Authentic voices: What Natural Language Processing reveals about your patients

For most health systems, rolling out a real-time feedback program reliably spurs a massive uptick in patient response rates. But it’s not the volumes that make such programs valuable. It’s the voices that they capture.

 Today’s forward-thinking organizations are using real-time feedback programs to solicit open-ended feedback. This means that more and more patients are being given the space to say whatever they like about their providers or the care experience at large—and their candid comments, delivered immediately after their episodes of care, reveal far more intimate and important details than conventional, closed-ended questionnaires.

However, open-ended comments also present a logistical problem. When comments come in at a rate of thousands per month, it becomes physically impossible for human reviewers to read them all. This is why today’s health systems also rely on Natural Language Processing (NLP) technology to help them grasp what their patients have to say.

For Harris Health System (HHS) in Houston, a real-time feedback solution—with NLP as its functional core—enabled an unprecedented level of insight into how their patients felt about their care. 

“The feedback is exponentially more useful,” says David Riddle, HHS’s Administrative Director of Patient Experience.

NLP, in short, gives organizations a way to capture the authentic voice of their patients. Here is a brief overview of how the technology works, along with specific features that made NRC Health’s NLP-augmented real-time feedback system so successful at HHS and 200 other organizations that, together, have collected more than 10 million patient reponses.

What is NLP?

At its simplest, NLP is a way for machines to process large volumes of verbal information. NLP programs algorithmically process, categorize, and apply sentiment to text, in order to make it more manageable for human readers.

One everyday NLP application is your email inbox’s spam filters. NLP algorithms often automatically flag emails containing the words “money” or “free,” because these are strong indicators of a spammer at work. This saves the user the hassle of having to click a delete button.

It’s easy to see how this might be valuable in a patient-feedback context. NLP, in effect, can render thousands (or, for that matter, hundreds of thousands) of comments newly legible in groups, sparing patient-experience teams the hours it would take to read through them all.

NLP solutions, however, are not one-size-fits-all. Healthcare is considerably more complex than an email inbox, and NLP capabilities aren’t always transferable across different contexts. To function in the healthcare space, an NLP solution must be specifically configured for the industry.


“Every NLP solution is an improvement over human review,” says Sanjay Motwani, VP of Product at NRC Health. “But without industry-specific input, generalized solutions can cause healthcare organizations to miss opportunities.”

Motwani notes that it’s not enough to simply extract verbal data from patient comments. “To be useful, voices from healthcare consumers and patients must be organized and presented in such a way as to drive meaningful action across a health system. Often, generic solutions can’t do that.”

The key to useful NLP in healthcare: Industry-specific configuration

A spam filter is binary: an email is either spam, or it is not. Patients’ opinions about their care providers, on the other hand, aren’t so black and white. They can love one aspect of a care experience (say, their provider’s warmth and compassion) while hating another (wait times are a frequent culprit). Any functional healthcare NLP system must be able to capture these gradations.

Two different NLP processes make this possible: comment categorization and sentiment analysis.

Comment categorization is the first step. It’s what identifies and catalogues patient comments in a way that’s useful for healthcare leaders. This process sorts and organizes comments by their domains of concern within a healthcare organization, putting them into categories like “Interpersonal communication,” “Access to care,” and “Follow-up.”

From there, sentiment analysis identifies the emotional tenor of a patients’ words. It looks for key words that indicate not only when patients discuss a topic, but also how they feel about it. If a comment includes the word “rude,” for instance, that hints at a service problem; the word “forever” probably indicates an overlong wait time. This shows leaders where patients are satisfied (or not), and therefore where they should direct their attention.

Together, these NLP capabilities bring several important benefits to health systems.

Resonating with clinicians

Almost as important as what patients say is how many patients are saying it. In healthcare, as in many other businesses, one-off complaints from disgruntled customers can be all too easy to dismiss. Accumulated complaints, however, are harder to ignore.

This is especially true for clinicians, whose work requires a scientific—and skeptical—approach to data. A small number of comments are unlikely to motivate clinicians to change how they work; but when numerous comments all converge on a certain point, that’s often enough to spur clinicians to action.

At the LBJ Hospital Specialty Clinic Platform, one of HHS’s facilities, that convergence point was clear. A plurality of patient comments pointed to an arena ripe for intervention: in one quarter, 44% of the entire Specialty Clinic’s negative patient comments were directed toward wait times at the ophthalmology clinic. This stark statistic helped ambulatory-care leadership build an unimpeachable case for change.

So says Dr. Mohammad Zare, HHS’s Ambulatory Chief of Staff: “Our biggest ‘ah-ha’ is that, globally, the feedback data now speaks for itself, stands on its own, and is credible. Physicians don’t question it.”

Unearthing genuine concerns

Without NLP, however, it’s unlikely that HHS’s leaders would ever have resolved the wait-time issue in the opthamology clinic.

This is due to the limitations of conventional surveys. Their narrow range of inquiry means that certain particulars can be easily overlooked. CAHPS surveys, for instance, ask some questions about care episodes, but these may not cover the patient’s true foremost concerns.

By contrast, open-ended comments give patients the opportunity to voice whatever they’re thinking. Well-designed NLP programming registers what they have to say, and alerts leaders to what they should be changing.

In HHS’s case, patients in the ophthalmology clinic consistently reported frustration with their wait times. However, reviewing patient comments showed that it wasn’t the delays themselves that irked HHS’s patients—it was the lack of communication surrounding those delays.

By creating a more effective communication strategy, the ophthalmology clinic helped patients better understand where they stood in the clinical process. As a result, negative comments about wait times dropped by 60%.


Results like this are what make the technology worthwhile, Riddle says. “The qualitative feedback provided is often around things that are not normally asked on a traditional survey,” he says. “It’s truly representative of what’s on the patient’s mind.”



Empowering service recovery

Finally, NLP has one more crucial function in health systems: rapidly identifying opportunities for clinical improvement and service recovery.

Certain words and phrases, when they appear in patient comments, are signifiers of extreme dissatisfaction. When patients report that they “hated” their experience, or that their providers seemed “dismissive,” they may well be frustrated enough to change providers altogether.

Still worse is when they report feeling “confused” about their discharge instructions, or “unclear” about follow-up care. These may indicate that the patient is in danger of non-adherence, which could put them at serious risk for harm.

A well-designed NLP program flags these charged comments and brings them to the attention of staff who can help. This way, providers can preserve their relationships with unhappy patients, and possibly even prevent negative health outcomes.

“NRC Health’s NLP solution provides us with more opportunities to react in real time to our patient’s concerns and provide appropriate recovery,” Riddle says.

In 2018, HHS used NLP to identify 307 instances of actionable comment alerts at HHS’s emergency centers. When appropriate, they were able to successfully intervene every time.

Small wonder that overall patient satisfaction (9–10 ratings, Would recommend) spiked from 56.0% to 67.7%—in just one year.

What the right data can do

As astonishing as these numerical victories can be, HHS’s leaders believe that the qualitative changes they’ve observed at their organization are even more important. Both Riddle and Dr. Zare will attest to NLP’s power to effect a cultural shift within an organization.

“It gives us insight on what matters most to our patients,” Riddle says—and clinicians at HHS rallied immediately around that insight, driving themselves to deliver what their customers needed most. A culture of healthy competition emerged at HHS’s ambulatory-care practice, as providers vied to see who could best satisfy their patients.

That’s the real value of healthcare-specific Natural Language Processing. When coupled with real-time open-ended comments, NLP shows organizations what patients want, in their own words and in meaningful numbers. That’s undeniably compelling. In its specificity, in its validity, and in its volume, the use of NLP technology empowers organizations to confidently proceed with ambitious programs to improve their patient experience.



News From Our Partners.

Copyright © 2024 Becker's Healthcare. All Rights Reserved. Privacy Policy. Cookie Policy. Linking and Reprinting Policy.


Featured Whitepapers

Featured Webinars