As Medicare Advantage (MA) plans improve their performance, enrollment rises—along with the need for quality data to help plans understand their member population. MA plans depend on medical payment data and patient data as part of medical data sets to maximize reimbursement, manage risk, and implement quality improvements.
But critical clinical data such as medical history and clinician notes may be locked in unstructured text. Not having access to this data puts health plans at risk of having an incomplete picture of their members’ health.
That’s where NLP (Natural Language Processing) comes in. Clinically intelligent NLP, is an important application of artificial intelligence (AI) technology, and the processing of natural language in clinician notes can help MA plans extract clinical meaning from unstructured data.
To tap the wealth of information in unstructured data, health plan executives need a strategy and solution for adopting this powerful technology in their organizations.
MA Plans can Extract the Meaning from Unstructured Data with NLP
Read the article to learn:
- Why value-based-care requires health plans to leverage unstructured data from medical data systems
- The importance of having a complete picture of members’ health
- How clinically intelligent NLP can help health plans achieve performance goals
- Six key steps health plan executives can take to implement NLP within their organizations
The need for value-based care in today’s healthcare environment is clear: one in three Medicare benefi ciaries has four or more chronic conditions, and 6% of healthcare users represent 75% of healthcare costs in the United States. Medicare Advantage plans have an important role in providing the highest quality of care for their benefi ciaries, and the Centers for Medicare & Medicaid Services (CMS) has been encouraging more diverse and more affordable Medicare Advantage choices for the Medicare population.
CMS is eager to ensure that vulnerable higher-risk populations get the care needed to improve health outcomes and at the same time demonstrate program integrity. The agency has been exploring how emerging technologies such as artifi cial intelligence (AI) can help ensure proper claims payment, reduce provider burden, and make program integrity activities more effi cient.
For Medicare Advantage plans, CMS changes mean more fl exibility and an opportunity to offer more services, take on more members, and maximize reimbursement. But plans have to better understand compliance requirements, including making sure that the risk adjustment data they submit to CMS is accurate and up to date.
Healthcare executives must explore innovative strategies and processes that address risk in order to offset the costs of providing care for those with chronic and, therefore, costly conditions. They must fi nd effi cient ways of extracting the necessary data to properly classify members in Hierarchical Condition Categories (HCCs), which are used for determining the resources needed to care for patient populations based on diagnoses of chronic and active conditions. And they must improve their overall performance in CMS’s Star Rating Program, which measures Medicare Advantage plans’ performance and which earns plans certain bonuses if they achieve higher ratings.
Because benefi ciaries can compare plans and Star Ratings through the Medicare Plan Finder tool, a plan’s performance is crucial to attracting and retaining benefi ciaries. According to fi ndings from McKinsey, the Star Rating Program has been effective in driving better overall performance. With that improved performance there has also been a surge in enrollment to 33% of Medicare benefi ciaries, or 22.2 million people, according to data from the Kaiser Family Foundation. That number is expected to reach 24.4 million in 2020.
Th e Need for Value-Based Care is Clear:
- One in three Medicare beneficiaries has four or more chronic conditions.
- 6% of healthcare users represent 75% of healthcare costs in the United States.
Gaining a Holistic Picture of Beneficiaries
Given both the opportunities and the challenges, the big question healthcare executives must answer is how to obtain a holistic picture of each member’s health status in a way that can meaningfully inform the organization about its overall performance.
A true, holistic picture of a patient requires more than just standard claims data. It also includes the free-text observations that may supplement the clinical data—such as patient-generated information and social determinants of health (SDoH)—to better understand how the conditions under which people are born, live, learn, work, play, and age can infl uence health and wellness.
The problem for plans is that most of the clinical insights needed for a full picture of a patient and the patient’s health status lie in the unstructured notes contained in electronic health records (EHRs). These can be medical history, progress notes, or pathology and imaging narratives that are captured by clinicians outside what’s codifi ed in the medical record. In fact, approximately 80% of healthcare data is unstructured.
Extracting insights and observations from the sea of unstructured data can be accomplished using advanced AI technologies such as natural language processing (NLP), combined with clinical expertise. NLP can help automate a labor-intensive, costly, and retrospective process.
This is not new to plans. Today, payer organizations employ clinicians to inspect physician notes to determine diagnoses, interventions, and other crucial information about a patient’s health. But that process is time-consuming and diffi cult to manage at scale.
Compounding the challenge is the diversity of terms, expressions, and abbreviations used by clinicians in EHRs. Similar or identical information may be written in different ways, using different words, in the record, and may be mixed in with other observations irrelevant to a given condition. To make sense of all the unstructured information, plans must normalize the data to standardized codes such as LOINC (Logical Observation Identifi ers Names and Codes) and SNOMED (Systematized Nomenclature of Medicine).
From there, connecting the dots to diagnoses (ICD-10), drugs (RxNorm), and other domains becomes possible.
Blending unstructured and structured data is even more complex. Plans can’t afford the time it takes to obtain the data, clean it, and map it to the correct format or standard. Manual efforts to do all of that are resource-intensive, and the data is largely out of date by the time decisions are being made.
Clinical NLP: Th e Missing Link
Today, machine learning and clinical NLP are automating many of these processes, helping normalize data, fi ltering out noise (irrelevant data), and bringing important patient information to the forefront. This lets plans view a full picture of a patient’s health in order to manage the gaps in care. AI is becoming an important tool to help plans manage risk and improve performance, but just how can AI make a difference?
According to a 2017 Accenture survey, 72% of payer executives said that within the year, AI would be one of the top three strategic priorities for their organizations.
Payers have the opportunity to unlock billions of dollars in total value in coming years by using AI-driven solutions. But while those payers are leveraging AI and machine learning in ways not possible just fi ve years ago, most organizations have struggled to fi nd the proper focus for their AI initiatives—in part because off-the-shelf AI and NLP technologies aren’t tuned to understand how clinicians represent problems, diagnoses, labs, and drugs within the medical record.
AI is an umbrella term for technologies that perform tasks that simulate aspects of human intelligence. NLP is one important branch of AI, and in the fi eld of healthcare, clinical NLP (cNLP) is an emerging specialized form of NLP that uses sophisticated algorithms to extract clinical meaning from the unstructured data found in a variety of patient medical records such as radiology reports, lab panels, medication lists, patient medical records, physician notes, and even an organization’s own data warehouse.
The challenge of extracting unstructured text and combining that data with broader patient data for a full picture of member health imposes signifi cant limitations for a typical payer, and puts them at a disadvantage in key quality areas: pinpointing at-risk members, modeling a more accurate level of risk, and reducing costs.
This is where cNLP can make a signifi cant difference. Medicare Advantage plans can use clinical NLP to process a patient’s complete medical record and extract key information, such as whether the patient has had a colonoscopy in the previous 10 years. Clinical NLP can summarize patient risk and spotlight high-risk situations based on predefi ned criteria. All of this helps yield clinical insights—in seconds—that are medically relevant, that optimize revenue, and that drive down costs for high-risk patients.
Uncovering Insights From Unstructured Data
Without cNLP, it’s nearly impossible to fully identify risk or proactively target gaps in care and therefore accurately meet the measures needed to improve performance, because so much of this information is stored in unstructured text.
For example, health plans want to target patient populations with multiple chronic conditions, to improve clinical outcomes, enhance patient experience, and reduce expenses. One way to achieve that is to offer patients cost-effective and convenient healthcare services through homebased care. Transparency is key to such initiatives because payment arrangements depend on whether a program is closing gaps in care.
Another important consideration for plans is access to information needed for their risk adjustment models, to accurately determine beneficiaries’ health risks. This is key to ensuring that plans are reimbursed properly for care provided to their patient populations. Using cNLP to automate the identification of relevant information within progress notes and medical histories, and build a workflow around the identification of key concepts in these narratives would enable plans to prioritize their reviews and focus on important areas for accurate risk adjustment.
Clinical NLP can also be a game changer when it comes to CMS initiatives in support of individuals under certain SDoH conditions, such as air conditioners for people with asthma, or home food delivery for homebound patients. How well plans can identify the needs of their beneficiaries— especially those who fly under the radar—could be an important advantage, but it’s also extremely difficult to accomplish. For example, one plan is offering to have a college student visit an older adult identified as lonely. Leveraging NLP to search for this type of information in medical notes and then target a beneficiary who would benefit from visits would be a powerful differentiator.
Clinical NLP can help extract meaningful data for mapping to standard healthcare terminologies for interoperability and integration into central databases, enabling real-time access and analytics. And by reconciling structured and unstructured information in seconds—rather than hours or days— clinical NLP dramatically accelerates access to a complete patient view, something that has eluded many health plans and providers.
One Step Ahead: Innovation Checklist for Health Plans
There are several key steps that health plan executives can take to capitalize on unstructured data by implementing clinical natural language processing.
- Break Down the Silos: This needs to be considered from both a data standpoint and a functional standpoint. Progress will be made when organizations can blend data from across all claims, clinical, administrative, and SDoH systems, and combine deep clinical expertise with data and analytics expertise in order to help unravel data tangles and shift the organization towards a prevention/intervention mindset.
- Compete on Knowledge, Not Data: View data as a shared asset with provider networks and other third parties that manage SDoH data because successful data management requires both collaboration between existing players and the recognition that no single source is big enough to provide all the answers. Payers have data about patient visits across various care settings that would be benefi cial for clinicians, and providers have granular clinical data that would benefi t payers. The goal should be to extract enough relevant data to help members make better choices about their care, optimize their plan benefi ts, and lower their costs of care.
- Move Toward a More Transparent Data Model: Payers are struggling with how to best deploy multiple value-based payment arrangements with providers aimed at caring for the same member population. Without data, how does a plan appropriately pay for value when the plan doesn’t know which of the programs affected the member’s health? Coordinating these programs is critical because payers will have to maintain, manage, and share all of the clinical and patient-generated data to be sure they’re not paying for the same value more than once.
- Define the Plan: Evaluate options to determine the best clinical NLP-enabled model, including a go-it-alone approach, a platform model, or the integration of cutting-edge technologies into existing systems. Identify opportunities to automate the identifi cation of specifi c and relevant information that lies within progress notes or medical histories.
- Prove It, Then Scale It: Launch pilots within specifi c functions so that the use cases for cNLP are well-defi ned and lead to value for the member, the health plan, and provider network.
- Align the Organization: Make sure a strong data governance structure is in place so that all functions are consistently managing the data, normalizing it, and mapping it to industry standards.
A Strong Foundation with Clinical NLP
A strong data management foundation translates into benefi ts for the payer, its provider network, and plan members. With an agile approach to the blending of data from across both structured and unstructured text and then making sense of that data using cNLP, teams will have the knowledge to help their organizations comply with regulations, improve quality ratings, better understand the health of their populations, and improve data-sharing practices across their provider networks.