基于多示例多标记学习发现艾滋病“方-证-效”动态靶关系的方法研究

Between the treatment plan and the population in TCM prescription-syndrome-efficacy, efficacy of individual patient is hard to be repeated in the population clinical trials, which is the key problem in TCM efficacy evaluation. For example, the prescription of famous TCM doctor is effective for the difficult miscellaneous diseases. However, the effective prescription cannot obtain the efficacy for the patients in the population clinical trials. The main reason is that the effective prescription of individual patient cannot deal with the individual and dynamic feature of the population in the clinical trials. Beginning with the immunological indicators, we propose that 20000 retrospective AIDS cases and 400 prospective AIDS cases are used iteratively to explore the effective prescription and the corresponding population feature based on. In this research, multiple-instance multiple-label learning is used to mine the structure, dose relation and dynamic relationship of the effective prescription and exploit the dynamic relationship, self-feeling and the effective patients’ feature. Then, the dynamic target relationship of AIDS prescription-syndrome-efficacy is able to be build reveal the implication of efficacy evaluation of TCM treating AIDS. This research will provide the data evidence of effective prescription for the AIDS clinical doctor. Moreover, this research will provide a feasible way to repeat the effective individual treatment in the population clinical trials.

中医临床“方-证-效”的治疗方案与研究人群之间，个体疗效难以在群体层次临床试验中完全精准重复，是中医疗效评价的关键科学问题之一。如名老中医经验方治疗疑难杂症显效，但在群体层次临床试验时难保重复出来。主要原因是个体层次取得的显效处方用于群体研究时，不能精细处理对应研究人群的个性化、动态化特征。该问题在中医药治疗艾滋病效果评价中尤为突出，导致艾滋病临床很多年未发现有效治疗方案。本研究以2万例回顾和400例前瞻艾滋病实际诊疗数据为基础，采用多示例多标记学习方法，从免疫功能指标出发，反复凝练艾滋病群体层次具有精准靶关系的有效处方和适宜人群，探索有效处方构成、剂量关系、加减变化等，探索对应适宜人群动态变化、自我感受、有效患者特征等，建立发现艾滋病“方-证-效”动态靶关系的方法，揭示艾滋病疗效评价的内涵，为临床发现艾滋病有效治疗方案提供数据支持，也为艾滋病群体研究时可重复显著个体疗效提供一条可行途径。

本人承担的2012年青年基金“基于图论方法建立艾滋病辨证论治辅助决策模型，在揭示艾滋病症-证-治之间潜在关联关系的最优决策路径研究中，发现“研究人群与治疗处方不对应”是导致中医临床评价结果失真的关键问题之一。在2016年面上基金“基于多示例多标记学习发现艾滋病“方-证-效”动态靶关系的方法研究”中，通过多示例多标记学习、深度学习等数据挖掘方法，深入分析了中医药治疗艾滋病的人群特征（证候）与治疗处方之间一一对应的靶关系，进一步明确了提高免疫功能指标（CD4）时，对应的适应人群和有效核心处方，找到了中医药治疗艾滋病的 “方-证-效”动态靶关系。该方法为中医药治疗艾滋病群体疗效研究时可重复显著个体疗效提供一条可行途径。