偏最小二乘法在肝炎后肝硬化中医证候分类建模中的应用

Classification and modeling of traditional Chinese medicine syndromes in patients with posthepatitic cirrhosis by partial least-squares

目的:探讨偏最小二乘法(partial least-squares,PLS)在肝炎后肝硬化中医证候分类模型建立中的应用及临床价值。方法:拟定中医四诊等生物信息采集基本要求,用临床流行病学调查方法,采集278例肝炎后肝硬化患者症状、体征、舌象、脉象及临床实验室指标等生物信息;采用逐步回归方法筛选各证候相关指标,设证候类型为因变量矩阵,相关临床信息为自变量矩阵,所有变量均经列最大化处理后导入PLS程序中进行计算,采用留一法(Leave-One-Out,LOO)完成建模和样本的预测。结果:逐步回归建立的5个证候类型回归方程的决定系数R2均在0.7以上;在此基础上,基于PLS建立的肝炎后肝硬化中医证候判识模型对肝郁脾虚、肝肾阴虚、脾肾阳虚、湿热内蕴和瘀热蕴结等主证候的预测准确率分别为73.02%、68.71%、92.45%、84.17%和68.71%。结论:PLS法能提高中医临床生物信息的利用度,在中医复杂证候建模的研究中有一定的推广应用价值。

Objective： To investigate the classification and modeling of traditional Chinese medicine （TCM） syndromes in posthepatitic cirrhosis by partial least-squares （PLS） method, and to study the clinical application of PLS method in TCM research. Methods： Inclusion criteria, exclusion criteria and criteria for harvesting and inputting of the biological information such as symptoms, signs, tongue, pulse and biological parameters were established. Stepwise regression was used to analyze the clinical information, including clinical symptoms and biological parameters obtained from 278 patients with posthepatitic cirrhosis by clinical epidemiological approach. TCM syndrome types were set up as dependent variable matrix, and the related clinical information was screened by stepwise regression as independent variable matrix. With the column maximization, all the variables were transformed into the PLS program. Mathematical models were obtained by Leave-One-Out in PLS program. Results： The determination coefficients R^2 of 5 regression equations were above 0.7. The practical accuracy rates of the models in TCM patterns including liver stagnation and spleen asthenia, liver-kidney yin deficiency, spleen-kidney yang deficiency, retention of dampness-heat, and blood stasis-heat accumulation were 74.02%, 68.71%, 92.45%, 84.17% and 70.50% by PLS, respectively. Conclusion： The utilization of clinical information can be improved by PLS, which is definitely useful in the classification and modeling of TCM complicated syndromes.