The Health Care System Cannot Earn Trust Until It Learns in Public

What if the deepest problem in health care is not that we lack information, but that we do not know how to use information without distorting it?

A drug can cost one price to manufacture, another price to acquire, a third price to insure, and a fourth price to prescribe. Each layer may have its own contract, incentive, and vocabulary. By the time the patient sees a bill, the original economic reality has disappeared beneath rebates, subsidiaries, coding systems, network rules, and opaque negotiations.

Now consider a seemingly unrelated idea from experimental science: the bandit experiment. Instead of assigning every participant equally to every treatment, an adaptive experiment gradually sends more participants toward the treatments that appear to work best. The system learns while it acts. It does not wait for perfect knowledge before making decisions, and it does not pretend that all options deserve equal investment forever.

The connection is more important than it first appears. Both problems concern how institutions make decisions under uncertainty. Both ask who gets to see the relevant evidence, who controls the allocation of resources, and whether the system can correct itself when reality contradicts its assumptions.

The central lesson is this: transparency is valuable not merely because it exposes bad behavior, but because it makes intelligent adaptation possible. A trustworthy health care system would not simply publish prices. It would create a visible feedback loop in which patients, communities, clinicians, and hospitals could observe outcomes, compare alternatives, and redirect resources toward what creates the most value.

The hidden cost of a system that cannot learn

Imagine a hospital serving a town of 100,000 people. It has a fixed budget for facilities, staff, equipment, and administration. The community needs emergency care, cancer treatment, maternity services, orthopedic surgery, mental health care, and routine primary care. But the exact mix changes over time.

A traditional system often handles this uncertainty through layers of intermediaries. Insurers negotiate contracts with hospitals. Pharmacy benefit managers negotiate with manufacturers and pharmacies. Employers select plans. Patients encounter networks, deductibles, prior authorization rules, and changing formularies. The hospital then builds an administrative machine to translate care into the codes and contractual categories that produce reimbursement.

This machine can become extremely sophisticated while remaining almost entirely disconnected from the question patients care about: What are we paying for, and is it helping?

The problem is not complexity by itself. Medicine is complex. The problem is complexity without a learning mechanism. When prices, supply chains, and decisions are hidden, poor performance can persist because no participant sees enough of the system to challenge it. A hospital may continue offering a service because it is profitable under a particular contract, even if the community would prefer another service. A drug may remain expensive because the revenue is distributed across related entities whose relationships are difficult to see. An insurer may change its network annually, generating administrative work that adds no clinical value but does preserve bargaining power.

Opacity therefore does more than weaken trust. It prevents correction.

A visible system creates the possibility of comparison. If the acquisition cost of a generic drug, the operating margin, the dispensing fee, and the final price are all clear, someone can ask whether each layer is justified. If a hospital publishes the services it offers, their costs, their outcomes, and the community demand behind them, residents can debate whether the allocation makes sense.

This is where the bandit model offers a useful mental model. An organization operating in uncertainty has two responsibilities: exploration and exploitation. It must explore alternatives to discover what works, while exploiting the options that currently appear most valuable. A health system that never experiments wastes resources. A system that endlessly experiments without committing also fails its patients.

The real challenge is designing a process that does both.

A system earns trust when people can see not only what it decided, but how it learns from being wrong.

From price transparency to decision transparency

Publishing a list price is not the same as making a system transparent. A consumer may be shown a number while remaining unable to understand what it includes, what alternatives exist, or who benefits from the transaction.

Useful transparency has at least four layers.

First, structural transparency: Who owns whom? Does a pharmacy benefit manager have relationships with an insurer, a pharmacy chain, or a rebate aggregator? Are supposedly independent entities actually parts of one economic group?

Second, price transparency: What did the product or service cost to acquire? What fees were added? Which discounts were offered, and to whom? A transparent price should be explainable in ordinary language.

Third, performance transparency: What happened after the money was spent? Did the medication improve health? Did the surgery reduce complications? Did the new clinic reduce emergency room visits?

Fourth, decision transparency: Why was this service offered, this contract selected, or this treatment prioritized over another? What evidence would cause the decision to change?

The fourth layer is the one most institutions neglect. They disclose information but not the rules for acting on it. As a result, transparency becomes a public relations exercise rather than an operating system.

Consider generic drugs. If one company raises the price of a common medication dramatically, the public may recognize the outrage. But outrage alone does not create a functioning alternative. The alternative requires an intelligible chain from manufacturer to distributor to pharmacy to patient. It requires a price that can be explained, a margin that can be defended, and a supply process that can be monitored.

A straightforward cost structure can be powerful precisely because it converts trust from a feeling into a testable claim. If the seller says, in effect, “Here is what we paid, here is our fixed markup, and here is what you owe,” competitors can challenge the result. Customers can compare it. Regulators can investigate it. The market can learn.

The same logic applies to hospitals. Suppose a community hospital says its annual fixed cost is 100 million dollars, and that it needs an additional amount for variable services. Instead of hiding that financial reality inside hundreds of reimbursement arrangements, it could present the community with a service portfolio: emergency care costs this much, maternity care costs this much, oncology requires this staffing level, and mental health demand has changed by this amount.

The point is not that every resident will audit the books. Most will not. The point is that the system becomes legible to the people capable of auditing it: journalists, clinicians, researchers, civic organizations, employers, and competing providers. Legibility changes the behavior of executives because decisions can now be challenged with evidence.

The subscription model is really a governance model

The comparison between health care and a streaming subscription is easy to misunderstand. Health care is not a collection of movies. Patients cannot casually sample chemotherapy, and medical capacity cannot be delivered with the same marginal cost as digital content.

The useful analogy is not unlimited consumption. It is the relationship between a visible shared budget and a defined service library.

A community could decide that a hospital needs a certain amount of annual revenue to maintain its core capacity. That amount would support a clearly described set of services. Additional services could be proposed, priced, and evaluated according to demand and outcomes. Residents would not be purchasing every possible intervention individually through a maze of contracts. They would be participating in the support of a local capability.

This changes the identity of the customer. In the current arrangement, the payer often behaves like the customer because the payer controls reimbursement. The patient is the nominal beneficiary but frequently the least informed and least powerful participant. In a community centered model, the people who depend on the system become its principal constituency.

That shift also changes what counts as success. Under a reimbursement centered model, an organization may optimize the number and type of billable encounters. Under a community centered model, it has stronger reasons to ask whether a service is needed, whether it produces good outcomes, and whether prevention can reduce expensive downstream care.

But a subscription structure should not become a blank check. That is where adaptive experimentation matters.

Suppose a community is deciding whether to expand behavioral health, urgent care, or home based chronic disease management. It should not rely solely on political enthusiasm or the loudest advocacy group. It can allocate a modest initial budget to each option, measure demand and outcomes, and then direct more resources toward the approaches that demonstrate value.

This is the bandit principle in civic form. Start with several plausible options. Give each enough support to generate evidence. As the evidence improves, increase investment in the strongest options while continuing to monitor the neglected ones. The objective is not to discover a universal winner. It is to discover what works for this population, at this time, under these constraints.

A hospital might test two ways of reducing readmissions. One program uses remote monitoring, while another uses home visits by nurses. If remote monitoring performs better for younger patients but home visits work better for older patients living alone, an adaptive system does not force a single answer. It learns which intervention belongs where.

This is more realistic than the fantasy of a perfect health care plan designed in advance. Communities are not static. Diseases change, technologies improve, and preferences evolve. A fixed system eventually becomes a fossil. A transparent system can remain stable in its principles while adapting in its allocation.

Why simple rules can outperform grand reform

Large institutions often respond to dysfunction with more complexity. A new problem generates a new form, a new exception, a new intermediary, and a new reporting requirement. Each addition may be defensible in isolation. Together, they create a system too intricate for anyone to understand.

The alternative is not simplistic thinking. It is simplifying the rules at the points where complexity produces no value.

A universal contract between hospitals and insurers could eliminate multiple versions of the same agreement. Limiting contract changes to every two years instead of every year could reduce the administrative churn caused by constant renegotiation. Stabilizing provider networks for a defined period could prevent patients from discovering that their doctors have vanished from coverage midway through treatment.

These reforms may sound minor beside proposals to redesign the entire health care system. Yet small reductions in friction can matter enormously when applied to an industry that consumes roughly a quarter of the economy. A one percent improvement is not trivial when the base is immense.

Adaptive experiments reinforce this point. The purpose of an experiment is not always to invent a dramatic new institution. Often it is to determine which small change produces a measurable improvement. A hospital can test a simpler referral process. A pharmacy can test whether a clear cost breakdown increases adherence. An insurer can compare a stable network with a frequently changing one and measure the effect on missed appointments and continuity of care.

The critical condition is that the experiment must be visible and governed by predeclared criteria. Otherwise, institutions can cherry pick favorable results or quietly abandon inconvenient evidence.

A practical framework would ask five questions:

1. What decision are we trying to improve? Not “How do we increase engagement?” but “How do we reduce avoidable emergency visits among patients with uncontrolled diabetes?”
2. What alternatives are we comparing? At least two credible options should be tested rather than merely measuring one program against hope.
3. What will we measure? Cost, health outcomes, access, patient experience, and unintended consequences should be considered together.
4. When will we shift resources? The rules for increasing or reducing investment should be defined before results arrive.
5. What would change our minds? A system that cannot name disconfirming evidence is not learning. It is defending a preference.

These rules apply beyond health care. Any organization with a large budget, uncertain outcomes, and multiple stakeholders can use them. But health care makes the stakes especially clear because opacity is paid for not only with money, but with delayed treatment, anxiety, and lost trust.

Key Takeaways

• Treat transparency as infrastructure, not disclosure. Publish ownership, prices, outcomes, and the reasoning behind major decisions.
• Separate exploration from exploitation. Test promising alternatives with limited resources, then expand the approaches that produce reliable value.
• Make the service portfolio visible. Communities should be able to see what their institutions provide, what it costs, and why those services were chosen.
• Simplify before adding new rules. Universal contracts, longer periods of stability, and fewer intermediaries can produce meaningful savings without redesigning everything.
• Define what would change your mind. Every major program should have measurable success criteria and a clear process for reallocating resources when evidence changes.

The most important reform may therefore be cultural rather than technical. Institutions need to stop treating the public as an audience that receives explanations after decisions have been made. They should treat patients and communities as participants in an ongoing learning process.

A trustworthy system will still make mistakes. No adaptive experiment begins with certainty, and no hospital can predict every future need. Trust does not require perfection. It requires that people can inspect the incentives, understand the tradeoffs, see the evidence, and observe whether the institution responds when the evidence changes.

The future of health care may not be a single national blueprint or a miraculous technology. It may be a set of local systems that are financially legible, experimentally minded, and accountable to the people they serve.

That reframes the question. We should not ask only, “What does health care cost?” We should ask, “Can this health care system learn in public, and can the people who fund it redirect it when it fails?” If the answer is yes, transparency becomes more than a cure for suspicion. It becomes the mechanism through which a complex institution gets better.