Managed Care 3.0: Rise of the Robots


Healthcare payment in the US has evolved in decades-long sweeps over the past fifty years, as both public programs and employers attempted to contain the relentless rise in health costs.  Managed care in the United States has gone through three distinct phases in that time- from physician- and hospital-led delegated risk to “shadow” capitation via virtual networks like ACOs to machine-governed payment systems, where intelligent agents (AI) using machine learning are managing the flow of dollars.  Increasingly, care is being managed not by physicians or health systems but by computerized documentation and payment systems governed by artificial intelligence (AI).  

Phase I- Health Maintenance Organizations and Delegated Risk

After a lengthy stretch of double-digit health cost inflation following the passage of Medicare in 1965, the Nixon administration launched a bold initiative- the HMO Act of 1973- to attempt to tame health costs.   The goal of the HMO Act was to power up prepaid health plans modeled on the Pacific Coast-based Kaiser Foundation Health plans nationwide.  The Act provided federal start-up funds and subsidies for HMOs.  It also compelled employers to offer HMOs as an alternative to Blue Cross and indemnity insurance.  These plans were intended to save money by managing a fixed annual budget for care (the number of subscribers multiplied by an annual premium) vs. the legacy, open-ended fee-for-service system.  

While a few HMOs either employed physicians directly on salary (staff models like the Group Health Co-Operatives), or contracted on an exclusive basis with an affiliated physician group (like Kaiser’s Permanente Medical Groups), many more delegated capitated risk to special purpose physician networks- Independent Practice Associations (IPAs)- whose physicians continued in private medical practice.  Capitation created a compelling incentive for physicians to economize in care provision.  

By 1996, according to the Kaiser/HRET employee benefits survey, HMOs covered 31% of the employer market (roughly 160 million employees and dependents).  The impact of HMO growth on overall US health spending remains uncertain, because health spending continued growing aggressively during the next fifteen years,  only abating during the mid-1990’s around the Clinton Health Reform debate.

Two things brought the HMO movement to a crashing halt in the late 1990’s.  One was a political backlash from workers and their families who were simply assigned to HMOs by their employers, rather than choosing them themselves.  This unilateral assignment violated a fundamental principle of HMO advocates like Paul Ellwood, who championed consumer choice.   Employees and their families so assigned found their access to care narrowed by the mechanical application of medical necessity criteria to their care.  Women, who are the pivotal actors in managing their families’ health and were growing increasingly confident of their political influence, went ballistic.  

The other political force that helped quash the HMO movement was angry pushback from physician communities, particularly specialists, who bitterly resented the invasion of their professional freedom by prior authorization and medical necessity reviews. Physicians also resisted pressure to discount their rates to the HMOs, which reduced their incomes. A major concurrent financial blow to HMOs was a sharp downward adjustment in Medicare payment rate for health plans in the Balanced Budget Act of 1998.  

In the aftermath of this reaction, closed panel, delegated-risk capitation gave way to open panel “preferred provider organization” (PPO) models which paid physicians and hospitals a discounted fee-for-service.  PPOs were basically an industrialized version of traditional Blue Cross, but with allegedly narrower provider networks. PPOs offered patients broad access to physician and hospital networks and somewhat less interference in physician decision making.   

As PPOs rolled out,  the threat of being excluded from PPO networks led to a lengthy and damaging provider pricing panic.    As PPOs spread, their networks broadened to be virtually indistinguishable from Blue Cross networks in many communities.  Providers who discounted their rates in the panic to avoid being excluded from PPOs discovered to their horror that their pricing concessions yielded no growth in volume or market share, just reduced revenues.   

By 2014, HMO’s share of the total commercial market had shrunk to only 13%, well less than half of its peak.  Where HMOs grew, it was through Medicare Advantage and managed Medicaid.  There was an important exception to this trend.   While the HMO industry shrank nationally, Kaiser saw its enrollment grow to more than 12 million, dominant on the Pacific Coast but a negligible presence nationally.   Strong regional HMOs sponsored by health systems and physician organizations competed with Kaiser in the West and in a scattering of non-Western markets against traditional health insurers.

HMOs are not “over”, however.  From 2016 to 2019, HMO penetration in the commercial market quietly grew to 19% mainly at the expense of PPO coverage, according to the very same Kaiser/HRET survey.  And in 2018, about two thirds of the 22 million Medicare Advantage subscribers are in HMOs.    While HMOs by no means disappeared, the “movement” fell short of national reach. HMOs remain a negligible presence in the most rapidly growing parts of the US- the Southwest, South and Mid-Atlantic regions.

Phase II- Managed Care Lite plus Rising Patient Cost Sharing 

After the 2008 recession, employers and their health plans shifted strategy from putting physicians and hospitals at risk to putting patients at risk.  In the wake of the recession, the number of patients with high deductible health plans nearly sextupled- to over sixty million lives (see earlier cited Kaiser/HRET survey).  By 2019, 30% of the “lives” in employer-based plans were in high deductible plans regardless of patient economic circumstances. 

At the same time, Medicare moved aggressively to get providers into a new, less politically inflammatory version of managed care for large regular Medicare market (e.g. the non-Medicare Advantage portion).  The 2010 Affordable Care Act catalyzed the formation of new, virtual, special-purpose managed care enterprises called Accountable Care Organizations.   Two key changes in approach vs HMOs were conditioned by the managed care backlash:  Medicare patients were not forced into managed care plans (or even told they were in them), and providers would be insulated from downside financial risk for a lengthy period.  

Unlike with HMOs, patients would not be asked to choose an alternative “value -based” care system, nor would they receive any of the savings generated by the presumably better care management.   Under ACOs, patients would be statistically assigned retrospectively to ACO panels based on whether their primary care physician was participating.  ACO membership was a statistical construct, not a consensual patient panel. 

Secondly, Medicare retained the risk, paying for care in the first instance based of the Medicare fee schedule, then retrospectively calculating whether ACOs achieved savings relative to local spending benchmarks.  The vast majority of ACOs bore no downside risk, meaning that if they overspent their spending targets, Medicare absorbed the losses.  ACOs would earn bonuses, however, based on beating statistically determined savings targets.    And as part of the bargain, ACOs had to submit voluminous “quality” information and meet predetermined quality metrics.  

Large commercial health plans serving employers shadowed the Medicare program, taking advantage of yet another hospital pricing panic to create new virtual ACO networks based on deeply discounted rates.   Most of the plans offered under ObamaCare’s insurance exchanges were of this type.   While it was assumed by providers that commercial ACOs would move rapidly toward true delegation of risk, a decade on, the risk mysteriously has not passed over to providers, who have spent at least $10 billion preparing for ACOs.  Moody’s Investor Service  found that the median US hospital received only 1.8% of their revenues from capitation, and another 1.9% from “two-sided” ACO style risk in 2018, proportions that barely rose in five years from 2014 to 2018.  The much hyped movement “from volume to value” has been largely illusory.

Though the ACO “industry”, mainly consultants and investors,  continue hyping the idea, the federal ACO program has so far not saved the Medicare program a penny, if one counts the bonuses paid out to successful ACOs, the cost overruns by the ACOs that missed their cost targets, and the cost to Medicare of setting up, administering and monitoring the program. Paid out bonuses tended to be highly concentrated in those fortunate ACOs operating in high Medicare cost markets.  Medicare’s ACOs have been, to quote MedPac, a “disappointment” (see “Medpac Medicare Advantage and ACOs Can’t Cut Costs and We’ll Never Know Why” And to no one’s surprise,  physician-sponsored ACOs decisively outperformed those sponsored by hospitals.   If there have been any savings to anyone or bonuses paid from commercial ACOs to provider networks, they remain a closely guarded secret.  Today, perhaps 10% of the US population is in some form of ACO.  

Phase III- Machine Driven Managed Care.  

The reason for the reluctance of health plans to delegate risk to providers is that not only would they have had to share profits with providers (from reducing the cost trend) , but, crucially, they would have lost much of the new data they were gathering on physicians and patients. This would have meant the loss of a crucial leverage point in containing medical costs.   “Value based care” has seen a shift in economic power from hospitals to health plans (see attached graph from Nate Kaufman). 

Health Insurance Premium Growth vs Hospital Payment Rate Growth

Much of this trend change was due to the pricing panic referred to earlier- as hospitals and physicians scrambled to avoid being excluded from new “narrow network” plans targeted at the ObamaCare health exchanges and Medicaid managed care.   But in addition, health plans markedly increased the use of contractors employing artificial intelligence (AI) –  data mining, algorithms and machine learning – to comb through the newly rich medical claims data sets to deny or reprice claims submitted by hospitals and physicians for their commercial, Medicare Advantage and Medicaid managed care patients. 

A shadowy industry populated with billion dollar high tech firms no one in the care system had ever heard of – with names like Emdeon (now Change Healthcare), Equian (now part of Optum) , MultiPlan and Cotiviti – emerged to service health plans with automated systems to review hospital and physician claims prior to payment.   Many of these firms were founded after the passage of the Health Insurance Portability and Accountability Act of 1996 (HIPAA) whose “administrative simplification” provisions catalyzed a push toward electronic data interchange of medical claims (EDI).  These firms ramped up rapidly as ubiquitous broadband replaced expensive dedicated T-1 lines. They marketed their expanded role as part of the war against medical “fraud and abuse”.    

A key factor in the wave of denials was the increased centrality of hospital emergency admissions as the main gateway to complex and expensive inpatient care.  With primary care physicians withdrawing from hospital practice, decisions to admit patients to hospitals were increasingly made by employed physicians or physician contractors to the hospital.     Upwards of 70% of patients in many health systems are admitted through the emergency room and care is rendered to those patients on an urgent basis.    “Prior authorization”, a forty year old HMO expense control tool for managing “elective care”, has given way to “prospective pre-payment review” applied after hospitals have admitted and cared for patients and submitted insurance claims.    

Hospitals saw, in some cases, a doubling of claims denials or “repricing” in just a twelve to eighteen-month period after 2016 based on these automated “prospective” reviews.   This surge of machine-driven denials played a major role in the mysterious 39% plummet in hospital operating earnings seen in 2016 and 2017.  These denials often result in unexpected higher bills to patients with high deductible plans as well as significant new administrative expenses for hospitals to track and contest the surge of denials.  Many patients were unable to pay these unexpectedly higher expenses, resulting in a rise of “insured self-pay” bad debts. 

Hospitals and health systems are getting help pushing back against automated payer-driven AI claims review systems.  A private equity funded startup, Olive, has developed  its own AI systems to audit hospital claims prior to submission )   Olive and other companies will use robotic process automation (RPA) to manage documentation to make sure they minimize the risk of a costly denial, and to assure “payment integrity”.  Thus, the next decade is likely to see an escalating war of robotic provider and payer AI systems managing the filing and adjudication of medical claims. 

COVID has produced what seems like a temporary cease fire in this “War of the Robots”.  This is because the shutdown of routine care in hospitals has produced a multi-hundred billion windfall in cash flow for health plans.  Denying care during a health emergency could produce a lot of ugly headlines, so plans appear to have turned the “denial machine” off, according to colleagues in the revenue cycle industry.  This will magically raise their medical expenses (the so-called Medical Loss Ratio or MLR) and lower their political profile. Plans do not want to be socked with a “windfall profits” tax to help fund federal hospital relief efforts.  However, when health plans medical expenses MLRs have risen again it is a dead certainty that the denial machinery will crank up again and that the war will resume.   

A managed care” movement” which began more than seventy years ago by empowering clinicians to manage care for populations within a fixed budget has devolved, by degrees, into an increasingly data driven payment system run by computers, that shifts costs to physicians and patients without their input.  The early stages of this devolution spawned successful, high quality integrated health systems and health plans in some  parts of the country.     However, the last decade has left the bulk of care providers drowning in documentation busywork and box checking, and burdened by a growing revenue cycle bureaucracy.  Their incomes are increasing managed by machines, not colleagues.  How this evolution will improve actual care to patients remains to be demonstrated.  

Jeff Goldsmith is a veteran health industry observer based in Charlottesville, Virginia and President of Health Futures Inc

11 thoughts on “Managed Care 3.0: Rise of the Robots

  1. I think the people in the know are – if you go by the reactions to this post on Twitter and LinkedIn.

    Like the EHR stories that you and I love to hate, this stuff is so complex (read boring for most normal human beings) that many tech reporters don’t want to touch them with a ten foot pole.

    That’s a pity because like electronic health records, this is one of the great under-reported stories in health care. Here are some questions we should all be thinking about:

    — Are algorithms discriminating (either intentionally or in effect) ? See the public uproar over the LAPD’s use of algorithms for community policing.

    — What is the error rate of this kind of algorithm? There’s already ample evidence that algorithms screw up. But with little to or no oversight, we need to know more.

    — With algorithms in effect practicing medicine it seems likely they are committing malpractice by making bad decisions. If a doctor can be sued for malpractice, it stands to reason that an algorithm should be.

    — What impact are coverage decisions having on the COVID19 pandemic?

    I have no idea what the answer to this question is, but I can speculate. I have a feeling there some interesting questions and answers out there.

    1. What happens is that whether a particular policy solution works or not, there has been enough money in our system to turn the solution into an “industry”- with consultants who help execute them, middlemen who service the new entities in the field, and lobbyists in DC to keep the money flowing.
      Evidence doesn’t matter.

      The ACO is a classic example. There is now 15 years worth of evidence that they work in a few places and circumstances, but overall the solution doesn’t scale or generate system wide savings.. The godfathers of this idea-the Dartmouth policy shop- envisioned it as a way to solve small area variation and to end Medicare FFS. EVERY hospital and physician community would become a little HMO. There was plentiful evidence from the previous thirty years of trying to do this that most hospitals/physicians communities were too porous, divided and diffuse to be able to make this leap.
      The success rate in creating risk bearing provider enterprises was barely one in five. But the historical evidence didn’t matter. And it still doesn’t. ACOs are now an industry and they will probably be with us forever.

  2. We’re in a robots arms race between payers and providers with people as collateral damage. The robots of both sides are like armed drones with all of the ethical issues that come with, but hey, the military that does not deploy armed drones will surely lose. When the drones screw up and kill civilians, the armies are still safe in their respective $1T waste and disparities bunkers.

    That the people called patients accept this is a tragedy. They’re not organized to listen and are like the Eloi in HG Wells The Time Machine. That the people called doctors accept this is like the Morlocks in the novel: people that chose to seek safety by hiding in the military bunker and watching the drones do their thing. The doctors are not listening either.

  3. If you leave out BOTH the doctor and the patient from what is unquestionably clinical decision making, eventually you pay a price. It starts in the political system and works its way back.

    Those with long memories may recall that the HMO backlash was fed by women reacting to the “driveby obstetrical delivery” problem. This was the mechanical application of Milliman’s guidelines for hospital length of stay after a normal delivery that were pushing down towards one night in the hospital. Now, the difference is: no medical director, no peer panels, just the algorithm.

    1. Who paid the price, eventually, over three decades? Not the providers, or the payers, or the doctors. The pandemic threatens to disrupt the administrative and political kleptocracy we call US healthcare but it might take an even bigger disaster. I ask again, who is organized to listen?

  4. Thanks for this information on AI programs, Jeff. It would be helpful to have more information about what the robots are fighting about.

    If AI is being used simply to ensure that all the necessary boxes on claim forms are checked off, that strikes me as at worst harmless and at best a minor improvement in efficiency. But if AI is being used to make utilization review (prior authorization, concurrent review, and retrospective review) cheaper for the insurance industry to inflict on doctors and hospitals, then it is not harmless. In that event, we should view AI as just the latest phase in the wasteful, half-century campaign waged by the insurance industry to reduce health care costs by second-guessing doctors and patients.

    1. I think the answer to your question is that they are doing both- both coding accuracy and enforcing medical policy, This is a sample of just a portion of the product from one company:

      A classic example would be downcoding an acute care admission to an observation stay (which carries hefty patient responsibility under Medicare AND dramatically reduces payment). There are a ton of others. The new “value based” contracts contained exacting data specifications and a quantum leap in the sheer volume of data required (ostensibly to qualify the provider for quality related bonus payments), This new documentation burden obviously introduces the opportunity for empty check boxes, but the increase in complexity of rules also introduces the opportunity for payers to save money by denying or repricing the claim.

      Talk to a local revenue cycle professional inside a health system finance office and they can give you a ton of examples. It is like a thicket of brambles. It costs hundreds of dollars of manual labor and clinical time to respond to a single denial. The increased cost of care is spread over everyone’s bill.

      1. I’m still having a hard time imagining how AI would help an insurance company reduce the number of claims it pays. It seems to me the excuses for not paying a claim are finite and obvious — the patient was never enrolled at Aetna, Aetna’s formulary does not include statin y but does include statin x, Aetna will pay for statin x but only if it is preauthorized but in this case Aetna has no record that prior auth was dispensed, etc. How AI could reduce claims any faster or more effectively than the simple software that’s been available for decades is not obvious to me.

        However, I can readily imagine how AI might help an insurance company find issues in an ACO contract to nitpick about and thereby reduce the “savings” that have to be shared with the ACO. ACO contracts require several excruciatingly complex calculations. Complex calculations are required to determine whether the patient is assigned to an ACO; what the benchmark is against which savings or penalties will be determined; and how the insurance company’s or Medicare’s payout will be adjusted based on dozens of bogus “quality” measures that might in turn need to be adjusted (ever so crudely) for patient health and other factors outside doctor or hospital control.

        I’ve been studying health policy for decades. My inability to see clearly what the new cyborgs are up to reminds me of the plot of the “Terminator” movies. In those movies, an AI system known as Skynet, originally built by humans at NORAD, decides “all humans are a threat” and orders their extinction. The new managed care 3.0 cyborgs are operating at a similar level of murk and unaccountability. The managed care 3.0 cyborgs aren’t going to drop a bomb on us, but they are going to make it even harder for ordinary mortals to hold all insurers — insurance companies, self-insured companies, and both FFS Medicare and Medicare Advantage — accountable for their behavior.

    2. I am a board certified infectious disease doctor and a Hospitalist. My job is not as enjoyable anymore. Much in part to the insane amount of rules, documentation requirements and algorithms that push the physician into non- patient interactions for the benefit of either optimized billing or not accurate coding with no real benefit to the patient. The less time the patient gets with me the less I get to listen to their concern. The harm is in the lack of innovation in this space for the patient, certainly not the lack of innovation in optimization of medical billing and denial AI as stated above, but what about using that to help patients change behaviors and lead to better outcomes. In ten years I am not sure I will be a doctor, and I may be happier if only I could get rid of this debt.

Leave a Reply