Yin Jianming, Zheng Chuanjie, Lin Xiaoqian, Huang Chaoqiang, Hu Zhanhui, Lin Shuyuan, Qu Yiqian
School of Basic Medical Sciences, Zhejiang Chinese Medical University, Hangzhou, China.
Front Endocrinol (Lausanne). 2025 Jan 23;15:1499417. doi: 10.3389/fendo.2024.1499417. eCollection 2024.
Previous studies have indicated an association between UHR and diabetes risk, but evidence from large-scale and diverse populations remains limited. This study aims to verify UHR's independent role in diabetes risk prediction in a large sample population and assess its applicability across different populations. We drew upon data from 30,813 participants collected during the 2005-2018 NHANES cycle. The association between UHR and the risk of diabetes was explored using multivariate logistic regression models, with key predictive factors identified through LASSO regression. Model effectiveness was evaluated through receiver operating characteristic (ROC) curves, decision curve analysis (DCA), and calibration metrics. Additionally, restricted cubic spline (RCS) and threshold effect assessments were applied to examine the nonlinear association between UHR and diabetes risk. The results showed that UHR levels were notably elevated in individuals with diabetes when compared to those without diabetes ( < 0.001). The occurrence of diabetes showed a marked increase across ascending UHR quartiles (6.63%, 10.88%, 14.15%, 18.02%; < 0.001). Results from multivariate logistic regression indicated that elevated UHR was strongly linked to a heightened risk of diabetes; participants in the highest UHR quartile were found to have nearly four times the risk compared to those in the lowest quartile (OR = 4.063, 95% CI: 3.536-4.669, < 0.001). Subgroup analyses demonstrated that the predictive effect of UHR was more pronounced in females. Key variables selected via LASSO regression improved the model's performance. Restricted cubic spline (RCS) analysis indicated an inflection point at UHR = 10; beyond this point, diabetes risk accelerated, and when UHR exceeded 18, the risk increased significantly (OR > 1). ROC curve analysis showed the baseline model (M1) had an area under the curve (AUC) of 0.797, while the multivariable model (M4) after LASSO selection had an AUC of 0.789. Decision curve analysis and calibration curves validated the model's predictive ability and consistency. This study indicates that UHR may be an independent predictor of diabetes risk, showing a positive correlation with diabetes and a more pronounced predictive effect in females.
先前的研究表明未分化甲状腺炎（UHR）与糖尿病风险之间存在关联，但来自大规模和多样化人群的证据仍然有限。本研究旨在验证UHR在大样本人群中对糖尿病风险预测的独立作用，并评估其在不同人群中的适用性。我们利用了2005 - 2018年美国国家健康与营养检查调查（NHANES）周期中收集的30813名参与者的数据。使用多变量逻辑回归模型探讨UHR与糖尿病风险之间的关联，并通过套索（LASSO）回归确定关键预测因素。通过受试者工作特征（ROC）曲线、决策曲线分析（DCA）和校准指标评估模型有效性。此外，应用受限立方样条（RCS）和阈值效应评估来检验UHR与糖尿病风险之间的非线性关联。结果显示，与未患糖尿病的个体相比，糖尿病患者的UHR水平显著升高（<0.001）。随着UHR四分位数的升高，糖尿病的发生率显著增加（6.63%、10.88%、14.15%、18.02%；<0.001）。多变量逻辑回归结果表明，UHR升高与糖尿病风险增加密切相关；UHR最高四分位数的参与者与最低四分位数的参与者相比，风险几乎高出四倍（OR = 4.063，95%CI：3.536 - 4.669，<0.001）。亚组分析表明，UHR的预测作用在女性中更为明显。通过LASSO回归选择的关键变量提高了模型的性能。受限立方样条（RCS）分析表明，UHR = 10时有一个拐点；超过这一点，糖尿病风险加速上升，当UHR超过18时，风险显著增加（OR>1）。ROC曲线分析显示，基线模型（M1）的曲线下面积（AUC）为0.797，而LASSO选择后的多变量模型（M4）的AUC为0.789。决策曲线分析和校准曲线验证了模型的预测能力和一致性。本研究表明，UHR可能是糖尿病风险的独立预测指标，与糖尿病呈正相关，且在女性中的预测作用更为明显。
