IBM Consulting Prior Authorization OpenAI Solution: 4 week MVP

Prior Authorization involves multiple stakeholders and requires both simple and complex processing to achieve end-to-end automation potential.

“Complex Processing” utilizing structured data pertaining to Provider, Patient, Procedure, etc AND unstructured EMR / clinical attachment data to digitize, understand, and rationalize clinical information to render decisions (i.e., Criteria-Level and Case-Level Decisioning). Data extracted can also enhance simple processing capability and drive improvement in 360 degree understanding of the patient.

For complex medical procedures which require Prior Authorization, payers have established clearly outlined "Medical Necessity" documents and criteria.

When providers submit patient medical records, finding specific information related to each medical necessity criteria is difficult and time consuming.

IBM Consulting's Prior Authorization solution provides the automation using Azure OpenAI for

• Reading the medical necessity criteria from policy documents
• Converting them to OpenAI prompts
• Getting the medical necessity criteria questions answered from patient medical records
• Provide the evidence for answers from the medical records (explainability and audit)
• Ability for Nurses and Doctors to ask questions dynamically and get answers from patient records

Value Proposition:

Payer/Provider:

• Speed to Value: Improve time to process prior authorization requests; reduce administrative burden

• Improve Clinician Experience: Decrease Clinician Burden (Provider & Payer) by reducing "think time" by utilizing job aid to support decisioning

• Enable MDs/Nurses to operate at Top of their License:  Focus on the complex criterion and cases requiring their expertise

• Assist in Approval / Denial Communication: Apply Generative AI to create an initial approval and/or denial letter for review

• Explainability: Provide reasoning and reference to the Clinical documentation. Deliver evidence-based traceability to build trust

• Conversational Engagement: Quickly find and display answers in natural language for each determination criteria

Foundation Model Benefits

• Model Training Acceleration: Improve time to train AI compared to “traditional” NLP (Clinical) models

• Criterion Extraction Acceleration: Automate extraction of criterion from policy documents

• Computational Efficiency: Ability to support process sequencing in parallel