Information discovery on electronic health records using authority flow techniques

Abstract

Background: As the use of electronic health records (EHRs) becomes more widespread, so does the need to search and provide effective information discovery within them. Querying by keyword has emerged as one of the most effective paradigms for searching. Most work in this area is based on traditional Information Retrieval (IR) techniques, where each document is compared individually against the query. We compare the effectiveness of two fundamentally different techniques for keyword search of EHRs.

Methods: We built two ranking systems. The traditional BM25 system exploits the EHRs' content without regard to association among entities within. The Clinical ObjectRank (CO) system exploits the entities' associations in EHRs using an authority-flow algorithm to discover the most relevant entities. BM25 and CO were deployed on an EHR dataset of the cardiovascular division of Miami Children's Hospital. Using sequences of keywords as queries, sensitivity and specificity were measured by two physicians for a set of 11 queries related to congenital cardiac disease.

Results: Our pilot evaluation showed that CO outperforms BM25 in terms of sensitivity (65% vs. 38%) by 71% on average, while maintaining the specificity (64% vs. 61%). The evaluation was done by two physicians.

Conclusions: Authority-flow techniques can greatly improve the detection of relevant information in EHRs and hence deserve further study.

Figure 1

A subset of Electronic Health Record Dataset. A subset of the relational-anonymized experimental EHR dataset of the Cardiac Division of Miami Children's Hospital, which contains clinical entities like hospitalization, patient, employee, medication, diagnosis, diagnostics, cardiac, events, and labs.

Figure 2

An example HL7 CDA XML medical document. An example HL7 CDA XML medical document that shows the use of ID/IDREF attributes in XML.

Figure 3

Schema of the EHR dataset. The schema graph of the EHR dataset is a directed graph that describes the structure of the EHR dataset in Figure 1.

Figure 4

Sample User Survey Page for query "respiratory distress". Figure 4 shows a sample user survey page for query "respiratory distress". It starts with a brief explanation about the survey and then query results are displayed. Each result displays a brief description of the clinical entity in tabular format along with links to the complete description and adjacent entities.

Figure 5

Sample "Full Description" of HospitalizationID 2406 for query "respiratory distress". Figure 5 shows the Full Description of HospitalizationID_2406 for query "respiratory distress" in tabular format with each row displaying an attribute-value. Each occurrence of the query phrase is highlighted.

Figure 6

Sample explaining sub graph of an ObjectRank2 result - HospitalizationID 2406 for query "respiratory distress". Figure 6 displays an explaining sub graph for HospitalizationID 2406 for query "respiratory distress". The figure gives a better picture of why this entity was ranked higher for query "respiratory distress" and its relationship with other entities that contain the query keywords. Detail descriptions of each entity are displayed as tool tip texts when the user points at them.