The Argument Nobody Is Actually Having

Every few years the country stages the same debate. One side says health care is a right and the market has failed it. The other says the market is the only thing that has ever produced abundance, and a government program inside a capitalist republic is a slow-motion rationing scheme. Both sides argue from values. Both sides are arguing about something that has never been the binding constraint.

The binding constraint is that a national health system is, mechanically, a records system. Not a philosophy. Not a funding formula. A records system — one that has to know who you are, what happened to you, who is allowed to see it, and who owes whom money for it, across roughly 340 million people, tens of thousands of institutions, fifty regulatory regimes, and a lifetime per person.

We have run this experiment several times now, at several scales, under both parties, in this country and abroad. The results do not sort by ideology. They sort by architecture.

What Happened When the Republic Tried

Start with the most-litigated example, because everyone already has an opinion about it and almost nobody knows the numbers.

HealthCare.gov launched on October 1, 2013 and went down within two hours. By the time the Government Accountability Office finished counting, roughly $840 million had been spent. The six major contracts grew from $464 million to $824 million. Obligations on the federal marketplace alone went from $56 million to more than $209 million; the data hub from $30 million to nearly $85 million.

The GAO’s finding is the part worth memorizing, because it is not a partisan sentence: CMS “undertook the development of HealthCare.gov and its related systems without effective planning or oversight practices.” A readiness assessment scheduled for March 2013 was moved to September 2013 — weeks before launch — and the required approvals were never obtained. The site went live without verification that it met its performance requirements.

That is not a failure of socialism. That is a failure that would have destroyed a private company, and in fact the prime contractor was replaced.

A surgical team performing a procedure together in an equipped operating room
The clinical end of the system is where the records failure eventually lands.

Then there is the part of the ACA that best illuminates the hybrid problem — a socialist-flavored program operating inside a market. The law created 24 nonprofit insurance CO-OPs to inject competition where incumbents dominated. Twenty-three of the 24 survived their first full year. Only 12 survived 2015. Most of the rest were gone within two years.

They did not die of socialism, and they did not die of the market. They died because the federal government promised risk-corridor payments to make early-year losses survivable, and then did not pay. The Supreme Court eventually ruled 8–1, in April 2020, that the government had improperly withheld more than $12 billion — of which the mostly-insolvent CO-OPs were owed about $1.6 billion. The check arrived years after the patients were dead.

That is the actual failure mode of a public program inside a capitalist republic, and it has nothing to do with which economic system is superior. A market participant can survive a bad year. It cannot survive a counterparty that changes the terms mid-contract and takes seven years to be told to honor them by the Supreme Court. The incumbents absorbed it. The new entrants — the competition the program existed to create — did not.

The Failure Is Not American

If you think this is a uniquely American incompetence, England ran the cleanest version of the experiment.

The NHS National Programme for IT launched in 2002 to build an integrated national care record. It was billed as the largest civilian IT program in the world. It was formally dismantled in 2011 after costs neared £12.7 billion, against government-admitted benefits of roughly £2.6 billion.

The post-mortems agree on the cause with unusual unanimity, and it is not funding: it was top-down. Contracts were awarded fast and huge, to a handful of vendors, with scope still undefined at signature. Clinicians — the people who would have to type into the thing every day — were not meaningfully consulted. Uniformity was prioritized over utility.

And lest anyone think this is a problem of national scale specifically, the United States is currently running the same play inside a single agency. The VA’s Oracle Cerner health record began as a $10 billion contract in May 2018. Lifecycle cost estimates now run about $37 billion, with a 2022 estimate closer to $50 billion. The VA inspector general counted more than 800 major performance incidents. A GAO review found that 13% of staff using it believed it made the VA as efficient as possible, and 58% believed it increased patient safety risk — with reports of clinical notes disappearing and prescriptions displaying incorrect dosages. Deployments paused for three years, resumed in 2026, and completion is now projected “as soon as 2031.”

One agency. One vendor. Thirteen years, if it goes well from here.

What Actually Works, and Why It’s Not a Talking Point

Now the uncomfortable part for the other side of the aisle: national health records demonstrably do work. Denmark and Estonia are not theoretical.

Denmark’s sundhed.dk presents a unified clinical picture drawn from more than 120 different sources — without storing or duplicating the data centrally. Estonia’s X-Road platform underpins what independent rankings put at a 99.1% national interoperability rate.

Read that Danish sentence again, because it is the entire engineering thesis of this article: federated, not centralized. Denmark did not build one big database and migrate a nation into it. It built an addressing and exchange layer over the systems that already existed, and let the sources stay where they were.

Two honest caveats, because this is exactly where advocacy usually cheats:

  1. Scale. Denmark has under 6 million people; Estonia about 1.3 million. The United States has roughly 340 million and a far more fragmented payer landscape. Population scale changes the constant factors; it does not change the architecture, but anyone who waves it away is selling something.
  2. The identifier. Both countries have a universal civil registration number that makes linking records deterministic. The United States has legally forbidden itself from building one.

That second point deserves its own section, because it is the most self-inflicted wound in American health IT.

We Banned the Thing That Makes It Work

HIPAA — the 1996 law — actually called for a unique national patient identifier. Since fiscal year 1999, Section 510 of the annual Labor-HHS appropriations bill has prohibited HHS from spending a dollar to adopt one. That rider has been renewed every year for more than a quarter century. The House has repeatedly stripped it out on a bipartisan basis; it keeps getting reinserted in final legislation.

The consequence is that American health care matches patients probabilistically — by name, date of birth, address, and fuzzy logic — and gets it wrong constantly. The industry coalition pushing repeal puts numbers on it: repeated care from duplicate records averaging about $1,950 per inpatient stay and $1,700+ per emergency department visit; 35% of denied claims traceable to inaccurate patient identification, costing the average hospital $2.5 million and the system over $6.7 billion annually.

Here is the ethical shape of that, stated plainly: we spend billions annually, and accept a measurable rate of clinical error, in order to preserve the deniability of not having a national health identifier. That is a values choice, and it is a defensible one — the surveillance concern is real and I’ll come back to it — but it should be argued honestly as a purchase, with the price tag visible, rather than as a free-floating principle.

The Ethics of Referral: We Criminalized What Coordination Requires

The American system contains an elaborate legal apparatus that exists solely because a referral is a financial event.

The Stark Law prohibits a physician from referring Medicare and Medicaid patients to an entity in which they hold a financial interest. The Anti-Kickback Statute makes it a crime to exchange anything of value to induce a referral. Both are strict, heavily lawyered, and generate a compliance industry. CMS has since had to carve out explicit value-based safe harbors, because the same laws that stop self-dealing also blocked legitimate care coordination — you cannot easily pay a specialist to coordinate with a primary care physician when payment-for-referral is per se illegal.

This is the genuinely interesting ethical knot, and it cuts against both camps:

  • Against the market side: the necessity of Stark and the AKS is an admission that in health care, unmanaged financial incentive reliably produces referrals that serve the referrer. We already legislated that. The market does not self-correct here, and we stopped pretending it did in 1989.
  • Against the single-payer side: in a system where the payer is also the record-keeper, referral policing becomes trivial — because the payer sees everything. The apparatus you no longer need is replaced by a capability you should be far more afraid of. The record that coordinates care is the same record that enforces, audits, and eventually rations it. There is no version of the technology where you get one and not the other. Anyone promising the coordination without the surveillance either hasn’t built one or isn’t telling you.

That is the real trade, and it is architectural rather than ideological: how much observability does the coordinating layer get, and who is allowed to query it for purposes other than care?

Federation is the answer that takes the question seriously. Denmark’s 120 sources stay put; the layer above resolves and presents. A centralized national data store answers the same clinical question and simultaneously creates an artifact that a future administration of any persuasion will find irresistible.

Benefits Distribution: The System Didn’t Judge Them, It Lost Them

If you want the single most clarifying fact in American health policy, it is this one.

When pandemic-era continuous Medicaid enrollment ended, more than 25 million people were disenrolled. About 69% of them lost coverage for procedural reasons — a renewal form that never arrived, an address that was stale, a deadline nobody understood — not because anyone determined they were ineligible. Total enrollment fell by roughly 13 million even though 25 million were disenrolled, which tells you plainly that millions of people who were still eligible lost coverage and then re-enrolled.

Sit with the ethics of that for a second, because it is not a left or right fact.

Nobody in that story made a moral judgment about who deserves care. No death panel convened. No market cleared. A records system lost track of people, and the losing was itself the policy outcome. Whatever your view on whether Medicaid should exist at that size, essentially nobody’s political philosophy includes “coverage should depend on whether the mail forwarded.”

This is why I find the standard debate so unserious. We argue about whether benefits should be distributed, at enormous volume and heat, while the actual distribution is being decided by address hygiene and form design. The most consequential health policy intervention available right now is not a bill. It is a competent renewal workflow with accurate contact data and a system that knows the same human being across three agencies.

The Verdict: It’s the Clock, Not the Creed

Put the cases side by side and the pattern is not subtle.

What failed: monolithic, centrally-specified, big-bang programs with scope undefined at contract signature, clinicians consulted late, and a delivery horizon longer than the political term of the people who authorized them. That describes NPfIT, HealthCare.gov’s launch, and the VA rollout equally well — two of those under governments that wanted them badly.

What worked: federated architectures over existing sources, a deterministic identity spine, incremental delivery, and — this matters more than anything else on the list — institutional continuity long enough to finish.

So: can a capitalist democratic republic build a national health record?

Yes — but not the way it keeps trying. A republic’s distinguishing feature is that power changes hands on a fixed schedule, which is precisely the property that kills a 12-year monolith. Every big-bang national program is implicitly a bet that the next four administrations will keep funding a thing that shows no visible benefit until year eight. That bet has never once paid off. Not here, not in England.

What a republic can build is the boring part: identity resolution, exchange standards, and mandatory interfaces — infrastructure that delivers compounding value each year and survives a change in government because each increment already works. That is what TEFCA and the QHIN framework are, whether or not they were designed with this in mind, and it’s why they’ve quietly outlasted flashier efforts.

And the honest concession to the skeptics: the same federal government that just spent $37 billion on one agency’s record, and that has been unable to stop reinserting a 27-year-old rider blocking the identifier that would make any of it work, is not obviously the entity to run the coordinating layer. The case for federation isn’t just technical elegance. It’s that federation is the only architecture that doesn’t require trusting a single operator for a generation.

What This Costs, In Hours, To The People Doing The Work

One last number, because every architectural argument above eventually lands on a human being with a keyboard.

The time-and-motion research is consistent and grim: for every hour of direct patient face time, physicians spend nearly two additional hours on the EHR and desk work during the workday, plus another one to two hours of personal time. Primary care physicians spend nearly 6 of 12 hours interacting with the record. Clerical and administrative tasks are about 44% of that time. “Pajama time” — charting after the kids are in bed — runs over an hour on clinic days and weekends alike.

The HITECH Act spent more than $35 billion in subsidies to put those systems in place. It succeeded completely at adoption and largely failed at the thing adoption was for: the records still don’t talk to each other well, and the clinician pays the tax in evenings.

That is what happens when you procure a system instead of designing a workflow. And it is the strongest available argument that the next attempt — national, state, or county — should be judged on a single question: does the clinician get time back? Every architecture discussed above is downstream of that. If the answer is no, it does not matter whose ideology built it.

Part 2 of this series gets specific about the delivery side — how a small, disciplined organization would actually implement a system at this scale, and why the business, people, and process work is where these programs are really won and lost: How You’d Actually Deliver It.

For the engineering underneath the argument, DevForge Academy has the technical series: the requirements worksheet that survives procurement, why the interface list is the project, and federating a nation’s health record.

And if this pattern feels familiar, it should — it’s the same accountability gap documented in The Open Secret of Government IT.

Sources