April 25, 2011
Natural Language Processing
By Lisa A. Eramo
For The Record
Vol. 23 No. 8 P. 20
By using natural language processing to dig deeper into documentation, healthcare organizations can analyze patient narrative to enhance care.
As part of meaningful use, hospitals must demonstrate that they can capture discrete data elements in an EHR, but how do they accomplish this when much of the information is tucked away in the narrative portion of the medical record?
Experts say the answer to this question may lie in the functionality of a familiar technology: natural language processing (NLP). Unlike speech recognition, which simply converts spoken words into digital text, NLP infers meaning behind the words.
“The technology tags and parses every word,” says Chris Cashwell, senior vice president of client development for Webmedx, which offers a product that combines intelligent speech and clinical analytics technology with traditional dictation and transcription services. “Every word can be put into a discrete format that can be used for reporting capabilities.”
As part of its tagging process, the technology can recognize related words and phrases, says Scott D. Faulkner, principal and CEO of InterFix, LLC, an international consulting, engineering, and technology solutions provider specializing in global solutions for the HIM and medical transcription industries. For example, NLP can tag several commonly used medical terms (eg, high blood pressure and hypertensive) and recognize them as fitting the overall description of the term “hypertension.”
NLP allows hospitals to mine data within the narrative without requiring additional effort on the part of physicians or abstractors, says Nick van Terheyden, MD, chief medical information officer with Nuance Communications, which offers products that provide a bridge from narrative to medical intelligence.
Using NLP to tap into the narrative not only allows hospitals to easily capture discrete data elements, but it’s also an important part of preserving this valuable part of the record, says Don Fallati, senior vice president of marketing and product management for M*Modal, which offers a platform that unifies speech recognition and NLP technologies.
Unless there are ways to extract structured data from the narrative, this particular style of documentation may be discouraged in the drive to deploy EHR structured documentation, cautions Fallati. If this happens, “We would lose very substantial, meaningful information,” he says.
What makes these data so valuable? Narrative information captures the essence of how doctors think, including valuable details about the rationale for medical decisions, additional evidence on which diagnoses are based, and a variety of other considerations the physician takes into account when diagnosing and treating a patient, says Fallati.
“NLP is essential to the preservation of the important narrative information that can get lost in structured documentation,” he says. “In this way, the narrative becomes a crucial complement to EHRs.”
The Wonders of NLP
Experts say the implications of NLP functionality are immeasurable, particularly as healthcare quality continues to take center stage.
“Hospitals are beginning to understand that there is a huge quality avalanche coming, [and they must have] the ability to manage data into meaningful outcomes,” says Cashwell.
Although NLP is commonly associated with computer-assisted coding, there’s also “a whole universe of [technologies] pertaining to physician dictation and transcription that are absolutely meaningful to the use of NLP,” says Faulkner.
The following are several ways in which hospitals can use NLP to improve quality and efficiency, enhance clinical outcomes, and take advantage of already-existing physician documentation.
Assist With EMR Content Fulfillment
First and foremost, NLP can assist with EMR content fulfillment, meaning it can extract discrete elements from any data source, including unstructured sources, and populate an EMR with this information, says Faulkner.
For example, a hospital with a newly acquired EMR can use NLP to identify problem lists from prior patient narratives and move this information to its database, says Fallati. Without NLP, populating data fields in real time within an EMR would likely require manual capture using a point-and-click template, which isn’t ideal for many clinical scenarios, says van Terheyden.
Enhance Abstracting and Reporting
NLP is also helpful in terms of abstracting and reporting information for larger quality-related initiatives, such as the Physician Quality Reporting Initiative and core measures. “The reporting demands are mushrooming, so the ability for NLP to potentially help automate greater degrees of abstracting offers great promise,” says Fallati.
Experts agree that meaningful use may be the single largest driving force behind NLP adoption to date.
“NLP can go through a giant universe of documents and extract information that specifically point to those meaningful use data elements,” says Faulkner. This could include a problem list, procedures, medications, allergies, vital signs, social history, and quality measures information.
Provide Real-Time Patient Data
Hospitals can use NLP concurrently to provide quality analysts and physicians with valuable information about patients while they are in the hospital receiving treatment, says Faulkner.
“As a record is completed, it is parsed and indexed for query searches,” he explains. For example, hospitals can use the technology to identify patients at risk for developing a urinary tract infection, which the Centers for Medicare & Medicaid Services considers a hospital-acquired condition.
“If in the H & P [history and physical] the doctor says he or she inserted a urinary catheter, you [can use NLP to monitor these patients] and possibly do something [preventive] while they’re still in-house. Usually these types of analyses are retrospective and hospitals can’t do anything about them,” Faulkner adds.
Likewise, hospitals can use NLP to identify all in-house patients with a past history of heart failure, says Faulkner. “These folks will be at risk for a recurrence when they’re admitted. This is important for surgeons to know prior to surgery. Right now, there’s no proactive way to know this information because it’s hidden in the narrative of the document,” he says, adding that NLP technology is intelligent enough to be able to decipher a past medical history of heart failure from a principal diagnosis of heart failure.
Cashwell says another way in which hospitals are using NLP is to track sepsis patients, who may be diagnosed with the condition at any given point during care. NLP helps uncover the symptoms immediately rather than retrospectively, allowing physicians to direct care toward these patients, which can indirectly reduce readmission rates.
Enhance Clinical Documentation Improvement Efforts
Some hospitals have taken advantage of NLP technology to help get a handle on their most problematic Medicare severity diagnosis-related groups. NLP analyzes documentation and searches for certain conditions in real time to determine whether clinical documentation improvement specialists should query physicians while the patient is still in-house, says Cashwell. Performing concurrent queries can effectively reduce the number of postdischarge queries and record holds, thereby improving cash flow and discharged, not final billed, numbers.
Facilities can use NLP to identify records—such as those involving one-day stays—that may be the target of a recovery audit contractor or other third-party auditors. For example, NLP can analyze discharge summaries and H & Ps to determine whether documentation meets admission and medical necessity criteria, says Cashwell. All this can be done prospectively while the patient is still in-house as opposed to postdischarge.
Perform Sophisticated Data Queries
“NLP can scan in fractions of a second literally hundreds of thousands of documents to find [terms] related to quality measures and patient safety that are otherwise strictly done through a very manual process. Because it’s manual, it’s also retrospective,” says Cashwell.
For example, staff members may need to comb through multiple clinical systems to identify all diabetes patients in the hospital at any given time, whereas NLP can generate the same list in a matter of seconds. The technology searches for explicit key terms (eg, diabetes) as well as related terms and implicit conditions based on lab results or other indicators.
NLP can help hospitals perform a “multivariable inquiry into a mass of information that comes from narrative and a variety of other sources,” Fallati says. An example in radiology might be to generate a list of all bone marrow cases that took place within the last 18 months to answer questions such as the number of reads performed for each case, the impressions generated for each read, and any conclusions drawn, all sorted by radiologist. Such information can drive radiologist quality and efficiency reviews and be analyzed in relation to other pertinent data to impact additional clinical analyses.
NLP allows hospitals to apply the intelligence “over a huge swath of narrative documentation to find the same results in infinitely quicker time,” says Cashwell. Anecdotal information revealed through case studies has uncovered that these NLP-generated queries may actually be more accurate than ones generated by humans, he adds.
The move toward widespread NLP adoption is inevitable, given the race to implement EMRs and discrete data measures, says van Terheyden. “As we move toward meaningful use and the digitization of records, we can’t just produce text as the industry has done,” he says. “We must produce data because everything is based on clinical data. We still need the narrative—it’s not going to disappear—but from that narrative we must extract discrete data. This will be assisted with NLP.”
Physician response about NLP has been positive thus far because it has minimal impact on workflow and allows for flexible dictation styles, according to Faulkner. “They don’t have to watch their words, and they don’t have to speak in a structured format,” he says.
However, much remains to be seen in terms of whether—and how—the technology may change workflow and required skill sets for medical transcriptionists (MTs).
From a workflow perspective, NLP requires someone to review the data and ensure that all clinical documentation is tagged and parsed into the correct fields within an EMR, says Faulkner. It’s unclear whether MTs, coders, abstracters, case managers, or quality improvement managers will embrace the role of data reviewer/validator—or whether an entirely new role will emerge, says van Terheyden.
Faulkner says MTs may be required to extract metadata, such as patient demographic information (race, ethnicity, etc), from physician dictation and insert it under the correct headings so NLP will parse and tag it. MTs may also read something in an accompanying report, such as a discharge summary or H & P, that is relevant to NLP searches and insert that information into metadata fields, he adds.
“It’s safe to say that the transcriptionist’s role is rapidly evolving. However, NLP underscores the value of things that medical transcriptionists have been doing for a long time,” says Faulkner.
“NLP makes transcription more relevant because it preserves the power of the narrative information,” Cashwell says, adding that the technology allows hospitals to take better advantage of MTs and/or speech recognition technology that they currently have onboard.
When evaluating vendors, industry experts suggest asking the following questions:
• Exactly how does NLP technology work with speech recognition? Some vendors may mislead customers to believe that they offer NLP when, in fact, they only technically offer speech recognition technology. Others offer true NLP that can be used in addition to speech recognition. In this scenario, a physician dictates using speech recognition and an MT then edits the document and sends it back to the physician for authentication. Finally, the hospital applies NLP to the document and a data reviewer validates the tagging and parsing of that information into the EMR.
However, some vendors provide both technologies (speech recognition and NLP) as part of a single solution, says Fallati. When choosing this option, physicians dictate using speech recognition and the resulting digital text is immediately tagged and parsed using NLP.
“If you can intelligently marry speech recognition with the NLP technology, then you can more dynamically look at the context of the words and sentences,” Fallati says.
• Will NLP be easy for physicians to use, and can they use it with front-end and back-end speech recognition?
• Does the NLP technology integrate with EMRs?
• Can the NLP technology easily incorporate data from other systems? Some vendors can process information regardless of where the information originates (eg, from another transcription company, a lab system, a radiology system). Other vendors have to process the data through their own platform first, says Cashwell.
• Can the vendor present data in an XML format? If so, this allows hospital IT departments to create their own customized dashboards, says Cashwell.
• What is the NLP vendor’s level of precision and accuracy?
— Lisa A. Eramo is a freelance writer and editor in Cranston, R.I., who specializes in healthcare regulatory topics, HIM, and medical coding.