Development and Validation of a Machine Learning-Based Risk Prediction Model for 60-Day Survival Status in Patients with Acute Pancreatitis Complicated with Acute Kidney Injury

Development and Validation of a Machine Learning-Based Risk Prediction Model for 60-Day Survival Status in Patients with Acute Pancreatitis Complicated with Acute Kidney Injury | 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 Validation of a Machine Learning-Based Risk Prediction Model for 60-Day Survival Status in Patients with Acute Pancreatitis Complicated with Acute Kidney Injury Wei Wang, Tongping Shen This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8896948/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Acute kidney injury (AKI) is a severe complication of severe acute pancreatitis (SAP) with an extremely poor prognosis. Our study aimed to develop and validate an interpretable machine learning (ML) model to predict 60-day survival in patients with acute pancreatitis (AP) complicated by AKI, and to identify key prognostic factors to support clinical decision-making. Methods This was a retrospective cohort study, with data extracted from the MIMIC-IV v3.1 database (released in October 2024). The inclusion criteria were as follows: patients aged 18 years or older with acute pancreatitis complicated with acute kidney injury confirmed by ICD diagnostic codes, a clinical data completeness rate of ≥ 80%, and a follow-up duration of ≥ 60 days or a definitive in-hospital death record within 60 days. Propensity score matching (PSM) was applied to address the class imbalance between the survival and death groups. The optimal features were screened from 31 candidate variables using three feature selection algorithms: the Boruta algorithm, random forest (RF), and recursive feature elimination with cross-validation (RFECV). A total of 22 machine learning models were constructed, and their predictive performance was evaluated based on the following metrics: area under the receiver operating characteristic curve (AUC), accuracy, Kappa coefficient, sensitivity, specificity, and Brier score. The Shapley Additive exPlanations (SHAP) method was adopted to interpret the optimal model, which elucidated feature importance, nonlinear relationships, and inter-feature interactions. Ultimately, an interactive web application was developed to facilitate the clinical application and popularization of the model. Results A total of 11,699 eligible patients were included (5,849 in the survival group and 5,850 in the death group after propensity score matching (PSM). Fifteen core features were screened for model construction, including Sequential Organ Failure Assessment (SOFA) score, liver function, cardiovascular function, renal function, length of hospital stay (LoS), admission age, and comorbidity of intracerebral hemorrhage (co_ICH). The random forest (RF) model exhibited the best performance, with an area under the receiver operating characteristic curve (AUC) of 0.9731, accuracy of 0.9068, sensitivity of 0.8755, specificity of 0.9392, and Brier score of 0.0722. SHAP analysis revealed that admission age, co_ICH, liver function, and cardiovascular function were the most important predictors of 60-day mortality. Nonlinear relationships were observed between key features (e.g., activated partial thromboplastin time [PTT], red cell distribution width [RDW], and length of hospital stay) and survival outcomes, along with threshold effects and U-shaped associations. An interactive web application ( https://medicalpredictor.shinyapps.io/Online_Prediction_of_60-Day_Survival/ ) has been successfully deployed for individualized risk assessment. Conclusions The random forest model developed in this study exhibits excellent performance and interpretability in predicting 60-day survival in patients with acute pancreatitis complicated by acute kidney injury. Key prognostic factors identified via SHAP analysis provide valuable clinical references, while the web application enhances the model's practical utility. This tool can assist clinicians in conducting early risk stratification and formulating personalized intervention strategies, thereby improving patient outcomes. Acute Kidney Injury Prognosis Machine Learning 60-Day Survival Prediction Clinical Prognostic Factors Shiny Application Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 Figure 13 Introduction AP is an inflammatory disease of the pancreas with a spectrum of severity ranging from mild pancreatitis to SAP [1]. It is caused by multiple factors, including gallstones, alcohol consumption, hypertriglyceridemia, and medications [2]. AP presents with clinical manifestations varying from mild abdominal pain to multiple organ dysfunction syndrome [3]. In 2021, an estimated 5.9 million people worldwide were living with pancreatitis, 2.7 million new cases were diagnosed, and 122,000 deaths were attributed to the disease [4]. Acute kidney injury (AKI) is one of the most common and severe complications of severe acute pancreatitis (SAP), with a reported incidence ranging from 14% to 42% [5]. Studies have shown that pancreatitis complicated with AKI significantly increases mortality [6,7,8]. The pathogenesis of AKI in SAP is multifactorial, involving hemodynamic changes, inflammatory mediators, oxidative stress, and direct nephrotoxicity [9]. Therefore, early identification and management of AKI in SAP patients are crucial for improving prognosis and reducing mortality [10]. As an important branch of artificial intelligence, machine learning (ML) can uncover complex nonlinear correlations between features and disease outcomes. It has been widely applied in disease diagnosis, complication monitoring, and prognosis prediction, providing auxiliary support for clinical decision-making [11,12,13,14,15,16]. Notably, numerous studies have utilized ML techniques to construct prediction tools for AP, with a primary focus on predicting disease severity, complications, and mortality [17,18,19,20,21]. However, most existing studies are limited to retrospective, single-center designs, with drawbacks such as small sample sizes and limited model interpretability. To address these gaps, this study aims to develop and validate an interpretable machine learning model to predict 60-day survival in patients with AP complicated by AKI. The model employs the Shapley Additive exPlanations (SHAP) method to elucidate the impact of key features and their potential interactions. Additionally, a Shiny-based interactive web application has been developed to improve the model’s clinical applicability. Methods Study Design and Study Population This study used data from the MIMIC-IV v3.1 database, released in October 2024 [22]. The author Tongping Shen completed the training required by the National Institutes of Health (NIH) and obtained database access permission (Record ID: 14348115). Since the data had been anonymized, written informed consent was waived, and the study adhered to the observational research norms outlined in the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines. Inclusion criteria were as follows: (1) Diagnosis of acute pancreatitis confirmed by ICD diagnostic codes; (2) Concurrent fulfillment of clinical diagnostic criteria for acute kidney injury (AKI); (3) Age ≥ 18 years; (4) Completeness rate of clinical data ≥ 80%, with no missing key variables (e.g., gender, age, underlying diseases, severity grading of pancreatitis, AKI staging, core laboratory indicators, treatment measures, etc.); (5) Follow-up duration ≥ 60 days or a clearly documented death outcome within 60 days. Exclusion criteria were as follows: (1) Acute exacerbation of chronic pancreatitis or other types of pancreatitis; (2) AKI complicated by end-stage chronic kidney disease (ESKD); (3) Cases with hospital stay < 24 hours; (4) Duplicate hospitalization records (for patients admitted multiple times due to acute pancreatitis complicated with AKI, only data from the first admission were included); (5) Data with obvious logical errors or outliers (e.g., laboratory indicators outside the clinically reasonable range without supporting original records). After screening against the inclusion and exclusion criteria, 10,218 eligible study participants were identified. Each participant was uniquely identified by subject_id to ensure sample independence. Clinical data for all included patients were derived from routine clinical diagnosis and treatment records, which were standardized and de-identified in the database in compliance with privacy protection regulations. Data Preprocessing and Feature Selection This study initially retrieved clinical data for 10,218 patients with acute pancreatitis complicated with acute kidney injury (AKI). After excluding 3,345 records with missing key variables, 6,873 complete datasets were obtained, comprising 5,898 cases in the 60-day survival group and 975 in the death group, indicating significant class imbalance. The propensity score matching (PSM) algorithm was applied to mitigate this imbalance: baseline covariates were used to calculate propensity scores, followed by 1:1 nearest-neighbor matching. A balanced dataset was ultimately generated, comprising 5,849 cases in the survival group and 5,850 cases in the death group. Meanwhile, a series of quality control procedures was performed, including outlier elimination and standardized coding of categorical variables. These preprocessing steps established a high-quality data foundation for subsequent machine learning model construction, and the results of data preprocessing are presented in Fig. 1 . We adopted three feature selection algorithms, namely the Boruta algorithm, Random Forest (RF), and cross-validation-based Recursive Feature Elimination (RFE). Multi-algorithm cross-validation was employed to improve the reliability of the selected features, and the Venn diagram distribution of the feature selection results is shown in Fig. 2 . Boruta is a feature selection method based on the random forest algorithm that screens for valid features by comparing the importance of genuine features with that of randomly generated "shadow features." Its advantage is that it can reduce bias in feature selection. As can be seen from the Venn diagram, Boruta identified no uniquely selected features, with all its results overlapping with those of RFE and RF: one feature was screened out in the intersection of Boruta and RFE, 11 features in the intersection of Boruta and RF, and an additional 15 core features were jointly selected by Boruta with both RFE and RF. RF is a feature selection method based on ensemble learning that screens for high-contribution features by calculating feature importance (e.g., Gini coefficient, out-of-bag (OOB) error). Its advantages lie in the ability to capture nonlinear relationships among features and a strong resistance to overfitting. As shown in the Venn diagram, RF identified 6 uniquely selected features, 11 features in its intersection with Boruta, and participated in screening the 15 core features shared by all three algorithms, thus serving as an intermediate algorithm linking Boruta and RFE. RFE iteratively removes the "least important features" and evaluates the performance of feature subsets through cross-validation to determine the optimal number of features. It is advantageous for its high degree of automation, which can reduce the subjectivity in feature selection. As indicated in the Venn diagram, RFE yielded 4 uniquely selected features, 1 feature in its intersection with Boruta, and participated in the screening of the 15 core features common to all three algorithms. Model Construction and Validation In this study, we selected 22 machine learning algorithms, including linear, tree-based, ensemble, and neural network models. The algorithm set specifically includes Random Forest、Support Vector Machine、Bayesian GLM、Naive Bayes、K-Nearest Neighbors、Neural Network、Flexible Discriminant Analysis、Gradient Boosting Machine、Classification and Regression Tree、Elastic Net、Logistic Regression、LASSO Regression、Ridge Regression、Linear Discriminant Analysis、Quadratic Discriminant Analysis、Conditional Inference Tree、Partial Least Squares DA、Bagging Decision Trees、XGBoost、AdaBoost、LightGBM、Sparse Linear Discriminant Analysis. To conduct an objective evaluation of multiple machine learning models under consistent conditions, the dataset was sampled via a random function and partitioned into a training set and a test set, with 70% of the samples allocated to the training set and 30% to the test set. Each model was subjected to 10-fold cross-validation and parameter optimization. The evaluation metrics included: AUC, AUC_SD, Accuracy, Kappa, Sensitivity, Specificity, Pos_Pred_Value, Neg_Pred_Value, Precision, Recall, and F1. In this study, a systematic performance evaluation of the final predictive model was conducted across three key dimensions: discrimination, calibration, and clinical utility. The discriminative ability was quantitatively characterized by the area under the receiver operating characteristic curve (AUC-ROC), which was used to assess the model’s efficacy in distinguishing between subjects who experienced the primary outcome (mortality) and those who did not (survival) within 60 days. Calibration ability was visually illustrated by calibration curves that reflected the consistency between the model’s predicted probabilities and actual clinical outcomes, and the integrated Brier score was further used to quantitatively evaluate calibration accuracy. Specifically, the integrated Brier score quantifies the deviation between the predicted survival probabilities and actual survival status, thereby comprehensively reflecting the model’s calibration efficacy and fitting quality [23]. With respect to clinical utility, Decision Curve Analysis (DCA) was performed to assess the model's clinical net benefit at different threshold probabilities, to verify its value in practical clinical decision-making scenarios. Model Interpretation and Deployment An explanation method based on SHAP technology was adopted to analyze and discuss the model with the optimal overall performance, aiming to reveal its prediction mechanism and feature contributions [24]. To elucidate the model’s prediction mechanism in depth, this study identified key predictive factors using summary and dependence plots from the SHAP (Shapley Additive exPlanations) method and explored their correlations with patients’ 60-day survival outcomes. Meanwhile, latent feature interactions were uncovered via SHAP interactive plots. To improve the model’s usability and accessibility, the optimized predictive model was ultimately deployed as a web-based, interactive Shiny application that supports personalized survival prediction and customized interpretive explanations. As a lightweight web framework in the R language ecosystem, Shiny enables rapid packaging of R-built models and visualization results into interactive web applications without professional programming knowledge, allowing clinical staff without a technical background to directly operate and use the model through a web browser. Statistical Analysis Categorical variables were described using counts and percentages, while continuous variables were summarized as median (interquartile range, IQR) or mean ± standard deviation (SD), depending on their distribution characteristics. For comparisons of data distribution differences between groups, the chi-square (χ²) test was used to analyze differences in categorical variables, whereas the Mann-Whitney U test was employed to evaluate differences in continuous variables. All tests were two-tailed, and a P-value < 0.05 was considered statistically significant. All statistical analyses and chart generation in this study were performed using R software (version 4.3.1). Core analytical tasks including the plotting of receiver operating characteristic (ROC) curves and the construction and training of machine learning models were completed with the assistance of packages such as plotROC, caret, pROC, and e1071. Figure 3 illustrates the development process of the explainable machine learning model. Results Patient Characteristics The total of 11,699 patients with acute pancreatitis complicated by acute kidney injury were included in this study for a detailed retrospective analysis. The study population was stratified into the survival group and the mortality group based on 60-day survival status, with each group accounting for 50% of the total cohort. Table 1 presents the clinical characteristics of the study population. Univariate and multivariate Logistic regression analyses revealed that advanced age, low body mass index (BMI), elevated Sequential Organ Failure Assessment (SOFA) score, increased liver function score, elevated cardiovascular function score, raised renal function score, comorbid intracerebral hemorrhage, prolonged partial thromboplastin time (PTT), elevated red cell distribution width (RDW) and increased white blood cell (WBC) count were independent risk factors for the mortality outcome. In contrast, higher mean corpuscular hemoglobin concentration (MCHC), elevated renal function score, and longer length of hospital stay were identified as potential protective factors. The effect of body weight on prognosis was not statistically significant after multivariate adjustment. The association between heparin duration and prognosis was substantially confounded by other factors and thus requires further verification. Among all the factors, the SOFA score, cardiovascular function score, and comorbid intracerebral hemorrhage exerted the most significant effects on mortality risk, providing empirical support for the identification and clinical intervention of high-risk populations. Table 1 Baseline Characteristics of Patients with Acute Pancreatitis Complicated by Acute Kidney Injury at 60 Days Characteristic Statistic Alive (N = 5849) Dead(N = 5850) OR (univariable) OR (multivariable) admission_age Mean ± SD 68.7 ± 13.1 72.5 ± 9.3 1.03 (1.03–1.03, p<.001) 1.02 (1.02–1.03, p<.001) BMI Mean ± SD 29.8 ± 7.5 28.0 ± 5.7 0.96 (0.95–0.96, p<.001) 0.95 (0.94–0.97, p<.001) sofa_score 2 1814 (31%) 434 (7.4%) 3 1321 (22.6%) 727 (12.4%) 2.30 (2.00-2.64, p<.001) 1.68 (1.42–1.98, p<.001) 4 1025 (17.5%) 1224 (20.9%) 4.99 (4.37–5.70, p<.001) 2.77 (2.34–3.28, p<.001) 5 613 (10.5%) 1203 (20.6%) 8.20 (7.11–9.46, p<.001) 3.23 (2.67–3.91, p<.001) 6 394 (6.7%) 866 (14.8%) 9.19 (7.84–10.77, p<.001) 2.84 (2.28–3.54, p<.001) 7 258 (4.4%) 559 (9.6%) 9.06 (7.56–10.85, p<.001) 4.11 (3.21–5.27, p<.001) 8 182 (3.1%) 497 (8.5%) 11.41 (9.35–13.93, p<.001) 5.36 (4.04–7.11, p<.001) 9 115 (2%) 228 (3.9%) 8.29 (6.47–10.61, p<.001) 4.96 (3.50–7.01, p<.001) 10 127 (2.2%) 112 (1.9%) 3.69 (2.80–4.85, p<.001) 2.31 (1.56–3.42, p<.001) liver 0 5403 (92.4%) 4071 (69.6%) 1 220 (3.8%) 1207 (20.6%) 7.28 (6.27–8.45, p<.001) 4.40 (3.64–5.32, p<.001) 2 165 (2.8%) 502 (8.6%) 4.04 (3.37–4.84, p<.001) 2.25 (1.75–2.90, p<.001) 3 37 (0.6%) 51 (0.9%) 1.83 (1.20–2.80, p=.005) 3.33 (1.87–5.90, p<.001) 4 24 (0.4%) 19 (0.3%) 1.05 (0.57–1.92, p=.872) 1.40 (0.66–2.99, p=.378) cardiovascular 0 1648 (28.2%) 693 (11.8%) 1 2712 (46.4%) 2106 (36%) 1.85 (1.66–2.05, p<.001) 1.26 (1.10–1.44, p<.001) 2 47 (0.8%) 860 (14.7%) 43.51 (32.02–59.13, p<.001) 24.48 (17.23–34.77, p<.001) 3 874 (14.9%) 1015 (17.4%) 2.76 (2.43–3.13, p<.001) 1.51 (1.26–1.80, p<.001) 4 568 (9.7%) 1176 (20.1%) 4.92 (4.31–5.63, p<.001) 2.07 (1.70–2.52, p<.001) renal 0 3606 (61.7%) 2125 (36.3%) 1 1279 (21.9%) 1819 (31.1%) 2.41 (2.21–2.64, p<.001) 1.64 (1.45–1.85, p<.001) 2 498 (8.5%) 1237 (21.1%) 4.22 (3.75–4.74, p<.001) 2.86 (2.43–3.36, p<.001) 3 213 (3.6%) 533 (9.1%) 4.25 (3.59–5.02, p<.001) 2.95 (2.37–3.67, p<.001) 4 253 (4.3%) 136 (2.3%) 0.91 (0.74–1.13, p=.403) 0.66 (0.50–0.87, p=.004) los_hospital Mean ± SD 16.9 ± 16.7 10.3 ± 7.9 0.94 (0.94–0.95, p<.001) 0.92 (0.91–0.93, p<.001) duration_pres_heparin Mean ± SD 9.4 ± 14.2 7.4 ± 6.4 0.98 (0.98–0.99, p<.001) 1.05 (1.04–1.05, p<.001) weight Mean ± SD 85.8 ± 23.9 81.9 ± 17.0 0.99 (0.99–0.99, p<.001) 1.00 (1.00-1.01, p=.277) co_ICH 0 4662 (79.7%) 3229 (55.2%) 1 1187 (20.3%) 2621 (44.8%) 3.19 (2.94–3.46, p<.001) 2.33 (2.08–2.60, p<.001) ptt Mean ± SD 39.6 ± 25.7 52.4 ± 32.6 1.02 (1.01–1.02, p<.001) 1.01 (1.01–1.01, p<.001) mchc Mean ± SD 32.8 ± 1.7 32.3 ± 1.3 0.80 (0.78–0.82, p<.001) 0.93 (0.90–0.97, p<.001) platelet Mean ± SD 211.6 ± 102.9 237.5 ± 94.1 1.00 (1.00–1.00, p<.001) 1.00 (1.00–1.00, p<.001) rdw Mean ± SD 14.9 ± 2.1 15.7 ± 1.9 1.24 (1.22–1.27, p<.001) 1.18 (1.15–1.21, p<.001) wbc Mean ± SD 12.1 ± 8.2 14.4 ± 6.5 1.06 (1.05–1.06, p<.001) 1.02 (1.02–1.03, p<.001) Feature Extraction Figure 4 depicts the process of optimal feature selection combining three algorithms: Boruta, RF, and RFE based on cross-validation. The feature selection process for each algorithm is presented in (Supplementary Fig. 1: Feature Selection Process of the Three Algorithms). Ultimately, we selected 15 features common to all three models (SOFA score, liver score, cardiovascular score, renal score, los_hospital, admission_age, duration_pres_aspirin, weight, co_ICH, ptt, mchc, platelet count, rdw, wbc, and BMI as input parameters for machine learning. Model Selection and Construction Using the 15 features common to the three models, we constructed 22 machine learning models to predict patients' 60-day survival status. As shown in Table 2 , the RF algorithm achieved the optimal predictive performance in the benchmark test, with the values of Accuracy, AUC, Kappa, Sensitivity, Specificity, PPV, NPV, Precision, Recall, and F1 reaching 0.9068, 0.9731, 0.8138, 0.8755, 0.9392, 0.9370, 0.8796, 0.9370, 0.8755, and 0.9052, respectively. The Random Forest model yielded the lowest Brier score of 0.0722, outperforming all other models (Figs. 5 and 6 and Supplementary Table 1: Model Brier Score). As illustrated in Fig. 7 , among the 22 machine learning models evaluated in this study, RF was selected as the core analytical framework owing to its excellent and well-balanced performance. RF not only achieved the highest AUC value, demonstrating outstanding discriminative ability in identifying high-risk cases, but also had the smallest standard deviation (SD = 0.005), confirming its robust generalization and extremely low performance fluctuation across different data subsets. This finding not only verified that RF was the most reliable model for the clinical prediction task in this study, but also laid a solid foundation for the subsequent SHAP-based interpretive analysis of feature contributions. Table 2 Various Performance Metrics of 22 Machine Learning Models Model AUC AUC_SD Accuracy Kappa Sensitivity Specificity PPV NPV Precision Recall F1 Random Forest 0.9731 0.0053 0.9068 0.8138 0.8755 0.9392 0.9370 0.8796 0.9370 0.8755 0.9052 Support Vector Machine 0.8487 0.0162 0.7909 0.5819 0.7757 0.8066 0.8055 0.7769 0.8055 0.7757 0.7903 Bayesian GLM 0.8489 0.0158 0.7889 0.5780 0.7695 0.8089 0.8061 0.7727 0.8061 0.7695 0.7874 Naive Bayes 0.8932 0.0137 0.8191 0.6391 0.7386 0.9021 0.8863 0.7698 0.8863 0.7386 0.8058 K-Nearest Neighbors 0.8591 0.0102 0.7823 0.5643 0.8020 0.7620 0.7768 0.7885 0.7768 0.8020 0.7892 Neural Network 0.8551 0.0212 0.7735 0.5474 0.7398 0.8083 0.7994 0.7505 0.7994 0.7398 0.7684 Flexible Discriminant Analysis 0.8849 0.0123 0.8071 0.6143 0.7964 0.8182 0.8189 0.7956 0.8189 0.7964 0.8075 Gradient Boosting Machine 0.9166 0.0104 0.8325 0.6652 0.8059 0.8599 0.8559 0.8110 0.8559 0.8059 0.8302 Classification and Regression Tree 0.7242 0.0245 0.7120 0.4241 0.6977 0.7267 0.7249 0.6996 0.7249 0.6977 0.7111 Elastic Net 0.8490 0.0160 0.7892 0.5785 0.7717 0.8072 0.8051 0.7740 0.8051 0.7717 0.7881 Logistic Regression 0.8488 0.0158 0.7889 0.5780 0.7695 0.8089 0.8061 0.7727 0.8061 0.7695 0.7874 LASSO Regression 0.8490 0.0131 0.7889 0.5780 0.7717 0.8066 0.8047 0.7739 0.8047 0.7717 0.7879 Ridge Regression 0.8484 0.0175 0.7917 0.5836 0.7762 0.8078 0.8065 0.7776 0.8065 0.7762 0.7911 Linear Discriminant Analysis 0.8464 0.0181 0.7863 0.5728 0.7712 0.8020 0.8008 0.7724 0.8008 0.7712 0.7857 Quadratic Discriminant Analysis 0.8498 0.0176 0.7761 0.5520 0.7835 0.7684 0.7774 0.7747 0.7774 0.7835 0.7804 Conditional Inference Tree 0.8843 0.0144 0.8017 0.6036 0.7835 0.8205 0.8184 0.7859 0.8184 0.7835 0.8006 Partial Least Squares DA 0.8467 0.0187 0.7892 0.5784 0.7830 0.7956 0.7982 0.7802 0.7982 0.7830 0.7905 Bagging Decision Trees 0.9550 0.0092 0.8934 0.7870 0.8727 0.9149 0.9137 0.8744 0.9137 0.8727 0.8927 XGBoost 0.9626 0.0060 0.8997 0.7995 0.8800 0.9201 0.9192 0.8813 0.9192 0.8800 0.8991 AdaBoost 0.9186 0.0110 0.8311 0.6623 0.8059 0.8570 0.8533 0.8105 0.8533 0.8059 0.8290 LightGBM 0.9403 0.0135 0.8624 0.7251 0.8222 0.9039 0.8983 0.8312 0.8983 0.8222 0.8586 Sparse Linear Discriminant Analysis 0.8220 0.0169 0.5100 0.0041 1.0000 0.0041 0.5090 1.0000 0.5090 1.0000 0.6746 The calibration curves of all models were close to the ideal dashed line, with high fitting for most models. The group-specific error bars (vertical bars) were generally short, indicating small fluctuations in the proportion of actual events within each group. The models' predicted probabilities could well reflect the actual risk of event occurrence, thus providing reliable probabilistic references for clinical decision-making (Fig. 8 ). Results of the DCA showed that RF and other machine learning models included in the study all demonstrated positive clinical application value across different probability threshold intervals. Owing to its robust performance, the RF model can serve as a fundamental tool for routine clinical decision-making, whereas models such as LGBM and XGB are better suited for scenarios that demand high precision. These results demonstrated that different machine learning models can provide multidimensional support for clinical decision-making, facilitating the implementation of individualized risk assessment and the formulation of targeted intervention strategies (Fig. 9 ). Model Interpretation and Deployment The SHAP method was employed to interpret the model by ranking the importance of candidate features in individual predictions. As illustrated in Figs. 10 and 11 , admission_age, co_ICH, liver score, and cardiovascular status were the most impactful predictive factors; their high values (yellow dots) consistently elevated the predicted risk, whereas duration_pres_heparin exhibited a potential protective effect with negative SHAP values. This visualization enhanced the transparency of the "black-box" model by quantifying the magnitude and direction of each feature’s contribution, and verified the clinical relevance of key predictive factors, which were highly consistent with the existing domain knowledge. At the global level, the bee plot revealed that admission_age, co_ICH, liver function score, and cardiovascular status (cardiovascular) were the most impactful predictive factors; their high values generally increased the predicted risk of mortality, whereas the duration_pres_heparin exhibited a potential protective effect. A detailed analysis of Sample 78 demonstrated that protective factors, particularly a SOFA score of 2 (SHAP value = -0.1) and a white blood cell count of 7.1 (wbc = 7.1, SHAP value = -0.09), predominated in the prediction, reducing the mortality probability from a baseline of 0.487 to 0.22. The countervailing effects of risk-increasing factors (e.g., red cell distribution width = 16.2 (rdw = 16.2) and length of hospital stay = 6.17 (los_hospital = 6.17)) were relatively weak. This also illustrated how the model generates individualized risk estimates by aggregating the contributions of each feature, and this process was highly consistent with the patterns of feature importance observed at the global level. Figure 12 illustrates the non-linear relationships between key features and predicted mortality risk, with the red trend lines highlighting the overall direction of each feature’s effect. For ptt and rdw, SHAP values rise with increasing feature values, peak at moderate levels, and then decline, indicating moderate values exert the strongest enhancing effect on mortality risk, whereas the impact of extreme values is instead attenuated. los_hospital exhibits an initial negative correlation (longer hospital stays reduce mortality risk), with the trend reversing at approximately 100 days, reflecting a complex time-dependent relationship. The SOFA score and wbc also show an initial positive impact that diminishes at high values. In contrast, the duration_pres_heparin follows a U-shaped pattern: both short-term and long-term heparin use are associated with reduced mortality risk, while moderate-duration use exerts a neutral to positive impact on risk. We integrated the final RF model into an interactive Shiny-based web application that provides individualized survival predictions and interpretations, as well as global model interpretation, as shown in Fig. 13 . This web application is accessible via the URL: https://medicalpredictor.shinyapps.io/Online_Prediction_of_60-Day_Survival/ . Focusing on the high-risk clinical scenario of acute pancreatitis complicated by acute kidney injury, this system takes the Random Forest algorithm as its core and constructs an integrated workflow encompassing parameter entry, one-click prediction, and result visualization. It can not only accurately output the 60-day survival probability of patients but also enable clinicians to intuitively understand the key influencing factors underlying the predictions through SHAP feature importance analysis. This not only improves the efficiency of clinical decision-making but also enhances the interpretability and credibility of the model, thereby providing strong support for the formulation of personalized treatment plans. Discussion Acute pancreatitis complicated by acute kidney injury (AP-AKI) is a high-risk disease complex in the field of critical care medicine, and its prognostic assessment has long been a key focus of clinical research. Based on large-sample data from the MIMIC-IV v3.1 database, this study identified the core predictive factors for 60-day survival in AP-AKI patients through multi-algorithmic feature selection and machine learning model construction, and developed an interpretable and generalizable clinical tool, thereby providing novel, evidence-based guidance for the disease management of AP-AKI. Multivariate Logistic regression analysis in this study revealed that admission age was an independent risk factor for 60-day mortality in AP-AKI patients (OR = 1.02, 95% confidence interval (95% CI: 1.02–1.03, P < 0.001). The mean age of patients in the mortality group (72.5 ± 9.3 years) was significantly higher than that in the survival group (68.7 ± 13.1 years), consistent with previous studies [25,26]. Elderly patients exhibit impaired organ reserve function and reduced tolerance to inflammatory stress, rendering them more susceptible to sequential multi-organ dysfunction injury. Low BMI was also associated with an increased risk of mortality (OR = 0.95, 95% CI: 0.94–0.97, P < 0.001), suggesting that nutritional status may regulate disease prognosis by modulating immune function, tissue repair capacity, and related pathways. This finding provides insights for the implementation of clinical nutritional support interventions. Furthermore, this study balanced baseline distributions between the survival and mortality groups using propensity score matching (PSM), ultimately yielding a well-balanced cohort of 11,699 patients (5,849 and 5,850 in each group, respectively). This approach effectively reduced biases arising from confounding demographic characteristics and enhanced the reliability of the study conclusions. AKI is one of the major complications in patients with AP. Its development is influenced by multiple factors and directly leads to organ damage and multiple organ failure [27]. Organ function-related scores were identified as the core prognostic features in this study. As a key indicator reflecting the degree of multi-organ dysfunction, the SOFA score exhibited a significant increasing trend in its risk effect with rising scores: at a SOFA score of 8, the odds ratio (OR) for mortality risk reached as high as 5.36 (95% CI: 4.04–7.11, P < 0.001), and the score also emerged as an important variable influencing prediction results in SHAP analysis. This finding highlights the central role of multi-organ function protection in the treatment of AP-AKI, which is highly consistent with the pathophysiological mechanism of acute pancreatitis. The systemic inflammatory response syndrome (SIRS) induced by pancreatic inflammation can cause the sequential impairment of liver, cardiovascular, renal, and other organ functions through the diffusion of inflammatory mediators. Elevated liver, cardiovascular, and renal function scores were all closely associated with an increased risk of mortality. Specifically, at a cardiovascular function score of 2, the odds ratio (OR) for mortality risk reached 24.48 (95% confidence interval (95% CI: 17.23–34.77, P < 0.001), and for patients with Stage Ⅲ renal function score, the OR was 2.95 (95% CI: 2.37–3.67, P < 0.001). These findings indicate that cardiovascular system failure and moderate to severe renal injury are strong predictive markers for poor prognosis in AP-AKI patients. In terms of biochemical indicators, ptt (OR = 1.01, 95% CI: 1.01–1.01, P < 0.001), elevated rdw (OR = 1.18, 95% CI: 1.15–1.21, P < 0.001), and increased wbc count (OR = 1.02, 95% CI: 1.02–1.03, P < 0.001) were all identified as independent risk factors, respectively reflecting coagulation dysfunction, hematopoietic system stress, and systemic inflammatory activation. The degree of abnormality of these indicators can serve as key entry points for clinical disease monitoring. The staging of renal function injury is directly correlated with prognosis [28,29]. The OR for mortality risk increased progressively from 1.64 to 2.95 as the renal function score advanced from Stage Ⅰ to Stage Ⅲ, demonstrating that the severity of renal injury is positively associated with poorer patient prognosis. The progressive deterioration of renal function exacerbates systemic metabolic disorders and triggers sequential cascades of multi-organ dysfunction. Additionally, patients with co_ICH had a markedly increased mortality risk (OR = 2.33, 95% CI: 2.08–2.60, P < 0.001). This may be attributed to elevated intracranial pressure and cerebral hypoperfusion induced by intracerebral hemorrhage, as well as the synergistic effect of systemic inflammation caused by the coexistence of intracerebral hemorrhage and AP-AKI. Such patients require more intensive multidisciplinary collaborative management. Among treatment-related factors, length of hospital stay (OR = 0.92, 95% CI: 0.91–0.93, P < 0.001) was identified as a potential protective factor, with longer hospital stays at the initial stage possibly reflecting the effectiveness of clinical interventions. In contrast, the association between heparin duration and prognosis was substantially confounded by other factors (multivariate OR = 1.05, 95% CI: 1.04–1.05, P < 0.001) and thus requires further verification in prospective studies. SHAP analysis revealed that the duration of heparin use was associated with negative SHAP values, suggesting that heparin may reduce the risk of thrombus-related complications (e.g., renal vein thrombosis) through its anticoagulant effect. This hypothesis provides a direction for optimizing treatment strategies in AP-AKI patients. Among the 22 machine learning models constructed in this study, the RF model demonstrated the optimal performance, with an AUC of 0.9731, an Accuracy of 0.9068, a Sensitivity of 0.8755, a Specificity of 0.9392, and the lowest Brier score (0.0722), exhibiting excellent discriminative and calibration performance. Compared with traditional linear models (e.g., Logistic regression with an AUC of 0.8488), the RF model can more accurately capture the non-linear relationships and interactions between features, which is fully verified by SHAP analysis: ptt and rdw show a threshold effect with mortality risk (the highest risk at moderate levels), length of hospital stay presents a time-dependent association (the trend reversed around 100 days), and the duration of heparin use follows a U-shaped pattern (both short-term and long-term use may reduce the risk of mortality). The revelation of these non-linear relationships breaks through the limitations of the linear assumption inherent in traditional prognostic models, providing a more clinically relevant basis for individualized risk assessment. The interactive Shiny Web application developed using the Random Forest (RF) model ( https://medicalpredictor.shinyapps.io/Online_Prediction_of_60-Day_Survival/ ) provides an integrated workflow from parameter entry to prognostic visualization. Without professional programming knowledge, clinicians can quickly obtain the 60-day survival probability and interpretations of key influencing factors by entering 15 core patient features (including SOFA score, liver and renal function, admission age, etc.). This has significantly enhanced the model's clinical utility, providing a convenient tool for early risk stratification and personalized intervention in AP-AKI patients. Although valuable results have been achieved in this study, certain limitations remain. First, the study adopted a retrospective design with data from a single database (MIMIC-IV v3.1). Derived from standardized clinical records, the data ensured a large sample size and completeness but may be subject to selection bias. The model's generalizability needs further verification across heterogeneous cohorts, including non-Western populations and community hospitals. Second, the database failed to fully capture certain clinical details, including specific treatment regimens (timing and dosage of renal replacement therapy), nutritional status assessment, and patients' long-term quality of life. These factors may affect prognosis and should be incorporated into subsequent studies. Third, this study focused on short-term 60-day survival prediction, while long-term outcomes of AP-AKI patients (e.g., progression to chronic kidney disease, readmission risk) have not been addressed. Future research can expand the model's outcome coverage. Finally, despite balancing baseline confounders through propensity score matching (PSM), residual confounding inherent to retrospective studies may persist. The validity and interventional value of the model need further verification via prospective studies. Conclusion This study developed and validated an interpretable Random Forest (RF) model to predict 60-day survival in patients with acute pancreatitis complicated by acute kidney injury (AP-AKI). The model exhibits excellent discriminative efficacy, calibration accuracy, and clinical utility. The interactive web application facilitates the integration of the model into clinical workflows, providing clinicians with a convenient tool for early risk stratification and a reference for personalized interventions, thereby supporting the optimization of diagnosis and treatment decisions. Future research should focus on external validation in multicenter cohorts to further enhance the model’s generalizability, incorporate more clinically relevant features (e.g., specific treatment strategies, nutritional status) to improve model performance, and clarify the model’s role in improving patients’ final outcomes through prospective studies. Declarations Acknowledgments We would like to acknowledge the researchers and participants who contributed to the MIMIC-IV database. Funding This research was funded by the Anhui Province Teaching Research Program (Granted No. 2023jyxm0347). Conflict of interest The authors declare no conflict of interest. Ethics Statement This study was based on data extracted from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database—a publicly accessible, de-identified critical care database. Ethical approval for the establishment and ongoing maintenance of MIMIC-IV was obtained from the Institutional Review Boards (IRBs) of both the Massachusetts Institute of Technology (MIT) and Beth Israel Deaconess Medical Center (BIDMC), and informed consent was waived due to the retrospective, fully anonymized nature of the data. For this secondary analysis of de-identified data, no further ethical approval was deemed necessary. The study was designed and conducted in strict adherence to the ethical standards of the institutional and/or national research committees, as well as the 1964 Declaration of Helsinki and its subsequent amendments or equivalent ethical guidelines. Authorship contribution Data Analysis: T.S., and W.W; Writing Original Draft: T.S., and WW. Data availability Statement The data used in this study are publicly available from the MIMIC-IV database. Access to these data requires completion of mandatory training and approval via PhysioNet (https://physionet.org/). References Kumar R, Pahwa N, Jain N. Acute kidney injury in severe acute pancreatitis: an experience from a tertiary care center. Saudi J Kidney Dis Transpl. 2015;26(1):56-60. doi:10.4103/1319-2442.148734 Yadav D, Lowenfels AB. The epidemiology of pancreatitis and pancreatic cancer. Gastroenterology. 2013;144(6):1252-1261. doi:10.1053/j.gastro.2013.01.068 Bollen TL, van Santvoort HC, Besselink MG, et al. The Atlanta Classification of acute pancreatitis revisited. Br J Surg. 2008;95(1):6-21. doi:10.1002/bjs.6010 Liu B, Zhang X, Li J, et al. The Global, Regional, and National Burden of Pancreatitis in 204 Countries and Territories, 1990-2021: A Systematic Analysis for the Global Burden of Disease Study 2021. Dig Dis Sci. 2025;70(7):2328-2339. doi:10.1007/s10620-025-08996-y Zhou J, Li Y, Tang Y, et al. Effect of acute kidney injury on mortality and hospital stay in patient with severe acute pancreatitis. Nephrology (Carlton). 2015;20(7):485-491. doi:10.1111/nep.12439 Selvanathan DK, Johnson PG, Thanikachalam DK, Rajendran P, Gopalakrishnan N. Acute Kidney Injury Complicating Severe Acute Pancreatitis: Clinical Profile and Factors Predicting Mortality. Indian J Nephrol. 2022;32(5):460-466. doi:10.4103/ijn.IJN_476_20 Kes P, Vucicević Z, Ratković-Gusić I, Fotivec A. Acute renal failure complicating severe acute pancreatitis. Ren Fail. 1996;18(4):621-628. doi:10.3109/08860229609047686 Naqvi R. Acute Kidney Injury in association with Acute Pancreatitis. Pak J Med Sci. 2018;34(3):606-609. doi:10.12669/pjms.343.14953 Nassar TI, Qunibi WY. AKI Associated with Acute Pancreatitis. Clin J Am Soc Nephrol. 2019;14(7):1106-1115. doi:10.2215/CJN.13191118 Bagshaw SM, George C, Dinu I, Bellomo R. A multi-centre evaluation of the RIFLE criteria for early acute kidney injury in critically ill patients. Nephrol Dial Transplant. 2008;23(4):1203-1210. doi:10.1093/ndt/gfm744 Saimon, S. I., Islam, I., Abir, S. I., Sultana, N., Hossain, M. S., & Al Shiam, S. A. Advancing Neurological Disease Prediction through Machine Learning Techniques. Journal of Computer Science and Technology Studies, 2025;7(1), 139-156. doi:10.32996/jcsts.2025.7.1.11 Ejiyi CJ, Qin Z, Nneji GU, et al. Enhanced Cardiovascular Disease Prediction Modelling using Machine Learning Techniques: A Focus on CardioVitalnet. Network. 2025;36(3):716-748. doi:10.1080/0954898X.2024.2343341 Sanju, P., Ahmed, N. S. S., Ramachandran, P., Sajid, P. M., & Jayanthi, R. (2025). Enhancing thyroid disease prediction and comorbidity management through advanced machine learning frameworks. Clinical eHealth,2025; 8:7-16. doi:10.1016/j.ceh.2025.01.002 Qi X, Wang S, Fang C, Jia J, Lin L, Yuan T. Machine learning and SHAP value interpretation for predicting comorbidity of cardiovascular disease and cancer with dietary antioxidants. Redox Biol. 2025;79:103470. doi:10.1016/j.redox.2024.103470 Thatha VN, Chalichalamala S, Pamula U, et al. Optimized machine learning mechanism for big data healthcare system to predict disease risk factor. Sci Rep. 2025;15(1):14327. Published 2025 Apr 24. doi:10.1038/s41598-025-98721-6 Tan Z, Li G, Zheng Y, et al. Advances in the clinical application of machine learning in acute pancreatitis: a review. Front Med (Lausanne). 2025;11:1487271. Published 2025 Jan 7. doi:10.3389/fmed.2024.1487271 Shi N, Lan L, Luo J, et al. Predicting the Need for Therapeutic Intervention and Mortality in Acute Pancreatitis: A Two-Center International Study Using Machine Learning. J Pers Med. 2022;12(4):616. Published 2022 Apr 11. doi:10.3390/jpm12040616 Kiss S, Pintér J, Molontay R, et al. Early prediction of acute necrotizing pancreatitis by artificial intelligence: a prospective cohort-analysis of 2387 cases. Sci Rep. 2022;12(1):7827. Published 2022 May 12. doi:10.1038/s41598-022-11517-w Lan L, Guo Q, Zhang Z, et al. Classification of Infected Necrotizing Pancreatitis for Surgery Within or Beyond 4 Weeks Using Machine Learning. Front Bioeng Biotechnol. 2020;8:541. Published 2020 Jun 4. doi:10.3389/fbioe.2020.00541 Luo Z, Shi J, Fang Y, et al. Development and evaluation of machine learning models and nomogram for the prediction of severe acute pancreatitis. J Gastroenterol Hepatol. 2023;38(3):468-475. doi:10.1111/jgh.16125 Yuan L, Ji M, Wang S, et al. Machine learning model identifies aggressive acute pancreatitis within 48 h of admission: a large retrospective study. BMC Med Inform Decis Mak. 2022;22(1):312. Published 2022 Nov 29. doi:10.1186/s12911-022-02066-3 Johnson, A., Bulgarelli, L., Pollard, T., Gow, B., Moody, B., Horng, S., Celi, L. A., & Mark, R. MIMIC-IV (version 3.1). PhysioNet. 2024, RRID:SCR_007345. https://doi.org/10.13026/kpb9-mt58 Park SY, Park JE, Kim H, Park SH. Review of Statistical Methods for Evaluating the Performance of Survival or Other Time-to-Event Prediction Models (from Conventional to Deep Learning Approaches). Korean J Radiol. 2021;22(10):1697-1707. doi:10.3348/kjr.2021.0223 Lundberg SM, Erion G, Chen H, et al. From Local Explanations to Global Understanding with Explainable AI for Trees. Nat Mach Intell. 2020;2(1):56-67. doi:10.1038/s42256-019-0138-9 Li H, Qian Z, Liu Z, Liu X, Han X, Kang H. Risk factors and outcome of acute renal failure in patients with severe acute pancreatitis. J Crit Care. 2010;25(2):225-229. doi:10.1016/j.jcrc.2009.07.009 Mederos MA, Reber HA, Girgis MD. Acute Pancreatitis: A Review. JAMA. 2021;325(4):382-390. doi:10.1001/jama.2020.20317 Mathur P, Vaishnav S. Pathophysiology of acute kidney injury in severe acute pancreatitis-An overview Gastroenterol Hepatol Open Access. 2019;10:242–5 Zhou J, Li Y, Tang Y, et al. Effect of acute kidney injury on mortality and hospital stay in patient with severe acute pancreatitis. Nephrology (Carlton). 2015;20(7):485-491. doi:10.1111/nep.12439 Kong L, Santiago N, Han TQ, Zhang SD. Clinical characteristics and prognostic factors of severe acute pancreatitis. World J Gastroenterol. 2004;10(22):3336-3338. doi:10.3748/wjg.v10.i22.3336 Additional Declarations No competing interests reported. Supplementary Files SupplementaryTable1.BrierScore.csv SupplementaryFigure1.tif Cite Share Download PDF Status: Posted Version 1 posted 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-8896948","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":594209934,"identity":"a25a4219-d689-4542-9ec4-5557921e51ad","order_by":0,"name":"Wei Wang","email":"","orcid":"","institution":"Anhui University of Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Wei","middleName":"","lastName":"Wang","suffix":""},{"id":594209935,"identity":"2aeac1d8-29a7-40d5-ad75-c5fa3d9bde75","order_by":1,"name":"Tongping Shen","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA9klEQVRIiWNgGAWjYDACCTBpw8PGzHyAGSpmQFjLAYY0OT72tgRmhgTitRw2luM5Y0CcFvnZzccef9zBnNgmkfPxc+EPu2gG9uZtEgw1d3BqMbhzLN3g4Bk2oJbczdIzEpJzG3iOlUkwHHuGW4tEjpnEwTYekJYN0jwJB3IbQCKMDYdxO2xG/jegFgmQwx7/BmuRf4NfC8ONHDagFgNjNp4zbFBbePBrMbiRZiZx9kyCHBt7m5k1T1pybhtPWrFFwjF8Dkt+JlG54z+PfDPz49s8Nna5/eyHN974UIPHYSDA2IDEYQMRCfg1oGkZBaNgFIyCUYAOAFaFU907QtoSAAAAAElFTkSuQmCC","orcid":"","institution":"Anhui University of Chinese Medicine","correspondingAuthor":true,"prefix":"","firstName":"Tongping","middleName":"","lastName":"Shen","suffix":""}],"badges":[],"createdAt":"2026-02-17 01:53:45","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8896948/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8896948/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":103349311,"identity":"512033cf-2f1c-4b5a-b51f-249c5a0e34d4","added_by":"auto","created_at":"2026-02-24 16:41:52","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":44367,"visible":true,"origin":"","legend":"\u003cp\u003eSample Size Comparison Before and After PSM\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8896948/v1/dc055a6f0e66f5be0a7f35f5.png"},{"id":103506960,"identity":"85b21273-d4da-42b4-a712-b906557265a8","added_by":"auto","created_at":"2026-02-26 13:40:04","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":126909,"visible":true,"origin":"","legend":"\u003cp\u003eVenn Diagram of Features Selected by Three Algorithms\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8896948/v1/80a747e51c655e92865a6535.png"},{"id":103349316,"identity":"3784fa74-b46a-4a0f-96bb-67201a311729","added_by":"auto","created_at":"2026-02-24 16:41:53","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":375862,"visible":true,"origin":"","legend":"\u003cp\u003eFramework Diagram of Prostate Cancer Treatment Decision Model\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-8896948/v1/b8b34efc3bfc988fd8da2440.png"},{"id":103349310,"identity":"c51b4a5e-eb9b-48bb-8c01-d5cf6aea0529","added_by":"auto","created_at":"2026-02-24 16:41:52","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":97319,"visible":true,"origin":"","legend":"\u003cp\u003eOptimal Feature Selection Process for the Merging of Three Algorithms\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-8896948/v1/619f6d7f23397395cf6d2901.png"},{"id":103349317,"identity":"6d109e54-0fe2-45a9-b742-7c951369fc31","added_by":"auto","created_at":"2026-02-24 16:41:53","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":67429,"visible":true,"origin":"","legend":"\u003cp\u003eAccuracy of 12 Machine Learning Models\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-8896948/v1/7d34616ad1c6c848d5cf0418.png"},{"id":103349320,"identity":"8b6076bf-352d-4c8d-8203-b12375440402","added_by":"auto","created_at":"2026-02-24 16:41:53","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":154562,"visible":true,"origin":"","legend":"\u003cp\u003eROC Curves on Test for 12 Machine Learning Models\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-8896948/v1/576a5c9a2c490099a9280d5a.png"},{"id":103349313,"identity":"840c6410-b166-4a95-a544-f2394087a00a","added_by":"auto","created_at":"2026-02-24 16:41:52","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":76916,"visible":true,"origin":"","legend":"\u003cp\u003ePredictive Performance and Stability of 22 Machine Learning Models\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-8896948/v1/712f04912620467db9d12605.png"},{"id":103506900,"identity":"aef1f038-93d1-491d-86f6-031877e5ceef","added_by":"auto","created_at":"2026-02-26 13:39:50","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":168581,"visible":true,"origin":"","legend":"\u003cp\u003eCalibration Curves in the test\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-8896948/v1/4f73adc3d7decd220083cd96.png"},{"id":103349323,"identity":"6c15ad41-51c8-47ed-a7a8-564c21a7417f","added_by":"auto","created_at":"2026-02-24 16:41:53","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":164896,"visible":true,"origin":"","legend":"\u003cp\u003eDecision Curve Analysis Plot of 10 Models\u003c/p\u003e","description":"","filename":"9.png","url":"https://assets-eu.researchsquare.com/files/rs-8896948/v1/d93f0ad70425133f50ef96bb.png"},{"id":103349321,"identity":"3b8afc5d-f26a-48bc-a5d7-57c3262e8f38","added_by":"auto","created_at":"2026-02-24 16:41:53","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":219857,"visible":true,"origin":"","legend":"\u003cp\u003eBeeswarm Plot of SHAP Values Based on Random Forest Model\u003c/p\u003e","description":"","filename":"10.png","url":"https://assets-eu.researchsquare.com/files/rs-8896948/v1/94bda07255fcb7dc51e723d5.png"},{"id":103506414,"identity":"36c2ad35-a51c-4d3e-8a87-ee3d01179bbe","added_by":"auto","created_at":"2026-02-26 13:36:11","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":99642,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP Feature Contribution Analysis Plot for a Single Sample\u003c/p\u003e","description":"","filename":"11.png","url":"https://assets-eu.researchsquare.com/files/rs-8896948/v1/18cb95f8e8d37e4897ad0611.png"},{"id":103507060,"identity":"b02ded92-205b-4e7e-9d36-c4c3defd9ba3","added_by":"auto","created_at":"2026-02-26 13:40:19","extension":"png","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":310384,"visible":true,"origin":"","legend":"\u003cp\u003eVisualization of SHAP Dependence Relationships for Multiple Features\u003c/p\u003e","description":"","filename":"12.png","url":"https://assets-eu.researchsquare.com/files/rs-8896948/v1/1bd31db168cb2fcfc04c0374.png"},{"id":103349319,"identity":"56fdc218-e057-4048-b156-e412ed2cadf0","added_by":"auto","created_at":"2026-02-24 16:41:53","extension":"png","order_by":13,"title":"Figure 13","display":"","copyAsset":false,"role":"figure","size":370393,"visible":true,"origin":"","legend":"\u003cp\u003eOnline Prediction System Interface for 60-Day Survival Status in Patients with Acute Pancreatitis Complicated by Acute Kidney Injury\u003c/p\u003e","description":"","filename":"13.png","url":"https://assets-eu.researchsquare.com/files/rs-8896948/v1/9e861480f1a4432587885387.png"},{"id":104835469,"identity":"39001109-a056-4596-bf4a-8da7aa20bac0","added_by":"auto","created_at":"2026-03-17 17:45:13","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2867963,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8896948/v1/1a55a627-4215-44fe-8254-1df1f68fc8ad.pdf"},{"id":103349314,"identity":"ceeedba5-70f5-4913-ad4e-9041bbb94428","added_by":"auto","created_at":"2026-02-24 16:41:53","extension":"csv","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":140,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTable1.BrierScore.csv","url":"https://assets-eu.researchsquare.com/files/rs-8896948/v1/f287d75b1c1335da907708ad.csv"},{"id":103506788,"identity":"1667a7bd-6c35-4416-9699-bd8cedd515bc","added_by":"auto","created_at":"2026-02-26 13:39:29","extension":"tif","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":1138150,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryFigure1.tif","url":"https://assets-eu.researchsquare.com/files/rs-8896948/v1/423915c7da2516ac69a1a3be.tif"}],"financialInterests":"No competing interests reported.","formattedTitle":"Development and Validation of a Machine Learning-Based Risk Prediction Model for 60-Day Survival Status in Patients with Acute Pancreatitis Complicated with Acute Kidney Injury","fulltext":[{"header":"Introduction","content":"\u003cp\u003eAP is an inflammatory disease of the pancreas with a spectrum of severity ranging from mild pancreatitis to SAP [1]. It is caused by multiple factors, including gallstones, alcohol consumption, hypertriglyceridemia, and medications [2]. AP presents with clinical manifestations varying from mild abdominal pain to multiple organ dysfunction syndrome [3]. In 2021, an estimated 5.9\u0026nbsp;million people worldwide were living with pancreatitis, 2.7\u0026nbsp;million new cases were diagnosed, and 122,000 deaths were attributed to the disease [4].\u003c/p\u003e \u003cp\u003eAcute kidney injury (AKI) is one of the most common and severe complications of severe acute pancreatitis (SAP), with a reported incidence ranging from 14% to 42% [5]. Studies have shown that pancreatitis complicated with AKI significantly increases mortality [6,7,8]. The pathogenesis of AKI in SAP is multifactorial, involving hemodynamic changes, inflammatory mediators, oxidative stress, and direct nephrotoxicity [9]. Therefore, early identification and management of AKI in SAP patients are crucial for improving prognosis and reducing mortality [10].\u003c/p\u003e \u003cp\u003eAs an important branch of artificial intelligence, machine learning (ML) can uncover complex nonlinear correlations between features and disease outcomes. It has been widely applied in disease diagnosis, complication monitoring, and prognosis prediction, providing auxiliary support for clinical decision-making [11,12,13,14,15,16]. Notably, numerous studies have utilized ML techniques to construct prediction tools for AP, with a primary focus on predicting disease severity, complications, and mortality [17,18,19,20,21].\u003c/p\u003e \u003cp\u003eHowever, most existing studies are limited to retrospective, single-center designs, with drawbacks such as small sample sizes and limited model interpretability. To address these gaps, this study aims to develop and validate an interpretable machine learning model to predict 60-day survival in patients with AP complicated by AKI. The model employs the Shapley Additive exPlanations (SHAP) method to elucidate the impact of key features and their potential interactions. Additionally, a Shiny-based interactive web application has been developed to improve the model\u0026rsquo;s clinical applicability.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStudy Design and Study Population\u003c/h2\u003e \u003cp\u003eThis study used data from the MIMIC-IV v3.1 database, released in October 2024 [22]. The author Tongping Shen completed the training required by the National Institutes of Health (NIH) and obtained database access permission (Record ID: 14348115). Since the data had been anonymized, written informed consent was waived, and the study adhered to the observational research norms outlined in the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.\u003c/p\u003e \u003cp\u003eInclusion criteria were as follows: (1) Diagnosis of acute pancreatitis confirmed by ICD diagnostic codes; (2) Concurrent fulfillment of clinical diagnostic criteria for acute kidney injury (AKI); (3) Age\u0026thinsp;\u0026ge;\u0026thinsp;18 years; (4) Completeness rate of clinical data\u0026thinsp;\u0026ge;\u0026thinsp;80%, with no missing key variables (e.g., gender, age, underlying diseases, severity grading of pancreatitis, AKI staging, core laboratory indicators, treatment measures, etc.); (5) Follow-up duration\u0026thinsp;\u0026ge;\u0026thinsp;60 days or a clearly documented death outcome within 60 days.\u003c/p\u003e \u003cp\u003eExclusion criteria were as follows: (1) Acute exacerbation of chronic pancreatitis or other types of pancreatitis; (2) AKI complicated by end-stage chronic kidney disease (ESKD); (3) Cases with hospital stay\u0026thinsp;\u0026lt;\u0026thinsp;24 hours; (4) Duplicate hospitalization records (for patients admitted multiple times due to acute pancreatitis complicated with AKI, only data from the first admission were included); (5) Data with obvious logical errors or outliers (e.g., laboratory indicators outside the clinically reasonable range without supporting original records).\u003c/p\u003e \u003cp\u003eAfter screening against the inclusion and exclusion criteria, 10,218 eligible study participants were identified. Each participant was uniquely identified by subject_id to ensure sample independence. Clinical data for all included patients were derived from routine clinical diagnosis and treatment records, which were standardized and de-identified in the database in compliance with privacy protection regulations.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eData Preprocessing and Feature Selection\u003c/h3\u003e\n\u003cp\u003eThis study initially retrieved clinical data for 10,218 patients with acute pancreatitis complicated with acute kidney injury (AKI). After excluding 3,345 records with missing key variables, 6,873 complete datasets were obtained, comprising 5,898 cases in the 60-day survival group and 975 in the death group, indicating significant class imbalance. The propensity score matching (PSM) algorithm was applied to mitigate this imbalance: baseline covariates were used to calculate propensity scores, followed by 1:1 nearest-neighbor matching. A balanced dataset was ultimately generated, comprising 5,849 cases in the survival group and 5,850 cases in the death group. Meanwhile, a series of quality control procedures was performed, including outlier elimination and standardized coding of categorical variables. These preprocessing steps established a high-quality data foundation for subsequent machine learning model construction, and the results of data preprocessing are presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eWe adopted three feature selection algorithms, namely the Boruta algorithm, Random Forest (RF), and cross-validation-based Recursive Feature Elimination (RFE). Multi-algorithm cross-validation was employed to improve the reliability of the selected features, and the Venn diagram distribution of the feature selection results is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eBoruta is a feature selection method based on the random forest algorithm that screens for valid features by comparing the importance of genuine features with that of randomly generated \"shadow features.\" Its advantage is that it can reduce bias in feature selection. As can be seen from the Venn diagram, Boruta identified no uniquely selected features, with all its results overlapping with those of RFE and RF: one feature was screened out in the intersection of Boruta and RFE, 11 features in the intersection of Boruta and RF, and an additional 15 core features were jointly selected by Boruta with both RFE and RF.\u003c/p\u003e \u003cp\u003eRF is a feature selection method based on ensemble learning that screens for high-contribution features by calculating feature importance (e.g., Gini coefficient, out-of-bag (OOB) error). Its advantages lie in the ability to capture nonlinear relationships among features and a strong resistance to overfitting. As shown in the Venn diagram, RF identified 6 uniquely selected features, 11 features in its intersection with Boruta, and participated in screening the 15 core features shared by all three algorithms, thus serving as an intermediate algorithm linking Boruta and RFE.\u003c/p\u003e \u003cp\u003eRFE iteratively removes the \"least important features\" and evaluates the performance of feature subsets through cross-validation to determine the optimal number of features. It is advantageous for its high degree of automation, which can reduce the subjectivity in feature selection. As indicated in the Venn diagram, RFE yielded 4 uniquely selected features, 1 feature in its intersection with Boruta, and participated in the screening of the 15 core features common to all three algorithms.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eModel Construction and Validation\u003c/h3\u003e\n\u003cp\u003eIn this study, we selected 22 machine learning algorithms, including linear, tree-based, ensemble, and neural network models. The algorithm set specifically includes Random Forest、Support Vector Machine、Bayesian GLM、Naive Bayes、K-Nearest Neighbors、Neural Network、Flexible Discriminant Analysis、Gradient Boosting Machine、Classification and Regression Tree、Elastic Net、Logistic Regression、LASSO Regression、Ridge Regression、Linear Discriminant Analysis、Quadratic Discriminant Analysis、Conditional Inference Tree、Partial Least Squares DA、Bagging Decision Trees、XGBoost、AdaBoost、LightGBM、Sparse Linear Discriminant Analysis.\u003c/p\u003e \u003cp\u003eTo conduct an objective evaluation of multiple machine learning models under consistent conditions, the dataset was sampled via a random function and partitioned into a training set and a test set, with 70% of the samples allocated to the training set and 30% to the test set. Each model was subjected to 10-fold cross-validation and parameter optimization. The evaluation metrics included: AUC, AUC_SD, Accuracy, Kappa, Sensitivity, Specificity, Pos_Pred_Value, Neg_Pred_Value, Precision, Recall, and F1.\u003c/p\u003e \u003cp\u003eIn this study, a systematic performance evaluation of the final predictive model was conducted across three key dimensions: discrimination, calibration, and clinical utility. The discriminative ability was quantitatively characterized by the area under the receiver operating characteristic curve (AUC-ROC), which was used to assess the model\u0026rsquo;s efficacy in distinguishing between subjects who experienced the primary outcome (mortality) and those who did not (survival) within 60 days. Calibration ability was visually illustrated by calibration curves that reflected the consistency between the model\u0026rsquo;s predicted probabilities and actual clinical outcomes, and the integrated Brier score was further used to quantitatively evaluate calibration accuracy. Specifically, the integrated Brier score quantifies the deviation between the predicted survival probabilities and actual survival status, thereby comprehensively reflecting the model\u0026rsquo;s calibration efficacy and fitting quality [23].\u003c/p\u003e \u003cp\u003eWith respect to clinical utility, Decision Curve Analysis (DCA) was performed to assess the model's clinical net benefit at different threshold probabilities, to verify its value in practical clinical decision-making scenarios.\u003c/p\u003e\n\u003ch3\u003eModel Interpretation and Deployment\u003c/h3\u003e\n\u003cp\u003eAn explanation method based on SHAP technology was adopted to analyze and discuss the model with the optimal overall performance, aiming to reveal its prediction mechanism and feature contributions [24].\u003c/p\u003e \u003cp\u003eTo elucidate the model\u0026rsquo;s prediction mechanism in depth, this study identified key predictive factors using summary and dependence plots from the SHAP (Shapley Additive exPlanations) method and explored their correlations with patients\u0026rsquo; 60-day survival outcomes. Meanwhile, latent feature interactions were uncovered via SHAP interactive plots. To improve the model\u0026rsquo;s usability and accessibility, the optimized predictive model was ultimately deployed as a web-based, interactive Shiny application that supports personalized survival prediction and customized interpretive explanations. As a lightweight web framework in the R language ecosystem, Shiny enables rapid packaging of R-built models and visualization results into interactive web applications without professional programming knowledge, allowing clinical staff without a technical background to directly operate and use the model through a web browser.\u003c/p\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003eStatistical Analysis\u003c/h2\u003e \u003cp\u003eCategorical variables were described using counts and percentages, while continuous variables were summarized as median (interquartile range, IQR) or mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation (SD), depending on their distribution characteristics. For comparisons of data distribution differences between groups, the chi-square (χ\u0026sup2;) test was used to analyze differences in categorical variables, whereas the Mann-Whitney U test was employed to evaluate differences in continuous variables. All tests were two-tailed, and a P-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was considered statistically significant.\u003c/p\u003e \u003cp\u003eAll statistical analyses and chart generation in this study were performed using R software (version 4.3.1). Core analytical tasks including the plotting of receiver operating characteristic (ROC) curves and the construction and training of machine learning models were completed with the assistance of packages such as plotROC, caret, pROC, and e1071.\u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e illustrates the development process of the explainable machine learning model.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003ePatient Characteristics\u003c/h2\u003e \u003cp\u003eThe total of 11,699 patients with acute pancreatitis complicated by acute kidney injury were included in this study for a detailed retrospective analysis. The study population was stratified into the survival group and the mortality group based on 60-day survival status, with each group accounting for 50% of the total cohort. Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e presents the clinical characteristics of the study population.\u003c/p\u003e \u003cp\u003eUnivariate and multivariate Logistic regression analyses revealed that advanced age, low body mass index (BMI), elevated Sequential Organ Failure Assessment (SOFA) score, increased liver function score, elevated cardiovascular function score, raised renal function score, comorbid intracerebral hemorrhage, prolonged partial thromboplastin time (PTT), elevated red cell distribution width (RDW) and increased white blood cell (WBC) count were independent risk factors for the mortality outcome. In contrast, higher mean corpuscular hemoglobin concentration (MCHC), elevated renal function score, and longer length of hospital stay were identified as potential protective factors. The effect of body weight on prognosis was not statistically significant after multivariate adjustment. The association between heparin duration and prognosis was substantially confounded by other factors and thus requires further verification. Among all the factors, the SOFA score, cardiovascular function score, and comorbid intracerebral hemorrhage exerted the most significant effects on mortality risk, providing empirical support for the identification and clinical intervention of high-risk populations.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eBaseline Characteristics of Patients with Acute Pancreatitis Complicated by Acute Kidney Injury at 60 Days\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCharacteristic\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStatistic\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAlive (N\u0026thinsp;=\u0026thinsp;5849)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eDead(N\u0026thinsp;=\u0026thinsp;5850)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOR (univariable)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eOR (multivariable)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eadmission_age\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e68.7\u0026thinsp;\u0026plusmn;\u0026thinsp;13.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e72.5\u0026thinsp;\u0026plusmn;\u0026thinsp;9.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.03 (1.03\u0026ndash;1.03, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.02 (1.02\u0026ndash;1.03, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBMI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e29.8\u0026thinsp;\u0026plusmn;\u0026thinsp;7.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e28.0\u0026thinsp;\u0026plusmn;\u0026thinsp;5.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.96 (0.95\u0026ndash;0.96, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.95 (0.94\u0026ndash;0.97, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003esofa_score\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1814 (31%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e434 (7.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1321 (22.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e727 (12.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2.30 (2.00-2.64, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.68 (1.42\u0026ndash;1.98, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1025 (17.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1224 (20.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.99 (4.37\u0026ndash;5.70, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.77 (2.34\u0026ndash;3.28, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e613 (10.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1203 (20.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e8.20 (7.11\u0026ndash;9.46, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e3.23 (2.67\u0026ndash;3.91, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e394 (6.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e866 (14.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e9.19 (7.84\u0026ndash;10.77, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.84 (2.28\u0026ndash;3.54, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e258 (4.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e559 (9.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e9.06 (7.56\u0026ndash;10.85, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e4.11 (3.21\u0026ndash;5.27, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e182 (3.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e497 (8.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e11.41 (9.35\u0026ndash;13.93, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e5.36 (4.04\u0026ndash;7.11, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e115 (2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e228 (3.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e8.29 (6.47\u0026ndash;10.61, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e4.96 (3.50\u0026ndash;7.01, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e127 (2.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e112 (1.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e3.69 (2.80\u0026ndash;4.85, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.31 (1.56\u0026ndash;3.42, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eliver\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e5403 (92.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4071 (69.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e220 (3.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1207 (20.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e7.28 (6.27\u0026ndash;8.45, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e4.40 (3.64\u0026ndash;5.32, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e165 (2.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e502 (8.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.04 (3.37\u0026ndash;4.84, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.25 (1.75\u0026ndash;2.90, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e37 (0.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e51 (0.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.83 (1.20\u0026ndash;2.80, p=.005)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e3.33 (1.87\u0026ndash;5.90, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e24 (0.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e19 (0.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.05 (0.57\u0026ndash;1.92, p=.872)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.40 (0.66\u0026ndash;2.99, p=.378)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ecardiovascular\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1648 (28.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e693 (11.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2712 (46.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2106 (36%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.85 (1.66\u0026ndash;2.05, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.26 (1.10\u0026ndash;1.44, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47 (0.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e860 (14.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e43.51 (32.02\u0026ndash;59.13, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e24.48 (17.23\u0026ndash;34.77, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e874 (14.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1015 (17.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2.76 (2.43\u0026ndash;3.13, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.51 (1.26\u0026ndash;1.80, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e568 (9.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1176 (20.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.92 (4.31\u0026ndash;5.63, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.07 (1.70\u0026ndash;2.52, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003erenal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3606 (61.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2125 (36.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1279 (21.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1819 (31.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2.41 (2.21\u0026ndash;2.64, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.64 (1.45\u0026ndash;1.85, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e498 (8.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1237 (21.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.22 (3.75\u0026ndash;4.74, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.86 (2.43\u0026ndash;3.36, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e213 (3.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e533 (9.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.25 (3.59\u0026ndash;5.02, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.95 (2.37\u0026ndash;3.67, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e253 (4.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e136 (2.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.91 (0.74\u0026ndash;1.13, p=.403)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.66 (0.50\u0026ndash;0.87, p=.004)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003elos_hospital\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e16.9\u0026thinsp;\u0026plusmn;\u0026thinsp;16.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e10.3\u0026thinsp;\u0026plusmn;\u0026thinsp;7.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.94 (0.94\u0026ndash;0.95, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.92 (0.91\u0026ndash;0.93, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eduration_pres_heparin\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e9.4\u0026thinsp;\u0026plusmn;\u0026thinsp;14.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e7.4\u0026thinsp;\u0026plusmn;\u0026thinsp;6.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.98 (0.98\u0026ndash;0.99, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.05 (1.04\u0026ndash;1.05, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eweight\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e85.8\u0026thinsp;\u0026plusmn;\u0026thinsp;23.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e81.9\u0026thinsp;\u0026plusmn;\u0026thinsp;17.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.99 (0.99\u0026ndash;0.99, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.00 (1.00-1.01, p=.277)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eco_ICH\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4662 (79.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3229 (55.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1187 (20.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2621 (44.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e3.19 (2.94\u0026ndash;3.46, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.33 (2.08\u0026ndash;2.60, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eptt\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e39.6\u0026thinsp;\u0026plusmn;\u0026thinsp;25.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e52.4\u0026thinsp;\u0026plusmn;\u0026thinsp;32.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.02 (1.01\u0026ndash;1.02, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.01 (1.01\u0026ndash;1.01, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003emchc\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e32.8\u0026thinsp;\u0026plusmn;\u0026thinsp;1.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e32.3\u0026thinsp;\u0026plusmn;\u0026thinsp;1.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.80 (0.78\u0026ndash;0.82, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.93 (0.90\u0026ndash;0.97, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eplatelet\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e211.6\u0026thinsp;\u0026plusmn;\u0026thinsp;102.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e237.5\u0026thinsp;\u0026plusmn;\u0026thinsp;94.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.00 (1.00\u0026ndash;1.00, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.00 (1.00\u0026ndash;1.00, p\u0026lt;.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=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e14.9\u0026thinsp;\u0026plusmn;\u0026thinsp;2.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e15.7\u0026thinsp;\u0026plusmn;\u0026thinsp;1.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.24 (1.22\u0026ndash;1.27, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.18 (1.15\u0026ndash;1.21, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ewbc\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e12.1\u0026thinsp;\u0026plusmn;\u0026thinsp;8.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e14.4\u0026thinsp;\u0026plusmn;\u0026thinsp;6.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.06 (1.05\u0026ndash;1.06, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.02 (1.02\u0026ndash;1.03, p\u0026lt;.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 \u003c/div\u003e\n\u003ch3\u003eFeature Extraction\u003c/h3\u003e\n\u003cp\u003eFigure \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e depicts the process of optimal feature selection combining three algorithms: Boruta, RF, and RFE based on cross-validation. The feature selection process for each algorithm is presented in (Supplementary Fig.\u0026nbsp;1: Feature Selection Process of the Three Algorithms). Ultimately, we selected 15 features common to all three models (SOFA score, liver score, cardiovascular score, renal score, los_hospital, admission_age, duration_pres_aspirin, weight, co_ICH, ptt, mchc, platelet count, rdw, wbc, and BMI as input parameters for machine learning.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eModel Selection and Construction\u003c/h2\u003e \u003cp\u003eUsing the 15 features common to the three models, we constructed 22 machine learning models to predict patients' 60-day survival status. As shown in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, the RF algorithm achieved the optimal predictive performance in the benchmark test, with the values of Accuracy, AUC, Kappa, Sensitivity, Specificity, PPV, NPV, Precision, Recall, and F1 reaching 0.9068, 0.9731, 0.8138, 0.8755, 0.9392, 0.9370, 0.8796, 0.9370, 0.8755, and 0.9052, respectively. The Random Forest model yielded the lowest Brier score of 0.0722, outperforming all other models (Figs.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e and \u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e and Supplementary Table\u0026nbsp;1: Model Brier Score). As illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e, among the 22 machine learning models evaluated in this study, RF was selected as the core analytical framework owing to its excellent and well-balanced performance. RF not only achieved the highest AUC value, demonstrating outstanding discriminative ability in identifying high-risk cases, but also had the smallest standard deviation (SD\u0026thinsp;=\u0026thinsp;0.005), confirming its robust generalization and extremely low performance fluctuation across different data subsets. This finding not only verified that RF was the most reliable model for the clinical prediction task in this study, but also laid a solid foundation for the subsequent SHAP-based interpretive analysis of feature contributions.\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\u003eVarious Performance Metrics of 22 Machine Learning Models\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"12\"\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 \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c12\" colnum=\"12\"\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\u003eAUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAUC_SD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eKappa\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSensitivity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eSpecificity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003ePPV\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eNPV\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c10\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c11\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c12\"\u003e \u003cp\u003eF1\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\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\u003e0.9731\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0053\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9068\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.8138\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8755\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9392\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.9370\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.8796\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.9370\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.8755\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.9052\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSupport Vector Machine\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8487\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0162\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7909\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.5819\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.7757\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.8066\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8055\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.7769\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8055\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.7757\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.7903\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBayesian GLM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8489\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0158\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7889\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.5780\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.7695\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.8089\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8061\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.7727\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8061\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.7695\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.7874\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNaive Bayes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8932\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0137\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8191\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.6391\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.7386\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9021\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8863\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.7698\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8863\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.7386\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8058\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eK-Nearest Neighbors\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8591\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0102\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7823\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.5643\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8020\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.7620\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.7768\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.7885\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.7768\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.8020\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.7892\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNeural Network\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8551\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0212\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7735\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.5474\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.7398\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.8083\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.7994\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.7505\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.7994\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.7398\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.7684\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFlexible Discriminant Analysis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8849\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0123\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8071\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.6143\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.7964\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.8182\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8189\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.7956\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8189\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.7964\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8075\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGradient Boosting Machine\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9166\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0104\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8325\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.6652\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8059\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.8599\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8559\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.8110\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8559\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.8059\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8302\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eClassification and Regression Tree\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.7242\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0245\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7120\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.4241\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.6977\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.7267\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.7249\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.6996\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.7249\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.6977\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.7111\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eElastic Net\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8490\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0160\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7892\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.5785\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.7717\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.8072\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8051\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.7740\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8051\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.7717\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.7881\u003c/p\u003e \u003c/td\u003e \u003c/tr\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\u003e0.8488\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0158\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7889\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.5780\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.7695\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.8089\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8061\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.7727\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8061\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.7695\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.7874\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLASSO Regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8490\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0131\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7889\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.5780\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.7717\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.8066\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8047\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.7739\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8047\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.7717\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.7879\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRidge Regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8484\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0175\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7917\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.5836\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.7762\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.8078\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8065\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.7776\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8065\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.7762\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.7911\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLinear Discriminant Analysis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8464\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0181\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7863\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.5728\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.7712\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.8020\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8008\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.7724\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8008\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.7712\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.7857\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eQuadratic Discriminant Analysis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8498\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0176\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7761\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.5520\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.7835\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.7684\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.7774\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.7747\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.7774\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.7835\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.7804\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eConditional Inference Tree\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8843\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0144\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8017\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.6036\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.7835\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.8205\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8184\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.7859\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8184\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.7835\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8006\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePartial Least Squares DA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8467\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0187\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7892\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.5784\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.7830\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.7956\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.7982\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.7802\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.7982\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.7830\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.7905\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBagging Decision Trees\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9550\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0092\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8934\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7870\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8727\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9149\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.9137\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.8744\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.9137\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.8727\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8927\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\u003e0.9626\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0060\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8997\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7995\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8800\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9201\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.9192\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.8813\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.9192\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.8800\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8991\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\u003e0.9186\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0110\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8311\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.6623\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8059\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.8570\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8533\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.8105\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8533\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.8059\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8290\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\u003e0.9403\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0135\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8624\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7251\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8222\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9039\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8983\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.8312\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8983\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.8222\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8586\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSparse Linear Discriminant Analysis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8220\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0169\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.5100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.0041\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.0000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.0041\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.5090\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e1.0000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.5090\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e1.0000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.6746\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\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe calibration curves of all models were close to the ideal dashed line, with high fitting for most models. The group-specific error bars (vertical bars) were generally short, indicating small fluctuations in the proportion of actual events within each group. The models' predicted probabilities could well reflect the actual risk of event occurrence, thus providing reliable probabilistic references for clinical decision-making (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eResults of the DCA showed that RF and other machine learning models included in the study all demonstrated positive clinical application value across different probability threshold intervals. Owing to its robust performance, the RF model can serve as a fundamental tool for routine clinical decision-making, whereas models such as LGBM and XGB are better suited for scenarios that demand high precision. These results demonstrated that different machine learning models can provide multidimensional support for clinical decision-making, facilitating the implementation of individualized risk assessment and the formulation of targeted intervention strategies (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eModel Interpretation and Deployment\u003c/h2\u003e \u003cp\u003eThe SHAP method was employed to interpret the model by ranking the importance of candidate features in individual predictions. As illustrated in Figs.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e and \u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e, admission_age, co_ICH, liver score, and cardiovascular status were the most impactful predictive factors; their high values (yellow dots) consistently elevated the predicted risk, whereas duration_pres_heparin exhibited a potential protective effect with negative SHAP values. This visualization enhanced the transparency of the \"black-box\" model by quantifying the magnitude and direction of each feature\u0026rsquo;s contribution, and verified the clinical relevance of key predictive factors, which were highly consistent with the existing domain knowledge.\u003c/p\u003e \u003cp\u003eAt the global level, the bee plot revealed that admission_age, co_ICH, liver function score, and cardiovascular status (cardiovascular) were the most impactful predictive factors; their high values generally increased the predicted risk of mortality, whereas the duration_pres_heparin exhibited a potential protective effect. A detailed analysis of Sample 78 demonstrated that protective factors, particularly a SOFA score of 2 (SHAP value = -0.1) and a white blood cell count of 7.1 (wbc\u0026thinsp;=\u0026thinsp;7.1, SHAP value = -0.09), predominated in the prediction, reducing the mortality probability from a baseline of 0.487 to 0.22. The countervailing effects of risk-increasing factors (e.g., red cell distribution width\u0026thinsp;=\u0026thinsp;16.2 (rdw\u0026thinsp;=\u0026thinsp;16.2) and length of hospital stay\u0026thinsp;=\u0026thinsp;6.17 (los_hospital\u0026thinsp;=\u0026thinsp;6.17)) were relatively weak. This also illustrated how the model generates individualized risk estimates by aggregating the contributions of each feature, and this process was highly consistent with the patterns of feature importance observed at the global level.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig12\" class=\"InternalRef\"\u003e12\u003c/span\u003e illustrates the non-linear relationships between key features and predicted mortality risk, with the red trend lines highlighting the overall direction of each feature\u0026rsquo;s effect. For ptt and rdw, SHAP values rise with increasing feature values, peak at moderate levels, and then decline, indicating moderate values exert the strongest enhancing effect on mortality risk, whereas the impact of extreme values is instead attenuated. los_hospital exhibits an initial negative correlation (longer hospital stays reduce mortality risk), with the trend reversing at approximately 100 days, reflecting a complex time-dependent relationship. The SOFA score and wbc also show an initial positive impact that diminishes at high values. In contrast, the duration_pres_heparin follows a U-shaped pattern: both short-term and long-term heparin use are associated with reduced mortality risk, while moderate-duration use exerts a neutral to positive impact on risk.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eWe integrated the final RF model into an interactive Shiny-based web application that provides individualized survival predictions and interpretations, as well as global model interpretation, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig13\" class=\"InternalRef\"\u003e13\u003c/span\u003e. This web application is accessible via the URL: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://medicalpredictor.shinyapps.io/Online_Prediction_of_60-Day_Survival/\u003c/span\u003e\u003cspan address=\"https://medicalpredictor.shinyapps.io/Online_Prediction_of_60-Day_Survival/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eFocusing on the high-risk clinical scenario of acute pancreatitis complicated by acute kidney injury, this system takes the Random Forest algorithm as its core and constructs an integrated workflow encompassing parameter entry, one-click prediction, and result visualization. It can not only accurately output the 60-day survival probability of patients but also enable clinicians to intuitively understand the key influencing factors underlying the predictions through SHAP feature importance analysis. This not only improves the efficiency of clinical decision-making but also enhances the interpretability and credibility of the model, thereby providing strong support for the formulation of personalized treatment plans.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eAcute pancreatitis complicated by acute kidney injury (AP-AKI) is a high-risk disease complex in the field of critical care medicine, and its prognostic assessment has long been a key focus of clinical research. Based on large-sample data from the MIMIC-IV v3.1 database, this study identified the core predictive factors for 60-day survival in AP-AKI patients through multi-algorithmic feature selection and machine learning model construction, and developed an interpretable and generalizable clinical tool, thereby providing novel, evidence-based guidance for the disease management of AP-AKI.\u003c/p\u003e \u003cp\u003eMultivariate Logistic regression analysis in this study revealed that admission age was an independent risk factor for 60-day mortality in AP-AKI patients (OR\u0026thinsp;=\u0026thinsp;1.02, 95% confidence interval (95% CI: 1.02\u0026ndash;1.03, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001). The mean age of patients in the mortality group (72.5\u0026thinsp;\u0026plusmn;\u0026thinsp;9.3 years) was significantly higher than that in the survival group (68.7\u0026thinsp;\u0026plusmn;\u0026thinsp;13.1 years), consistent with previous studies [25,26]. Elderly patients exhibit impaired organ reserve function and reduced tolerance to inflammatory stress, rendering them more susceptible to sequential multi-organ dysfunction injury.\u003c/p\u003e \u003cp\u003eLow BMI was also associated with an increased risk of mortality (OR\u0026thinsp;=\u0026thinsp;0.95, 95% CI: 0.94\u0026ndash;0.97, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001), suggesting that nutritional status may regulate disease prognosis by modulating immune function, tissue repair capacity, and related pathways. This finding provides insights for the implementation of clinical nutritional support interventions. Furthermore, this study balanced baseline distributions between the survival and mortality groups using propensity score matching (PSM), ultimately yielding a well-balanced cohort of 11,699 patients (5,849 and 5,850 in each group, respectively). This approach effectively reduced biases arising from confounding demographic characteristics and enhanced the reliability of the study conclusions.\u003c/p\u003e \u003cp\u003eAKI is one of the major complications in patients with AP. Its development is influenced by multiple factors and directly leads to organ damage and multiple organ failure [27].\u003c/p\u003e \u003cp\u003eOrgan function-related scores were identified as the core prognostic features in this study. As a key indicator reflecting the degree of multi-organ dysfunction, the SOFA score exhibited a significant increasing trend in its risk effect with rising scores: at a SOFA score of 8, the odds ratio (OR) for mortality risk reached as high as 5.36 (95% CI: 4.04\u0026ndash;7.11, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001), and the score also emerged as an important variable influencing prediction results in SHAP analysis. This finding highlights the central role of multi-organ function protection in the treatment of AP-AKI, which is highly consistent with the pathophysiological mechanism of acute pancreatitis. The systemic inflammatory response syndrome (SIRS) induced by pancreatic inflammation can cause the sequential impairment of liver, cardiovascular, renal, and other organ functions through the diffusion of inflammatory mediators.\u003c/p\u003e \u003cp\u003eElevated liver, cardiovascular, and renal function scores were all closely associated with an increased risk of mortality. Specifically, at a cardiovascular function score of 2, the odds ratio (OR) for mortality risk reached 24.48 (95% confidence interval (95% CI: 17.23\u0026ndash;34.77, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001), and for patients with Stage Ⅲ renal function score, the OR was 2.95 (95% CI: 2.37\u0026ndash;3.67, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001). These findings indicate that cardiovascular system failure and moderate to severe renal injury are strong predictive markers for poor prognosis in AP-AKI patients. In terms of biochemical indicators, ptt (OR\u0026thinsp;=\u0026thinsp;1.01, 95% CI: 1.01\u0026ndash;1.01, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001), elevated rdw (OR\u0026thinsp;=\u0026thinsp;1.18, 95% CI: 1.15\u0026ndash;1.21, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001), and increased wbc count (OR\u0026thinsp;=\u0026thinsp;1.02, 95% CI: 1.02\u0026ndash;1.03, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001) were all identified as independent risk factors, respectively reflecting coagulation dysfunction, hematopoietic system stress, and systemic inflammatory activation. The degree of abnormality of these indicators can serve as key entry points for clinical disease monitoring.\u003c/p\u003e \u003cp\u003eThe staging of renal function injury is directly correlated with prognosis [28,29]. The OR for mortality risk increased progressively from 1.64 to 2.95 as the renal function score advanced from Stage Ⅰ to Stage Ⅲ, demonstrating that the severity of renal injury is positively associated with poorer patient prognosis. The progressive deterioration of renal function exacerbates systemic metabolic disorders and triggers sequential cascades of multi-organ dysfunction. Additionally, patients with co_ICH had a markedly increased mortality risk (OR\u0026thinsp;=\u0026thinsp;2.33, 95% CI: 2.08\u0026ndash;2.60, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001). This may be attributed to elevated intracranial pressure and cerebral hypoperfusion induced by intracerebral hemorrhage, as well as the synergistic effect of systemic inflammation caused by the coexistence of intracerebral hemorrhage and AP-AKI. Such patients require more intensive multidisciplinary collaborative management.\u003c/p\u003e \u003cp\u003eAmong treatment-related factors, length of hospital stay (OR\u0026thinsp;=\u0026thinsp;0.92, 95% CI: 0.91\u0026ndash;0.93, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001) was identified as a potential protective factor, with longer hospital stays at the initial stage possibly reflecting the effectiveness of clinical interventions. In contrast, the association between heparin duration and prognosis was substantially confounded by other factors (multivariate OR\u0026thinsp;=\u0026thinsp;1.05, 95% CI: 1.04\u0026ndash;1.05, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and thus requires further verification in prospective studies. SHAP analysis revealed that the duration of heparin use was associated with negative SHAP values, suggesting that heparin may reduce the risk of thrombus-related complications (e.g., renal vein thrombosis) through its anticoagulant effect. This hypothesis provides a direction for optimizing treatment strategies in AP-AKI patients.\u003c/p\u003e \u003cp\u003eAmong the 22 machine learning models constructed in this study, the RF model demonstrated the optimal performance, with an AUC of 0.9731, an Accuracy of 0.9068, a Sensitivity of 0.8755, a Specificity of 0.9392, and the lowest Brier score (0.0722), exhibiting excellent discriminative and calibration performance. Compared with traditional linear models (e.g., Logistic regression with an AUC of 0.8488), the RF model can more accurately capture the non-linear relationships and interactions between features, which is fully verified by SHAP analysis: ptt and rdw show a threshold effect with mortality risk (the highest risk at moderate levels), length of hospital stay presents a time-dependent association (the trend reversed around 100 days), and the duration of heparin use follows a U-shaped pattern (both short-term and long-term use may reduce the risk of mortality). The revelation of these non-linear relationships breaks through the limitations of the linear assumption inherent in traditional prognostic models, providing a more clinically relevant basis for individualized risk assessment.\u003c/p\u003e \u003cp\u003eThe interactive Shiny Web application developed using the Random Forest (RF) model (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://medicalpredictor.shinyapps.io/Online_Prediction_of_60-Day_Survival/\u003c/span\u003e\u003cspan address=\"https://medicalpredictor.shinyapps.io/Online_Prediction_of_60-Day_Survival/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) provides an integrated workflow from parameter entry to prognostic visualization. Without professional programming knowledge, clinicians can quickly obtain the 60-day survival probability and interpretations of key influencing factors by entering 15 core patient features (including SOFA score, liver and renal function, admission age, etc.). This has significantly enhanced the model's clinical utility, providing a convenient tool for early risk stratification and personalized intervention in AP-AKI patients.\u003c/p\u003e \u003cp\u003eAlthough valuable results have been achieved in this study, certain limitations remain. First, the study adopted a retrospective design with data from a single database (MIMIC-IV v3.1). Derived from standardized clinical records, the data ensured a large sample size and completeness but may be subject to selection bias. The model's generalizability needs further verification across heterogeneous cohorts, including non-Western populations and community hospitals. Second, the database failed to fully capture certain clinical details, including specific treatment regimens (timing and dosage of renal replacement therapy), nutritional status assessment, and patients' long-term quality of life. These factors may affect prognosis and should be incorporated into subsequent studies. Third, this study focused on short-term 60-day survival prediction, while long-term outcomes of AP-AKI patients (e.g., progression to chronic kidney disease, readmission risk) have not been addressed. Future research can expand the model's outcome coverage. Finally, despite balancing baseline confounders through propensity score matching (PSM), residual confounding inherent to retrospective studies may persist. The validity and interventional value of the model need further verification via prospective studies.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis study developed and validated an interpretable Random Forest (RF) model to predict 60-day survival in patients with acute pancreatitis complicated by acute kidney injury (AP-AKI). The model exhibits excellent discriminative efficacy, calibration accuracy, and clinical utility. The interactive web application facilitates the integration of the model into clinical workflows, providing clinicians with a convenient tool for early risk stratification and a reference for personalized interventions, thereby supporting the optimization of diagnosis and treatment decisions. Future research should focus on external validation in multicenter cohorts to further enhance the model\u0026rsquo;s generalizability, incorporate more clinically relevant features (e.g., specific treatment strategies, nutritional status) to improve model performance, and clarify the model\u0026rsquo;s role in improving patients\u0026rsquo; final outcomes through prospective studies.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe would like to acknowledge the researchers and participants who contributed to the MIMIC-IV database.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research was funded by the Anhui Province Teaching Research Program (Granted No. 2023jyxm0347).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no conflict of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was based on data extracted from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database—a publicly accessible, de-identified critical care database. Ethical approval for the establishment and ongoing maintenance of MIMIC-IV was obtained from the Institutional Review Boards (IRBs) of both the Massachusetts Institute of Technology (MIT) and Beth Israel Deaconess Medical Center (BIDMC), and informed consent was waived due to the retrospective, fully anonymized nature of the data.\u003c/p\u003e\n\u003cp\u003eFor this secondary analysis of de-identified data, no further ethical approval was deemed necessary. The study was designed and conducted in strict adherence to the ethical standards of the institutional and/or national research committees, as well as the 1964 Declaration of Helsinki and its subsequent amendments or equivalent ethical guidelines.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthorship contribution\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData Analysis: T.S., and W.W; Writing Original Draft: T.S., and WW.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data used in this study are publicly available from the MIMIC-IV database. Access to these data requires completion of mandatory training and approval via PhysioNet (https://physionet.org/).\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eKumar R, Pahwa N, Jain N. Acute kidney injury in severe acute pancreatitis: an experience from a tertiary care center. Saudi J Kidney Dis Transpl. 2015;26(1):56-60. doi:10.4103/1319-2442.148734\u003c/li\u003e\n \u003cli\u003eYadav D, Lowenfels AB. The epidemiology of pancreatitis and pancreatic cancer. Gastroenterology. 2013;144(6):1252-1261. doi:10.1053/j.gastro.2013.01.068\u003c/li\u003e\n \u003cli\u003eBollen TL, van Santvoort HC, Besselink MG, et al. The Atlanta Classification of acute pancreatitis revisited. Br J Surg. 2008;95(1):6-21. doi:10.1002/bjs.6010\u003c/li\u003e\n \u003cli\u003eLiu B, Zhang X, Li J, et al. The Global, Regional, and National Burden of Pancreatitis in 204 Countries and Territories, 1990-2021: A Systematic Analysis for the Global Burden of Disease Study 2021. Dig Dis Sci. 2025;70(7):2328-2339. doi:10.1007/s10620-025-08996-y\u003c/li\u003e\n \u003cli\u003eZhou J, Li Y, Tang Y, et al. Effect of acute kidney injury on mortality and hospital stay in patient with severe acute pancreatitis. Nephrology (Carlton). 2015;20(7):485-491. doi:10.1111/nep.12439\u003c/li\u003e\n \u003cli\u003eSelvanathan DK, Johnson PG, Thanikachalam DK, Rajendran P, Gopalakrishnan N. Acute Kidney Injury Complicating Severe Acute Pancreatitis: Clinical Profile and Factors Predicting Mortality. Indian J Nephrol. 2022;32(5):460-466. doi:10.4103/ijn.IJN_476_20\u003c/li\u003e\n \u003cli\u003eKes P, Vucicević Z, Ratković-Gusić I, Fotivec A. Acute renal failure complicating severe acute pancreatitis. Ren Fail. 1996;18(4):621-628. doi:10.3109/08860229609047686\u003c/li\u003e\n \u003cli\u003eNaqvi R. Acute Kidney Injury in association with Acute Pancreatitis. Pak J Med Sci. 2018;34(3):606-609. doi:10.12669/pjms.343.14953\u003c/li\u003e\n \u003cli\u003eNassar TI, Qunibi WY. AKI Associated with Acute Pancreatitis. Clin J Am Soc Nephrol. 2019;14(7):1106-1115. doi:10.2215/CJN.13191118\u003c/li\u003e\n \u003cli\u003eBagshaw SM, George C, Dinu I, Bellomo R. A multi-centre evaluation of the RIFLE criteria for early acute kidney injury in critically ill patients. Nephrol Dial Transplant. 2008;23(4):1203-1210. doi:10.1093/ndt/gfm744\u003c/li\u003e\n \u003cli\u003eSaimon, S. I., Islam, I., Abir, S. I., Sultana, N., Hossain, M. S., \u0026amp; Al Shiam, S. A. Advancing Neurological Disease Prediction through Machine Learning Techniques. Journal of Computer Science and Technology Studies, 2025;7(1), 139-156. doi:10.32996/jcsts.2025.7.1.11\u003c/li\u003e\n \u003cli\u003eEjiyi CJ, Qin Z, Nneji GU, et al. Enhanced Cardiovascular Disease Prediction Modelling using Machine Learning Techniques: A Focus on CardioVitalnet. Network. 2025;36(3):716-748. doi:10.1080/0954898X.2024.2343341\u003c/li\u003e\n \u003cli\u003eSanju, P., Ahmed, N. S. S., Ramachandran, P., Sajid, P. M., \u0026amp; Jayanthi, R. (2025). Enhancing thyroid disease prediction and comorbidity management through advanced machine learning frameworks. Clinical eHealth,2025; 8:7-16. doi:10.1016/j.ceh.2025.01.002\u003c/li\u003e\n \u003cli\u003eQi X, Wang S, Fang C, Jia J, Lin L, Yuan T. Machine learning and SHAP value interpretation for predicting comorbidity of cardiovascular disease and cancer with dietary antioxidants. Redox Biol. 2025;79:103470. doi:10.1016/j.redox.2024.103470\u003c/li\u003e\n \u003cli\u003eThatha VN, Chalichalamala S, Pamula U, et al. Optimized machine learning mechanism for big data healthcare system to predict disease risk factor. Sci Rep. 2025;15(1):14327. Published 2025 Apr 24. doi:10.1038/s41598-025-98721-6\u003c/li\u003e\n \u003cli\u003eTan Z, Li G, Zheng Y, et al. Advances in the clinical application of machine learning in acute pancreatitis: a review. Front Med (Lausanne). 2025;11:1487271. Published 2025 Jan 7. doi:10.3389/fmed.2024.1487271\u003c/li\u003e\n \u003cli\u003eShi N, Lan L, Luo J, et al. Predicting the Need for Therapeutic Intervention and Mortality in Acute Pancreatitis: A Two-Center International Study Using Machine Learning. J Pers Med. 2022;12(4):616. Published 2022 Apr 11. doi:10.3390/jpm12040616\u003c/li\u003e\n \u003cli\u003eKiss S, Pint\u0026eacute;r J, Molontay R, et al. Early prediction of acute necrotizing pancreatitis by artificial intelligence: a prospective cohort-analysis of 2387 cases. Sci Rep. 2022;12(1):7827. Published 2022 May 12. doi:10.1038/s41598-022-11517-w\u003c/li\u003e\n \u003cli\u003eLan L, Guo Q, Zhang Z, et al. Classification of Infected Necrotizing Pancreatitis for Surgery Within or Beyond 4 Weeks Using Machine Learning. Front Bioeng Biotechnol. 2020;8:541. Published 2020 Jun 4. doi:10.3389/fbioe.2020.00541\u003c/li\u003e\n \u003cli\u003eLuo Z, Shi J, Fang Y, et al. Development and evaluation of machine learning models and nomogram for the prediction of severe acute pancreatitis. J Gastroenterol Hepatol. 2023;38(3):468-475. doi:10.1111/jgh.16125\u003c/li\u003e\n \u003cli\u003eYuan L, Ji M, Wang S, et al. Machine learning model identifies aggressive acute pancreatitis within 48 h of admission: a large retrospective study. BMC Med Inform Decis Mak. 2022;22(1):312. Published 2022 Nov 29. doi:10.1186/s12911-022-02066-3\u003c/li\u003e\n \u003cli\u003eJohnson, A., Bulgarelli, L., Pollard, T., Gow, B., Moody, B., Horng, S., Celi, L. A., \u0026amp; Mark, R. MIMIC-IV (version 3.1). PhysioNet. 2024, RRID:SCR_007345. https://doi.org/10.13026/kpb9-mt58\u003c/li\u003e\n \u003cli\u003ePark SY, Park JE, Kim H, Park SH. Review of Statistical Methods for Evaluating the Performance of Survival or Other Time-to-Event Prediction Models (from Conventional to Deep Learning Approaches). Korean J Radiol. 2021;22(10):1697-1707. doi:10.3348/kjr.2021.0223\u003c/li\u003e\n \u003cli\u003eLundberg SM, Erion G, Chen H, et al. From Local Explanations to Global Understanding with Explainable AI for Trees. Nat Mach Intell. 2020;2(1):56-67. doi:10.1038/s42256-019-0138-9\u003c/li\u003e\n \u003cli\u003eLi H, Qian Z, Liu Z, Liu X, Han X, Kang H. Risk factors and outcome of acute renal failure in patients with severe acute pancreatitis. J Crit Care. 2010;25(2):225-229. doi:10.1016/j.jcrc.2009.07.009\u003c/li\u003e\n \u003cli\u003eMederos MA, Reber HA, Girgis MD. Acute Pancreatitis: A Review. JAMA. 2021;325(4):382-390. doi:10.1001/jama.2020.20317\u003c/li\u003e\n \u003cli\u003eMathur P, Vaishnav S. Pathophysiology of acute kidney injury in severe acute pancreatitis-An overview Gastroenterol Hepatol Open Access. 2019;10:242\u0026ndash;5\u003c/li\u003e\n \u003cli\u003eZhou J, Li Y, Tang Y, et al. Effect of acute kidney injury on mortality and hospital stay in patient with severe acute pancreatitis. Nephrology (Carlton). 2015;20(7):485-491. doi:10.1111/nep.12439\u003c/li\u003e\n \u003cli\u003eKong L, Santiago N, Han TQ, Zhang SD. Clinical characteristics and prognostic factors of severe acute pancreatitis. World J Gastroenterol. 2004;10(22):3336-3338. doi:10.3748/wjg.v10.i22.3336\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"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":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Acute Kidney Injury Prognosis, Machine Learning, 60-Day Survival Prediction, Clinical Prognostic Factors, Shiny Application","lastPublishedDoi":"10.21203/rs.3.rs-8896948/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8896948/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eAcute kidney injury (AKI) is a severe complication of severe acute pancreatitis (SAP) with an extremely poor prognosis. Our study aimed to develop and validate an interpretable machine learning (ML) model to predict 60-day survival in patients with acute pancreatitis (AP) complicated by AKI, and to identify key prognostic factors to support clinical decision-making.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eThis was a retrospective cohort study, with data extracted from the MIMIC-IV v3.1 database (released in October 2024). The inclusion criteria were as follows: patients aged 18 years or older with acute pancreatitis complicated with acute kidney injury confirmed by ICD diagnostic codes, a clinical data completeness rate of \u0026ge;\u0026thinsp;80%, and a follow-up duration of \u0026ge;\u0026thinsp;60 days or a definitive in-hospital death record within 60 days. Propensity score matching (PSM) was applied to address the class imbalance between the survival and death groups. The optimal features were screened from 31 candidate variables using three feature selection algorithms: the Boruta algorithm, random forest (RF), and recursive feature elimination with cross-validation (RFECV). A total of 22 machine learning models were constructed, and their predictive performance was evaluated based on the following metrics: area under the receiver operating characteristic curve (AUC), accuracy, Kappa coefficient, sensitivity, specificity, and Brier score. The Shapley Additive exPlanations (SHAP) method was adopted to interpret the optimal model, which elucidated feature importance, nonlinear relationships, and inter-feature interactions. Ultimately, an interactive web application was developed to facilitate the clinical application and popularization of the model.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eA total of 11,699 eligible patients were included (5,849 in the survival group and 5,850 in the death group after propensity score matching (PSM). Fifteen core features were screened for model construction, including Sequential Organ Failure Assessment (SOFA) score, liver function, cardiovascular function, renal function, length of hospital stay (LoS), admission age, and comorbidity of intracerebral hemorrhage (co_ICH). The random forest (RF) model exhibited the best performance, with an area under the receiver operating characteristic curve (AUC) of 0.9731, accuracy of 0.9068, sensitivity of 0.8755, specificity of 0.9392, and Brier score of 0.0722. SHAP analysis revealed that admission age, co_ICH, liver function, and cardiovascular function were the most important predictors of 60-day mortality. Nonlinear relationships were observed between key features (e.g., activated partial thromboplastin time [PTT], red cell distribution width [RDW], and length of hospital stay) and survival outcomes, along with threshold effects and U-shaped associations. An interactive web application (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://medicalpredictor.shinyapps.io/Online_Prediction_of_60-Day_Survival/\u003c/span\u003e\u003cspan address=\"https://medicalpredictor.shinyapps.io/Online_Prediction_of_60-Day_Survival/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) has been successfully deployed for individualized risk assessment.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eThe random forest model developed in this study exhibits excellent performance and interpretability in predicting 60-day survival in patients with acute pancreatitis complicated by acute kidney injury. Key prognostic factors identified via SHAP analysis provide valuable clinical references, while the web application enhances the model's practical utility. This tool can assist clinicians in conducting early risk stratification and formulating personalized intervention strategies, thereby improving patient outcomes.\u003c/p\u003e","manuscriptTitle":"Development and Validation of a Machine Learning-Based Risk Prediction Model for 60-Day Survival Status in Patients with Acute Pancreatitis Complicated with Acute Kidney Injury","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-24 16:41:44","doi":"10.21203/rs.3.rs-8896948/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"afa8d4dc-51b9-41b6-9356-dbbf23800444","owner":[],"postedDate":"February 24th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-03-17T08:32:06+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-24 16:41:44","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8896948","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8896948","identity":"rs-8896948","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}