Abstract: Background　Partially observable Markov decision-making process is a mathematical algorithm for solving sequential decision-making problems.It is often used in the dynamic optimization of complex treatment schemes in the medical field.Now it has been tried to be used in the optimization of clinical treatment for unstable angina pectoris.Objective　To introduce in detail the modeling method of the optimized process of clinical treatment for unstable angina pectoris with integrated traditional Chinese and western medicine，and to preliminarily verify the reliability of the model.Methods　A total of 111 060 structured medical records were collected from 2 212 patients with unstable angina pectoris from 2008 to 2010 in Xiyuan Hospital，Guang'anmen Hospital，China-Japan Friendship Hospital，Beijing Hospital of Traditional Chinese Medicine，Tongzhou District Hospital of Traditional Chinese Medicine，Dongzhimen Hospital，Huairou District Hospital of Traditional Chinese Medicine，and Peking University People's Hospital.Patients with unstable angina pectoris of Qi deficiency and blood stasis were randomly divided into the test group（n=732） and the verification group（n=732）.The core prescriptions of the optimal treatment scheme were obtained by observing Markov decision model in the test group.The patients eligible for the scheme in the verification group were selected as the optimization group，and those who did not conform to the scheme were selected as the control group.The blood stasis scores and the occurrence of long-term end-point events were compared to verify the clinical efficacy of the scheme.The primary end-point events included cardiovascular death，nonfatal myocardial infarction，revascularization，and secondary end-point events included stroke，readmission due to acute coronary syndrome，cardiac insufficiency，and other thrombotic complications.Results　The decrease of blood stasis scores in Qi deficiency optimization group（n=152） and blood stasis optimization group（n=127） was （5.29±2.64） and （4.87±3.04）respectively，which were higher than those in the control group，with statistical significance（P<0.05）.Compared with the control group，the long-term end-point event rates of qi deficiency optimization group and blood stasis optimization group were 16.4%（25 / 152） and 12.6%（16 / 127），respectively，with no significant difference（P>0.05）.Conclusion　Partially observable Markov decision-making process model can realize the optimization of clinical treatment for unstable angina pectoris with integrated traditional Chinese and western medicine.The model has been proved to be stable and reliable.It could be possible to provide more detailed and in-depth optimization programs for other diseases clinically.

Key words: Angina, unstable；Coronary heart disease；Partially observable Markov decision-making process；Modeling method；Validation

摘要： 背景　部分可观察马尔科夫决策过程（POMDP）是一种解决序列决策问题的数学算法，在医学领域中常应用于复杂治疗方案的动态优化。其在不稳定型心绞痛（UA）中西医结合临床治疗方案优化中的应用已进行了初步尝试。目的　详细描述POMDP在UA中西医结合临床治疗方案优化过程中的建模方法并对模型的可靠性做出初步验证。方法　收集中国中医科学院西苑医院、中国中医科学院广安门医院、中日友好医院、北京市中医院、通州区中医院、东直门医院、怀柔中医医院、北京大学人民医院2008—2010年2 212例UA患者结构化住院病历111 060条记录。将UA证候要素类型为气虚、血瘀的患者随机分为测试组732例和验证组732例，对测试组患者应用POMDP得出一组最优治疗方案的核心处方，在验证组患者中筛选符合方案的患者（作为优化组），与不符合方案的患者（作为对照组）进行血瘀证积分、远期终点事件（主要终点事件包括心血管死亡、非致命性心肌梗死、血运重建术，次要终点事件包括脑卒中、因急性冠脉综合征再住院、心功能不全、其他血栓并发症）发生情况的比较，验证方案的临床疗效。结果　气虚优化组（n=152）、血瘀优化组（n=127）患者血瘀证积分降分分别为（5.29±2.64）、（4.87±3.04）分，均高于对照组，差异有统计学意义（P<0.05）。气虚优化组、血瘀优化组远期终点事件发生率分别为16.4%（25/152）、12.6%（16/127），与对照组比较，差异无统计学意义（P>0.05）。结论　POMDP可以实现UA中西医结合临床治疗方案的优化，模型经验证较为稳定可靠，可尝试为其他疾病临床方案的制定提供更为细化、深入的优化方案。

关键词: 心绞痛, 不稳定型；冠心病；部分可观察马尔科夫决策过程；建模；验证