The VA's Oracle-Cerner Record System: 11,000 Lost Order...

There is a federal project that set out to give every American veteran a single medical record that would follow them from their first day in uniform to their last visit to a VA clinic, and to do it by buying the same commercial software the Pentagon had already chosen. It is one of the largest information-technology purchases the government has ever made. It has also, at its first hospital, silently dropped more than eleven thousand doctors' orders into a queue that no one could see, been tied by the VA's own inspector general to scores of harm events, and been stopped in its tracks by the department that ordered it. The program is the VA's Electronic Health Record Modernization, and whether you call it a catastrophe or a course correction depends almost entirely on whether you are measuring the rollout or the goal.

What it is, and who actually runs it

For decades the Department of Veterans Affairs ran its hospitals on VistA, a records system it built in-house starting in the 1970s. VistA is widely admired and, by now, widely agreed to be aging and expensive to keep alive. The modernization plan is to retire it and move the entire VA onto a commercial platform called Cerner Millennium, the same core system the Department of Defense adopted for its own MHS Genesis rollout. The point of matching the Pentagon is continuity: a service member's record would mesh with the VA's the day they leave active duty, instead of being re-keyed or faxed across a gap that has failed veterans for years.

The VA signed the deal on May 17, 2018. It was a sole-source award to Cerner, meaning no competition, on the theory that only Cerner's platform could match what DoD had already bought, signed by then-Acting Secretary Robert Wilkie. One naming point matters before going further. Cerner won the contract, but Oracle completed its acquisition of Cerner in June 2022, as the Government Accountability Office notes across its EHR reporting. The vendor is now branded Oracle Health, and you will see it called Cerner, Oracle Cerner, and Oracle Health across the timeline. It is the same contract and the same system throughout: a rename, not a new supplier.

Where the money goes, and why the number keeps moving

The single most important thing to understand about the cost of this program is that there is no single number, and the figures in circulation measure different things. It is worth separating them cleanly, because blurring them is how the story gets distorted in both directions.

The contract ceiling: the May 2018 award carried a ceiling of nearly 10 billion dollars over ten years, with about 782 million dollars in initial funding already on hand for fiscal 2018. That is the cap on what can be paid to the vendor, not the cost of the whole undertaking.
The program estimate: in January 2019 the VA put the full program at about 16.1 billion dollars, the roughly 10 billion dollar contract plus about 6.1 billion dollars for the infrastructure upgrades and program-management work the software would sit on top of. GAO now calls that original estimate severely outdated and incomplete.
The independent lifecycle estimate: in October 2022 the Institute for Defense Analyses, an outside body, put the full lifecycle cost at about 49.8 billion dollars, which GAO reports as 32.7 billion dollars for thirteen years of implementation plus 17.1 billion dollars for fifteen years of sustainment. GAO calls that estimate more realistic than the VA's.

Two cautions on that jump. The 49.8 billion dollar figure covers roughly twenty-eight years of building and then running the system, so setting it against the ten-year, 16.1 billion dollar program estimate as if they measured the same thing overstates the growth. The apples-to-apples comparison is implementation against implementation, and there the cost still roughly doubled: GAO reports the independent estimate found the cost to implement rose from the 16.1 billion dollar baseline to somewhere between about 33.6 billion and 38.9 billion dollars. That precise band appears in GAO's full report and secondary summaries rather than the report's public abstract, so treat the exact endpoints as reported via GAO rather than a figure this piece verified line by line. The direction is not in doubt; the implementation cost at least doubled.

There is a newer number too. VA Deputy Secretary Paul Lawrence told a House subcommittee in September 2025 that the VA now estimates the lifecycle at roughly 37.2 billion dollars, a figure lawmakers openly questioned at a December 2025 hearing. Label it for what it is: a VA estimate aired at a hearing, not a GAO-validated or independent one. In fact GAO's standing complaint is precisely that the VA still lacks a fully validated, updated cost estimate at all.

Where it went wrong

The money is the smaller half of the story. The larger half is what happened when the software met patients.

The first go-live was at the Mann-Grandstaff VA Medical Center in Spokane, Washington, in October 2020. From that go-live through June 2021, the new system failed to deliver more than eleven thousand orders for requested clinical services, routing them instead into what the VA Office of Inspector General called an "unknown queue", a holding location that neither the clinician who sent the order nor anyone who might fill it could see, and without alerting the sender that the order had gone nowhere. In plain terms, a doctor could order a test, a referral, or a follow-up, watch the system accept it, and have it vanish into a place no one was told to look.

The harm that followed is where careful language matters. A clinical review associated with the OIG's work identified 149 adverse events: 2 categorized as major harm, 52 as moderate, and 95 as minor. These are events the inspector general's review identified and associated with the queue failure, not a proven causal death toll, and it is worth being exact about the most-cited case. An at-risk patient's psychiatric follow-up order was among those lost to the queue; that patient contacted the Veterans Crisis Line and was psychiatrically hospitalized. The person survived. The 2022 unknown-queue report documents serious harm, but it does not document a death.

A death does appear in the record, and it belongs to a different, later report. In March 2024, separate OIG work concerning the VA in central Ohio found that a scheduling and routing error may have contributed to a patient's disengagement from care, relapse, and fatal overdose in 2022. The watchdog's language is "may have contributed," and the case is distinct from the Spokane queue failure. The honest way to state the harm record is that two separate inspector-general findings, roughly two years apart, tie the new system to patient harm, that one of them involves a death, and that both use associative rather than flatly causal language.

Nor did the problems resolve into a system clinicians came to like. As of September 2024, after years of fixes, GAO reported that 75 percent of surveyed users, 1,247 of 1,670, disagreed or strongly disagreed that the new system made them as efficient as possible. Low usability was not a launch-week complaint that faded; it persisted at the live sites well after the rollout stopped.

The reset, not the cancellation

By April 2023 the VA had seen enough. It paused all future deployments of the system under what it called a larger program reset, choosing to fix what it had at the five medical centers already live rather than push the software onto new hospitals. The restart conditions the VA set were explicit: no new deployments until it was confident the system was highly functioning, demonstrated by improved clinician and veteran experience, sustained performance and reliability, and improved productivity at the sites already running it.

The word to hold onto is paused, not cancelled. The five live sites kept running. A sixth, the Lovell Federal Health Care Center in North Chicago, went live in 2024 as a deliberate exception, because it is a joint DoD and VA facility where the two departments share one building and one patient population. And the VA used its leverage. In May 2023 it renegotiated the contract, adding accountability measures, restructuring the term from a single five-year period into five one-year terms so it could walk away more easily, and building in stronger performance metrics with larger monetary credits and penalties. Then it began, cautiously, to move again: on December 20, 2024 the VA announced it would resume deployments at four Michigan facilities, and by late 2025 it was readying a broader restart in 2026, still over the objections of skeptical lawmakers.

Through all of it, GAO has kept a running tally of what the VA has not done. In one of its March 2025 reports, GAO recorded three new recommendations, two of them priority, on top of fifteen prior recommendations of which only one had been implemented, leaving fourteen open. The specifics of the count vary slightly between GAO's two related 2025 products, which is why it is worth citing a report rather than merging them, but the shape is consistent: the VA keeps agreeing with GAO and keeps not closing the recommendations, most of which circle back to the same missing pieces, a validated cost estimate and an integrated schedule.

Why it looks like a loss

Judged as a procurement, this program fails on almost every axis that procurements are supposed to be judged on. It was a sole-source award, so there was no competitive price discovery at the start. The cost to implement at least doubled from the VA's own baseline. The first hospital to run the software produced a defect that hid clinical orders from the people who wrote them, a failure mode that is close to the worst thing a medical records system can do, and the inspector general tied it to scores of harm events. Years into the effort, three out of four users at the live sites still said it slowed them down. The department had to stop the whole rollout for what became a roughly twenty-month pause, rewrite the contract to give itself escape hatches, and it still, by GAO's account, cannot produce a fully validated estimate of what the finished system will cost. Every one of those is a real, sourced failure, and none of them is spin from a critic. They come from the VA's inspector general and from the government's own auditors.

Ratio-adjusted for the mission

Now change the question from "how badly was it run" to "was the thing worth trying," and the ledger reads differently.

VistA is genuinely old. It is admired precisely because it was ahead of its time, but it was built in the 1970s and it grows more expensive and more fragile to sustain every year. Standing still is not a free option; it is a slow, quiet cost that never shows up as a headline. And the problem the modernization is meant to solve is real and specific. A veteran's care does not start at the VA. It starts in the military health system, and for decades the hand-off between the two has been a place where records were lost, re-entered by hand, and delayed. A single record on the same platform DoD already uses is not a vanity project; it is aimed squarely at the seam where veterans have actually been harmed by fragmentation for years. That is a public good measured in millions of people.

The status matters here too, and it cuts against the easy verdict. This program was paused and reset, not cancelled. The distinction is the whole argument. A cancelled program is a total loss, money spent for nothing. A paused-and-reset program is the government doing something it is often accused of never doing: stopping a troubled project before it spread, keeping the sites that work, renegotiating from a position of leverage, extracting penalties and off-ramps it did not have before, and restarting only under stated conditions. You can believe the VA restarted too soon, and many lawmakers do, while still recognizing that "fix it at six sites before touching a seventh" is a more disciplined response than either blindly pressing on or walking away and writing off the billions already spent.

There is also no clean alternative on the shelf. "Just scrap it and go back to VistA" ignores that VistA is the thing everyone agrees needs replacing, and that DoD is already on the commercial platform, so a VA retreat would re-open the very DoD-to-VA seam the program exists to close. The honest critique is not that the goal is wrong. It is that the execution has been bad enough to put the goal at risk.

The ledger reading

The VA's health-record modernization is one of those programs that is accurately described as a debacle and accurately described as a mission worth finishing, because those are answers to two different questions. Measured as a rollout, it is a hard case to defend: a hidden-order defect tied to patient harm, a cost that doubled, users who still say it slows them down, and auditors who still cannot get a validated number out of the department. Measured as a purpose, replacing a fifty-year-old system and closing the records gap that has failed transitioning service members for decades, it is close to unanswerable, because the thing it is trying to buy has no good substitute.

What makes it survive is that the second case keeps rescuing the first. Congress has floated bills to terminate the program, and the failures are damning enough to make that tempting, but the mission is real enough, and the sunk investment large enough, that the pattern each time is a reset rather than an ending. The lesson underneath is the one that recurs wherever public money meets a genuinely hard problem. A program can fail the test of how well it was executed and still pass the test of whether it should have been attempted, and the number you lead with, the eleven thousand lost orders or the one lifetime record, usually tells you which verdict you had already reached before you opened the file.

Fact-check notes and sources

The May 17, 2018 sole-source award to Cerner to replace VistA and match DoD's platform, with a ceiling of nearly 10 billion dollars over ten years and about 782 million dollars in initial fiscal-2018 funding: Nextgov/FCW on the signing and the VA's own contract announcement. The 10 billion dollars is a contract ceiling, not the total program cost.
Oracle's completion of its Cerner acquisition in June 2022 and the rebrand to Oracle Health, same contract and system: Government Accountability Office.
The VA's January 2019 program estimate of about 16.1 billion dollars (roughly 10 billion for the contract plus about 6.1 billion for infrastructure and program management), now called severely outdated by GAO: Government Accountability Office, GAO-25-108091.
The Institute for Defense Analyses' October 2022 lifecycle estimate of about 49.8 billion dollars (32.7 billion for thirteen years of implementation plus 17.1 billion for fifteen years of sustainment), which GAO calls more realistic: Government Accountability Office, GAO-25-106874. Note this lifecycle figure spans roughly twenty-eight years and is not directly comparable to the ten-year program estimate.
The implementation cost roughly doubling, from the 16.1 billion dollar baseline to between about 33.6 billion and 38.9 billion dollars: Government Accountability Office, GAO-25-108091. The precise 33.6-to-38.9-billion band appears in the full report and secondary summaries rather than the public abstract; treat the exact endpoints as reported via GAO.
The roughly 37.2 billion dollar newer lifecycle figure: presented by VA Deputy Secretary Paul Lawrence and questioned at a December 2025 House hearing, Nextgov/FCW. This is a VA estimate aired at a hearing, not a GAO-validated or independent one; GAO separately says the VA still lacks a fully validated updated estimate.
The "unknown queue" at Mann-Grandstaff in Spokane: more than 11,000 orders undelivered from the October 2020 go-live through June 2021, sent to a location neither senders nor receivers could see and without alerts: VA Office of Inspector General, July 2022 report.
The 149 adverse events identified by clinical review (2 major, 52 moderate, 95 minor harm), and the at-risk patient whose psychiatric follow-up order was lost and who was psychiatrically hospitalized (survived, not killed): Healthcare IT News on the OIG report and the VA OIG report itself. These are events the OIG review identified and associated with the failure, not a proven causal tally.
The separate March 2024 OIG finding, concerning the VA in central Ohio, that a scheduling and routing error may have contributed to a patient's relapse and fatal overdose in 2022: Nextgov/FCW. This is a distinct report from the 2022 unknown-queue findings; the language is "may have contributed."
The 75 percent (1,247 of 1,670) user-dissatisfaction figure as of September 2024: Government Accountability Office, GAO-25-106874.
The April 2023 pause and program reset, the five live sites (a sixth, Lovell Federal Health Care Center, added in 2024 as a joint DoD-VA exception), the December 20, 2024 Michigan resumption, and GAO's recommendation tally (three new plus fifteen prior with only one implemented): Government Accountability Office, GAO-25-106874. The VA's stated restart conditions: Federal News Network.
The May 2023 contract renegotiation adding accountability measures, restructuring into five one-year terms, and adding stronger metrics and larger penalties: Fierce Healthcare. The late-2025 preparations to restart deployments in 2026 amid lawmaker unease: Nextgov/FCW.

This post is informational and journalistic, describing a public program and public records. It is not legal, financial, medical, or policy advice. Figures are drawn from government reports, inspector-general findings, and public reporting, with cost tiers, hedged estimates, and associative harm findings labeled as such; where a figure is a VA estimate aired at a hearing rather than an independently validated one, it is named that way.

The VA's Oracle-Cerner Record System: 11,000 Lost Orders and a Program That Was Paused, Not Cancelled