The Legal Implications of Using AI and Algorithms to Combat Health Insurance Administrative Costs

Posted on Feb 17, 2025

Matthew Goodman

 

Introduction

            It is a statistic so commonly repeated that it has almost become a trope: the United States spends more than any other country on healthcare while receiving subpar results. For the most part, this is true. The United States spends nearly $5 trillion on healthcare, accounting for 17.6% of its total GDP.[1] On a per capita basis, this is nearly 20% more than the second highest spending country in the world, and over 40% more than most other similarly developed nations.[2] Meanwhile, the US lags behind other countries in health outcomes, including life expectancy and access to care.[3]

The health insurance market portrays a very similar story. Almost half of Americans are concerned with their ability to afford their monthly health insurance premium.[4] These costs have consistently increased each year, and are likely to continue rising.[5] These growing cost of health insurance is due to a number of factors, including insurer consolidation[6], the rising cost of medical services, and an aging population.[7] However, perhaps the most hidden – and easily addressable – contributor to these excessive costs are administrative costs.[8]

            Administrative costs as a whole account for up to 30% of total healthcare costs, over a trillion dollars annually.[9] Among the largest contributors, and one of most easily reducible administrative costs, is utilization management (UM). UM generally refers to the process by which insurers receive and adjudicate claims and requests for care. The costs associated with these processes comprise an estimated 20% of administrative costs.[10]

            Artificial Intelligence (AI) and other forms of algorithms present promising solutions to the problem of bloated administrative costs. The premise is simple: health insurers can use programs to determine, based on a patient’s medical records, whether a given claim is properly reimbursable or whether a prior authorization request should be approved.[11] This has substantial implications, with one study estimating potential savings between $150 million and $300 million per $10 billion of revenue.[12] This can ultimately save hundreds of millions, if not billions, for large health insurers, many of which have revenue in the tens or hundreds of billions.[13] By reducing these costs, health insurers can optimize profits while also passing on savings to customers via reduced premiums[14], thereby increasing access to coverage.

The use of these tools also poses several problems, particularly when they are being used to deny claims. The first is the technological concern – i.e., are these tools able assess claims at least as accurately as humans.[15] The second is the legal landscape and what legal exposure insurers might incur by implementing AI and algorithms to streamline UM processes.

Legal Considerations

Much of the uncertainty surrounding the legality of using AI or algorithms for utilization management is whether it can take the place of utilization review specialists and physicians who are typically required by law to conduct the review. Even if AI is used to make the initial determination, there is still a question as to whether an actual person must conduct their own detailed review of the AI decision.

Federally, the pertinent laws and regulations are those that govern Medicare and Medicaid, the largest federal health insurance payors, in addition to Medicare Advantage.[16] In October 2023, President Biden issued the Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence (EO 14110). This required, among other things, the Department of Health and Human Services to develop a plan to assess and deploy the use of AI in healthcare.[17] In tandem, the Centers for Medicare & Medicaid Services (CMS) offered guidance on the current usage of AI. They clarified that while AI and algorithms can be used, determinations must consider a patient’s specific circumstances and medical records and cannot be made solely based on an algorithmic analysis of a larger data set.[18] UM decisions must also comply with existing regulations. For example, insurers must involve a medical director who is responsible to ensure accuracy of all determinations of medical necessity.[19] Furthermore, all denials of coverage based on medical necessity must be reviewed by a “physician or other healthcare professional” before being issued to the patient.[20]

There is, however, gray area as to how deeply a medical professional must review the patient’s medical history and circumstances before the denial. For example, in 2023 Cigna - a large insurer that administers both private and Medicare Advantage insurance - was accused of using an algorithm to make coverage determinations, with medical reviewers allegedly reviewing 300,000 denials over the course of two months, spending an average of 1.2 seconds reviewing each case.[21] Cigna claims it merely uses the algorithm to match diagnosis codes and, based on what codes are submitted, determine whether the medical service was authorized.[22] Class action litigation over this allegation is ongoing, so to date it remains unclear as to whether these denials were compliant with Medicare Advantage regulation.[23] Furthermore, the Trump Administration repealed EO 14110.[24] It is thus unclear how the attitude on the use of AI in healthcare will shift.

On a state level, laws and regulations that affect the use of AI by private insures vary widely. California, for example, recently passed a law that apparently condones the use of AI to make some coverage determinations, but specifically prohibits the use of AI or algorithms to “deny, delay, or modify health care services based, in whole or in part, on medical necessity”, requiring a licensed physician or healthcare provider to make the determination.[25] Georgia has proposed similar legislation,[26] as has Illinois.[27] Other states are taking more moderate approaches. For example, New York’s proposed legislation does not expressly forbid the use of AI but requires clinical review of AI determinations measures to increase transparency such as submission of the algorithm for certification.[28] For these states, insurers can certainly leverage AI to reduce administrative costs and customer experience via faster claims approvals. However, it remains unclear what administrative burden the requirement for physicians to either make or review adverse determinations will impose. The landscape is even more uncertain in states that have no express legislation. There, insurers are likely proceeding at their own risk, hoping that their conduct does not run afoul of any general business or unfair practices laws.[29] The outcome of current litigation, much of which is ongoing, will provide further clarity on how courts may evaluate this issue.[30]

Looking Forward

The currently uncertain legal landscape provides several options for health insurers looking to use AI or algorithms. Insurers may choose to take a “wait and see” approach, monitoring current litigation and legislation before making investments in new technology. Other insurers may begin to use these tools in a conservative manner. To minimize risk, these insurers may choose to borrow from Blackstone’s view of criminal law, crafting algorithms with the preference that ten improper claims be reimbursed before one proper claim be denied. Still other insurers might choose an aggressive approach, employing these tools to reduce costs, offer lower prices, and capture market share. Ultimately, it will be fascinating to see exactly how this plays out moving forward. Much will rest on the outcome of current litigation as well as this administration’s appetite to allow the use of AI in this space. If done correctly and within the limits of current law and regulations, these tools offer an enticing opportunity to both improve the health insurance industry and better serve healthcare customers.

 

--------------------------------------------------------

[1] Historical National Health Expenditure Data, Centers for Medicare & Medicaid Services (Dec. 18, 2024), https://www.cms.gov/data-research/statistics-trends-and-reports/national-health-expenditure-data/historical.

[2] Health Expenditure per Capita, The World Bank (Apr. 15, 2024), https://data.worldbank.org/indicator/SH.XPD.CHEX.PC.CD?end=2022&most_recent_value_desc=true&start=2022.

[3] David Blumenthal et al., Mirror, Mirror 2024: How the U.S. Health System Compares Internationally, The Commonwealth Fund (Sept. 19, 2024), https://www.commonwealthfund.org/publications/fund-reports/2024/sep/mirror-mirror-2024#:~:text=The%20U.S.%20ranks%20last%20on,for%20people%20under%20age%2075.

[4] Lunna Lopes et al., Americans’ Challenges with Health Care Costs, Henry J. Kaiser Family Foundation (Mar. 1, 2024), https://www.kff.org/health-costs/issue-brief/americans-challenges-with-health-care-costs/.

[5] Michael Erman, U.S. Employers Expect Nearly 6% Spike in Health Insurance Costs in 2025: Mercer, Reuters (Sept. 12, 2024), https://www.reuters.com/markets/us/us-employers-expect-nearly-6-spike-health-insurance-costs-2025-mercer-says-2024-09-12/.

[6] Health Insurance Costs Are Increasing as Markets Become More Concentrated, with Fewer Insurance Companies: Interactive Map, U.S. Gov't Accountability Office (Dec. 5, 2024), https://www.gao.gov/blog/health-insurance-costs-are-increasing-markets-become-more-concentrated-fewer-insurance-companies-interactive-map.

[7] Rising Health Care Costs, Marsh & McLennan Agencies (Jun. 18, 2024), https://www.marshmma.com/us/insights/details/rising-health-care-costs.html.

[8] Administrative costs refer to the costs of managing health insurance plans other than paying medical claims, including claims adjudication, marketing, customer service, and employment costs.

[9] The Role of Administrative Waste in Excess U.S. Health Spending, Health Affairs (Oct. 6, 2022, 2024), https://www.healthaffairs.org/content/briefs/role-administrative-waste-excess-us-health-spending.

[10] David M. Cutler, Reducing Health Care Costs and Improving Health Outcomes, The Hamilton Project (Mar. 2020), https://www.hamiltonproject.org/assets/files/Cutler_PP_LO.pdf.

[11] Prior authorization request refers to the process by which insurers need to approve a patient’s request for care before covering the care. For example, an insurer might need to approve a prior authorization request before covering an extended inpatient hospital stay.

[12] An AI Opportunity for Health Insurers, McKinsey & Company (Jul. 5, 2024), https://www.mckinsey.com/featured-insights/sustainable-inclusive-growth/charts/an-ai-opportunity-for-health-insurers.

[13] Preeti Vankar, Largest Health Insurance Companies in the U.S. by Revenue, Statista (Feb. 27, 2024), https://www.statista.com/statistics/1451735/largest-health-insurance-companies-in-us-by-revenue/.

[14] Eugene Chang & John Kasey, Focusing on Health Plan Administrative Cost, Milliman (Dec. 29, 2022) https://www.milliman.com/en/insight/focusing-on-health-plan-administrative-cost (noting administrative costs are typically supported by premiums).

[15] An assessment of the state of current technology is outside the scope of this analysis. For the purposes of this analysis, we will assume technological capability.

[16] Generally, Medicare Advantage plans are offered by private insurers that administer Medicare plans.

[17] See Exec. Order No. 14110, 8 FR 75191 (2023).

[18] See Frequently Asked Questions related to Coverage Criteria and Utilization Management Requirements in CMS Final Rule (CMS-4201-hoF), Question 2, Centers for Medicare & Medicaid Services (Feb. 6, 2024), https://www.aha.org/system/files/media/file/2024/02/faqs-related-to-coverage-criteria-and-utilization-management-requirements-in-cms-final-rule-cms-4201-f.pdf.

[19] See 42 C.F.R. § 422.101(c)(1)(i)(D); 42 C.F.R. § 422.562(a)(4). “Medical necessity” refers to determinations that a procedure is warranted based on the patient’s medical conditions. A procedure generally must be deemed medically necessary to be covered.

[20] 42 C.F.R. § 422.566(d).

[21] Patrick Rucker et al., How Cigna Saves Millions by Having Its Doctors Reject Claims Without Reading Them, ProPublica (Mar. 25, 2023), https://www.propublica.org/article/cigna-pxdx-medical-health-insurance-rejection-claims.

[22] See The Facts about Cigna Healthcare's Claims Review Process, The Cigna Group, https://newsroom.thecignagroup.com/pxdx.

[23] See Kisting-Leung, et al, v. Cigna Corp., et al, 2:23CV01477 (E.D. Cal. 2023).

[24] See Exec. Order No. 14148, 90 FR 8237 (Jan. 2025).

[25] See CA LEGIS 879 (2024), 2024 Cal. Legis. Serv. Ch. 879 (S.B. 1120(k)(2)) (WEST).

[26] See H.B. 887, Reg. Sess. (Ga. 2024).

[27] See H.B. 2472, 103rd Gen. Assemb., Reg. Sess. (Ill. 2024).

[28] See N.Y. Legis. Assemb. A-1456, Reg. Sess. 2025-26 (2025).

[29] Some have argued that failure to disclose the use of AI may constitute a deceptive and unfair business practice. See Compl. ¶ 63-82, Kisting-Leung, 2:23CV01477

[30] See id. See also Compl. ¶ 251-75, Estate of Gene B. Lokken, The et al v. UnitedHealth Group, Inc. et al, 0:23CV03514 (D. Minn. 2023).