Overcoming prior authorization challenges in healthcare Payer plans
with CitiusTech and Snowflake Cortex AI
Dec - 24
Article
Overview
CitiusTech, a leading healthcare technology and consulting provider, plays a pivotal role in driving digital innovation and improving outcomes across the healthcare ecosystem. As a strategic partner to some of the world's largest healthcare organizations, CitiusTech has deep expertise in prior authorization workflows and associated utilization management (UM) applications, delivering innovative solutions that accelerate the prior authorization process for Payers.
By leveraging Snowflake's Cortex AI Service, CitiusTech has developed a Prior Authorization (PA) Generative AI solution designed to help UM staffs, physicians, and clinical reviewers efficiently handle PA requests. In this blog post, we'll explore how this solution addresses the challenges of the prior authorization process and how it can transform healthcare operations.
The problem: Challenges of prior authorization in Health Plans
Prior authorization is a crucial cost control mechanism for health plans, requiring clinical reviewers to meticulously assess members' clinical history to determine the medical necessity of prescribed procedures or medications. Despite its importance, the PA process is plagued by several challenges:
- Manual document review: Reviewing documents related to PA requests is a time-consuming and labour-intensive process.
- High administrative costs: The manual nature of the process increases administrative expenses, reducing overall efficiency.
- Risk of human error: Human involvement increases the likelihood of inaccuracies, leading to potential errors in authorization decisions.
- Longer Turnaround Times (TAT): Patients experience delays in receiving necessary authorizations, affecting their care journey.
- Increased burden on clinical reviewers: The workload for medical reviewers becomes overwhelming, reducing productivity and impacting decision-making.
The CitiusTech solution: Enhancing Prior Authorization with Cortex AI
CitiusTech's Generative AI Payer Prior Authorization solution, built using Snowflake's Cortex AI service, addresses these pain points by generating a summarized and relevant synopsis of members’ clinical history. The solution aims to:
- Reduce the administrative overhead required for PA approvals.
- Enhance member experience by improving turnaround times (TAT).
- Provide clinical reviewers contextualized information at speed.
Key features of the CitiusTech solution
- Clinical data summarization: Extracts and summarizes clinical information from case documents, aligning with health plan coverage guidelines to aid reviewers in making accurate and faster determinations.
- Guidelines and responses: Provides PA request guidelines based on CPT codes and generates responses to determine if a patient meets the guidelines.
- Clinical reviewer Q&A Assistant (Chatbot): Offers a chat-based interface to answer specific questions about members, reducing the time needed for processing PA requests.
- Clinical documents reference and letter generation: Offers citations from clinical documents to validate responses and helps generate approval or denial letters based on reviewers' decisions, including specific reasons for each outcome.
Solution architecture:
The architecture is divided into two key stages that work seamlessly to handle clinical document processing: Data Preprocessing and Embeddings, and Prompt Construction and Inference.
1. Data preprocessing and embeddings:
In this stage, we have established a robust prior authorization data pipeline to efficiently preprocess clinical documents and generate vector embeddings. This event-driven pipeline leverages Snowflake streams and tasks, ensuring near real-time processing of documents as soon as they become available in external storage locations like AWS S3 or Azure ADLS.
To achieve comprehensive document processing, we utilize Snowpark python APIs alongside PyPDF2 and Langchain Python libraries. The resulting data is indexed and served through the Snowflake Cortex Search Service, which enables fast and accurate retrieval within the architecture.
2. Prompt construction and inference:
The second stage harnesses the power of Snowflake Cortex LLMs to generate detailed clinical document summaries, clinical guidelines and responses, and automate letter generation tasks.
We have implemented a clinical reviewer Q&A assistant using the Retrieval-Augmented Generation (RAG) architecture, which ensures that the chatbot maintains chat history and provides highly contextualized responses by leveraging the Cortex Search Service and LLMs.
Leveraging Snowflake Cortex AI
Snowflake Cortex AI is at the core of the CitiusTech Prior Auth solution, enabling healthcare organizations to build generative AI applications quickly using managed Large Language Models (LLMs) and Cortex Search service. The AI-powered platform offers multiple interfaces, such as no-code, SQL, and Python, allowing different users to engage with AI models seamlessly.
Key Snowflake features used in the solution
Cortex LLM functions: Serverless SQL/Python functions that run inference on performant LLMs across every model-size category. We have used these functions to generate document summary, guidelines, and responses. The solution leverages advanced LLMs from Mistral and Reka to process clinical data accurately and generate insightful summaries, enhancing the ability of clinical reviewers to make informed decisions efficiently.
Snowpark: Set of libraries and code execution environments that run Python and other programming languages next to your data in Snowflake. We have leveraged Snowpark Python libraries to pre-process the clinical documents(pdf) and built automated pipeline.
Streamlit: Turn data and ML models into interactive apps with Python—now all in Snowflake. We have built the front-end for PA using Streamlit.
Streams and tasks: We utilized Snowflake streams and tasks to seamlessly orchestrate our CDC (Change Data Capture) pipelines. Streams effectively track the arrival of new clinical documents in the external stage, while tasks are automatically triggered to preprocess these documents as soon as they are detected, ensuring near real-time and efficient processing.
Snowflake Cortex search: Enhancing data retrieval and analysis
The CitiusTech solution integrates Snowflake Cortex Search to power high-quality, low latency "fuzzy" search across Snowflake data. This feature enables efficient search experiences, especially when leveraging Retrieval Augmented Generation (RAG) applications with Large Language Models (LLMs). Cortex Search offers a hybrid search engine that handles both vector and keyword searches, ensuring rapid access to relevant clinical data.
Key benefits of Cortex search in this solution
Low-latency search: Ensures quick retrieval of clinical information, reducing delays in the PA process.
High-quality search experiences: Enables clinical reviewers to interact with data more efficiently, enhancing the quality of review and decision-making.
Value delivered by CitiusTech and Snowflake Cortex AI
A. Payer value proposition
- Reduced administrative overhead: Automatically extracts and summarizes relevant clinical information from case attachments, reducing manual efforts for medical necessity evaluations based on health plan guidelines.
- Faster information retrieval: Clinical reviewers can access answers to specific questions about members via a chat-based interface, significantly speeding up the PA process.
B. Member value proposition
- Reduced turnaround time (TAT): Faster processing of PA requests leads to an improved member experience and builds a stronger Net Promoter Score (NPS) for the healthcare payer.
Use case: How CitiusTech and Snowflake Cortex AI accelerate prior authorization
Imagine a scenario where a clinical reviewer needs to evaluate a PA request for a patient with complex medical history. Traditionally, this process would involve sifting through multiple documents to gather necessary information, resulting in delays.
CitiusTech’s Gen AI solution enables:
- Automatic summarization: The AI extracts key details from clinical documents, presenting a concise summary aligned with health plan coverage guidelines.
- Guideline compliance: The system checks the patient’s data against PA policies/guidelines and generates responses to ensure compliance.
- Interactive Q&A: The reviewer engages with the AI-powered chatbot to get quick answers, speeding up decision-making.
The result? The PA request is processed in a fraction of the time, reducing the administrative burden, and enhancing patient care.
Conclusion
CitiusTech, in collaboration with Snowflake Cortex AI, offers a transformative solution to the challenges of prior authorization in healthcare payer plans. By leveraging AI-driven capabilities, the solution minimizes administrative overhead, reduces turnaround times, and enhances the accuracy of PA approvals, ultimately improving the experience for both clinical reviewers and members.