中医观察性研究中混杂因素识别及控制的方法及挑战

Methods and Challenges for Identifying and Controlling Confounding Factors in Traditional Chinese Medicine Observational Studies

摘要:

作为随机对照试验的补充，观察性研究可为中医治疗措施的疗效提供证据，还可用于疾病影响因素、病因及预后等问题的研究。但其由于其缺少随机化，普遍存在混杂效应，严重影响了研究因素与结局之间的因果推断，造成了混杂偏倚，因此识别和控制混杂因素是中医观察性研究中需重点解决的难题。根据因果关系网和中医理论特点，可以将混杂因素分为已测量的混杂和未测量的混杂。此外，还需关注混杂因素与中间变量的鉴别，以及与研究因素的交互作用。对于混杂因素的控制方法，可通过分层、多因素分析、倾向性评分和疾病风险评分等方法控制已测量的混杂因素；可使用工具变量法、双重差分法和底事件率比校正法对不可测量及未知混杂因素进行校正。通过对混杂因素的校正和控制，可保证组间的均衡性，减少混杂偏倚。此外，敏感性分析、判定交互作用等方法使得对混杂的控制更加全面。因中医自身特点，中医观察性研究在识别和控制混杂因素时面临着独特挑战，包括患者接受的中医治疗措施不断变化、中医四诊信息的混杂效应常被忽略及中医证候诊断缺少客观化标准。目前已有部分学者针对这些问题进行创新性的探索，为开展更高质量的中医观察性研究提供方法学基础，以获得更严谨的中医真实世界证据，并逐步形成符合中医特色的观察性研究质量评价标准。

Abstract:

As a supplement to randomized controlled trials， observational studies can provide evidence for the effectiveness of traditional Chinese medicine （TCM） treatment measures. They can also study influencing factors of diseases， etiology， and prognosis. However， there is a confounding effect due to the lack of randomization， which seriously affects the causal inference between the study factors and the outcome， resulting in confounding bias. Therefore， identifying and controlling confounding factors are key issues to be addressed in TCM observational studies. According to the causal network and the characteristics of TCM theory， confounding factors can be categorized into measured and unmeasured confounding factors. In addition， attention must be paid to identifying confounding factors and intermediate variables， as well as the interaction between confounding factors and study factors. For methods of controlling confounding factors， measured confounding factors can be controlled by stratification， multifactor analysis， propensity scores， and disease risk scores. Unmeasured and unknown confounding factors can be corrected using instrumental variable methods， difference-in-difference methods， and correction for underlying event rate ratios. Correcting and controlling confounding factors can ensure a balance between groups， and confounding bias can be reduced. In addition， methods such as sensitivity analysis and determination of interactions make the control of confounding factors more comprehensive. Due to the unique characteristics of TCM， observational studies of TCM face unique challenges in identifying and controlling confounding factors， including the ever-changing TCM treatment measures received by patients， the often-overlooked confounding effects in the four diagnostic information of TCM， and the lack of objective criteria for TCM evidence-based diagnosis. Some scholars have already conducted innovative explorations to address these issues， providing a methodological basis for conducting higher-quality TCM observational studies， so as to obtain more rigorous real-world evidence of TCM and gradually develop quality evaluation criteria for OS that are consistent with the characteristics of TCM.