RadLex: The Plumbing that Enables Practice

Like so many elves—plumbers is the analogy used by Daniel L. Rubin, MD—at work, a few dozen radiologists took a terminology for coding procedures created by the ACR for teaching purposes and developed it over the past decade into not just a common language but a complete ontology for the specialty of radiology: RadLex®.

Complicated to the degree that any system would be that intends to encompass not just the language but also the metaphysics of the specialty, RadLex celebrated its 10th anniversary at last year’s meeting of the Radiological Society of North America (RSNA) and is poised to become a key piece of the infrastructure of value creation in radiology.

RadLex was spearheaded by the RSNA and involved more than 30 specialty societies, including the ACR. Dozens of radiologists have participated in and contributed to its development, but RadLex is governed by a steering committee first helmed by Curtis Langlotz, MD, PhD, recently succeeded by Rubin, and an assortment of organ system and modality committees.

Rubin, associate professor of radiology and medicine, Stanford University School of Medicine, and Charles E. Kahn, Jr, MD, MS, professor and vice chairman of radiology, University of Pennsylvania, provided an overview of RadLex, its parts and its applications at the 2014 meeting of the RSNA in Chicago, December 2, 2014.  At the outset, Rubin described RadLex as plumbing that enables radiology practice, helping radiologists to exploit its information sources for decision support, reporting, data analysis, teaching and participation in quality initiatives.

“Daniel’s plumbing analogy is very apropos, because when plumbing works, we don’t even think about it,” Kahn noted.

Analogies aside, RadLex is a machine-searchable controlled terminology. Rubin underscored the need for a common language in the specialty by reeling off some of the 16 different terms for interstitial edema/infiltrate, adding that the same variation exists in procedure names. Without the tagging mechanism that a controlled terminology provides, synonymous terms cannot be unified.

Controlled terminologies

Half a dozen computer-searchable medical lexicons are pertinent to radiology, but none of those encompass the complete radiological nomenclature, Rubin observed. The Systematized Nomenclature of Medicine (SNOMED) and the Logical Observation Identifiers Names and Codes (LOINC) have gained importance for all providers because both are required in the federal meaningful use program. SNOMED, the International Classification of Diseases (ICD) and even the Unified Medical Language System (UMLS) include the radiology domain, but mainly as they relate to disease—not at all in terms of radiology observations and incompletely in terms of radiology procedures. Hence, the need, recognized more than 10 years ago.1

RadLex both supplements and links to other lexicons. Its main role, however, is filling gaps that other lexicons have when it comes to radiology.  A comprehensive list of terms and relationships that describe imaging, it is composed of a “Playbook” of imaging methods; and an exhaustive list of indications, findings, anatomy and diagnoses that encompass imaging results.

The Playbook—important for billing and registry participation—

provides a standard nomenclature for naming radiological procedures and procedure steps. Each term has an identifier—an RPID, for rich presence information data—based on a “post-coordinated” grammar that maps to terms in the broader RadLex lexicon, each of which is associated with an RID, or resource identifier.

The Playbook grammar for CT, for instance, includes: [modality], [modality modifier], [body region], [anatomic focus], [population], [laterality], [reason for exam], [projection], [contrast]. Grammar components are optional, allowing for long or short names, and feeding the hierarchy of terms and term subtypes.

The main consumer of this is computer applications that can look up a term, tell whether it is a preferred RadLex or ad hoc, synonymous term and associate it with the proper RPID. “It’s more than just a lexicon,” Rubin said of RadLex. “It has this knowledge about other variants of preferred terminology. What’s important about this is that this structure allows a great deal of flexibility in computer applications.”

How it works

When Playbook was initially built, the Playbook committee assembled a collection of chargemasters from hospitals, fed the procedure names into a spreadsheet and built Playbook from the bottom up. All terms providers use for medical imaging procedures and procedure steps were mapped to a preferred RadLex Playbook term.

Both the University of Pennsylvania and Stanford University are in the process of migrating their chargemasters to the Playbook terms.  “[It’s] a matter of adding, to whatever IT system you have, another column,” Rubin explained. Current IT systems may have columns for the exam code, the CPT code, the exam description, and how much you bill for them.

By adding an RPID column, you can associate your current procedure name with the Playbook term, passing it throughout your information system. “That will enable applications to transmit this identifier instead of your ad hoc term to, for instance, the ACR Dose Registry, which now can recognize this as a standard name,” Rubin explained.

“One of the remarkable, successful applications of RadLex has been in the ACR Dose Registry,” Kahn said. “There are institutions around the country that use this as part of their quality management reporting: They upload their dose index information to the ACR registry and receive reports that show how dose index at your institution compares to other institution. Central to that is comparing apples to apples. RadLex gives that level of detail and harmonizes the data.”

“Academic hospitals, community hospitals and imaging centers are consolidating into health systems and they all have their different chargemasters,” Kahn added. “Moving to a common, nationally accepted set of procedure names helps avoid some of the political battles.”

Penn is using RadLex to harmonize all procedure names across what are now seven different institutions that are part of the network, Kahn reported. “Again, this is plumbing,” he said. “Unless you have an interest in working with the details, you simply will benefit by the investment made by RSNA and a lot of people.”

The findings piece

RadLex terms also include the observations radiologists make and the associated anatomical entities, organized into hierarchies. Through these and other relationships, RadLex is transformed from terminology to ontology.

“You see more specific terms are subsumed underneath less specific terms and this becomes important for search, to do query expansions along different axes of the ontology,” Rubin explained.  “If my teaching files are annotated by anatomy, and I want to find all of the images of the lungs, then I can use RadLex to do the query expansion to find images that will, for example, be annotated as ‘right upper lobe’ or ‘lingual.’ RadLex is more than a lexicon, it is a controlled terminology with definitions and dimensions.”

Rubin said the bigger vision for RadLex is to use it for annotating radiology data, such as images. By associating whatever the radiologist observes or does while dictating a report, and using RadLex terms to describe this, including image annotations, then everything in the report is query-able.

“With RadLex you can type a term, mucinous cystic neoplasm or serous cystadenoma, and RadLex will help you recognize that those are both types of cystic pancreatic neoplasm,” Rubin explained.  “That’s an example of query expansion. Also, RadLex has the synonymous relationships and those all get unified to a common term.”

He underscored the criticality of RadLex for quality assurance activities, including business intelligence. “If you want to look at utilization of radiology, you’ve got to have standardized ways of talking about that,” he said. “Order entry decision support is all about linking what the referring physician has ordered with the indication—the clinical context—to an appropriateness score. You’ve got to have a controlled terminology for what’s being ordered, otherwise you can’t execute an order entry decision-support application.”

Initially, radiology decision-support vendors used their own terms for radiology procedures, but as they adopt RadLex, they are replacing their internal names with RadLex Playbook RPIDs, Rubin noted. “Adopting Playbook is all about enabling interoperability, but the core infrastructure is the same,” he said. “We are just replacing an ad hoc name with a standardized name.”

Kahn identified several other resources that utilize the RadLex ontology and are freely available.

The RSNA Radiology Reporting Initiative. More than 200 report templates—downloaded more than 2.4 million times—have been developed and are intended as best practice examples to be used and adapted to local practice patterns. To illustrate the challenges associated with the traditional prose report, Kahn offered the example of a renal cell carcinoma follow-up study.

To understand what’s happening, the radiologist must read the report and try to find the measurement to know that the index lesion in the lung has grown from 12 mm to 17 mm and then extract the information. “The reporting initiative has created these templates whereby information can be plugged in, you have discrete fields and you can find values more easily,” he said.

Under the hood, Kahn said, the various elements in the reports can be linked to RadLex terms, improving the ability to retrieve the information from the report. Profiles utilizing RadLex RPIDs also have been developed within the Integrating the Healthcare Enterprise initiative, enabling coded information to flow out into the EMR where it can be data mined and intermingled with other information.

Imaging Signs. While the term pneumoperitoneum can be found in 11 of the 201 vocabularies in the BioPortal at the National Center for Biomedical Ontology, the names used to describe many of these findings—the Rigler sign and the football sign—were nowhere to be found. Matthew Shore, MD, (now a radiology resident) initiated a project as a first-year medical student to add these visual metaphors to RadLex, numbering 743 signs and 392 synonyms. “There is now a hierarchy of imaging signs, related to modalities and conditions,” Kahn said.

Imaging Gamuts. Kahn initiated a project to catalog radiology gamuts taken from classic gamuts textbooks of differential diagnoses and, using RadLex terms and the ARRS Goldminer search tool, pull in a set of images. “It is a huge knowledge model,” Kahn said. “It has more than 16,000 entities that represent imaging observations, disease observations, all linked by causal relationships and hierarchical relationships.”

RadLex continues to evolve and grow. A current project involves mapping between CPT codes and Playbook, which will make it easier to develop billing systems with native CPT codes. There also are plans to harmonize RadLex with LOINC, the lexicon for laboratory tests. “This is important, because LOINC is one of the meaningful use terminologies, and this is the path by which RadLex will become part of that ecosystem,” Rubin said.

Reference:
• Langlotz CP, Caldwell SA. The completeness of existing lexicons for representing radiology report information.  J Digit Imaging. 2002; 15(1S):201-5

Resources: ima
• RadLex - RadLex.org
• Playbook - Playbook.RadLex.org
• Radiology Gamuts - Gamuts.net
• RSNA Radiology Reporting Initiative - RadReport.org