Forward Looking Opportunities of NLP in Healthcare

NLP in Healthcare eBook Health Fidelity Adam Gronsky

Building on my colleague Chris Gluhak’s work unpacking NLP technology in healthcare, I’ve written our next installment in the Health Fidelity ebook series: Natural Language Processing in Healthcare: The Clinical and Financial Opportunity of Suspecting. This next piece explains how our technology stack processes information from our NLP engine, Lumanent Insights, to derive clinical suspects and deliver them to providers.

First, a point of clarification on the difference between suspects and suggestions. Suggestions and suspects both use NLP technology to comb through enormous volumes of disparate-sourced clinical data and propose a conclusion to healthcare experts. The traditional outputs from NLP technology around risk adjustment are suggestions. These are provided to coders after care is delivered to amend claims for completeness and accuracy prior to submission to catch missing or under-coded conditions. Beyond that, at least in Lumanent, suggestions aren’t exclusively additive, they often create an opportunity for linked clinical evidence of codes as well as redaction where necessary or sought after.

Suspecting, on the other hand, is an exciting next step in applying the use of NLP to unstructured clinical data. It’s a prospective-looking view of potential member conditions and treatments based on a wide variety of data sources to surface incomplete and under-documented diagnoses, or suspects. These suspected conditions are presented for clinical review to either be addressed with patients at an upcoming visit or, proactively, used to reach out to patients to schedule a visit, depending on clinical indication and individual patient needs.

It’s worth noting that as much as this is a fresh iteration, suspecting has already been a part of payer risk adjustment, albeit from a more limited data source. Generally known as “prospective review” or risk analytics, (as opposed to retrospective, bringing suggestions to coders), it was exclusively drawn from claims data, whereas suspects are now drawn from a wider array of sources. The novel aspect centers on deploying suspects in a care setting, especially within the realm of risk adjustment. The data already gathered is now being put to work in an actionable way directly for providers, instead of exclusively after the fact for coders. In the past we’ve discussed re-centering risk around the patient, that while risk adjustment is traditionally considered an actuarial tool, there’s opportunities to use it to positively impact care. This is a great example of that. Under models that emphasize delivery of necessary care at the individual patient level, populations are then better understood and treated thanks to NLP-powered clinical suspecting.

For example, imagine a physician had the time and resources to review every piece of available documentation on a patient prior to a visit, find that they are receiving a medication that indicates a condition, but there’s no note of the condition in their local EHR, and easily determine why. Of course, that’s not feasible, but the right AI can do just that for the clinician, effectively prepping the chart before the patient arrives. Through clinical suspecting, all available data is reviewed, and any risk adjustable conditions with a high degree of evidence (but not coded) are presented to the care team for verification or rejection, through documentation or alteration to that patient’s care.

Additionally, suspecting helps eliminate documentation gaps on a go-forward basis through a fairly simple truth: the more chart preparation is utilized, the more gaps are narrowed, and suspecting makes intensive chart prep go from a wish to a reality. And through that, care can be more accurately provided by clinicians with less administrative work. At the same time, the record and reimbursement better reflect the total disease burden the patients (and their care providers) are carrying. Finally, even care management and patient cross cover/handoff is smoother.

NLP derived clinical suspecting is an exciting, emergent application of technology in healthcare. It’s use by health systems is an opportunity to deliver more refined care and have greater financial stability through a complete, accurate look at both individual patient and population disease burdens.

To learn more about clinical suspecting, its benefits, uses, and challenges in healthcare, click through to read the next installment of our ebook series: Natural Language Processing in Healthcare: The Clinical and Financial Opportunity of Suspecting.