Explainable and Visualizable Machine Learning Model Development and Validation for 5-Year Postoperative Survival Prediction in Prostate Cancer Patients Aged ≥ 65 Years

Explainable and Visualizable Machine Learning Model Development and Validation for 5-Year Postoperative Survival Prediction in Prostate Cancer Patients Aged ≥ 65 Years | 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 Explainable and Visualizable Machine Learning Model Development and Validation for 5-Year Postoperative Survival Prediction in Prostate Cancer Patients Aged ≥ 65 Years Huaying Chen, TONGPING SHEN This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8665949/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 11 You are reading this latest preprint version Abstract Background Machine Learning (ML) models have achieved outstanding performance in predicting post-surgical survival. However, the “black-box” nature of ML models restricts their clinical application. This study aims to develop and validate a clinically feasible, machine learning-based online prediction system for predicting the 5-year survival status of patients with prostate cancer (PCa) after surgery. Methods This study conducted a retrospective analysis of clinical data from 300 elderly patients with PCa aged ≥ 65 years. LASSO regression, random forest (RF), and recursive feature elimination (RFE) were employed to screen for clinical parameters. Subsequently, the cohort was split into a training set and a test set at a 7:3 ratio, and 25 machine learning approaches were utilized for comparative assessment to determine the optimal model. Calibration curves were applied to evaluate the performance of each model, while decision curve analysis (DCA) was adopted to assess their clinical usefulness. In addition, SHAP (Shapley Additive exPlanations) values were used to interpret the model features. Finally, the Shiny framework was employed to develop an online prediction system. Results The intersection of the three feature selection algorithms identified 13 clinical parameters: Age, ALB, BUN, CRE, HB, PLT, PT, PSA, GS, PNI, PSM, pT and pN. Comparing 25 machine learning algorithms, LightGBM gave the best results. Its performance metrics were: Accuracy 0.9328, Sensitivity 0.9074, Specificity 0.9538, Positive Predictive Value (PPV) 0.9418, Negative Predictive Value (NPV) 0.9118, AUC 0.9778, Recall 0.9074, F1 Score 0.9143. SHAP values revealed the contribution of each feature in the LightGBM model. Conclusion This study successfully developed a 5-year survival prediction model for prostate cancer patients after surgery, providing guidance for clinical decision-making, optimizing treatment efficacy, and improving quality of life. Prostate Cancer machine learning elderly people SHAP Shiny Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Introduction PCa is the most common malignant tumor of the urinary system and the second most common malignant tumor in men, ranking fifth in terms of mortality rate[ 1 , 2 , 3 ]. GBD 2023 data show that the global death toll from prostate cancer was 21,461,648 in 1990 and 47,295,326 in 2023, indicating that it has become an important public health issue[ 4 ]. PCa is a complex disease that poses significant challenges to both patients and clinicians. High heterogeneity is one of its core characteristics. Within the same patient's prostate tissue, different lesions may exhibit distinct biological behaviors (e.g., aggressive vs. indolent). Among different patients, the mutation profiles, epigenetic modifications, and tumor molecular subtypes also show significant differences [ 5 ]. This heterogeneity leads to significant uncertainty in the natural course of PCa. While the majority of PCa presents as slowly progressive indolent tumors, a minority of patients may develop into highly malignant diseases characterized by strong invasiveness and distant metastasis. Treatment decisions are primarily based on a comprehensive assessment of the patient’s prognostic risk, including clinical-pathological features and individual patient health conditions. Clinical pathological features include prostate-specific antigen (PSA) levels, tumor Gleason scores, and TNM staging. Individual patient health conditions encompass sociodemographic and clinical characteristics such as age, overall health status (e.g., presence of other chronic diseases), and life expectancy [ 6 , 7 ]. Additionally, in recent years, notable advances have been achieved in the diagnosis and treatment of PCa, bringing substantial innovations to clinical practice. Considering the marked variations in disease progression stages, individual physiological conditions, and quality-of-life expectations among PCa patients, both clinical research and practical experience have verified that surgical treatment can effectively prevent or delay tumor recurrence and progression, thereby significantly enhancing patients' disease-free survival and overall survival rates. In addition to surgical treatment, clinical practice offers a variety of non-surgical options, including external beam radiotherapy, brachytherapy, chemotherapy, hormone therapy, and targeted drug therapy, providing patients with a diverse range of treatment choices [ 8 ]. It is crucial to formulate an individualized optimal treatment plan for each patient. This decision directly impacts disease control efficacy and patient prognosis, while also helping to optimize medical resource allocation and alleviate patients' daily and financial care burdens [ 9 ]. Prostate cancer has an extremely high incidence rate among elderly men. The incidence and mortality rates of this disease are strongly age-related, with the average age at diagnosis for most patients centered around 65 years old. The formation of this age-related difference may be attributed to the joint effects of multiple factors, including genetic susceptibility, environmental exposure, and socioeconomic factors [ 10 ]. In 2015, 72,000 new prostate cancer cases and roughly 30,700 deaths were documented in China. While the incidence of prostate cancer in China is substantially lower than that in European and American countries, it has exhibited an upward tendency in recent years, with a faster growth rate compared to developed nations in Europe and America[ 11 ]. By 2022, the number of new cases and deaths from prostate cancer in China is projected to increase to 134,200 and 47,500 cases, respectively[ 12 ]. China still confronts numerous challenges in the early diagnosis and standardized treatment of prostate cancer, exhibiting a notable disparity compared with Western developed countries. Taking the United States as an illustration, nearly 91% of newly diagnosed prostate cancer cases in the country are clinically localized, with radical prostatectomy or radical radiotherapy serving as the first-line treatment. Patients receiving standardized therapy tend to have a favorable prognosis, with a 5-year overall survival rate approaching 100%. In contrast, only 30% of newly diagnosed cases in China are clinically localized prostate cancer, while the remaining 70% are patients with locally advanced or metastatic disease. These patients are ineligible for local curative treatment, resulting in a relatively poor prognosis [ 13 ]. At present, treatment guidelines for PCa primarily rely on traditional linear models—such as survival analysis and Cox proportional hazards models—for risk stratification. Nevertheless, owing to the intricate nonlinear interactions among various prognostic factors in PCa biology, relying exclusively on these linear approaches may not be adequate to precisely forecast individual survival outcomes. As artificial intelligence (AI) and ML technologies continue to advance, enabling the processing of massive datasets in relatively short periods, their applications in the medical field are increasingly widespread. Machine learning can be applied to various areas, including drug development, gene expression profiling, biomarker research in multi-omics construction and analysis, and digital pathology slide analysis [ 14 , 15 ]. In the era of personalized medicine, researchers are developing machine learning algorithms based on individual data to achieve precise predictions of survival outcomes for prostate cancer patients. Machine learning technology has enormous potential for designing personalized treatment plans [ 16 , 17 , 18 ]. By systematically integrating individual factors such as age, health status, and lifestyle, machine learning models can help researchers capture subtle differences in tumor characteristics among prostate cancer patients, thereby providing robust technical support for the formulation of precise, personalized treatment strategies. Nevertheless, this field still confronts considerable challenges. First, the variables utilized in prior studies are lacking in dimensionality and scope [ 20 ], which undermines the predictive performance of the models. Consequently, this affects the reliability of experimental findings and the comprehension of complex treatment decision models [ 21 ]. Second, the small sample sizes and concentrated data sources from single research centers have led to generally diminished representativeness and generalizability of previous research. Finally, the machine learning models currently employed often depend on a single or limited set of model types, with a dearth of comprehensive comparisons across multiple models. This may restrict their generalizability and adaptability in practical, complex clinical scenarios [ 22 ]. This study seeks to establish a more accurate and reliable prostate cancer treatment recommendation model using patient data from the First Affiliated Hospital of the University of Science and Technology of China. It integrates machine learning algorithms with SHAP value interpretation techniques, with a specific focus on optimizing survival outcomes for Chinese patients. SHAP technology clarifies the complex associations between features and predictions, enhances model interpretability, and provides valuable references for clinical decision-making. This, in turn, facilitates the optimization of treatment efficacy and ultimately improves the quality of life and long-term survival of Chinese prostate cancer patients. Materials and Methods Data Sources and Study Population This study consecutively recruited patients who underwent radical prostatectomy (RP) from the First Affiliated Hospital of the University of Science and Technology of China between January 2013 and June 2020. Postoperative 5-year survival outcomes were collected via telephone follow-up on July 1, 2025. During data cleaning, 86 cases with incomplete clinical records or missing key information that prevented valid analysis were excluded, resulting in a final analysis cohort of 300 patients with complete follow-up data. The minimum age was 65 years, the maximum age was 85 years, and the mean age was 71.59 years. The study was approved by the institutional review board (IRB), and all subjects provided written informed consent. Data Preprocessing and Feature Selection To balance sample distributions across groups, reduce confounding bias, and improve model performance, this study used propensity score matching (PSM) for data processing. The sample size comparison results indicate a significant difference between Group 0 (234 cases) and Group 1 (66 cases) in the original data, which could affect subsequent analyses. After PSM processing, the sample sizes of both groups were balanced and adjusted to 198 cases (as shown in Figs. 1 (a) and 1(b)), successfully achieving balanced matching between groups and laying a more reasonable dataset foundation for subsequent model analysis. We employed three feature selection algorithms, namely LASSO regression, RF, and RFE. To enhance the reliability of the selected features, we used cross-validation across multiple algorithms. The Venn diagram distribution of the feature selection results is shown in Fig. 2 . LASSO incorporates an L1 regularization penalty term built upon classical linear regression, which forces the coefficients of less relevant features to zero, thereby achieving feature selection and dimensionality reduction. Its advantage lies in the efficient elimination of redundant features. There were 4 features uniquely identified by LASSO alone. In addition, LASSO also participated in multi-algorithm overlapping feature selection, with 4 features jointly identified with RF, 1 feature jointly identified with RFE, and 13 core features jointly identified by all three algorithms. RF is an ensemble-based feature selection method that selects high-contribution features by calculating feature importance metrics (e.g., the Gini coefficient). Its advantages include the ability to capture nonlinear relationships between features and strong resistance to overfitting. RF serves as an intermediate algorithm connecting LASSO and RFE. It not only participated in the overlapping feature selection of multiple algorithms but also, together with the other two algorithms, determined the core feature set. RFE iteratively removes the “least important features” and evaluates the performance of feature subsets using cross-validation to determine the optimal number of features. Its advantage is high automation, which can reduce the subjectivity of feature selection. The number of features uniquely identified by RFE alone was 0. All features selected by RFE were included in the multi-algorithm overlapping segments, with 1 feature jointly identified with LASSO, 5 features jointly identified with RF, and 13 core features jointly identified by all three algorithms. Model Development and Evaluation In this study, we selected 25 machine learning algorithms, covering linear models, tree-based models, ensemble models, and neural network models, including 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, C50 Decision Tree, Conditional Inference Tree, Partial Least Squares DA, Bagging Decision Trees, XGBoost, Gaussian Process, Sparse Linear Discriminant Analysis, AdaBoost, Deep Neural Network, and LightGBM. Our objective is to address the limitations of oversimplified models in current research by comprehensively exploring data features and capturing complex relationships. All analyses in this paper were primarily performed using R software (version 4.3.0). Packages such as plotROC, caret, pROC, and e1071 were utilized for tasks including plotting ROC curves and training machine learning models. Additionally, SHAP plots were generated to visualize clinical features and display each feature's contribution to the model. Finally, an online prediction system was developed using the Shiny framework to assess patients' 5-year postoperative survival. The dataset was randomly split into a training set and a test set, with 70% 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 comprised: AUC (Area Under the Curve), AUC_SD (AUC Standard Deviation), Accuracy, Kappa, Sensitivity, Specificity, PPV, NPV, Precision, Recall, and F1. Model Interpretation and Clinical Applications We employed a SHAP-based interpretation approach to analyze and explore the model with the optimal overall performance, aiming to elucidate its predictive mechanism and feature contributions. Derived from cooperative game theory, SHAP theory offers a rigorous and highly interpretable tool to quantify the specific impact and relative importance of each feature on the model’s predictions. Shiny is a lightweight web framework within the R language ecosystem. It allows rapid encapsulation of R-based models and visualization results into interactive web applications without specialized knowledge, enabling non-technical personnel to access them directly in a web browser. We have developed a “Prostate Cancer Patient 5-Year Postoperative Survival Prediction System” based on Shiny. Its core functionality enables users to input clinical indicators (such as age, PSA, etc.) and obtain, with a single click, predictions of 5-year postoperative survival outcomes and visualizations of feature impacts. This system assists physicians in making clinical decisions. Statistical Analysis Rigorous statistical methods were employed for data analysis in this study. Continuous variables were presented as mean ± standard deviation or median (interquartile range) based on their distribution characteristics, with intergroup comparisons performed using Student's t-test or Wilcoxon rank-sum test; categorical variables were expressed as case numbers and percentages, and intergroup differences were evaluated via chi-square test or Fisher's exact test. Model performance was assessed using discriminative metrics, including sensitivity, specificity, accuracy, and area under the curve (AUC), and 95% confidence intervals for these metrics were constructed using the DeLong method. For survival analysis, the Cox proportional hazards model was adopted, and time-dependent AUC was calculated. The statistical significance threshold was set at P < 0.05. Figure 3 depicts the establishment process of the prostate cancer treatment decision model and the principles of survival analysis. Results Baseline Characteristics This study included 396 patients with prostate cancer (PCa) for a comprehensive retrospective analysis. According to the 5-year postoperative survival status, the cohort was evenly divided into the survival group and the death group, with 198 cases in each group (50% each). Table 1 summarizes the baseline clinical characteristics of the included PCa patients. Univariate and multivariate logistic regression analyses were performed to identify factors associated with 5-year mortality. The results demonstrated that elevated age (multivariate OR = 1.17, 95% CI: 1.06–1.30, p = 0.001), increased blood urea nitrogen (BUN) (multivariate OR = 2.02, 95% CI: 1.49–2.75, p < 0.001), prostate-specific membrane antigen positivity (PSM = 1) (multivariate OR = 6.56, 95% CI: 2.28–18.84, p < 0.001), seminal vesicle invasion (SVI = 1) (multivariate OR = 3.61, 95% CI: 1.31–10.00, p = 0.013), advanced tumor stage (pT3/pT4) (pT3: multivariate OR = 3.93, 95% CI: 1.56–9.92, p = 0.004; pT4: multivariate OR = 26.82, 95% CI: 4.29–167.86, p < 0.001) and lymph node metastasis (pN = 1) (multivariate OR = 9.57, 95% CI: 1.77–51.67, p = 0.009) were independent risk factors for 5-year mortality in PCa patients. In contrast, higher body mass index (BMI) (multivariate OR = 0.77, 95% CI: 0.66–0.91, p = 0.002), elevated albumin (ALB) (multivariate OR = 0.77, 95% CI: 0.67–0.88, p < 0.001) and increased serum creatinine (CRE) (multivariate OR = 0.92, 95% CI: 0.89–0.95, p < 0.001) were confirmed as independent protective factors for 5-year survival (all p < 0.05). Notably, nutritional risk index positivity (PNI = 1) and hemoglobin (HB) were associated with survival outcomes in univariate analysis but failed to exhibit independent associations after adjusting for confounding factors (multivariate p = 0.114 and 0.346, respectively). Total bilirubin (TBIL) showed no significant correlation with survival status in either univariate or multivariate analysis (p > 0.05). Although uric acid (UA) and certain subgroups of Gleason score (GS) were correlated with survival in univariate analysis, their independent predictive value was not verified in the multivariate model. Regarding comorbidities, hypertension was not significantly associated with 5-year survival status in PCa patients (p = 0.417). In contrast, diabetes mellitus was identified as an independent risk factor for 5-year mortality (multivariate OR = 6.77, 95% CI: 1.98–23.18, p = 0.002), rather than showing no statistical relevance. Table 1 Prostate Cancer Patients' Characteristics Stratstate Cancer Patients' Characteristics Stratified by Survival Status Characteristic Statistic alive (N = 198) dead (N = 198) OR (univariable) OR (multivariable) Age Mean ± SD 71.5 ± 4.5 73.0 ± 4.4 1.08 (1.03–1.13, p=.001) 1.17 (1.06–1.30, p=.001) BMI Mean ± SD 23.5 ± 2.8 22.3 ± 2.7 0.86 (0.80–0.93, p<.001) 0.77 (0.66–0.91, p=.002) TBIL Mean ± SD 14.0 ± 4.9 13.3 ± 3.8 0.97 (0.92–1.01, p=.134) 0.96 (0.86–1.07, p=.493) DBIL Mean ± SD 4.7 ± 2.0 4.5 ± 1.6 0.94 (0.84–1.05, p=.271) IBIL Mean ± SD 9.4 ± 3.7 9.0 ± 2.8 0.96 (0.91–1.02, p=.228) ALB Mean ± SD 40.7 ± 5.4 37.5 ± 3.6 0.84 (0.79–0.89, p<.001) 0.77 (0.67–0.88, p<.001) BUN Mean ± SD 6.2 ± 1.5 7.1 ± 2.0 1.45 (1.26–1.68, p<.001) 2.02 (1.49–2.75, p<.001) CRE Mean ± SD 75.7 ± 14.0 66.6 ± 14.8 0.95 (0.94–0.97, p<.001) 0.92 (0.89–0.95, p<.001) UA Mean ± SD 349.3 ± 84.3 311.0 ± 87.4 0.99 (0.99-1.00, p<.001) 1.00 (0.99-1.00, p=.626) A_G Mean ± SD 1.5 ± 0.2 1.5 ± 0.2 0.39 (0.15–1.04, p=.059) 1.32 (0.12–14.08, p=.819) WBC Mean ± SD 6.0 ± 1.6 6.1 ± 1.4 1.04 (0.91–1.19, p=.551) NE Mean ± SD 3.6 ± 1.3 3.7 ± 1.4 1.06 (0.91–1.22, p=.464) LYM Mean ± SD 3.0 ± 12.5 1.7 ± 0.5 0.78 (0.55–1.12, p=.177) 1.46 (0.40–5.36, p=.565) NOM Mean ± SD 0.5 ± 0.2 0.5 ± 0.1 6.48 (1.80-23.32, p=.004) 1.43 (0.01-318.98, p=.896) LMR Mean ± SD 6.5 ± 22.6 3.6 ± 1.6 0.81 (0.71–0.92, p<.001) 0.81 (0.40–1.64, p=.553) NLR Mean ± SD 2.3 ± 1.8 2.5 ± 1.5 1.07 (0.94–1.21, p=.301) PLR Mean ± SD 114.4 ± 62.9 122.7 ± 48.7 1.00 (1.00-1.01, p=.150) 1.00 (1.00-1.01, p=.331) RBC Mean ± SD 4.4 ± 0.5 4.1 ± 0.4 0.24 (0.14–0.40, p<.001) 1.86 (0.26–13.16, p=.535) HB Mean ± SD 134.8 ± 13.4 125.9 ± 12.0 0.94 (0.93–0.96, p<.001) 0.97 (0.90–1.04, p=.346) PLT Mean ± SD 184.3 ± 59.7 187.7 ± 39.8 1.00 (1.00-1.01, p=.507) PT Mean ± SD 11.5 ± 1.9 11.3 ± 1.3 0.96 (0.85–1.08, p=.488) FIB Mean ± SD 2.9 ± 0.7 2.9 ± 0.7 0.92 (0.68–1.23, p=.565) PSA Mean ± SD 33.2 ± 28.1 57.0 ± 69.2 1.02 (1.01–1.02, p<.001) 1.01 (0.99–1.02, p=.390) GS 1 43 (21.7%) 15 (7.6%) 2 37 (18.7%) 37 (18.7%) 2.87 (1.36–6.03, p=.005) 0.65 (0.17–2.50, p=.526) 3 72 (36.4%) 26 (13.1%) 1.04 (0.49–2.17, p=.927) 0.21 (0.05–0.85, p=.029) 4 19 (9.6%) 45 (22.7%) 6.79 (3.06–15.04, p<.001) 0.72 (0.16–3.19, p=.667) 5 26 (13.1%) 70 (35.4%) 7.72 (3.68–16.18, p<.001) 1.95 (0.47–8.13, p=.358) 6 1 (0.5%) 5 (2.5%) 14.33 (1.55-132.77, p=.019) 4.24 (0.10-173.02, p=.446) PNI 0 157 (79.3%) 85 (42.9%) 1 41 (20.7%) 113 (57.1%) 5.09 (3.26–7.94, p<.001) 1.99 (0.85–4.68, p=.114) PSM 0 178 (89.9%) 116 (58.6%) 1 20 (10.1%) 82 (41.4%) 6.29 (3.66–10.82, p<.001) 6.56 (2.28–18.84, p<.001) SVI 0 175 (88.4%) 112 (56.6%) 1 23 (11.6%) 86 (43.4%) 5.84 (3.48–9.80, p<.001) 3.61 (1.31-10.00, p=.013) pT 2 128 (64.6%) 39 (19.7%) 3 68 (34.3%) 107 (54%) 5.16 (3.23–8.26, p<.001) 3.93 (1.56–9.92, p=.004) 4 2 (1%) 52 (26.3%) 85.33 (19.89-366.16, p<.001) 26.82 (4.29-167.86, p<.001) pN 0 194 (98%) 143 (72.2%) 1 4 (2%) 55 (27.8%) 18.65 (6.61–52.66, p<.001) 9.57 (1.77–51.67, p=.009) Hypertension 0 108 (54.5%) 116 (58.6%) 1 90 (45.5%) 82 (41.4%) 0.85 (0.57–1.26, p=.417) Diabetes_mellitus 0 180 (90.9%) 156 (78.8%) 1 18 (9.1%) 42 (21.2%) 2.69 (1.49–4.87, p=.001) 6.77 (1.98–23.18, p=.002) Feature Extraction Figure 4 depicts the optimal feature selection process combining LASSO regression, RF, and RFE. The feature selection process for each individual algorithm is presented in Supplementary Fig. 1: Feature Selection Process of the Three Algorithms. We ultimately selected 10 features common to all three models (Age, ALB, BUN, CRE, HB, PLT, PT, PSA, GS, PNI, PSM, pT and pN) as input parameters for machine learning. Model Construction and Selection The model results indicate that LightGBM and Random Forest achieved the highest Accuracy value, reaching a comprehensive best performance metric of 0.9328, as shown in Fig. 5 . LightGBM, AdaBoost, and Random Forest demonstrated stronger generalization ability on the test set, with AUC values of 0.9778, 0.9758, and 0.9746, respectively, as shown in Fig. 6 . (Table 2 ). Table 2 Various Performance Metrics of 12 Machine Learning Models Model AUC AUC_SD Accuracy Kappa Sensitivity Specificity PPV NPV Precision Recall F1 Gradient Boosting Machine 0.9632 0.0160 0.9244 0.8640 0.9074 0.9538 0.9423 0.9254 0.9423 0.9074 0.9245 LightGBM 0.9778 0.0189 0.9328 0.8467 0.9074 0.9538 0.9412 0.9118 0.9412 0.9074 0.9143 Random Forest 0.9746 0.0127 0.9244 0.8472 0.9074 0.9385 0.9245 0.9242 0.9245 0.9074 0.9159 XGBoost 0.9658 0.0148 0.9244 0.8477 0.9259 0.9231 0.9091 0.9375 0.9091 0.9259 0.9174 Bagging Decision Trees 0.9641 0.0341 0.9160 0.8305 0.9074 0.9231 0.9074 0.9231 0.9074 0.9074 0.9074 AdaBoost 0.9758 0.0113 0.9076 0.8127 0.8704 0.9385 0.9216 0.8971 0.9216 0.8704 0.8952 Bayesian GLM 0.9282 0.0675 0.8824 0.7612 0.8333 0.9231 0.9000 0.8696 0.9000 0.8333 0.8654 Logistic Regression 0.9265 0.0654 0.8824 0.7612 0.8333 0.9231 0.9000 0.8696 0.9000 0.8333 0.8654 Support Vector Machine 0.9296 0.0608 0.8824 0.7619 0.8519 0.9077 0.8846 0.8806 0.8846 0.8519 0.8679 C5.0 Decision Tree 0.9510 0.0192 0.8739 0.7453 0.8519 0.8923 0.8679 0.8788 0.8679 0.8519 0.8598 Deep Neural Network 0.9561 0.0499 0.8739 0.7453 0.8519 0.8923 0.8679 0.8788 0.8679 0.8519 0.8598 Elastic Net 0.9342 0.0658 0.8739 0.7437 0.8148 0.9231 0.8980 0.8571 0.8980 0.8148 0.8544 Naive Bayes 0.9450 0.0350 0.8739 0.7429 0.7963 0.9385 0.9149 0.8472 0.9149 0.7963 0.8515 Ridge Regression 0.9348 0.0686 0.8739 0.7437 0.8148 0.9231 0.8980 0.8571 0.8980 0.8148 0.8544 Flexible Discriminant Analysis 0.9425 0.0333 0.8655 0.7279 0.8333 0.8923 0.8654 0.8657 0.8654 0.8333 0.8491 Gaussian Process 0.9564 0.0528 0.8655 0.7271 0.8148 0.9077 0.8800 0.8551 0.8800 0.8148 0.8462 LASSO Regression 0.9219 0.0599 0.8655 0.7271 0.8148 0.9077 0.8800 0.8551 0.8800 0.8148 0.8462 Linear Discriminant Analysis 0.9370 0.0728 0.8655 0.7262 0.7963 0.9231 0.8958 0.8451 0.8958 0.7963 0.8431 Partial Least Squares DA 0.9405 0.0756 0.8655 0.7262 0.7963 0.9231 0.8958 0.8451 0.8958 0.7963 0.8431 Neural Network 0.9239 0.0484 0.8571 0.7141 0.8889 0.8308 0.8136 0.9000 0.8136 0.8889 0.8496 Conditional Inference Tree 0.8709 0.0803 0.8067 0.6191 0.9259 0.7077 0.7246 0.9200 0.7246 0.9259 0.8130 Quadratic Discriminant Analysis 0.8999 0.0589 0.8067 0.6033 0.6852 0.9077 0.8605 0.7763 0.8605 0.6852 0.7629 K-Nearest Neighbors 0.8177 0.0925 0.7647 0.5268 0.7593 0.7692 0.7321 0.7937 0.7321 0.7593 0.7455 Classification and Regression Tree 0.7709 0.1243 0.6807 0.3829 0.9074 0.4923 0.5976 0.8649 0.5976 0.9074 0.7206 Sparse Linear Discriminant Analysis 0.9328 0.0910 0.5882 0.1003 0.0926 1.0000 1.0000 0.5702 1.0000 0.0926 0.1695 Clinical application of the model All models' calibration curves were close to the ideal dashed line, and the curve-fitting quality of most models was high (see Fig. 7 ). The error bars (vertical lines) corresponding to each group were generally short, indicating smaller fluctuations in the actual event proportions within each group. This suggests that these predictive models had good calibration in the test set, with a high match between the model's predicted death risk and patients' actual 5-year mortality rates, further verifying the reliability of the model's predictions. The quantitative indicators of calibration performance for each model, such as Expected Calibration Error (ECE) and Maximum Calibration Error (MCE), are shown in Supplementary Table 1.The Brier scores of the 12 models are shown in Supplementary Table 2. Figure 8 shows the results of clinical decision curve analysis for the 12 models. All models showed significantly higher net benefits than the two extreme strategies of “no intervention for any patient” (None) and “intervention for all patients” (All) within most clinical threshold probability ranges. Only a few models (e.g., FDA) showed slight decreases in net benefits at low thresholds. These machine learning models all provided substantial net benefits in the clinical decision scenarios of this study, and there were small differences in their clinical practical value, indicating that the practical value of the models in guiding clinical decision-making was superior to traditional experiential judgment. Model interpretability and clinical interactive application The SHAP value analysis of model feature importance (Fig. 9 ) showed that CRE had the greatest impact (+ 0.0953), followed by PSA (+ 0.0782), PSM (+ 0.0769), and pT (+ 0.0598). ALB, PLT, HB, GS, and BUN made smaller contributions to the model's prediction but remained statistically significant. This interactive web application developed with R Shiny (the tool shown in Fig. 10 ) is a predictive system designed specifically for prostate cancer patients to assess their 5-year postoperative survival status, equipped with multiple practical functions: it supports medical staff in entering 13 clinical and pathological parameters and conducts comprehensive analysis of these indicators through machine learning algorithms. In addition to outputting predictive conclusions on patients' 5-year postoperative survival status, it also generates a visualization report of feature importance based on the SHAP values of the LightGBM model, helping medical staff intuitively identify which indicators (such as tumor stage, PSA level, etc.) have a more significant impact on survival prognosis. The online prediction system features a concise, user-friendly interface. After completing parameter input, the medical staff only needs to click the "Predict" button to quickly obtain a comprehensive evaluation report including predictive results and feature analysis. Currently, this application has been officially launched (access address: https://medicalpredictor.shinyapps.io/Online_Prediction_of_Risk_Factors/ ) and is free for medical staff to use. It can be directly integrated into the clinical evaluation workflow, providing data-driven support for clinical prognostic assessment.. Discussion Based on the data of prostate cancer patients from The First Affiliated Hospital of the University of Science and Technology of China (USTC), this study first systematically applied a variety of feature selection algorithms to screen 13 key predictive variables from a large number of potential predictors, and then constructed and validated a machine learning-based predictive model for the 5-year postoperative survival of prostate cancer patients. This streamlined and efficient survival assessment tool integrates multiple machine learning algorithms, enabling accurate, reliable, and personalized predictions. In rigorous performance evaluation, the LightGBM model achieved excellent predictive performance for prostate cancer survival status, with an AUC of 0.9778, sensitivity of 0.9074, and specificity of 0.9538. This study conducted an in-depth analysis of the importance and influencing mechanisms of each variable using the SHAP algorithm, further clarifying the impact of different indicators on survival status. The model demonstrates robust predictive performance in both validation and verification. It not only provides a feasible solution for individualized treatment evaluation of prostate cancer but also, through its post-operative 5-year survival prediction system, constructed using the LightGBM model, offers significant guidance for clinical practice. This system can provide data to help doctors formulate more precise, rational, individualized treatment plans for prostate cancer. In recent years, some researchers have conducted AI-based prognostic studies on prostate cancer, though they mainly utilize clinical MRI and histopathological data. Wong et al. analyzed 338 patients with localized prostate cancer who underwent robot-assisted prostatectomy. They employed three supervised machine learning algorithms and 19 distinct training variables (covering demographic, clinical, imaging, and surgical data) to construct a model in a hypothesis-free manner that predicts the risk of biochemical recurrence within 1 year after surgery [ 23 ]. Zhou et al. retrospectively collected data from 399 prostate cancer patients across three medical centers. They extracted the maximum regions of interest (ROI) from three MRI sequences, trained a deep learning model using H&E-stained sections, and built a nomogram for the combined model. By integrating multimodal data, they developed an AI-based predictive model for the progression of castration-resistant prostate cancer (CRPC) [ 24 ]. Lee et al. established a model integrating clinical data and deep learning to predict long-term biochemical recurrence-free survival (BCRFS) in PCa patients following radical prostatectomy (RP), utilizing multiparametric magnetic resonance imaging (mpMRI) [ 25 ]. Hou et al. initially developed MRI radiomics features (RadS) for biochemical recurrence (BCR), and employed two pre-specified AI models to forecast the tumor's T3 stage and lymph node metastasis (LN+). Subsequently, clinical, imaging, and histopathological variables were incorporated into iBCR-Net for BCR prediction [ 26 ]. Polymeri et al. developed an AI-driven algorithm to automatically quantify the prostate and its tumor content in PET/CT images of 145 patients. This algorithm automatically acquired prostate tumor volume, tumor-to-prostate ratio, whole-tumor lesion uptake rate, and SUVmax values. Subsequently, the Cox proportional hazards regression model was applied to explore the correlations between these measurements and age, PSA, Gleason score, as well as prostate cancer-specific survival (PCSS) [ 27 ]. Koo et al. adopted an artificial neural network (ANN) model to predict survival outcomes under different initial treatment regimens and established an online decision-support system. Compared with the Cox proportional hazards regression model, the ANN model exhibited superior predictive performance in forecasting 5-year and 10-year progression-free survival (PFS) in castration-resistant prostate cancer (CRPC), cancer-specific survival (CSS), and overall survival (OS) [ 28 ]. The aforementioned literature has laid a solid foundation for prostate cancer research by integrating clinical, imaging, and pathological data to provide diverse approaches for model construction and clinical references for different prognostic objectives. Compared with previous studies, this study simplified data sources, developed a more user-friendly survival prediction model, and conducted a comprehensive and detailed comparison and optimization of various machine learning models. Through the study, it was found that T-stage, PSA, BUN, CRE, and Age have a significant impact on the 5-year survival status of prostate cancer patients after surgery. From Table 1 , it was observed that at the T2 stage, the survival and death rates were 186 (62.8%) and 63 (21.3%), respectively, with the number of survivors significantly higher than that of non-survivors. However, at the T4 stage, the survival and death rates were 1 (0.3%) and 75 (25.3%), respectively, indicating an extremely poor prognosis. PSA levels serve as a key basis for prostate cancer monitoring and are critical to its progression [ 29 ]. Fluctuations in PSA levels may act as an early indicator of BCR, rendering continuous PSA monitoring a vital step in tracking changes in prostate cancer status [ 30 ]. However, PSA level changes are not always directly associated with cancer recurrence. For instance, following certain specific treatments (e.g., high-intensity focused ultrasound, HIFU), PSA levels may surge sharply, which does not directly reflect the actual prostate cancer status [ 31 , 32 ]. Thus, it is essential to acknowledge the limitations of PSA testing, as these can lead to unnecessary additional examinations and treatments, further intensifying patients' psychological distress and increasing medical expenses [ 33 ]. A large-scale study of 160,787 U.S. men who underwent radical prostatectomy revealed that patients diagnosed at an older age had a higher prostate cancer mortality rate [ 34 ]. Among prostate cancer patients aged 75 years and above, cryotherapy demonstrated superior survival outcomes compared to other surgical methods for this specific age group [ 35 ]. Nevertheless, academic debate persists regarding the impact of age on prostate cancer mortality risk. A study by Pettersson et al. found no significant inherent effect of age on prostate cancer mortality risk; instead, it indicated that in current clinical practice, elderly prostate cancer patients may not receive adequate diagnostic evaluations and subsequent radical treatments [ 36 ]. Therefore, when formulating treatment plans for prostate cancer, it is necessary to develop individualized and precise plans based on the specific conditions of each patient. Particularly for the elderly prostate cancer population, given the extension of life expectancy, it is crucial to strictly adhere to the screening and treatment guidelines for prostate cancer and determine the most suitable individualized treatment strategies for them. In this study, the LightGBM model exhibited superior performance to other machine learning algorithms. As a highly efficient gradient boosting framework, LightGBM not only excels in capturing complex nonlinear relationships and interactions among variables but also attains faster training speeds and reduced memory usage via histogram-based algorithms and one-sided gradient sampling techniques [ 37 ]. Machine learning methods offer significant advantages over traditional statistical methods for addressing complex, nonlinear clinical problems. The core innovation of our approach lies in the organic integration of model interpretability and clinical utility. Through SHAP value analysis, we can clearly identify the key factors influencing individual risk prediction, thereby enhancing clinicians' trust in the model's results and supporting the development of targeted intervention strategies [ 38 ]. This interpretability is crucial for the clinical implementation of machine learning models in decision support systems. Furthermore, the interactive web app we developed converts complex machine learning outputs into intuitive clinical tools, enabling real-time risk evaluation and visualization of feature significance, thereby effectively addressing the "black box" issue of machine learning models [ 39 ]. Notably, this user-friendly interface design is consistent with Shah et al.'s findings, highlighting its pivotal role in promoting the clinical implementation of machine learning technologies [ 40 ]. Machine learning methods offer significant advantages over traditional statistical methods for addressing complex, nonlinear clinical problems. The core innovation of our approach lies in the organic integration of model interpretability and clinical utility. Through SHAP value analysis, we can clearly identify the key factors influencing individual risk prediction, thereby enhancing clinicians' trust in the model's results and supporting the development of targeted intervention strategies [ 38 ]. This interpretability is crucial for the clinical implementation of machine learning models in decision support systems. In addition, the interactive web application developed by our team converts the complex outputs of machine learning into an intuitive clinical tool, which enables real-time risk assessment and feature importance visualization. It effectively mitigates the black-box problem of machine learning models in clinical prediction, though there is still room for improvement regarding the comprehensiveness of interpretation [ 39 ]. Notably, this user-friendly interface design is consistent with Shah et al.'s findings, highlighting its pivotal role in promoting the clinical implementation of machine learning technologies [ 40 ]. This study employs machine learning methods combined with the SHAP tool to construct a predictive model, thereby enhancing its interpretability. By providing real-time predictions and clinical recommendations, it helps clinicians make more accurate, personalized treatment decisions. However, this study has the following limitations: First, the dataset was based on retrospective clinical data, which may lack dynamic monitoring indicators, such as long-term follow-up data on patients’ quality of life, treatment compliance, and adverse reaction records. Meanwhile, incomplete inclusion of lost-to-follow-up patients has led to potential selection bias. Furthermore, the total sample size of this study was 300 cases, which is relatively limited, and all samples were collected from a single center. This may compromise the external validity and generalizability of the model, whose applicability to multi-center and multi-ethnic populations requires further verification. Moreover, the current model was constructed based on routine clinical indicators (e.g., T stage, prostate-specific antigen, blood urea nitrogen, creatinine, and age), without incorporating imaging data (e.g., magnetic resonance imaging and computed tomography) and genomic data (e.g., gene mutation status). Thus, the comprehensiveness and depth of the feature variables need to be further improved. Finally, this study focused on predicting the 5-year survival status, and the predictive efficacy of the model for long-term survival (e.g., 10-year survival) has not been verified. Future research should extend the follow-up duration and supplement follow-up data to optimize the model. To address these aforementioned issues, the research team aims to expand and enrich the dataset by integrating and harmonizing multi-source data, encompassing imaging data, genomic profiles, and other granular clinical covariates to construct a more comprehensive and precise predictive model. In turn, this model will offer more robust evidentiary support for prostate cancer treatment decision-making, thereby enhancing the scientific rigor and clinical translatability of individualized treatment strategies and prognostic assessment [ 41 ]. Conclusion This study conducted relevant explorations on predicting optimal treatment regimens for prostate cancer patients using multiple machine learning models. Among these machine learning models, the LightGBM model exhibited the best comprehensive performance with an AUC value of 0.9778, indicating that this model has good efficacy in predicting treatment regimens for prostate cancer within the scope of the dataset used in this study. This study analyzed the internal mechanism of the LightGBM model using the SHAP tool and found that five indicators, namely T stage, CRE, PSA, PSM, and ALB, were significantly correlated with the 5-year postoperative survival rate of prostate cancer patients. By integrating machine learning algorithms with SHAP interpretability analysis, this study constructed a model applicable for predicting treatment regimens for prostate cancer. Its predictive efficacy has been verified to reach a high level in this study, providing potential reference and technical support for the clinical implementation of precise individualized decision-making. Declarations Data Availability Statement Data are available from the corresponding author upon reasonable request. Conflicts of interest The authors declare no competing interests. Ethics approval This study was conducted in accordance with the principles of the Declaration of Helsinki and its subsequent revisions. The study was approved by the Bioethics Committee of The First Affiliated Hospital of University of Science and Technology of China (No.2024-RE-272). Informed consent or substitute for it was obtained from all patients for being included in the study. Clinical trial number Not applicable. Consent for publication Not applicable. Funding This research was funded by the Anhui Province Teaching Research Program (Granted No. 2023jyxm0347). Author Contribution Data Analysis: T.S., and C.H; Writing Original Draft: T.S., and C.H. References Sung H, Ferlay J, Siegel RL, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin. 2021;71(3):209–49. 10.3322/caac.21660 . Miller KD, Nogueira L, Devasia T, et al. Cancer treatment and survivorship statistics, 2022. CA Cancer J Clin. 2022;72(5):409–36. 10.3322/caac.21731 . Siegel RL, Miller KD, Wagle NS, Jemal A, Cancer statistics. 2023. CA Cancer J Clin. 2023;73(1):17–48. 10.3322/caac.21763 Global Burden of Disease Collaborative Network. Global Burden of Disease Study 2023 (GBD 2023). Seattle, United States: Institute for Health Metrics and Evaluation (IHME), 2025. Available from: https://vizhub.healthdata.org/gbd-results/ Wu B, Lu X, Shen H, et al. Intratumoral heterogeneity and genetic characteristics of prostate cancer. Int J Cancer. 2020;146(12):3369–78. 10.1002/ijc.32961 . Panaiyadiyan S, Kumar R. Prostate cancer nomograms and their application in Asian men: a review. Prostate Int. 2024;12(1):1–9. 10.1016/j.prnil.2023.07.004 . López-Abad A, Ramírez Backhaus M, Server Gómez G, et al. Real-world prostate-specific antigen reduction and survival outcomes of metastatic hormone-sensitive prostate cancer patients treated with apalutamide: An observational, retrospective, and multicentre study. Prostate Int. 2024;12(1):20–6. 10.1016/j.prnil.2023.10.003 . Litwin MS, Tan HJ. The Diagnosis and Treatment of Prostate Cancer: A Review. JAMA. 2017;317(24):2532–42. 10.1001/jama.2017.7248 . Chen AB, Niu J, Cronin AM, Shih YT, Giordano S, Schrag D. Variation in Use of High-Cost Technologies for Palliative Radiation Therapy by Radiation Oncologists. J Natl Compr Canc Netw. 2021;19(4):421–31. 10.6004/jnccn.2020.7633 . Published 2021 Feb 12. Ferlay JEM, Lam F, Colombet M, Mery L, Pineros et al. Global Cancer Observatory: Cancer Tomorrow; International Agency for Research on Cancer: Lyon, France; Available online: https://gco.iarc.fr/tomorrow Fu ZT, Guo XL, Zhang SW, et al. Analysis of incidence and death of prostate cancer in China in 2015. Chin J Oncol. 2020;42(9):718–22. Zheng RS, Chen R, Han BF, et al. Zhonghua Zhong Liu Za Zhi. 2024;46(3):221–31. 10.3760/cma.j.cn112152-20240119-00035 . He J, Chen WQ, Li N, et al. Guidelines for screening, early diagnosis, and early treatment of prostate cancer in China (2022, Beijing). Chin J Oncol. 2022;44(1):29–53. Satturwar S, Parwani AV. Artificial Intelligence-Enabled Prostate Cancer Diagnosis and Prognosis: Current State and Future Implications. Adv Anat Pathol. 2024;31(2):136–44. 10.1097/PAP.0000000000000425 . Ren S, Li J, Dorado J, et al. From molecular mechanisms of prostate cancer to translational applications: based on multi-omics fusion analysis and intelligent medicine. Health Inf Sci Syst. 2023;12(1):6. 10.1007/s13755-023-00264-5 . Published 2023 Dec 18. Bulten W, Kartasalo K, Chen PC, et al. Artificial intelligence for diagnosis and Gleason grading of prostate cancer: the PANDA challenge. Nat Med. 2022;28(1):154–63. 10.1038/s41591-021-01620-2 . Kench JG, Amin MB, Berney DM, et al. WHO Classification of Tumours fifth edition: evolving issues in the classification, diagnosis, and prognostication of prostate cancer. Histopathology. 2022;81(4):447–58. 10.1111/his.14711 . Tang S, Zhang H, Liang J, et al. Prostate cancer treatment recommendation study based on machine learning and SHAP interpreter. Cancer Sci. 2024;115(11):3755–66. 10.1111/cas.16327 . Alataş E, Tanyıldızı Kökkülünk H, Tanyıldızı H, Alcın G. Treatment prediction with machine learning in prostate cancer patients. Comput Methods Biomech Biomed Engin. 2025;28(4):572–80. 10.1080/10255842.2023.2298364 . Hu S, Chang CP, Snyder J, et al. Comparing Active Surveillance and Watchful Waiting With Radical Treatment Using Machine Learning Models Among Patients With Prostate Cancer. JCO Clin Cancer Inf. 2023;7:e2300083. 10.1200/CCI.23.00083 . Syed K, Sleeman W, Soni P, et al. Machine-Learning Models for Multicenter Prostate Cancer Treatment Plans. J Comput Biol. 2021;28(2):166–84. 10.1089/cmb.2020.0188 . Auffenberg GB, Ghani KR, Ramani S, et al. askMUSIC: Leveraging a Clinical Registry to Develop a New Machine Learning Model to Inform Patients of Prostate Cancer Treatments Chosen by Similar Men. Eur Urol. 2019;75(6):901–7. 10.1016/j.eururo.2018.09.050 . Wong NC, Lam C, Patterson L, Shayegan B. Use of machine learning to predict early biochemical recurrence after robot-assisted prostatectomy. BJU Int. 2019;123(1):51–7. 10.1111/bju.14477 . Zhou C, Zhang YF, Guo S et al. Multimodal data integration for predicting progression risk in castration-resistant prostate cancer using deep learning: a multicenter retrospective study. Front Oncol. 2024;14:1287995. Published 2024 Mar 14. 10.3389/fonc.2024.1287995 Lee HW, Kim E, Na I, Kim CK, Seo SI, Park H. Novel Multiparametric Magnetic Resonance Imaging-Based Deep Learning and Clinical Parameter Integration for the Prediction of Long-Term Biochemical Recurrence-Free Survival in Prostate Cancer after Radical Prostatectomy. Cancers (Basel). 2023;15(13):3416. 10.3390/cancers15133416 . Published 2023 Jun 29. Hou Y, Jiang KW, Wang LL, et al. Biopsy-free AI-aided precision MRI assessment in prediction of prostate cancer biochemical recurrence. Br J Cancer. 2023;129(10):1625–33. 10.1038/s41416-023-02441-5 . Polymeri E, Kjölhede H, Enqvist O, et al. Artificial intelligence-based measurements of PET/CT imaging biomarkers are associated with disease-specific survival of high-risk prostate cancer patients. Scand J Urol. 2021;55(6):427–33. 10.1080/21681805.2021.1977845 . Koo KC, Lee KS, Kim S, et al. Long short-term memory artificial neural network model for prediction of prostate cancer survival outcomes according to initial treatment strategy: development of an online decision-making support system. World J Urol. 2020;38(10):2469–76. 10.1007/s00345-020-03080-8 . Hoey C, Ahmed M, Fotouhi Ghiam A, et al. Circulating miRNAs as non-invasive biomarkers to predict aggressive prostate cancer after radical prostatectomy. J Transl Med. 2019;17(1):173. 10.1186/s12967-019-1920-5 . Published 2019 May 23. Planas J, Morote J. Criterios de recidiva bioquímica después de prostatectomía radical. Historia natural de la enfermedad [Biochemical recurrence criteria after radical prostatectomy. Natural history of the disease]. Arch Esp Urol. 2012;65(1):4–11. Inamoto T, Komura K, Watsuji T, Azuma H. Rapid Increase of the Serum PSA Level in Response to High-Intensity Focused Ultrasound Therapy may be a Potential Indicator of Biochemical Recurrence of Low- and Intermediate-Risk Prostate Cancer. Clin Med Insights Oncol. 2011;5:101–6. 10.4137/CMO.S7073 . Lamallem H, et al. Elévation du PSA après irradiation prostatique: rebond ou récidive biologique? [Elevation of PSA after prostate radiotherapy: rebound or biochemical recurrence?]. Prog Urol. 2008;18(9):557–61. 10.1016/j.purol.2008.06.001 . Xia SJ, Jing YF. Biochemical recurrence of prostate cancer: the controversial recognition and management. Chin Med J (Engl). 2011;124(9):1283–5. Abdollah F, Sun M, Thuret R, et al. A competing-risks analysis of survival after alternative treatment modalities for prostate cancer patients: 1988–2006. Eur Urol. 2011;59(1):88–95. 10.1016/j.eururo.2010.10.003 . Zheng Y, Ye Y, Chen L, et al. Prevalence and outcomes of focal ablation versus prostatectomy for elderly patients with prostate cancer: a population-based study. J Natl Cancer Cent. 2021;2(1):25–32. 10.1016/j.jncc.2021.11.005 . Published 2021 Nov 26. Pettersson A, Robinson D, Garmo H, Holmberg L, Stattin P. Age at diagnosis and prostate cancer treatment and prognosis: a population-based cohort study. Ann Oncol. 2018;29(2):377–85. 10.1093/annonc/mdx742 . GuolinKe QM, Finley T, Wang T, Chen W, Ma W, Ye Q, Liu TY, Lightgbm. A highly efficient gradient boosting decision tree. Adv Neural Inf Process Syst. 2017;30:52. Bernard D, Doumard E, Ader I, et al. Explainable machine learning framework to predict personalized physiological aging. Aging Cell. 2023;22(8):e13872. 10.1111/acel.13872 . Rudin C. Stop Explaining Black Box Machine Learning Models for High Stakes Decisions and Use Interpretable Models Instead. Nat Mach Intell. 2019;1(5):206–15. 10.1038/s42256-019-0048-x . Shah NH, Milstein A, Bagley SC. Making machine learning models clinically useful. JAMA. 2019;322(14):1351–2. Liu M, Luo J, Li L et al. Design and development of a disease-specific clinical database system to increase the availability of hospital data in China. Health Inf Sci Syst. 2023;11(1):11. Published 2023 Jan 30. 10.1007/s13755-023-00211-4 Additional Declarations No competing interests reported. Supplementary Files SupplementaryTable1CalibrationModelECE.csv SupplementaryTable2brierresults.csv SupplementaryFigure1.tif cleanedExplainableandVisualizableMachineLearningModelDevelopmentandValidationfor5YearPostoperativeSurvivalPredictioninProstateCancerPatientsAged65Years.docx Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 06 Mar, 2026 Reviews received at journal 24 Feb, 2026 Reviewers agreed at journal 14 Feb, 2026 Reviewers agreed at journal 13 Feb, 2026 Reviews received at journal 10 Feb, 2026 Reviewers agreed at journal 02 Feb, 2026 Reviewers invited by journal 02 Feb, 2026 Editor assigned by journal 29 Jan, 2026 Editor invited by journal 28 Jan, 2026 Submission checks completed at journal 28 Jan, 2026 First submitted to journal 28 Jan, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8665949","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":584205426,"identity":"256969ac-63b6-4a18-85a3-f79b212bcf93","order_by":0,"name":"Huaying Chen","email":"","orcid":"","institution":"The First Affiliated Hospital of University of Science and Technology of China","correspondingAuthor":false,"prefix":"","firstName":"Huaying","middleName":"","lastName":"Chen","suffix":""},{"id":584205427,"identity":"8ac2587c-aa51-439f-9cc0-9601891db799","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":"2026-01-22 06:24:17","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8665949/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8665949/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":101850817,"identity":"5d72ab7a-474b-4c0b-8d7b-919db477a488","added_by":"auto","created_at":"2026-02-04 09:59:25","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":398635,"visible":true,"origin":"","legend":"\u003cp\u003eSample Size Comparison Before and After PSM\u003c/p\u003e","description":"","filename":"Figure1.SampleSizeComparisonBeforeandAfterPSM.png","url":"https://assets-eu.researchsquare.com/files/rs-8665949/v1/23b0a5d41f11a47df7b90322.png"},{"id":101850827,"identity":"fa6093d6-8905-436e-b1f0-eb5ee6e63e32","added_by":"auto","created_at":"2026-02-04 09:59:29","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1014298,"visible":true,"origin":"","legend":"\u003cp\u003eVenn Diagram of Features Selected by Three Algorithms\u003c/p\u003e","description":"","filename":"Figure2.VennDiagramofFeaturesSelectedbyThreeAlgorithms.png","url":"https://assets-eu.researchsquare.com/files/rs-8665949/v1/c165204eefae4310fd08e48d.png"},{"id":101850876,"identity":"a02ad7c0-ca8f-4c82-8ddb-73bd6e0f6c1b","added_by":"auto","created_at":"2026-02-04 09:59:41","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":4701271,"visible":true,"origin":"","legend":"\u003cp\u003eFramework Diagram of Prostate Cancer Treatment Decision Model\u003c/p\u003e","description":"","filename":"Figure3.FrameworkDiagramofProstateCancerTreatmentDecisionModel.png","url":"https://assets-eu.researchsquare.com/files/rs-8665949/v1/8b01e0ea2c5cb2d32ebc2e47.png"},{"id":101850727,"identity":"8005d272-54e0-4c80-a3c6-312a16186ef8","added_by":"auto","created_at":"2026-02-04 09:59:07","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":36638611,"visible":true,"origin":"","legend":"\u003cp\u003eOptimal Feature Selection Process for the Merging of Three Algorithms\u003c/p\u003e","description":"","filename":"Figure4.OptimalFeatureSelectionProcessfortheMergingofThreeAlgorithms.png","url":"https://assets-eu.researchsquare.com/files/rs-8665949/v1/9f344104e009e2376e8c569a.png"},{"id":101881459,"identity":"cc268f95-e2c2-466e-8fd7-68947a7d75da","added_by":"auto","created_at":"2026-02-04 15:12:16","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":1564884,"visible":true,"origin":"","legend":"\u003cp\u003eAccuracy of 12 Machine Learning Models\u003c/p\u003e","description":"","filename":"Figure5.ROCCurvesonTestfor12MachineLearningModels.png","url":"https://assets-eu.researchsquare.com/files/rs-8665949/v1/4e998eece26e6722b1753c5c.png"},{"id":101850806,"identity":"269ccf01-57c5-4137-838e-15907aefa53b","added_by":"auto","created_at":"2026-02-04 09:59:16","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":2748132,"visible":true,"origin":"","legend":"\u003cp\u003eROC Curves on Test for 12 Machine Learning Models\u003c/p\u003e","description":"","filename":"Figure6.Accuracyof12MachineLearningModels.png","url":"https://assets-eu.researchsquare.com/files/rs-8665949/v1/47aef43b1893a4502376a301.png"},{"id":101850865,"identity":"d294b3ab-d8cc-49d5-b00c-997051d6adcc","added_by":"auto","created_at":"2026-02-04 09:59:36","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":3934303,"visible":true,"origin":"","legend":"\u003cp\u003eCalibration Curves in the test\u003c/p\u003e","description":"","filename":"Figure7.CalibrationCurvesinthetest.png","url":"https://assets-eu.researchsquare.com/files/rs-8665949/v1/86d01eb7285997826c6a02a7.png"},{"id":101850878,"identity":"5640bcc5-c1b4-4c88-b964-45a468a07ebf","added_by":"auto","created_at":"2026-02-04 09:59:41","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":2402928,"visible":true,"origin":"","legend":"\u003cp\u003eDecision Curve Analysis Plot of 12 Models\u003c/p\u003e","description":"","filename":"Figure8.DecisionCurveAnalysisPlotof12Models.png","url":"https://assets-eu.researchsquare.com/files/rs-8665949/v1/3157cf45d3410e59773543d8.png"},{"id":101850812,"identity":"e8b5888a-0d33-470c-85d4-fd862f54eaad","added_by":"auto","created_at":"2026-02-04 09:59:22","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":3163171,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP Values of Feature Variables and Their Impact on Prediction Results\u003c/p\u003e","description":"","filename":"Figure9.SHAPValuesofFeatureVariablesandTheirImpactonPredictionResults.png","url":"https://assets-eu.researchsquare.com/files/rs-8665949/v1/965e0e34034bba04b0764b2f.png"},{"id":101850891,"identity":"9b492a0f-fcd9-4d62-837a-092b72563c53","added_by":"auto","created_at":"2026-02-04 09:59:49","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":10459512,"visible":true,"origin":"","legend":"\u003cp\u003eOnline Prediction System for Postoperative 5-Year Survival Status of elder Prostate Cancer Patients\u003c/p\u003e","description":"","filename":"Figure10.OnlinePredictionSystemforPostoperative5YearSurvivalStatusofPros.png","url":"https://assets-eu.researchsquare.com/files/rs-8665949/v1/1b8f078849330854c1408747.png"},{"id":101850412,"identity":"cdb17424-d3ee-40dc-9094-931d3dc05158","added_by":"auto","created_at":"2026-02-04 09:57:55","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1083479,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8665949/v1/1cf3d6c0-5c49-40f1-ad20-37f1bd883485.pdf"},{"id":101850808,"identity":"9ade2b2e-2bb1-4925-9fd4-a8244533a5e4","added_by":"auto","created_at":"2026-02-04 09:59:16","extension":"csv","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":250,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTable1CalibrationModelECE.csv","url":"https://assets-eu.researchsquare.com/files/rs-8665949/v1/2ee10e5a1665bed81791c406.csv"},{"id":101850877,"identity":"cdc262d1-c3f1-4b22-9fbe-8f142977c26a","added_by":"auto","created_at":"2026-02-04 09:59:41","extension":"csv","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":235,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTable2brierresults.csv","url":"https://assets-eu.researchsquare.com/files/rs-8665949/v1/0169e37a1accb921fc2511a7.csv"},{"id":101850832,"identity":"9e377439-952d-44df-b28a-c9e40e38fdc7","added_by":"auto","created_at":"2026-02-04 09:59:31","extension":"tif","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":2386144,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryFigure1.tif","url":"https://assets-eu.researchsquare.com/files/rs-8665949/v1/ac56522444ab8d061e6ff223.tif"},{"id":101850840,"identity":"83af0541-b56c-4907-9930-86b90853bc21","added_by":"auto","created_at":"2026-02-04 09:59:36","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":12457672,"visible":true,"origin":"","legend":"","description":"","filename":"cleanedExplainableandVisualizableMachineLearningModelDevelopmentandValidationfor5YearPostoperativeSurvivalPredictioninProstateCancerPatientsAged65Years.docx","url":"https://assets-eu.researchsquare.com/files/rs-8665949/v1/2570dab406131b37790bd744.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Explainable and Visualizable Machine Learning Model Development and Validation for 5-Year Postoperative Survival Prediction in Prostate Cancer Patients Aged ≥ 65 Years","fulltext":[{"header":"Introduction","content":"\u003cp\u003ePCa is the most common malignant tumor of the urinary system and the second most common malignant tumor in men, ranking fifth in terms of mortality rate[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eGBD 2023 data show that the global death toll from prostate cancer was 21,461,648 in 1990 and 47,295,326 in 2023, indicating that it has become an important public health issue[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e \u003cp\u003ePCa is a complex disease that poses significant challenges to both patients and clinicians. High heterogeneity is one of its core characteristics. Within the same patient's prostate tissue, different lesions may exhibit distinct biological behaviors (e.g., aggressive vs. indolent). Among different patients, the mutation profiles, epigenetic modifications, and tumor molecular subtypes also show significant differences [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. This heterogeneity leads to significant uncertainty in the natural course of PCa. While the majority of PCa presents as slowly progressive indolent tumors, a minority of patients may develop into highly malignant diseases characterized by strong invasiveness and distant metastasis.\u003c/p\u003e \u003cp\u003eTreatment decisions are primarily based on a comprehensive assessment of the patient\u0026rsquo;s prognostic risk, including clinical-pathological features and individual patient health conditions. Clinical pathological features include prostate-specific antigen (PSA) levels, tumor Gleason scores, and TNM staging. Individual patient health conditions encompass sociodemographic and clinical characteristics such as age, overall health status (e.g., presence of other chronic diseases), and life expectancy [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAdditionally, in recent years, notable advances have been achieved in the diagnosis and treatment of PCa, bringing substantial innovations to clinical practice. Considering the marked variations in disease progression stages, individual physiological conditions, and quality-of-life expectations among PCa patients, both clinical research and practical experience have verified that surgical treatment can effectively prevent or delay tumor recurrence and progression, thereby significantly enhancing patients' disease-free survival and overall survival rates.\u003c/p\u003e \u003cp\u003eIn addition to surgical treatment, clinical practice offers a variety of non-surgical options, including external beam radiotherapy, brachytherapy, chemotherapy, hormone therapy, and targeted drug therapy, providing patients with a diverse range of treatment choices [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIt is crucial to formulate an individualized optimal treatment plan for each patient. This decision directly impacts disease control efficacy and patient prognosis, while also helping to optimize medical resource allocation and alleviate patients' daily and financial care burdens [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eProstate cancer has an extremely high incidence rate among elderly men. The incidence and mortality rates of this disease are strongly age-related, with the average age at diagnosis for most patients centered around 65 years old. The formation of this age-related difference may be attributed to the joint effects of multiple factors, including genetic susceptibility, environmental exposure, and socioeconomic factors [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn 2015, 72,000 new prostate cancer cases and roughly 30,700 deaths were documented in China. While the incidence of prostate cancer in China is substantially lower than that in European and American countries, it has exhibited an upward tendency in recent years, with a faster growth rate compared to developed nations in Europe and America[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. By 2022, the number of new cases and deaths from prostate cancer in China is projected to increase to 134,200 and 47,500 cases, respectively[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eChina still confronts numerous challenges in the early diagnosis and standardized treatment of prostate cancer, exhibiting a notable disparity compared with Western developed countries. Taking the United States as an illustration, nearly 91% of newly diagnosed prostate cancer cases in the country are clinically localized, with radical prostatectomy or radical radiotherapy serving as the first-line treatment. Patients receiving standardized therapy tend to have a favorable prognosis, with a 5-year overall survival rate approaching 100%. In contrast, only 30% of newly diagnosed cases in China are clinically localized prostate cancer, while the remaining 70% are patients with locally advanced or metastatic disease. These patients are ineligible for local curative treatment, resulting in a relatively poor prognosis [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e].\u003c/p\u003e \u003cp\u003e At present, treatment guidelines for PCa primarily rely on traditional linear models\u0026mdash;such as survival analysis and Cox proportional hazards models\u0026mdash;for risk stratification. Nevertheless, owing to the intricate nonlinear interactions among various prognostic factors in PCa biology, relying exclusively on these linear approaches may not be adequate to precisely forecast individual survival outcomes.\u003c/p\u003e \u003cp\u003eAs artificial intelligence (AI) and ML technologies continue to advance, enabling the processing of massive datasets in relatively short periods, their applications in the medical field are increasingly widespread. Machine learning can be applied to various areas, including drug development, gene expression profiling, biomarker research in multi-omics construction and analysis, and digital pathology slide analysis [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. In the era of personalized medicine, researchers are developing machine learning algorithms based on individual data to achieve precise predictions of survival outcomes for prostate cancer patients.\u003c/p\u003e \u003cp\u003eMachine learning technology has enormous potential for designing personalized treatment plans [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. By systematically integrating individual factors such as age, health status, and lifestyle, machine learning models can help researchers capture subtle differences in tumor characteristics among prostate cancer patients, thereby providing robust technical support for the formulation of precise, personalized treatment strategies.\u003c/p\u003e \u003cp\u003eNevertheless, this field still confronts considerable challenges. First, the variables utilized in prior studies are lacking in dimensionality and scope [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e], which undermines the predictive performance of the models. Consequently, this affects the reliability of experimental findings and the comprehension of complex treatment decision models [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. Second, the small sample sizes and concentrated data sources from single research centers have led to generally diminished representativeness and generalizability of previous research. Finally, the machine learning models currently employed often depend on a single or limited set of model types, with a dearth of comprehensive comparisons across multiple models. This may restrict their generalizability and adaptability in practical, complex clinical scenarios [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThis study seeks to establish a more accurate and reliable prostate cancer treatment recommendation model using patient data from the First Affiliated Hospital of the University of Science and Technology of China. It integrates machine learning algorithms with SHAP value interpretation techniques, with a specific focus on optimizing survival outcomes for Chinese patients. SHAP technology clarifies the complex associations between features and predictions, enhances model interpretability, and provides valuable references for clinical decision-making. This, in turn, facilitates the optimization of treatment efficacy and ultimately improves the quality of life and long-term survival of Chinese prostate cancer patients.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eData Sources and Study Population\u003c/h2\u003e \u003cp\u003eThis study consecutively recruited patients who underwent radical prostatectomy (RP) from the First Affiliated Hospital of the University of Science and Technology of China between January 2013 and June 2020. Postoperative 5-year survival outcomes were collected via telephone follow-up on July 1, 2025. During data cleaning, 86 cases with incomplete clinical records or missing key information that prevented valid analysis were excluded, resulting in a final analysis cohort of 300 patients with complete follow-up data. The minimum age was 65 years, the maximum age was 85 years, and the mean age was 71.59 years. The study was approved by the institutional review board (IRB), and all subjects provided written informed consent.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eData Preprocessing and Feature Selection\u003c/h3\u003e\n\u003cp\u003eTo balance sample distributions across groups, reduce confounding bias, and improve model performance, this study used propensity score matching (PSM) for data processing. The sample size comparison results indicate a significant difference between Group 0 (234 cases) and Group 1 (66 cases) in the original data, which could affect subsequent analyses. After PSM processing, the sample sizes of both groups were balanced and adjusted to 198 cases (as shown in Figs.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e(a) and 1(b)), successfully achieving balanced matching between groups and laying a more reasonable dataset foundation for subsequent model analysis.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eWe employed three feature selection algorithms, namely LASSO regression, RF, and RFE. To enhance the reliability of the selected features, we used cross-validation across multiple algorithms. 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\u003eLASSO incorporates an L1 regularization penalty term built upon classical linear regression, which forces the coefficients of less relevant features to zero, thereby achieving feature selection and dimensionality reduction. Its advantage lies in the efficient elimination of redundant features. There were 4 features uniquely identified by LASSO alone. In addition, LASSO also participated in multi-algorithm overlapping feature selection, with 4 features jointly identified with RF, 1 feature jointly identified with RFE, and 13 core features jointly identified by all three algorithms.\u003c/p\u003e \u003cp\u003eRF is an ensemble-based feature selection method that selects high-contribution features by calculating feature importance metrics (e.g., the Gini coefficient). Its advantages include the ability to capture nonlinear relationships between features and strong resistance to overfitting. RF serves as an intermediate algorithm connecting LASSO and RFE. It not only participated in the overlapping feature selection of multiple algorithms but also, together with the other two algorithms, determined the core feature set.\u003c/p\u003e \u003cp\u003eRFE iteratively removes the \u0026ldquo;least important features\u0026rdquo; and evaluates the performance of feature subsets using cross-validation to determine the optimal number of features. Its advantage is high automation, which can reduce the subjectivity of feature selection. The number of features uniquely identified by RFE alone was 0. All features selected by RFE were included in the multi-algorithm overlapping segments, with 1 feature jointly identified with LASSO, 5 features jointly identified with RF, and 13 core features jointly identified by all three algorithms.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eModel Development and Evaluation\u003c/h3\u003e\n\u003cp\u003eIn this study, we selected 25 machine learning algorithms, covering linear models, tree-based models, ensemble models, and neural network models, including 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, C50 Decision Tree, Conditional Inference Tree, Partial Least Squares DA, Bagging Decision Trees, XGBoost, Gaussian Process, Sparse Linear Discriminant Analysis, AdaBoost, Deep Neural Network, and LightGBM.\u003c/p\u003e \u003cp\u003eOur objective is to address the limitations of oversimplified models in current research by comprehensively exploring data features and capturing complex relationships. All analyses in this paper were primarily performed using R software (version 4.3.0). Packages such as plotROC, caret, pROC, and e1071 were utilized for tasks including plotting ROC curves and training machine learning models. Additionally, SHAP plots were generated to visualize clinical features and display each feature's contribution to the model. Finally, an online prediction system was developed using the Shiny framework to assess patients' 5-year postoperative survival.\u003c/p\u003e \u003cp\u003eThe dataset was randomly split into a training set and a test set, with 70% 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 comprised: AUC (Area Under the Curve), AUC_SD (AUC Standard Deviation), Accuracy, Kappa, Sensitivity, Specificity, PPV, NPV, Precision, Recall, and F1.\u003c/p\u003e\n\u003ch3\u003eModel Interpretation and Clinical Applications\u003c/h3\u003e\n\u003cp\u003eWe employed a SHAP-based interpretation approach to analyze and explore the model with the optimal overall performance, aiming to elucidate its predictive mechanism and feature contributions. Derived from cooperative game theory, SHAP theory offers a rigorous and highly interpretable tool to quantify the specific impact and relative importance of each feature on the model\u0026rsquo;s predictions.\u003c/p\u003e \u003cp\u003eShiny is a lightweight web framework within the R language ecosystem. It allows rapid encapsulation of R-based models and visualization results into interactive web applications without specialized knowledge, enabling non-technical personnel to access them directly in a web browser. We have developed a \u0026ldquo;Prostate Cancer Patient 5-Year Postoperative Survival Prediction System\u0026rdquo; based on Shiny. Its core functionality enables users to input clinical indicators (such as age, PSA, etc.) and obtain, with a single click, predictions of 5-year postoperative survival outcomes and visualizations of feature impacts. This system assists physicians in making clinical decisions.\u003c/p\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003eStatistical Analysis\u003c/h2\u003e \u003cp\u003eRigorous statistical methods were employed for data analysis in this study. Continuous variables were presented as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation or median (interquartile range) based on their distribution characteristics, with intergroup comparisons performed using Student's t-test or Wilcoxon rank-sum test; categorical variables were expressed as case numbers and percentages, and intergroup differences were evaluated via chi-square test or Fisher's exact test. Model performance was assessed using discriminative metrics, including sensitivity, specificity, accuracy, and area under the curve (AUC), and 95% confidence intervals for these metrics were constructed using the DeLong method. For survival analysis, the Cox proportional hazards model was adopted, and time-dependent AUC was calculated. The statistical significance threshold was set at P\u0026thinsp;\u0026lt;\u0026thinsp;0.05. Figure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e depicts the establishment process of the prostate cancer treatment decision model and the principles of survival analysis.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003eBaseline Characteristics\u003c/h2\u003e \u003cp\u003eThis study included 396 patients with prostate cancer (PCa) for a comprehensive retrospective analysis. According to the 5-year postoperative survival status, the cohort was evenly divided into the survival group and the death group, with 198 cases in each group (50% each). Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e summarizes the baseline clinical characteristics of the included PCa patients.\u003c/p\u003e \u003cp\u003eUnivariate and multivariate logistic regression analyses were performed to identify factors associated with 5-year mortality. The results demonstrated that elevated age (multivariate OR\u0026thinsp;=\u0026thinsp;1.17, 95% CI: 1.06\u0026ndash;1.30, p\u0026thinsp;=\u0026thinsp;0.001), increased blood urea nitrogen (BUN) (multivariate OR\u0026thinsp;=\u0026thinsp;2.02, 95% CI: 1.49\u0026ndash;2.75, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), prostate-specific membrane antigen positivity (PSM\u0026thinsp;=\u0026thinsp;1) (multivariate OR\u0026thinsp;=\u0026thinsp;6.56, 95% CI: 2.28\u0026ndash;18.84, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), seminal vesicle invasion (SVI\u0026thinsp;=\u0026thinsp;1) (multivariate OR\u0026thinsp;=\u0026thinsp;3.61, 95% CI: 1.31\u0026ndash;10.00, p\u0026thinsp;=\u0026thinsp;0.013), advanced tumor stage (pT3/pT4) (pT3: multivariate OR\u0026thinsp;=\u0026thinsp;3.93, 95% CI: 1.56\u0026ndash;9.92, p\u0026thinsp;=\u0026thinsp;0.004; pT4: multivariate OR\u0026thinsp;=\u0026thinsp;26.82, 95% CI: 4.29\u0026ndash;167.86, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and lymph node metastasis (pN\u0026thinsp;=\u0026thinsp;1) (multivariate OR\u0026thinsp;=\u0026thinsp;9.57, 95% CI: 1.77\u0026ndash;51.67, p\u0026thinsp;=\u0026thinsp;0.009) were independent risk factors for 5-year mortality in PCa patients.\u003c/p\u003e \u003cp\u003eIn contrast, higher body mass index (BMI) (multivariate OR\u0026thinsp;=\u0026thinsp;0.77, 95% CI: 0.66\u0026ndash;0.91, p\u0026thinsp;=\u0026thinsp;0.002), elevated albumin (ALB) (multivariate OR\u0026thinsp;=\u0026thinsp;0.77, 95% CI: 0.67\u0026ndash;0.88, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and increased serum creatinine (CRE) (multivariate OR\u0026thinsp;=\u0026thinsp;0.92, 95% CI: 0.89\u0026ndash;0.95, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) were confirmed as independent protective factors for 5-year survival (all p\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e \u003cp\u003eNotably, nutritional risk index positivity (PNI\u0026thinsp;=\u0026thinsp;1) and hemoglobin (HB) were associated with survival outcomes in univariate analysis but failed to exhibit independent associations after adjusting for confounding factors (multivariate p\u0026thinsp;=\u0026thinsp;0.114 and 0.346, respectively). Total bilirubin (TBIL) showed no significant correlation with survival status in either univariate or multivariate analysis (p\u0026thinsp;\u0026gt;\u0026thinsp;0.05). Although uric acid (UA) and certain subgroups of Gleason score (GS) were correlated with survival in univariate analysis, their independent predictive value was not verified in the multivariate model.\u003c/p\u003e \u003cp\u003eRegarding comorbidities, hypertension was not significantly associated with 5-year survival status in PCa patients (p\u0026thinsp;=\u0026thinsp;0.417). In contrast, diabetes mellitus was identified as an independent risk factor for 5-year mortality (multivariate OR\u0026thinsp;=\u0026thinsp;6.77, 95% CI: 1.98\u0026ndash;23.18, p\u0026thinsp;=\u0026thinsp;0.002), rather than showing no statistical relevance.\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\u003eProstate Cancer Patients' Characteristics Stratstate Cancer Patients' Characteristics Stratified by Survival Status\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;198)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003edead (N\u0026thinsp;=\u0026thinsp;198)\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\u003eAge\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\u003e71.5\u0026thinsp;\u0026plusmn;\u0026thinsp;4.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e73.0\u0026thinsp;\u0026plusmn;\u0026thinsp;4.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.08 (1.03\u0026ndash;1.13, p=.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.17 (1.06\u0026ndash;1.30, p=.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\u003e23.5\u0026thinsp;\u0026plusmn;\u0026thinsp;2.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e22.3\u0026thinsp;\u0026plusmn;\u0026thinsp;2.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.86 (0.80\u0026ndash;0.93, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.77 (0.66\u0026ndash;0.91, p=.002)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTBIL\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\u003e14.0\u0026thinsp;\u0026plusmn;\u0026thinsp;4.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e13.3\u0026thinsp;\u0026plusmn;\u0026thinsp;3.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.97 (0.92\u0026ndash;1.01, p=.134)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.96 (0.86\u0026ndash;1.07, p=.493)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDBIL\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\u003e4.7\u0026thinsp;\u0026plusmn;\u0026thinsp;2.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4.5\u0026thinsp;\u0026plusmn;\u0026thinsp;1.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.94 (0.84\u0026ndash;1.05, p=.271)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIBIL\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;3.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e9.0\u0026thinsp;\u0026plusmn;\u0026thinsp;2.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.96 (0.91\u0026ndash;1.02, p=.228)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eALB\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\u003e40.7\u0026thinsp;\u0026plusmn;\u0026thinsp;5.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e37.5\u0026thinsp;\u0026plusmn;\u0026thinsp;3.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.84 (0.79\u0026ndash;0.89, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.77 (0.67\u0026ndash;0.88, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBUN\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\u003e6.2\u0026thinsp;\u0026plusmn;\u0026thinsp;1.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e7.1\u0026thinsp;\u0026plusmn;\u0026thinsp;2.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.45 (1.26\u0026ndash;1.68, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.02 (1.49\u0026ndash;2.75, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCRE\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\u003e75.7\u0026thinsp;\u0026plusmn;\u0026thinsp;14.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e66.6\u0026thinsp;\u0026plusmn;\u0026thinsp;14.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.95 (0.94\u0026ndash;0.97, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.92 (0.89\u0026ndash;0.95, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUA\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\u003e349.3\u0026thinsp;\u0026plusmn;\u0026thinsp;84.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e311.0\u0026thinsp;\u0026plusmn;\u0026thinsp;87.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.99 (0.99-1.00, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.00 (0.99-1.00, p=.626)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eA_G\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\u003e1.5\u0026thinsp;\u0026plusmn;\u0026thinsp;0.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.5\u0026thinsp;\u0026plusmn;\u0026thinsp;0.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.39 (0.15\u0026ndash;1.04, p=.059)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.32 (0.12\u0026ndash;14.08, p=.819)\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\u003e6.0\u0026thinsp;\u0026plusmn;\u0026thinsp;1.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e6.1\u0026thinsp;\u0026plusmn;\u0026thinsp;1.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.04 (0.91\u0026ndash;1.19, p=.551)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNE\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\u003e3.6\u0026thinsp;\u0026plusmn;\u0026thinsp;1.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3.7\u0026thinsp;\u0026plusmn;\u0026thinsp;1.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.06 (0.91\u0026ndash;1.22, p=.464)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLYM\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\u003e3.0\u0026thinsp;\u0026plusmn;\u0026thinsp;12.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.7\u0026thinsp;\u0026plusmn;\u0026thinsp;0.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.78 (0.55\u0026ndash;1.12, p=.177)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.46 (0.40\u0026ndash;5.36, p=.565)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNOM\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\u003e0.5\u0026thinsp;\u0026plusmn;\u0026thinsp;0.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.5\u0026thinsp;\u0026plusmn;\u0026thinsp;0.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e6.48 (1.80-23.32, p=.004)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.43 (0.01-318.98, p=.896)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLMR\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\u003e6.5\u0026thinsp;\u0026plusmn;\u0026thinsp;22.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3.6\u0026thinsp;\u0026plusmn;\u0026thinsp;1.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.81 (0.71\u0026ndash;0.92, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.81 (0.40\u0026ndash;1.64, p=.553)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNLR\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\u003e2.3\u0026thinsp;\u0026plusmn;\u0026thinsp;1.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2.5\u0026thinsp;\u0026plusmn;\u0026thinsp;1.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.07 (0.94\u0026ndash;1.21, p=.301)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePLR\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\u003e114.4\u0026thinsp;\u0026plusmn;\u0026thinsp;62.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e122.7\u0026thinsp;\u0026plusmn;\u0026thinsp;48.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.00 (1.00-1.01, p=.150)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.00 (1.00-1.01, p=.331)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRBC\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\u003e4.4\u0026thinsp;\u0026plusmn;\u0026thinsp;0.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4.1\u0026thinsp;\u0026plusmn;\u0026thinsp;0.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.24 (0.14\u0026ndash;0.40, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.86 (0.26\u0026ndash;13.16, p=.535)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHB\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\u003e134.8\u0026thinsp;\u0026plusmn;\u0026thinsp;13.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e125.9\u0026thinsp;\u0026plusmn;\u0026thinsp;12.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.94 (0.93\u0026ndash;0.96, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.97 (0.90\u0026ndash;1.04, p=.346)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePLT\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\u003e184.3\u0026thinsp;\u0026plusmn;\u0026thinsp;59.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e187.7\u0026thinsp;\u0026plusmn;\u0026thinsp;39.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.00 (1.00-1.01, p=.507)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePT\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\u003e11.5\u0026thinsp;\u0026plusmn;\u0026thinsp;1.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e11.3\u0026thinsp;\u0026plusmn;\u0026thinsp;1.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.96 (0.85\u0026ndash;1.08, p=.488)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFIB\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\u003e2.9\u0026thinsp;\u0026plusmn;\u0026thinsp;0.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2.9\u0026thinsp;\u0026plusmn;\u0026thinsp;0.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.92 (0.68\u0026ndash;1.23, p=.565)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePSA\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\u003e33.2\u0026thinsp;\u0026plusmn;\u0026thinsp;28.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e57.0\u0026thinsp;\u0026plusmn;\u0026thinsp;69.2\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 (0.99\u0026ndash;1.02, p=.390)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e43 (21.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e15 (7.6%)\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\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e37 (18.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e37 (18.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2.87 (1.36\u0026ndash;6.03, p=.005)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.65 (0.17\u0026ndash;2.50, p=.526)\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\u003e72 (36.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e26 (13.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.04 (0.49\u0026ndash;2.17, p=.927)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.21 (0.05\u0026ndash;0.85, p=.029)\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\u003e19 (9.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e45 (22.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e6.79 (3.06\u0026ndash;15.04, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.72 (0.16\u0026ndash;3.19, p=.667)\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\u003e26 (13.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e70 (35.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e7.72 (3.68\u0026ndash;16.18, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.95 (0.47\u0026ndash;8.13, p=.358)\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\u003e1 (0.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5 (2.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e14.33 (1.55-132.77, p=.019)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e4.24 (0.10-173.02, p=.446)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePNI\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\u003e157 (79.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e85 (42.9%)\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\u003e41 (20.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e113 (57.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e5.09 (3.26\u0026ndash;7.94, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.99 (0.85\u0026ndash;4.68, p=.114)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePSM\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\u003e178 (89.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e116 (58.6%)\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\u003e20 (10.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e82 (41.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e6.29 (3.66\u0026ndash;10.82, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e6.56 (2.28\u0026ndash;18.84, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSVI\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\u003e175 (88.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e112 (56.6%)\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\u003e23 (11.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e86 (43.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e5.84 (3.48\u0026ndash;9.80, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e3.61 (1.31-10.00, p=.013)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003epT\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\u003e128 (64.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e39 (19.7%)\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\u003e68 (34.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e107 (54%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e5.16 (3.23\u0026ndash;8.26, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e3.93 (1.56\u0026ndash;9.92, p=.004)\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\u003e2 (1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e52 (26.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e85.33 (19.89-366.16, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e26.82 (4.29-167.86, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003epN\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\u003e194 (98%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e143 (72.2%)\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\u003e4 (2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e55 (27.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e18.65 (6.61\u0026ndash;52.66, p\u0026lt;.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e9.57 (1.77\u0026ndash;51.67, p=.009)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHypertension\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\u003e108 (54.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e116 (58.6%)\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\u003e90 (45.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e82 (41.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.85 (0.57\u0026ndash;1.26, p=.417)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDiabetes_mellitus\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\u003e180 (90.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e156 (78.8%)\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\u003e18 (9.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e42 (21.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2.69 (1.49\u0026ndash;4.87, p=.001)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e6.77 (1.98\u0026ndash;23.18, p=.002)\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 optimal feature selection process combining LASSO regression, RF, and RFE. The feature selection process for each individual algorithm is presented in Supplementary Fig.\u0026nbsp;1: Feature Selection Process of the Three Algorithms. We ultimately selected 10 features common to all three models (Age, ALB, BUN, CRE, HB, PLT, PT, PSA, GS, PNI, PSM, pT and pN) as input parameters for machine learning.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eModel Construction and Selection\u003c/h2\u003e \u003cp\u003eThe model results indicate that LightGBM and Random Forest achieved the highest Accuracy value, reaching a comprehensive best performance metric of 0.9328, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e. LightGBM, AdaBoost, and Random Forest demonstrated stronger generalization ability on the test set, with AUC values of 0.9778, 0.9758, and 0.9746, respectively, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e. (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\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 12 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\u003eGradient Boosting Machine\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9632\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\u003e0.9244\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.8640\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9074\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9538\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.9423\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.9254\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.9423\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.9074\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.9245\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\u003e0.9778\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0189\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9328\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.8467\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9074\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9538\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.9412\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.9118\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.9412\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.9074\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.9143\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9746\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0127\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9244\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.8472\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9074\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9385\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.9245\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.9242\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.9245\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.9074\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.9159\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\u003e0.9658\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0148\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9244\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.8477\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9259\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9231\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.9091\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.9375\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.9091\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.9259\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.9174\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\u003e0.9641\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0341\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9160\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.8305\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9074\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9231\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.9074\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.9231\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.9074\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.9074\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.9074\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\u003e0.9758\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0113\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9076\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.8127\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8704\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9385\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.9216\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.8971\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.9216\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.8704\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8952\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\u003e0.9282\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0675\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8824\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7612\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8333\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9231\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.9000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.8696\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.9000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.8333\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8654\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\u003e0.9265\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0654\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8824\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7612\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8333\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9231\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.9000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.8696\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.9000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.8333\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8654\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\u003e0.9296\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0608\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8824\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7619\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8519\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9077\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8846\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.8806\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8846\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.8519\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8679\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eC5.0 Decision Tree\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9510\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0192\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8739\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7453\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8519\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.8923\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8679\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.8788\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8679\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.8519\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8598\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDeep Neural Network\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9561\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0499\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8739\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7453\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8519\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.8923\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8679\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.8788\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8679\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.8519\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8598\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\u003e0.9342\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0658\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8739\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7437\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8148\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9231\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8980\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.8571\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8980\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.8148\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8544\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\u003e0.9450\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0350\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8739\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7429\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.7963\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9385\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.9149\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.8472\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.9149\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.7963\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8515\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\u003e0.9348\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0686\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8739\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7437\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8148\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9231\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8980\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.8571\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8980\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.8148\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8544\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\u003e0.9425\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0333\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8655\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7279\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8333\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.8923\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8654\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.8657\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8654\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.8333\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8491\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGaussian Process\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9564\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0528\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8655\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7271\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8148\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9077\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8800\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.8551\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8800\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.8148\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8462\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\u003e0.9219\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0599\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8655\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7271\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8148\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9077\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8800\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.8551\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8800\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.8148\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8462\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\u003e0.9370\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0728\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8655\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7262\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.7963\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9231\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8958\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.8451\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8958\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.7963\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8431\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\u003e0.9405\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0756\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8655\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7262\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.7963\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9231\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8958\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.8451\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8958\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.7963\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8431\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\u003e0.9239\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0484\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8571\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7141\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8889\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.8308\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8136\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.9000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8136\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.8889\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8496\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\u003e0.8709\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0803\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8067\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.6191\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9259\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.7077\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.7246\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.9200\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.7246\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.9259\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.8130\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\u003e0.8999\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0589\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8067\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.6033\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.6852\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.9077\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.8605\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.7763\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.8605\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.6852\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.7629\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\u003e0.8177\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0925\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7647\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.5268\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.7593\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.7692\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.7321\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.7937\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.7321\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.7593\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.7455\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\u003e0.7709\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.1243\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.6807\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.3829\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9074\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.4923\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.5976\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.8649\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.5976\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.9074\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.7206\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\u003e0.9328\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0910\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.5882\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.1003\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.0926\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e1.0000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.0000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.5702\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e1.0000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.0926\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.1695\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 \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eClinical application of the model\u003c/h2\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAll models' calibration curves were close to the ideal dashed line, and the curve-fitting quality of most models was high (see Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e). The error bars (vertical lines) corresponding to each group were generally short, indicating smaller fluctuations in the actual event proportions within each group. This suggests that these predictive models had good calibration in the test set, with a high match between the model's predicted death risk and patients' actual 5-year mortality rates, further verifying the reliability of the model's predictions. The quantitative indicators of calibration performance for each model, such as Expected Calibration Error (ECE) and Maximum Calibration Error (MCE), are shown in Supplementary Table\u0026nbsp;1.The Brier scores of the 12 models are shown in Supplementary Table\u0026nbsp;2.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e shows the results of clinical decision curve analysis for the 12 models. All models showed significantly higher net benefits than the two extreme strategies of \u0026ldquo;no intervention for any patient\u0026rdquo; (None) and \u0026ldquo;intervention for all patients\u0026rdquo; (All) within most clinical threshold probability ranges. Only a few models (e.g., FDA) showed slight decreases in net benefits at low thresholds. These machine learning models all provided substantial net benefits in the clinical decision scenarios of this study, and there were small differences in their clinical practical value, indicating that the practical value of the models in guiding clinical decision-making was superior to traditional experiential judgment.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eModel interpretability and clinical interactive application\u003c/h2\u003e \u003cp\u003eThe SHAP value analysis of model feature importance (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e) showed that CRE had the greatest impact (+\u0026thinsp;0.0953), followed by PSA (+\u0026thinsp;0.0782), PSM (+\u0026thinsp;0.0769), and pT (+\u0026thinsp;0.0598). ALB, PLT, HB, GS, and BUN made smaller contributions to the model's prediction but remained statistically significant.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThis interactive web application developed with R Shiny (the tool shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e) is a predictive system designed specifically for prostate cancer patients to assess their 5-year postoperative survival status, equipped with multiple practical functions: it supports medical staff in entering 13 clinical and pathological parameters and conducts comprehensive analysis of these indicators through machine learning algorithms. In addition to outputting predictive conclusions on patients' 5-year postoperative survival status, it also generates a visualization report of feature importance based on the SHAP values of the LightGBM model, helping medical staff intuitively identify which indicators (such as tumor stage, PSA level, etc.) have a more significant impact on survival prognosis.\u003c/p\u003e \u003cp\u003eThe online prediction system features a concise, user-friendly interface. After completing parameter input, the medical staff only needs to click the \"Predict\" button to quickly obtain a comprehensive evaluation report including predictive results and feature analysis.\u003c/p\u003e \u003cp\u003eCurrently, this application has been officially launched (access address: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://medicalpredictor.shinyapps.io/Online_Prediction_of_Risk_Factors/\u003c/span\u003e\u003cspan address=\"https://medicalpredictor.shinyapps.io/Online_Prediction_of_Risk_Factors/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) and is free for medical staff to use. It can be directly integrated into the clinical evaluation workflow, providing data-driven support for clinical prognostic assessment..\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eBased on the data of prostate cancer patients from The First Affiliated Hospital of the University of Science and Technology of China (USTC), this study first systematically applied a variety of feature selection algorithms to screen 13 key predictive variables from a large number of potential predictors, and then constructed and validated a machine learning-based predictive model for the 5-year postoperative survival of prostate cancer patients. This streamlined and efficient survival assessment tool integrates multiple machine learning algorithms, enabling accurate, reliable, and personalized predictions.\u003c/p\u003e \u003cp\u003eIn rigorous performance evaluation, the LightGBM model achieved excellent predictive performance for prostate cancer survival status, with an AUC of 0.9778, sensitivity of 0.9074, and specificity of 0.9538. This study conducted an in-depth analysis of the importance and influencing mechanisms of each variable using the SHAP algorithm, further clarifying the impact of different indicators on survival status.\u003c/p\u003e \u003cp\u003eThe model demonstrates robust predictive performance in both validation and verification. It not only provides a feasible solution for individualized treatment evaluation of prostate cancer but also, through its post-operative 5-year survival prediction system, constructed using the LightGBM model, offers significant guidance for clinical practice. This system can provide data to help doctors formulate more precise, rational, individualized treatment plans for prostate cancer.\u003c/p\u003e \u003cp\u003eIn recent years, some researchers have conducted AI-based prognostic studies on prostate cancer, though they mainly utilize clinical MRI and histopathological data.\u003c/p\u003e \u003cp\u003eWong et al. analyzed 338 patients with localized prostate cancer who underwent robot-assisted prostatectomy. They employed three supervised machine learning algorithms and 19 distinct training variables (covering demographic, clinical, imaging, and surgical data) to construct a model in a hypothesis-free manner that predicts the risk of biochemical recurrence within 1 year after surgery [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Zhou et al. retrospectively collected data from 399 prostate cancer patients across three medical centers. They extracted the maximum regions of interest (ROI) from three MRI sequences, trained a deep learning model using H\u0026amp;E-stained sections, and built a nomogram for the combined model. By integrating multimodal data, they developed an AI-based predictive model for the progression of castration-resistant prostate cancer (CRPC) [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eLee et al. established a model integrating clinical data and deep learning to predict long-term biochemical recurrence-free survival (BCRFS) in PCa patients following radical prostatectomy (RP), utilizing multiparametric magnetic resonance imaging (mpMRI) [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Hou et al. initially developed MRI radiomics features (RadS) for biochemical recurrence (BCR), and employed two pre-specified AI models to forecast the tumor's T3 stage and lymph node metastasis (LN+). Subsequently, clinical, imaging, and histopathological variables were incorporated into iBCR-Net for BCR prediction [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e].\u003c/p\u003e \u003cp\u003ePolymeri et al. developed an AI-driven algorithm to automatically quantify the prostate and its tumor content in PET/CT images of 145 patients. This algorithm automatically acquired prostate tumor volume, tumor-to-prostate ratio, whole-tumor lesion uptake rate, and SUVmax values. Subsequently, the Cox proportional hazards regression model was applied to explore the correlations between these measurements and age, PSA, Gleason score, as well as prostate cancer-specific survival (PCSS) [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. Koo et al. adopted an artificial neural network (ANN) model to predict survival outcomes under different initial treatment regimens and established an online decision-support system. Compared with the Cox proportional hazards regression model, the ANN model exhibited superior predictive performance in forecasting 5-year and 10-year progression-free survival (PFS) in castration-resistant prostate cancer (CRPC), cancer-specific survival (CSS), and overall survival (OS) [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe aforementioned literature has laid a solid foundation for prostate cancer research by integrating clinical, imaging, and pathological data to provide diverse approaches for model construction and clinical references for different prognostic objectives. Compared with previous studies, this study simplified data sources, developed a more user-friendly survival prediction model, and conducted a comprehensive and detailed comparison and optimization of various machine learning models. Through the study, it was found that T-stage, PSA, BUN, CRE, and Age have a significant impact on the 5-year survival status of prostate cancer patients after surgery. From Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, it was observed that at the T2 stage, the survival and death rates were 186 (62.8%) and 63 (21.3%), respectively, with the number of survivors significantly higher than that of non-survivors. However, at the T4 stage, the survival and death rates were 1 (0.3%) and 75 (25.3%), respectively, indicating an extremely poor prognosis.\u003c/p\u003e \u003cp\u003ePSA levels serve as a key basis for prostate cancer monitoring and are critical to its progression [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. Fluctuations in PSA levels may act as an early indicator of BCR, rendering continuous PSA monitoring a vital step in tracking changes in prostate cancer status [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. However, PSA level changes are not always directly associated with cancer recurrence. For instance, following certain specific treatments (e.g., high-intensity focused ultrasound, HIFU), PSA levels may surge sharply, which does not directly reflect the actual prostate cancer status [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. Thus, it is essential to acknowledge the limitations of PSA testing, as these can lead to unnecessary additional examinations and treatments, further intensifying patients' psychological distress and increasing medical expenses [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eA large-scale study of 160,787 U.S. men who underwent radical prostatectomy revealed that patients diagnosed at an older age had a higher prostate cancer mortality rate [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. Among prostate cancer patients aged 75 years and above, cryotherapy demonstrated superior survival outcomes compared to other surgical methods for this specific age group [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]. Nevertheless, academic debate persists regarding the impact of age on prostate cancer mortality risk. A study by Pettersson et al. found no significant inherent effect of age on prostate cancer mortality risk; instead, it indicated that in current clinical practice, elderly prostate cancer patients may not receive adequate diagnostic evaluations and subsequent radical treatments [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTherefore, when formulating treatment plans for prostate cancer, it is necessary to develop individualized and precise plans based on the specific conditions of each patient. Particularly for the elderly prostate cancer population, given the extension of life expectancy, it is crucial to strictly adhere to the screening and treatment guidelines for prostate cancer and determine the most suitable individualized treatment strategies for them.\u003c/p\u003e \u003cp\u003eIn this study, the LightGBM model exhibited superior performance to other machine learning algorithms. As a highly efficient gradient boosting framework, LightGBM not only excels in capturing complex nonlinear relationships and interactions among variables but also attains faster training speeds and reduced memory usage via histogram-based algorithms and one-sided gradient sampling techniques [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eMachine learning methods offer significant advantages over traditional statistical methods for addressing complex, nonlinear clinical problems. The core innovation of our approach lies in the organic integration of model interpretability and clinical utility. Through SHAP value analysis, we can clearly identify the key factors influencing individual risk prediction, thereby enhancing clinicians' trust in the model's results and supporting the development of targeted intervention strategies [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. This interpretability is crucial for the clinical implementation of machine learning models in decision support systems.\u003c/p\u003e \u003cp\u003eFurthermore, the interactive web app we developed converts complex machine learning outputs into intuitive clinical tools, enabling real-time risk evaluation and visualization of feature significance, thereby effectively addressing the \"black box\" issue of machine learning models [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]. Notably, this user-friendly interface design is consistent with Shah et al.'s findings, highlighting its pivotal role in promoting the clinical implementation of machine learning technologies [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eMachine learning methods offer significant advantages over traditional statistical methods for addressing complex, nonlinear clinical problems. The core innovation of our approach lies in the organic integration of model interpretability and clinical utility. Through SHAP value analysis, we can clearly identify the key factors influencing individual risk prediction, thereby enhancing clinicians' trust in the model's results and supporting the development of targeted intervention strategies [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. This interpretability is crucial for the clinical implementation of machine learning models in decision support systems.\u003c/p\u003e \u003cp\u003eIn addition, the interactive web application developed by our team converts the complex outputs of machine learning into an intuitive clinical tool, which enables real-time risk assessment and feature importance visualization. It effectively mitigates the black-box problem of machine learning models in clinical prediction, though there is still room for improvement regarding the comprehensiveness of interpretation [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]. Notably, this user-friendly interface design is consistent with Shah et al.'s findings, highlighting its pivotal role in promoting the clinical implementation of machine learning technologies [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThis study employs machine learning methods combined with the SHAP tool to construct a predictive model, thereby enhancing its interpretability. By providing real-time predictions and clinical recommendations, it helps clinicians make more accurate, personalized treatment decisions. However, this study has the following limitations: First, the dataset was based on retrospective clinical data, which may lack dynamic monitoring indicators, such as long-term follow-up data on patients\u0026rsquo; quality of life, treatment compliance, and adverse reaction records. Meanwhile, incomplete inclusion of lost-to-follow-up patients has led to potential selection bias. Furthermore, the total sample size of this study was 300 cases, which is relatively limited, and all samples were collected from a single center. This may compromise the external validity and generalizability of the model, whose applicability to multi-center and multi-ethnic populations requires further verification. Moreover, the current model was constructed based on routine clinical indicators (e.g., T stage, prostate-specific antigen, blood urea nitrogen, creatinine, and age), without incorporating imaging data (e.g., magnetic resonance imaging and computed tomography) and genomic data (e.g., gene mutation status). Thus, the comprehensiveness and depth of the feature variables need to be further improved. Finally, this study focused on predicting the 5-year survival status, and the predictive efficacy of the model for long-term survival (e.g., 10-year survival) has not been verified. Future research should extend the follow-up duration and supplement follow-up data to optimize the model.\u003c/p\u003e \u003cp\u003eTo address these aforementioned issues, the research team aims to expand and enrich the dataset by integrating and harmonizing multi-source data, encompassing imaging data, genomic profiles, and other granular clinical covariates to construct a more comprehensive and precise predictive model. In turn, this model will offer more robust evidentiary support for prostate cancer treatment decision-making, thereby enhancing the scientific rigor and clinical translatability of individualized treatment strategies and prognostic assessment [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e].\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis study conducted relevant explorations on predicting optimal treatment regimens for prostate cancer patients using multiple machine learning models. Among these machine learning models, the LightGBM model exhibited the best comprehensive performance with an AUC value of 0.9778, indicating that this model has good efficacy in predicting treatment regimens for prostate cancer within the scope of the dataset used in this study. This study analyzed the internal mechanism of the LightGBM model using the SHAP tool and found that five indicators, namely T stage, CRE, PSA, PSM, and ALB, were significantly correlated with the 5-year postoperative survival rate of prostate cancer patients. By integrating machine learning algorithms with SHAP interpretability analysis, this study constructed a model applicable for predicting treatment regimens for prostate cancer. Its predictive efficacy has been verified to reach a high level in this study, providing potential reference and technical support for the clinical implementation of precise individualized decision-making.\u003c/p\u003e "},{"header":"Declarations","content":"\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eData Availability Statement\u003c/h2\u003e \u003cp\u003eData are available from the corresponding author upon reasonable request.\u003c/p\u003e \u003c/div\u003e \u003cp\u003e \u003cstrong\u003eConflicts of interest\u003c/strong\u003e \u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eEthics approval\u003c/strong\u003e \u003cp\u003e This study was conducted in accordance with the principles of the Declaration of Helsinki and its subsequent revisions. The study was approved by the Bioethics Committee of The First Affiliated Hospital of University of Science and Technology of China (No.2024-RE-272). Informed consent or substitute for it was obtained from all patients for being included in the study.\u003c/p\u003e \u003c/p\u003e\u003cp\u003e \u003ch2\u003eClinical trial number\u003c/h2\u003e \u003cp\u003eNot applicable.\u003c/p\u003e \u003c/p\u003e\u003cp\u003e \u003ch2\u003eConsent for publication\u003c/h2\u003e \u003cp\u003eNot applicable.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThis research was funded by the Anhui Province Teaching Research Program (Granted No. 2023jyxm0347).\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eData Analysis: T.S., and C.H; Writing Original Draft: T.S., and C.H.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eSung H, Ferlay J, Siegel RL, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin. 2021;71(3):209\u0026ndash;49. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3322/caac.21660\u003c/span\u003e\u003cspan address=\"10.3322/caac.21660\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMiller KD, Nogueira L, Devasia T, et al. Cancer treatment and survivorship statistics, 2022. CA Cancer J Clin. 2022;72(5):409\u0026ndash;36. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3322/caac.21731\u003c/span\u003e\u003cspan address=\"10.3322/caac.21731\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSiegel RL, Miller KD, Wagle NS, Jemal A, Cancer statistics. 2023. CA Cancer J Clin. 2023;73(1):17\u0026ndash;48. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3322/caac.21763\u003c/span\u003e\u003cspan address=\"10.3322/caac.21763\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGlobal Burden of Disease Collaborative Network. Global Burden of Disease Study 2023 (GBD 2023). Seattle, United States: Institute for Health Metrics and Evaluation (IHME), 2025. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://vizhub.healthdata.org/gbd-results/\u003c/span\u003e\u003cspan address=\"https://vizhub.healthdata.org/gbd-results/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu B, Lu X, Shen H, et al. Intratumoral heterogeneity and genetic characteristics of prostate cancer. Int J Cancer. 2020;146(12):3369\u0026ndash;78. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/ijc.32961\u003c/span\u003e\u003cspan address=\"10.1002/ijc.32961\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePanaiyadiyan S, Kumar R. Prostate cancer nomograms and their application in Asian men: a review. Prostate Int. 2024;12(1):1\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.prnil.2023.07.004\u003c/span\u003e\u003cspan address=\"10.1016/j.prnil.2023.07.004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eL\u0026oacute;pez-Abad A, Ram\u0026iacute;rez Backhaus M, Server G\u0026oacute;mez G, et al. Real-world prostate-specific antigen reduction and survival outcomes of metastatic hormone-sensitive prostate cancer patients treated with apalutamide: An observational, retrospective, and multicentre study. Prostate Int. 2024;12(1):20\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.prnil.2023.10.003\u003c/span\u003e\u003cspan address=\"10.1016/j.prnil.2023.10.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLitwin MS, Tan HJ. The Diagnosis and Treatment of Prostate Cancer: A Review. JAMA. 2017;317(24):2532\u0026ndash;42. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1001/jama.2017.7248\u003c/span\u003e\u003cspan address=\"10.1001/jama.2017.7248\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen AB, Niu J, Cronin AM, Shih YT, Giordano S, Schrag D. Variation in Use of High-Cost Technologies for Palliative Radiation Therapy by Radiation Oncologists. J Natl Compr Canc Netw. 2021;19(4):421\u0026ndash;31. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.6004/jnccn.2020.7633\u003c/span\u003e\u003cspan address=\"10.6004/jnccn.2020.7633\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Published 2021 Feb 12.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFerlay JEM, Lam F, Colombet M, Mery L, Pineros et al. Global Cancer Observatory: Cancer Tomorrow; International Agency for Research on Cancer: Lyon, France; Available online: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://gco.iarc.fr/tomorrow\u003c/span\u003e\u003cspan address=\"https://gco.iarc.fr/tomorrow\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFu ZT, Guo XL, Zhang SW, et al. Analysis of incidence and death of prostate cancer in China in 2015. Chin J Oncol. 2020;42(9):718\u0026ndash;22.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZheng RS, Chen R, Han BF, et al. Zhonghua Zhong Liu Za Zhi. 2024;46(3):221\u0026ndash;31. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3760/cma.j.cn112152-20240119-00035\u003c/span\u003e\u003cspan address=\"10.3760/cma.j.cn112152-20240119-00035\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHe J, Chen WQ, Li N, et al. Guidelines for screening, early diagnosis, and early treatment of prostate cancer in China (2022, Beijing). Chin J Oncol. 2022;44(1):29\u0026ndash;53.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSatturwar S, Parwani AV. Artificial Intelligence-Enabled Prostate Cancer Diagnosis and Prognosis: Current State and Future Implications. Adv Anat Pathol. 2024;31(2):136\u0026ndash;44. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/PAP.0000000000000425\u003c/span\u003e\u003cspan address=\"10.1097/PAP.0000000000000425\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRen S, Li J, Dorado J, et al. From molecular mechanisms of prostate cancer to translational applications: based on multi-omics fusion analysis and intelligent medicine. Health Inf Sci Syst. 2023;12(1):6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s13755-023-00264-5\u003c/span\u003e\u003cspan address=\"10.1007/s13755-023-00264-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Published 2023 Dec 18.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBulten W, Kartasalo K, Chen PC, et al. Artificial intelligence for diagnosis and Gleason grading of prostate cancer: the PANDA challenge. Nat Med. 2022;28(1):154\u0026ndash;63. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41591-021-01620-2\u003c/span\u003e\u003cspan address=\"10.1038/s41591-021-01620-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKench JG, Amin MB, Berney DM, et al. WHO Classification of Tumours fifth edition: evolving issues in the classification, diagnosis, and prognostication of prostate cancer. Histopathology. 2022;81(4):447\u0026ndash;58. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/his.14711\u003c/span\u003e\u003cspan address=\"10.1111/his.14711\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTang S, Zhang H, Liang J, et al. Prostate cancer treatment recommendation study based on machine learning and SHAP interpreter. Cancer Sci. 2024;115(11):3755\u0026ndash;66. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/cas.16327\u003c/span\u003e\u003cspan address=\"10.1111/cas.16327\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlataş E, Tanyıldızı K\u0026ouml;kk\u0026uuml;l\u0026uuml;nk H, Tanyıldızı H, Alcın G. Treatment prediction with machine learning in prostate cancer patients. Comput Methods Biomech Biomed Engin. 2025;28(4):572\u0026ndash;80. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1080/10255842.2023.2298364\u003c/span\u003e\u003cspan address=\"10.1080/10255842.2023.2298364\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHu S, Chang CP, Snyder J, et al. Comparing Active Surveillance and Watchful Waiting With Radical Treatment Using Machine Learning Models Among Patients With Prostate Cancer. JCO Clin Cancer Inf. 2023;7:e2300083. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1200/CCI.23.00083\u003c/span\u003e\u003cspan address=\"10.1200/CCI.23.00083\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSyed K, Sleeman W, Soni P, et al. Machine-Learning Models for Multicenter Prostate Cancer Treatment Plans. J Comput Biol. 2021;28(2):166\u0026ndash;84. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1089/cmb.2020.0188\u003c/span\u003e\u003cspan address=\"10.1089/cmb.2020.0188\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAuffenberg GB, Ghani KR, Ramani S, et al. askMUSIC: Leveraging a Clinical Registry to Develop a New Machine Learning Model to Inform Patients of Prostate Cancer Treatments Chosen by Similar Men. Eur Urol. 2019;75(6):901\u0026ndash;7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.eururo.2018.09.050\u003c/span\u003e\u003cspan address=\"10.1016/j.eururo.2018.09.050\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWong NC, Lam C, Patterson L, Shayegan B. Use of machine learning to predict early biochemical recurrence after robot-assisted prostatectomy. BJU Int. 2019;123(1):51\u0026ndash;7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/bju.14477\u003c/span\u003e\u003cspan address=\"10.1111/bju.14477\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhou C, Zhang YF, Guo S et al. Multimodal data integration for predicting progression risk in castration-resistant prostate cancer using deep learning: a multicenter retrospective study. Front Oncol. 2024;14:1287995. Published 2024 Mar 14. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fonc.2024.1287995\u003c/span\u003e\u003cspan address=\"10.3389/fonc.2024.1287995\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee HW, Kim E, Na I, Kim CK, Seo SI, Park H. Novel Multiparametric Magnetic Resonance Imaging-Based Deep Learning and Clinical Parameter Integration for the Prediction of Long-Term Biochemical Recurrence-Free Survival in Prostate Cancer after Radical Prostatectomy. Cancers (Basel). 2023;15(13):3416. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/cancers15133416\u003c/span\u003e\u003cspan address=\"10.3390/cancers15133416\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Published 2023 Jun 29.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHou Y, Jiang KW, Wang LL, et al. Biopsy-free AI-aided precision MRI assessment in prediction of prostate cancer biochemical recurrence. Br J Cancer. 2023;129(10):1625\u0026ndash;33. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41416-023-02441-5\u003c/span\u003e\u003cspan address=\"10.1038/s41416-023-02441-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePolymeri E, Kj\u0026ouml;lhede H, Enqvist O, et al. Artificial intelligence-based measurements of PET/CT imaging biomarkers are associated with disease-specific survival of high-risk prostate cancer patients. Scand J Urol. 2021;55(6):427\u0026ndash;33. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1080/21681805.2021.1977845\u003c/span\u003e\u003cspan address=\"10.1080/21681805.2021.1977845\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKoo KC, Lee KS, Kim S, et al. Long short-term memory artificial neural network model for prediction of prostate cancer survival outcomes according to initial treatment strategy: development of an online decision-making support system. World J Urol. 2020;38(10):2469\u0026ndash;76. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00345-020-03080-8\u003c/span\u003e\u003cspan address=\"10.1007/s00345-020-03080-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHoey C, Ahmed M, Fotouhi Ghiam A, et al. Circulating miRNAs as non-invasive biomarkers to predict aggressive prostate cancer after radical prostatectomy. J Transl Med. 2019;17(1):173. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12967-019-1920-5\u003c/span\u003e\u003cspan address=\"10.1186/s12967-019-1920-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Published 2019 May 23.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePlanas J, Morote J. Criterios de recidiva bioqu\u0026iacute;mica despu\u0026eacute;s de prostatectom\u0026iacute;a radical. Historia natural de la enfermedad [Biochemical recurrence criteria after radical prostatectomy. Natural history of the disease]. Arch Esp Urol. 2012;65(1):4\u0026ndash;11.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eInamoto T, Komura K, Watsuji T, Azuma H. Rapid Increase of the Serum PSA Level in Response to High-Intensity Focused Ultrasound Therapy may be a Potential Indicator of Biochemical Recurrence of Low- and Intermediate-Risk Prostate Cancer. Clin Med Insights Oncol. 2011;5:101\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.4137/CMO.S7073\u003c/span\u003e\u003cspan address=\"10.4137/CMO.S7073\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLamallem H, et al. El\u0026eacute;vation du PSA apr\u0026egrave;s irradiation prostatique: rebond ou r\u0026eacute;cidive biologique? [Elevation of PSA after prostate radiotherapy: rebound or biochemical recurrence?]. Prog Urol. 2008;18(9):557\u0026ndash;61. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.purol.2008.06.001\u003c/span\u003e\u003cspan address=\"10.1016/j.purol.2008.06.001\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXia SJ, Jing YF. Biochemical recurrence of prostate cancer: the controversial recognition and management. Chin Med J (Engl). 2011;124(9):1283\u0026ndash;5.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbdollah F, Sun M, Thuret R, et al. A competing-risks analysis of survival after alternative treatment modalities for prostate cancer patients: 1988\u0026ndash;2006. Eur Urol. 2011;59(1):88\u0026ndash;95. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.eururo.2010.10.003\u003c/span\u003e\u003cspan address=\"10.1016/j.eururo.2010.10.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZheng Y, Ye Y, Chen L, et al. Prevalence and outcomes of focal ablation versus prostatectomy for elderly patients with prostate cancer: a population-based study. J Natl Cancer Cent. 2021;2(1):25\u0026ndash;32. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jncc.2021.11.005\u003c/span\u003e\u003cspan address=\"10.1016/j.jncc.2021.11.005\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Published 2021 Nov 26.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePettersson A, Robinson D, Garmo H, Holmberg L, Stattin P. Age at diagnosis and prostate cancer treatment and prognosis: a population-based cohort study. Ann Oncol. 2018;29(2):377\u0026ndash;85. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/annonc/mdx742\u003c/span\u003e\u003cspan address=\"10.1093/annonc/mdx742\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuolinKe QM, Finley T, Wang T, Chen W, Ma W, Ye Q, Liu TY, Lightgbm. A highly efficient gradient boosting decision tree. Adv Neural Inf Process Syst. 2017;30:52.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBernard D, Doumard E, Ader I, et al. Explainable machine learning framework to predict personalized physiological aging. Aging Cell. 2023;22(8):e13872. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/acel.13872\u003c/span\u003e\u003cspan address=\"10.1111/acel.13872\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRudin C. Stop Explaining Black Box Machine Learning Models for High Stakes Decisions and Use Interpretable Models Instead. Nat Mach Intell. 2019;1(5):206\u0026ndash;15. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s42256-019-0048-x\u003c/span\u003e\u003cspan address=\"10.1038/s42256-019-0048-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShah NH, Milstein A, Bagley SC. Making machine learning models clinically useful. JAMA. 2019;322(14):1351\u0026ndash;2.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu M, Luo J, Li L et al. Design and development of a disease-specific clinical database system to increase the availability of hospital data in China. Health Inf Sci Syst. 2023;11(1):11. Published 2023 Jan 30. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s13755-023-00211-4\u003c/span\u003e\u003cspan address=\"10.1007/s13755-023-00211-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-geriatrics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bgtc","sideBox":"Learn more about [BMC Geriatrics](http://bmcgeriatr.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bgtc/default.aspx","title":"BMC Geriatrics","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Prostate Cancer, machine learning, elderly people, SHAP, Shiny","lastPublishedDoi":"10.21203/rs.3.rs-8665949/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8665949/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eMachine Learning (ML) models have achieved outstanding performance in predicting post-surgical survival. However, the \u0026ldquo;black-box\u0026rdquo; nature of ML models restricts their clinical application. This study aims to develop and validate a clinically feasible, machine learning-based online prediction system for predicting the 5-year survival status of patients with prostate cancer (PCa) after surgery.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eThis study conducted a retrospective analysis of clinical data from 300 elderly patients with PCa aged\u0026thinsp;\u0026ge;\u0026thinsp;65 years. LASSO regression, random forest (RF), and recursive feature elimination (RFE) were employed to screen for clinical parameters. Subsequently, the cohort was split into a training set and a test set at a 7:3 ratio, and 25 machine learning approaches were utilized for comparative assessment to determine the optimal model. Calibration curves were applied to evaluate the performance of each model, while decision curve analysis (DCA) was adopted to assess their clinical usefulness. In addition, SHAP (Shapley Additive exPlanations) values were used to interpret the model features. Finally, the Shiny framework was employed to develop an online prediction system.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eThe intersection of the three feature selection algorithms identified 13 clinical parameters: Age, ALB, BUN, CRE, HB, PLT, PT, PSA, GS, PNI, PSM, pT and pN. Comparing 25 machine learning algorithms, LightGBM gave the best results. Its performance metrics were: Accuracy 0.9328, Sensitivity 0.9074, Specificity 0.9538, Positive Predictive Value (PPV) 0.9418, Negative Predictive Value (NPV) 0.9118, AUC 0.9778, Recall 0.9074, F1 Score 0.9143. SHAP values revealed the contribution of each feature in the LightGBM model.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eThis study successfully developed a 5-year survival prediction model for prostate cancer patients after surgery, providing guidance for clinical decision-making, optimizing treatment efficacy, and improving quality of life.\u003c/p\u003e","manuscriptTitle":"Explainable and Visualizable Machine Learning Model Development and Validation for 5-Year Postoperative Survival Prediction in Prostate Cancer Patients Aged ≥ 65 Years","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-04 09:55:59","doi":"10.21203/rs.3.rs-8665949/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-03-06T12:50:14+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-02-24T21:09:46+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"124417475159380125223760386787113014498","date":"2026-02-14T16:47:02+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"60700160016301651466318702774073278375","date":"2026-02-13T12:53:17+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-02-10T18:22:12+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"30509533563998291219390703400878125898","date":"2026-02-02T09:40:19+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-02-02T09:16:10+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-01-29T11:51:06+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-01-29T04:27:39+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-01-29T02:05:31+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Geriatrics","date":"2026-01-29T01:55:17+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-geriatrics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bgtc","sideBox":"Learn more about [BMC Geriatrics](http://bmcgeriatr.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bgtc/default.aspx","title":"BMC Geriatrics","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"afa8d4dc-51b9-41b6-9356-dbbf23800444","owner":[],"postedDate":"February 4th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-04-21T08:11:09+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-04 09:55:59","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8665949","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8665949","identity":"rs-8665949","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}