The rise of electronic health records with natural language processing technology is transforming provider workflow and clinical documentation.
Though NLP is not without its challenges, it can offer valuable benefits when used wisely, said Anupam Goel, vice president of clinical information at Chicago-based Advocate Health Care.
While NLP can offer advanced diagnostic benefits, it depends heavily on the specifics of how clinicians enter their documentation, Goel said this week at the Healthcare IT News and HIMSS Big Data and Healthcare Analytics Forum in San Francisco.
The moment is ripe for NLP-enabled charting, he said. The technology has sufficiently evolved to be useful rather than counterproductive, and the benefits – ease-of-use, a shorter window between clinical documentation and the activation of care teams – are clear.
Goel said he expects more clinicians will be using voice commands to log data rather than filling out forms. NLP is useful for identifying clinical gaps, he said, and can help organizations reduce labor costs.
That said, “there are lots of ways NLP can screw up,” he said.
NLP enables misinterpretations related to both content (the acronym “SOB” has a different denotation in a clinical setting (shortness of breath) than it does in common parlance, for instance) and context (it could miss crucial distinctions between family history and personal history in a patient’s EHR).
“This is the reality of NLP,” said Goel. “If you can set expectations low, you’ll be better positioned for success in the future.”
NLP is an algorithm, and it depends intrinsically on the input and corrections it gets from its users to improve over time, he said. “It’s not perfect out of the gate. It’s a tool that is going to evolve and get better as you feed it more information.”
Vendors of NLP technology are focusing most these days on clinical documentation improvement for more complex diagnoses, real-time feedback for better-quality charting and improved billing accuracy.
“That’s all got great value,” said Goel.
A more complex deployment of NLP – but the one with the “most potential for healthcare,” he said – is to link it with rules engines to help with decision support.
Doing so can enable providers to more easily flag specific data for review, route certain assignments to condition-based worklists or suggest certain treatment courses.
But getting it right is not necessarily easy. And it depends heavily on physicians and clinical staff following best practices.
The risks for NLP from a clinician’s perspective include “too little information,” said Goel.
The more data for the technology to make use of, the better. Minimal documentation limits the algorithm’s ability to separate “wanted” from “unwanted” cases, he said, adding that cases that are difficult for humans to distinguish will likewise be hard for NLP programs to make sense of.
As he highlighted NLP systems such as R+OpenNLP and Apache cTAKES, Goel said natural language algorithms hold big promise for targeted case-finding and streamlined clinical efficiencies.
He suggested beginners bolster their NLP proficiency by first deploying it in “low-risk scenarios,” that could help build up confidence in the tools while avoiding the risk of adverse events.
He also cautioned that smart integration of NLP with clinicians’ workflow is critical to realizing any productivity gains.
“If you don’t integrate it with workflow, you’re just creating more work without more value.”