Welcome to the first installment of a four-part blog series that explores the broad and growing value of using clinical natural language processing (cNLP) to deliver results for both payer and provider organizations.
I hope you read our last blog introducing you to the concept of Clinical Natural Language Processing or cNLP.
In the introduction, we highlighted the common need for payers and providers to manually comb through volumes of patient records in order to unlock the clinical detail found in the unstructured narrative of the medical record. There are several reasons an organization would want to invest in the resources required to do this. Arguably, the one we hear about most is the need to capture the level of acuity found in the patient population that they serve through risk adjusted contracts.
The methodology used by the Centers for Medicare & Medicaid Services (CMS) to calculate risk scores is one familiar to coders and CFO’s alike. Hierarchical Condition Categories or HCC’s are used in contracts with Medicare Advantage (MA) plans. The long and short of it is that higher HCC scores identify more complex patients and allow for additional payment from CMS to care for those patients.
I think we all recognize that free text narrative will continue to be an important part of the patient story. It is simply not possible to capture rich clinical detail any other way. Busy clinicians should not be asked to apply the complexity of coding rules and guidelines required to meet the high standards of ICD-10 coding, let alone the additional complication of capturing HCC’s in a code. After all, it should already be in the documentation. This is where those highly skilled resources are deployed. It is their responsibility to translate the documentation into ICD-10 codes that risk adjust when the patient condition and documentation warrant it. When the documentation is not complete or detailed enough, these professionals need to query the treating clinician in order to fill in the gaps. Currently this process is mostly done retrospectively, but wouldn’t it be nice to prospectively identify gaps in documentation and prompt the clinician at the point of care to fill those gaps? But I digress.
Without this retrospective review, healthcare organizations may find themselves out of compliance with HCC coding guidelines, which could mean they are leaving money on the table.
Understanding what's at stake with risk adjustment
Health insurers use risk adjustment methods to offset the cost of high-risk populations—generally those with chronic conditions. As discussed earlier, for Medicare populations CMS uses an HCC model to rank diagnoses into categories that represent conditions with similar cost patterns.
Long-term conditions such as diabetes, chronic obstructive pulmonary disease (COPD), and chronic heart failure (CHF) will “risk adjust,” or fall within a specific HCC; whereas acute illnesses and injuries will generally not because acute conditions are not reliably predictive of ongoing healthcare costs. Risk scores measure individual beneficiaries’ relative risk and are used to adjust payments for each beneficiary’s expected expenditures.
To factor into risk adjustment, a diagnosis must be based on clinical medical record documentation from a face-to-face encounter, documented at least once per year, and coded according to ICD-10-CM guidelines.
In these days of a public health crisis, visits may now be performed via telehealth when medically appropriate. Let’s hope that continues long after the public health crisis ends. Accurate documentation and the ability to capture that information in ICD-10-CM codes is foundational to optimizing the level of reimbursement associated with risk adjustment. All ICD-10 codes must be captured within a calendar year. This means that the clinician must document each condition each year to be considered in the risk adjustment calculation.
Examples of risk calculation adjustments
Let’s take the example of a below-the-knee amputation. In year one of the risk adjustment contract a patient has a documented and codified amputation status. In year two, if the clinician does not document that same amputation status, then the amputation does not technically exist for the purposes of HCC calculations for year two.
Other examples of common data elements that are often found in free text but not codified to ICD-10-CM, and therefore could impact a risk adjustment score are:
- Parkinson’s Disease
- Rheumatoid Arthritis
- Atherosclerosis of Aorta
- Alcohol & Drug Dependency (even in remission)
- Morbid Obesity (BMI >40)
- Transplant Status
The challenge of extracting insights from unstructured data
As previously mentioned, current methods for extracting valuable information from free text are often characterized by manual, error-prone processes that require excessive resources. Commonly done by teams of clinicians and coding professionals utilizing keywords to comb through patient charts and documents, manual chart reviews for a single patient record can take anywhere from 30 minutes to three hours depending on the information collected. In addition, manual processes by their very nature are error prone.
Two common risk adjustment use cases
1.) Information on a diabetic patient can come from the structured ICD-10 code in the EHR or claim, and present as “E11.9, diabetes without complications.” By understanding HCC content and structure, cNLP can be applied to help determine whether there are additional non-coded complications present in the notes, such as dry mouth.
Once identified and validated by the clinician as a complication of Diabetes, this information empowers the correct designation of HCC18—diabetes with a chronic complication. This process not only improves the accuracy of HCC reporting but also elevates reimbursement to the appropriate level based on the patient’s condition severity.
2.) The clinical presentation of Chronic Kidney Disease (CKD) is in most cases considered, nonspecific. This condition may present with somewhat generic symptoms such as hematuria, anemia, and peripheral edema. Progression into later stages of CKD go unnoticed in general documentation resulting in the condition being coded as N18.9 Chronic Kidney Disease, unspecified.
Properly trained cNLP technology can identify the commonly used test, glomerular filtration rate or GFR, and identify the value associated with that lab test to trigger the clinician to document stage two or three CKD and increase the level of reimbursement and care quality for these patients.
How Health Language solutions can help
The Health Language cNLP Solution delivers the industry’s most comprehensive approach to identifying and leveraging unstructured data and is easily integrated into health IT environments through application programming interfaces (APIs). Through functionality that enables extraction, conversion and mapping of free text to industry standards, the application ensures interoperability and meaningful exchange of data assets that have the power to provide deeper analytics insights.
The solution features industry-leading functionality such as:
- Proprietary cNLP Lexicon library containing over 1 million provider-friendly terms including clinical synonyms, acronyms and common misspellings
- Embedded contextual awareness to support proper identification of the context surrounding clinical information
- Best-of-breed tool sets configured to desired client use cases, all coded to a single API
- Knowledge of multiple terminology domains out of the box
In addition, clients have access to the Health Language team of clinical informaticists who can rapidly assess needs and recommend optimal configuration for immediate results.
When payers and provides deploy infrastructures and processes to achieve high data quality, they can more effectively extract value from the wealth of information that resides in free text. Speak to an expert to learn more about how the Health Language cNLP Solution can help your organization leverage unstructured text.
Learn more in our next blog in the cNLP series, where we will dive into how cNLP can support healthcare organizations with increased accuracy and efficiency for quality measure reporting.