We all remember when IBM’s Watson beat the two all time grand champions of Jeopardy (if you don’t then see the video above). However, as Watson moves into various industries we wonder what Watson would look like in the healthcare industry. Today we got the answer to that question from Randall Wilcox of IBM at a HIMSS sponsored webinar entitled “IBM Watson and Medical Record Text Analytics”.
Watson within the healthcare industry is focused on diagnostic decision support. Errors in diagnosis often occur due to not considering all possibilities. In order to practice medicine at the highest possible level, doctors need to know everything that is known today. If you consider the amount of clinical data available, then this is a hefty task. Watson’s value proposition leans upon providing a physician with the ability to ask questions in natural language and have a weighted decision back within a short timeframe. Given that over 80% of information is unstructured and based on natural language, legacy “search” methods have failed. Watson’s solution is based on natural language processing, machine learning, and data mining.
Watson does this through extracting keywords and normalizing with medical terminology. Watson then uses a series of algorithms to generate a hypothesis around a particular diagnosis or medical question. A subset of viable answers is then generated. Each answer then goes through a “massively parallel” deep evidence scoring. For example, if one possible disease is caused by an insect, but the patient has not visited or does not live in a location where that insect is found, then that disease can be routed out as a candidate based on the hypothesis. In the end, Watson provides a competence weighted differential diagnosis. This process is shown below in where natural language is processed and results in a diagnosis of UTI. Watson also provides a helpful flashing red button to help you find the correct ICD code.
However, disease diagnosis is not the only use for Watson in healthcare. Watson can be used for quality reporting, revenue cycle management, ACO support, and appeals and grievances on the payor side. This is done by converting information from notes and lab results into a structured format to be analyzed and reviewed through 1) automated data collection, which frees up nurse time 2) a clinical dashboard that provides the most relevant patient information, and 3) a performance dashboard to better enforce standards of care by providing information against core measures, etc.
Last but not least, we were introduced to Watson’s mobile capabilities that allows physicians to use dictation to stream audio to text that can be sent to a content analytics platform to be codified.
