Interactive Visualization of a Patient’s Electronic Health Information to Assist Manual Chart Review Using Tableau® Software

Background: Identification of many important health conditions requires synthesis of subtle clinical observations diffusely recorded in electronic health record (EHR) notes, making manual chart review the only viable method of collecting such information. Clinical documentation of problem use of prescription opioids, including abuse and addiction, is an example. A research project we conducted required determining whether problem opioid use was documented in the charts of 2,000 patients receiving chronic opioid therapy during a 9-year observation period. To enhance thoroughness and efficiency, we provided chart abstractors interactive graphical summaries of selected information from each patient’s EHR as an abstraction aid. Methods: We used pilot chart reviews and expert opinion to identify four types of information considered useful for identifying clinical documentation of problem opioid use: 1) the timing, type and days’ supply of opioid fills; 2) encounters; 3) diagnoses including behavioral health, substance abuse and chronic pain conditions; and 4) mentions in clinical notes of terms related to problem opioid use. We obtained structured data from the virtual data warehouse. We used natural language processing (NLP) to extract information from clinical notes. We created longitudinal graphical displays of each content area using Tableau® (www.Tableau.com), a business intelligence software product that simplifies the creation of graphical representations and real-time exploration of complex data. Graphics were juxtaposed on “dashboards” with shared time scales and drillable details (eg, details of medications fill, text surrounding NLP-extracted terms) facilitating a visual synthesis of multiple types of potentially relevant information. Results: Five experienced chart abstractors used the interactive graphics on one screen and the Epic® EHR interface on a second screen to conduct each chart review. Abstractors reported that the graphics facilitated efficiency through more rapid detection of periods of care in which problem opioid use may be documented (eg, emergency room encounters coinciding with early opioid refills). Particularly valuable was the ability to “see the larger picture” while also being able to drill into the details of specific events. Conclusion: Graphical visualization of information from EHRs can assist manual abstraction of health conditions when determinations about the presence of those conditions require synthesis of diffusely recorded content in voluminous charts