5 Considerations For Health Plan Leaders Using AI-Enabled Prior Authorization And Utilization Management

Amber Nigam is CEO and cofounder of basys.ai, a Harvard-based company streamlining prior authorization for health plans with generative AI.

Health plan leaders uphold responsible guardrails that enable affordable and quality care. However, with the rising cost of healthcare in the United States, this can be challenging. Two essential tools for overcoming this challenge are utilization management (UM) and prior authorization (PA). UM is an umbrella of processes in which health plans review the necessity and appropriateness of medical services. PA is a UM process whereby plans review and approve requests for treatment coverage. My company uses artificial intelligence (AI) to streamline this process, and our health plan partners have seen firsthand some of the benefits. However, there are also potential risks that leaders need to carefully navigate when deploying AI for PA.

Here are five considerations health plan leaders can take into account to sustainably deliver affordable and effective care with AI for PA:

Reducing Administrative Burden: Streamlining Workflows While Retaining a Human Touch

McKinsey & Company reports that anywhere from 50–75% of manual tasks associated with PA can be automated. By handling routine authorizations and streamlining workflows, AI not only accelerates the PA process but also minimizes the risk of manual data entry errors. However, when using AI to streamline the PA process, there is a risk of over-automation and short-circuiting expert opinions for sensitive cases. To address this, health plan leaders can consider a “human in the loop” approach. While retaining a human reviewer may slightly reduce efficiency, it is necessary to maintain balanced judgment, especially for complex cases.

Promoting Affordability and Access: Balancing Medical Expenditures with Care Quality

By triaging care and recommending cost-effective alternatives for unnecessary treatments, AI can improve affordability and accessibility. Using AI to streamline PA presents the opportunity to save up to $418 million per year for members. However, a risk lies in sacrificing care quality for affordability. Population trends of medical necessity can miss life-threatening edge cases, especially for conditions with many subtypes, such as cancer. To prioritize member outcomes while maintaining affordability, health plans can train AI to personalize the care journey to each member’s needs.

Encouraging Provider Buy-In and Member Transparency: Avoiding Hallucinations while Automating Explanations of Decisions

Generative AI can resolve member and provider queries while delivering explanations for coverage decisions that are grounded in medical and scientific literature. While this can improve member and provider alignment with the PA process, generative AI sometimes has a propensity to “hallucinate” false information. The risk of hallucination is considerable: ChatGPT’s hallucination rate is currently between 20-25%. According to the New York Times
NYT
, tech companies are attempting to combat hallucination by grounding chatbot answers in a body of knowledge. This complicated process requires ongoing oversight, but incorporating rules on the corpus of relevant literature that AI may reference and procedures to evaluate the algorithm’s output are steps to maintain factual PA explanations.

Achieving Accuracy: Addressing The Needs of Both The Overall Population and Marginalized Communities

AI can analyze vast amounts of data to recommend clinically appropriate care pathways with high accuracy. For example, an AI model predicting breast cancer risk from mammograms made the news in 2019 for achieving a greater diagnostic accuracy (paywall) than the prior medical standard statistical model. The caveat is that AI can appear accurate on a macro level while ignoring the needs of marginalized or underrepresented populations. Strategic initiatives for health equity can include defining training guardrails for AI-enabled PA and UM to prevent algorithmic bias from creating disparities in care access. One such guardrail would be training AI on not only population-level historical data but also attributes of social determinants of health.

Enabling Interoperability and Agile Integration: Balancing Data Exchange With Security

By driving rapid integration through data standardization and normalization, AI can enable faster communication across health plan and provider IT systems and equip plans with robust data for PA and UM. Recognizing this potential, major EMR providers like Cerner
CERN
, Epic and Athena are integrating AI into their systems and international efforts are ongoing to consolidate billions of medical records through natural language processing (NLP).

Although AI enables agile integration, this agility can create vulnerabilities in data security. Over 40 million members were impacted by data breaches in 2022, with health plan infrastructures being the primary targets of 14% of breaches. It’s important that health plan leaders follow clear data exchange governance standards, such as DaVinci’s HL7 FHIR, and implement robust security protocols to provide a seamless integration experience without risking data breaches.

In conclusion, applying AI to PA and UM can improve both the affordability and quality of care for members. However, carefully designed protocols are necessary for effective AI implementation and include strategies like balancing automation with human expertise through a “human-in-the-loop” approach, personalizing care to member needs, grounding generative AI output in a knowledge base, overcoming algorithmic bias through ethical AI training guardrails and maintaining data security through rigorous data exchange standards. With such implementation strategies, health plan leaders can leverage AI-enabled PA and UM to optimize both member savings and outcomes.

Forbes Business Council is the foremost growth and networking organization for business owners and leaders. Do I qualify?