RISE Risk Adjustment Forum 2024: Fathom panel on autonomous risk adjustment coding

At the RISE Risk Adjustment Forum in Louisville (Nov. 17-19, 2024), Andrew Lockhart (Fathom) moderated a panel with RaeAnn Grossman (HLTHworks), Carrie Horn (Baylor Scott & White Health), and Dr. Sunita Varghees (Baylor Scott & White Health) about the latest developments in RA coding tech. This recap synthesizes key takeaways and advice from the discussion.

Key takeaways

1. MA expansion and persistent workforce challenges create an urgent need for automation. Traditional approaches can't keep pace with rising demand and quality requirements.
2. Technology has matured beyond basic automation. Deep learning enables 90%+ chart automation at high accuracy levels. AI can now handle complex risk-adjustment scenarios effectively.
3. Successful adoption requires structured implementation. Define clear objectives based on deep understanding of pain points. Start with a pilot to validate performance. Maintain rigorous QA through deployment and scale-up.
‍

Summary

1. Industry updates

Medicare Advantage enrollment driving demand. MA enrollment has more than doubled in the last 10 years, and is forecast to grow another 35% by 2030. This growth, combined with expanding ACO and value-based contracts, increases pressure on risk-adjustment operations.

Workforce crisis strains quality of coding. A 30% coder shortage, coupled with vendor consolidation and inflation, has caused labor costs to increase. At the same time, coding accuracy has decreased as organizations have had to increasingly rely on less experienced coders and higher productivity metrics to compensate for staffing gaps.

Regulatory oversight is intensifying. The growth in MA spending has led CMS to increase its auditing efforts and compliance requirements, motivating health plans to scale up review and validation of coding before it goes out the door. As RaeAnn noted, "Extrapolation is a scary thing. We need to be complete and accurate."
‍

2. Technology evolution

To support risk-adjustment coding, technology has developed in three main phases.

2000s to mid-2010s: Early NLP-based tools show promise. These technologies enable automation for simple coding scenarios, but struggle with the complexity and chart length involved in risk-adjustment coding.

2015-2018: ICD-10 creates major disruption. The ICD-10 transition in 2015 increases the number of codes tenfold, causing automation to fall off a cliff – even for basic cases. Organizations begin limiting technology usage due to poor coding quality and compliance risk.

2018 onward: Deep learning transforms capabilities. First end-to-end automation of coding arrives in the market, starting with provider-side coding before expanding to cover risk-adjustment for payers in 2022. This approach unlocks 90%+ automation at high accuracy.
‍

3. Implementation considerations

Delineate explicit outcome(s) from the beginning. Understand specific pain points and goals (e.g., increase quality scores, improve HCC capture, reduce costs, etc.) at a deep level before considering technology solutions. As Andrew noted, "You get what you measure. If you don't have the outcomes clearly outlined, you won't have the basis for a strong contract."

Pressure-test underlying technology and workflows. Ask vendors to explain how their tech works, since "not everyone is using terms the same way," and pressure-test workflows. Push for transparency around rules and logic. Overall, "AI should be doing more than what it's doing right now" to meet the growing demands on health plan leaders, as one panelist noted.

Validate performance with a pilot before full rollout. Successful adoption requires a methodical approach, engaging compliance teams during the pre-sale process and starting with a thoughtfully designed pilot. In risk-based provider organizations, beginning with a focused group of engaged physicians (e.g., one department) to validate system performance and move past initial skepticism has worked well, in one panelist's experience.

Systematically monitor performance and maintain vendor touchpoints. Organizations need regular touchbases with the vendor to raise and fix concerns and to obtain timely, actionable reporting. Daily touchpoints may be recommended for the first couple of weeks after go-live, with less frequent touchpoints once performance stabilizes. As Andrew shared, "The same work you would put into monitoring any vendor should be put into monitoring an AI system."
‍

4. Impact and outcomes

Align KPIs to stakeholder needs. Tailor reporting to each applicable audience, such as finance, clinical, and specialty groups. This enables organizations to track and improve performance across the different dimensions that stakeholders care about. As one panelist recommended, "Depending on who this tech is touching, have a KPI for each one of those groups – a budget KPI for CFO, a different KPI for provider engagement, and so on." This thoughtful approach can also help to increase credibility for future endeavors.

Early adopters demonstrate clear ROI. A health plan with approximately 2 million members, including 400,000+ MA members, transformed its risk-adjustment operations with autonomous coding. This plan achieved 91.2% full automation of chart reviews, resulting in a 47.8% reduction in vendor coding spend. In contrast to its accuracy challenges with manual vendors, the plan saw a 38% reduction in coding errors and a 27% increase in ICD capture.
‍

5. Strategic implications

Teams evolve toward higher-value work. Autonomous RA frees staff from the daily pressure of mounting chart volumes, unlocking bandwidth to think strategically about coding guidelines and to perform deeper analysis. As one panelist noted, this enables teams to start "working at the top of their license," shifting focus to higher priorities such as quality.

Change management facilitates adoption. Whether in a health plan or provider setting, implementation success hinges on effective stakeholder engagement. Clinical teams generally start skeptical and want to see proof that the technology works, with experiences varying significantly across specialties and organizations. A comprehensive change management strategy must address these varying levels of readiness and surface the needed evidence.

‍

If you'd like to learn more about Fathom, schedule a meeting here.