Prior Authorization NLP and Tenasol

Prior authorization NLP (or “prior auth NLP“) is the use of language processing in the assistance of expediting prior authorization. If you are unfamiliar with prior authorization, please view our blog on the subject as a prerequisite.

To start:

• Prior authorization NLP promotes better star rating. By improving approval timelines, member experience metrics are improved, as well as reduction of member churn trends:

• Prior authorization timelines have recently been accelerated. This has been mandated by the federal government to accelerate patient care and possibly to reduce health plan float. Previously prior authorizations were permitted to take 14 days, but are now being shortened to 7 days.

• Automated denials with NLP are not something we do. There is substantial legal action taking place in regards to automated denials of patient care. Please view our blog on the paradox of healthcare AI to better understand why these practices are so dangerous. Instead Tenasol takes the approach of finding approval or denial evidences, and presenting them to a user and/or participating in some automated approval processes.

Auto-Approval with Prior Authorization NLP

Auto approvals offer substantially less patient risk than auto-denials. The majority of the risk involved in auto-approvals is related to either overpayment by the health plan or adverse effects associated with a treatment that conflicts with a patients medical history, for example the mixing of medications. Usually this second risk is evaluated by the practitioner, heavily mitigating it.

Auto approvals work as follows:

For Tenasol to pursue an auto approval or to detect evidence that can be presented for an approval or denial scenario, we require two ingredients:

• Clinical Guidelines for prior authorization: These are a form of explanation of benefits (EOB) whereby legal thresholds are defined for when a patient is covered by a health plan for a medication or procedure. Oscar Health’s are listed here. It also references billing codes (diagnosis codes like ICD-10) that are associated with a patient being approved, as well as codes that are examples of a patient who would not be approved.

• Patient medical records: These may come in the form of PDF, image, HL7 CDA, or HL7 FHIR. In the event of PDF records, Tenasol first performs OCR. Medical records contain billing information, structured data, and unstructured data, all of which have narrative that is important for the prior authorization process. It is important to note that medical records are collected for prior authorization on a best efforts basis. It is possible that a health plan is unable to acquire all past medical records of a patient required to perform a prior authorization due to limits of its interoperability network.

With these two resources in hand, Tenasol can commence a prior authorization evidence NLP detection process.

Data Extraction for Prior Authorization NLP

Tenasol first runs the patients medical records, regardless of volume and format (medical records can go as high as 50,000 pages in length), to perform data extraction. Information extracted includes:

• Patient information

• Code detection

• Machine learning / LLM detection that is specific to the prior authorization.

Tenasols multi-modal NLP engine

Data extracted is then:

• evaluated to validate that the records are for the same patient as the prior authorization is being evaluated for

• reduced to information that is relevant to the prior authorization and

• consolidated to either:

  • JSON (optionally FHIR as USCDI or Base R4) or

  • Highlighted PDF that has highlights indicating approval or denial evidences in color-specific formatting based on the typed of evidence (green for approval, red for denial). Reviewers are able to jump through processed records quickly to see these evidences. OR

  • Custom format. Depending on the scenario, Tenasol can construct custom export types to accommodate client needs.

Conclusion

Tenasol’s Prior Authorization NLP represents a significant advancement in healthcare technology, designed to streamline and enhance the prior authorization process while prioritizing patient safety and regulatory compliance. By leveraging clinical guidelines and patient medical records, Tenasol’s multi-modal NLP engine extracts, validates, and consolidates relevant data into actionable formats such as JSON, FHIR, or highlighted PDFs, facilitating swift and informed decision-making for approvals.

Unlike controversial automated denial systems, Tenasol’s approach focuses on evidence detection for approvals and denials, emphasizing transparency and ethical standards. This method not only aligns with federal mandates to accelerate prior authorization timelines but also improves patient outcomes and health plan efficiency. Tenasol’s dedication to precision and flexibility underscores its commitment to supporting healthcare organizations in navigating the complexities of prior authorization, ultimately fostering better member experiences and compliance in an ever-evolving regulatory landscape.

Contact us for more details.