HIMSS Webinar: IBMs Watson to bring NLP to healthcare
 |
In February, Watson made its initial splash on the Jeopardy show, besting a pair of champions. “Jeopardy provided a nice platform to [showcase natural language processing],” explained Randall Wilcox, senior managing consultant of IBM.
The show provided Watson the opportunity to demonstrate its broad domain of knowledge, understanding of the nuances of question categories, ability to formulate a game strategy and estimate a confidence level about its knowledge.
However, Watson’s destiny extends beyond game shows.
IBM expects to transfer the technology to healthcare in the next 12 to 18 months, shared Wilcox. One of the first use cases of the technology will be diagnostic assistance.
Diagnostic errors often come from not considering all of the possibilities, shared Herbert Chase, MD, professor of clinical medicine in biomedical informatics at Columbia University School of Medicine in New York City. “To practice medicine at its highest level we need to know what is known today. This is the holy grail.”
With Watson, physicians can access up to date information that will inform the patient’s diagnosis and therapeutic plan, enabling physicians to practice state of the art medicine on any given day, explained Chase.
The key is translating unstructured data from medical literature, guidelines and patient notes into structured elements and employing algorithms to present answers to physicians.
Watson uses parallel evidence-based architecture to deconstruct questions into key words that are inputted into algorithms to generate hypotheses and generate viable candidate answers, which are scored via evidence and weighted to present the physician with a differential diagnosis.
The first commercial outgrowth of Watson is likely to be Medical Records Text Analytics Solution, which incorporates three components: content analytics, health language medical terminology management and IBM industry solution services healthcare annotator assets.
Wilcox presented a use case of Medical Records Text Analytics Solution: a patient who presents to the emergency department with chest pain and symptoms of acute coronary syndrome. After consulting with the cardiologist on call, the emergency department physician admits the patient and sends him for a nuclear stress test which reveals blocked arteries and results in bypass surgery. Although the patient is diagnosed with a post-surgical urinary tract infection, he recovers and is discharged to cardiac rehabilitation.
At the one week follow-up visit, the system extracts data from thousands of documents to present the cardiologist a dashboard view of test results, diagnoses, medication lists, procedures, allergies and alerts regarding deviations from standards of care. The cardiologist can use the dashboard to inform decision making.
The same process applies to a performance dashboard that can be used by hospital administrators to inform administrative decision making, continued Wilcox.
The system, concluded Wilcox, could enable hospitals to achieve four objectives. With these capabilities, hospitals can automate data collection, improve productivity, provide dashboard views and enforce standards of care.