Abstract：To find out the changing trend of medical behaviors in clinical pathways (CPs) can present convincing evidence for the improvement of clinical pathway template and help to improve the medical quality of CPs. In this study, we proposed a twostep approach to detect medical behavior changes in CPs over time. In the first step, we adopted a wellknown probabilistic topic model, i.e., latent Dirichlet allocation (LDA), to discover yearly treatment patterns w.r.t risk stratifications of patients from electronic medical records (EMR). With the discovered treatment patterns, the changes of medical behaviors were classified into six content change patterns (i.e., emerged, faded, stable, updown, downup and jumping) and three occurring time change patterns (i.e., earlyoccurred, stable and delayoccurred). The proposed method was evaluated via 12152 EMRS pertaining to the unstable angina pathway with time arranges of 10 years. Experimental results indicated that the proposed approach is effective to mine the significant changes of content and time of medical behaviors and therefore presents convincing evidences for scheduling better practice of CPs.
［1］Adeyemi S, Demir E, Thierry C. Towards an evidencebased decision making healthcare system management: Modelling patient pathways to improve clinical outcomes ［J］. Decision Support System, 2013, 55(1):117-125.
［2］Lin Furen, Chou Shienchao, Pan Shuangmei, et al. Mining time dependency patterns in clinical pathways［J］. International Journal of Medical Informatics, 2001,62(1):11-25.［3］Peleg M, Mulyar N, Van Der Aalst WMP. Patternbased analysis of computerinterpretable guidelines: Don’t forget the context［J］. Artificial Intelligence in Medicine, 2012, 54(1):73-74.
［4］Huang Zhengxing, Lv Xudong, Duan Huilong. Latent treatment pattern discovery for clinical processes ［J］. Journal of Medical Systems, 2013, 37(2): 9915.［5］Van de Klundert J, Gorissen P, Zeemering S. Measuring clinical pathway adherence［J］. Journal of Biomedical Informatics, 2010, 43(6):861-872.
［6］Blei D, Ng A, Jordan M. Latent Dirichlet allocation ［J］. Journal of Machine Learning Research, 2003,
3: 993-1022.
［7］Catherwood E, O’Rouke DJ. Critical pathway management of unstable angina ［J］. Progress in Cardiovascular Diseases, 1994, 37(3):121-148.
［8］Dong Guozhu, Li Jinyan. Efficient mining of emerging patterns: Discovering trends and differences ［C］ // Proceedings of the fifth International Conference on Knowledge Discovery and Data Mining. San Diego: ACM Digital Library, 1999: 43-52.
［9］Wang Chong, Blei D, Heckerman D. Continuous time dynamic topic models ［C］ // Proceedings of the TwentyFourth Conference on Uncertainty in Artificial Intelligence. Helsinki: arXiv, 2008:579-586.
［10］Huang Zhengxing, Dong Wei, Peter B, et al. On mining latent medical behavior patterns from electronic medical records［EB/OL］. http://link.springer.com/article/101007/s10618-014-0381-y, 2014-09-24/2014-12-30.
［11］Hall D, Jurafsky D, Manning C. Studying the history of ideas using topic models ［C］ // Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing. Waikiki: ACM Digital Library, 2003: 363-371.
［12］Granger CB, Goldberg RJ, Dabbous O, et al. Predictors of Hospital Mortality in the Global Registry of Acute Coronary Events ［J］. JAMA, 2003, 163(19):2345-2353.
［13］Eagle KA, Lim MJ, Dabbous OH, et al. A validated prediction model for all forms of acute coronary syndrome estimating the risk of 6month postdischarge death in an international registry ［J］. JAMA, 2004, 291(22):2727-2733.
［14］Antman EM, Cohen M, Bernink PJ, et al. The TIMI risk score for unstable angina/nonST elevation MI: A method for prognostication and therapeutic decision making ［J］. Circulation, 2000, 284(7): 835-842.
［15］Morrow DA, Antman EM, Bernink PJLM, et al. TIMI risk score for STelevation myocardial infarction: a convenient, bedside, clinical score for risk assessment at presentation ［J］. Circulation, 2000, 102: 2031-2037.
［16］沈剑锋，汪威，杜平，等. 基于卫生信息平台的电子病历信息共享研究和实现［J］.中国生物医学工程学报, 2013,32(4):504-507.
