By: Joseph Valcazar
As of 2024, 32.8 million Americans received health insurance through a Medicare Advantage plan. This accounts for over half of all Medicare recipients. Covering some of the most vulnerable members of the populace. Including senior citizens aged 65 and older, individuals with disabilities, and those with end stage renal disease. It should come as no surprise that these groups are reliant on insurance to cover necessary treatments that would otherwise be too costly. Even with coverage, 13.6% of a Medicare family’s total expenses are health-related. In contrast, for non-Medicare families this figure is 6.5%. Now, health insurance carriers are integrating AI driven predictive models to calculate care plans, which is raising concerns among medical professionals that patients are being denied necessary care, leading to legal action.
What is Medicare Advantage?
Traditional Medicare encompasses inpatient treatment through Medicare Part A, and outpatient treatment through Medicare Part B. Eligible recipients of Medicare are automatically enrolled to receive coverage under part A. Part B coverage is voluntary. Users who choose to participate in Part B pay a monthly premium, determined by an individual’s household income.
In 1997, Congress passed the Balanced Budget Act (BBA) which introduced Medicare Part C, later named Medicare Advantage (MA). The BBA permitted the Center for Medicare & Medicaid Services (CMS) to contract with private health insurance carriers to provide health insurance plans to eligible Medicare recipients. In turn, MA participants would receive full Part A and Part B coverage, just as they would under traditional Medicare, but through a private insurance carrier (think UnitedHealthcare, Blue Cross Blue Shield, etc.). In addition, MA plans could offer supplemental benefits not offered under traditional Medicare, such as dental and vision coverage or gym memberships.
However, under Medicare Advantage, these private companies control all MA related claims, determining how much of received or expected care is covered. This is where the controversial nH Predict model enters the picture.
The nH Predict Model.
Created by NaviHealth (now owned by UnitedHealth Group), the nH Predict model is designed to predict post-acute care needs. Post-acute care refers to treatment for a severe injury, illness, or surgery, typically caused by trauma. The most common post-acute treatments involve visits to skilled nursing facilities (SNF), and home health agencies (HHA).
Investigations of the nH predict model have indicated the model has become “increasingly influential in decisions about patient care and coverage.” While the specifics of the model are unknown, the nH predict model functions by utilizing databases containing millions of medical records, evaluating demographic information such as age, preexisting health conditions, and other factors to determine custom care plans, including duration of treatments.
The utilization of predictive models has garnered concerns from medical professionals and patients alike, who are concerned that an increasing reliance on such models fail to account for the unique individual factors that contribute to a patient’s recovery, leading to inaccurate results. An ongoing class action lawsuit claims the nH predict model has a 90% error rate. The lawsuit also accuses UnitedHealthcare of having knowledge of this error rate and still using the model to override treating physicians’ determinations.
Class Action Lawsuits
Since its creation, multiple health insurance providers have integrated the error-prone nH predict model into their claims process. Many MA patients have filed federal class action lawsuits against major health insurance companies, including UnitedHealthcare and Humana, alleging breach of contract, breach of implied covenant of good faith and fair dealing, and unjust enrichment. The plaintiffs claim that by using the faulty nH predict model, these companies have unfairly denied claims which have directly and proximately caused their damages.
In one claim against UnitedHealthcare, Dale Henry Tezletoff, a 74 year old MA recipient suffered a stroke that required hospital admission. Mr. Tezletoff’s doctor recommended he seek post-acute treatment at a SNF for 100 days. After 20 days of treatment at an SNF, he was informed by UnitedHealthcare that any future treatment would not be covered. It required two separate appeals before a UnitedHealthcare doctor reviewed Mr. Tezletoff’s medical records and concluded additional recovery time was needed. Yet, after 20 more days at the SNF, Mr. Tezletoff was again informed that any future post-acute care had been denied. And this time, even with an opposing opinion from Mr. Tezletoff’s doctor, UnitedHealthcare refused to overrule its decision. As a result, Mr. Tezletoff was required to pay out-of-pocket expenses totalling $70,000 to receive the necessary treatment.
These lawsuits shine a spotlight on the ethical and legal ambiguities of AI in its current state. The legal system is not well equipped to respond on the whim to new complex technological advancements. When a court has the opportunity to hear a case on an emerging issue, it is placed in a position to serve as a voice of authority. A ruling in the plaintiff’s favor would act as a deterrent to similar future conduct. Providing the legislature an additional buffer as they tackle the unenviable task of regulating this new technology.
The fact is, Mr. Tezletoff’s story is not unique, and the implications of these lawsuits are apparent; people’s quality of life is on the line. The outcome of these lawsuits, and the response from the government, will help shape how AI is integrated into the healthcare industry and others like it.
The Government’s Initial Reactions
The federal government has begun to respond to these concerns. On January 1, 2024, the Department of Health and Human Services enacted new rules requiring specialized health care professionals to review any denial involving a determination of a service’s medical necessity. A change that is viewed as fixing “a big hole” in managing the use of AI predictive models.
More recently, on September 28, 2024, California passed SB 1120, requiring health care service plans utilizing AI to determine necessary medical treatments to meet and comply with specific requirements. The objective of this new legislation is to increase the transparency of these models, prevent discrimination, and limit supplantation of health care providers decision making.
The introduction of AI in the healthcare industry is novel, and further reactions from governments on a state and federal level are likely to follow.
Conclusion
Proponents of AI predictive models believe that these systems will speed up the claims process, detect unusual billing patterns, and allow health insurance companies to make more accurate risk assessments. In turn, this will allow these companies to utilize their resources more efficiently and offer better treatment plans. But at what cost to the insured? If AI proves to be as reliable as its proponents believe, then perhaps a future exists where predictive models are commonplace, and serve to benefit not only the insurance companies, but those covered as well. However, many of these models are in their infancy. Currently relying on the outputs of these models, especially when it involves the health and wellbeing of individuals, is a slippery slope that can, and has harmed people physically and financially.
Washington Journal of Law, Technology & Arts 