Generative AI for Prior Authorization Automation

The prior authorization process is a time-consuming and frustrating process that often delays patient care. Furthermore it has increased administrative burden for healthcare providers and is labor intensive for health insurance plans who are required to staff teams, manage manual reviews, and run other administration of complicated cases and appeals. By leveraging both Large Language Models (LLMs) and Large Medical Models (LMMs) provider and health plans can automate large portions of the prior authorization process, allowing for faster and more accurate approval of treatments. This will improve patient satisfaction, reduce administrative burden, and decrease costs for healthcare organizations (payers and providers alike).

For Health Plans and Payers

Compliance Automation

In the Interoperability and Patient Access final rule (85 FR 25510), CMS finalized a policy to require impacted payers to implement a HL7 FHIR Patient Access API. CMS is proposing that payers make prior authorization requests and decisions available to in-network providers via the Patient Access API beginning January 1, 2026. This proposed rule would also require impacted payers to report annual metrics to CMS about patient use of the Patient Access API. The task of translating the current coverage and payment requirements into a format that is compatible with the Patient Access Application Programming Interface (API) is substantial. This translation process is not simple as it involves restructuring and reformatting a large volume of complex and constantly updating information. LLMs can automate a large portion of this mapping process and significantly reduce the manual effort required.

Simulated Prior Authorization for Complex Cases

Prior authorization in complex cases where clear care guidelines are lacking can be very challenging for both healthcare providers and health plans. Unpredictable medical scenarios and inherent ambiguity in care guidelines can lead to delays in treatment approval, which in turn could adversely affect the patient's health. GenHealth’s large medical model (LMM) can be used in this scenario to simulate patient futures based on yes/no prior authorization decisions to provide a visual roadmap of potential outcomes based on different authorization decisions. This not only helps in making informed and timely decisions but also in understanding the possible long-term implications of those decisions on a patient's health and on the healthcare system as a whole.

Communicating Reasons for Prior Authorization Denials

When prior authorizations are denied, providers or their administrative staff often reach out to the payers or payor portals to receive more information about the denial, a process that tends to be complex and laborious for health insurers. While claim denials have deterministic reasons, translating the denial reasons from their raw format output from claims processing engines into plain english can be very difficult. LLM’s can dramatically simplify this translation process. This can simplify the appeal and re-submission process for providers and their administrative staff, and automate a highly manual part of a claims denial-appeal process.

For Providers

Large language models (LLMs) can be trained to parse dense coverage determination documents to provide answers to specific questions. For example, answers to the most commonly asked questions by providers can be answered definitely in seconds by a large language model fine-tuned on a set of coverage determination documents. This not only saves time but also reduces the risk of errors and denials due to misunderstanding or misinterpretation of the documents. These are questions like:

• Is prior authorization required for this medication/procedure?
• What is the eligibility criteria?
• What documentation is needed?

These questions can be directly answered by a large language model for any procedure or prescription with a reference document. The LLM can also provide source citations to the original coverage documents to allow a provider to verify AI-generated suggestions.

Prior Authorization Pre-Check

Large language models can also easily turn complicated coverage documents into simple questionnaires which can be filled out by providers to quickly receive a response as to whether a particular procedure or medication will be covered by a plan.

Easy Interoperability with FHIR

Coverage determinations can be effectively formatted and stored as a simple Questionnaire resources using the FHIR (Fast Healthcare Interoperability Resources) standard. The completed Questionnaire can then be stored in a standard FHIR server for future reference, interoperability, and data analytics. Certified Electronic Health Record (EHR) systems are mandated to support the FHIR standard by the Office of the National Coordinator for Health Information Technology (ONC) and legislation like the 21st Century Cures Act. Using a FHIR Questionnaire for coverage determination not only standardizes the eligibility process, but simplifies implementation using existing infrastructure and systems .

Understanding Coverage Determinations is Time Consuming

Understanding prior authorization requirements is a complex and time-consuming task because they are often published in disparate formats, frequently updated, and vary significantly from one payer to another. This lack of consistency and predictability introduces a significant administrative burden as healthcare providers must continually adapt to changing requirements, decipher complex documents, and navigate the unique stipulations of each individual payer. This complexity not only consumes valuable time and resources but also increases the risk of errors, denials, and delays in patient care.

Staying Up-To-Date With Changing Guidelines

One other major challenge in prior authorization is the need to stay informed and keep up to date with the frequently evolving guidelines and regulations. New updates and changes can occur on a monthly, or even weekly basis, making it essential for healthcare professionals to constantly monitor and adapt to these modifications. Failure to stay up to date can result in delays, denials, and potential compliance issues. For payers, failing to communicate updates to coverage or payment requirements promptly can result in additional work for manual review teams. New large language models, in combination with large medical models can dramatically simplify this process.