A High-Precision Machine Learning-Based Prediction Model for Delayed Graft functon(DGF) in Chinese Kidney Transplant Patients: A Multicenter Study

A High-Precision Machine Learning-Based Prediction Model for Delayed Graft functon(DGF) in Chinese Kidney Transplant Patients: A Multicenter Study | 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 Article A High-Precision Machine Learning-Based Prediction Model for Delayed Graft functon(DGF) in Chinese Kidney Transplant Patients: A Multicenter Study Ying Cheng, he sun, ping sun, zheng ding, xi wang, long he, ke xin ma, and 6 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5617823/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Delayed graft function (DGF) is a severe complication following kidney transplantation, and currently, there is a lack of accurate prediction tools tailored for the Chinese population. This study integrates data from 1,093 kidney transplant cases across four medical centers in China (2016–2024) to develop and validate a machine learning-based model for DGF prediction. By comparing nine machine learning algorithms, we found that the LightGBM model performed best in external validation (AUC = 0.80, accuracy = 0.73). SHAP analysis identified donor GFR, donor hemoglobin, and recipient plasma BNP levels as the primary predictive factors, while also highlighting novel predictors such as donor microscopic hematuria and APTT. Cox regression analysis showed that preoperative dialysis duration in recipients (HR = 1.006, 95% CI: 1.001–1.012) was an independent predictor of DGF recovery. In the follow-up study, we observed that while the DGF mortality group exhibited the most significant kidney function impairment (serum creatinine β = 200.57, eGFR β = -39.91), the prognosis of the DGF survival group was comparable to that of the non-DGF survival group. Additionally, the duration of DGF (16.66 ± 13.73 vs. 15.44 ± 14.62 days) and the number of dialysis treatments (8.13 ± 7.39 vs. 7.78 ± 7.22 sessions) were not significantly associated with prognosis. Based on these findings, we developed an online prediction platform (www.kidney-dgf-match.cn) to support clinical decision-making. This study not only establishes the first high-precision DGF prediction model for the Chinese population but also reveals the potential for favorable outcomes in DGF patients with proper management, offering new insights for optimizing post-transplant management strategies. Health sciences/Nephrology Health sciences/Medical research/Translational research Health sciences/Medical research/Outcomes research Health sciences/Medical research/Experimental models of disease Delayed Graft Function (DGF) Kidney Transplantation Machine Learning Multicenter Study Post-Transplant Outcomes Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 1 Introduction Delayed Graft Function (DGF) is a common and severe complication following kidney transplantation, typically defined as the need for dialysis within the first week post-transplantation [ 1 ]. Over the past decade, the incidence of DGF has shown a significant upward trend, reaching 26.3% among adult kidney transplant recipients in 2022 [ 2 ]. The occurrence of DGF is influenced by a variety of complex factors, including donor- and recipient-related variables, with its molecular mechanisms primarily involving pathological processes such as ischemia-reperfusion injury and immune responses. Clinically, the most direct and typical symptom of DGF is anuria or oliguria [ 3 ]. For patients, DGF not only significantly increases hospitalization costs but also severely undermines their hope of escaping dialysis, heightening anxiety about the success of the transplant. For clinicians, DGF complicates perioperative management, making it more challenging and increasing the risk of graft loss and patient mortality. Even more concerning is the poor prognosis associated with DGF. Numerous studies have shown that DGF significantly reduces long-term graft survival rates and patient survival rates. Moreover, DGF increases the risk of complications such as rejection and infection, subjecting patients to prolonged physical and psychological distress, which can even be life-threatening [ 4 ]. Given the significance of DGF, researchers in recent years have proposed several scoring systems to predict and reduce its incidence. For instance, Irish et al. developed a prediction model based on five risk factors for the North American population [ 5 ], while Marion Chapal et al. proposed a similar model for the French population [ 6 ]. However, these existing models have limitations, including relatively low accuracy and reliance on single-source data, preventing them from gaining widespread consensus. Notably, as the country with the second-highest number of kidney transplants globally, China still lacks a localized DGF prediction model. This highlights the urgent need to develop a DGF prediction model tailored to the Chinese population to enable accurate prediction of DGF occurrence and duration, guide early clinical interventions, and improve outcomes for kidney transplant patients. In recent years, machine learning (ML) technology has demonstrated immense potential and applicability in the medical field, particularly in the development of predictive models [ 7 – 8 ]. Machine learning algorithms can efficiently process large-scale, multidimensional medical datasets and identify complex nonlinear relationships, giving them significant advantages in addressing clinical prediction challenges. Various machine learning algorithms, such as random forests, support vector machines, and neural networks, have shown exceptional performance in medical prediction models. These methods not only improve the accuracy of predictive models but also enhance their generalizability, enabling consistent performance across diverse populations. Particularly when utilizing multicenter data, machine learning models exhibit greater stability and applicability, offering new possibilities for the development of personalized medical decision support systems. However, in the field of kidney transplantation, especially in DGF prediction, the application of machine learning methods is still in its infancy, with existing studies limited by small sample sizes and single-center data. Given China's prominent status as a major contributor to global kidney transplantation, developing a machine learning-based DGF prediction model using multicenter data from China is both innovative and necessary. This study aims to fill existing research gaps and provide more accurate individualized prediction and treatment strategies for kidney transplant patients in China and beyond. We developed an innovative pre-transplant evaluation and matching system, integrating various advanced machine learning algorithms. Drawing on years of accumulated clinical experience, we identified nearly 100 relevant variables encompassing multiple aspects of donors and recipients to comprehensively assess and predict donor-recipient matching. To ensure the robustness and generalizability of the model, we not only utilized data from our center but also collaborated with three other large kidney transplant centers, conducting extensive data training and rigorous external validation. To the best of our knowledge, this is the first pre-transplant matching prediction model based on a large multicenter dataset in China. The development of this model represents a significant breakthrough in this field in China and provides a powerful decision-support tool to improve transplant success rates and patient outcomes. By integrating multicenter data with advanced machine learning techniques, our model is expected to enhance the accuracy of DGF prediction while offering a scientific basis for developing personalized kidney transplantation strategies, thereby potentially optimizing overall transplant outcomes. 2 Materials 2.1 Patients and Inclusion criteria This study employed a multicenter, large-sample research design, enrolling a total of 1,093 kidney transplant recipients. Data were collected from four major medical centers in China, including the First Affiliated Hospital of China Medical University (366 cases, January 2016 to October 2024), Jilin Hospital (209 cases), Dalian Hospital (211 cases), and the Northern Army General Hospital (307 cases). For the latter three hospitals, the data collection period spanned from March 2021 to October 2024. To ensure data quality, the following inclusion and exclusion criteria were applied: Donors: Only cases involving deceased donors who were Chinese citizens and met the Chinese national donor standards were included. Cases involving living donors were excluded. Recipients: Cases were excluded if the recipient was under 18 years old, undergoing a second or multiple transplants, involved in dual kidney transplants, or undergoing multi-organ transplants. Cases with key parameter data missing for more than 50% or with a recipient or graft survival time of fewer than 7 days were also excluded. The study adopted the internationally accepted definition of DGF, which is the requirement for dialysis within the first week post-transplantation. The study protocol was approved by the institutional review board (IRB) of the First Affiliated Hospital of China Medical University and the ethics committees of all participating institutions (IRB No.: AF-SOP-07-1.2-01). As a retrospective study, informed consent from patients was waived. All methods adhered to relevant guidelines and regulations. 2.2 Variables The primary outcome variable of this study was the occurrence of DGF following kidney transplantation (KTx). A comprehensive set of variables was collected, including recipient-, donor-, graft-, matching-, and transplant immunology-related factors. In addition to baseline characteristics and routine blood and urine laboratory tests, the following parameters were specifically included: • Donor-Related Variables: Donor Blood Type, Donor Hypertension (Doner Hp), Donor Infect, Kidney Donor Profile Index (KDPI), Kidney Donor Risk Index (KDRI). • Graft-Related Variables: Renal Artery Variation, Renal Vein Variation, Cold ischemia time (CIT). • Recipient-Related Variables: Recipient Blood Type, Recipient Nephropathy Etiology, Recipient Preoperative Dialysis Modality, Recipient Preoperative Systolic Blood Pressure, Recipient Preoperative Diastolic Blood Pressure, Recipient Preoperative Pulse Pressure, Recipient Preoperative mean arterial pressure, Recipient Preoperative Urine. • Immunology-Related Variables: Immunoinduction Protocols, HLA Mismatches (HLA MM), Panel reactive antibody levels (PRA). • Matching-Related Variables: ABO mismatch (ABO MM), Sex mismatch (Sex MM), Age mismatch (Age MM), BMI ratio (BMI Ratio), Deviation of BMI ratio from 1 (BMI Ratio to 1), Body surface area ratio (BSA Ratio), Deviation of BSA ratio from 1 (BSA Ratio to 1). 2.3 Feature Engineering In multicenter kidney transplantation studies, data missingness is a common challenge due to differences in management practices and organ sources from various locations. To address this issue, we carefully evaluated multiple approaches for handling missing data, including traditional methods such as mean or median imputation, as well as more advanced techniques like IterativeImputer and KNNImputer [ 9 – 10 ]. Considering that certain laboratory values might be entirely absent in specific centers, we opted for IterativeImputer and KNNImputer to ensure model accuracy while minimizing bias and the risk of overfitting. The IterativeImputer method is advantageous in its ability to capture complex relationships between features. Through an iterative process, the method refines imputation results step by step, making it particularly suitable for clinical datasets with intricate interdependencies. This approach simulates the clinical decision-making process where physicians consider multiple related factors, thereby preserving the internal logical structure of the data. On the other hand, the KNNImputer method leverages the K-nearest neighbors algorithm, estimating and imputing missing values based on the information from the K most similar neighbors. This method works particularly well for clinical data exhibiting local similarities, effectively capturing subtle yet significant variations within patient populations. 2.4 Data Preprocessing Given that the incidence of DGF across the participating centers was consistently around 20%, we first addressed the issue of class imbalance in the dataset. Using the RandomUnderSampler, we undersampled the majority class to balance the proportions of positive and negative classes [ 11 ]. This step was crucial for preventing the model from being biased toward the majority class, thereby improving its ability to predict the minority class.We also applied the StandardScaler to standardize the features, ensuring that all variables were on the same scale. Standardization is critical for enhancing the performance of many machine learning algorithms by preventing features with larger ranges from dominating those with smaller ranges. For model selection, we compared the performance of multiple machine learning algorithms, including: Logistic Regression (LR), Gaussian Naive Bayes (GNB), Decision Tree, Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Neural Network, XGBoost, and LightGBM. This multi-model comparison strategy allowed us to comprehensively evaluate the performance of different algorithms on this specific task [ 12 – 21 ]. For Logistic Regression and Random Forest, we further applied Bayesian optimization (via BayesSearchCV) to fine-tune hyperparameters. Compared to traditional grid search or random search methods, Bayesian optimization is more efficient and can identify optimal hyperparameter combinations in a shorter amount of time. This helped enhance the performance of the selected models while reducing computational overhead. 2.5 Model Evaluation To account for regional population differences, we divided the dataset from the First Affiliated Hospital of China Medical University, Dalian Hospital, and Jilin Hospital into an internal dataset for model training and testing, while using data from the remaining hospital as an external test set to validate the model.For the internal dataset, we initially employed a 70% training set and 30% test set split to evaluate model performance. The evaluation metrics included the area under the receiver operating characteristic curve (ROC-AUC) and accuracy. To further ensure a comprehensive and robust assessment of each model’s performance, we implemented a 10-fold cross-validation approach. This method not only maximizes the use of limited data but also effectively reduces the risk of overfitting, providing a more reliable estimate of model performance. Using the StratifiedKFold method, we randomly divided the dataset into 10 equal-sized subsets. For each model, we performed 10 training and testing cycles, where 9 subsets were used as the training set and the remaining 1 subset served as the validation set. For performance evaluation during cross-validation, we calculated multiple metrics [ 22 ], including Accuracy, Precision, Recall, F1 Score, Specificity, ROC-AUC, and Precision-Recall AUC (PR-AUC). After completing the cross-validation, we selected the model that demonstrated the best performance based on these metrics for external test set validation and final performance evaluation. 2.6 Cox Proportional Hazards Regression Analysis To assess the impact of various factors on the time to DGF recovery, we conducted a Cox proportional hazards regression analysis. First, we performed univariate Cox regression analysis on all variables to identify potential significant predictors. Using the survival package in R, we analyzed DGF recovery time as the outcome variable and the DGF status as the event indicator. For each variable, we calculated Hazard Ratio (HR), 95% Confidence Interval (CI) and p-value. Variables with a p-value < 0.1 in the univariate analysis were considered potential significant predictors and were included in the subsequent multivariate Cox regression analysis. The multivariate analysis allowed us to determine the independent effects of these significant predictors on DGF recovery time while adjusting for the influence of other variables. Finally, we visualized the results, providing insights into the key predictors and their relative contributions to DGF recovery. 2.7 Follow-up Analysis This study conducted long-term follow-up of patients from the transplantation center at the First Affiliated Hospital of China Medical University, concluding in October 2024, using a combination of outpatient follow-ups and telephone interviews, with patients considered lost to follow-up if they could not be contacted after three consecutive phone attempts or if they voluntarily withdrew from the study. Patients were categorized into four groups based on their DGF status and prognosis: DGF Death Group, DGF Survival Group, Non-DGF Death Group, and Non-DGF Survival Group. The primary endpoints included overall patient survival rate, graft survival rate, and mortality rate, while secondary endpoints focused on the longitudinal changes in renal function indicators, including serum creatinine, eGFR, blood urea, and cystatin C, as well as the incidence of clinical complications. Follow-up data were collected at postoperative intervals of 1, 3, 6, 12, 24, 36, and 48 months. Longitudinal data analysis was performed using the Generalized Estimating Equations (GEE) method with an exchangeable correlation structure, using the Non-DGF Survival Group as the control group, as this method accounts for the correlation between repeated measurements within individuals, is robust to missing data, and incorporates both time effects and group effects [ 23 ]. The impact of DGF status and prognosis on renal function trends was assessed using coefficient estimates, confidence intervals, and model fit indices such as the Quasi-likelihood under the Independence Model Criterion (QIC), enabling a robust evaluation of how DGF status and prognosis influence long-term renal function and complication occurrences. 2.8 Website Visualization To enhance the clinical applicability and accessibility of our predictive models, we developed an interactive web-based visualization tool. This tool, accessible at http://www.kidney-dgf-match.cn/ , allows healthcare professionals to input patient-specific data and obtain real-time predictions for DGF risk. The interface was designed to be user-friendly, featuring separate sections for donor characteristics, recipient characteristics, organ preservation details, and immunology-related factors. Users can input a wide range of variables, including demographic information, laboratory test results, and transplant-specific parameters. The tool utilizes best models to provide predictions, offering a comparative view of the results from these two high-performing algorithms. This web-based tool not only serves as a practical application of our research findings but also as a potential decision support system for clinicians involved in kidney transplantation. 2.9 Statistical analysis All our data can be classified as continuous and categorical variables. We use Shapiro–Wilk test to detect normal distribution in continuous variables, the two independent samples t-test for the normal distribution, and the Mann-Whitney U test for unnormal distribution. In categorical variables, Chi-Squared Test is used to calculate statistical significance. If P-value < 0.05, we consider it remains statistical significance. All model weights are visualized through the SHapley Additive exPlanations (SHAP) [ 24 ]. All methods are generated in Jupiter-notebook with kernel Python 3.9 with the scikit package [ 25 ]. 3 Result 3.1 Basic clinical characteristics comparation According to Table 1 , this study compared the baseline clinical characteristics of kidney transplant recipients with DGF (292 cases) and without DGF (801 cases). Regarding donor characteristics, there were no significant differences between the two groups in terms of age, sex, height, weight, BMI, and body surface area (BSA) (P > 0.05). However, significant differences were observed in donor blood type (P < 0.001), infection status (P < 0.001), and renal artery variation (P < 0.001). Specifically, the DGF group had a higher rate of donor infections (12.3% vs. 3.5%) but a lower incidence of renal artery variation (1.8% vs. 9.6%). For recipient characteristics, no significant differences were found between the two groups in blood type distribution, sex, height, weight, BMI, or urine output (P > 0.05). However, notable differences were identified in ABO matching (P < 0.001), age matching (P < 0.001), preoperative systolic blood pressure (P < 0.001), pulse pressure (P = 0.001), and mean arterial pressure (P = 0.001), with the DGF group showing lower preoperative systolic and mean arterial pressures. In terms of immunological features, the DGF group had a higher number of HLA mismatches (P < 0.001), a higher rate of PRA positivity (12.3% vs. 7.0%, P = 0.007), and was more likely to receive ATG as the induction therapy (77.2% vs. 83.1%, P < 0.001). Additionally, there were significant differences in the etiology of kidney disease (P < 0.001); glomerulonephritis was more prevalent in the DGF group (28.1% vs. 24.0%), while hypertensive nephropathy and diabetic nephropathy were more common in the non-DGF group. These findings suggest that certain donor and recipient characteristics, such as donor infection, HLA mismatches, and preoperative blood pressure, may be associated with the occurrence of DGF, providing important clues for further analysis. Table 1 Baseline characteristics of kidney transplant recipients with and without DGF. DGF(n = 292) Non-DGF(n = 801) P value Donor Donor Age, median ± SD 54.0 ± 10.2 53 ± 11.2 0.844 Donor Sex, n(%) 0.971 Male 248.0(85.1) 677(84.5) Female 44.0(14.9) 124(15.5) Donor Height_(cm), median ± SD 170.0 ± 5.4 170 ± 6.9 0.539 Donor Weight_(kg), median ± SD 70.0 ± 10.7 73 ± 15.3 0.237 DBMI, median ± SD 24.0 ± 3.3 24.4 ± 4.3 0.268 DBSA, median ± SD 1.8 ± 0.2 1.8 ± 0.2 0.314 Donor Blood Type, n(%) < 0.001 O 92.0(31.6) 268.0(33.5) A 74.0(25.4) 205.0(25.6) B 105.0(36.0) 261.0(32.6) AB 21.0(7.0) 67.0(8.3) Donor Hp, n(%) 0.235 N 131.0(44.7) 307.0(38.3) Y 161.0(55.3) 494.0(61.7) Donor infect, n(%) < 0.001 N 256.0(87.7) 773.0(96.5) Treponema 31.0(10.5) 13.0(1.6) HBV 0.0(0.0) 0.0(0.0) HCV 0.0(0.0) 0.0(0.0) Coinfection 5.0(1.8) 15.0(1.9) KDPI, median ± SD 66.0 ± 18.0 64.0 ± 20.3 0.530 KDRI, median ± SD 1.2 ± 0.3 1.2 ± 0.3 0.795 Renal Artery Variation, n(%) 0.000 N 287.0(98.2) 724.0(90.4) Y 5.0(1.8) 77.0(9.6) Renal Vein Variation, n(%) 0.147 N 292.0(100.0) 793.0(99.0) Y 0.0(0.0) 8.0(1.0) Recipient CIT, median ± SD 7.6 ± 3.3 7.0 ± 3.9 0.592 Blood Type, n(%) 0.469 O 84.0(28.9) 269.0(33.6) A 74.0(25.4) 200.0(24.9) B 105.0(36.0) 266.0(33.2) AB 29.0(9.7) 66.0(8.3) ABO MM, n(%) < 0.001 ABO-identical 277.0(94.7) 745.0(93.0) ABO-compatible 15.0(5.3) 56.0(7.0) Recipient Sex, n(%) 0.625 Male 205.0(70.2) 586.0(73.2) Female 87.0(29.8) 215.0(26.8) Sex MM, n(%) 0.646 Male to male 187.0(64.0) 512.0(63.9) Male to female 61.0(21.1) 163.0(20.4) Female to male 18.0(6.1) 75.0(9.3) Female to female 26.0(8.8) 51.0(6.4) Recipient Age, median ± SD 46.0 ± 11.1 46 ± 10.3 0.797 Age MM, n(%) 50 to >50 49.0(16.7) 259.0(32.3) >50 to <50 156.0(53.5) 225.0(28.1) 50 36.0(12.3) 79.0(9.9) <50 to <50 51.0(17.5) 238.0(29.7) Height_(cm), median ± SD 171.0 ± 8.0 172 ± 8.0 0.481 Weight_(kg), median ± SD 68.0 ± 15.1 69 ± 13.3 0.630 BMI, median ± SD 23.2 ± 4.02 23.6 ± 3.6 0.547 BMIr, median ± SD 0.9 ± 0.2 1.0 ± 0.2 0.171 BMIr to 1, n(%) 0.089 N 159.0(54.4) 360.0(45.0) Y 133.0(45.6) 441.0(55.0) RBSA, median ± SD 1.8 ± 0.2 1.8 ± 0.2 0.571 BSAr, median ± SD 1.0 ± 0.1 1.0 ± 0.2 0.181 BSAr to 1, n(%) 0.007 N 84.0(28.9) 405.0(50.5) Y 208.0(71.1) 396.0(49.5) RNE, n(%) < 0.001 Glomerulonephritis 82.0(28.1) 192.0(24.0) Hypertensive nephropathy 10.0(3.5) 54.0(6.7) Diabetic nephropathy 28.0(9.6) 54.0(6.7) Polycystic kidney 8.0(2.6) 30.0(3.8) Unknown 164.0(56.2) 471.0(58.8) Pdm, n(%) 0.135 N 3.0(0.9) 28.0(3.5) Hemodialysis 261.0(89.4) 658.0(82.1) Peritoneal dialysis 28.0(9.7) 115.0(14.4) PSBP, median ± SD 144.0 ± 21.4 152 ± 21.2 < 0.001 PDP, median ± SD 89 ± 13.0 90 ± 13.8 0.029 PSDP, median ± SD 54.0 ± 15.9 60 ± 15.2 0.001 PMAP, median ± SD 116 ± 16.2 120 ± 16.7 0.001 RPU, n(%) 0.312 N 230.0(78.9) 578.0(72.2) Y 62.0(21.1) 223.0(27.8) HLA MM, n(%) 4 218.0(74.5) 440.0(54.9) PRA, n(%) 0.007 Negative 256.0(87.7) 745.0(93.0) Positive 36.0(12.3) 56.0(7.0) Immunoinduction protocols, n(%) < 0.001 Basiliximab 67.0(22.8) 135.0(16.9) ATG 225.0(77.2) 666.0(83.1) 3.2 Visualization of Model Performance We divided the training dataset into a 7:3 ratio for model development, and as shown in Fig. 2 , the Random Forest and LightGBM models demonstrated the best performance with an AUC of 0.79, followed closely by the XGBoost model with an AUC of 0.76. However, in terms of accuracy, the LightGBM model outperformed all other models, achieving an accuracy of 80%, followed by XGBoost and Random Forest, both with an accuracy of 79%. Considering that a single train-test split may introduce randomness and that the Random Forest model is prone to overfitting, potentially limiting its generalization ability, we further employed a 10-fold cross-validation method to evaluate model performance. Figure 3 illustrates the results of the 10-fold cross-validation, revealing the consistently strong and stable performance of the Random Forest, XGBoost, and LightGBM models across multiple validation rounds. Table 2 provides detailed performance metrics for each model during 10-fold cross-validation. Notably, the Random Forest, LightGBM, and XGBoost models achieved significantly higher average accuracy and AUC compared to other models, showcasing their robust predictive power and stability. 3.3 External Validation Results After performing 10-fold cross-validation on the training set and selecting the top three models, we further validated the generalization ability of the Random Forest, XGBoost, and LightGBM models using an external dataset. As shown in the ROC curves and performance metrics (Fig. 4 & Table 3 ), the LightGBM model demonstrated the best performance during external validation. It achieved the highest scores across all evaluation metrics, with an ROC-AUC of 0.80, accuracy of 0.73, precision of 0.69, recall of 0.35, F1 score of 0.47, and specificity of 0.92. These results indicate that the LightGBM model performed exceptionally well in distinguishing between positive and negative samples, showcasing strong generalization ability. The XGBoost model followed closely behind, delivering solid performance with an ROC-AUC of 0.73 and accuracy of 0.71. While it outperformed the Random Forest model across most metrics, it fell slightly short of the LightGBM model. In contrast, the Random Forest model showed relatively weaker performance on the external dataset. Its ROC AUC was 0.63, and its accuracy was 0.67, with noticeably lower precision, recall, and F1 scores compared to the other two models. However, it is worth noting that the Random Forest model achieved a high specificity of 0.98, indicating strong performance in identifying negative samples. 3.4 Feature Importance Visualization To better understand the decision-making process of the models and enhance their interpretability, we performed SHAP (SHapley Additive exPlanations) analysis on the top-performing XGBoost and LightGBM models. SHAP values quantify the contribution of each feature to the model’s predictions, providing insights into the key factors driving DGF prediction. Figure 5 presents the SHAP summary plot for the LightGBM model, clearly highlighting the top five features influencing DGF prediction: pBNP, DeGFR, DHb, DuEC, and DuRBC. The analysis revealed that lower DeGFR values significantly increased the risk of DGF, which aligns with clinical understanding, as decreased glomerular filtration rate typically reflects poorer kidney function. Additionally, higher pBNP levels were associated with an increased risk of DGF, potentially reflecting the impact of underlying cardiovascular issues on transplant outcomes. Interestingly, higher DHb values and mild abnormalities in routine urine tests, such as elevated DuEC and DuRBC, also showed a trend of increased DGF risk. Figure 6 displays the SHAP summary plot for the XGBoost model, which further corroborates the importance of several key features. Similar to the LightGBM model, DeGFR, DHb, and pBNP were confirmed as critical predictors. Additionally, the XGBoost model identified DAPPT and pBUN as important features. Notably, lower DAPPT values were associated with an increased risk of DGF, offering a new perspective on the role of donor and recipient coagulation status in transplant outcomes. The consistency between the two models in identifying major predictors, particularly DeGFR, DHb, and pBNP, strengthens confidence in the significance of these features for DGF prediction. At the same time, the unique features identified by each model, such as DAPPT and pBUN, provide diverse perspectives for risk assessment, helping to capture the intricate network of factors influencing DGF. 3.5 DGF Recovery Time Univariate Cox regression analysis (Fig. 7 A) identified several factors associated with DGF recovery time (p < 0.1). Among these, donor type (HR = 1.66, 95% CI: 1.233–2.236, p = 0.0008), recipient serum creatinine level (HR = 1.002, 95% CI: 1.000–1.003, p = 0.0044), and donor serum calcium level (HR = 0.246, 95% CI: 0.088–0.685, p = 0.0073) had statistically significant impacts on recovery time. Other factors with significant associations included recipient weight, primary kidney disease type, recipient BMI, body surface area, dialysis duration, and donor serum magnesium levels. In multivariate Cox regression analysis (Fig. 7 B), two independent predictors were strongly associated with DGF recovery time after stepwise selection. First, dialysis duration (Pdt) showed statistical significance (HR = 1.006, 95% CI: 1.001–1.012, p = 0.021), indicating that each additional unit of dialysis time increased the risk of prolonged DGF recovery by 0.6%. Second, recipient preoperative systolic blood pressure (PsBp) exhibited marginal significance (HR = 0.094, 95% CI: 0.998–1.032, p = 0.094), suggesting that it may have a modest influence on DGF recovery time, though its p-value slightly exceeded the traditional threshold of 0.05. 3.6 Long-Term Prognosis of DGF Longitudinal analysis of renal function parameters over time (Fig. 8 ) revealed distinct trends among the four patient groups. The Non-DGF Survival Group showed the most stable renal function, with a steady increase in eGFR (from 60.90 to 74.89 mL/min/1.73m²) and consistently low serum creatinine levels (101.00–133.30 µmol/L). In contrast, the DGF Survival Group exhibited poorer early renal function but gradual improvement over time, with serum creatinine decreasing from 255.26 to 122.93 µmol/L, although a significant increase was observed at 36 months (289.25 µmol/L). The Death Groups (both DGF Death and Non-DGF Death) showed highly variable and generally worse trends, particularly in the DGF Death Group, where serum creatinine and urea levels were markedly elevated in the early stages. To quantify the differences among groups, a GEE model was applied using the Non-DGF Survival Group as the reference (Table 3 ). The analysis showed that the DGF Death Group exhibited the most severe renal dysfunction, with significantly higher serum creatinine (β = 200.57, 95% CI: 78.30–322.84, p = 0.001), significantly lower eGFR (β = -39.91, 95% CI: -47.61 to -32.21, p < 0.001), and significantly elevated levels of urea (β = 13.51, 95% CI: 7.40–19.61, p < 0.001) and cystatin C (β = 1.64, 95% CI: 0.92–2.36, p < 0.001). The Non-DGF Death Group showed significant elevation only in urea levels (β = 8.89, 95% CI: 2.83–14.94, p = 0.004). Meanwhile, the DGF Survival Group exhibited no statistically significant differences in most renal function parameters compared to the control group, except for a slight increase in cystatin C levels (β = 0.58, 95% CI: 0.09–1.06, p = 0.020). Time effect analysis indicated that urea (β = -0.11, p < 0.001) and cystatin C (β = -0.02, p = 0.020) levels decreased significantly over time, suggesting that renal function in surviving patients gradually improved. These findings suggest that while DGF patients experience slower early recovery, their long-term prognosis in terms of renal function is similar to that of Non-DGF patients, provided they survive the early stages. Table 4 Comparison of Dialysis-related Characteristics Between DGF Survival and Death Groups. Variable DGF Survival (n = 38) DGF Death (n = 9) P value DGF Duration (days) 16.66 ± 13.73 15.44 ± 14.62 0.735 Number of Dialysis 8.13 ± 7.39 7.78 ± 7.22 0.807 Mean Dialysis Interval (days) 1.89 ± 0.58 1.75 ± 0.70 0.649 3.7 Interactive Web-Based Tool To enhance the clinical utility and accessibility of our predictive model, we developed an interactive web-based visualization tool ( http://www.kidney-dgf-match.cn/ ). The website features an intuitive and user-friendly interface that organizes the predictive variables into multiple modules, including donor characteristics, recipient characteristics, organ preservation, and immunology (Fig. 9 ). Clinicians can input patient demographic information, laboratory test results, and transplant-related parameters, and the system will compute the risk probability of DGF in real-time. The tool incorporates the significant predictive variables identified in this study, with their respective weight coefficients integrated into the calculation model, providing transplant physicians with a convenient decision support tool. This online platform represents a practical application of our research findings and offers healthcare professionals in the field of kidney transplantation an objective risk assessment method. By enabling personalized perioperative management strategies, it aims to improve patient outcomes and optimize transplant success rates. 4 Discussion Among all the machine learning models evaluated, the ensemble learning models XGBoost and LightGBM demonstrated outstanding predictive performance on the external validation dataset, which can be attributed to their unique algorithmic features [ 26 ]. XGBoost (eXtreme Gradient Boosting) incorporates a second-order Taylor expansion to approximate the objective function and includes a regularization term to control model complexity, enhancing its robustness in handling nonlinear relationships. Furthermore, its innovative split-point finding algorithm efficiently processes sparse data, making it particularly advantageous in addressing the prevalent issue of missing values in medical datasets. In contrast, LightGBM's standout features include the "Gradient-based One-Side Sampling" (GOSS) and "Exclusive Feature Bundling" (EFB) strategies. GOSS retains samples with large gradients while randomly sampling those with small gradients, significantly reducing computational complexity while maintaining training accuracy. This was particularly effective in handling our imbalanced dataset, as DGF is a relatively rare clinical event. Additionally, LightGBM's leaf-wise growth strategy, which prioritizes optimal splits, allowed the model to better capture the complex nonlinear relationships in medical data, resulting in superior predictive performance on the external validation set (ROC-AUC = 0.80, accuracy = 0.73). In terms of data preprocessing, the IterativeImputer outperformed KNNImputer in both prediction accuracy and model robustness. This may be due to the inherent limitations of KNNImputer, which relies on finding the most similar neighbors. As data dimensions increase, the definition of similarity becomes less reliable, and the heterogeneity of populations in kidney transplantation amplifies the errors introduced by this method. Conversely, IterativeImputer focuses on the complex interrelationships between features, leveraging these connections to handle missing values more effectively, which resulted in better outcomes. Thus, for large-scale medical machine learning, IterativeImputer appears to be a more suitable choice. SHAP visualizations of the two models identified several groups of features with significant predictive value for DGF, spanning organ function, blood status, and immunological factors. Among donor-related factors, DeGFR was the most significant predictor, with lower donor GFR values substantially increasing the risk of DGF. This finding aligns with clinical experience, as poor baseline kidney function directly impacts early graft recovery [ 27 ]. Donor hemoglobin (DHb) levels showed a complex bidirectional influence: moderately elevated Hb levels may improve oxygen delivery to tissues, but excessively high Hb levels might indicate hemoconcentration, inadequate fluid maintenance, or poor organ perfusion, increasing the risk of perfusion-related injury. This highlights the importance of precise management of donor hemoglobin and fluid levels to optimize organ perfusion quality during the maintenance period. Additionally, recipient preoperative blood urea nitrogen (pBUN) levels showed significant predictive value, likely reflecting the impact of poor preoperative dialysis quality, metabolic imbalances, and nitrogen retention [ 28 ]. In terms of cardiovascular function, recipient preoperative plasma BNP (pBNP) levels emerged as a strong predictor. Elevated pBNP levels suggest impaired cardiac function, potentially reducing cardiac output and transplant kidney perfusion pressure, thereby increasing the risk of DGF [ 29 ]. This finding underscores the importance of comprehensive cardiovascular assessments for both donors and recipients during pre-transplant evaluations. Interestingly, the study also identified several unconventional but clinically significant predictors. Microscopic hematuria (DuEC, DuRBC), which is well-established as a marker for chronic kidney disease progression, has been relatively understudied in transplantation but was found to be a valuable predictor for DGF [ 30 ]. Additionally, decreased donor APTT levels, indicating hypercoagulability, may increase the risk of microthrombosis, potentially exacerbating ischemia-reperfusion injury and impairing graft microcirculation [ 31 ]. This finding highlights the potential interaction between coagulation and immune responses in the development of DGF, although further experimental evidence is needed to validate this hypothesis [ 32 ]. Potassium ion (K⁺) levels also demonstrated clinical relevance, as hypokalemia not only reflects simple electrolyte imbalance but may also indicate impaired tubular reabsorption and membrane instability, which could negatively impact early graft function recovery [ 33 ]. Further insights into the clinical timeline of DGF were provided by Cox regression and long-term prognosis analyses. Univariate Cox analysis revealed that donor type, recipient serum creatinine levels, and donor serum calcium levels significantly influenced DGF recovery time, highlighting the critical roles of organ quality and recipient baseline status in DGF duration [ 34 ]. Multivariate analysis identified preoperative dialysis duration as an independent predictor, revealing a strong association between prolonged dialysis burden and delayed DGF recovery [ 35 ]. This emphasizes the need for tailored management strategies for long-term dialysis patients in clinical practice. Long-term prognosis analysis demonstrated the complex impacts of DGF on transplant outcomes. By categorizing patients into DGF Death, DGF Survival, Non-DGF Death, and Non-DGF Survival groups, several key findings emerged. The DGF Death Group exhibited the most severe renal dysfunction, with significantly higher serum creatinine (β = 200.57) and lower eGFR (β = -39.91), reflecting the poor prognosis associated with severe DGF. However, an important observation was that the DGF Survival Group showed renal function comparable to the Non-DGF Survival Group in most metrics, with only slightly elevated cystatin C levels (β = 0.58). This underscores that the occurrence of DGF does not necessarily lead to poor long-term outcomes, provided that early identification and timely interventions are implemented. Notably, comparisons between the DGF Survival and Death groups revealed no significant differences in DGF duration (16.66 ± 13.73 vs. 15.44 ± 14.62 days, p = 0.735), number of dialysis sessions (8.13 ± 7.39 vs. 7.78 ± 7.22, p = 0.807), or mean dialysis interval (1.89 ± 0.58 vs. 1.75 ± 0.70 days, p = 0.649). This unexpected finding suggests that these characteristics may not be decisive factors in determining long-term prognosis, which is more likely influenced by overall patient condition and perioperative management quality. Longitudinal analysis showed significant declines in urea and cystatin C levels over time, supporting the notion that, even in the presence of DGF, most surviving patients can achieve gradual renal function improvement with appropriate management. These findings carry important clinical implications. Firstly, they emphasize the importance of early identification of high-risk patients. Secondly, they suggest that clinicians should not view DGF as a definitive indicator of poor prognosis but rather focus on early intervention and individualized treatment plans. Finally, these results provide transplant physicians with more detailed evidence for communicating long-term outcomes to patients [ 36 ]. Although this study has made significant progress in developing the first DGF prediction model based on multicenter data from China, several limitations require further exploration and refinement in future research. First, given the vast geographic diversity of China, regional differences in population characteristics (e.g., living environment, dietary habits) and medical practices may introduce heterogeneity, impacting the model's applicability. While this study used multicenter data, it primarily involved institutions from northern China. Expanding data collection to include more diverse regions and identifying population-specific factors will improve the model's generalizability. Second, the study spans a long data collection period, during which kidney transplantation techniques and perioperative management protocols may have evolved. Temporal effects may influence model accuracy and require additional attention in future research. Furthermore, ethical considerations in artificial intelligence, particularly ensuring unbiased model application in clinical practice and enhancing interpretability, represent critical challenges. Incorporating novel biomarkers, such as genomic and transcriptomic data, into feature selection could further improve predictive power. Finally, prospective validation studies are essential for evaluating the model's effectiveness in real-world clinical applications, which represents a key direction for future research. 5 Conclusion This study developed the first DGF prediction model based on multicenter data from China, integrating machine learning techniques with clinical data to achieve accurate post-transplant DGF risk prediction. In external validation, the LightGBM model demonstrated the best predictive performance (AUC = 0.80), providing a reliable decision-support tool for clinicians. SHAP analysis confirmed the importance of traditional risk factors such as donor eGFR and recipient preoperative plasma BNP levels while highlighting the potential value of novel predictors like microscopic hematuria and donor APTT. Cox regression analysis identified dialysis duration as a key factor influencing DGF recovery, while long-term follow-up results indicated that, although DGF increases early mortality risk, it does not significantly affect long-term outcomes in surviving patients. Importantly, DGF duration, number of dialysis sessions, and mean dialysis interval were not decisive factors for long-term prognosis. Based on these findings, the web-based prediction platform ( http://www.kidney-dgf-match.cn/ ) was developed to assist clinicians in risk assessment and personalized treatment planning. This predictive model not only fills a research gap in the field of kidney transplantation in China but also provides new tools and insights for improving transplant outcomes. Declarations Acknowledge We thank everyone in the department for their excellent work and selfless help. Author contributions PS and ZD conceptualized and designed the study. PS, XW, JL, KW, ZY, and WZ were responsible for data collection and curation. ZD performed the computational analysis, developed the machine learning models, and generated the figures. PS and ZD conducted the data analysis and interpretation. PS, ZD, and XW drafted the initial manuscript. HS, YC, LH, KM, and GW provided critical revision of the manuscript for important intellectual content and contributed to data collection from their respective centers. TG contributed to the study design. All authors reviewed and approved the final version of the manuscript. HS and YC supervised the project and provided administrative support. Funding information Central Government Support for Local Development 2023: Special Project for Construction of Scarce Disciplines in Strategic Emerging Industries - Experimental Surgery (Project No. 3110230005) Data availability To protect patient privacy and confidentiality, the raw data containing sensitive patient information used in this study are not publicly available. However, these data can be made available upon reasonable request to the first author or corresponding author, subject to appropriate data sharing agreements and ethical approvals. The code used for data analysis and model development in this study is publicly accessible. All relevant scripts and code files can be found in our GitHub repository at https://github.com/Mr-SiSi/DGF-prediction-model-in-kidney-transplantation. Ethics Statement The study protocol was reviewed and approved by the Ethics Committee of the First Affiliated Hospital of China Medical University (approval number: AF-SOP-07-1.2-01). Due to the retrospective nature of this study, which involved the analysis of existing clinical data, the requirement for individual patient consent was waived by the ethics committee. Conflict of Interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. References Halloran PF, Hunsicker LG (2000) Delayed graft function: state of the art, November 10–11, Summit meeting, Scottsdale, Arizona, USA. Am J Transplant . 2001;1(2):115–120 Lentine KL, Smith JM, Hart A et al (2022) OPTN/SRTR 2020 Annual Data Report: Kidney. Am J Transpl 22(Suppl 2):21–136 Perico N, Cattaneo D, Sayegh MH, Remuzzi G Delayed graft function in kidney transplantation. Lancet 2004 Nov 13–19;364(9447):1814–1827 Siedlecki A, Irish W, Brennan DC (2011) Delayed graft function in the kidney transplant. Am J Transpl 11(11):2279–2296 Irish WD, Ilsley JN, Schnitzler MA, Feng S, Brennan DC (2010) A risk prediction model for delayed graft function in the current era of deceased donor renal transplantation. Am J Transpl 10(10):2279–2286 Chapal M, Le Borgne F, Legendre C et al (2014) A useful scoring system for the prediction and management of delayed graft function following kidney transplantation from cadaveric donors. Kidney Int 86(6):1130–1139 Yoo D, Divard G, Raynaud M et al (2024) A Machine Learning-Driven Virtual Biopsy System For Kidney Transplant Patients. Nat Commun 15(1):554 Published 2024 Jan 16 Lee KH, Choi GH, Yun J et al (2024) Machine learning-based clinical decision support system for treatment recommendation and overall survival prediction of hepatocellular carcinoma: a multi-center study. NPJ Digit Med. ;7(1):2. Published 2024 Jan 5 Platias C, Petasis G (2020) A comparison of machine learning methods for data imputation[C]//11th Hellenic Conference on Artificial Intelligence. : 150–159 Karamti H, Alharthi R, Anizi AA et al (2023) Improving Prediction of Cervical Cancer Using KNN Imputed SMOTE Features and Multi-Model Ensemble Learning Approach. Cancers (Basel) 15(17):4412 Published 2023 Sep 4 Hancock IIIJT, Khoshgoftaar TM (2023) Exploring maximum tree depth and random undersampling in ensemble trees to optimize the classification of imbalanced big data[J]. SN Comput Sci 4(5):462 Meurer WJ, Tolles J (2017) Logistic Regression Diagnostics: Understanding How Well a Model Predicts Outcomes. JAMA 317(10):1068–1069 Ontivero-Ortega M, Lage-Castellanos A, Valente G, Goebel R, Valdes-Sosa M (2017) Fast Gaussian Naïve Bayes for searchlight classification analysis. NeuroImage 163:471–479 Flayer CH, Perner C, Sokol CL (2021) A decision tree model for neuroimmune guidance of allergic immunity. Immunol Cell Biol 99(9):936–948 Yang L, Wu H, Jin X, Zheng P, Hu S, Xu X, Yu W, Yan J (2020) Study of cardiovascular disease prediction model based on random forest in eastern China. Sci Rep 10(1):5245 Huang S, Cai N, Pacheco PP, Narrandes S, Wang Y, Xu W (2018) Applications of Support Vector Machine (SVM) Learning in Cancer Genomics. Cancer Genomics Proteomics 15(1):41–51 Handelman GS, Kok HK, Chandra RV, Razavi AH, Lee MJ, Asadi H (2018) eDoctor: machine learning and the future of medicine. J Intern Med 284(6):603–619 Lee JY, Styczynski MP (2018) NS-kNN: a modified k-nearest neighbors approach for imputing metabolomics data. Metabolomics: Official J Metabolomic Soc 14(12):153 Kriegeskorte N, Golan T (2019) Neural network models and deep learning. Curr biology: CB 29(7):R231–R236 Ogunleye A, Wang QG (2020) XGBoost Model for Chronic Kidney Disease Diagnosis. IEEE/ACM Trans Comput Biol Bioinf 17(6):2131–2140 Chen T, Guestrin C (2016), August Xgboost: A scalable tree boosting system. In Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining (pp. 785–794) Herron P (2004) Machine learning for medical decision support: evaluating diagnostic performance of machine learning classification algorithms. INLS 110:1–15 Wang M (2014) Generalized estimating equations in longitudinal data analysis: a review and recent developments. Adv Stat 2014(1):303728 Rao S, Mehta S, Kulkarni S, Dalvi H, Katre N, Narvekar M (2022), December A study of LIME and SHAP model explainers for autonomous disease predictions. In 2022 IEEE Bombay Section Signature Conference (IBSSC) (pp. 1–6). IEEE Hao J, Ho TK (2019) Machine learning made easy: a review of scikit-learn package in python programming language. J Educational Behav Stat 44(3):348–361 Zhang D, Gong Y (2020) The comparison of LightGBM and XGBoost coupling factor analysis and prediagnosis of acute liver failure. Ieee Access 8:220990–221003 Mogulla MR, Bhattacharjya S, Clayton PA (2019) Risk factors for and outcomes of delayed graft function in live donor kidney transplantation–a retrospective study. Transpl Int 32(11):1151–1160 Lee SW, Yang YM, Kim HY, Cho H, Nam SW, Kim SM, Kwon SK (2022) Predialysis Urea Nitrogen Is a Nutritional Marker of Hemodialysis Patients. Chonnam Med J 58(2):69–74 Jarolim P, Claggett BL, Conrad MJ, Carpenter MA, Ivanova A, Bostom AG, Kusek JW, Hunsicker LG, Jacques PF, Gravens-Mueller L, Finn P, Solomon SD, Weiner DE, Levey AS, Pfeffer MA (2017) B-Type Natriuretic Peptide and Cardiac Troponin I Are Associated With Adverse Outcomes in Stable Kidney Transplant Recipients. Transplantation 101(1):182–190 Okada S, Samejima KI, Matsui M et al (2020) Microscopic hematuria is a risk factor for end-stage kidney disease in patients with biopsy-proven diabetic nephropathy. BMJ Open Diabetes Res Care 8(2):e001863 Sood P, Randhawa PS, Mehta R, Hariharan S, Tevar AD (2015) Donor kidney microthrombi and outcomes of kidney transplant: a single-center experience. Clin Transplant 29(5):434–438 Wilhelm G, Mertowska P, Mertowski S, Przysucha A, Strużyna J, Grywalska E, Torres K (2023) The Crossroads of the Coagulation System and the Immune System: Interactions and Connections. Int J Mol Sci 24(16):12563 Kardalas E, Paschou SA, Anagnostis P, Muscogiuri G, Siasos G, Vryonidou A (2018) Hypokalemia: a clinical update. Endocr connections 7(4):R135–R146 Siedlecki A, Irish W, Brennan DC (2011) Delayed graft function in the kidney transplant. Am J transplantation: official J Am Soc Transplantation Am Soc Transpl Surg 11(11):2279–2296 Freitas MHB, Lima LC, Couceiro TCM, Silva WBD, Andrade JM, Freitas MHB (2018) Perioperative factors associated with delayed graft function in renal transplant patients. Jornal brasileiro de nefrologia 40(4):360–365 Neuberger JM, Bechstein WO, Kuypers DR, Burra P, Citterio F, De Geest S, Duvoux C, Jardine AG, Kamar N, Krämer BK, Metselaar HJ, Nevens F, Pirenne J, Rodríguez-Perálvarez ML, Samuel D, Schneeberger S, Serón D, Trunečka P, Tisone G, van Gelder T (2017) Practical Recommendations for Long-term Management of Modifiable Risks in Kidney and Liver Transplant Recipients: A Guidance Report and Clinical Checklist by the Consensus on Managing Modifiable Risk in Transplantation (COMMIT) Group. Transplantation 101(4):S1–S56 Tables Tables 2 and 3 are available in the Supplementary Files section. Additional Declarations There is NO Competing Interest. Supplementary Files Tables.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5617823","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":397267300,"identity":"b5e1f635-a8db-4be4-8a7f-f167162a118d","order_by":0,"name":"Ying Cheng","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA5UlEQVRIiWNgGAWjYDACCRgp//7hAwYDkrQw5DAbkKIFBHLYJPCoQwD+2c3HHn5ts8iTdzh7rPJHwR15BvbDRzfgteTOsXRj2TaJYsODfWm3eQyeGTbwpKXdwKfFQCLHTFqyTSJxYzOD2W0Gg8OMDRI8ZgS05H+DaGljMCv8YXDYnggtOWySH4Fa5vPwmDHwGBxOJKhF4kaamTTDOYnEDRJsydJALclthPzCPyP5meSPsrrE+TOYD3788eewbT/74WN4tYAAMy8b0IUHoDw2QspBgPHHH2B6aSBG6SgYBaNgFIxIAADsR0mc5YW/IwAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0002-0657-7551","institution":"Department of Organ Transplantation and Hepatobiliary, The First Hospital of China Medical University","correspondingAuthor":true,"prefix":"","firstName":"Ying","middleName":"","lastName":"Cheng","suffix":""},{"id":397267301,"identity":"8154adc5-9c12-4c04-b052-3aec58f32f63","order_by":1,"name":"he sun","email":"","orcid":"","institution":"Department of Organ transplantation \u0026 Hepatobiliary, The First Affiliated Hospital of China Medical University, Liaoning, Shenyang, China","correspondingAuthor":false,"prefix":"","firstName":"he","middleName":"","lastName":"sun","suffix":""},{"id":397267302,"identity":"bf4ecf75-952e-4f26-aeeb-af7b62f13ad7","order_by":2,"name":"ping sun","email":"","orcid":"","institution":"The First Affiliated Hospital of China Medical University","correspondingAuthor":false,"prefix":"","firstName":"ping","middleName":"","lastName":"sun","suffix":""},{"id":397267303,"identity":"c78129fd-787b-4418-8319-f36e36dcf727","order_by":3,"name":"zheng ding","email":"","orcid":"","institution":"Department of Cardiac Surgery, The First Affiliated Hospital of China Medical University, Liaoning, Shenyang, China","correspondingAuthor":false,"prefix":"","firstName":"zheng","middleName":"","lastName":"ding","suffix":""},{"id":397267304,"identity":"7f45a276-cf9b-4b5a-911a-07e418c87e80","order_by":4,"name":"xi wang","email":"","orcid":"","institution":"Department of Organ transplantation \u0026 Hepatobiliary, The First Affiliated Hospital of China Medical University, Liaoning, Shenyang, China","correspondingAuthor":false,"prefix":"","firstName":"xi","middleName":"","lastName":"wang","suffix":""},{"id":397267305,"identity":"18036b31-604d-48fd-9863-fc00165a7880","order_by":5,"name":"long he","email":"","orcid":"","institution":"Organ transplantation center, General Hospital of Northern Theater Command, Shenyang, China","correspondingAuthor":false,"prefix":"","firstName":"long","middleName":"","lastName":"he","suffix":""},{"id":397267306,"identity":"2343c30b-b068-4df0-b0d8-709a8444e14a","order_by":6,"name":"ke xin ma","email":"","orcid":"","institution":"Department of General Surgery, Division of Hepatobiliary and Pancreatic Surgery, The Second Affiliated Hospital of Dalian Medical University, Dalian, Liaoning, China","correspondingAuthor":false,"prefix":"","firstName":"ke","middleName":"xin","lastName":"ma","suffix":""},{"id":397267307,"identity":"c02ef25f-2641-4a4b-8cc0-f655e4bf51b2","order_by":7,"name":"gang wang","email":"","orcid":"","institution":"Department of Organ transplantation, The First Hospital of Jilin University, Jilin, Changchun, China","correspondingAuthor":false,"prefix":"","firstName":"gang","middleName":"","lastName":"wang","suffix":""},{"id":397267308,"identity":"e30c9adb-a21f-434c-9024-f222ced2d5fd","order_by":8,"name":"jing yun li","email":"","orcid":"","institution":"Department of Organ transplantation \u0026 Hepatobiliary, The First Affiliated Hospital of China Medical University, Liaoning, Shenyang, China","correspondingAuthor":false,"prefix":"","firstName":"jing","middleName":"yun","lastName":"li","suffix":""},{"id":397267309,"identity":"41afa0e2-f420-4841-a8cb-2ad8bed6c4c5","order_by":9,"name":"Kangchun Wang","email":"","orcid":"","institution":"Nanjing University","correspondingAuthor":false,"prefix":"","firstName":"Kangchun","middleName":"","lastName":"Wang","suffix":""},{"id":397267310,"identity":"8be614cd-ac41-4d46-bd31-0407671dde1f","order_by":10,"name":"zitong yu","email":"","orcid":"","institution":"Department of Organ transplantation \u0026 Hepatobiliary, The First Affiliated Hospital of China Medical University, Liaoning, Shenyang, China","correspondingAuthor":false,"prefix":"","firstName":"zitong","middleName":"","lastName":"yu","suffix":""},{"id":397267311,"identity":"3ed89d4f-e599-4e5a-b244-ae682d17ec25","order_by":11,"name":"weichen zhang","email":"","orcid":"","institution":"Department of Organ transplantation \u0026 Hepatobiliary, The First Affiliated Hospital of China Medical University, Liaoning, Shenyang, China","correspondingAuthor":false,"prefix":"","firstName":"weichen","middleName":"","lastName":"zhang","suffix":""},{"id":397267312,"identity":"35d8ce56-ee67-4357-a284-b9f577533634","order_by":12,"name":"tianxiang gu","email":"","orcid":"","institution":"Department of Cardiac Surgery, The First Affiliated Hospital of China Medical University, Liaoning, Shenyang, China","correspondingAuthor":false,"prefix":"","firstName":"tianxiang","middleName":"","lastName":"gu","suffix":""}],"badges":[],"createdAt":"2024-12-10 15:01:12","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5617823/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5617823/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":73083277,"identity":"47ea1bbe-2c49-4dce-8d60-f95881454f49","added_by":"auto","created_at":"2025-01-06 14:35:52","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":594150,"visible":true,"origin":"","legend":"\u003cp\u003eStudy flowchart.\u003c/p\u003e","description":"","filename":"Figure1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5617823/v1/017e88c960999c522e7489fc.jpg"},{"id":73084780,"identity":"77361e9f-8f35-4ea2-b6a8-fda9456ca2e1","added_by":"auto","created_at":"2025-01-06 14:43:53","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":593469,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of Receiver Operating Characteristic (ROC) curves for nine machine learning models\u003c/p\u003e","description":"","filename":"Figure2AUC.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5617823/v1/7650914b5493b50d80dbe2db.jpg"},{"id":73083303,"identity":"dc8691b0-2bfb-435c-a241-60b26336d9e7","added_by":"auto","created_at":"2025-01-06 14:35:53","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":440752,"visible":true,"origin":"","legend":"\u003cp\u003eLine graph comparing the average accuracy and ROC-AUC values of nine machine learning models during 10-fold cross-validation\u003c/p\u003e","description":"","filename":"Figure310foldcrossvalidation.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5617823/v1/d3aaab584388b1dd039fff75.jpg"},{"id":73083295,"identity":"0f5bc026-fd58-44b6-ad3a-869b5c469197","added_by":"auto","created_at":"2025-01-06 14:35:53","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":295378,"visible":true,"origin":"","legend":"\u003cp\u003eROC Curves Comparison of Machine Learning Models on External Validation Dataset.\u003c/p\u003e","description":"","filename":"Figure4XXXXXXXXXXXX.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5617823/v1/ac2b7bce70b06668807a8c09.jpg"},{"id":73083301,"identity":"ca0f98b1-5bef-44ae-bdca-6bf7f2fcfb20","added_by":"auto","created_at":"2025-01-06 14:35:53","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":427140,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP summary plot for the LightGBM model\u003c/p\u003e","description":"","filename":"Figure5LightGBM.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5617823/v1/3d6cf08ac0f763b6bb8cb041.jpg"},{"id":73083297,"identity":"1d26da64-474d-4570-bd44-a7ab0d973a64","added_by":"auto","created_at":"2025-01-06 14:35:53","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":490611,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP summary plot for the XGBoost model\u003c/p\u003e","description":"","filename":"Figure6XGboost.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5617823/v1/33b22c803a008ac71292a0f5.jpg"},{"id":73083326,"identity":"8f7beb2e-5378-4c8d-8f33-556a0826748e","added_by":"auto","created_at":"2025-01-06 14:35:54","extension":"jpg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":166677,"visible":true,"origin":"","legend":"\u003cp\u003eUnivariate (A) and Multivariate (B) Cox regression analysis of factors associated with DGF recovery time.\u003c/p\u003e","description":"","filename":"Figure7.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5617823/v1/b9c009914e7e7d57c9acb8a6.jpg"},{"id":73083323,"identity":"928fc78f-b908-4d9c-95ae-a36f0bb40664","added_by":"auto","created_at":"2025-01-06 14:35:54","extension":"jpg","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":332380,"visible":true,"origin":"","legend":"\u003cp\u003eLongitudinal Changes in Renal Function Parameters Among Different Groups. A. Urea (mmol/L). B. Serum Creatinine (μmol/L). C. Cystatin C (mg/L). D. eGFR (mL/min/1.73m²).\u003c/p\u003e","description":"","filename":"Figure8.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5617823/v1/363fd8f71359b3851cee4b17.jpg"},{"id":73084778,"identity":"09cae858-f058-46f0-a175-a9255d3e0c23","added_by":"auto","created_at":"2025-01-06 14:43:52","extension":"jpg","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":584834,"visible":true,"origin":"","legend":"\u003cp\u003eWeb Interface of the Online Risk Assessment Tool for DGF Prediction in Kidney Transplantation\u003c/p\u003e","description":"","filename":"Figure9.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5617823/v1/27989ba0cf336f1c6556d8cd.jpg"},{"id":80409384,"identity":"61a15255-d442-491c-ac52-29a807896500","added_by":"auto","created_at":"2025-04-11 15:28:57","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":4961719,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5617823/v1/245ac934-b910-4ad3-9722-4d53799bf85f.pdf"},{"id":73083278,"identity":"f0525c86-ec24-458e-bc7f-b9ac1882b6cc","added_by":"auto","created_at":"2025-01-06 14:35:52","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":619413,"visible":true,"origin":"","legend":"","description":"","filename":"Tables.docx","url":"https://assets-eu.researchsquare.com/files/rs-5617823/v1/11bf68075e0baf3caf72c23a.docx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"A High-Precision Machine Learning-Based Prediction Model for Delayed Graft functon(DGF) in Chinese Kidney Transplant Patients: A Multicenter Study","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eDelayed Graft Function (DGF) is a common and severe complication following kidney transplantation, typically defined as the need for dialysis within the first week post-transplantation [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Over the past decade, the incidence of DGF has shown a significant upward trend, reaching 26.3% among adult kidney transplant recipients in 2022 [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. The occurrence of DGF is influenced by a variety of complex factors, including donor- and recipient-related variables, with its molecular mechanisms primarily involving pathological processes such as ischemia-reperfusion injury and immune responses. Clinically, the most direct and typical symptom of DGF is anuria or oliguria [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. For patients, DGF not only significantly increases hospitalization costs but also severely undermines their hope of escaping dialysis, heightening anxiety about the success of the transplant. For clinicians, DGF complicates perioperative management, making it more challenging and increasing the risk of graft loss and patient mortality. Even more concerning is the poor prognosis associated with DGF. Numerous studies have shown that DGF significantly reduces long-term graft survival rates and patient survival rates. Moreover, DGF increases the risk of complications such as rejection and infection, subjecting patients to prolonged physical and psychological distress, which can even be life-threatening [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eGiven the significance of DGF, researchers in recent years have proposed several scoring systems to predict and reduce its incidence. For instance, Irish et al. developed a prediction model based on five risk factors for the North American population [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e], while Marion Chapal et al. proposed a similar model for the French population [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. However, these existing models have limitations, including relatively low accuracy and reliance on single-source data, preventing them from gaining widespread consensus. Notably, as the country with the second-highest number of kidney transplants globally, China still lacks a localized DGF prediction model. This highlights the urgent need to develop a DGF prediction model tailored to the Chinese population to enable accurate prediction of DGF occurrence and duration, guide early clinical interventions, and improve outcomes for kidney transplant patients.\u003c/p\u003e \u003cp\u003eIn recent years, machine learning (ML) technology has demonstrated immense potential and applicability in the medical field, particularly in the development of predictive models [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Machine learning algorithms can efficiently process large-scale, multidimensional medical datasets and identify complex nonlinear relationships, giving them significant advantages in addressing clinical prediction challenges. Various machine learning algorithms, such as random forests, support vector machines, and neural networks, have shown exceptional performance in medical prediction models. These methods not only improve the accuracy of predictive models but also enhance their generalizability, enabling consistent performance across diverse populations. Particularly when utilizing multicenter data, machine learning models exhibit greater stability and applicability, offering new possibilities for the development of personalized medical decision support systems. However, in the field of kidney transplantation, especially in DGF prediction, the application of machine learning methods is still in its infancy, with existing studies limited by small sample sizes and single-center data. Given China's prominent status as a major contributor to global kidney transplantation, developing a machine learning-based DGF prediction model using multicenter data from China is both innovative and necessary.\u003c/p\u003e \u003cp\u003eThis study aims to fill existing research gaps and provide more accurate individualized prediction and treatment strategies for kidney transplant patients in China and beyond. We developed an innovative pre-transplant evaluation and matching system, integrating various advanced machine learning algorithms. Drawing on years of accumulated clinical experience, we identified nearly 100 relevant variables encompassing multiple aspects of donors and recipients to comprehensively assess and predict donor-recipient matching. To ensure the robustness and generalizability of the model, we not only utilized data from our center but also collaborated with three other large kidney transplant centers, conducting extensive data training and rigorous external validation. To the best of our knowledge, this is the first pre-transplant matching prediction model based on a large multicenter dataset in China. The development of this model represents a significant breakthrough in this field in China and provides a powerful decision-support tool to improve transplant success rates and patient outcomes. By integrating multicenter data with advanced machine learning techniques, our model is expected to enhance the accuracy of DGF prediction while offering a scientific basis for developing personalized kidney transplantation strategies, thereby potentially optimizing overall transplant outcomes.\u003c/p\u003e"},{"header":"2 Materials","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Patients and Inclusion criteria\u003c/h2\u003e \u003cp\u003eThis study employed a multicenter, large-sample research design, enrolling a total of 1,093 kidney transplant recipients. Data were collected from four major medical centers in China, including the First Affiliated Hospital of China Medical University (366 cases, January 2016 to October 2024), Jilin Hospital (209 cases), Dalian Hospital (211 cases), and the Northern Army General Hospital (307 cases). For the latter three hospitals, the data collection period spanned from March 2021 to October 2024.\u003c/p\u003e \u003cp\u003eTo ensure data quality, the following inclusion and exclusion criteria were applied:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eDonors: Only cases involving deceased donors who were Chinese citizens and met the Chinese national donor standards were included. Cases involving living donors were excluded.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eRecipients: Cases were excluded if the recipient was under 18 years old, undergoing a second or multiple transplants, involved in dual kidney transplants, or undergoing multi-organ transplants. Cases with key parameter data missing for more than 50% or with a recipient or graft survival time of fewer than 7 days were also excluded.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cp\u003eThe study adopted the internationally accepted definition of DGF, which is the requirement for dialysis within the first week post-transplantation. The study protocol was approved by the institutional review board (IRB) of the First Affiliated Hospital of China Medical University and the ethics committees of all participating institutions (IRB No.: AF-SOP-07-1.2-01). As a retrospective study, informed consent from patients was waived. All methods adhered to relevant guidelines and regulations.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Variables\u003c/h2\u003e \u003cp\u003eThe primary outcome variable of this study was the occurrence of DGF following kidney transplantation (KTx). A comprehensive set of variables was collected, including recipient-, donor-, graft-, matching-, and transplant immunology-related factors. In addition to baseline characteristics and routine blood and urine laboratory tests, the following parameters were specifically included:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003e\u0026bull; Donor-Related Variables: Donor Blood Type, Donor Hypertension (Doner Hp), Donor Infect, Kidney Donor Profile Index (KDPI), Kidney Donor Risk Index (KDRI).\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e\u0026bull; Graft-Related Variables: Renal Artery Variation, Renal Vein Variation, Cold ischemia time (CIT).\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e\u0026bull; Recipient-Related Variables: Recipient Blood Type, Recipient Nephropathy Etiology, Recipient Preoperative Dialysis Modality, Recipient Preoperative Systolic Blood Pressure, Recipient Preoperative Diastolic Blood Pressure, Recipient Preoperative Pulse Pressure, Recipient Preoperative mean arterial pressure, Recipient Preoperative Urine.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e\u0026bull; Immunology-Related Variables: Immunoinduction Protocols, HLA Mismatches (HLA MM), Panel reactive antibody levels (PRA).\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e\u0026bull; Matching-Related Variables: ABO mismatch (ABO MM), Sex mismatch (Sex MM), Age mismatch (Age MM), BMI ratio (BMI Ratio), Deviation of BMI ratio from 1 (BMI Ratio to 1), Body surface area ratio (BSA Ratio), Deviation of BSA ratio from 1 (BSA Ratio to 1).\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Feature Engineering\u003c/h2\u003e \u003cp\u003eIn multicenter kidney transplantation studies, data missingness is a common challenge due to differences in management practices and organ sources from various locations. To address this issue, we carefully evaluated multiple approaches for handling missing data, including traditional methods such as mean or median imputation, as well as more advanced techniques like IterativeImputer and KNNImputer [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Considering that certain laboratory values might be entirely absent in specific centers, we opted for IterativeImputer and KNNImputer to ensure model accuracy while minimizing bias and the risk of overfitting.\u003c/p\u003e \u003cp\u003eThe IterativeImputer method is advantageous in its ability to capture complex relationships between features. Through an iterative process, the method refines imputation results step by step, making it particularly suitable for clinical datasets with intricate interdependencies. This approach simulates the clinical decision-making process where physicians consider multiple related factors, thereby preserving the internal logical structure of the data.\u003c/p\u003e \u003cp\u003eOn the other hand, the KNNImputer method leverages the K-nearest neighbors algorithm, estimating and imputing missing values based on the information from the K most similar neighbors. This method works particularly well for clinical data exhibiting local similarities, effectively capturing subtle yet significant variations within patient populations.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Data Preprocessing\u003c/h2\u003e \u003cp\u003e Given that the incidence of DGF across the participating centers was consistently around 20%, we first addressed the issue of class imbalance in the dataset. Using the RandomUnderSampler, we undersampled the majority class to balance the proportions of positive and negative classes [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. This step was crucial for preventing the model from being biased toward the majority class, thereby improving its ability to predict the minority class.We also applied the StandardScaler to standardize the features, ensuring that all variables were on the same scale. Standardization is critical for enhancing the performance of many machine learning algorithms by preventing features with larger ranges from dominating those with smaller ranges.\u003c/p\u003e \u003cp\u003eFor model selection, we compared the performance of multiple machine learning algorithms, including: Logistic Regression (LR), Gaussian Naive Bayes (GNB), Decision Tree, Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Neural Network, XGBoost, and LightGBM. This multi-model comparison strategy allowed us to comprehensively evaluate the performance of different algorithms on this specific task [\u003cspan additionalcitationids=\"CR13 CR14 CR15 CR16 CR17 CR18 CR19 CR20\" citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eFor Logistic Regression and Random Forest, we further applied Bayesian optimization (via BayesSearchCV) to fine-tune hyperparameters. Compared to traditional grid search or random search methods, Bayesian optimization is more efficient and can identify optimal hyperparameter combinations in a shorter amount of time. This helped enhance the performance of the selected models while reducing computational overhead.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5 Model Evaluation\u003c/h2\u003e \u003cp\u003eTo account for regional population differences, we divided the dataset from the First Affiliated Hospital of China Medical University, Dalian Hospital, and Jilin Hospital into an internal dataset for model training and testing, while using data from the remaining hospital as an external test set to validate the model.For the internal dataset, we initially employed a 70% training set and 30% test set split to evaluate model performance. The evaluation metrics included the area under the receiver operating characteristic curve (ROC-AUC) and accuracy.\u003c/p\u003e \u003cp\u003eTo further ensure a comprehensive and robust assessment of each model\u0026rsquo;s performance, we implemented a 10-fold cross-validation approach. This method not only maximizes the use of limited data but also effectively reduces the risk of overfitting, providing a more reliable estimate of model performance. Using the StratifiedKFold method, we randomly divided the dataset into 10 equal-sized subsets. For each model, we performed 10 training and testing cycles, where 9 subsets were used as the training set and the remaining 1 subset served as the validation set.\u003c/p\u003e \u003cp\u003eFor performance evaluation during cross-validation, we calculated multiple metrics [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e], including Accuracy, Precision, Recall, F1 Score, Specificity, ROC-AUC, and Precision-Recall AUC (PR-AUC). After completing the cross-validation, we selected the model that demonstrated the best performance based on these metrics for external test set validation and final performance evaluation.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.6 Cox Proportional Hazards Regression Analysis\u003c/h2\u003e \u003cp\u003eTo assess the impact of various factors on the time to DGF recovery, we conducted a Cox proportional hazards regression analysis. First, we performed univariate Cox regression analysis on all variables to identify potential significant predictors. Using the survival package in R, we analyzed DGF recovery time as the outcome variable and the DGF status as the event indicator. For each variable, we calculated Hazard Ratio (HR), 95% Confidence Interval (CI) and p-value.\u003c/p\u003e \u003cp\u003eVariables with a p-value\u0026thinsp;\u0026lt;\u0026thinsp;0.1 in the univariate analysis were considered potential significant predictors and were included in the subsequent multivariate Cox regression analysis. The multivariate analysis allowed us to determine the independent effects of these significant predictors on DGF recovery time while adjusting for the influence of other variables. Finally, we visualized the results, providing insights into the key predictors and their relative contributions to DGF recovery.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e2.7 Follow-up Analysis\u003c/h2\u003e \u003cp\u003eThis study conducted long-term follow-up of patients from the transplantation center at the First Affiliated Hospital of China Medical University, concluding in October 2024, using a combination of outpatient follow-ups and telephone interviews, with patients considered lost to follow-up if they could not be contacted after three consecutive phone attempts or if they voluntarily withdrew from the study. Patients were categorized into four groups based on their DGF status and prognosis: DGF Death Group, DGF Survival Group, Non-DGF Death Group, and Non-DGF Survival Group. The primary endpoints included overall patient survival rate, graft survival rate, and mortality rate, while secondary endpoints focused on the longitudinal changes in renal function indicators, including serum creatinine, eGFR, blood urea, and cystatin C, as well as the incidence of clinical complications. Follow-up data were collected at postoperative intervals of 1, 3, 6, 12, 24, 36, and 48 months. Longitudinal data analysis was performed using the Generalized Estimating Equations (GEE) method with an exchangeable correlation structure, using the Non-DGF Survival Group as the control group, as this method accounts for the correlation between repeated measurements within individuals, is robust to missing data, and incorporates both time effects and group effects [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. The impact of DGF status and prognosis on renal function trends was assessed using coefficient estimates, confidence intervals, and model fit indices such as the Quasi-likelihood under the Independence Model Criterion (QIC), enabling a robust evaluation of how DGF status and prognosis influence long-term renal function and complication occurrences.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e2.8 Website Visualization\u003c/h2\u003e \u003cp\u003eTo enhance the clinical applicability and accessibility of our predictive models, we developed an interactive web-based visualization tool. This tool, accessible at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.kidney-dgf-match.cn/\u003c/span\u003e\u003cspan address=\"http://www.kidney-dgf-match.cn/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e, allows healthcare professionals to input patient-specific data and obtain real-time predictions for DGF risk. The interface was designed to be user-friendly, featuring separate sections for donor characteristics, recipient characteristics, organ preservation details, and immunology-related factors. Users can input a wide range of variables, including demographic information, laboratory test results, and transplant-specific parameters. The tool utilizes best models to provide predictions, offering a comparative view of the results from these two high-performing algorithms. This web-based tool not only serves as a practical application of our research findings but also as a potential decision support system for clinicians involved in kidney transplantation.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e2.9 Statistical analysis\u003c/h2\u003e \u003cp\u003eAll our data can be classified as continuous and categorical variables. We use Shapiro\u0026ndash;Wilk test to detect normal distribution in continuous variables, the two independent samples t-test for the normal distribution, and the Mann-Whitney U test for unnormal distribution. In categorical variables, Chi-Squared Test is used to calculate statistical significance. If P-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05, we consider it remains statistical significance. All model weights are visualized through the SHapley Additive exPlanations (SHAP) [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. All methods are generated in Jupiter-notebook with kernel Python 3.9 with the scikit package [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e"},{"header":"3 Result","content":"\u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Basic clinical characteristics comparation\u003c/h2\u003e \u003cp\u003eAccording to Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, this study compared the baseline clinical characteristics of kidney transplant recipients with DGF (292 cases) and without DGF (801 cases). Regarding donor characteristics, there were no significant differences between the two groups in terms of age, sex, height, weight, BMI, and body surface area (BSA) (P\u0026thinsp;\u0026gt;\u0026thinsp;0.05). However, significant differences were observed in donor blood type (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001), infection status (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001), and renal artery variation (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Specifically, the DGF group had a higher rate of donor infections (12.3% vs. 3.5%) but a lower incidence of renal artery variation (1.8% vs. 9.6%).\u003c/p\u003e \u003cp\u003eFor recipient characteristics, no significant differences were found between the two groups in blood type distribution, sex, height, weight, BMI, or urine output (P\u0026thinsp;\u0026gt;\u0026thinsp;0.05). However, notable differences were identified in ABO matching (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001), age matching (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001), preoperative systolic blood pressure (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001), pulse pressure (P\u0026thinsp;=\u0026thinsp;0.001), and mean arterial pressure (P\u0026thinsp;=\u0026thinsp;0.001), with the DGF group showing lower preoperative systolic and mean arterial pressures.\u003c/p\u003e \u003cp\u003eIn terms of immunological features, the DGF group had a higher number of HLA mismatches (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001), a higher rate of PRA positivity (12.3% vs. 7.0%, P\u0026thinsp;=\u0026thinsp;0.007), and was more likely to receive ATG as the induction therapy (77.2% vs. 83.1%, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Additionally, there were significant differences in the etiology of kidney disease (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001); glomerulonephritis was more prevalent in the DGF group (28.1% vs. 24.0%), while hypertensive nephropathy and diabetic nephropathy were more common in the non-DGF group.\u003c/p\u003e \u003cp\u003eThese findings suggest that certain donor and recipient characteristics, such as donor infection, HLA mismatches, and preoperative blood pressure, may be associated with the occurrence of DGF, providing important clues for further analysis.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eBaseline characteristics of kidney transplant recipients with and without DGF.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"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=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDGF(n\u0026thinsp;=\u0026thinsp;292)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNon-DGF(n\u0026thinsp;=\u0026thinsp;801)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eP value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eDonor\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDonor Age, median\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e54.0\u0026thinsp;\u0026plusmn;\u0026thinsp;10.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e53\u0026thinsp;\u0026plusmn;\u0026thinsp;11.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.844\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDonor Sex, n(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.971\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e248.0(85.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e677(84.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e44.0(14.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e124(15.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDonor Height_(cm), median\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e170.0\u0026thinsp;\u0026plusmn;\u0026thinsp;5.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e170\u0026thinsp;\u0026plusmn;\u0026thinsp;6.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.539\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDonor Weight_(kg), median\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e70.0\u0026thinsp;\u0026plusmn;\u0026thinsp;10.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e73\u0026thinsp;\u0026plusmn;\u0026thinsp;15.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.237\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDBMI, median\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e24.0\u0026thinsp;\u0026plusmn;\u0026thinsp;3.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e24.4\u0026thinsp;\u0026plusmn;\u0026thinsp;4.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.268\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDBSA, median\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.8\u0026thinsp;\u0026plusmn;\u0026thinsp;0.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.8\u0026thinsp;\u0026plusmn;\u0026thinsp;0.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.314\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDonor Blood Type, n(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eO\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e92.0(31.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e268.0(33.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e74.0(25.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e205.0(25.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eB\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e105.0(36.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e261.0(32.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAB\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e21.0(7.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e67.0(8.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDonor Hp, n(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.235\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e131.0(44.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e307.0(38.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eY\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e161.0(55.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e494.0(61.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDonor infect, n(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e256.0(87.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e773.0(96.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTreponema\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e31.0(10.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e13.0(1.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHBV\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.0(0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.0(0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHCV\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.0(0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.0(0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCoinfection\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e5.0(1.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e15.0(1.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKDPI, median\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e66.0\u0026thinsp;\u0026plusmn;\u0026thinsp;18.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e64.0\u0026thinsp;\u0026plusmn;\u0026thinsp;20.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.530\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKDRI, median\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.2\u0026thinsp;\u0026plusmn;\u0026thinsp;0.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.2\u0026thinsp;\u0026plusmn;\u0026thinsp;0.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.795\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRenal Artery Variation, n(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.000\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e287.0(98.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e724.0(90.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eY\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e5.0(1.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e77.0(9.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRenal Vein Variation, n(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.147\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e292.0(100.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e793.0(99.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eY\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.0(0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e8.0(1.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eRecipient\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCIT, median\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7.6\u0026thinsp;\u0026plusmn;\u0026thinsp;3.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7.0\u0026thinsp;\u0026plusmn;\u0026thinsp;3.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.592\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBlood Type, n(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.469\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eO\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e84.0(28.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e269.0(33.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e74.0(25.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e200.0(24.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eB\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e105.0(36.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e266.0(33.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAB\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e29.0(9.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e66.0(8.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eABO MM, n(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eABO-identical\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e277.0(94.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e745.0(93.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eABO-compatible\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e15.0(5.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e56.0(7.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRecipient Sex, n(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.625\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e205.0(70.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e586.0(73.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e87.0(29.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e215.0(26.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSex MM, n(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.646\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale to male\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e187.0(64.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e512.0(63.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale to female\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e61.0(21.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e163.0(20.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemale to male\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e18.0(6.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e75.0(9.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemale to female\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e26.0(8.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e51.0(6.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRecipient Age, median\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e46.0\u0026thinsp;\u0026plusmn;\u0026thinsp;11.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e46\u0026thinsp;\u0026plusmn;\u0026thinsp;10.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.797\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge MM, n(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026gt;50 to \u0026gt;50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e49.0(16.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e259.0(32.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026gt;50 to \u0026lt;50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e156.0(53.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e225.0(28.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026lt;50 to \u0026gt;50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e36.0(12.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e79.0(9.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026lt;50 to \u0026lt;50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e51.0(17.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e238.0(29.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHeight_(cm), median\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e171.0\u0026thinsp;\u0026plusmn;\u0026thinsp;8.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e172\u0026thinsp;\u0026plusmn;\u0026thinsp;8.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.481\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWeight_(kg), median\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e68.0\u0026thinsp;\u0026plusmn;\u0026thinsp;15.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e69\u0026thinsp;\u0026plusmn;\u0026thinsp;13.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.630\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBMI, median\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e23.2\u0026thinsp;\u0026plusmn;\u0026thinsp;4.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e23.6\u0026thinsp;\u0026plusmn;\u0026thinsp;3.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.547\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBMIr, median\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.9\u0026thinsp;\u0026plusmn;\u0026thinsp;0.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.0\u0026thinsp;\u0026plusmn;\u0026thinsp;0.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.171\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBMIr to 1, n(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.089\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e159.0(54.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e360.0(45.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eY\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e133.0(45.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e441.0(55.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRBSA, median\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.8\u0026thinsp;\u0026plusmn;\u0026thinsp;0.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.8\u0026thinsp;\u0026plusmn;\u0026thinsp;0.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.571\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBSAr, median\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.0\u0026thinsp;\u0026plusmn;\u0026thinsp;0.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.0\u0026thinsp;\u0026plusmn;\u0026thinsp;0.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.181\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBSAr to 1, n(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.007\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e84.0(28.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e405.0(50.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eY\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e208.0(71.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e396.0(49.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRNE, n(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGlomerulonephritis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e82.0(28.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e192.0(24.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHypertensive nephropathy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e10.0(3.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e54.0(6.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDiabetic nephropathy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e28.0(9.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e54.0(6.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePolycystic kidney\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e8.0(2.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e30.0(3.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUnknown\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e164.0(56.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e471.0(58.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePdm, n(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.135\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3.0(0.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e28.0(3.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHemodialysis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e261.0(89.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e658.0(82.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePeritoneal dialysis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e28.0(9.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e115.0(14.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePSBP, median\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e144.0\u0026thinsp;\u0026plusmn;\u0026thinsp;21.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e152\u0026thinsp;\u0026plusmn;\u0026thinsp;21.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePDP, median\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e89\u0026thinsp;\u0026plusmn;\u0026thinsp;13.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e90\u0026thinsp;\u0026plusmn;\u0026thinsp;13.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.029\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePSDP, median\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e54.0\u0026thinsp;\u0026plusmn;\u0026thinsp;15.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e60\u0026thinsp;\u0026plusmn;\u0026thinsp;15.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePMAP, median\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e116\u0026thinsp;\u0026plusmn;\u0026thinsp;16.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e120\u0026thinsp;\u0026plusmn;\u0026thinsp;16.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRPU, n(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.312\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e230.0(78.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e578.0(72.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eY\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e62.0(21.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e223.0(27.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHLA MM, n(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026le;\u0026thinsp;3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e74.0(25.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e361.0(45.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026gt;4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e218.0(74.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e440.0(54.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePRA, n(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.007\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNegative\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e256.0(87.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e745.0(93.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePositive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e36.0(12.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e56.0(7.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eImmunoinduction protocols, n(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBasiliximab\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e67.0(22.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e135.0(16.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eATG\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e225.0(77.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e666.0(83.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Visualization of Model Performance\u003c/h2\u003e \u003cp\u003eWe divided the training dataset into a 7:3 ratio for model development, and as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, the Random Forest and LightGBM models demonstrated the best performance with an AUC of 0.79, followed closely by the XGBoost model with an AUC of 0.76. However, in terms of accuracy, the LightGBM model outperformed all other models, achieving an accuracy of 80%, followed by XGBoost and Random Forest, both with an accuracy of 79%.\u003c/p\u003e \u003cp\u003eConsidering that a single train-test split may introduce randomness and that the Random Forest model is prone to overfitting, potentially limiting its generalization ability, we further employed a 10-fold cross-validation method to evaluate model performance. Figure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e illustrates the results of the 10-fold cross-validation, revealing the consistently strong and stable performance of the Random Forest, XGBoost, and LightGBM models across multiple validation rounds.\u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e provides detailed performance metrics for each model during 10-fold cross-validation. Notably, the Random Forest, LightGBM, and XGBoost models achieved significantly higher average accuracy and AUC compared to other models, showcasing their robust predictive power and stability.\u003c/p\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e3.3 External Validation Results\u003c/h2\u003e \u003cp\u003eAfter performing 10-fold cross-validation on the training set and selecting the top three models, we further validated the generalization ability of the Random Forest, XGBoost, and LightGBM models using an external dataset.\u003c/p\u003e \u003cp\u003eAs shown in the ROC curves and performance metrics (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e \u0026amp; Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e3\u003c/span\u003e), the LightGBM model demonstrated the best performance during external validation. It achieved the highest scores across all evaluation metrics, with an ROC-AUC of 0.80, accuracy of 0.73, precision of 0.69, recall of 0.35, F1 score of 0.47, and specificity of 0.92. These results indicate that the LightGBM model performed exceptionally well in distinguishing between positive and negative samples, showcasing strong generalization ability.\u003c/p\u003e \u003cp\u003e The XGBoost model followed closely behind, delivering solid performance with an ROC-AUC of 0.73 and accuracy of 0.71. While it outperformed the Random Forest model across most metrics, it fell slightly short of the LightGBM model. In contrast, the Random Forest model showed relatively weaker performance on the external dataset. Its ROC AUC was 0.63, and its accuracy was 0.67, with noticeably lower precision, recall, and F1 scores compared to the other two models. However, it is worth noting that the Random Forest model achieved a high specificity of 0.98, indicating strong performance in identifying negative samples.\u003c/p\u003e\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e3.4 Feature Importance Visualization\u003c/h2\u003e \u003cp\u003eTo better understand the decision-making process of the models and enhance their interpretability, we performed SHAP (SHapley Additive exPlanations) analysis on the top-performing XGBoost and LightGBM models. SHAP values quantify the contribution of each feature to the model\u0026rsquo;s predictions, providing insights into the key factors driving DGF prediction.\u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e presents the SHAP summary plot for the LightGBM model, clearly highlighting the top five features influencing DGF prediction: pBNP, DeGFR, DHb, DuEC, and DuRBC. The analysis revealed that lower DeGFR values significantly increased the risk of DGF, which aligns with clinical understanding, as decreased glomerular filtration rate typically reflects poorer kidney function. Additionally, higher pBNP levels were associated with an increased risk of DGF, potentially reflecting the impact of underlying cardiovascular issues on transplant outcomes. Interestingly, higher DHb values and mild abnormalities in routine urine tests, such as elevated DuEC and DuRBC, also showed a trend of increased DGF risk.\u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e displays the SHAP summary plot for the XGBoost model, which further corroborates the importance of several key features. Similar to the LightGBM model, DeGFR, DHb, and pBNP were confirmed as critical predictors. Additionally, the XGBoost model identified DAPPT and pBUN as important features. Notably, lower DAPPT values were associated with an increased risk of DGF, offering a new perspective on the role of donor and recipient coagulation status in transplant outcomes.\u003c/p\u003e \u003cp\u003eThe consistency between the two models in identifying major predictors, particularly DeGFR, DHb, and pBNP, strengthens confidence in the significance of these features for DGF prediction. At the same time, the unique features identified by each model, such as DAPPT and pBUN, provide diverse perspectives for risk assessment, helping to capture the intricate network of factors influencing DGF.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e3.5 DGF Recovery Time\u003c/h2\u003e \u003cp\u003eUnivariate Cox regression analysis (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eA) identified several factors associated with DGF recovery time (p\u0026thinsp;\u0026lt;\u0026thinsp;0.1). Among these, donor type (HR\u0026thinsp;=\u0026thinsp;1.66, 95% CI: 1.233\u0026ndash;2.236, p\u0026thinsp;=\u0026thinsp;0.0008), recipient serum creatinine level (HR\u0026thinsp;=\u0026thinsp;1.002, 95% CI: 1.000\u0026ndash;1.003, p\u0026thinsp;=\u0026thinsp;0.0044), and donor serum calcium level (HR\u0026thinsp;=\u0026thinsp;0.246, 95% CI: 0.088\u0026ndash;0.685, p\u0026thinsp;=\u0026thinsp;0.0073) had statistically significant impacts on recovery time. Other factors with significant associations included recipient weight, primary kidney disease type, recipient BMI, body surface area, dialysis duration, and donor serum magnesium levels.\u003c/p\u003e \u003cp\u003eIn multivariate Cox regression analysis (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eB), two independent predictors were strongly associated with DGF recovery time after stepwise selection. First, dialysis duration (Pdt) showed statistical significance (HR\u0026thinsp;=\u0026thinsp;1.006, 95% CI: 1.001\u0026ndash;1.012, p\u0026thinsp;=\u0026thinsp;0.021), indicating that each additional unit of dialysis time increased the risk of prolonged DGF recovery by 0.6%. Second, recipient preoperative systolic blood pressure (PsBp) exhibited marginal significance (HR\u0026thinsp;=\u0026thinsp;0.094, 95% CI: 0.998\u0026ndash;1.032, p\u0026thinsp;=\u0026thinsp;0.094), suggesting that it may have a modest influence on DGF recovery time, though its p-value slightly exceeded the traditional threshold of 0.05.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e3.6 Long-Term Prognosis of DGF\u003c/h2\u003e \u003cp\u003eLongitudinal analysis of renal function parameters over time (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e) revealed distinct trends among the four patient groups. The Non-DGF Survival Group showed the most stable renal function, with a steady increase in eGFR (from 60.90 to 74.89 mL/min/1.73m\u0026sup2;) and consistently low serum creatinine levels (101.00\u0026ndash;133.30 \u0026micro;mol/L). In contrast, the DGF Survival Group exhibited poorer early renal function but gradual improvement over time, with serum creatinine decreasing from 255.26 to 122.93 \u0026micro;mol/L, although a significant increase was observed at 36 months (289.25 \u0026micro;mol/L). The Death Groups (both DGF Death and Non-DGF Death) showed highly variable and generally worse trends, particularly in the DGF Death Group, where serum creatinine and urea levels were markedly elevated in the early stages.\u003c/p\u003e\u003cp\u003eTo quantify the differences among groups, a GEE model was applied using the Non-DGF Survival Group as the reference (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e3\u003c/span\u003e). The analysis showed that the DGF Death Group exhibited the most severe renal dysfunction, with significantly higher serum creatinine (β\u0026thinsp;=\u0026thinsp;200.57, 95% CI: 78.30\u0026ndash;322.84, p\u0026thinsp;=\u0026thinsp;0.001), significantly lower eGFR (β = -39.91, 95% CI: -47.61 to -32.21, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), and significantly elevated levels of urea (β\u0026thinsp;=\u0026thinsp;13.51, 95% CI: 7.40\u0026ndash;19.61, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and cystatin C (β\u0026thinsp;=\u0026thinsp;1.64, 95% CI: 0.92\u0026ndash;2.36, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). The Non-DGF Death Group showed significant elevation only in urea levels (β\u0026thinsp;=\u0026thinsp;8.89, 95% CI: 2.83\u0026ndash;14.94, p\u0026thinsp;=\u0026thinsp;0.004). Meanwhile, the DGF Survival Group exhibited no statistically significant differences in most renal function parameters compared to the control group, except for a slight increase in cystatin C levels (β\u0026thinsp;=\u0026thinsp;0.58, 95% CI: 0.09\u0026ndash;1.06, p\u0026thinsp;=\u0026thinsp;0.020).\u003c/p\u003e \u003cp\u003eTime effect analysis indicated that urea (β = -0.11, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and cystatin C (β = -0.02, p\u0026thinsp;=\u0026thinsp;0.020) levels decreased significantly over time, suggesting that renal function in surviving patients gradually improved. These findings suggest that while DGF patients experience slower early recovery, their long-term prognosis in terms of renal function is similar to that of Non-DGF patients, provided they survive the early stages.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparison of Dialysis-related Characteristics Between DGF Survival and Death Groups.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDGF Survival (n\u0026thinsp;=\u0026thinsp;38)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDGF Death\u003c/p\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;9)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eP value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDGF Duration (days)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e16.66\u0026thinsp;\u0026plusmn;\u0026thinsp;13.73\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e15.44\u0026thinsp;\u0026plusmn;\u0026thinsp;14.62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.735\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNumber of Dialysis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e8.13\u0026thinsp;\u0026plusmn;\u0026thinsp;7.39\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e7.78\u0026thinsp;\u0026plusmn;\u0026thinsp;7.22\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.807\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMean Dialysis Interval (days)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e1.89\u0026thinsp;\u0026plusmn;\u0026thinsp;0.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e1.75\u0026thinsp;\u0026plusmn;\u0026thinsp;0.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.649\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 \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003e3.7 Interactive Web-Based Tool\u003c/h2\u003e \u003cp\u003eTo enhance the clinical utility and accessibility of our predictive model, we developed an interactive web-based visualization tool (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.kidney-dgf-match.cn/\u003c/span\u003e\u003cspan address=\"http://www.kidney-dgf-match.cn/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). The website features an intuitive and user-friendly interface that organizes the predictive variables into multiple modules, including donor characteristics, recipient characteristics, organ preservation, and immunology (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eClinicians can input patient demographic information, laboratory test results, and transplant-related parameters, and the system will compute the risk probability of DGF in real-time. The tool incorporates the significant predictive variables identified in this study, with their respective weight coefficients integrated into the calculation model, providing transplant physicians with a convenient decision support tool.\u003c/p\u003e \u003cp\u003eThis online platform represents a practical application of our research findings and offers healthcare professionals in the field of kidney transplantation an objective risk assessment method. By enabling personalized perioperative management strategies, it aims to improve patient outcomes and optimize transplant success rates.\u003c/p\u003e\u003c/div\u003e"},{"header":"4 Discussion","content":"\u003cp\u003eAmong all the machine learning models evaluated, the ensemble learning models XGBoost and LightGBM demonstrated outstanding predictive performance on the external validation dataset, which can be attributed to their unique algorithmic features [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. XGBoost (eXtreme Gradient Boosting) incorporates a second-order Taylor expansion to approximate the objective function and includes a regularization term to control model complexity, enhancing its robustness in handling nonlinear relationships. Furthermore, its innovative split-point finding algorithm efficiently processes sparse data, making it particularly advantageous in addressing the prevalent issue of missing values in medical datasets. In contrast, LightGBM's standout features include the \"Gradient-based One-Side Sampling\" (GOSS) and \"Exclusive Feature Bundling\" (EFB) strategies. GOSS retains samples with large gradients while randomly sampling those with small gradients, significantly reducing computational complexity while maintaining training accuracy. This was particularly effective in handling our imbalanced dataset, as DGF is a relatively rare clinical event. Additionally, LightGBM's leaf-wise growth strategy, which prioritizes optimal splits, allowed the model to better capture the complex nonlinear relationships in medical data, resulting in superior predictive performance on the external validation set (ROC-AUC\u0026thinsp;=\u0026thinsp;0.80, accuracy\u0026thinsp;=\u0026thinsp;0.73).\u003c/p\u003e \u003cp\u003eIn terms of data preprocessing, the IterativeImputer outperformed KNNImputer in both prediction accuracy and model robustness. This may be due to the inherent limitations of KNNImputer, which relies on finding the most similar neighbors. As data dimensions increase, the definition of similarity becomes less reliable, and the heterogeneity of populations in kidney transplantation amplifies the errors introduced by this method. Conversely, IterativeImputer focuses on the complex interrelationships between features, leveraging these connections to handle missing values more effectively, which resulted in better outcomes. Thus, for large-scale medical machine learning, IterativeImputer appears to be a more suitable choice.\u003c/p\u003e \u003cp\u003eSHAP visualizations of the two models identified several groups of features with significant predictive value for DGF, spanning organ function, blood status, and immunological factors. Among donor-related factors, DeGFR was the most significant predictor, with lower donor GFR values substantially increasing the risk of DGF. This finding aligns with clinical experience, as poor baseline kidney function directly impacts early graft recovery [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. Donor hemoglobin (DHb) levels showed a complex bidirectional influence: moderately elevated Hb levels may improve oxygen delivery to tissues, but excessively high Hb levels might indicate hemoconcentration, inadequate fluid maintenance, or poor organ perfusion, increasing the risk of perfusion-related injury. This highlights the importance of precise management of donor hemoglobin and fluid levels to optimize organ perfusion quality during the maintenance period. Additionally, recipient preoperative blood urea nitrogen (pBUN) levels showed significant predictive value, likely reflecting the impact of poor preoperative dialysis quality, metabolic imbalances, and nitrogen retention [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn terms of cardiovascular function, recipient preoperative plasma BNP (pBNP) levels emerged as a strong predictor. Elevated pBNP levels suggest impaired cardiac function, potentially reducing cardiac output and transplant kidney perfusion pressure, thereby increasing the risk of DGF [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. This finding underscores the importance of comprehensive cardiovascular assessments for both donors and recipients during pre-transplant evaluations.\u003c/p\u003e \u003cp\u003eInterestingly, the study also identified several unconventional but clinically significant predictors. Microscopic hematuria (DuEC, DuRBC), which is well-established as a marker for chronic kidney disease progression, has been relatively understudied in transplantation but was found to be a valuable predictor for DGF [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Additionally, decreased donor APTT levels, indicating hypercoagulability, may increase the risk of microthrombosis, potentially exacerbating ischemia-reperfusion injury and impairing graft microcirculation [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. This finding highlights the potential interaction between coagulation and immune responses in the development of DGF, although further experimental evidence is needed to validate this hypothesis [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. Potassium ion (K⁺) levels also demonstrated clinical relevance, as hypokalemia not only reflects simple electrolyte imbalance but may also indicate impaired tubular reabsorption and membrane instability, which could negatively impact early graft function recovery [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eFurther insights into the clinical timeline of DGF were provided by Cox regression and long-term prognosis analyses. Univariate Cox analysis revealed that donor type, recipient serum creatinine levels, and donor serum calcium levels significantly influenced DGF recovery time, highlighting the critical roles of organ quality and recipient baseline status in DGF duration [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. Multivariate analysis identified preoperative dialysis duration as an independent predictor, revealing a strong association between prolonged dialysis burden and delayed DGF recovery [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]. This emphasizes the need for tailored management strategies for long-term dialysis patients in clinical practice.\u003c/p\u003e \u003cp\u003eLong-term prognosis analysis demonstrated the complex impacts of DGF on transplant outcomes. By categorizing patients into DGF Death, DGF Survival, Non-DGF Death, and Non-DGF Survival groups, several key findings emerged. The DGF Death Group exhibited the most severe renal dysfunction, with significantly higher serum creatinine (β\u0026thinsp;=\u0026thinsp;200.57) and lower eGFR (β = -39.91), reflecting the poor prognosis associated with severe DGF. However, an important observation was that the DGF Survival Group showed renal function comparable to the Non-DGF Survival Group in most metrics, with only slightly elevated cystatin C levels (β\u0026thinsp;=\u0026thinsp;0.58). This underscores that the occurrence of DGF does not necessarily lead to poor long-term outcomes, provided that early identification and timely interventions are implemented. Notably, comparisons between the DGF Survival and Death groups revealed no significant differences in DGF duration (16.66\u0026thinsp;\u0026plusmn;\u0026thinsp;13.73 vs. 15.44\u0026thinsp;\u0026plusmn;\u0026thinsp;14.62 days, p\u0026thinsp;=\u0026thinsp;0.735), number of dialysis sessions (8.13\u0026thinsp;\u0026plusmn;\u0026thinsp;7.39 vs. 7.78\u0026thinsp;\u0026plusmn;\u0026thinsp;7.22, p\u0026thinsp;=\u0026thinsp;0.807), or mean dialysis interval (1.89\u0026thinsp;\u0026plusmn;\u0026thinsp;0.58 vs. 1.75\u0026thinsp;\u0026plusmn;\u0026thinsp;0.70 days, p\u0026thinsp;=\u0026thinsp;0.649). This unexpected finding suggests that these characteristics may not be decisive factors in determining long-term prognosis, which is more likely influenced by overall patient condition and perioperative management quality.\u003c/p\u003e \u003cp\u003eLongitudinal analysis showed significant declines in urea and cystatin C levels over time, supporting the notion that, even in the presence of DGF, most surviving patients can achieve gradual renal function improvement with appropriate management. These findings carry important clinical implications. Firstly, they emphasize the importance of early identification of high-risk patients. Secondly, they suggest that clinicians should not view DGF as a definitive indicator of poor prognosis but rather focus on early intervention and individualized treatment plans. Finally, these results provide transplant physicians with more detailed evidence for communicating long-term outcomes to patients [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAlthough this study has made significant progress in developing the first DGF prediction model based on multicenter data from China, several limitations require further exploration and refinement in future research. First, given the vast geographic diversity of China, regional differences in population characteristics (e.g., living environment, dietary habits) and medical practices may introduce heterogeneity, impacting the model's applicability. While this study used multicenter data, it primarily involved institutions from northern China. Expanding data collection to include more diverse regions and identifying population-specific factors will improve the model's generalizability. Second, the study spans a long data collection period, during which kidney transplantation techniques and perioperative management protocols may have evolved. Temporal effects may influence model accuracy and require additional attention in future research. Furthermore, ethical considerations in artificial intelligence, particularly ensuring unbiased model application in clinical practice and enhancing interpretability, represent critical challenges. Incorporating novel biomarkers, such as genomic and transcriptomic data, into feature selection could further improve predictive power. Finally, prospective validation studies are essential for evaluating the model's effectiveness in real-world clinical applications, which represents a key direction for future research.\u003c/p\u003e"},{"header":"5 Conclusion","content":"\u003cp\u003eThis study developed the first DGF prediction model based on multicenter data from China, integrating machine learning techniques with clinical data to achieve accurate post-transplant DGF risk prediction. In external validation, the LightGBM model demonstrated the best predictive performance (AUC\u0026thinsp;=\u0026thinsp;0.80), providing a reliable decision-support tool for clinicians. SHAP analysis confirmed the importance of traditional risk factors such as donor eGFR and recipient preoperative plasma BNP levels while highlighting the potential value of novel predictors like microscopic hematuria and donor APTT. Cox regression analysis identified dialysis duration as a key factor influencing DGF recovery, while long-term follow-up results indicated that, although DGF increases early mortality risk, it does not significantly affect long-term outcomes in surviving patients. Importantly, DGF duration, number of dialysis sessions, and mean dialysis interval were not decisive factors for long-term prognosis. Based on these findings, the web-based prediction platform (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.kidney-dgf-match.cn/\u003c/span\u003e\u003cspan address=\"http://www.kidney-dgf-match.cn/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) was developed to assist clinicians in risk assessment and personalized treatment planning. This predictive model not only fills a research gap in the field of kidney transplantation in China but also provides new tools and insights for improving transplant outcomes.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledge\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe thank everyone in the department for their excellent work and selfless help.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePS and ZD conceptualized and designed the study. PS, XW, JL, KW, ZY, and WZ were responsible for data collection and curation. ZD performed the computational analysis, developed the machine learning models, and generated the figures. PS and ZD conducted the data analysis and interpretation. PS, ZD, and XW drafted the initial manuscript. HS, YC, LH, KM, and GW provided critical revision of the manuscript for important intellectual content and contributed to data collection from their respective centers. TG contributed to the study design. All authors reviewed and approved the final version of the manuscript. HS and YC supervised the project and provided administrative support.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding information\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCentral Government Support for Local Development 2023: Special Project for Construction of Scarce Disciplines in Strategic Emerging Industries - Experimental Surgery (Project No. 3110230005)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo protect patient privacy and confidentiality, the raw data containing sensitive patient information used in this study are not publicly available. However, these data can be made available upon reasonable request to the first author or corresponding author, subject to appropriate data sharing agreements and ethical approvals.\u003c/p\u003e\n\u003cp\u003eThe code used for data analysis and model development in this study is publicly accessible. All relevant scripts and code files can be found in our GitHub repository at\u0026nbsp;https://github.com/Mr-SiSi/DGF-prediction-model-in-kidney-transplantation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study protocol was reviewed and approved by the Ethics Committee of the First Affiliated Hospital of China Medical University (approval number: AF-SOP-07-1.2-01). Due to the retrospective nature of this study, which involved the analysis of existing clinical data, the requirement for individual patient consent was waived by the ethics committee.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of Interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eHalloran PF, Hunsicker LG (2000) Delayed graft function: state of the art, November 10\u0026ndash;11, Summit meeting, Scottsdale, Arizona, USA. \u003cem\u003eAm J Transplant\u003c/em\u003e. 2001;1(2):115\u0026ndash;120\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLentine KL, Smith JM, Hart A et al (2022) OPTN/SRTR 2020 Annual Data Report: Kidney. Am J Transpl 22(Suppl 2):21\u0026ndash;136\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePerico N, Cattaneo D, Sayegh MH, Remuzzi G Delayed graft function in kidney transplantation. Lancet 2004 Nov 13\u0026ndash;19;364(9447):1814\u0026ndash;1827\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSiedlecki A, Irish W, Brennan DC (2011) Delayed graft function in the kidney transplant. Am J Transpl 11(11):2279\u0026ndash;2296\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIrish WD, Ilsley JN, Schnitzler MA, Feng S, Brennan DC (2010) A risk prediction model for delayed graft function in the current era of deceased donor renal transplantation. Am J Transpl 10(10):2279\u0026ndash;2286\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChapal M, Le Borgne F, Legendre C et al (2014) A useful scoring system for the prediction and management of delayed graft function following kidney transplantation from cadaveric donors. Kidney Int 86(6):1130\u0026ndash;1139\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYoo D, Divard G, Raynaud M et al (2024) A Machine Learning-Driven Virtual Biopsy System For Kidney Transplant Patients. Nat Commun 15(1):554 Published 2024 Jan 16\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee KH, Choi GH, Yun J et al (2024) Machine learning-based clinical decision support system for treatment recommendation and overall survival prediction of hepatocellular carcinoma: a multi-center study. NPJ Digit Med. ;7(1):2. Published 2024 Jan 5\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePlatias C, Petasis G (2020) A comparison of machine learning methods for data imputation[C]//11th Hellenic Conference on Artificial Intelligence. : 150\u0026ndash;159\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKaramti H, Alharthi R, Anizi AA et al (2023) Improving Prediction of Cervical Cancer Using KNN Imputed SMOTE Features and Multi-Model Ensemble Learning Approach. Cancers (Basel) 15(17):4412 Published 2023 Sep 4\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHancock IIIJT, Khoshgoftaar TM (2023) Exploring maximum tree depth and random undersampling in ensemble trees to optimize the classification of imbalanced big data[J]. SN Comput Sci 4(5):462\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMeurer WJ, Tolles J (2017) Logistic Regression Diagnostics: Understanding How Well a Model Predicts Outcomes. JAMA 317(10):1068\u0026ndash;1069\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOntivero-Ortega M, Lage-Castellanos A, Valente G, Goebel R, Valdes-Sosa M (2017) Fast Gaussian Na\u0026iuml;ve Bayes for searchlight classification analysis. NeuroImage 163:471\u0026ndash;479\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFlayer CH, Perner C, Sokol CL (2021) A decision tree model for neuroimmune guidance of allergic immunity. Immunol Cell Biol 99(9):936\u0026ndash;948\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang L, Wu H, Jin X, Zheng P, Hu S, Xu X, Yu W, Yan J (2020) Study of cardiovascular disease prediction model based on random forest in eastern China. Sci Rep 10(1):5245\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang S, Cai N, Pacheco PP, Narrandes S, Wang Y, Xu W (2018) Applications of Support Vector Machine (SVM) Learning in Cancer Genomics. Cancer Genomics Proteomics 15(1):41\u0026ndash;51\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHandelman GS, Kok HK, Chandra RV, Razavi AH, Lee MJ, Asadi H (2018) eDoctor: machine learning and the future of medicine. J Intern Med 284(6):603\u0026ndash;619\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee JY, Styczynski MP (2018) NS-kNN: a modified k-nearest neighbors approach for imputing metabolomics data. Metabolomics: Official J Metabolomic Soc 14(12):153\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKriegeskorte N, Golan T (2019) Neural network models and deep learning. Curr biology: CB 29(7):R231\u0026ndash;R236\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOgunleye A, Wang QG (2020) XGBoost Model for Chronic Kidney Disease Diagnosis. IEEE/ACM Trans Comput Biol Bioinf 17(6):2131\u0026ndash;2140\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen T, Guestrin C (2016), August Xgboost: A scalable tree boosting system. In Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining (pp. 785\u0026ndash;794)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHerron P (2004) Machine learning for medical decision support: evaluating diagnostic performance of machine learning classification algorithms. INLS 110:1\u0026ndash;15\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang M (2014) Generalized estimating equations in longitudinal data analysis: a review and recent developments. Adv Stat 2014(1):303728\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRao S, Mehta S, Kulkarni S, Dalvi H, Katre N, Narvekar M (2022), December A study of LIME and SHAP model explainers for autonomous disease predictions. In \u003cem\u003e2022 IEEE Bombay Section Signature Conference (IBSSC)\u003c/em\u003e (pp. 1\u0026ndash;6). IEEE\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHao J, Ho TK (2019) Machine learning made easy: a review of scikit-learn package in python programming language. J Educational Behav Stat 44(3):348\u0026ndash;361\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang D, Gong Y (2020) The comparison of LightGBM and XGBoost coupling factor analysis and prediagnosis of acute liver failure. Ieee Access 8:220990\u0026ndash;221003\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMogulla MR, Bhattacharjya S, Clayton PA (2019) Risk factors for and outcomes of delayed graft function in live donor kidney transplantation\u0026ndash;a retrospective study. Transpl Int 32(11):1151\u0026ndash;1160\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee SW, Yang YM, Kim HY, Cho H, Nam SW, Kim SM, Kwon SK (2022) Predialysis Urea Nitrogen Is a Nutritional Marker of Hemodialysis Patients. Chonnam Med J 58(2):69\u0026ndash;74\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJarolim P, Claggett BL, Conrad MJ, Carpenter MA, Ivanova A, Bostom AG, Kusek JW, Hunsicker LG, Jacques PF, Gravens-Mueller L, Finn P, Solomon SD, Weiner DE, Levey AS, Pfeffer MA (2017) B-Type Natriuretic Peptide and Cardiac Troponin I Are Associated With Adverse Outcomes in Stable Kidney Transplant Recipients. Transplantation 101(1):182\u0026ndash;190\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOkada S, Samejima KI, Matsui M et al (2020) Microscopic hematuria is a risk factor for end-stage kidney disease in patients with biopsy-proven diabetic nephropathy. BMJ Open Diabetes Res Care 8(2):e001863\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSood P, Randhawa PS, Mehta R, Hariharan S, Tevar AD (2015) Donor kidney microthrombi and outcomes of kidney transplant: a single-center experience. Clin Transplant 29(5):434\u0026ndash;438\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWilhelm G, Mertowska P, Mertowski S, Przysucha A, Strużyna J, Grywalska E, Torres K (2023) The Crossroads of the Coagulation System and the Immune System: Interactions and Connections. Int J Mol Sci 24(16):12563\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKardalas E, Paschou SA, Anagnostis P, Muscogiuri G, Siasos G, Vryonidou A (2018) Hypokalemia: a clinical update. Endocr connections 7(4):R135\u0026ndash;R146\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSiedlecki A, Irish W, Brennan DC (2011) Delayed graft function in the kidney transplant. Am J transplantation: official J Am Soc Transplantation Am Soc Transpl Surg 11(11):2279\u0026ndash;2296\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFreitas MHB, Lima LC, Couceiro TCM, Silva WBD, Andrade JM, Freitas MHB (2018) Perioperative factors associated with delayed graft function in renal transplant patients. Jornal brasileiro de nefrologia 40(4):360\u0026ndash;365\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNeuberger JM, Bechstein WO, Kuypers DR, Burra P, Citterio F, De Geest S, Duvoux C, Jardine AG, Kamar N, Kr\u0026auml;mer BK, Metselaar HJ, Nevens F, Pirenne J, Rodr\u0026iacute;guez-Per\u0026aacute;lvarez ML, Samuel D, Schneeberger S, Ser\u0026oacute;n D, Trunečka P, Tisone G, van Gelder T (2017) Practical Recommendations for Long-term Management of Modifiable Risks in Kidney and Liver Transplant Recipients: A Guidance Report and Clinical Checklist by the Consensus on Managing Modifiable Risk in Transplantation (COMMIT) Group. Transplantation 101(4):S1\u0026ndash;S56\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTables 2 and 3 are available in the Supplementary Files section.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Delayed Graft Function (DGF), Kidney Transplantation, Machine Learning, Multicenter Study, Post-Transplant Outcomes","lastPublishedDoi":"10.21203/rs.3.rs-5617823/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5617823/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eDelayed graft function (DGF) is a severe complication following kidney transplantation, and currently, there is a lack of accurate prediction tools tailored for the Chinese population. This study integrates data from 1,093 kidney transplant cases across four medical centers in China (2016–2024) to develop and validate a machine learning-based model for DGF prediction. By comparing nine machine learning algorithms, we found that the LightGBM model performed best in external validation (AUC = 0.80, accuracy = 0.73). SHAP analysis identified donor GFR, donor hemoglobin, and recipient plasma BNP levels as the primary predictive factors, while also highlighting novel predictors such as donor microscopic hematuria and APTT. Cox regression analysis showed that preoperative dialysis duration in recipients (HR = 1.006, 95% CI: 1.001–1.012) was an independent predictor of DGF recovery. In the follow-up study, we observed that while the DGF mortality group exhibited the most significant kidney function impairment (serum creatinine β = 200.57, eGFR β = -39.91), the prognosis of the DGF survival group was comparable to that of the non-DGF survival group. Additionally, the duration of DGF (16.66 ± 13.73 vs. 15.44 ± 14.62 days) and the number of dialysis treatments (8.13 ± 7.39 vs. 7.78 ± 7.22 sessions) were not significantly associated with prognosis. Based on these findings, we developed an online prediction platform (www.kidney-dgf-match.cn) to support clinical decision-making. This study not only establishes the first high-precision DGF prediction model for the Chinese population but also reveals the potential for favorable outcomes in DGF patients with proper management, offering new insights for optimizing post-transplant management strategies.\u003c/p\u003e","manuscriptTitle":"A High-Precision Machine Learning-Based Prediction Model for Delayed Graft functon(DGF) in Chinese Kidney Transplant Patients: A Multicenter Study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-01-06 14:35:47","doi":"10.21203/rs.3.rs-5617823/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"3dbc6aaa-8655-4710-bfea-33c58486eb70","owner":[],"postedDate":"January 6th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":42324253,"name":"Health sciences/Nephrology"},{"id":42324254,"name":"Health sciences/Medical research/Translational research"},{"id":42324255,"name":"Health sciences/Medical research/Outcomes research"},{"id":42324256,"name":"Health sciences/Medical research/Experimental models of disease"}],"tags":[],"updatedAt":"2025-04-11T15:20:48+00:00","versionOfRecord":[],"versionCreatedAt":"2025-01-06 14:35:47","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5617823","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5617823","identity":"rs-5617823","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}