Predicting Clinical Outcomes in Membranous Nephropathy Using Machine Learning

Predicting Clinical Outcomes in Membranous Nephropathy Using Machine Learning | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Predicting Clinical Outcomes in Membranous Nephropathy Using Machine Learning Yuming Lu, Xiquan Lu, Cuiying Yu, Pengjie Sha, Yue Cao, Dan Wang, and 6 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9019166/v1 This work is licensed under a CC BY 4.0 License Status: Under Revision Version 1 posted 12 You are reading this latest preprint version Abstract Objective This study aimed to develop and compare machine learning models for predicting clinical outcomes—Complete Remission (CR), partial relief (PR), and Relapse—in patients with idiopathic membranous nephropathy (IMN). We specifically evaluated whether incorporating temporal summary features derived from longitudinal laboratory data could enhance predictive performance beyond baseline measurements. Methods We conducted a retrospective cohort study of 336 IMN patients from Guangdong Provincial Hospital of Traditional Chinese Medicine, Guangzhou (2021–2024). Predictors included demographic characteristics, lifestyle factors (smoking, alcohol consumption), and laboratory parameters. For each laboratory variable, we constructed a comprehensive feature set comprising baseline values, mean, maximum, minimum, and change from 0–12 weeks. The dataset was partitioned into training and test sets, and we compared three machine learning approaches: Random Forest (RF), XGBoost, and Multinomial Logistic Regression (MLR), using 5-fold cross-validation for hyperparameter tuning. Results The tree-based ensemble models demonstrated superior predictive capability. XGBoost achieved the highest overall accuracy (0.754), followed closely by random forest. Multinomial Logistic Regression showed moderately lower performance. Critically, models utilizing the temporal summary feature set consistently outperformed those relying solely on baseline data across all algorithms. Conclusions Tree-based ensemble models, particularly XGBoost and Random Forest, effectively predict clinical outcomes in idiopathic membranous nephropathy when incorporating temporal feature engineering from longitudinal laboratory data. XGBoost demonstrated superior performance in relapse prediction (AUC = 0.948), while Random Forest achieved balanced multiclass performance (Macro-AUC = 0.935). These approaches offer promising avenues for risk stratification and personalized treatment planning in IMN management, warranting further validation in multi-center settings. idiopathic membranous nephropathy machine learning clinical outcome prediction random forest XGBoost Figures Figure 1 Figure 2 Figure 3 Introduction Idiopathic membranous nephropathy (IMN) is a common cause of nephrotic syndrome and chronic kidney disease 1 . Clinical outcomes such as Complete Remission (CR), partial relief (PR), and Relapse are critical endpoints in assessing treatment response, often defined by specific proteinuria thresholds, with relapse indicating the reemergence of significant proteinuria 2 . Despite their clinical importance, well-established prognostic models for predicting these remission outcomes in IMN remain limited 3 . The application of machine learning (ML) for predictive modeling in nephrology is an emerging field. Previous studies, such as that by Duo et al. 3 , have developed ML-based nomograms for predicting IMN remission using techniques like LASSO and Cox regression 5 . However, classification among the specific outcomes of CR, PR, and Relapse has not been extensively explored. Furthermore, comparative analyses often indicate that advanced ML methods, such as Random Forest (RF) and XGBoost, may perform similarly to traditional regression models in clinical prediction tasks. For instance, Takkavatakarn et al. found that RF and XGBoost achieved AUC values comparable to LASSO regression in predicting chronic kidney disease progression 6 , and a systematic review highlighted no consistent superiority of ML over logistic regression in various clinical contexts 7 . This underscores the necessity of empirically comparing different modeling approaches for the specific task of IMN outcome prediction. The selection of predictive features is crucial. In many clinical datasets, predictor values evolve over time, and models relying solely on baseline measurements may miss informative temporal trends 8 . To address this, our study incorporates longitudinal information by deriving summary features—including baseline values, mean, maximum, minimum, and change from baseline to 12 weeks—from laboratory measurements, a method analogous to landmark analysis in time-to-event modeling 9 . This approach effectively "compresses" longitudinal data into a static format usable by standard classification algorithms. This study aims to compare the performance of Random Forest, XGBoost, and Multinomial Logistic Regression in predicting CR, PR, and Relapse in IMN patients. The models will utilize both baseline demographic data and compressed longitudinal laboratory features. Performance will be comprehensively assessed using ROC-AUC and standard classification metrics. Methods Data Collection and Processing This retrospective cohort study included patients diagnosed with idiopathic membranous nephropathy (IMN) at Guangdong Provincial Hospital of Traditional Chinese Medicine between January 2021 and December 2024. The diagnosis of IMN and clinical evaluations followed the Kidney Disease: Improving Global Outcomes (KDIGO) 2021 Clinical Practice Guideline for Glomerular Diseases.This guideline provides standardized criteria for diagnosis, treatment, and outcome assessment in glomerular diseases, including IMN 10 . Eligible patients were aged ≥ 18 years and had available baseline demographic information as well as longitudinal biochemical measurements, including 24-hour urinary total protein (24hUTP), serum albumin (ALB), serum creatinine (SCR), urine protein-to-creatinine ratio (UPCR), and anti-phospholipase A2 receptor (PLA2R) antibody status. Laboratory data were collected at baseline and during routine follow-up visits up to 96 weeks. Patients were excluded if baseline clinical data were incomplete or if follow-up duration was insufficient to ascertain clinical outcomes. Patients without adequate follow-up information to determine remission or relapse status were classified as having missing outcome data (NoEvent) and were excluded from outcome-based analyses and model development. After applying these criteria, patients with complete outcome information were included in the final analysis. Outcome Definition Clinical outcomes were defined based on longitudinal proteinuria measurements during follow-up, consistent with established criteria for membranous nephropathy.Complete remission (CR) was defined as 24-hour urinary protein excretion < 300 mg/day with stable renal function. Partial relief (PR) was defined as a ≥ 50% reduction in proteinuria from baseline with a final value between 300 and 3500 mg/day. Relapse was defined as the reappearance of nephrotic-range proteinuria (> 3500 mg/day) after a patient had previously achieved CR or PR. Patients for whom these outcome criteria could not be determined due to insufficient follow-up data were considered to have missing outcome information and were not included in baseline comparisons or predictive modeling. For modeling purposes, each patient was assigned a single mutually exclusive outcome category based on their final clinical status during follow-up: complete remission, partial relief, or relapse. Variable Encoding For machine learning modeling, all predictors were transformed into categorical or binary variables based on established clinical reference thresholds to enhance interpretability and reduce the influence of extreme values. This discretization strategy aligns with routine clinical decision-making and avoids assumptions of linear relationships between biochemical indicators and clinical outcomes.To avoid information leakage, only laboratory measurements obtained within the first 12 weeks after baseline were used for temporal feature engineering. Binary variables—including anti-PLA2R antibody status, smoking status, and alcohol consumption—were encoded as 0/1. Sex was treated as a categorical variable and dummy encoded. Biochemical indicators were categorized according to widely accepted nephrology guidelines. Specifically, 24-hour urinary total protein was categorized into 3500 mg/day, corresponding to normal, moderate, and nephrotic-range proteinuria. UPCR was classified using parallel thresholds. Serum albumin (ALB) was dichotomized at 35 g/L to indicate hypoalbuminemia (0 = normal, 1 = low). Anti-PLA2R antibody status was dichotomized using validated immunoassay criteria. The same engineered feature set was applied across all machine learning models to ensure fair comparison of predictive performance.The inclusion of anti-PLA2R antibody status as a predictor aligns with established evidence demonstrating its association with disease activity and clinical outcomes in IMN 13 . Machine Learning Methods Three predictive models were developed to classify clinical outcomes—complete remission (CR), partial relief (PR), and relapse—including Random Forest, XGBoost, and Multinomial Logistic Regression. Random Forest was implemented using Breiman’s ensemble-learning framework 14 . XGBoost employed a gradient boosting architecture with regularization under the multi:softprob objective to generate multiclass probability estimates, consistent with the methodology of Chen and Guestrin 15 . Multinomial Logistic Regression estimated category-specific log-odds via maximum likelihood. Model training was performed using a stratified 70/30 train-test split to preserve class balance. Hyperparameters were optimized using cross-validation, following established machine-learning development practices 12 . Model Performance Evaluation Model performance on the test set was evaluated using overall accuracy, macro-averaged area under the ROC curve (Macro-AUC), and class-specific AUCs. Sensitivity and specificity were computed for each clinical outcome category. Confusion matrices and ROC curves were generated to visualize prediction errors and comparative model performance, following established diagnostic modeling principles 16 . Model Interpretation and Feature Identification Feature importance was computed for all models to support interpretability. For Random Forest and XGBoost, importance scores were derived from impurity reduction and gain metrics 14 , 15 . Standardized coefficients were examined for Multinomial Logistic Regression. Integrated interpretation across these models enabled identification of key biochemical dynamics and baseline characteristics associated with distinct clinical outcomes in membranous nephropathy. Results Baseline Characteristics The baseline characteristics of the study population stratified by clinical outcome are presented in Table 1 . The study included 336 patients with idiopathic membranous nephropathy. Overall, the mean age was 56.65 years (SD 14.62), and no statistically significant differences in age were observed across the outcome groups (p = 0.193), indicating that age did not vary substantially by clinical outcome. Table 1 Baseline characteristics stratified by clinical outcome n Overall complete remission partial relief relapse p 336 49 53 44 age (mean (SD)) 56.65 (14.62) 52.90 (14.88) 56.57 (13.88) 59.23 (15.90) 0.193 24hUTP (%) < 0.001 3500 mg/day 147 (43.8) 20 (40.8) 52 (98.1) 0 (0.0) ALB = 1 (%) 117 (34.8) 16 (32.7) 2 (3.8) 25 (56.8) < 0.001 SCR = 1 (%) 199 (59.2) 33 (67.3) 31 (58.5) 25 (56.8) 0.66 UPCR (%) < 0.001 <300 mg/g 48 (14.3) 5 (10.2) 2 (3.8) 8 (18.2) sex = 2 (%) 132 (39.3) 18 (36.7) 15 (28.3) 18 (40.9) 0.289 Significant differences were observed in several baseline laboratory parameters among the different outcome groups. Patients who achieved complete remission (CR) were more likely to present with lower baseline proteinuria categories, including normal or moderately elevated 24-hour urinary total protein levels ( 3500 mg/day) at baseline (p < 0.001). A similar distribution pattern was observed for the urine protein-to-creatinine ratio (UPCR), with relapse patients more frequently classified in the highest UPCR category, while CR patients were enriched in the lower categories (p < 0.001). In addition, hypoalbuminemia (ALB < 35 g/L) was more prevalent among patients who experienced relapse compared with those achieving CR or partial relief (p < 0.001), indicating a greater baseline disease severity in the relapse group. Baseline anti-PLA2R antibody positivity was also more common among relapse patients than among patients in the remission groups (p = 0.009), consistent with prior evidence linking PLA2R status to disease activity and prognosis in idiopathic membranous nephropathy. No statistically significant differences were observed among the outcome groups with respect to age (p = 0.193), baseline serum creatinine category (p = 0.663), sex distribution (p = 0.289), smoking status (p = 0.441), or alcohol consumption (p = 0.400). These findings suggest that, within this cohort, baseline laboratory indicators—particularly proteinuria severity, serum albumin status, and anti-PLA2R antibody positivity—play a more prominent role in outcome stratification than demographic or lifestyle factors. Model Performance Three supervised learning models—Random Forest (RF), XGBoost (XGB), and Multinomial Logistic Regression (MLR)—were trained using longitudinally engineered features. Overall performance is summarized in Table 2 . XGBoost achieved the highest accuracy (0.754), suggesting strong discrimination power across the three outcome classes. Random Forest exhibited the highest Macro-AUC (0.935), indicating balanced prediction performance even for the minority class (complete remission). MLR exhibited moderate overall performance but maintained stable behavior across classes.The AUCs for each outcome (CR, PR, relapse) revealed several important patterns.PR consistently achieved the highest AUC (0.898–0.980) across all models, indicating that intermediate outcomes with stable biochemical patterns were easier to classify.XGBoost demonstrated the best relapse prediction (AUC 0.948), suggesting superior ability to capture complex nonlinear trends preceding relapse.RF achieved the strongest performance for PR (AUC 0.980), consistent with its sensitivity results.Overall, the combination of high accuracy, robust AUC, and stable class-wise performance supports the suitability of ensemble tree-based algorithms for predicting clinical trajectories in membranous nephropathy. Table 2 Performance metrics Model Accuracy Macro-AUC AUC_CR AUC_PR AUC_Relapse Random Forest 0.677 0.935 0.882 0.980 0.843 XGBoost 0.754 0.928 0.861 0.974 0.948 Multinomial Logit regression 0.708 0.875 0.831 0.898 0.896 ROC Curves One-vs-rest ROC curves for each model are shown in Fig. 1. All models performed best in identifying PR, reflecting the stability and distinctiveness of biochemical marker trajectories in this group. XGBoost showed the strongest performance for relapse (AUC 0.948), underscoring its suitability for detecting early biochemical signals preceding clinical recurrence.Random Forest achieved the highest AUC for PR (0.980), consistent with its high PR sensitivity in the confusion matrix.MLR demonstrated lower but acceptable AUC values, supporting its role as a transparent baseline model. Taken together, these findings suggest that ensemble-based methods provide superior discrimination capability for complex longitudinal biochemical data in membranous nephropathy. Confusion Matrices of the Three Models Confusion matrices based on the independent test set are presented in Fig. 2 to illustrate the class-wise classification performance of the three models. Overall, all models demonstrated good discrimination for partial relief (PR), with high proportions of correctly classified PR cases, reflecting the relative stability and distinct biochemical patterns of this outcome group. The Random Forest model showed strong sensitivity for PR and relapse, correctly identifying the majority of relapse cases in the test set. However, a small number of complete remission (CR) cases were misclassified as PR, suggesting overlap in early biochemical patterns between these outcome categories. XGBoost exhibited comparable class-wise classification performance to Random Forest in the confusion matrix, with similar numbers of misclassified cases across outcome categories. Although the absolute number of misclassifications for relapse was not markedly lower than that of Random Forest in this test set, XGBoost demonstrated superior discriminative ability for relapse at the probability level, as reflected by its higher AUC. Multinomial Logistic Regression showed more evenly distributed misclassifications across the three outcome categories, with a tendency to misclassify CR cases as PR. This pattern is consistent with the linear decision boundaries of logistic regression, which may be less capable of capturing complex nonlinear relationships among discretized biochemical predictors. Taken together, the confusion matrices highlight differences in hard classification behavior across models, while complementary AUC analyses provide a more comprehensive assessment of their overall discriminative performance. Feature importance Feature importance derived from the Random Forest model is presented in Fig. 3 . The analysis indicates that proteinuria-related indicators, serum albumin status, and anti-PLA2R antibody positivity were among the most influential features contributing to model predictions. Baseline proteinuria category and hypoalbuminemia (ALB < 35 g/L) ranked highest, reflecting the importance of baseline disease severity in outcome classification. In addition, early biochemical measurements contributed substantially to the model, suggesting that short-term changes in proteinuria-related indicators provide complementary information beyond baseline characteristics. Feature importance values reflect the relative contribution of variables to classification performance and do not imply causal relationships. Model Robustness Assessed by Cross-Validation To evaluate the robustness and stability of the model selection process, we performed 5-fold cross-validation on the training set. The results, presented in Table 3 , show the mean and standard deviation of performance across different data splits.The lower accuracy observed on the independent test set compared with cross-validation reflects the stricter evaluation under unseen data and the impact of class imbalance, particularly for the complete remission group. Table 3 5-Fold Cross-Validation Performance Summary Model Mean Accuracy Accuracy SD Mean Macro-AUC Macro-AUC SD Random Forest 0.865 0.023 0.930 0.054 XGBoost 0.817 0.038 0.920 0.054 Multinomial Logit 0.701 0.085 0.867 0.042 Random Forest demonstrated the greatest robustness during cross-validation, evidenced by its high mean accuracy (0.865) and lowest standard deviation (0.023). This indicates that its performance was consistent and less sensitive to variations in the training data. In contrast, while XGBoost's test set performance was strong, its higher standard deviation in CV accuracy (0.038) suggests greater variability across different training subsets. Multinomial Logistic Regression exhibited the lowest mean CV accuracy (0.701) and the highest standard deviation (0.085), confirming its relative instability and lower predictive capability observed in the final test. Discussion This study developed and compared three machine learning models for predicting clinical outcomes in idiopathic membranous nephropathy (IMN), leveraging longitudinal biochemical data through comprehensive feature engineering. Our findings demonstrate that all evaluated models exhibited reasonably strong predictive ability, with distinct strengths emerging across different outcome categories and performance metrics. The tree-based ensemble methods, XGBoost and Random Forest, consistently outperformed the conventional Multinomial Logistic Regression approach. XGBoost achieved the highest overall accuracy (0.754) and demonstrated exceptional discriminative capability for relapse prediction (AUC = 0.948), suggesting its particular effectiveness in capturing complex, nonlinear interactions among dynamic biochemical features. This strong performance in identifying relapse cases holds significant clinical relevance, as early detection of relapse risk could prompt timely intervention and treatment adjustments. Meanwhile, Random Forest, while slightly lower in accuracy, achieved the highest macro-AUC (0.935), indicating more balanced performance across all outcome classes. Its remarkable AUC for partial relief (PR = 0.980) highlights the model's robustness to class imbalance and measurement noise—characteristic advantages of ensemble tree methods that aggregate predictions across multiple decision trees. In contrast, Multinomial Logistic Regression showed comparatively lower discriminative ability (Macro-AUC = 0.875), reflecting its limitations in capturing complex feature interactions inherent in clinical outcomes determination. However, it retained superior interpretability, maintaining value for understanding linear relationships between individual biochemical parameters and clinical status. Analysis of confusion matrices revealed that logistic regression frequently misclassified cases between complete and partial relief categories, suggesting oversimplification of the potentially nonlinear boundaries separating these outcome states. Clinical Implications and Model Selection The complementary strengths of these models inform their potential clinical applications. XGBoost appears preferable when maximizing predictive accuracy is paramount, particularly for identifying high-risk relapse cases. Random Forest offers advantages when balanced performance across all outcome categories is desired, demonstrating resilience to dataset irregularities. Multinomial Logistic Regression remains valuable in settings where model interpretability and transparency are prioritized over marginal gains in predictive power. This spectrum of performance characteristics supports the incorporation of multiple modeling approaches in future clinical decision-support frameworks for membranous nephropathy, allowing flexibility based on specific clinical needs and institutional resources. Accurate early prediction of remission status or relapse risk could significantly impact patient management by guiding treatment intensification or modification. For instance, patients identified as high-risk for relapse might benefit from closer monitoring or adjusted immunosuppression regimens. If validated prospectively, our models could be implemented within electronic health record systems to provide real-time risk stratification, potentially enhancing personalized treatment approaches in IMN. Limitations and Future Directions Several limitations of our study merit consideration. The moderate sample size, single-center design, and retrospective nature constrain the generalizability of our findings. While excluding "NoEvent" patients helped focus our analysis on discriminative outcomes, this selection approach may introduce spectrum bias. Furthermore, we did not incorporate pathological parameters from kidney biopsies or serial anti-PLA2R antibody levels, both established prognostic factors in IMN, which might have enhanced predictive performance. The absence of external validation and potential for overfitting, despite using cross-validation, remain concerns that require addressing before clinical implementation. Future research should focus on integrating additional predictors, including genetic markers, detailed pathological findings, and extended biomarker profiles. Larger, multi-center prospective cohorts would enable more robust model validation and enhance generalizability. Methodologically, exploring advanced time-series modeling approaches, such as joint models for longitudinal and time-to-event data or deep learning architectures specifically designed for temporal patterns, may capture disease dynamics more effectively than our feature engineering approach. Ultimately, integrating these refined models into clinical workflows for prospective validation represents a crucial next step toward translating predictive analytics into improved patient outcomes in membranous nephropathy. In conclusion, our study demonstrates that machine learning models, particularly tree-based ensembles incorporating longitudinal feature engineering, show substantial promise for predicting clinical outcomes in IMN. The complementary strengths of different algorithms provide clinicians with multiple viable pathways toward enhanced risk stratification, supporting the ongoing integration of data-driven approaches into nephrology practice. Conclusion In this study, we developed and validated machine learning models for predicting clinical outcomes in idiopathic membranous nephropathy using engineered features derived from longitudinal laboratory data and demographic characteristics. Our findings demonstrate that tree-based ensemble methods, particularly XGBoost and Random Forest, achieved superior discriminative performance compared to conventional multinomial logistic regression, consistent with prior evidence supporting the advantages of ensemble learning in clinical prediction tasks. The exceptional capability of XGBoost in identifying relapse cases (AUC = 0.948) and the balanced performance of Random Forest across all outcome categories (Macro-AUC = 0.935) highlight their complementary strengths for biomedical classification problems. Critically, our feature engineering approach—incorporating temporal summary statistics and early dynamic changes—proved to be a computationally efficient strategy for leveraging longitudinal data, aligning with previous work demonstrating that early biochemical changes strongly predict disease trajectories in nephrology 11 , 17 . This methodology successfully captured the evolving nature of renal pathology while remaining practically implementable within standard clinical workflows. The robust performance of these models, maintained through cross-validation, supports their potential utility as decision-support tools for risk stratification in membranous nephropathy management, echoing recent studies advocating for machine-learning–based clinical decision support in nephrology 18 . Accurate early identification of patients at high risk of relapse could enable timely treatment adjustments and personalized monitoring protocols. Several limitations warrant consideration, including the single-center design, moderate sample size, and absence of external validation. Future research should focus on multi-center prospective validation, integration of additional prognostic factors such as pathological markers and genetic data, and exploration of more sophisticated temporal modeling approaches, such as deep learning–based dynamic prediction models 19 . Ultimately, the integration of data-driven prediction tools into electronic health record systems represents a promising direction for advancing personalized nephrology care and improving long-term outcomes for patients with membranous nephropathy 20 . Declarations Conflicts of Interest It is declared that the authors have no conflict of interest in this paper’s publication. Ethics Approval and Consent to Participate This retrospective study was approved by the Ethics Committee of Guangdong Provincial Hospital of Traditional Chinese Medicine (Guangzhou, China). The study was conducted in accordance with the principles of the Declaration of Helsinki. As this study used anonymized retrospective clinical data, the requirement for written informed consent was waived by the institutional review board. Funding This research was funded by State Key Laboratory of Dampness Syndrome of Chinese Medicine, The Second Affiliated Hospital of Guangzhou University of Chinese Medicine, grant number SZ2021ZZ09; State Key Laboratory of Dampness Syndrome of Chinese Medicine, The Second Affiliated Hospital of Guangzhou University of Chinese Medicine, grant number SZ2024KF11.The Second Affiliated Hospital of Guangzhou University of Chinese Medicine, grant number YN2023HL03. Author Contribution Conception and design: Yuming Lu, Xiquan Lu, Rongrong Wang. Data collection: Pengjie Sha, Yue Cao, Qi Zuo, Yanhua Tian. Analysis and interpretation of data: Yuming Lu, Cuiying Yu, Pengjie Sha, Sha Jiang, Kun Bao. Writing, review, and/or revision of the manuscript: Yuming Lu, Xiquan Lu, Cuiying Yu, Yue Cao, Xiaofan Hong, Rongrong Wang. Study supervision: Xiaofan Hong, Rongrong Wang. All authors commented on previous versions of the manuscript and read and approved the final version of the manuscript. Data Availability Please contact the corresponding author if you would like to inquiries the original data from this study. References Ronco P, Beck L, Debiec H, Fervenza FC, Hou FF, Jha V, Zhao MH. Membranous nephropathy. Nat Reviews Disease Primers. 2021;7(1):69. https://doi.org/10.1038/s41572-021-00296-1 . Cattran DC, Brenchley PE. Membranous nephropathy: Integrating basic science into improved clinical management. Kidney Int Rep. 2017;2(4):536–48. https://doi.org/10.1016/j.ekir.2017.02.010 . Duo L, Chen L, Zuo Y, Guo J, He M, Zhao H, Kang Y, Tang W. Machine learning model to estimate probability of remission in patients with idiopathic membranous nephropathy. Int Immunopharmacol. 2023;125(Pt A):111126. https://doi.org/10.1016/j.intimp.2023.111126 . Shin DH, Enkhbayar D, Park SY, Gwan JT, Park HY, Lee JA, Yang JW, Kim J, Urtnasan E. Ensemble learning approaches for early prediction of chronic kidney disease based on polysomnographic phenotype analysis. NPJ Digit Med. 2025;8(1):675. https://doi.org/10.1038/s41746-025-02039-6 . Christodoulou E, Ma J, Collins GS, Steyerberg EW, Verbakel JY, Van Calster B. A systematic review shows no performance benefit of machine learning over logistic regression for clinical prediction models. J Clin Epidemiol. 2019;110:12–22. https://doi.org/10.1016/j.jclinepi.2019.02.004 . Takkavatakarn K, Oh W, Cheng E, Nadkarni GN, Chan L. Machine learning models to predict end-stage kidney disease in chronic kidney disease stage 4. BMC Nephrol. 2023;24:376. https://doi.org/10.1186/s12882-023-03369-8 . Steyerberg EW, Vergouwe Y. Towards better clinical prediction models: Seven steps for development and an ABCD for validation. Eur Heart J. 2014;35(29):1925–31. https://doi.org/10.1093/eurheartj/ehu207 . Li L, Luo S, Hu B, Wang J, Chen S, Xu X. Dynamic prediction of renal failure using longitudinal biomarkers in a cohort study of chronic kidney disease. Stat Biosci. 2017;9:357–78. https://doi.org/10.1007/s12561-016-9167-x . Yao Y, Li L, Astor B, Yang W, Greene T. Predicting the risk of a clinical event using longitudinal data: The generalized landmark analysis. BMC Med Res Methodol. 2023;23(1):5. https://doi.org/10.1186/s12874-022-01867-y . Kidney Disease: Improving Global Outcomes (KDIGO) Glomerular Diseases Work Group. KDIGO 2021 clinical practice guideline for the management of glomerular diseases. Kidney Int. 2021;100(1):S1–276. https://doi.org/10.1016/j.kint.2021.05.021 . Tangri N, Stevens LA, Griffith J, Tighiouart H, Wagner M, Weiner DE, et al. A predictive model for progression of chronic kidney disease. JAMA. 2011;305(15):1553–9. https://doi.org/10.1001/jama.2011.451 . Kuhn M, Johnson K. (2013). Applied predictive modeling. Springer. https://doi.org/10.1007/978-1-4614-6849-3 Hofstra JM, Beck LH, Beck DM, Wetzels JFM, Salant DJ. Anti-PLA2R antibodies and clinical outcomes in idiopathic membranous nephropathy. J Am Soc Nephrol. 2011;22(6):1167–73. https://doi.org/10.1681/ASN.2010090967 . Breiman L. Random forests. Mach Learn. 2001;45(1):5–32. https://doi.org/10.1023/A:1010933404324 . Chen T, Guestrin C. (2016). XGBoost: A scalable tree boosting system. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (pp. 785–794). https://doi.org/10.1145/2939672.2939785 Fawcett T. An introduction to ROC analysis. Pattern Recognit Lett. 2006;27(8):861–74. https://doi.org/10.1016/j.patrec.2005.10.010 . van de Logt A, Wetzels JF. The clinical relevance of PLA2R autoantibodies in membranous nephropathy. Nephrol Dialysis Transplantation. 2018;33(2):196–203. https://doi.org/10.1093/ndt/gfx001 . Tomasev N, Glorot X, Rae JW, Zielinski M, Askham H, Saraiva A, Suleyman M. A clinically applicable approach to continuous prediction of future acute kidney injury. Nature. 2019;572(7767):116–9. https://doi.org/10.1038/s41586-019-1390-1 . Rajkomar A, Dean J, Kohane I. Machine learning in medicine. N Engl J Med. 2019;380(14):1347–58. https://doi.org/10.1056/NEJMra1814259 . Sendak M, Gao M, Brajer N, Balu S. Real-world integration of artificial intelligence in healthcare: Challenges and opportunities. BMJ Health Care Inf. 2020;27(3):e100193. https://doi.org/10.1136/bmjhci-2020-100193 . Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Revision Version 1 posted Editorial decision: Revision requested 05 May, 2026 Reviews received at journal 03 May, 2026 Reviews received at journal 01 May, 2026 Reviewers agreed at journal 30 Apr, 2026 Reviewers agreed at journal 26 Apr, 2026 Reviews received at journal 26 Mar, 2026 Reviewers agreed at journal 23 Mar, 2026 Reviewers invited by journal 18 Mar, 2026 Editor invited by journal 16 Mar, 2026 Editor assigned by journal 13 Mar, 2026 Submission checks completed at journal 13 Mar, 2026 First submitted to journal 03 Mar, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9019166","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":608214701,"identity":"c5cfcf55-4895-4445-951e-d263e8f5b55a","order_by":0,"name":"Yuming Lu","email":"","orcid":"","institution":"Guangdong Provincial Hospital of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Yuming","middleName":"","lastName":"Lu","suffix":""},{"id":608214703,"identity":"d11b4525-fcfe-4af0-9a56-664653664d0e","order_by":1,"name":"Xiquan Lu","email":"","orcid":"","institution":"Guangdong Provincial Hospital of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Xiquan","middleName":"","lastName":"Lu","suffix":""},{"id":608214705,"identity":"fbc1a206-fdf4-4a6a-8a60-993fb8ca356a","order_by":2,"name":"Cuiying Yu","email":"","orcid":"","institution":"Guangdong Provincial Hospital of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Cuiying","middleName":"","lastName":"Yu","suffix":""},{"id":608214706,"identity":"bebe452d-0a1e-430a-b508-b833b0b0beb1","order_by":3,"name":"Pengjie Sha","email":"","orcid":"","institution":"Guangdong Provincial Hospital of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Pengjie","middleName":"","lastName":"Sha","suffix":""},{"id":608214710,"identity":"d0447876-dbea-4d0e-9d7b-0e03449d93b4","order_by":4,"name":"Yue Cao","email":"","orcid":"","institution":"Guangdong Provincial Hospital of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Yue","middleName":"","lastName":"Cao","suffix":""},{"id":608214711,"identity":"7ea5daa8-4fe9-4159-b170-aebf45fe6eb6","order_by":5,"name":"Dan Wang","email":"","orcid":"","institution":"Guangdong Provincial Hospital of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Dan","middleName":"","lastName":"Wang","suffix":""},{"id":608214712,"identity":"49d0396b-69f0-4e55-9966-8ff9f12427c8","order_by":6,"name":"Qi Zuo","email":"","orcid":"","institution":"Guangdong Provincial Hospital of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Qi","middleName":"","lastName":"Zuo","suffix":""},{"id":608214714,"identity":"d41642a9-896b-4386-abb6-e289d7e87245","order_by":7,"name":"Yanhua Tian","email":"","orcid":"","institution":"Guangdong Provincial Hospital of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Yanhua","middleName":"","lastName":"Tian","suffix":""},{"id":608214715,"identity":"65cb515d-2304-45ab-9b3d-e1526be4b572","order_by":8,"name":"Sha Jiang","email":"","orcid":"","institution":"Guangdong Provincial Hospital of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Sha","middleName":"","lastName":"Jiang","suffix":""},{"id":608214717,"identity":"f69d57a4-86d4-4756-8011-0033a42d0af7","order_by":9,"name":"Kun Bao","email":"","orcid":"","institution":"Guangdong Provincial Hospital of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Kun","middleName":"","lastName":"Bao","suffix":""},{"id":608214718,"identity":"b646e140-76ac-4069-b57e-f670dc5522b8","order_by":10,"name":"Xiaofan Hong","email":"","orcid":"","institution":"Guangdong Provincial Hospital of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Xiaofan","middleName":"","lastName":"Hong","suffix":""},{"id":608214722,"identity":"9892572a-1397-468a-a569-ac9ca505d0ce","order_by":11,"name":"Rongrong Wang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABA0lEQVRIiWNgGAWjYNCDDzAGD7E6GGeQrIUZrhKfFvP23sOvKyru2M2fkZ382eZXXWL/tAOMD962Mcib49Aic+ZcmuWZM8+SN9zI3Sad28eWOON2ArPh3DYGw50N2LVISOSYGTa2HU42kMjdxpzbw5O4QTqBTZq3jSHB4AABLfIzcjd/tuyRAGlh/01Ai/FDoBY7hhu5G6QZfhiAbWHGq4XnjBljw5nDCQZn3m6T7G1IMJ5xO7FZcs45CcMNuLSw9xh/bKg4bC/fnrv5w48/dbL9s5MPfnhTZiOPyxYgYJMAEokNICZjG5gEsSVwqgcCZlAysYew/+BTOApGwSgYBSMVAAB+tl34Oge07AAAAABJRU5ErkJggg==","orcid":"","institution":"Guangdong Provincial Hospital of Traditional Chinese Medicine","correspondingAuthor":true,"prefix":"","firstName":"Rongrong","middleName":"","lastName":"Wang","suffix":""}],"badges":[],"createdAt":"2026-03-03 10:23:24","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9019166/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9019166/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":105195696,"identity":"ed67d4f5-e2cf-4078-b728-95bd217c1491","added_by":"auto","created_at":"2026-03-23 10:14:01","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":73619,"visible":true,"origin":"","legend":"\u003cp\u003eROC curves for CR, PR, and Relapse (one-vs-rest) for each model.\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9019166/v1/c595d1c388e0a2eb4fec2c0f.jpg"},{"id":105195695,"identity":"28a2d714-ed80-4092-9c4a-1b0ab28706ca","added_by":"auto","created_at":"2026-03-23 10:14:01","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":102593,"visible":true,"origin":"","legend":"\u003cp\u003eConfusion matrices were generated using the independent test set.\u003c/p\u003e","description":"","filename":"2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9019166/v1/78e8267fb7fb5b2f06584709.jpg"},{"id":105195694,"identity":"6eee1d08-5e0b-4930-bcba-c024046e24fc","added_by":"auto","created_at":"2026-03-23 10:14:01","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":77625,"visible":true,"origin":"","legend":"\u003cp\u003eFeature importance of the Random Forest model.\u003c/p\u003e","description":"","filename":"3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9019166/v1/72c3967b97c8604fac5cd2f0.jpg"},{"id":105195698,"identity":"e2842de0-c4d0-49af-bb63-3852f63119ff","added_by":"auto","created_at":"2026-03-23 10:14:06","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":989510,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9019166/v1/f0813e06-7673-42fa-b98a-dc3ef8b65410.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Predicting Clinical Outcomes in Membranous Nephropathy Using Machine Learning","fulltext":[{"header":"Introduction","content":"\u003cp\u003eIdiopathic membranous nephropathy (IMN) is a common cause of nephrotic syndrome and chronic kidney disease \u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. Clinical outcomes such as Complete Remission (CR), partial relief (PR), and Relapse are critical endpoints in assessing treatment response, often defined by specific proteinuria thresholds, with relapse indicating the reemergence of significant proteinuria\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. Despite their clinical importance, well-established prognostic models for predicting these remission outcomes in IMN remain limited \u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThe application of machine learning (ML) for predictive modeling in nephrology is an emerging field. Previous studies, such as that by Duo et al.\u003csup\u003e3\u003c/sup\u003e, have developed ML-based nomograms for predicting IMN remission using techniques like LASSO and Cox regression \u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. However, classification among the specific outcomes of CR, PR, and Relapse has not been extensively explored. Furthermore, comparative analyses often indicate that advanced ML methods, such as Random Forest (RF) and XGBoost, may perform similarly to traditional regression models in clinical prediction tasks. For instance, Takkavatakarn et al. found that RF and XGBoost achieved AUC values comparable to LASSO regression in predicting chronic kidney disease progression \u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e, and a systematic review highlighted no consistent superiority of ML over logistic regression in various clinical contexts \u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. This underscores the necessity of empirically comparing different modeling approaches for the specific task of IMN outcome prediction.\u003c/p\u003e \u003cp\u003eThe selection of predictive features is crucial. In many clinical datasets, predictor values evolve over time, and models relying solely on baseline measurements may miss informative temporal trends \u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. To address this, our study incorporates longitudinal information by deriving summary features\u0026mdash;including baseline values, mean, maximum, minimum, and change from baseline to 12 weeks\u0026mdash;from laboratory measurements, a method analogous to landmark analysis in time-to-event modeling \u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. This approach effectively \"compresses\" longitudinal data into a static format usable by standard classification algorithms.\u003c/p\u003e \u003cp\u003eThis study aims to compare the performance of Random Forest, XGBoost, and Multinomial Logistic Regression in predicting CR, PR, and Relapse in IMN patients. The models will utilize both baseline demographic data and compressed longitudinal laboratory features. Performance will be comprehensively assessed using ROC-AUC and standard classification metrics.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eData Collection and Processing\u003c/h2\u003e \u003cp\u003eThis retrospective cohort study included patients diagnosed with idiopathic membranous nephropathy (IMN) at Guangdong Provincial Hospital of Traditional Chinese Medicine between January 2021 and December 2024. The diagnosis of IMN and clinical evaluations followed the Kidney Disease: Improving Global Outcomes (KDIGO) 2021 Clinical Practice Guideline for Glomerular Diseases.This guideline provides standardized criteria for diagnosis, treatment, and outcome assessment in glomerular diseases, including IMN\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eEligible patients were aged\u0026thinsp;\u0026ge;\u0026thinsp;18 years and had available baseline demographic information as well as longitudinal biochemical measurements, including 24-hour urinary total protein (24hUTP), serum albumin (ALB), serum creatinine (SCR), urine protein-to-creatinine ratio (UPCR), and anti-phospholipase A2 receptor (PLA2R) antibody status. Laboratory data were collected at baseline and during routine follow-up visits up to 96 weeks.\u003c/p\u003e \u003cp\u003ePatients were excluded if baseline clinical data were incomplete or if follow-up duration was insufficient to ascertain clinical outcomes. Patients without adequate follow-up information to determine remission or relapse status were classified as having missing outcome data (NoEvent) and were excluded from outcome-based analyses and model development. After applying these criteria, patients with complete outcome information were included in the final analysis.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eOutcome Definition\u003c/h3\u003e\n\u003cp\u003eClinical outcomes were defined based on longitudinal proteinuria measurements during follow-up, consistent with established criteria for membranous nephropathy.Complete remission (CR) was defined as 24-hour urinary protein excretion\u0026thinsp;\u0026lt;\u0026thinsp;300 mg/day with stable renal function. Partial relief (PR) was defined as a\u0026thinsp;\u0026ge;\u0026thinsp;50% reduction in proteinuria from baseline with a final value between 300 and 3500 mg/day. Relapse was defined as the reappearance of nephrotic-range proteinuria (\u0026gt;\u0026thinsp;3500 mg/day) after a patient had previously achieved CR or PR.\u003c/p\u003e \u003cp\u003ePatients for whom these outcome criteria could not be determined due to insufficient follow-up data were considered to have missing outcome information and were not included in baseline comparisons or predictive modeling. For modeling purposes, each patient was assigned a single mutually exclusive outcome category based on their final clinical status during follow-up: complete remission, partial relief, or relapse.\u003c/p\u003e\n\u003ch3\u003eVariable Encoding\u003c/h3\u003e\n\u003cp\u003eFor machine learning modeling, all predictors were transformed into categorical or binary variables based on established clinical reference thresholds to enhance interpretability and reduce the influence of extreme values. This discretization strategy aligns with routine clinical decision-making and avoids assumptions of linear relationships between biochemical indicators and clinical outcomes.To avoid information leakage, only laboratory measurements obtained within the first 12 weeks after baseline were used for temporal feature engineering.\u003c/p\u003e \u003cp\u003eBinary variables\u0026mdash;including anti-PLA2R antibody status, smoking status, and alcohol consumption\u0026mdash;were encoded as 0/1. Sex was treated as a categorical variable and dummy encoded.\u003c/p\u003e \u003cp\u003e Biochemical indicators were categorized according to widely accepted nephrology guidelines. Specifically, 24-hour urinary total protein was categorized into \u0026lt;\u0026thinsp;300 mg/day, 300\u0026ndash;3500 mg/day, and \u0026gt;\u0026thinsp;3500 mg/day, corresponding to normal, moderate, and nephrotic-range proteinuria. UPCR was classified using parallel thresholds. Serum albumin (ALB) was dichotomized at 35 g/L to indicate hypoalbuminemia (0\u0026thinsp;=\u0026thinsp;normal, 1\u0026thinsp;=\u0026thinsp;low). Anti-PLA2R antibody status was dichotomized using validated immunoassay criteria.\u003c/p\u003e \u003cp\u003eThe same engineered feature set was applied across all machine learning models to ensure fair comparison of predictive performance.The inclusion of anti-PLA2R antibody status as a predictor aligns with established evidence demonstrating its association with disease activity and clinical outcomes in IMN\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\n\u003ch3\u003eMachine Learning Methods\u003c/h3\u003e\n\u003cp\u003eThree predictive models were developed to classify clinical outcomes\u0026mdash;complete remission (CR), partial relief (PR), and relapse\u0026mdash;including Random Forest, XGBoost, and Multinomial Logistic Regression. Random Forest was implemented using Breiman\u0026rsquo;s ensemble-learning framework\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e. XGBoost employed a gradient boosting architecture with regularization under the multi:softprob objective to generate multiclass probability estimates, consistent with the methodology of Chen and Guestrin\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e. Multinomial Logistic Regression estimated category-specific log-odds via maximum likelihood.\u003c/p\u003e \u003cp\u003eModel training was performed using a stratified 70/30 train-test split to preserve class balance. Hyperparameters were optimized using cross-validation, following established machine-learning development practices \u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\n\u003ch3\u003eModel Performance Evaluation\u003c/h3\u003e\n\u003cp\u003eModel performance on the test set was evaluated using overall accuracy, macro-averaged area under the ROC curve (Macro-AUC), and class-specific AUCs. Sensitivity and specificity were computed for each clinical outcome category. Confusion matrices and ROC curves were generated to visualize prediction errors and comparative model performance, following established diagnostic modeling principles \u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eModel Interpretation and Feature Identification\u003c/h2\u003e \u003cp\u003eFeature importance was computed for all models to support interpretability. For Random Forest and XGBoost, importance scores were derived from impurity reduction and gain metrics\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e,\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e. Standardized coefficients were examined for Multinomial Logistic Regression. Integrated interpretation across these models enabled identification of key biochemical dynamics and baseline characteristics associated with distinct clinical outcomes in membranous nephropathy.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eBaseline Characteristics\u003c/h2\u003e \u003cp\u003eThe baseline characteristics of the study population stratified by clinical outcome are presented in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. The study included 336 patients with idiopathic membranous nephropathy. Overall, the mean age was 56.65 years (SD 14.62), and no statistically significant differences in age were observed across the outcome groups (p\u0026thinsp;=\u0026thinsp;0.193), indicating that age did not vary substantially by clinical outcome.\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 stratified by clinical outcome\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003en\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOverall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ecomplete remission\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003epartial relief\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003erelapse\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003ep\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e336\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e49\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e53\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e44\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eage (mean (SD))\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e56.65 (14.62)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e52.90 (14.88)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e56.57 (13.88)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e59.23 (15.90)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.193\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e24hUTP (%)\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 \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\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\u0026lt;300 mg/day\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e35 (10.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1 (1.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e5 (11.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e300\u0026ndash;3500 mg/day\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e154 (45.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e29 (59.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e39 (88.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026gt;3500 mg/day\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e147 (43.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e20 (40.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e52 (98.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eALB\u0026thinsp;=\u0026thinsp;1 (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e117 (34.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e16 (32.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2 (3.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e25 (56.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\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\u003eSCR\u0026thinsp;=\u0026thinsp;1 (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e199 (59.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e33 (67.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e31 (58.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e25 (56.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.66\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUPCR (%)\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 \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\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\u0026lt;300 mg/g\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e48 (14.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e5 (10.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2 (3.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e8 (18.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003esex\u0026thinsp;=\u0026thinsp;2 (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e132 (39.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e18 (36.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e15 (28.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e18 (40.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.289\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eSignificant differences were observed in several baseline laboratory parameters among the different outcome groups. Patients who achieved complete remission (CR) were more likely to present with lower baseline proteinuria categories, including normal or moderately elevated 24-hour urinary total protein levels (\u0026lt;\u0026thinsp;300 or 300\u0026ndash;3500 mg/day), whereas patients who subsequently experienced relapse predominantly exhibited nephrotic-range proteinuria (\u0026gt;\u0026thinsp;3500 mg/day) at baseline (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). A similar distribution pattern was observed for the urine protein-to-creatinine ratio (UPCR), with relapse patients more frequently classified in the highest UPCR category, while CR patients were enriched in the lower categories (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001).\u003c/p\u003e \u003cp\u003eIn addition, hypoalbuminemia (ALB\u0026thinsp;\u0026lt;\u0026thinsp;35 g/L) was more prevalent among patients who experienced relapse compared with those achieving CR or partial relief (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), indicating a greater baseline disease severity in the relapse group. Baseline anti-PLA2R antibody positivity was also more common among relapse patients than among patients in the remission groups (p\u0026thinsp;=\u0026thinsp;0.009), consistent with prior evidence linking PLA2R status to disease activity and prognosis in idiopathic membranous nephropathy.\u003c/p\u003e \u003cp\u003eNo statistically significant differences were observed among the outcome groups with respect to age (p\u0026thinsp;=\u0026thinsp;0.193), baseline serum creatinine category (p\u0026thinsp;=\u0026thinsp;0.663), sex distribution (p\u0026thinsp;=\u0026thinsp;0.289), smoking status (p\u0026thinsp;=\u0026thinsp;0.441), or alcohol consumption (p\u0026thinsp;=\u0026thinsp;0.400). These findings suggest that, within this cohort, baseline laboratory indicators\u0026mdash;particularly proteinuria severity, serum albumin status, and anti-PLA2R antibody positivity\u0026mdash;play a more prominent role in outcome stratification than demographic or lifestyle factors.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eModel Performance\u003c/h2\u003e \u003cp\u003eThree supervised learning models\u0026mdash;Random Forest (RF), XGBoost (XGB), and Multinomial Logistic Regression (MLR)\u0026mdash;were trained using longitudinally engineered features. Overall performance is summarized in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eXGBoost achieved the highest accuracy (0.754), suggesting strong discrimination power across the three outcome classes. Random Forest exhibited the highest Macro-AUC (0.935), indicating balanced prediction performance even for the minority class (complete remission). MLR exhibited moderate overall performance but maintained stable behavior across classes.The AUCs for each outcome (CR, PR, relapse) revealed several important patterns.PR consistently achieved the highest AUC (0.898\u0026ndash;0.980) across all models, indicating that intermediate outcomes with stable biochemical patterns were easier to classify.XGBoost demonstrated the best relapse prediction (AUC 0.948), suggesting superior ability to capture complex nonlinear trends preceding relapse.RF achieved the strongest performance for PR (AUC 0.980), consistent with its sensitivity results.Overall, the combination of high accuracy, robust AUC, and stable class-wise performance supports the suitability of ensemble tree-based algorithms for predicting clinical trajectories in membranous nephropathy.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformance metrics\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMacro-AUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAUC_CR\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAUC_PR\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAUC_Relapse\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.677\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.935\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.882\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.980\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.843\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eXGBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.754\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.928\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.861\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.974\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.948\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMultinomial Logit regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.708\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.875\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.831\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.898\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.896\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=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eROC Curves\u003c/h2\u003e \u003cp\u003eOne-vs-rest ROC curves for each model are shown in Fig.\u0026nbsp;1. All models performed best in identifying PR, reflecting the stability and distinctiveness of biochemical marker trajectories in this group. XGBoost showed the strongest performance for relapse (AUC 0.948), underscoring its suitability for detecting early biochemical signals preceding clinical recurrence.Random Forest achieved the highest AUC for PR (0.980), consistent with its high PR sensitivity in the confusion matrix.MLR demonstrated lower but acceptable AUC values, supporting its role as a transparent baseline model. Taken together, these findings suggest that ensemble-based methods provide superior discrimination capability for complex longitudinal biochemical data in membranous nephropathy.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eConfusion Matrices of the Three Models\u003c/h2\u003e \u003cp\u003eConfusion matrices based on the independent test set are presented in Fig.\u0026nbsp;2 to illustrate the class-wise classification performance of the three models. Overall, all models demonstrated good discrimination for partial relief (PR), with high proportions of correctly classified PR cases, reflecting the relative stability and distinct biochemical patterns of this outcome group.\u003c/p\u003e \u003cp\u003eThe Random Forest model showed strong sensitivity for PR and relapse, correctly identifying the majority of relapse cases in the test set. However, a small number of complete remission (CR) cases were misclassified as PR, suggesting overlap in early biochemical patterns between these outcome categories.\u003c/p\u003e \u003cp\u003eXGBoost exhibited comparable class-wise classification performance to Random Forest in the confusion matrix, with similar numbers of misclassified cases across outcome categories. Although the absolute number of misclassifications for relapse was not markedly lower than that of Random Forest in this test set, XGBoost demonstrated superior discriminative ability for relapse at the probability level, as reflected by its higher AUC.\u003c/p\u003e \u003cp\u003eMultinomial Logistic Regression showed more evenly distributed misclassifications across the three outcome categories, with a tendency to misclassify CR cases as PR. This pattern is consistent with the linear decision boundaries of logistic regression, which may be less capable of capturing complex nonlinear relationships among discretized biochemical predictors.\u003c/p\u003e \u003cp\u003eTaken together, the confusion matrices highlight differences in hard classification behavior across models, while complementary AUC analyses provide a more comprehensive assessment of their overall discriminative performance.\u003c/p\u003e \u003cp\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eFeature importance\u003c/h2\u003e \u003cp\u003eFeature importance derived from the Random Forest model is presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e3\u003c/span\u003e. The analysis indicates that proteinuria-related indicators, serum albumin status, and anti-PLA2R antibody positivity were among the most influential features contributing to model predictions. Baseline proteinuria category and hypoalbuminemia (ALB\u0026thinsp;\u0026lt;\u0026thinsp;35 g/L) ranked highest, reflecting the importance of baseline disease severity in outcome classification.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn addition, early biochemical measurements contributed substantially to the model, suggesting that short-term changes in proteinuria-related indicators provide complementary information beyond baseline characteristics. Feature importance values reflect the relative contribution of variables to classification performance and do not imply causal relationships.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eModel Robustness Assessed by Cross-Validation\u003c/h2\u003e \u003cp\u003eTo evaluate the robustness and stability of the model selection process, we performed 5-fold cross-validation on the training set. The results, presented in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, show the mean and standard deviation of performance across different data splits.The lower accuracy observed on the independent test set compared with cross-validation reflects the stricter evaluation under unseen data and the impact of class imbalance, particularly for the complete remission group.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e5-Fold Cross-Validation Performance Summary\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean Accuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAccuracy SD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMean Macro-AUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eMacro-AUC SD\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.865\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.023\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.930\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.054\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eXGBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.817\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.038\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.920\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.054\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMultinomial Logit\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.701\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.085\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.867\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.042\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eRandom Forest demonstrated the greatest robustness during cross-validation, evidenced by its high mean accuracy (0.865) and lowest standard deviation (0.023). This indicates that its performance was consistent and less sensitive to variations in the training data. In contrast, while XGBoost's test set performance was strong, its higher standard deviation in CV accuracy (0.038) suggests greater variability across different training subsets. Multinomial Logistic Regression exhibited the lowest mean CV accuracy (0.701) and the highest standard deviation (0.085), confirming its relative instability and lower predictive capability observed in the final test.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study developed and compared three machine learning models for predicting clinical outcomes in idiopathic membranous nephropathy (IMN), leveraging longitudinal biochemical data through comprehensive feature engineering. Our findings demonstrate that all evaluated models exhibited reasonably strong predictive ability, with distinct strengths emerging across different outcome categories and performance metrics.\u003c/p\u003e \u003cp\u003eThe tree-based ensemble methods, XGBoost and Random Forest, consistently outperformed the conventional Multinomial Logistic Regression approach. XGBoost achieved the highest overall accuracy (0.754) and demonstrated exceptional discriminative capability for relapse prediction (AUC\u0026thinsp;=\u0026thinsp;0.948), suggesting its particular effectiveness in capturing complex, nonlinear interactions among dynamic biochemical features. This strong performance in identifying relapse cases holds significant clinical relevance, as early detection of relapse risk could prompt timely intervention and treatment adjustments. Meanwhile, Random Forest, while slightly lower in accuracy, achieved the highest macro-AUC (0.935), indicating more balanced performance across all outcome classes. Its remarkable AUC for partial relief (PR\u0026thinsp;=\u0026thinsp;0.980) highlights the model's robustness to class imbalance and measurement noise\u0026mdash;characteristic advantages of ensemble tree methods that aggregate predictions across multiple decision trees.\u003c/p\u003e \u003cp\u003eIn contrast, Multinomial Logistic Regression showed comparatively lower discriminative ability (Macro-AUC\u0026thinsp;=\u0026thinsp;0.875), reflecting its limitations in capturing complex feature interactions inherent in clinical outcomes determination. However, it retained superior interpretability, maintaining value for understanding linear relationships between individual biochemical parameters and clinical status. Analysis of confusion matrices revealed that logistic regression frequently misclassified cases between complete and partial relief categories, suggesting oversimplification of the potentially nonlinear boundaries separating these outcome states.\u003c/p\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eClinical Implications and Model Selection\u003c/h2\u003e \u003cp\u003eThe complementary strengths of these models inform their potential clinical applications. XGBoost appears preferable when maximizing predictive accuracy is paramount, particularly for identifying high-risk relapse cases. Random Forest offers advantages when balanced performance across all outcome categories is desired, demonstrating resilience to dataset irregularities. Multinomial Logistic Regression remains valuable in settings where model interpretability and transparency are prioritized over marginal gains in predictive power. This spectrum of performance characteristics supports the incorporation of multiple modeling approaches in future clinical decision-support frameworks for membranous nephropathy, allowing flexibility based on specific clinical needs and institutional resources.\u003c/p\u003e \u003cp\u003eAccurate early prediction of remission status or relapse risk could significantly impact patient management by guiding treatment intensification or modification. For instance, patients identified as high-risk for relapse might benefit from closer monitoring or adjusted immunosuppression regimens. If validated prospectively, our models could be implemented within electronic health record systems to provide real-time risk stratification, potentially enhancing personalized treatment approaches in IMN.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003eLimitations and Future Directions\u003c/h2\u003e \u003cp\u003eSeveral limitations of our study merit consideration. The moderate sample size, single-center design, and retrospective nature constrain the generalizability of our findings. While excluding \"NoEvent\" patients helped focus our analysis on discriminative outcomes, this selection approach may introduce spectrum bias. Furthermore, we did not incorporate pathological parameters from kidney biopsies or serial anti-PLA2R antibody levels, both established prognostic factors in IMN, which might have enhanced predictive performance. The absence of external validation and potential for overfitting, despite using cross-validation, remain concerns that require addressing before clinical implementation.\u003c/p\u003e \u003cp\u003eFuture research should focus on integrating additional predictors, including genetic markers, detailed pathological findings, and extended biomarker profiles. Larger, multi-center prospective cohorts would enable more robust model validation and enhance generalizability. Methodologically, exploring advanced time-series modeling approaches, such as joint models for longitudinal and time-to-event data or deep learning architectures specifically designed for temporal patterns, may capture disease dynamics more effectively than our feature engineering approach. Ultimately, integrating these refined models into clinical workflows for prospective validation represents a crucial next step toward translating predictive analytics into improved patient outcomes in membranous nephropathy.\u003c/p\u003e \u003cp\u003eIn conclusion, our study demonstrates that machine learning models, particularly tree-based ensembles incorporating longitudinal feature engineering, show substantial promise for predicting clinical outcomes in IMN. The complementary strengths of different algorithms provide clinicians with multiple viable pathways toward enhanced risk stratification, supporting the ongoing integration of data-driven approaches into nephrology practice.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn this study, we developed and validated machine learning models for predicting clinical outcomes in idiopathic membranous nephropathy using engineered features derived from longitudinal laboratory data and demographic characteristics. Our findings demonstrate that tree-based ensemble methods, particularly XGBoost and Random Forest, achieved superior discriminative performance compared to conventional multinomial logistic regression, consistent with prior evidence supporting the advantages of ensemble learning in clinical prediction tasks. The exceptional capability of XGBoost in identifying relapse cases (AUC\u0026thinsp;=\u0026thinsp;0.948) and the balanced performance of Random Forest across all outcome categories (Macro-AUC\u0026thinsp;=\u0026thinsp;0.935) highlight their complementary strengths for biomedical classification problems.\u003c/p\u003e \u003cp\u003eCritically, our feature engineering approach\u0026mdash;incorporating temporal summary statistics and early dynamic changes\u0026mdash;proved to be a computationally efficient strategy for leveraging longitudinal data, aligning with previous work demonstrating that early biochemical changes strongly predict disease trajectories in nephrology \u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e,\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e. This methodology successfully captured the evolving nature of renal pathology while remaining practically implementable within standard clinical workflows.\u003c/p\u003e \u003cp\u003eThe robust performance of these models, maintained through cross-validation, supports their potential utility as decision-support tools for risk stratification in membranous nephropathy management, echoing recent studies advocating for machine-learning\u0026ndash;based clinical decision support in nephrology\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e. Accurate early identification of patients at high risk of relapse could enable timely treatment adjustments and personalized monitoring protocols.\u003c/p\u003e \u003cp\u003eSeveral limitations warrant consideration, including the single-center design, moderate sample size, and absence of external validation. Future research should focus on multi-center prospective validation, integration of additional prognostic factors such as pathological markers and genetic data, and exploration of more sophisticated temporal modeling approaches, such as deep learning\u0026ndash;based dynamic prediction models \u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. Ultimately, the integration of data-driven prediction tools into electronic health record systems represents a promising direction for advancing personalized nephrology care and improving long-term outcomes for patients with membranous nephropathy\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e "},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eConflicts of Interest\u003c/h2\u003e \u003cp\u003eIt is declared that the authors have no conflict of interest in this paper\u0026rsquo;s publication.\u003c/p\u003e \u003c/p\u003e\u003cp\u003e \u003ch2\u003eEthics Approval and Consent to Participate\u003c/h2\u003e \u003cp\u003e This retrospective study was approved by the Ethics Committee of Guangdong Provincial Hospital of Traditional Chinese Medicine (Guangzhou, China). The study was conducted in accordance with the principles of the Declaration of Helsinki. As this study used anonymized retrospective clinical data, the requirement for written informed consent was waived by the institutional review board.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThis research was funded by State Key Laboratory of Dampness Syndrome of Chinese Medicine, The Second Affiliated Hospital of Guangzhou University of Chinese Medicine, grant number SZ2021ZZ09; State Key Laboratory of Dampness Syndrome of Chinese Medicine, The Second Affiliated Hospital of Guangzhou University of Chinese Medicine, grant number SZ2024KF11.The Second Affiliated Hospital of Guangzhou University of Chinese Medicine, grant number YN2023HL03.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eConception and design: Yuming Lu, Xiquan Lu, Rongrong Wang. Data collection: Pengjie Sha, Yue Cao, Qi Zuo, Yanhua Tian. Analysis and interpretation of data: Yuming Lu, Cuiying Yu, Pengjie Sha, Sha Jiang, Kun Bao. Writing, review, and/or revision of the manuscript: Yuming Lu, Xiquan Lu, Cuiying Yu, Yue Cao, Xiaofan Hong, Rongrong Wang. Study supervision: Xiaofan Hong, Rongrong Wang. All authors commented on previous versions of the manuscript and read and approved the final version of the manuscript.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003ePlease contact the corresponding author if you would like to inquiries the original data from this study.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eRonco P, Beck L, Debiec H, Fervenza FC, Hou FF, Jha V, Zhao MH. Membranous nephropathy. Nat Reviews Disease Primers. 2021;7(1):69. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41572-021-00296-1\u003c/span\u003e\u003cspan address=\"10.1038/s41572-021-00296-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCattran DC, Brenchley PE. Membranous nephropathy: Integrating basic science into improved clinical management. Kidney Int Rep. 2017;2(4):536\u0026ndash;48. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ekir.2017.02.010\u003c/span\u003e\u003cspan address=\"10.1016/j.ekir.2017.02.010\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDuo L, Chen L, Zuo Y, Guo J, He M, Zhao H, Kang Y, Tang W. Machine learning model to estimate probability of remission in patients with idiopathic membranous nephropathy. Int Immunopharmacol. 2023;125(Pt A):111126. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.intimp.2023.111126\u003c/span\u003e\u003cspan address=\"10.1016/j.intimp.2023.111126\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShin DH, Enkhbayar D, Park SY, Gwan JT, Park HY, Lee JA, Yang JW, Kim J, Urtnasan E. Ensemble learning approaches for early prediction of chronic kidney disease based on polysomnographic phenotype analysis. NPJ Digit Med. 2025;8(1):675. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41746-025-02039-6\u003c/span\u003e\u003cspan address=\"10.1038/s41746-025-02039-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChristodoulou E, Ma J, Collins GS, Steyerberg EW, Verbakel JY, Van Calster B. A systematic review shows no performance benefit of machine learning over logistic regression for clinical prediction models. J Clin Epidemiol. 2019;110:12\u0026ndash;22. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jclinepi.2019.02.004\u003c/span\u003e\u003cspan address=\"10.1016/j.jclinepi.2019.02.004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTakkavatakarn K, Oh W, Cheng E, Nadkarni GN, Chan L. Machine learning models to predict end-stage kidney disease in chronic kidney disease stage 4. BMC Nephrol. 2023;24:376. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12882-023-03369-8\u003c/span\u003e\u003cspan address=\"10.1186/s12882-023-03369-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSteyerberg EW, Vergouwe Y. Towards better clinical prediction models: Seven steps for development and an ABCD for validation. Eur Heart J. 2014;35(29):1925\u0026ndash;31. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/eurheartj/ehu207\u003c/span\u003e\u003cspan address=\"10.1093/eurheartj/ehu207\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi L, Luo S, Hu B, Wang J, Chen S, Xu X. Dynamic prediction of renal failure using longitudinal biomarkers in a cohort study of chronic kidney disease. Stat Biosci. 2017;9:357\u0026ndash;78. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s12561-016-9167-x\u003c/span\u003e\u003cspan address=\"10.1007/s12561-016-9167-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYao Y, Li L, Astor B, Yang W, Greene T. Predicting the risk of a clinical event using longitudinal data: The generalized landmark analysis. BMC Med Res Methodol. 2023;23(1):5. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12874-022-01867-y\u003c/span\u003e\u003cspan address=\"10.1186/s12874-022-01867-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKidney Disease: Improving Global Outcomes (KDIGO) Glomerular Diseases Work Group. KDIGO 2021 clinical practice guideline for the management of glomerular diseases. Kidney Int. 2021;100(1):S1\u0026ndash;276. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.kint.2021.05.021\u003c/span\u003e\u003cspan address=\"10.1016/j.kint.2021.05.021\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTangri N, Stevens LA, Griffith J, Tighiouart H, Wagner M, Weiner DE, et al. A predictive model for progression of chronic kidney disease. JAMA. 2011;305(15):1553\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1001/jama.2011.451\u003c/span\u003e\u003cspan address=\"10.1001/jama.2011.451\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKuhn M, Johnson K. (2013). Applied predictive modeling. Springer. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/978-1-4614-6849-3\u003c/span\u003e\u003cspan address=\"10.1007/978-1-4614-6849-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHofstra JM, Beck LH, Beck DM, Wetzels JFM, Salant DJ. Anti-PLA2R antibodies and clinical outcomes in idiopathic membranous nephropathy. J Am Soc Nephrol. 2011;22(6):1167\u0026ndash;73. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1681/ASN.2010090967\u003c/span\u003e\u003cspan address=\"10.1681/ASN.2010090967\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBreiman L. Random forests. Mach Learn. 2001;45(1):5\u0026ndash;32. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1023/A:1010933404324\u003c/span\u003e\u003cspan address=\"10.1023/A:1010933404324\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen T, Guestrin C. (2016). 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). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1145/2939672.2939785\u003c/span\u003e\u003cspan address=\"10.1145/2939672.2939785\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFawcett T. An introduction to ROC analysis. Pattern Recognit Lett. 2006;27(8):861\u0026ndash;74. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.patrec.2005.10.010\u003c/span\u003e\u003cspan address=\"10.1016/j.patrec.2005.10.010\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evan de Logt A, Wetzels JF. The clinical relevance of PLA2R autoantibodies in membranous nephropathy. Nephrol Dialysis Transplantation. 2018;33(2):196\u0026ndash;203. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/ndt/gfx001\u003c/span\u003e\u003cspan address=\"10.1093/ndt/gfx001\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTomasev N, Glorot X, Rae JW, Zielinski M, Askham H, Saraiva A, Suleyman M. A clinically applicable approach to continuous prediction of future acute kidney injury. Nature. 2019;572(7767):116\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41586-019-1390-1\u003c/span\u003e\u003cspan address=\"10.1038/s41586-019-1390-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRajkomar A, Dean J, Kohane I. Machine learning in medicine. N Engl J Med. 2019;380(14):1347\u0026ndash;58. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1056/NEJMra1814259\u003c/span\u003e\u003cspan address=\"10.1056/NEJMra1814259\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSendak M, Gao M, Brajer N, Balu S. Real-world integration of artificial intelligence in healthcare: Challenges and opportunities. BMJ Health Care Inf. 2020;27(3):e100193. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/bmjhci-2020-100193\u003c/span\u003e\u003cspan address=\"10.1136/bmjhci-2020-100193\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-nephrology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bnep","sideBox":"Learn more about [BMC Nephrology](http://bmcnephrol.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bnep/default.aspx","title":"BMC Nephrology","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"idiopathic membranous nephropathy, machine learning, clinical outcome prediction, random forest, XGBoost","lastPublishedDoi":"10.21203/rs.3.rs-9019166/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9019166/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eObjective\u003c/h2\u003e \u003cp\u003eThis study aimed to develop and compare machine learning models for predicting clinical outcomes\u0026mdash;Complete Remission (CR), partial relief (PR), and Relapse\u0026mdash;in patients with idiopathic membranous nephropathy (IMN). We specifically evaluated whether incorporating temporal summary features derived from longitudinal laboratory data could enhance predictive performance beyond baseline measurements.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eWe conducted a retrospective cohort study of 336 IMN patients from Guangdong Provincial Hospital of Traditional Chinese Medicine, Guangzhou (2021\u0026ndash;2024). Predictors included demographic characteristics, lifestyle factors (smoking, alcohol consumption), and laboratory parameters. For each laboratory variable, we constructed a comprehensive feature set comprising baseline values, mean, maximum, minimum, and change from 0\u0026ndash;12 weeks. The dataset was partitioned into training and test sets, and we compared three machine learning approaches: Random Forest (RF), XGBoost, and Multinomial Logistic Regression (MLR), using 5-fold cross-validation for hyperparameter tuning.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eThe tree-based ensemble models demonstrated superior predictive capability. XGBoost achieved the highest overall accuracy (0.754), followed closely by random forest. Multinomial Logistic Regression showed moderately lower performance. Critically, models utilizing the temporal summary feature set consistently outperformed those relying solely on baseline data across all algorithms.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eTree-based ensemble models, particularly XGBoost and Random Forest, effectively predict clinical outcomes in idiopathic membranous nephropathy when incorporating temporal feature engineering from longitudinal laboratory data. XGBoost demonstrated superior performance in relapse prediction (AUC\u0026thinsp;=\u0026thinsp;0.948), while Random Forest achieved balanced multiclass performance (Macro-AUC\u0026thinsp;=\u0026thinsp;0.935). These approaches offer promising avenues for risk stratification and personalized treatment planning in IMN management, warranting further validation in multi-center settings.\u003c/p\u003e","manuscriptTitle":"Predicting Clinical Outcomes in Membranous Nephropathy Using Machine Learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-23 10:12:50","doi":"10.21203/rs.3.rs-9019166/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-05-05T16:30:58+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-03T17:39:14+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-01T13:47:51+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"225339725542994555102818839248459951326","date":"2026-04-30T09:08:54+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"158567241890661640090385845540544574472","date":"2026-04-27T00:07:35+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-27T01:36:48+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"241385034800075898373146120607417775268","date":"2026-03-23T15:00:46+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-03-18T11:07:56+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-03-16T08:57:17+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-13T06:49:49+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-13T06:49:35+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Nephrology","date":"2026-03-03T10:09:07+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-nephrology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bnep","sideBox":"Learn more about [BMC Nephrology](http://bmcnephrol.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bnep/default.aspx","title":"BMC Nephrology","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"0c5c458e-aa14-4b78-8cf7-af664689efc7","owner":[],"postedDate":"March 23rd, 2026","published":true,"recentEditorialEvents":[{"type":"decision","content":"Revision requested","date":"2026-05-05T16:30:58+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-03T17:39:14+00:00","index":108,"fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-01T13:47:51+00:00","index":107,"fulltext":""},{"type":"reviewerAgreed","content":"225339725542994555102818839248459951326","date":"2026-04-30T09:08:54+00:00","index":106,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"in-revision","subjectAreas":[],"tags":[],"updatedAt":"2026-05-05T16:39:03+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-23 10:12:50","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9019166","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9019166","identity":"rs-9019166","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}