使用结合UHR的机器学习模型预测心力衰竭患者的短期复合结局风险：一项回顾性队列研究

摘要(中文译文)

尿酸与高密度脂蛋白胆固醇比值（UHR）是一种反映氧化应激与抗氧化失衡的复合生物标志物。既往研究已将UHR与心力衰竭（HF）不良预后相关联，但证据主要依赖传统回归模型，缺乏独立验证，且尚未整合到现代个体化风险预测工具中。本研究旨在相对较大的心力衰竭队列中独立验证UHR与180天复合结局（全因死亡或心力衰竭再住院）之间的关联，并构建一个纳入UHR及常规临床指标的可解释机器学习模型。本研究为单中心回顾性队列研究，连续纳入2024年1月1日至2025年6月30日于首都医科大学附属北京安贞医院住院的2,737例心力衰竭患者。所有患者随访至2025年12月31日，以确保每例患者均有完整的180天随访数据。主要结局为180天内全因死亡或心力衰竭再住院的复合终点。采用多因素Cox回归和限制性立方样条（RCS）分析UHR与复合结局之间的关联及剂量-反应关系。采用最小绝对收缩和选择算子（LASSO）回归结合多因素Logistic回归筛选预测变量。在训练集（=1,917）中构建七种机器学习模型，并在内部验证集（=820）中进行验证。采用SHapley加法解释（SHAP）框架对最优模型进行解释，并将其部署为在线计算器。随访期间，926例患者（33.8%）发生复合结局。UHR被证实为复合结局的独立危险因素。RCS分析显示存在非线性关联。基于体重指数（BMI）、高脂血症病史、同型半胱氨酸（Hcy）、糖化白蛋白（GA）、小密低密度脂蛋白胆固醇（sdLDL-C）和UHR六个变量构建的极端梯度提升（XGBoost）模型表现最佳。在内部验证集中，其AUC为0.866（95%CI，0.839-0.890），校准良好，决策曲线分析显示具有较高的临床净获益。SHAP分析确定sdLDL-C和UHR为两个最重要的预测特征，并揭示了变量的非线性效应。该模型已部署为在线计算器（https://machine11.shinyapps.io/xgboostforUHR/），目前作为研究演示平台，在临床应用前尚需外部验证。UHR是心力衰竭患者180天复合结局的独立危险因素。基于UHR及其他五项常规临床指标构建的XGBoost模型具有良好的区分度、校准度和临床净获益。该在线计算器为短期风险分层提供了探索性工具，但在临床使用前仍需进一步外部验证。

Abstract

The uric acid-to-high-density lipoprotein cholesterol ratio (UHR) is a composite biomarker that reflects both oxidative stress and antioxidant imbalance. Previous studies have linked UHR to poor prognosis in heart failure (HF), but the evidence relies largely on traditional regression models; it lacks independent validation and has not been integrated into modern, individualized risk-prediction tools. This study aimed to independently validate the association between UHR and a 180-day composite outcome (all-cause death or HF rehospitalization) in a relatively large HF cohort, and to construct an interpretable machine learning model that incorporates UHR and routine clinical indicators. This was a single-center retrospective cohort study that consecutively enrolled 2,737 patients with heart failure hospitalized at Beijing Anzhen Hospital, Capital Medical University, from January 1, 2024, to June 30, 2025. All patients were followed up until December 31, 2025, to ensure complete 180-day follow-up data for each patient. The primary outcome was a composite endpoint of all-cause death or heart failure rehospitalization within 180 days. Multivariate Cox regression and restricted cubic splines (RCS) were used to analyze the association and dose-response relationship between UHR and the composite outcome. Least Absolute Shrinkage and Selection Operator (LASSO) regression combined with multivariate Logistic regression was employed to screen predictive variables. Seven machine learning models were developed in the training set ( = 1,917) and validated in an internal validation set ( = 820). The optimal model was interpreted using the SHapley Additive exPlanations (SHAP) framework and deployed as an online calculator. During follow-up, 926 patients (33.8%) experienced the composite outcome. UHR was confirmed as an independent risk factor for the composite outcome. RCS analysis revealed a nonlinear association. The eXtreme Gradient Boosting (XGBoost) model, constructed based on six variables: body mass index (BMI), history of hyperlipidemia, homocysteine (Hcy), glycated albumin (GA), small dense low-density lipoprotein cholesterol (sdLDL-C) and UHR, demonstrated the best performance. In the internal validation set, it achieved an AUC of 0.866 (95%CI, 0.839-0.890), with good calibration and high clinical net benefit in decision curve analysis. SHAP analysis identified sdLDL-C and UHR as the two most important predictive features and revealed nonlinear effects of the variables. The model was deployed as an online calculator (https://machine11.shinyapps.io/xgboostforUHR/), which currently serves as a research demonstration platform and requires external validation before any clinical application. UHR is an independent risk factor for the 180-day composite outcome in HF patients. The XGBoost model based on UHR and five other routine clinical indicators shows good discrimination, calibration, and clinical net benefit. The online calculator provides an exploratory tool for short-term risk stratification but requires further external validation before clinical use.

如何引用

Frontiers in nutrition 的其他文献

• 阿斯巴甜可能通过DPP4介导的内皮功能障碍和细胞凋亡促进勃起功能障碍：来自网络毒理学、分子动力学模拟和实验验证的证据
• 不同饮料消费与炎症性肠病风险之间的关联：剂量反应荟萃分析
• 口服左旋肉碱补充降低血清尿酸并与高尿酸血症小鼠的肾脏和肠道尿酸盐处理反应相关
• 心脏代谢指标对成人糖尿病前期进展为糖尿病和血糖恢复正常的预测性能：一项队列研究
• 白藜芦醇对绝经后慢性冠心病妇女内皮祖细胞和凋亡生物标志物的影响：一项随机对照试验
• 5:2饮食通过激活GHSR诱导小鼠雄性特异性体重减轻和非性别依赖性脂质储存增强
• 多囊卵巢综合征二十年的营养研究：新兴模式和科学产出（2004-2024）
• 美国成年人红细胞分布宽度与白蛋白比值与幽门螺杆菌血清阳性之间的关联：1999-2000年NHANES的一项观察性研究，并经外部验证
• 肥胖对男性不育影响的全球研究趋势：文献计量学分析
• 饮食多糖在炎症性肠病管理中的应用：最新进展