关键词: 高血压肾病, Cox回归, 随机生存森林, 预测模型

Abstract: Objective To develop a survival prediction model for hypertensive nephropathy patients based on Random Survival Forests (RSF) and Cox regression. Methods Data were collected from MIMIC-IV V3.0, with in-hospital mortality and post-treatment survival time as outcome variables. Predictor variables included demographics, laboratory indicators, vital signs, comorbidities, scoring metrics, acute kidney injury (AKI), and continuous renal replacement therapy (CRRT). Patients were randomly divided into training (70%) and validation (30%) sets. Important variables were extracted from the training set to construct Cox regression and RSF models. Model accuracy and discrimination were evaluated using concordance index, time-dependent ROC curves (tROC), and AUC in both training and validation sets. Results A total of 7 369 hypertensive nephropathy patients were included, with 3 543 deaths (48.1%). In the training set, the Cox model achieved AUC values of 0.783 6, 0.770 7, and 0.760 7 at 1, 5, and 10 years, respectively, with a concordance index of 0.719 0 (95% CI: 0.709 2-0.728 8). The RSF model achieved AUC values of 0.794 7, 0.783 0, and 0.769 5 at 1, 5, and 10 years, respectively, with a concordance index of 0.720 0 (95% CI: 0.710 2-0.729 8). In the validation set, the Cox model achieved AUC values of 0.794 7, 0.783 0, and 0.769 5 at 1, 5, and 10 years, respectively, with a concordance index of 0.731 0 (95% CI: 0.717 3-0.744 7). The RSF model achieved similar AUC values and concordance index as in the training set. Conclusion The survival prediction model for hypertensive nephropathy patients based on RSF and Cox regression demonstrates good predictive performance and can serve as a reference for clinical decision-making.

Key words: Hypertensive nephropathy, Cox risk proportional regression, Random survival forest, Prediction model