Construction of a risk prediction model for weakness in maintenance hemodialysis patients
维持性血液透析患者衰弱风险预测模型的构建

摘要

目的 建立维持性血液透析（Maintenance Hemodialysis, MHD）患者衰弱风险预测模型，为早期识别高危人群提供评价工具。方法 选取2022年12月至2023年9月在川北医学院附属医院行MHD的350例患者，用Fried衰弱量表等调查，分为建模组（245例）和验证组（105例）。利用Logistic回归分析筛选出独立影响因素，建立列线图模型，用受试者工作特征曲线（Receiver Operating Characteristic Curve , ROC）、校准曲线评价模型效能。结果 MHD患者衰弱发生率为33.1%。年龄（OR=1.072）、血清白蛋白（OR=0.861）、Charlson合并症指数（OR=1.483）、抑郁（OR=1.151）、营养不良-炎症评分（OR=1.309）是衰弱的独立影响因素（P<0.05）。建模组药时曲线下面积(Area Under the Curve, AUC)为0.887，灵敏度为82.6%，特异度为81.3%；验证组AUC为0.872，灵敏度为80.0%，特异度为79.6%。校准度好（P>0.05）。结论 该预测模型区分度高、校准度好，可以作为临床识别MHD患者衰弱高危人群的辅助工具。

Abstract

Objective To establish a frailty risk prediction model for patients undergoing maintenance hemodialysis (MHD) and provide an evaluation tool for early identification of high-risk populations.
Methods A total of 350 patients receiving MHD at the Affiliated Hospital of North Sichuan Medical University from December 2022 to September 2023 were enrolled. Using the Fried Frailty Scale and other assessments, patients were divided into a modeling group (n=245) and a validation group (n=105). Logistic regression analysis was employed to identify independent influencing factors, and a nomogram model was constructed. Model performance was evaluated using ROC curves and calibration curves.
Results The frailty incidence rate among MHD patients was 33.1%. Age (OR=1.072), serum albumin (OR=0.861), Charlson Comorbidity Index (OR=1.483), depression (OR=1.151), and malnutrition-inflammation score (OR=1.309) were identified as independent predictors of frailty (P<0.05). The modeling group achieved an AUC of 0.887, with sensitivity of 82.6% and specificity of 81.3%; the validation group showed an AUC of 0.872, sensitivity of 80.0%, and specificity of 79.6%. Calibration results were favorable (P>0.05).
Conclusion   This prediction model demonstrates high discriminative power and calibration accuracy, serving as an auxiliary tool for clinical identification of high-risk MHD patients for frailty.