Risk Adjustment NLP with Tenasol
Risk adjustment NLP is the use of NLP for risk adjustment, a large component of the US health economy whereby federal resources are distributed to health plans based on membership quantity and critical health conditions.
Tenasol, as a core part of its business offerings, performs risk adjustment NLP and AI evidence detection via records collected through interoperability. Our services include NLP risk adjustment language gap detection, prioritization of records, deduplication of medical records (not shown), code gap detection, and future risk adjustment gap prediction. We will cover each in detail. We do this for Medicaid, Affordable Care Act (ACA), and Medicare Advantage (MA).
NOTE: Tenasol does not simply use keyword searching for its risk adjustment NLP, but rather a wide range of 5 different types of NLP.
Risk Adjustment NLP Gap Detection
HCC NLP Gap Detection entails detection of language that is associated with a hierarchical condition category (HCC) that is not already represented by claims in a patient history.
To perform this, Tenasol uses millions of medical records to train language models for each year, for each line of business (ACA, MA, and ACA), and for each HCC respectively to build one ensemble, per year, per system. Tenasol is capable of systematically training these systems as soon as they are released.
Language that is associated Risk adjustment data may come from:
HL7 ADT messages,
HL7 C-CDA medical records,
HL7 FHIR messages, or
unstructured medical records (PDF/TIF/TIFF/JPG/JPEG/GIF/RTF/DOCX)
Tenasol uses risk adjustment NLP to parse all fields of all record types.
Tenasol performs both global data extraction that is in a later step reduced to risk adjustment-specific evidences, as well as the previously mentioned risk-adjustment NLP machine learning models that independently seek language associated with each HCC.
The below is an example of a detected finding using Tenasol risk adjustment gap NLP for the V2424 model detecting unstructured language associated with breast cancer (HCC 12).
Risk Adjustment NLP Filtering
Once the data is aggregated by Tenasol in the JSON-based extracted and normalized format shown, the output data then may be run up against the associated risk system to determine what information is relevant. This means:
When clients supply known information, those HCCs detected with NLP may be filtered out
Information may be restricted to years the client is interested, masking non-relevant timeframes.
We shape the data into however it is desired, included but not limited to API feed, API with FHIR formatting, CSV, or XLSX.
Tenasol is also capable of highlighting findings in PDF format for medical record review, with multiple color bindings.
Risk Adjustment NLP Prioritization
In addition to the generation of gaps based on risk adjustment NLP, Tenasol also is capable of prioritizing records based on clinical value to a risk review team. This maximizes the amount of risk gaps found within a fixed period of time by a risk review team.
Risk Adjustment Data Deduplication
This offers significant time savings if trying to mobilize a risk adjustment team by reducing the volume of data. Tenasol uniquely performs 5 document similarity measures across all medical records. We have seen some plans deliver datasets with unknowingly as much as 40% of duplicate records in their backlog.
Please see our main blog on medical record data deduplication here.
Risk Adjustment Claims Gap Detection with AI
Somewhat adjacent to the above topics, Tenasol is also able to deduce gaps in care via claims code history. See our blog on coding systems to learn more about these.
In the above example, a patient with diabetes may be detected, even though the patient is undiagnosed within the year for diabetes with or without complications. This is performed by picking up associated codes and values, such as LOINC or CPT codes. For example, the patient may be indicated to have a high blood sugar.
For human evidence review, the Tenasol system further supplies
confidence of each HCC suspected
supporting evidence of each HCC suspected
a boolean indicator describing if the gap is already known
Future Risk Adjustment NLP
Outside of the immediate scope, Tenasol also offers experimental capabilities to determine patients who may possibly have conditions in the future based on past evidence, using constructed timelines of other patients.
While these confidence levels, which are calibrated, are relatively low, they offer insight into patients most at risk of having chronic conditions that will be risk adjustable in the future. Both medical record language and claims code can be used for this process.
Conclusion
Risk adjustment NLP is a critical component of the U.S. healthcare economy, enabling equitable resource distribution by identifying and addressing gaps in medical documentation and claims. Tenasol excels in this domain by leveraging advanced NLP techniques across multiple data formats, such as HL7, unstructured medical records, and FHIR. Through processes like language gap detection, prioritization, deduplication, and claims gap analysis, Tenasol ensures clients can optimize their risk adjustment efforts with precision and efficiency.
Our models, trained on millions of records for Medicaid, ACA, and Medicare Advantage, identify Hierarchical Condition Categories (HCCs) and support human reviewers with high-confidence, evidence-backed findings. Beyond current gaps, Tenasol also explores predictive capabilities to assess future risks. By transforming complex healthcare data into actionable insights, Tenasol empowers health plans to improve compliance, enhance care quality, and maximize resource allocation in an increasingly data-driven healthcare environment.
Reach out to our team for more information!