A look at natural language processing (NLP) technology’s top potential use cases reveals how its destined to disrupt healthcare.
It’s no secret that the big data wave had been building in healthcare for some time. What has somewhat remained a bit mysterious, however, is how innovators planned to stave the flood and make use of the sizeable data to help healthcare providers create efficiencies and improve outcomes. According to a recent Chilmark report, natural language processing (NLP) may be the technology destined to turn the tide, or at least turn it into something more useful.
NLP emerged in the market with its algorithmic ability to interpret and manipulate human language more than a decade ago, enabling many voice-to-text technologies to simplify clinical documentation. As innovations such as artificial intelligence (AI) continue to develop so has the application and sophistication of NLP, despite the criticism that it’s underutilized. Nevertheless, the NLP market is expected to grow at a CAGR of 16.1 percent until 2012, creating approximately $16 billion in market opportunity.
According to a recent report released from Chilmark, there are multiple use cases for NLP (some obvious mainstays, some newly emerging, and some queued up for the next generation). The use cases vary, but they all seem to stem from three core drivers: supporting the needs of value-based care (VBC) and population health management (PHM), coding and analyzing encounters more effectively, and decreasing physician workload and burnout.
As the motivation to digitize healthcare and modernize clinical practice continues, so does the shift in healthcare to make better use of unstructured data. Unfortunately, many of the technologies introduced to ease workflows, such as electronic health records (EHRs), have actually made more work for healthcare providers. NLP has the potential to take advantage of 80 percent of unstructured data captured by IT systems and turn it into actionable insights.
Here’s a look of eight use cases for NLP technology in healthcare:
What’s Here to Stay
1. Speech Recognition
NLP has developed its roots in healthcare with speech recognition, allowing clinicians to transcribe notes for efficient EHR data entry for nearly two decades. Front-end speech recognition frees clinicians to dictate notes instead of having to sit at the computer at the point of care, while back-end recognition works to correct any transcript errors before passing it along or human proofing.
Speech recognition is considered one of the most cost-effective approaches, reducing the need and high cost of medical transcriptionists. There are very few providers who haven’t invested in an NLP technology such as the Nuance or M*Modal. The near ubiquitous adoption of such products will likely drive the market to develop new ones, particularly by sharpening its back-end functionality to eliminate the need for transcriptionists.
Although the market is saturated with speech recognition technologies, the disruption from startups incorporating deep-learning algorithms to text processing and mining applications is a real possibility. Mobius MD, for example, is developing a front-end speech recognition application that integrates deep learning.
2. Clinical Documentation
NLP’s impact in speech-recognition goes hand in hand with clinical documentation—freeing clinicians from the manual and confining structure of EHRs and allowing them to focus more on the patient—
thanks to speech-to-text dictation and formulated data entry. Both Nuance and M*Modal have technologies that work in tandem with their speech recognition technologies to capture structured data at the point of care and standardized terminologies for future use.
As NLP technologies continue to pull pertinent data from speech-recognition tools, as well as other emerging forms of data, this will significantly improve analytic data used to drive VBC and PHM efforts, allowing clinicians to make better decisions to ultimately improve outcomes. In the future, NLP tools could be applied to social media and other public data sets to determine social determinants of health (SDOH) as well as the effectiveness of wellness-based programs and initiatives.
3. Computer-Assisted Coding (CAC)
When NLP-driven CAC emerged in 2013, it brought with it the promise to ease the woes that came with ICD-10 coding and was predicted to improve coder accuracy by both the Health Information Systems Society and DAK Systems Consulting.
CAC extracts information regarding procedures and therapies to capture every possible code to maximize claims. Although CAC may be one of the most common established uses of NLP, its adoption rate is only at about 30 percent, perhaps due to the technology falling short of its promise. CAC may have improved the speed of coding but hasn’t done much in improving accuracy. For example, a study by the Cleveland Clinic found that CAC did reduce the time to code, but the use of CAC alone—without the intervention of a credentialed coder—had a lower recall and precision rate.
According to Chilmark report, 3M is the current vendor of choice, accounting for an estimated two-thirds of the market. However, there is competition from Optum 360, Dolby Systems, nThrive, M*Modal, and AMI (Artificial Medical Intelligence). All of these vendors will need to shift their solutions to meet the challenges of a value-based paradigm to ensure they are working as expected.
4. Clinical Trial Matching
Perhaps the most buzzworthy in the “emerging” use-case category is clinical trial matching. Using NLP to identify eligible patients for clinical trials is not only exciting but essential. According to the report, 20 percent of U.S oncology trials fail to meet their enrollment targets, while pharmaceutical and life science companies invest millions in this manual form of recruitment and managing trials.
Companies like Linguamatics Health and Clinithink are changing that, having developed NLP engines to address the challenges that come with trial matching as well as IBM Watson Health and Inspirata, which have dedicated tremendous resources to using NLP to support oncology trials. It seems NLP has the potential to make clinical trial matching a seamless and automated process in the very near future.
5. Prior Authorization
Physicians often know what’s best when it comes adequately treating their patients. The issue of whether payers will agree and authorize reimbursement is an entirely different matter, perhaps not for long. The idea of instantly determining coverage using high throughput methods at the point of care may soon come to fruition. IBM Watson Health and Anthem are working on an NLP module used by the payer’s network of providers to rapidly determine prior authorization.
Ubiquitous solutions are years away. However, the drive to solve the issue will surely motivate the growth of this use case. According to a survey conducted by the Medical Group Management Association, 86 percent of providers stated that prior authorization requirements have increased in the past year; the preapproval process was also identified as the top regulatory burden for providers in 2018.
6. Clinical Decision Support
NLP’s presence in technologies will bolster clinical decision support. However, solutions are being devised to support clinical decisions more acutely. There are certain areas of practice that are in need of better methods of surveillance, such as medical errors.
According to the report, recent research has demonstrated the effective use of NLP for automatic infection detection. M*Modal and IBM Watson Health are the leading vendors for NLP-powered CDS. Isabel Healthcare has used NLP to aid clinicians in diagnosis and symptom checking.
7. Artificial Intelligence: Virtual Scribe & Chatbots
There is no solution to turn to yet, but speech recognition applications that could do away with the need for human scribes altogether would be a game changer in clinical documentation. The ideal tool for the job would be something similar to Amazon’s Alexa or Google’s Assistant; Microsoft and Google have recently partnered in the pursuit of this specific goal, and it’s safe to assume that Amazon and IBM will do the same.
Virtual personal assistants also are known as chatbots are rapidly making their presence in the digital world, and the healthcare industry is of no exception. Currently, these automated interactions can capture symptoms and triage patients to the most appropriate provider. Startups developing chatbots include Bright.md, which has created Smart Exam, a “virtual physician assistant” that uses conversational NLP to collect personal health information, compare the data to evidence-based guidelines, and package the data, along with diagnostic recommendations, for the provider.
Woebot, another motivated startup has developed a “virtual therapist” that engages patients via Facebook Messenger. According to a two-week trial, Woebot successfully lowered anxiety and depression in 82 percent of the college students who participated.
8. Computational Phenotyping and Biomarker Discovery Phenotyping
Much the way NLP is changing clinical trial matching, it also had the potential to aid clinicians with the complexities of phenotyping patients for analysis. For example, NLP will allow phenotypes to be defined by the patients’ presented conditions as opposed to the knowledge of experts.
NLP could also be used to analyze speech patterns, which could prove to have diagnostic potential when it comes to neurocognitive injuries such as Alzheimer’s, dementia, or other psychological or cardiovascular diseases. A lot of startup activity is emerging around this use case, including from BeyondVerbal who collaborated with the Mayo Clinic to identify vocal biomarkers for coronary artery disease and Winterlight Labs who is uncovering distinct linguistic patterns in the speech of Alzheimer’s patients.
What Stands in the Way?
No question the future looks bright for NLP technology in healthcare, but these use cases won’t happen overnight, and some may not happen at all. Real barriers do stand in the way of NLP’s presumed fate. Challenges for adoption include the high costs of software and recruiting the talent necessary to prepare and manage the datasets; future offerings of classes dedicated to understanding NLP, and AI technology may help mitigate part of this challenge. Another barrier is that system usability continues to be a problem.
As already noted, some of NLP’s intended use cases have yet to be perfected, leaving love lost among healthcare providers and organizations, even the speech- recognition tools have flaws that make them difficult to use at times. Data quality is a similar issue that needs a valuable overhaul. No matter how sophisticated the application, if the quality of the data is subpar, so will be the innovation’s performance.
Naturally, the threat of cyber attacks is another real an unpredictable barrier for NLP. In 2017, the NotPetya cyber-attack preyed upon Nuance’s Dragon and cost the company nearly $100 million to date. As a result, every vendor will have to ensure it has an incident-response plan to handle potential cyber attacks.
Despite the barriers, the evolution of NLP in healthcare is already progressing. With NLP leading the way, the AI market is set to rise to more than $6.14 Billion by 2022. Although challenges with NLP exist, so does the desire to get past them and play on its potential. Big data isn’t going anywhere and now is the time for healthcare to capitalize on it with NLP.
Featured image via www.vpnsrus.com