Navigating Medicare Advantage Risk Adjustment Plan Changes

Learn about significant changes in the Medicare Advantage Risk Adjustment model (transitioning from version 24 to 28) for payers and providers and the importance of accurate coding and the use of technologies like AI and natural language processing from Calum Yacoubian, MD, director of NLP Healthcare Strategy at IQVIA.  

Interview Transcript: 

Please share your name, title, affiliation, and a bit about your background. 

My name is Calum Yacoubian. I'm the director for Healthcare Strategy at IQVIA in our natural language processing business unit. I have been here for 3 years, and I've spent the last 8 years working in the intersection of artificial intelligence and health care organizations. I'm also a physician and I previously worked and trained in the UK in hospital medicine.  

Could you share a brief overview of the Medicare Advantage 2024 Advance Notice changes from Risk Adjustment Model V24 to V28? 

The transition from version 24 of the Medicare Advantage Risk Adjustment model to version 28 represents the single biggest change that there has been in the risk adjustment model on the market since its inception. From a very high-level perspective, Medicare Advantage risk adjustment is the process by which payments are tied into value-based care and the arrangement of underlying comorbidities of the patients that health plans and providers are caring for. It's essentially a predictive model making care cost predictions based on submitted diagnosis codes. 

We’ve operated under version 24 for a few years and there are some significant changes in version 28. We're dropping the total number of ICD 10 codes which are risk adjustable. Those codes impact the payments from nearly 10,000 down to 7,700. So nearly a third of the potential codes that impact reimbursement are going away.  
Also, the categories that the rest fall into is changing significantly. We're going from around 86 categories to 115. The way that the risk is modeled and calculated is significantly different. They’re also predicted to have a greater than 3% impact on the Medicare Advantage risk scores, which translates to $11 billion in savings for the Medicare trust in 2024.  

What are the main factors and motivations that may have brought about this change?  

The driver behind these changes is coding and model accuracy. The risk adjustment predictive model for version 24 was based previously on historical years of service and was founded around ICD 9 coding practices. So, a major motivation is trying to base the new model on more recent years of service, coding, and the disease burden of today's populations.  

However, I think this is not isolated but sits alongside a growing feeling in the Medicare Advantage market that there's potentially inappropriate reimbursement happening. The version changes sit alongside the issued final rule for the risk adjustment data validation audits, which are increasing the financial penalties of organizations that submit unsubstantiated diagnoses as part of their risk adjustment submissions.  

What do you think are the most impactful changes in the refined V28 model for payers and providers?  

The obvious place to start is where the diagnosis codes have dropped and added to the model. The number of codes newly added is about 270 and nearly half of these represent somewhat rare conditions that most of the medical advantage population won't have. So, these new codes may not have as much of an impact.  
Codes that have been dropped I think will have a more significant impact. Disease areas like vascular disease, congestive heart failure, diabetes, psychiatric health, and major depressive disorder are well documented as being at the highest risk of their scores being impacted. These are significant comorbid conditions with a high cost of care. In vascular disease, for instance, in the previous model, there were over 400 related diagnosis codes. That's gone down to 293 codes.  

There are also changes in how the coefficients are now calculated. The associated complications of diseases like diabetes and congestive heart failure no longer impact the risk score used to protect per member per month costs. Previously if you had diabetes with chronic complications or a diagnosis code that matched to that HCC, there was a higher reimbursement than if you had diabetes with no complications. This clinically and logically makes sense because management is more complicated and expensive. But in both diabetes and congestive heart failure, there's no impact on having chronic complications of the disease on the net payment.  
This is going to be significant for providers who need to be well-educated about accurate coding. Payers in relationships with providers also need to ensure coding practices represent these diagnosis changes that will result in reimbursement for chronic conditions.  

What are the implications for RAF scores and financial outcomes?  

The implications depend very much on the case mix of a payer and whether these changes are going to result in them receiving the same, less, or more reimbursement for their population. But, as the overall trend leads to notable savings, then we can reason that the impact is going to be fewer payments on average to health plans and therefore less downstream for the providers. As I say, it depends if you have a case mix that has many patients with chronic conditions that no longer impact the RAS score. This is meant to better represent chronic conditions, so it's important that providers are accurately coding and documenting chronic conditions that are continuously treated and monitored.  

Why are technologies that support precise HCC identification and facilitate the retrieval of robust supporting evidence from unstructured medical records essential during these framework shifts?  

For payers, the use of technologies like artificial intelligence and natural language processing are probably the most widely adopted in risk adjustment because it's so important to get the right clinical condition documented and coded for appropriate reimbursement. Coding accurately has never been more important. And it is also important to use these technologies for a compliance framework.  

We need to support coders in identifying supporting evidence. In the last 9 audits from the office of the general inspector, all found overpayments to health plans based on diagnosis scores being submitted without supporting documentation. AI can be used not just to identify conditions but to ensure supporting documentation is available. Unless you're using this technology, you're going to miss opportunities and leave yourself susceptible to penalties. 

Is there anything else you would like to add?  

Our conversation today has centered around the changes in risk adjustment and the use of technology to augment and ensure accuracy in this process. But I think increasingly as payers and health care providers consider their use of technology, there's an appetite to see how we can get the most out of tech and use it across other areas. NLP can be applied to many different processes within a payer organization, particularly with burdensome chart review. Quality measures, understanding member-level social determinants of health; there's a lot more that can be done with NLP beyond traditional supporting of coding efforts and risk adjustment. I think that maximizing investment in technology and NLP is becoming increasingly important.