Home >> ALL ISSUES >> 2022 Issues >> Newsbytes

Newsbytes

image_pdfCreate PDF

Pathology lab uses homegrown tool to search AP report text

June 2022—Laboratories seeking a means to query their repository of archival anatomic pathology reports would do well to follow the advice of former tennis star Arthur Ashe: “Start where you are, use what you have, do what you can.”

Adhering to the same guidance, pathologists at the University of Texas Southwestern Medical Center developed a tool to query the unstructured text in AP reports using the business intelligence software Tableau, for which the hospital already had a subscription. “From the get-go, the things we had on campus to build the tool ranked higher,” says Ellen Araj, MD, a pathology informaticist at UT Southwestern.

Finding a solution to perform retrospective AP report queries became a priority in 2018, when the pathology lab transitioned from Sunquest’s CoPath laboratory information system to Epic Beaker. Unlike CoPath, Beaker cannot perform retrospective full-text searches of AP reports. However, Dr. Araj and her colleagues were able to input data from Epic’s Clarity clinical data warehouse directly into Tableau, which has the ability to conduct such text searches without using middleware.

The software tool can query UT Southwestern’s repository of more than 333,000 AP reports. It functions much like the “control F” command in Microsoft Excel, Dr. Araj says, which allows users to search for characters, text, and phrases within a spreadsheet. The tool’s ease of use and similarity to popular software programs are a boon for pathology department faculty, she adds. The platform is used by about one-third of the department, and these staff have employed it to perform a cumulative average of about 100 searches per month.

Dr. Arvisais-Anhalt

Simone Arvisais-Anhalt, MD, a former resident at UT Southwestern and coauthor (with Dr. Araj and others) of a Journal of Pathology Informatics article about the tool (doi.org/10.1016/j.jpi.2022.100014), says “this is one of the first times Tableau has been used effectively like a search engine.” Many health care organizations use Tableau as a business intelligence tool, she adds, making it easy for other institutions to replicate the project. “If you connect with your enterprisewide team that is using Tableau, it’s easy to extend the licensing and servers that are set up for Tableau to a project like this.” UT Southwestern’s quality and operational excellence group had set up a Tableau server for its own purposes, and it allowed the pathology department to allocate a small subsection for the search tool. “I say they built the ground we walked on,” Dr. Araj jokes.

To program the tool, Dr. Araj used a custom SQL script to extract from the clinical data warehouse the unstructured text from AP reports and the structured data that can be included in end users’ searches. She input both into Tableau’s data model, which operates in much the same manner as a relational database, using the software’s built-in extract, transform, and load, or ETL, functionality. She built the tool’s search capabilities within Tableau using parameters for user input and nested calculated variables, as well as the software’s regular expression functionality. The latter is a syntax built on pattern-matching strings of characters.

It took 80 hours, which included user feedback sessions, to develop the tool, plus an additional 40 hours to create data-governance documentation and a user tutorial, according to the Journal of Pathology Informatics article. Maintenance typically takes one hour per month and primarily involves adding or removing users. The search dashboard itself requires little maintenance, as the tool automatically performs a monthly ETL data refresh.

Dr. Araj

Much more difficult than the programming component was ensuring the tool would comply with the medical center’s data-governance standards. “It was something I underestimated,” Dr. Araj says. The process, which involved getting buy-in from the hospital’s security and HIPAA teams and chief medical information officer, took about four months. Only residents, fellows, and pathology department faculty can use the tool, and they must sign an agreement specifying how they will use it. Those who intend to use the tool for research must also provide an institutional review board number for a project that requires retrospective reviews of pathology reports and patient data.

Institutions interested in replicating the project should think about data governance early, Dr. Araj says. “Who’s going to use it? Is it going to be only your department? Will it be researchers? How much data will you allow them to see? Will it be aggregated? Line level?” Tableau automatically authenticates users and tracks which data are being accessed and by whom, fulfilling HIPAA compliance, but a similar tool developed from scratch would have to address user authentication, she notes.

Dr. Araj and her colleagues performed a vendor comparison and cross-benefit analysis before settling on Tableau. Though other solutions they considered had more powerful natural language processing functionality, they determined that Tableau’s simple text-search functionality was sufficient for the pathology department’s needs.

By building on Tableau, the developers of the tool were able to create an interface that allows users to enter up to three search concepts logically combined with the Boolean operator AND. For each of the search concepts, users can input up to four synonyms logically combined with the Boolean operator OR. Users can search a full pathology note or ICD-9 or ICD-10 diagnoses for each concept. The matching cases generated are displayed in the results bar, along with the patient’s medical record number, pathology case number, time the order was placed, full text of the pathology note, and a section called “lines found,” which displays the exact line of text within the report that matched the search concept provided. The “lines found” section, which arose from the initial feedback sessions, allows users to determine if a result is relevant without reading the entire report.

Tableau was also deemed a good fit for the project because it can be accessed through a Web browser, thereby avoiding the need to install or maintain software on end users’ computers. Furthermore, the user interface can be modified quickly and easily, and viewer user licenses are affordable. Speed, too, is a factor. Though it isn’t fast—Dr. Araj tells users to expect waits of about 30 seconds—it’s comparable to CoPath. “I didn’t expect it [Tableau] to scale at all,” she says. “That was one of the biggest surprises.”

Dr. Araj and her colleagues are in the early planning stages for the second version of the tool. The first step, she says, is to interview users about what they would like to see in version two. Dr. Araj, herself, hopes to incorporate into it additional discrete data elements from Epic’s clinical data warehouse. Though users can include some structured data in their searches, such as the authorizing provider and test name, they cannot include structured data from new data fields in Epic that contain such information as whether a case is malignant. Incorporating these data would allow users to quickly find all malignant cases associated with their search by checking a box and exclude benign cases from the results, she explains. Dr. Araj also wants to create a method to search synoptic reporting.

For those interested in developing a similar tool, Dr. Araj advises beginning the process by interviewing potential stakeholders. “Obtain feedback as early in the process as possible,” she says, and apply that to product versioning. “Version one should be the most slimmed down,” with a minimal amount of functionality, she notes. The goal, she adds, is to get a version in front of end users as quickly as possible to observe how they use it and how it could be improved.

After Dr. Araj developed a demo of the tool, based on how she thought it should work, she solicited feedback from Justin Bishop, MD, chief of anatomic pathology, and pathologist Jyoti Balani, MD (coauthors of the Journal of Pathology Informatics article). “We went through multiple iterations, rapid fire, for about two weeks,” she says, determining the bare minimum functionality needed for the tool to work.

CAP TODAY
X