictors like urine output and mean heart rate. SHAP and LIME significantly enhanced the model's interpretability.

Inguinal hernia represents a clinically significant yet underreported complication of robot-assisted radical prostatectomy (RARP) for localized prostate cancer, with a notably high incidence within the first postoperative year. Despite its adverse impact on quality of life and potential for severe sequelae, predictive tools for this outcome remain limited. To develop and validate the first machine learning (ML)-based clinical prediction model for inguinal hernia within 1 year after RARP, leveraging explainable artificial intelligence (AI) techniques for clinical interpretability. This retrospective study analyzed localized prostate cancer patients who underwent RARP between June 1, 2021 and May 1, 2023 at our center. Least absolute shrinkage and selection operator (LASSO) regression identified five key predictors from multiple clinical parameters. Five ML algorithms were developed and evaluated on a 70:30 training–test split. Model performance was assessed via area under the curve (AUC), accuracy, specificity, and decision curve analysis (DCA). SHapley Additive exPlanations (SHAP) methodology provided interpretable feature attribution. The final analysis included 652 eligible patients. Extreme gradient boosting (XGBoost) demonstrated superior discriminative ability, with an AUC of 0.833 (95% CI: 0.770–0.895) in the validation set and 0.791 (95% CI: 0.734–0.848) in the test set. SHAP analysis identified five critical predictors ranked by impact: age, body mass index (BMI), preoperative albumin level, T stage, and history of abdominal surgery. This study established the first ML-driven predictive model for post-RARP inguinal hernia, with XGBoost demonstrating optimal performance. High-risk patients identified by the model warrant personalized proactive interventions.

Hemorrhage is a prevalent and critical condition in the intensive care unit (ICU), characterized by high incidence, elevated mortality rates, and substantial therapeutic challenges. Accurate prediction of mortality in patients with hemorrhage is essential for developing personalized prevention and treatment strategies. Nevertheless, the implementation of effective predictive models in clinical practice remains limited, primarily due to the lack of robust and interpretable tools. This study aimed to develop an interpretable model for predicting mortality risk in critically ill patients with hemorrhage admitted to ICUs. The SHapley Additive exPlanations (SHAP) method was applied to interpret the eXtreme Gradient Boosting (XGBoost)model, identifying key prognostic factors in this population. In this retrospective cohort study, we derived data from the eICU Collaborative Research Database (eICU-CRD) to develop and evaluate a predictive model. ​Clinical data from the first 24h of ICU admission were extracted, and the dataset was randomly split into training (80%) and validation (20%) sets. Model performance was compared​ to four other machine learning algorithms using the area under the curve (AUC). ​SHAP was utilized to interpret the XGBoost model. External validation was subsequently performed using data from the ​Chinese REFRAIN cohort, which focuses on hemorrhage and coagulopathy in critically ill patients.​​. The study protocol was retrospectively registered in the Chinese Clinical Trial Registry (ChiCTR) on December 17, 2024 (Registration number ChiCTR2400094140). A total of 10,306 eligible patients with hemorrhage were included. The observed in-hospital mortality rate was 11.5%.Among the five models compared, XGBoost demonstrated the highest predictive performance (AUC = 0.81), whereas logistic regression (LR) showed the lowest generalizability(AUC = 0.726). Decision curve analysis revealed that the XGBoost model provided a greater net benefit than other models at threshold probabilities of 10-30%. SHAP analysis identified the top 15 predictors of mortality, with bilirubin level ranked as the most influential variable.​​ External validation using the REFRAIN cohort confirmed the robustness of model(AUC = 0.776). The interpretable predictive model improves mortality risk stratification in ICU patients with hemorrhage, supporting clinicians in optimizing treatment plans and resource allocation. Enhanced model transparency through SHAP explanations may facilitate clinical adoption by improving trust in model reliability.