Unlock Patient Data With Natural Language Understanding
Compiled by Susan Kreimer, contributing editor
Natural language understanding (NLU) technology is helping Lancaster General Health improve physician documentation, prepare for ICD-10, and translate patient notes into discrete data measurable by analytics.
Voice recognition technology has evolved to become much more than a speech-to-text dictation tool. Today’s voice-driven solutions can understand context, prompt physicians for clarification, and analyze unstructured data contained in a patient note. These capabilities are a powerful clinical resource for physicians, contributing to a more complete patient record and, ultimately, improving care.
With locations throughout southeastern PA, Lancaster General Health is one provider that understands the benefits this new breed of voice recognition can offer. The health system currently leverages the M*Modal Fluency Natural Language Understanding solutions suite. In this Q&A, Gary Davidson, senior vice president and CIO at Lancaster General Health, discusses the health system’s experiences with the technology to date.
Q: How are you currently leveraging voice recognition/NLU technology at Lancaster General?
Davidson: Today, we utilize both front end and back end speech recognition tools. Running through back end speech speeds up the process and makes it more efficient for the transcriptionist. The new method (front end speech) is geared more toward entering documentation into the system in real time. Physicians utilize this technology to enter information into the system during the clinical documentation process. They use templates, as well as free speech, rather than traditional transcription, for which the turnaround can take anywhere from 1 to 24 hours. We have more than 1,100 copies of front end speech in use at Lancaster General. We also offer the ability for physicians to use traditional transcription. At this point, slightly more than 10 percent of documents go into the system using the traditional format.
Q: How has your use of voice recognition/ NLU technology evolved over the years?
Davidson: Five years ago, we were using a more traditional form of hospital transcription. After listening to physicians’ dictations, a transcriptionist would complete the notes and send them back to the doctors for their signatures. Physicians’ involvement was limited to that. About three years ago, we moved to utilizing back end speech translation. This increased the efficiency of our transcriptionists, enabling them to convert voice clips into digital electronic media at a faster pace. However, that was an interim step.
In the past year, we have made a major push toward having physicians use front end speech. The templates help capture discrete values and ensure that physicians complete all necessary documentation. Physicians’ notes are completed as part of that process, and they no longer have to wait for a typed version to come back for their review. Instead, they’re able to access a document and make their own corrections. Physicians also can use partial dictation in lieu of full-note dictation. This enables them to dictate shorter paragraphs around structured items previously documented in the EMR such as medications, allergies, and a patient’s medical history; then, a transcriptionist types those short paragraphs, and they are automatically inserted into the final note. We found that for complex cases, or for certain specialties, a narrative is much better than an entire note based on a standard template. The standard template works for the majority of cases, but some patients are more complex, and a narrative is a more effective way to convey what’s going on with them. That’s why we have both options available to our physicians.
Q: How has NLU technology advanced beyond simple voice-to-text dictation?
Davidson: NLU now has the ability to review what a physician has put into text, to analyze the data, and to compare it against other structured and unstructured documentation in the EMR. Then NLU can prompt physicians in real time for clarification of what they have documented. For example, if they are documenting congestive heart failure (CHF), then NLU can request that the physician be more specific. Is it acute CHF, acute on chronic CHF, or systolic heart failure? And what’s really powerful is that NLU can do this in real time during the dictation, when it’s fresh in the provider’s mind.
This newly created information can be used to drive physician alerts and rules, thereby improving the quality of care. There’s no better way to get clarification than right there on the computer. With our old method, we would query physicians after the fact, when they moved on to other patients, and physicians would have to refamiliarize themselves with the records. The new method is much more efficient. In addition, it allows for more accurate patient documentation in the system in preparation for ICD-10 and ACO (accountable care organization) requirements, and it helps the hospital receive appropriate reimbursement based on the care provided. NLU also can examine large sets of data such as old transcribed reports. The system can scan through this nondiscrete data and extract selective values to create discrete values for use in the EMR. This would also help with quickly adding structured data, such as medications, allergies, or other concerns, from a practice that is converting its documentation process from transcription to EHR.
Q: How important is the clinical narrative to the patient record, and what role should NLU technology play in its documentation?
Davidson: The clinical narrative is much more than just words in a paragraph. NLU can understand them and put them into context by searching other databases and systems. A lot of NLU translation occurs while a physician is documenting. And we want to make sure the record is complete and accurate on the spot. That’s something you can’t accomplish without this type of technology. This innovation also can help with ICD-10. With almost 70,000 codes, NLU will be able to prompt physicians to clarify exactly which code they need in real time. This ensures that physicians are documenting appropriately. It’s a change and a learning process for them, too. However, in the long run, this streamlines the process for physicians and leads to more complete medical records for patients.
Q: What benefits has Lancaster General experienced from its use of NLU?
Davidson: NLU has helped to significantly reduce our transcription costs as an organization, saving us more than $2 million, and thereby lowering the expense of healthcare delivery. The other advantage is timeliness of records. Turnaround time is now instantaneous. We can convert voice into discrete values that can be run against analytics, or we can analyze things that have been transcribed and pick out discrete values, so that we can run analytics against it later and understand more about a patient’s condition. Paper records or physicians’ notes today contain an untapped wealth of material, but you can’t do anything with them. We’re hoping to apply NLU and turn this wealth of data into information.