
Additional Resource
Experiential Civic Learning for American Democracy
A new report provides a clear, actionable framework for effective experiential civic learning—what it is, why it matters, and how to do it well.
Additional Resource
As a part of the Allen Lab’s Political Economy of AI Essay Collection, Ajeet Singh explores how AI technologies deployed in the health care sector often orient towards the extraction of greater surplus revenues at the expense of patient health.
As algorithmic tools grow increasingly ubiquitous in healthcare, patients and providers call for innovations to improve the access to and quality of care. In recent years, such models have demonstrated the ability to automate retinal disease surveillance in remote clinics, reduce in-hospital death through early sepsis detection, and discover the first new structural class of antibiotics in decades to wield against growing threats of drug-resistant bacteria.1 2 3 While this laudable progress inspires hope, healthcare’s magnified corporatization raises important questions. If underlying incentives mandate further financialization of care, how will this shape investments in algorithmic innovation?
Any interrogation of algorithmic developments in healthcare must first confront the political economy in which they will be built and deployed. The most successful actors in the payor landscape have spent the last two decades developing novel models to restrict reimbursements and deny care, consolidating the market and amplifying profits to reinforce their market growth. If algorithms can accelerate the extraction of greater surplus revenues at the cost of patient health, payors will inevitably orient investments in computation to further power and obfuscate these practices. Though some of these models are under federal investigation, regulators must resist the temptation to separate innovation from its context if we are to guide the development of algorithms towards socially beneficial ends.
The drugstore CVS purchased the pharmacy benefits manager (PBM), an intermediary drug price negotiator, Caremark in 2007. With their acquisition of payer Aetna in 2018, CVS cemented its position as the largest PBM in the country, while continuing to buy various physician practices and home-based primary care services.4 UnitedHealth Group, the largest insurer and employer of physicians, merged its pharmacy and analytics subsidiaries into Optum in 2011- now the third-largest PBM in the country- having since purchased dominant physician groups DaVita Medical Group in 2019 and Atrius Health in 2021.5 6 Humana became the largest provider of senior-based primary care and post-acute home-based services after acquiring Kindred at Home in 2021.7 The remaining national insurers, including Cigna, Centene, and Elevance, along with outsiders, like private equity firms and Amazon, continue to expand their positions across this landscape.8 9
While industry proponents argue that mergers enable efficiencies, the reality reveals a range of financial engineering and market manipulation tactics to raise costs and extract greater profits, enabling the accumulation of capital and consolidation of market power. These motives underpinned the development of algorithms to catalyze and conceal profit extraction – until, inevitably, stories of their harms made national news, and payers faced calls for accountability. In response, the U.S. Department of Justice created the Task Force on Health Care Monopolies and Collusion.10
Much of the industry’s consolidation stems from the shift toward risk-sharing in Medicare and Medicaid. The Medicare Modernization Act of 2003 created Medicare Advantage (MA), a privatized form of Medicare intended to curtail alleged fraud and waste in traditional Fee-For-Service (FSS) Medicare. Using capitation-based financing, payers received a lump sum budget for a population’s care, where savings at the end of each term would be pocketed as profits. Theoretically, this would incentivize investments in preventative care to avoid costly hospitalizations. From its inception, MA set higher payment rates than FFS Medicare, and in 2023, MA outgrew its predecessor, managing over half of the eligible Medicare population.11
These plans have proven exceedingly profitable for payers. In 2023, the nonpartisan government agency Medicare Payment Advisory Commission (MedPAC) reported that the government paid MA plans 23% more per beneficiary than FFS Medicare, with overpayments costing $82 billion last year and $612 billion since 2007.12 This difference in revenue cannot be attributed to more efficient care delivery. Instead, payers deployed a range of strategies, most notably upcoding and favorable selection, to maximize extraction and enable further coercive tactics at scale.
MA’s capitated payments are informed by illness severity, accommodating higher costs for sicker patients. The Centers for Medicare & Medicaid Services (CMS) introduced risk adjustments in 2004, expanding the range of diagnosis codes used to inform risk scores. Immediately, Congress identified a wave of upcoding, a practice to inflate the quantity and severity of diagnoses per enrollee to increase CMS payments. The 2005 Deficit Reduction Act instructed CMS to constrain this by modifying risk calculations in 2008. However, CMS did not act until 2010, when it estimated a 3.41% score inflation due to coding differences between MA and FFS plans for comparable beneficiaries. CMS then implemented an equivalent downward MA score reduction to bluntly “correct” the risk. Meanwhile, the Government Accountability Office estimated this excess risk scoring at 7.1% for the same year.13 Subsequent legislation mandated further adjustments, but since 2018, the score adjustment has remained stagnant at 5.9%.14 15 While generously characterized as clever arithmetic, these practices entailed fraudulent behavior, with insurers coercing health care workers to add diagnoses, even when they were inappropriate or unrelated to treatment.16 Unsurprisingly, MedPAC found that MA’s risk score inflation reached 13% in 2023.17
In addition to making patients appear sicker, MA insurers also practiced favorable selection to avoid enrolling costly beneficiaries with more severe illnesses, as they could complicate the payer’s attempts to engineer greater profits. MedPAC’s 2012 report noted that payers were likely gaming risk adjustment by recruiting patients who were inexpensive relative to their risk score.18 They presented an example using congestive heart failure, where 5% of the sickest patients had costs 322-fold higher than the healthiest 5%, demonstrating clear opportunities to stratify and avoid costlier patients.
MedPAC’s June 2023 report emphasized that since 2008, new MA enrollees consistently incurred below-average costs compared to traditional Medicare patients of comparable risk, and enrollees who demonstrated higher costs were more likely to switch back to traditional Medicare.19 While MA cannot deny applicants, they levy other mechanisms to push patients out.20 Tactics include restricting in-network care for resources associated with higher costs (like specialized cancer centers), recategorizing expensive drugs to payment tiers that require higher copays, and imposing onerous prior authorization requirements that obstruct access to care and encourage patients to “choose” to switch back to FFS Medicare. These prohibitive practices enabled excess revenues on top of existing upcoding practices, resulting in a combined 23% in overpayments last year.
Furthermore, MA spends 9% less on medical care than FFS Medicare for comparable patients.21 While MA plans boast supplementary benefits like eyeglass and dental coverage, these are overshadowed by steep restrictions in care options. As a result, MA patients do not enjoy better access to dental care or lower out-of-pocket costs than FFS Medicare patients.22 Furthermore, MA patients requiring complex cancer surgeries face longer delays, are less likely to be treated at specialized centers, and bear higher mortality than comparable FFS Medicare patients.23
MA programs have also taken advantage of dual enrollment with the Veterans Health Administration (VHA). In 2020, over 20% of all Veterans, accounting for over 1 million patients, were dual VHA/MA enrolled. If they sought care at the VHA, the VHA would cover their bill. From 2011-2020, the VHA paid over $78 billion in health services for dual enrollees, while their MA plans continued to receive the full capitated payments, regardless of whether or not veterans used their care, effectively doubling government expenses in covering the same beneficiaries.24 Recent findings on “High-Veteran” MA plans, which targeted VA patients for enrollment, found that about one in five of these VHA/MA beneficiaries did not use any MA services in a given year, letting private payers pocket the entire premium from taxpayer dollars.25
While FFS Medicare allocates 2% of expenditures to overhead, MA overhead costs reach 14%, just under the federal medical loss ratio (MLR) limit, meant to ensure that at least 85% of revenue is spent on patient care.26 This feeds the bureaucracy of financial engineering that enables excess revenue extraction. From 2007 to 2024, MA overhead is estimated to total $592 billion.27 However, this is likely an underestimate, as further profits are hidden through intercompany transfers. For example, UnitedHealth evades MLR caps by moving a quarter of its revenue to subsidiaries, concealing profits by paying itself.28
This consolidation of capital and market power enables a cascade of downstream coercive practices, such as cutting labor costs, shortening visit times, replacing physicians with less costly advanced practice providers, increasing surprise fees, prolonging wait times, worsening health outcomes, and self-preferencing to divert business away from unaffiliated pharmacies, clinics, hospitals, rehabilitation facilities, physician practices, home-based care providers, and more.29, 30 31 32 Notably, many of the independent companies losing revenue themselves become targeted acquisitions to avoid bankruptcy.
This June, a bipartisan and bicameral letter to the Centers for Medicare and Medicaid Services reiterated concerns about MA plans’ ongoing use of algorithms to “erroneously deny care and contradict provider assessment findings.”33 For example, in 2023, Cigna was found enforcing a “PxDx” algorithm to bulk deny thousands of claims without clinical review, correctly anticipating few appeals and shifting significant costs to patients.34 The algorithm’s developer had come to Cigna having “solved the problem once before” at UnitedHealth, which has yet to acknowledge the use of any such system. More recently, a Cigna company EviCore, which contracts with over 100 insurers to reach 100 million patients nationally, was found using an algorithm called “the dial” to expedite denials in coverage for care. They boasted a 3-to-1 return on investment, informing patients their care was “not medically necessary” to help insurers reimburse less care and yield higher profits.35
UnitedHealth separately faces a class action lawsuit for pressuring employees to use another algorithm, “nH Predict,” to systematically deny rehabilitative care in skilled nursing facilities to MA patients.36 The suit claims UnitedHealth knew the model had a 90% error rate, and that only 0.2% of patients would file appeals to overturn the decisions. Demands for explanations are met with denials in the name of intellectual property, and these cases are just a few that have faced public scrutiny. Payers continue to aggregate data to build models that guide strategies in beneficiary enrollment, pricing, coverage, and more. UnitedHealth’s acquisition of Change Healthcare in 2022, the largest billing exchange in the country, handed them a 94% share over the clearinghouse market.37 While CMS clarified that predictive algorithms cannot be used in these ways, this will likely drive innovation towards other novel, opaque methods for extracting greater revenues.
Technologies are political in both their design and deployment, and to anticipate and understand the political economy of AI, one must first confront the material reality of the industry as it functions today.![]()
Ajeet Singh
Physician Instructor and Clinical Informaticist, Rush University Medical Center
There is nothing inherent in the logics of computation that demands the innovation and deployment of such predatory models. However, under the imperative to shareholders, payers will overwhelmingly orient technologies toward obfuscating and maximizing the extraction of surplus value from health care, accumulating capital and increasing market share. Market dominance translates into political power to reinforce a favorable regulatory landscape while constraining the possibilities for reform. Technologies are political in both their design and deployment, and to anticipate and understand the political economy of AI, one must first confront the material reality of the industry as it functions today.
It is no surprise that every major MA payer has been accused of fraud by the U.S. government or whistleblowers.38 The dominant MA payer strategies, augmented by algorithms, overwhelmingly demonstrate their framing of patients and care providers as financial assets whose purpose is to facilitate the ever-increasing extraction of taxpayer dollars. After 20 years of reactive and incremental reforms attempting to protect patient health, we must recognize the failure of MA and abolish it with haste. If we hope to ideate alternative technologies that support human flourishing, we must critically examine and challenge underlying power relations that shape the environment for any computational infrastructure. Only then can we hope to reorient the politics of AI toward a future that is more equitable and beneficial for all.
Earlier this year, the Allen Lab for Democracy Renovation hosted a convening on the Political Economy of AI. This collection of essays from leading scholars and experts raise critical questions surrounding power, governance, and democracy as they consider how technology can better serve the public interest.
Additional Resource
A new report provides a clear, actionable framework for effective experiential civic learning—what it is, why it matters, and how to do it well.
Additional Resource
The bipartisan Utah Digital Choice Act aims to reform the social media ecosystem by giving users more choice and ownership over their personal data, while encouraging platform innovation and competition.
Policy Brief
The GETTING-Plurality Research Network submitted a public comment on the Development of a 2025 National Artificial Intelligence Research and Development Strategic Plan.
Policy Brief
The GETTING-Plurality Research Network submitted a public comment on the Development of a 2025 National Artificial Intelligence Research and Development Strategic Plan.
Feature
What kind of democracy do legislators want? This question was at the center of a recent discussion with Melody Crowder-Meyer, associate professor of political science at Davidson College, as part of the American Politics Speaker Series.
Policy Brief
The GETTING-Plurality Research Network submitted a comment to Representative Trahan’s Request for Information to modernize the Privacy Act of 1974.