Recently, we discussed the imperative of HCC recapture, especially in a year with encounter volumes heavily disrupted by COVID-19. However, HCC recapture is only a small part of the larger goal: discerning a complete and accurate risk profile of a covered population. With an accurate risk score, reimbursement can match the total cost of necessary care, ensuring adequate funding to ensure the quality of life for patients and financial health of the institutions covering and treating them.
Let’s get into why.
Effective HCC recapture is a reconfirmation process, somewhere between a census and an audit. It establishes that patients are still covered, and their conditions are still managed year over year. What it doesn’t do is account for new conditions. It is also subject to constant attrition even when performed perfectly; patients move to other areas, obtain other methods of covering the costs of their care, or they pass away.
This is why a recapture program is absolutely necessary but, at the same time, only a first step. HCC recapture does not necessarily account for increased levels of coding severity for previously documented conditions through comorbidities or progression. It doesn’t capture conditions of new members of a population at all. To fully grasp the breadth and depth of conditions within a population, it requires experts in their fields to double and triple up their expertise, with doctors often learning basic coder practices, and coders making up any gap with thorough research through clinical notes. Unfortunately, with 80% of all clinical data being unstructured, this is a profoundly intensive process which is difficult to undertake manually, even with an army of clinically trained researchers and coders.
This is where an NLP-powered technology like Lumanent Retrospective Review can be a considerable asset. Every coder has stories of conditions indicated by claims, or even openly referred to, but not directly captured in any medical documentation and therefore never submitted on a claim. For example, say there’s a patient with a DME (durable medical equipment) claim for a chair lift, but no condition that requires the device. That claim, in and of itself, will not impact risk score, the diagnosis creating the need for it is not documented or on any claim. At the same time, if the DME claim is there the patient has a need for the chair. Asking a coder to dive through the entire corpus of records associated with that patient to find a risk adjustable condition that aligns with the need for the DME is either too expensive or too arduous. NLP, on the other hand, can take that (and other) DME claims into account, as well as pharmacy, labs, and prior diagnoses from previous years, and submit an appropriate code, as well as clinical evidence, for coder approval.
More specifically, in a case where a patient does have a documented condition of unspecified cirrhosis of liver, a traditional reconfirmation is relatively straightforward. Cirrhosis is chronic, and its treatment is largely around removing the cause and mitigating complications. Through a traditional HCC recapture at an annual wellness visit, current complications that are not acute may not be readily apparent, especially if the patient record does not reflect relevant diagnoses by a specialist in the prior year. In that circumstance, an NLP powered review of the patient would reveal an endoscopy and a prescription for Nadolol, both of which reveal portal hypertension and varices as a result of cirrhosis. The impact on adding that clinically validated diagnosis to the submitted claim impacts RAF for that patient by 0.94.
The pattern repeats throughout healthcare: diabetic supplies but no documented HCC for diabetes; oxygen delivery but no emphysema. Or, in cases where there are codes for diabetes, there’s evidence of treatment for peripheral vascular disease, but no code, effectively resulting in under-coding.
The inverse can also be true: any claims that lack clinical evidence of their severity or presence at all can be flagged for deeper review with the patient’s clinician. They can also be outright redacted, further insulating against audits through a vigilant internal compliance protocol supported by NLP and previously inaccessible unstructured clinical data.
Of the above examples, very little, if anything will be found in even the most robust recapture program. They are still, nonetheless, dealing with a greater disease burden than is likely realized, and incurring the associated costs. A recapture program is also easily interrupted by patients either refusing or not being able to receive care for clinical reconfirmation to take place.
We champion HCC recapture because it’s absolutely critical, but it’s only as good as the initial confirmation process. Those overseeing a recapture program are likely very well aware of the tiny voice whispering in their head, “What makes us think we caught everything to begin with?”
The confidence of knowing that the unstructured clinical data that had been captured over the years is now going to work to ensure that the greatest possible capture rate is achieved should bring comfort to that concern. All leaders want to know they’re seeing a more complete and accurate risk profile than before and that the data is finally going to work for them.