Newsbytes

Create PDF

Homegrown search platform addresses the need for speed

May 2021–Laboratory information systems contain a wealth of diagnostic patient case data, but accessing it promptly can be difficult if not impossible—without a workaround, that is.

While LISs are searchable, they don’t necessarily offer the type of instantaneous search functionality that would allow pathologists to quickly compare current cases to past cases, explains Scott Robertson, MD, PhD, a pathologist at Cleveland Clinic. “In my experience in the places I’ve worked, it has always been that way,” he says. Therefore, Dr. Robertson built an inexpensive search platform for the Cleveland Clinic, which could be replicated by other institutions.

Dr. Robertson

The first key step in such a project, Dr. Robertson says, is to move a copy of relevant LIS data to a separate database that is not restricted by vendor-specific tools or protocols. “You need your data where you can access it in a programmatic way outside of the production database,” he explains. “What I mean is, you have to be very careful on the database that is running your day-to-day business because there is a risk that the searches will stress the system. Once you have a copy of your data outside of that, you can do what you want with it and there’s no risk to your clinical operations.”

The search platform that Dr. Robertson constructed, called Pathtools, uses a standalone SQL server database. Cleveland Clinic allocated space for the platform on a SQL server hosted in its secure data center. Pathtools users at Cleveland Clinic can now search more than 4 million anatomic pathology cases dating back up to four decades and receive search results in seconds. (An automated process via Linux uploads new case information from the LIS into Pathtools every 24 hours.) In the past, Dr. Robertson says, a small team of analysts conducted all data searches, carefully managing the load on the production database to minimize negative effects on routine operations. “The analyst-centric workflow was slow,” he adds, “and it took more than 24 hours to return results to the requester.”

Dr. Robertson designed the Web-based graphical user interface for Pathtools and set up the database in his spare time over the course of two years. However, some medical institutions may have to hire a software developer on a project or part-time basis, he says. Because LIS data includes protected health information, the Web interface for a Pathtools-type platform would need to be built using a professional-grade Web framework and include extensive security features, Dr. Robertson explains. Therefore, he chose Django, a Web framework used by major corporations. Django is open source and based on the Python programming language, with which he is familiar.

For any project that affects how data is accessed, it is essential to collaborate with institutional oversight teams to ensure appropriate data protections are in place, Dr. Robertson says. He worked closely with Cleveland Clinic’s institutional review board and the clinic’s legal experts to ensure the platform met the medical center’s data governance standards.

To safeguard data, Pathtools has a three-step registration process. Users must complete a registration form online and then click an email confirmation sent to their institutional email address. The Pathtools administrator, who in this case is Dr. Robertson, must then approve the registration form to grant access. Only select pathology department staff, such as pathology residents and fellows, and clinic researchers who have been vetted are allowed to use the system. This last registration step allows Dr. Robertson to verify that only approved personnel gain access.

Security considerations also influenced how Pathtools’ search functionality was structured, resulting in two search modes with different degrees of data access. “From a data governance perspective, you want to give people the least amount of information needed for their purpose,” Dr. Robertson says. Therefore, the default, or preview, mode returns search results with case numbers and diagnostic data and no possibility of accessing protected health information. Preview mode searches also return only the most recent 100 cases, in reverse chronological order, and are intended for users who want to run quick searches during the workday. Cleveland Clinic, Dr. Robertson notes, is in the process of transitioning to a new LIS that takes more than five minutes to get results for searches that, on average, take less than a second using Pathtools.

Approximately 75 percent of searches in preview mode are conducted by pathology residents and fellows who want to compare past and present cases, Dr. Robertson says. The remaining 25 percent are primarily conducted by attending pathologists seeking case information for teaching, research, and presentation purposes.

Residents typically use Pathtools to look up factors considered in past cases when they preview current cases as part of an anatomic pathology training process that requires trainees to “preview” or write preliminary diagnoses that are reviewed and revised by attending pathologists. Before Pathtools, Dr. Robertson explains, there was no way to access this information, so writing preview diagnoses was more of a trial-and-error undertaking.

In a Pathtools study conducted by Dr. Robertson and published in the Journal of Pathology Informatics (doi.org/​10.4103/jpi.jpi_43_20), 18 trainees were surveyed and all “strongly agreed” or “agreed” that Pathtools helped them write diagnoses that required less editing by attending pathologists.

For pathologists who want access to more detailed information, Dr. Robertson created the multifaceted research mode in Pathtools. The initial search results in research mode are similar to those in preview mode, except that in addition to returning 100 case results, this functionality provides a count of the total number of cases that meet the search criteria. “When you are starting a research project,” Dr. Robertson says, “you want to be able to quickly get an idea of the scope of the project and its feasibility. For pathology projects, a lot of that centers around [determining] how many cases of diagnosis X we have at Cleveland Clinic and that exist in our archive.”

Research mode also allows users to store their searches on a portal page and begin formal requests for more detailed data, including sensitive, protected health information. All formal data requests are routed to analysts who review the requests, perform the data downloads, and, for requests that are approved, send the data file to the user. These analysts also fulfill the requests for data from the LIS. However, unlike with LIS data requests that can take 24 hours to fulfill, Pathtools search requests are usually returned in an hour or two, Dr. Robertson says. Pathtools has eased analysts’ workloads, he adds, because basic data searches can be run without their involvement and detailed searches for potentially sensitive information can be processed more quickly.

Despite its benefits, Dr. Robertson acknowledges that Pathtools isn’t perfect. One problem is that it is very literal. Therefore, a database search for “Crohn’s disease” would not capture “Crohns disease” if it were spelled without the apostrophe. Natural language technology could correct such an issue, he says, but it is not yet built into the system.

More importantly, although Pathtools contains approximately 4.3 million cases, it is not a complete copy of Cleveland Clinic’s LIS, Dr. Robertson says. Some cases cannot be imported into the Pathtools SQL server because of details such as odd text characters. Between one and two percent of more current cases cannot be transferred to Pathtools, and the percentage of older cases that cannot be transferred is higher, he notes.

However, Pathtools’ incomplete data set may soon be remedied by a data warehousing project underway at Cleveland Clinic. The medical center is organizing data from across the institution into a large centralized data repository that operates independent of the clinic’s vendor-specific systems, Dr. Robertson says. Pathology department data were expected to be loaded into the data warehouse by CAP TODAY press time.

“Once that has occurred,” Dr. Robertson says, “instead of searching my own SQL database that I built, I can point Pathtools to search the data warehouse, which will contain a much more comprehensive library of data.” —Renee Caruthers

IICC announces publication of new LIVD specification

The IVD Industry Connectivity Consortium has released LOINC to Vendor IVD (LIVD) Specification V2.0.