Textual analysis and visualization of research trends in data mining for electronic health records

摘要

Abstract Objectives Medical data mining is one of the most widely used techniques for discovering latent knowledge from databases, which in turn contributes to clinical decisions. In the past decade, medical data mining has advanced rapidly. The objective of this study is to analyse research trends and explore the general research framework in data mining for electronic health records (EHRs). Methods We first conducted a literature retrieval in PubMed, the Web of Science (WOS) core collection, and the Association for Computing Machinery (ACM) digital library for peer-reviewed records (n = 2516) related to data mining for EHRs from 2000 to 2016. Then, we adopted the Latent Dirichlet Allocation (LDA) and Topics over Time (TOT) models to extract topics and analyse topic evolution trends in the retrieved records. The former mainly analysed topic generation, division, mergers and extinction, while the latter analysed the evolution of topic intensity over time. Results We extracted the important topics and analysed topic evolution. We present the general research framework of data mining for EHRs by combining the topic co-occurrence relations and domain knowledge, including the data, methods, knowledge, and decision levels. Conclusions Our work can provide high-level insight for scholars in this emerging field and guide their choices of medical data mining techniques in healthcare knowledge discovery, medical decision support, and public health management. Highlights ? This study used topic models to analyze topic evolution of EHR data mining research. ? This study built a medical synonym dictionary and proposed a topic similarity method. ? This study presented a general research framework of EHR data mining research.