文献速递糖尿病肥胖与代谢综合征

中国香港糖尿病前期华人患者心血管、肾脏及代谢事件10年预测模型的开发与验证

Development and Validation of 10‐Year Prediction Models for Cardiovascular, Kidney and Metabolic Events Among Chinese Patients With Prediabetes in Hong Kong
RSS 订阅 · Diabetes, Obesity and Metabolism · 2026-11-10

本研究基于香港129 834名成人糖尿病前期患者,构建并验证了7类主要临床事件及死亡的10年预测模型,涵盖CVD、高血压、T2DM、CKD、CVD亚型及全因死亡。Cox模型区分度中等至良好,C指数0.673–0.830,校准良好且净获益为正,机器学习模型表现相当。

要点

  • 基于香港糖尿病前期华人队列开发10年心血管、肾脏及代谢事件预测模型。
  • Cox模型C指数:CVD 0.702、高血压0.697、T2DM 0.704、CKD 0.790、死亡0.816。
  • HbA1c与FPG版本模型表现相当,机器学习模型区分度与Cox模型相当。

核心结论

  • 研究开发了可解释的预测模型,用于香港华人糖尿病前期患者的临床可干预事件,以辅助临床策略和资源分配。

研究设计与样本

  • 基于人群的回顾性队列研究,纳入2012年129 834名成人糖尿病前期患者,随访至2022年。
  • 采用后向逐步Cox回归开发一个基于诊室非实验室模型和三个扩展模型(含实验室结果、变换、交互项和药物)。
  • 使用四种生存机器学习模型进行比较,每个结局分别开发HbA1c和FPG两个版本模型。

主要发现

  • 中位随访10.4年期间,19.4%的患者发生CVD。
  • 关键预测因子包括年龄、性别、吸烟状态、Charlson合并症指数、BMI、收缩压、血脂谱、eGFR和药物。
  • Cox模型区分度中等至良好:CVD C指数0.702、高血压0.697、T2DM 0.704、CKD 0.790、死亡0.816、CVD亚型范围0.673–0.830,校准良好且净获益为正。
  • 四种选定的机器学习方法表现与Cox模型相当。
  • 基于HbA1c和基于FPG的模型表现相当。

局限性

  • 原文未报告局限性。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

原文

ABSTRACT Aims Risk stratification can support frontline doctors and patients by enabling care and empowering early prevention of cardiovascular, kidney and metabolic disease. We aimed to develop and validate 10-year prediction models for seven major clinical events and mortality among Chinese patients with prediabetes in Hong Kong. Methods This population-based, retrospective cohort study identified 129 834 adults with prediabetes in 2012 and followed them through 2022. Backward stepwise Cox regression was used to develop an office-based non-laboratory model and three extended models with laboratory results, transformations, interactions and medications. Outcomes included cardiovascular disease (CVD), hypertension, Type 2 diabetes mellitus (T2DM), chronic kidney disease (CKD), CVD subtypes and all-cause mortality. Four survival machine learning (ML) models were used for comparison. For each outcome, two models for haemoglobin A1c (HbA1c) and fasting plasma glucose (FPG) were developed. Model performance was assessed by discrimination, calibration and decision curve analysis. Results During a median follow-up of 10.4 years, 19.4% of patients developed CVD. Key predictors included age, sex, smoking status, Charlson comorbidity index, body mass index, systolic blood pressure, lipid profiles, estimated glomerular filtration rate and medications. Our Cox models demonstrated moderate-to-good discrimination: CVD (C-index 0.702), hypertension (0.697), T2DM (0.704), CKD (0.790), mortality (0.816) and CVD subtypes (range: 0.673–0.830), with adequate calibration and positive net benefits. Four selected ML approaches showed comparable performance. HbA1c- and FPG-based models showed comparable performance. Conclusions We developed interpretable prediction models for clinically actionable events among Chinese patients with prediabetes in Hong Kong to inform clinical strategies and resource allocation.

来源:Diabetes, Obesity and Metabolism,查看原文。版权归原作者所有。

引用

  1. Diabetes, Obesity and Metabolism:Development and Validation of 10‐Year Prediction Models for Cardiovascular, Kidney and Metabolic Events Among Chinese Patients With Prediabetes in Hong Kong

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。