School of Statistics, Renmin University of China, Beijing, 100872, China.
Stat Med. 2012 Mar 30;31(7):643-52. doi: 10.1002/sim.4382. Epub 2012 Feb 21.
In the Traditional Chinese Medicine (TCM) cross-sectional survey conducted by our team, we were interested in determining the risk factors of osteoporosis. To analyze this TCM study, we had to deal with three statistical problems: (1) a very large number of potential risk factors, (2) interactions among potential risk factors, and (3) nonlinear effects of some continuous-scale risk factors. To address these analytic issues, we used two data mining methods, support vector machine recursive feature elimination and random forest; to deal with the curse of high-dimensional risk factors, we applied another data mining technique of association rule learning to discover the potential associations among risk factors. Finally, we employed the generalized partial linear model (GPLM) to determine nonlinear effects of an important continuous-scale risk factor. The final GPLM model shows that TCM symptoms play an important role in assessing the risk of osteoporosis. The GPLM also reveals a nonlinear effect of the important risk factor, menopause years, which might be missed by the generalized linear model.
在我们团队进行的中医（TCM）横断面调查中，我们有兴趣确定骨质疏松症的危险因素。为了分析这项 TCM 研究，我们必须处理三个统计问题：（1）大量潜在的危险因素；（2）潜在危险因素之间的相互作用；（3）一些连续尺度危险因素的非线性效应。为了解决这些分析问题，我们使用了两种数据挖掘方法，支持向量机递归特征消除和随机森林；为了应对高维危险因素的诅咒，我们应用了关联规则学习的另一种数据挖掘技术来发现危险因素之间的潜在关联。最后，我们采用广义部分线性模型（GPLM）来确定重要连续尺度危险因素的非线性效应。最终的 GPLM 模型表明，中医症状在评估骨质疏松症风险方面起着重要作用。GPLM 还揭示了重要危险因素——绝经年限的非线性效应，这可能会被广义线性模型所忽略。
