基于机器学习的老年原发性膜性肾病预后预测模型构建OA

Objective Elderly patients with primary membranous nephropathy(PMN)exhibit significant prognostic heterogeneity and poor tolerance to immunotherapy.However,there is a lack of early prog-nostic prediction tools specifically for this population.This study aimed to develop a prognostic prediction model applicable to elderly PMN patients.Methods This study retrospectively included elderly patients with PMN con-firmed by renal biopsy.The primary endpoint was a adverse composite outcome including end-stage renal disease(ESRD),a≥50%decline in estimated glomerular filtration rate(eGFR),or all-cause death.Patients were randomly divided into a training cohort and a validation cohort at a ratio of 7∶3.Key prognostic features were i-dentified using least absolute shrinkage and selection operator(LASSO)regression combined with random survival forest,and a predictive model was constructed based on penalized Cox regression.Model performance was evaluated using the concordance index(C-index),time-dependent area under the receiver operating characteristic curve(AUROC),calibration curves,and decision curve analysis.The SurvSHAP(t)method was employed for interpretability analysis of the model.Results A total of 309 elderly patients with PMN were included in this study,with a median age of 65.00 years(IQR,62.00-68.00)and a male predominance 61.2%(189/309).During a median follow-up of 47.00 months(IQR,25.00-89.00),38.2%(118/309)reached the endpoint event.The final model included nine key features,including eGFR,total protein(TP),glomerular capsu-lar adhesion,urine glucose,segmental glomerulosclerosis proportion,fibrinogen,urea,age,and activated partial thromboplastin time(APTT).In the validation cohort,the model demonstrated good discrimination,with a C-index of 0.731(95%CI:0.652-0.797).The time-dependent AUROCs for predicting adverse outcomes at 3,5,and 10 years were 0.758(95%CI:0.614-0.901),0.781(95%CI:0.646-0.916),and 0.866(95%CI:0.740-0.993),respectively.Calibration curves demonstrated a high degree of concordance between predicted probabilities and actual event rates.Decision curve analysis confirmed the net clinical benefit of the model.SurvSHAP(t)analy-sis showed that eGFR,TP,glomerular capsular adhesion,urine glucose,and the proportion of segmental glomerular sclerosis were the top five variables contributing to the model.Conclusions This prognostic model effectively pre-dicts the risk of adverse outcomes in elderly patients with PMN in the internal validation cohort,offering a potential scientific basis for individualized risk stratification and treatment decision-making in this population.