CMS Clarifies Boundaries on AI in Coverage Determination for Medicare Advantage Organizations

On February 6, the Centers for Medicare and Medicaid (CMS) published an important FAQ for Medicare Advantage Organizations (MA) and Medicare-Medicaid Plans intended to clarify ambiguities identified by the public in the “Medicare Program; Contract Year 2024 Policy and Technical Changes to the Medicare Advantage Program, Medicare Prescription Drug Benefit Program, Medicare Cost Plan Program, and Programs of All-Inclusive Care for the Elderly” which provisions went into effect January 1, 2024. 

MA organizations have historically employed certain software tools, including artificial intelligence (AI)-powered decision-making tools, to assist with informing care-related decisions, such as predicting lengths of stay, making diagnoses, or offering recommendations. Traditionally, these systems utilize prior patient experiences and current patient conditions to inform care recommendations. However, official CMS coverage criteria are necessary for determining coverage. These criteria may or may not be in line with the criteria used by increasingly sophisticated algorithms and AI-powered applications. Assuming MA organizations are permitted to continue to use these powerful tools to support their operations, the question becomes – to what extent can algorithmic or AI determinations be relied upon by MA organizations in making care and coverage decisions? 

In the recently published FAQ, CMS addressed this question. Specifically, CMS answered whether MA organizations are permitted, under the new rules on clinical coverage criteria for basic Medicare benefits, to use “algorithms or artificial intelligence to make coverage decisions.”  Ultimately, CMS concluded that algorithms or AI may be used in making coverage determinations but emphasized that the onus remains on the MA organization “to ensure [AI and algorithms] comply with all applicable rules for how coverage determinations…are made” and that neither tool should be the only foundation upon which MA organizations rely when determining coverage. CMS expressed concern that prevalent and unrestricted use of algorithms or other software tools risks “shift[ing] coverage criteria over time,” which effectively puts the “static and unchanging” coverage criteria publicly posted under CMS’s final rules in opposition with what is happening on the ground. Therefore, CMS reminded MA organizations that the hallmark of any coverage determination must be the “permissible applicable coverage criteria” set forth in the “publicly posted” rules. Such determination cannot be made based on “internal coverage criteria that have not been explicitly made public and adopted in compliance with the evidentiary standard[s]" set forth in CMS’s final rules.

Finally, CMS reminded MA organizations that MA organizations remain subject to Section 1557 of the Affordable Care Act, which “prohibits discrimination on the basis of race, color, national origin, sex, age, or disability in certain health programs and activities” and that such organizations must, prior to implementation, “ensure that the tool is not perpetuating or exacerbating existing bias or introducing new biases.”

CMS partners and vendors, including MA organizations, must be cognizant of these requirements when seeking to leverage the power of AI in advancing their service offerings and improving care quality. To better enable this effort, CMS has published guidance materials and aggregated foundational governance documents that incorporate many of the compliance requirements for AI applications. Notably, the CMS AI Playbook and HHS Trustworthy AI Playbook review principles of Responsible AI, best practices, and step-by-step instructions for building mission-critical, trustworthy AI solutions.

MA organizations should also pay close attention to other questions and topics addressed by CMS in the FAQ. The full FAQ can be found linked above and here.