基于中医治疗COPD临床试验建立数据混杂因素的控制方法研究

The curative effect evaluation of traditional Chinese medicine aims to judge the relationship between treatment factors and its outcomes. The objective existence of many confounding factors can lead to bias analysis of the results, and further influence the validity of outcomes. Traditional methods for controlling confounding factors (hierarchical, correspondence, standardized rates, covariance analysis, unconditional logistic regression, etc.) would be limited to use when there are too many confounding factors and data collinearity.While Propensity Scores and Diseases Risk Scores may avoid the defects mentioned above, and need further test and verify. We point out the hypothesis that Propensity Scores and Diseases Risk Scores will be better to control the confounding factors of research data comparing with unconditional Logistic regression. We will carry out the following study. Take TCM clinical observational trial for example, within 1000 COPD patients, establish model variable selection criteria and parameters through the expert questionnaire, then recognize and screen potential confounding factors that influences outcome evaluatation of large sample and multi-center clinical trials, further compare the effect of nconditional Logistic regression, Propensity Scores and Diseases Risk Scores with effectiveness and applicability in controlling confounding factors, and establish the best methods and then get truth results. The established best methods will provide the inspirations and methods for other chronic disease researches.

中医疗效评价的目的是判断处理因素和结局之间的因果关系。临床试验中大量混杂因素导致了结果偏倚，并影响结局评价的效度和真实性。传统的混杂因素控制方法（分层、匹配、标准化率、协方差分析和logistic回归等）在混杂因素高维和共线性时的应用受到限制，倾向性评分法和疾病风险评分法可能避免上述缺点，亟需进一步验证。本研究提出"倾向性评分法和疾病风险评分法较非条件logistic回归能更好的控制数据的混杂因素"的假说，拟开展如下研究：以中医治疗1000例COPD临床观察试验为示范，通过专家问卷建立模型变量选择标准及参数，识别、筛选影响结局评价的潜在混杂因素，评价非条件logistic回归、倾向性评分和疾病风险评分三种方法控制混杂因素的效果和适用性，建立适合中医临床研究的最佳混杂因素控制方法，为获得低偏倚、高质量的研究结果提供方法学支持。所建立的混杂因素控制方法将为其他研究提供可借鉴的思路和方法。

中医临床试验中大量混杂因素导致了结果偏倚，并影响结局评价的效度和真实性。有效的混杂因素控制方法亟需建立。本研究以中医治疗COPD临床试验为示范，完成以下内容：（1）完成了专家调查，建立了混杂因素模型变量筛选标准及参数：基于文献研究分析，制定了变量筛选咨询问卷，遴选全国50名专家，采用邮件信函法进行调查。共回收有效问卷46份，采用德尔菲法进行计算后，专家意见集中在：既往急性加重次数、急性加重持续时间、咳嗽、咯痰、胸闷、喘息、呼吸困难指数、病史、生存质量积分、六分钟步行距离、费用、年龄、性别、身高、体重、医保情况；同时明确了分析模型使用标准和具体参数。（2）完成了COPD稳定期中西医结合治疗方案的疗效评价研究：基于“真实世界”研究的多中心、大样本观察性研究方法，观察了1326例COPD稳定期患者，并进行了相关数据采集，形成了数据分析库。疗效显示中西医结合治疗COPD在急性加重次数、中医证候积分、呼吸困难分级、生存质量等方面优于西医常规治疗。（3）完成了COPD稳定期中西医结合治疗方案的混杂因素分析研究：采用线性回归模型对影响中西医结合治疗方案、西医常规治疗方案的疗效的因素进行了分析，共采集了97个混杂因素变量。将混杂因素变量分为连续型、二分类、多分类。采用倾向评分匹配方式，试验组和对照组2:1匹配，限定倾向值评分差异在0.5以内，计算倾向值评分和匹配结果；并采用疾病风险评分泊松回归进行混杂因素分析，验证了混杂因素。倾向性评分联合疾病风险评分能够较好的控制混杂因素。