JMIR Publications

ABSTRACT

Background:

Clinical information models enabling both technical and semantic interoperability are crucial for electronic health record (EHR) data use and reuse. Collaborative application of clinical information models helps ensure the interoperability of EHRs. Therefore, how to discover clinical information models from an open repository online to help health information system developers represent EHR data in a standard manner become important.

Objective:

This study aimed to develop a retrieval method to identify clinical information models online to represent EHR data.

Methods:

We proposed a graphical retrieval method. We probabilistic graphical represented of clinical information models using an extended Bayesian network with two concept-layers. In the inference process, we calculated the probability of relevance of each clinical information model for the given query, and sorted the result by decreasing order of probability. In evaluation, we compared our graphic-based method with three typical retrieval methods (BM25F, simple Bayesian network and the Clinical Knowledge Manager) in the medication, laboratory test and diagnosis retrieval tasks, using evaluation metrics of mean average precision (MAP), average precision (AP) and precision at cut-off points 10 (P@10).

Results:

We downloaded 526 clinical information models from an open repository of openEHR. Then, a clinical resources network was constructed. The network consisted of 5,513 nodes, which were 3,982 information element nodes (T), 504 concept nodes (C), 504 duplicated concept nodes (C’), and 523 clinical information model nodes (A), with 6366 edges from T to C, 2958 edges from C to C’, and 543 edges from C’ to A. The results showed that our method achieved the best mean average precision (MAP = 0.32). When retrieving information models related to diagnosis, our method could successfully identify the models covering diagnostic reports, problem list, patients background, clinical decision and etc., as well as models which other retrieval methods could not find, such as problems and diagnoses.

Conclusions:

The extended Bayesian network retrieval method we proposed is an effective approach to meet the uncertainty of finding clinical information models. Our method can help health information system developers identify clinical information models to represent EHR data in a standard manner, enabling EHR data exchangeable and interoperable.