Issue link: https://viewer.e-digitaledition.com/i/288786
By Jennifer Webster NATURAL LANGUAGE PROCESSING (NLP), THE TECHNOLOGY OF CONVERTING HUMAN DISCOURSE INTO STRUCTURED DATA, PROMISES TO UNLOCK A WEALTH OF INSIGHTS INTO THE PRACTICE AND BUSINESS OF MEDICINE. HAS THIS PROMISE BEEN FULFILLED? M*Modal collaborate with a variety of vendors to enhance EHR functionality. "We work with vendors such as Epic and Cerner to incorporate NLP into their EHR," says Juergen Fritsch, MSc, PhD, Chief Scientist with M*Modal. Fritsch predicts an ongoing incorporation of NLP capabilities into EHR as EHR systems continue to add functionality. "They are not developing [NLP] them- selves, but integ rat ing t h ird-pa r t y platforms," he says. "[These platforms are] mostly used for data mining and analytics." Following the Money In the healthcare fi eld, NLP became most successful in revenue cycle manage- ment support. In coding and billing, computer-assisted coding (CAC) relies on a medical coding ontology to fi nd words within a document and relate them to billable diagnoses and procedures. After abstracting a term and determining its relevance, CAC software assigns a billing code to it. "This is the part of the fi eld where NLP has had the greatest impact so far," Fritsch says. "Suggesting certain codes speeds up the process." Hastening the laborious process of hand coding isn't the only application of revenue- cycle CAC. Healthcare organizations hope clinical notes and other unstructured texts will reveal insights to identify ineffi ciencies and create promising workfl ow solutions. M*Modal, for instance, offers a robust array of products ranging from automated transcription to NLP systems that populate EHRs from dictation — saving physicians valuable time. More advanced capabilities include software that identifies and prompts for missing information in clinical documentation, as well as capabilities for sharing information easily throughout an organization. From Business to Medicine Watson's publicity promised more than enhanced billing via NLP, however. St r uct ur ing more infor mat ion a nd ma king it searchable was meant to improve health care, too. In places, this is approaching reality. For exa mple, t he Depa r tment of Veterans Affairs (VA) has begun using NLP to identify references to previous suicide attempts in clinical narrative. Working with the University of Washington, the VA Puget Sound Health Care System created an automated program for searching free text that returned 80 percent accurate r e s u lt s i n t r i a l s , c omp a r e d w it h manual scoring. New companies in the healthcare NLP field offer innovative uses for mining unstructured discourse. Linguamatics, for instance, an NLP company since 2001, recently moved into healthcare IT, offering software that extracts information, per- forms predictive modeling and links external sources with organizational healthcare data. "[Some of our functionality] runs pro- grammatically at regular intervals for compliance models," says Simon Beaulah, Director of Hea lthca re Strateg y at Linguamatics. "In others, you would do a batch retrieval, continuing someone else's data with new information." Real-time software applications that scan physician notes and nudge for compli- ance factors, such as prescribing medication or scheduling a test, have the potential to improve patient care by prompting "the right treatment at the right time," Beaulah says. Another function of Linguamatics Health delves into chart data for the pur- pose of making comparisons with external sources, such as clinical trials. "We take patient characteristics … and match exclusion/inclusion criteria, genetic aberrations, age, gender ... and bring back a set of matches," Beaulah says. "That could be the next wave of NLP applications … providing very tailored, precise results in a specifi c context." Does your organization use NLP? If so, how useful have you found it? Join the discussion at mdnews.com. ■ Natural An Ongoing Narrative Language Processing: At present, IBM and MD Anderson Cancer Center are collaborating on the problem of childhood leukemia, using Watson's capabilities for digesting large amounts of data and generating hypotheses for valuable assistance. What other roles is NLP playing today? The Transformation of Data to Improve Care Health information management profes- sionals seek to gather, organize and analyze healthcare data in order to improve patient care. In this paradigm, physician narratives represent an unlit treasure cave, full of nuanced information about the patient's history and the physician's thought pro- cess — so far unavailable as searchable data with clearly defi ned categories, meanings or relationships. In the realm of data governance, NLP makes up part of the data enrich- ment process, which takes place after aggregating and before storing data. By translating free narrative into diagnos- t i c o r p r o c e d u r a l c o d e s , N L P streamlines important routine tasks, such as ensuring patient encounters are f ully billed. Data enrichment a lso makes possible predictive modeling and other forms of analytics in the service of organizational objectives. Who's Providing NLP? An NLP vendor may offer services ranging from transcription to a variety of business analytics packages. At its most basic, NLP may take physician narrative — whether dictation or typed notes — a nd map it into a n EHR . Traditional dictation software vendors, such as StenTel and M*Modal, offer a spectrum of language transformation services, from automated transcription to machine language transformation and deep-delving analysis. Some EHR vendors are working with NLP providers to add that capability into their offerings, easing the burden on c u stomers. Compa n ies such a s IN 2011 AND 2012, healthcare informa- tion technolog y publications brimmed with excitement over IBM's Watson, which was prom- ised to use NLP combined with predictive analytics to make care recommendations for patients by comparing them against large data sets. NLP gave Watson access to a far wider and more nuanced range of information — in the form of clinicians' freeform notes — than would be available through the discrete text fi elds of electronic health records (EHRs). CORE CORE