CMS’s Enforcement on AI Utilization in Medicare Advantage
In the realm of Medicare Advantage (MA), the integration of artificial intelligence (AI) and algorithms by insurers to facilitate coverage determinations has been a subject of scrutiny by the Centers for Medicare & Medicaid Services (CMS). While acknowledging the potential benefits of AI in decision-making processes, CMS has issued a stern warning to MA insurers, emphasizing the imperative adherence to the agency’s internal benefits prerequisites and anti-discrimination regulations mandated by the Affordable Care Act (ACA) to ensure lawful care denial practices.
The recent cautionary directive directed towards MA insurers pertains to apprehensions regarding coverage criteria and utilization management prerequisites outlined in CMS’s final rule policy, initially disseminated in April last year.
Expressing apprehension, CMS articulated in a memo, “We are concerned that algorithms and various new artificial intelligence technologies have the capacity to exacerbate biases and discrimination.” The memo further iterated, “MA organizations must ascertain, prior to the implementation of any algorithm or software tool, that such tools do not perpetuate existing biases or introduce novel biases.”
Deciphering CMS’s Message
Within the memorandum, CMS elucidates several pivotal points, notably drawing a distinction between an algorithm and an AI model.
AI is defined as “a machine-based system capable of making predictions, recommendations, or decisions that influence real or virtual environments based on predefined objectives,” whereas algorithms are characterized as technological entities that “can imply a decisional flow chart comprising a series of if-then statements.”
CMS underscored that although algorithms can assist in forecasting post-acute service duration, they are not authorized to dictate coverage termination. Concerning inpatient admissions, the agency clarified that algorithms and AI in isolation are insufficient grounds for admission denial or observation stay downgrade.
Furthermore, CMS stipulated that while payers may leverage algorithms to aid coverage determinations, they must ensure compliance of the technology with the agency’s coverage decision prerequisites. MA payers are prohibited from basing decisions solely on extensive datasets; instead, decisions must be grounded in individual patient medical histories, provider endorsements, and clinician annotations.
Navigating the Trajectory
In the year 2023, the federal government repeatedly emphasized to MA insurers the prohibition of AI deployment for care denial purposes. Notably, prominent entities such as Cigna, Humana, and UnitedHealthcare faced legal actions for purportedly misusing AI to unjustly repudiate claims.
This issue has garnered congressional attention, with Senator Elizabeth Warren condemning MA insurers for exploiting AI to bolster profits through wrongful claims denials. Stressing the necessity for stringent regulations, Warren highlighted the potential detrimental impact of unchecked algorithms on patients and insurers’ financial gains.
At a recent hearing, Senator Warren probed Michelle M. Mello, JD, PhD, a health policy and law professor at Stanford University, on the actions CMS should undertake to prevent insurers from illicitly leveraging AI tools for care denials. Mello advocated for enhanced CMS audits in 2024, focusing on denials, and suggested the establishment of clear guidelines for appropriate AI utilization.
In light of the evolving regulatory landscape, MA payers must diligently adhere to these directives to avert future scrutiny regarding denials and AI utilization. CMS has also introduced a dedicated AI resource website for insurers to navigate these guidelines effectively.
The integration of AI is just one facet of the transformative shifts occurring in the MA domain. Apart from technological advancements, MA payers must remain vigilant regarding other key developments shaping the sector in the current year.
Marie DeFreitas serves as an associate content specialist at HealthLeaders.
