Corollary Order Recommendation Using Bayesian Networks

With the popularization of information systems, healthcare organizations accumulated masses of clinical data, e.g. diagnoses, orders, examinations and lab tests. These data reflect past decision information, are up-to-date, therefore are suited to be the foundation of building a clinical decision system. Over the recent years, research on using various data analyzing techniques to build models on medical data is on the rise, showing great theoretical & practical value. This work focus on building a clinical decision making system that recommends corollary orders, based on order & diagnoses history data. With carefully filtered raw data and properly tweaked models, it can be expected that such system produces context-aware order recommendations, which better meets care givers’ needs.In the paper, a literature review was carried out on corollary order recommendation systems, summarizing the shortcomings of current research. Then, we systemized clinical decision processes by reference to clinical pathways. Based on Bayesian Networks and other data mining methods, we constructed a clinical decision support model, with the addressing of problems of insufficient training data and biased recommendation lists as innovation points of this study. By summarizing the evaluation methods of previous studies, we developed new ways evaluate the system recommendation’s ‘effectiveness’ along with its ‘accuracy’, providing new insights for evaluation methods of clinical recommendation systems. Further, we developed a prototype order recommendation system for inpatient treatment, and carried out multiple tests and result analysis of the system using a real world data set. Comparison tests showed that the proposed Bayesian Network method outperforms benchmark models, indicating the effectiveness and correctness of the proposed method.