June 13, 2007
| MINNEAPOLIS — “I’m really a closet ontologist,” says Christopher Chute, chair of biomedical informatics at the Mayo Clinic in Rochester, Minnesota, speaking in April before a crowd of data-mining experts at a meeting held by the Society for Industrial and Applied Mathematics.
Chute studies vocabularies, and for good reason. “We still cling to syndromic characterization of disease,” he says. But Chute wants medicine to work from evidence — providing earlier diagnoses — instead of waiting for symptoms to materialize. To do that though, Chute and colleagues must first win a battle of words.
Chute’s team developed the Lexical Grid, or LexGrid. Chute describes LexGrid as “a way of taking terminologies and ontologies that are widely used in clinical medicine and biological research and, first and foremost, putting those terminologies and ontologic resources into a common framework — a common model. Then, secondarily, allowing these standard methods and standard computer protocols — grid-like tools if you will — to access that content in a scalable and syncronizable way.”
Currently, medical information comes in various forms, from dictated notes about patient visits to terminologies such as the Systematized Nomenclature of Medicine (SNOMED) and the Unified Medical Language System (UMLS).
LexGrid’s main applications started with Health Level Seven (HL7), a standards organization aimed at developing better ways to exchange, manage, and integrate healthcare information. In drawing from various vocabularies, HL7 hit a terminology problem. Says Chute: “You start seeing very quickly that all of these terminologies look different, feel different, and you want to use them in an interoperable, real-time service system.” LexGrid links these terminologies and lets clinicians and researchers work with the information.
“Let’s be clear,” Chute says. “LexGrid does not make semantic inferences. It stores inferences made by others — for example, in the UMLS community — in a format that’s readily usable. So, we’re more or less the pipe and wire builders, rather than the great thinkers.” LexGrid is “pure infrastructure,” says Chute, “but it’s infrastructure that’s scalable and usable and meets the use-case needs of the biology and medicine community — and arguably other communities — in an open-source tool.”
James Buntrock, unit head of IT at Mayo, explains: “We take a variety of imports, including ontology web language [OWL], represent them in a common model, and then have a set of services access, search, and navigate vocabulary concepts and relationships via a common API [application programming interface]. LexGrid provides implementations of HL7 Common Terminology Services API and Lex-specific APIs to access vocabularies in the common model.” He also points out, “You can add in things like reasoning and inferencing.” So far, though, the Mayo team has steered clear of these higher-level tasks. Just communicating between terminologies is challenging enough.
On the clinical side, much information comes arranged in common document architecture (CDA), which is a patient form. It includes boxes for the history of the present illness, past medical history, physical examination, and so on. Much of it is free text — a physician’s description of the patient’s visit. To address what the words in each box mean, says Guergana Savova, a medical informatics researcher at Mayo, “We want to process that data in a systematic way and make it semantically unified... We want to map whatever information is found in that data to an ontological classification that is a standard.” Savova uses natural language processing to encode the data, then information can be searched to find patients with a partic ular disease and see how treatments affected the outcome.
So far, one of the most significant uses of LexGrid involves the National Cancer Institute’s caBIG (cancer bioinformatics grid), which uses some of LexGrid’s infrastructure (See “Putting LexGrid to Work.”) In essence, caBIG makes clinical trials and biological data interoperable across all cancer centers. The information might be entered as OWL, for example, but working with that information requires additional data modifications.
“The semantic-web resources are woefully underspecified to serve these kinds of use cases,” says Chute. “OWL is kind of the aluminum foil, and LexGrid is the wire frame. It doesn’t have a shape or a substance until you wrap the aluminum foil around a frame and specify what the model looks like.” In essence, he’s complementing the Semantic Web. He says, “We see ourselves adding considerable value and utility to the Semantic Web’s very — almost too-generalized — tools.” In particular, the tool for caBIG is called LexBIG. Simply speaking, LexBIG lets caBIG access LexGrid capabilities. For example, it works with vocabularies in UMLS Rich Release Format (RDF) and OWL.
Perhaps LexGrid’s greatest asset is its ability to connect the basic and applied research communities. This is just how LexGrid is being used by the National Center for Biomedical Ontology (NCBO). There, LexGrid can bring together terminologies such as Gene Ontology (GO) and SNOMED.
In the end, Chute knows that science must bridge the gap between the lab to the clinic and back, finding ways to use medical information and knowledge from model organisms in the same system. “Biology and medicine are moving toward a big-science paradigm,” says Chute. “So LexGrid becomes one of the enabling infrastructures to let this happen.”
LexGrid is available online, fully open source and downloadable.
Sidebar: Putting LexGrid to Work
LexGrid applications already exist beyond caBIG. For example, LexGrid is being used in the Public Health Information Network (PHIN), operated by the Centers for Disease Control & Prevention (CDC). The PHIN website describes this network as “CDC’s vision for advancing fully capable and interoperable information systems in the many organizations that participate in public health.”
The World Health Organization’s International Classification of Diseases (ICD) will soon be adding LexGrid. Chute is chairing the WHO revision for the International Classification of Disease, which will use LexGrid infrastructure “to build that context for the next century.”
Chute envisions even broader applications of LexGrid. “The fantasy, if you carry it to the limit,” he says, “is that every many, woman, and child on the planet would have every health encounter that they engage in stored as a structured representation of that event, so that you’d be able to ask useful questions across nine billion people.” If that information was available, a database could be searched for a patient’s specific condition to see if anyone had ever experienced the same thing. “If we’ve seen it,” Chute asks, “what does it mean to be a patient like that?”
Answering that question demands not only that all the data are collected, but that they also get stored in interoperable ways. “You need concordance between the way you describe the patient and the way the data are stored,” Chute says. “None of that is true right now, but we can dream.” Instead of turning to data, Chute says that physicians often turn to colleagues instead — simply asking if they’ve ever seen a certain type of case. “It’s imperfect human recall,” Chute says.
But if the applications of LexGrid continue to grow, it could eventually be possible to turn medicine from anecdotes to data-driven diagnoses. -- M.M.
|LexGrid connects incoming data in many forms -- including RDF and OWL -- and makes that information searchable and accessible to various services, such as Java and .Net.|
Subscribe to Bio-IT World magazine.