Development and validation of a predictive model for the progression of diabetic kidney disease to kidney failure.
Development and validation of a predictive model for the progression of diabetic kidney disease to kidney failure.
复制标题
糖尿病肾病进展为肾衰竭的预测模型的开发和验证
DOI：
10.1080/0886022x.2020.1772294
复制
发表时间：
2020-11
期刊：
影响因子：
3
通讯作者：
Zhao Z
中科院分区：
文献类型：
--
作者：
Cheng Y;Shang J;Liu D;Xiao J;Zhao Z
摘要导言一个好的预测模型对糖尿病肾病的进展起着重要的作用。我们的目标是建立一种模型来预测糖尿病肾病患者进展为肾衰竭的情况。方法以641例2型糖尿病肾病患者为衍生队列，280例患者为体外超时验证队列。我们使用临床指导和单变量Logistic回归相结合的方法来选择相关变量。我们计算了不同模型的识别率和校准率。根据判别和校正的最优组合选择最优模型。结果在3个 年的随访中，衍生队列中有272个结果(42%)，外部验证队列中有138个结果(49%)。多因素Logistic回归分析的最终变量为年龄、性别、血红蛋白、NLR、血清胱抑素C、表皮生长因子受体、24小时尿蛋白定量和口服降糖药的使用。根据这些独立的危险因素，我们开发了临床模型、实验室模型、实验室用药模型和全模型。在所有模型中，实验室模型在区分和校正方面表现良好(C-统计量：外部验证0.863；霍斯默-莱姆斯豪斯模型的p值，.817)。实验室模型、实验室-药物模型和全模型之间的NRI无显著差异(p >； .05)。因此，我们选择实验室模型作为最优模型。结论我们构建了一个包含Hb、NLR、血清胱抑素C、表皮生长因子受体和24 h尿蛋白的诺模图来预测3 年内糖尿病肾病患者行肾脏置换手术的风险。
Abstract Introduction A good prediction model plays an important role in determining the progression to diabetic kidney disease. We aimed to create a model to predict progression to kidney failure in patients with diabetic kidney disease. Methods We retrospectively assessed 641 patients with type 2 diabetic kidney disease as derivation cohort and 280 patients as external out time validation cohort. We used a combination of clinical guidance and univariate logistic regression to select the relevant variables. We calculated the discrimination and calibration of different models. The best model was selected according to the optimal combination of discrimination and calibration. Results During the 3 years follow up, there were 272 outcomes (42%) in derivation cohort and 138 outcomes (49%) in external validation cohort. The final variables selected in the multivariate logistics regression were age, gender, hemoglobin, NLR, serum cystatin C, eGFR, 24-h urine protein, and the use of oral hypoglycemic drugs. We developed four different models as clinical, laboratory, lab-medication, and full models according to these independent risk factors. Laboratory model performed well in both discrimination and calibration among all the models (C-statistics: external validation 0.863; p value of the Hosmer–Lemeshow, .817). There was no significant difference in NRI among laboratory model, lab-medication model, and full model (p > .05). So, we chose the laboratory model as the optimal model. Conclusion We constructed a nomogram which contained hemoglobin, NLR, serum cystatin C, eGFR, and 24-h urine protein to predict the risk of patients with diabetic kidney disease initiating renal replacement in 3 years.