基于LASSO回归的骨质疏松肾阳虚状态辨识模型研究  

Study on the state identification model of kidney yang deficiency in osteoporosis patients based on LASSO regression

作　　者：车前子 王晶[2,3,4] 白卫国[1] 刘孟宇[1] 田雅欣[1] 王拥军 杨伟[1] CHE Qian-zi;WANG Jing;BAI Wei-guo;LIU Meng-yu;TIAN Ya-xin;WANG Yong-jun;YANG Wei(Institute of Basic Research in Clinical Medicine,China Academy of Chinese Medical Sciences,Beijing 100700,China;Institute of Spine Disease,Shanghai University of Traditional Chinese Medicine,Shanghai 200032,China;Longhua Hospital,Shanghai University of Traditional Chinese Medicine,Shanghai 200032,China;Key Laboratory of Theory and Therapy of Muscles and Bones,Ministry of Education,Shanghai 200032,China)

机构地区：[1]中国中医科学院中医临床基础医学研究所,北京100700 [2]上海中医药大学脊柱病研究所,上海200032 [3]上海中医药大学附属龙华医院,上海200032 [4]筋骨理论与治法教育部重点实验室,上海200032

出　　处：《中华中医药杂志》2022年第10期5928-5933,共6页China Journal of Traditional Chinese Medicine and Pharmacy

基　　金：国家重点研发计划(No.2018YFC1704306,No.2018YFC1704300);国家自然科学基金面上项目(No.61771491)。

摘　　要：目的:构建骨质疏松患者的肾阳虚状态辨识模型,为辅助中医临床辨证分型提供参考。方法:共纳入993例骨质疏松患者。基于患者一般情况、病因病机、临床症状和体征,通过套索(LASSO)回归筛选肾阳虚状态相关的主要特征;以Logistic回归进一步建立预测模型,生成风险评分工具。结果:LASSO回归筛选出10个对肾阳虚状态具有诊断和鉴别价值的辨识要素,包括年老体衰、口干、心烦、畏寒、肢冷、手足心热、大便溏、小便清、夜尿频多和舌红。Logistic回归模型曲线下面积(AUC)在训练集为0.919,测试集为0.925,校准度统计学检验无显著性。根据回归系数构建的风险评分工具在训练集和测试集的AUC均在0.90以上,以1分为风险截断值,灵敏度、特异度均在0.80以上。结论:联合LASSO变量选择与Logistic回归构建的模型可识别骨质疏松肾阳虚状态,风险评分有望辅助肾阳虚状态的辨识。Objective: Aimed to establish a syndrome prediction model of kidney yang deficiency(KYD) to provide a reference for syndrome differentiation in osteoporosis patients. Methods: A total of 993 patients with osteoporosis were recruited in the research. Based on data on demographics, etiology and pathogenesis, and clinical symptoms for osteoporosis,the main characteristics related to KYD were selected by least absolute shrinkage and selection operator(LASSO) regression.We established KYD prediction model with selected features using Logistic regression, and further developed scoring system to compute the risk estimates. Results: Ten diagnostic or distinguishing factors features were finally selected for modeling after LASSO screening, including frail, dry mouth, irritability, chills, cold limbs, feverish sensation over the palm and sole, diarrhea,clear urine, frequent night urination and reddish tongue. The AUC of the Logistic regression model reached 0.919 in training set and 0.925 in testing set, the model was also well calibrated. The scoring system had an AUC over 0.90 in both training and testing set, and sensitivity and specificity were both over 0.80 when the risk cut-off was set to 1 point. Conclusion: LASSO variable selection combined with Logistic regression model can analyze and identify not only how the correlative identification factors affect KYD state of osteoporosis but also developed well performed prediction model and risk scoring tool, providing an evidencebased basis for further TCM syndrome differentiation and treatment of combination of disease and syndrome.

关 键 词：骨质疏松 肾阳虚 LASSO回归 状态辨识模型 风险评分工具 

分 类 号：R259[医药卫生—中西医结合]