基于两种模型的绝经后骨质疏松性骨折早期预警方法机制研究

Osteoporotic fractures are the most devastating clinical outcomes, and postmenopausal women are a population with high risk of osteoporotic fractures. It is very important to early warn the risk of postmenopausal osteoporotic (PMOP) fractures occurance and take preventative intervention. In the finished project (NSFC, 2009-2011) , a PMOP risk evaluation model (Normal bone→Osteopenia→Osteoporosis) has been established on the 3 years epidemiological survey for the risk factors, traditional Chinese medicine(TCM) syndrome essential factors related information and bone mineral density(BMD) testing on 843 PMOP high-risk persons aged 40 to 65 years old, using the non-homogeneous Markov model. In this project, we aim to research on method mechanism for the early warning of PMOP fractures. A further 3 years epidemiological survey will be conducted for the risk factors, TCM syndrome essential factors related information and BMD testing on 1000 PMOP high-risk persons, analyse the TCM syndrome characteristics of the PMOP fractures population, then screen out important the risk factors and TCM syndrome essential factors closely related with PMOP→fractures through Group Lasso method, make clear the early warning indexes for the PMOP→fractures according to the PMOP clinical practice.Two different early warning models for PMOP fractures will be established on the base of the non-homogeneous Markov model and generalized linear single index semiparameteric survival model (GLSISSM). The predictive accuracy of the two early-warning models will be compared through the ROC curve method, and then a PMOP fracture early warning tool will be established with both accuracy and practicability through parameters adjustment and optimization, as scientific basis for the early prevention of PMOP fractures.

骨质疏松性骨折是骨质疏松症最具破坏性的临床结局，绝经后女性是其高发人群，其早期预警是警示高危人群骨折发生的重要途径，可为有效干预提供客观依据。在完成2009-2011年国家自然基金843例PMOP高危人群连续3年危险因素、中医证候要素测试及骨密度检测并建立"骨量正常→骨量减少→骨质疏松"风险评估模型基础上，本项目旨在探索"骨质疏松→骨折"早期预警新方法建构机制，拟新开展1000例社区PMOP高危人群连续3年测试，分析PMOP骨折人群中医证候要素特征，运用Group Lasso方法筛选出PMOP→骨折相关高危因素和中医证候要素，结合临床经验确定PMOP→骨折的预警指标，基于非齐性Markov模型和GLSISSM模型建立PMOP骨折早期预警模型，运用ROC曲线比较两种模型的预测准确性，经参数调整和优化，形成兼具准确性和实用性的PMOP骨折早期预警工具，进而明确PMOP骨折早期预警方法建构机制。

1.研究背景：绝经后骨质疏松性骨折是绝经后骨质疏松症最具破坏性的结局之一。如何有效的早期预测绝经后骨质疏松性骨折的发生风险极具意义。目前鲜有基于我国人口学特征，同时包含危险因素及中医证候要素构建的绝经后骨质疏松性骨折风险预测工具。.2 研究内容：2.1运用隐树模型、Group Lasso、决策树分类模型方法分别筛选绝经后骨质疏松症患者发生骨折的现代危险因素和中医证候要素，构建绝经后骨质疏松性骨折风险预警模型。2.2 根据ROC曲线下面积判断所建立风险模型的预测性能，构建兼具准确性、实用性风险评估工具并明确构建机制。.3 研究结果：3.1经Group Lasso模型筛选得出，具体组变量包括：骨密度、年龄、食物类、身高、月经情况、孕产次数和肝肾阴虚。决策树模型共筛选得出6个解释变量，各指标按重要程度从大到小依次为骨密度、目眩、肉类、生产次数、视物模糊和乏力。 3.2基于决策树分类模型与Logistic回归分析模型对于PMOP骨折的预测概率绘制ROC曲线。经计算，决策树模型的AUC值为0.871（95%CI = 0.8226 — 0.9211）；Logistic回归模型的AUC值为0.8775（95%CI = 0.8412 — 0.9138）。.4 研究结论：初步建立了基于北京、上海人口学特征的40-65岁女性绝经后骨质疏松性骨折早期风险预测工具。在危险因素的基础上，加入了中医证候要素的相关内容，提高了模型的早期预警能力，为临床实践提供可参考的工具。对比两种预警模型的构建过程与结果，初步明晰了病证结合风险预警模型的构建机制，为今后疾病预警模型的构建提供了范例与指导。.5 科学价值：在中医药领域开展绝经后骨质疏松性骨折患者3 年随访研究，以骨折发生与否作为终点结局指标；运用Group Lasso分析、决策树分类模型等方法，从大量协变量数据中筛选可能影响绝经后骨质疏松性骨折发生的重要现代危险因素和中医证候要素；在绝经后骨质疏松性骨折发生风险预测模型中，融入中医证候要素的特色内容，增加中医临床的意义。