Using Healthcare NLP for Risk Adjustment

Adoption of natural language processing (NLP) technology has accelerated across many industries in recent years as stakeholders seek to improve the speed and accuracy of documentation reviews. In healthcare, however, NLP adoption has been a bit slower.  

Yet several trends are now emerging that make the use of NLP more essential for healthcare organizations seeking to improve risk adjustment and other critical business functions. For example, patient populations are increasing and the number of patients eligible to enroll in risk-bearing programs such as Medicare Advantage is growing. In addition, the amount of healthcare data is exploding and outpacing many other industries, according to RBC Capital Markets. In this blog, I will outline some key changes in the Risk Adjustment market that make NLP a necessity in this space. First – a bit more on NLP in healthcare. 

NLP’s role in healthcare 

In short, NLP is a technology that ingests unstructured text, processes it using artificial intelligence (AI) and other techniques, and converts that text into structured information suitable for analysis by algorithms or humans.  

However, it’s important to note that one person's NLP is not necessarily another person's NLP. The technology encompasses far more than just finding words and key terms – it is much more sophisticated than that. For example, NLP is able to understand all of the different synonyms, words, abbreviations, and misspellings in medical records, as well as the many different contexts that doctors write in patient notes – such as negation and family history.  

At Linguamatics, we combine best-in-class artificial intelligence with NLP, then package it within a lean and scalable solution. Our solution can ingest a variety of different healthcare document formats that the industry processes on a daily basis to deliver improved accuracy of coding. This technology provides a significant depth of understanding and processing of records to put as much relevant information in front of coders to substantially reduce the time spent analyzing manual charts. 

NLP for risk adjustment and beyond 

Risk Adjustment is one area in healthcare where the uptake of NLP has been faster than others. This rate of adoption is only going to increase, thanks largely to two changes in the risk adjustment market.  

Firstly, perhaps the most pressing trend is the recent final Risk Adjustment Data Validation (RADV) rule issued by the U.S. Centers for Medicare and Medicaid Services (CMS), which has increased regulatory pressure on healthcare organizations to ensure accurate risk adjustment. The rule is intended to make it easier for CMS to claw back overpayments to healthcare organizations that were awarded as a result of inaccurate risk adjustment. Therefore – the use of accurate NLP to identify clinical conditions and their supporting evidence (as per the Monitor Evaluate Assess Treat – MEAT framework) is vital. Secondly, the Medicare Advantage risk adjustment model is due to change from V24 to V28 over the upcoming three years. These changes will significantly reduce the number of risk adjustable conditions, therefore, technologies which support accurate and complete capture of a member’s health are a necessity for organizations looking to ensure they don’t lose funding needed to provide care for their chronically ill members.    

It is important that NLP is not thought of as a tool to just add codes to members. It is a tool, that if used appropriately and fairly, can:  

• Ensure accurate coding – in terms of both “additions” and “deletions” (removal of claims where there is no substantiating evidence) 
• Increase regulatory compliance – and the evidence of compliant risk adjustment coding 
• Support audit teams in the compliance reviews 
• Augment a stretched workforce  

In addition to risk adjustment, there are numerous healthcare use cases that can benefit from the implementation of NLP. Here are three notable examples: 

Improving STARS ratings: NLP can scour clinical documentation to find both denominator and numerator criteria from unstructured medical records in quality programs such as HEDIS. For example – identifying patients who had had Falls Screenings or mammograms. 

Closing care gaps: NLP algorithms can mine clinical data to find specific disease features that indicate growing patient risk, enabling earlier interventions that can sometimes be lifesaving for patients.  

Identifying social determinants of health: On a daily basis in healthcare settings, doctors capture huge volumes of clinically important information that provides insights into patients’ social circumstances and risk, such as transportation access, employment status, and living situation. NLP surfaces this data, which is growing increasingly important in value-based care arrangements – and will soon become part of NCQA quality measures. 

Given an environment characterized by a growing patient population, a vast expansion of healthcare data, and a tightening regulatory climate, now is the time for organizations to consider how they can adopt and implement NLP technology to optimize risk adjustment.