When the Metric Becomes the Medicine
A woman sits on the examination table with her feet on the step and waits for the door. She has twenty minutes.
She does not know that the twenty minutes is not a medical judgment. She also does not know who her physician works for, and on that second question the profession cannot agree with itself.
The American Medical Association reports that 42.2 percent of American physicians worked in private practice in 2024. In 2012 the figure was 60.1 percent — a drop of eighteen percentage points, and roughly eighty thousand fewer physicians in private practice than at the start of the period.
The Physicians Advocacy Institute reports that as of January 2026, 82 percent of practicing physicians were employed by hospitals or corporate entities. Only 18 percent remained in physician-owned settings.
Both numbers are correct.
They are not measuring the same thing.
The AMA counts physicians working in practices wholly owned by physicians. A doctor in a forty-partner group with a sliver of equity counts as private practice. The Physicians Advocacy Institute, working from Avalere Health’s analysis of practice-ownership data, counts practices that have been acquired — by hospitals, insurers, private equity firms, pharmacy chains — and counts the physicians inside them as employed.
The distance between 42 percent and 18 percent is not a statistical dispute. It is the measurable footprint of what this essay is about:
ownership remaining on paper long after control has moved somewhere else.
The AMA’s own data shows it from the other side. Between 2012 and 2024, the share of physicians describing themselves as practice owners fell from 53.2 percent to 35.4 percent, while the share describing themselves as employees rose from 41.8 percent to 57.5 percent.
American medicine changed hands in a single decade. It did so quietly, without legislation, and largely without public notice.
What follows is not an argument that corporate ownership is inherently malignant, and I want to make that point before anything else. Large organizations can supply capital, information systems, specialist depth, call coverage, and quality infrastructure that a four-physician office cannot finance. Private practice was a business with conflicts of its own, and none of this is nostalgia for it.
The argument is narrower and harder to dismiss. It is the argument the four essays before this one have been making about the Pentagon, the federal agencies, the national accounts, and the schools, and it arrives in the examining room in the same form it arrived everywhere else.
The instrument was built to describe. It was promoted to decide.
There is a literature on this and I have found it less useful than the cases.
- Charles Goodhart observed in 1975 that a statistical regularity relied on for policy will collapse once pressure is placed on it.
- Donald Campbell, working the same decade from a different direction, said a quantitative indicator used for social decision-making will corrupt the process it was meant to monitor — and, as the fourth essay in this series showed, listed the Vietnam body count and school achievement testing side by side as instances of one thing.
- Jerry Muller has since traced the reproduction of the pattern across policing, education, the military, and medicine.
All of that is correct and none of it prepares a reader for the feature the maxims leave out:
at every step, in every institution, the people installing the measure were serious people solving a real disorder, and what they installed was better than what it replaced.
That is what makes this hard, and it is why the essays are histories rather than warnings.
● ● ●
The first essay in this series went to the Pentagon at length. Here is only the part a physician needs.
Attrition became the American strategy in Vietnam because nobody could say what winning would look like, and attrition requires one input: enemy dead.
The count did not begin as a target. It began as a description — what a unit had encountered, reported upward so somebody could see the shape of the fighting. Then attrition became the strategy, the description became the definition of success, and an officer’s record became a function of it.
What followed was not blindness at headquarters, and the popular version of this story gets that wrong. MACV tracked more than a hundred and eighty indicators. The command was not starved of measurement; it was drowning in it, and everyone at the top knew the count was soft. The damage was done a level down, where the officer producing the number was also the officer graded on it, and where an estimate made in the dark with a helicopter waiting went in the direction a career pointed. When a general officer surveyed the 173 Army generals who had commanded there, sixty-one percent said the count had often been inflated.
The men producing those figures were not liars and were not fools. They were professionals inside a system that had told them, unmistakably, what counted. Intelligence offers no protection against the wrong metric. Integrity offers less than people expect. The metric is upstream of both.
Now consider a physician at a dashboard.
● ● ●
In the mid-1980s a health economist at Harvard took on a problem nobody else wanted.
Medicare paid physicians out of history. What a service was worth was whatever had been charged for it, adjusted against whatever had been charged before that, which meant the program paid several times more to cut than to think and had no defensible reason for any of it. An internist could spend forty minutes untangling a complicated old man and be paid a fraction of what the same hour returned in a procedure room. Everyone knew. Nobody could price it differently, because nobody had any other basis for a price.
William Hsiao and his colleagues at the Harvard School of Public Health went and built one, under contract to what was then the Health Care Financing Administration. They surveyed thousands of physicians and asked them to set the work in one service against the work in another — the time it took, the mental effort, the judgment required, the stress of a thing that can go wrong under your hands while you are doing it. They took that apart into components and put it back together as a scale.
The resource-based relative value scale was adopted for Medicare in 1992. Each service receives relative value units in three parts — physician work, practice expense, and professional liability insurance — adjusted geographically and multiplied by a conversion factor to produce a payment. Cognitive work went up. A generation-old injustice inside the fee schedule was corrected by measuring the thing that had been asserted and never counted.
The instrument worked, and it worked on an injustice that decades of professional complaint had never touched. It is Kuznets in the third essay, doing the same thing in the same spirit — a man building an instrument because the alternative was a country, or a payment system, running on assertion.
Be precise about what the instrument was for. It was built to set relative prices — what Medicare should pay for one coded service against another. It was never designed to evaluate a physician, and its architect has said so since, in public and at length.
There is an irony inside it worth naming. The relative values are recommended by the Relative Value Scale Update Committee — the RUC — convened by the American Medical Association and its specialty societies since 1991. The profession built the machinery now used to manage the profession.
The promotion happened anyway, and it happened the way it happened at HEW and in Texarkana, which is to say without anybody deciding it.
Physician survey data compiled by Medscape indicates roughly 42 percent of physicians say RVUs affect their compensation, about a quarter report compensation based entirely on them, and more than three-quarters of those eligible for bonuses must hit RVU targets to qualify.
Around 70 percent say RVUs measure productivity poorly or very poorly, against roughly 4 percent who say very well. These are self-reported survey figures rather than audited records, and should be read as physician judgment about the instrument — which, given who is being measured, is not a trivial category of evidence.
The clinical consequence is an asymmetry no dashboard can correct, because the asymmetry is structural rather than technical.
A procedure performed is an event. It has a code, a value, a timestamp. It appears.
A procedure correctly declined is nothing at all. The physician who reviews an outside pathology report, recognizes the requested intervention is unnecessary, and spends eleven minutes explaining why has produced the most valuable work of that afternoon and produced it invisibly.
The system cannot see restraint. It was not built to. And what an institution cannot see, it cannot reward — and will eventually, without ever deciding to, discourage.
● ● ●
The appointment slot is where the productivity target becomes operational. The fifteen-to-twenty-minute visit is not a clinical finding. It is arithmetic — a target divided by the available hours.
Clinical need does not distribute itself evenly. A routine postoperative check, a newly discovered pelvic mass, abnormal bleeding in a medically complex patient, a trauma-informed examination, and a conversation following an abnormal cervical screening result are not the same encounter. A uniform slot treats time as an organizational input to be standardized rather than a clinical resource to be allocated by complexity.
Research on preventive care shows what the constraint costs. Asked to manage hypothetical patients, physicians expected to address roughly four to five preventive services in a twenty-minute encounter, compared with nine to eleven in a forty-minute one. What falls off the list is exactly what a metric cannot detect: the counseling, the screening question, the second concern the patient mentions while reaching for her coat.
Then there is documentation. The electronic health record was sold as a clinical instrument and arrived as a billing and compliance instrument. The Surgeon General’s advisory on health worker burnout summarized time-motion work — most influentially Sinsky and colleagues in 2016 — indicating roughly two hours of administrative and record work for every hour of direct patient care. That figure is now a decade old and has been cited past the point of usefulness, but subsequent audit-log research confirms its direction. Work that will not fit inside the encounter migrates to the evening. The profession has a name for it: pajama time.
Honesty requires a counterweight. Survey research on ambulatory clinicians finds that chaotic work conditions and lack of control over workload explain substantially more variation in burnout than record design alone.
The software is not the disease.
It is the most visible symptom of a management system that assigns more work than the schedule contains and leaves the remainder to the physician’s own hours.
● ● ●
Patient experience deserves measurement. Patients are the only people positioned to report whether they were treated with respect, whether they understood what they were told, whether anyone listened. The Agency for Healthcare Research and Quality is explicit that its CAHPS surveys are designed for precisely the things patients are qualified to assess.
The problem is what happens when a global consumer score becomes a verdict on clinical quality. Here the empirical record is more damaging than most administrators appear to realize.
Researchers examining Press Ganey surveys in a primary care setting analyzed twelve months of responses across 652 providers, each with at least three hundred responses, covering the ten questions specific to the care provider. The ten questions correlated with one another between 0.86 and 0.96. Factor analysis found no independent dimensions — a one-factor model was sufficient. Only eight of the 652 providers showed statistically significant differences among their own ten scores.
Read that carefully, because it is the whole case. An instrument marketed as diagnostic, capable of identifying which specific behaviors a physician should improve, is in practice measuring one thing and reporting it ten times. Whether the patient liked the doctor. It has almost no power to distinguish among physicians and essentially none to distinguish among the dimensions it claims to assess.
Separate work in emergency medicine found the same physicians receiving significantly different scores depending on which of their system’s buildings they worked in that day — newer facilities with private rooms scoring markedly higher than older shared-space ones. The physician was the constant. The architecture was the variable.
Then the mechanism that matters most. In 1,319 outpatient visits by 1,141 adults to 56 family physicians, published in JAMA Internal Medicine in 2018, denial of a request was associated with lower satisfaction after adjustment for a broad range of patient characteristics. A denied referral cost roughly twenty percentile points. A new non-pain medication, roughly twenty. A laboratory test, roughly nine.
The null findings matter as much as the positive ones. Denials of antibiotic and radiology requests showed no significant penalty. Patients do not punish every refusal. The pattern is specific, and the specificity makes it more credible.
A related study of a direct-to-consumer telemedicine platform found physicians who prescribed antibiotics more frequently received better satisfaction ratings — reaching the top quartile generally required prescribing in at least half of respiratory encounters. It could not assess appropriateness and covered a single platform, but it documents a measurable conflict between stewardship and consumer scoring.
A word about the patient who arrives having diagnosed herself online, because this is where the argument is usually made badly. Roughly two-thirds of patients consult the internet before an appointment, and physician discomfort with this is well documented. But the population-level data is mixed — a Flemish study of general practice found online searching correlated mostly with increased confidence in physicians. Physician perception and population outcome diverge here, and an essay that leans on the anecdote is leaning on the weaker evidence.
The stronger claim does not require patients to be foolish, and I would not make it if it did. It is this: the instrument used to evaluate physicians reliably penalizes the word “no,” and it cannot distinguish a justified clinical refusal from dismissiveness.
Where the request originated is beside the point. A physician whose compensation or standing depends on that score is being paid, at the margin, to say yes.
● ● ●
Before this becomes a case against measuring medicine, the other side has to win an argument, and in medicine it wins more of that argument than anywhere else in this series.
Anesthesia belongs first. Over roughly three decades the specialty drove its mortality down by better than an order of magnitude, and it did it by counting: closed-claims analysis to establish what was actually killing people rather than what everyone assumed was killing them, monitoring standards written and then required, outcomes tracked afterward to see whether the standards had worked. No physician’s judgment was displaced by any of that. Judgment was told, for the first time, where it had been reliably wrong.
Infection control has the same shape. When intensive care units began counting central-line bloodstream infections and reporting the counts, the counts came down, and they came down because counting made a diffuse failure visible to the people positioned to stop it. No clinician can perceive an infection rate. It is not available to the senses or to experience. An institution can see it, and only if it counts.
That is measurement doing precisely what the builders in all five of these essays hoped it would do, and it should be conceded without hedging. The line I am drawing is not between measured medicine and unmeasured medicine, which would be a foolish line and I would lose the argument. It is between an instrument that tells a clinician something he could not otherwise know and an instrument that tells an administrator what a clinician is worth.
The honest complication is that the boundary moves, and the same instrument can cross it. Publish a cardiac surgeon’s risk-adjusted mortality and you have given the public real information it had no other way to get. You have also given that surgeon a reason to decline the sickest patient on the list. How much of that actually happened is contested in the literature and I am not going to pretend it is settled. What is not contested is that the question is live — which is more than can be said for the RVU, where nobody appears to have asked it at all.
● ● ●
The American Society for Colposcopy and Cervical Pathology publishes risk-based management guidelines for abnormal cervical cancer screening. They are excellent science, and that should be said plainly.
Their governing principle — equal management for equal risk — holds that management should follow a patient’s estimated risk of cervical precancer within a defined interval rather than the result of any single test.
The 2019 guidelines refined this further, incorporating screening history alongside current results and defining explicit clinical action thresholds. Physicians who follow it catch disease earlier.
Read the guidelines themselves and you find something a compliance dashboard cannot accommodate. Among the guiding principles is a plain statement that guidelines cannot cover all clinical situations and that clinical judgment is advised, particularly in circumstances the guidelines do not address. The authors also acknowledge that some recommendations rest on expert consensus where evidence is limited.
The society that wrote the guideline built the exception into it. A corporate audit that flags every departure as presumptive failure is therefore more rigid than the professional body whose science it claims to be enforcing.
That sentence is the whole series in one line, transposed to a clinic. It is the Job Corps cost ceiling in the second essay, where a statute capped one definition of cost and the useful work moved to the definition. It is the Texarkana contract in the fourth, where a reading test built to tell a teacher what a child could do stopped being able to tell anyone anything the moment eighty dollars depended on the score. In each case the instrument was fine.
What changed was what the institution decided to do with it.
Two functions look identical on a screen. Decision support presents relevant evidence and leaves accountable judgment with the clinician. Managerial command assumes the metric is the care and treats contextual deviation as a defect.
- Guidelines are evidence organized for the physician.
- Compliance systems can turn them into evidence used against the physician.
The remedy is not to abandon standardization, which would be its own failure. It is comply-or-explain. A system should identify unexplained departures, missed surveillance, and undertreatment — that is patient safety. But the physician must be able to document why an exception was appropriate: uncertain pathology, incomplete outside records, pregnancy, competing medical risk, patient preference, inability to secure follow-up. And review should be conducted by qualified peers with access to the full record, not by an automated numerator over a denominator.
● ● ●
Physicians in employed practice frequently describe an asymmetry in test ordering. Primary care orders freely, specialists find their orders constrained by prior authorization or credentialing scope, and the revenue appears to follow the order.
The intuitive version of this complaint — that the ordering physician is personally paid for the order — is usually wrong, and stating it that way invites a correction that discredits everything around it. The federal self-referral statute, the Stark Law, restricts precisely that arrangement for designated health services including laboratory and imaging.
The accurate version is more troubling. What moves is not physician compensation but organizational revenue: facility fees, laboratory and imaging payments, downstream capture when the service stays inside the network. A Massachusetts study of vertical integration found that when primary care physicians entered relationships with large health systems, specialist visits rose roughly 23 percent and total medical spending roughly 6 percent, with in-system utilization increasing and readmission outcomes not improving. The Government Accountability Office has documented Medicare spending increases driven by services migrating from physician offices to higher-paid hospital outpatient settings.
The physicians are not competing with one another for that money. It is moving above them both. What has been distributed to them is the friction — the prior authorization, the ordering restriction, the patient’s frustration when the specialist explains she cannot order what she knows the patient needs.
● ● ●
Obstetrics and gynecology illustrates the whole pattern with unusual clarity.
Independent ownership in the specialty has declined more steeply than in internal medicine or family practice. Private equity entered women’s health early, and the sector now includes multiple national platforms built by roll-up acquisition. Practices above roughly five million dollars in normalized EBITDA have commanded multiples of ten to fourteen times, which creates powerful incentives to consolidate whether or not there is clinical rationale for it.
The detail that should stop a reader is the exit horizon. Private equity ownership typically operates on a three-to-seven-year timeline to sale.
Set that beside the clinical calendar. Surveillance after an abnormal cervical screening result runs one, three, or five years under the ASCCP framework. A woman’s relationship with her gynecologist can span three decades — contraception, pregnancy, delivery, perimenopause, cancer screening across the whole arc. The financial planning horizon of the practice’s owner is now shorter than the medical interval it exists to serve.
Research published in Obstetrics & Gynecology has warned that private equity ownership in the specialty tends toward higher-margin service utilization and substitution of advanced practice clinicians for physicians. Consolidation research has associated hospital mergers with increased travel distance for Medicaid labor-and-delivery admissions, a rising share of deliveries at safety-net hospitals, and a declining share at hospitals with neonatal intensive care capacity. Most obstetric outcomes did not change significantly in that work, and it examined an earlier merger period — it is not a verdict on every transaction. But it shows that consolidation alters not only price but where a woman delivers and what is in the building when something goes wrong.
One more finding. Adjusted analyses consistently show women physicians at meaningfully higher burnout risk than men. Obstetrics and gynecology is a majority-female specialty combining high communication demands with compressed scheduling. The burden lands where the exposure is greatest.
● ● ●
In 2018, Simon Talbot and Wendy Dean argued that the profession had misdiagnosed itself. The word burnout locates the failure in the physician — insufficient resilience, inadequate self-care, a deficiency correctable by a wellness module. The term they proposed instead came from combat psychiatry: moral injury. The wound sustained by knowing what the patient in front of you needs and being structurally prevented from providing it.
The distinction is not semantic. It determines where an institution looks for the cause. Burnout points at the individual. Moral injury points at the system, which is where the evidence points as well.
The battalion commanders in the first essay were not insufficiently resilient either. They were highly trained professionals inside an architecture that had told them what counted and graded them on producing it. No quantity of fortitude would have corrected that, and asking for fortitude was a way of not looking at the architecture.
● ● ●
None of this argues for dismantling large health systems, which is neither achievable nor desirable. It argues for subordinating the managerial instruments to the clinical purpose, and the specifics are not mysterious.
- On compensation: an adequate salary floor, adjustment for complexity, and explicit credit for the work that generates no code — messages, coordination, records review, the conversation that prevents a procedure. Raw work RVUs should not dominate.
- On ratings: use them to find communication and access problems, which is what they actually measure, and separate the system-caused complaints — scheduling, billing, prior authorization — from anything a physician controls. Never let a small unadjusted sample become a dominant compensation input.
- On guidelines: comply-or-explain, with qualified peer review of documented exceptions rather than automated flagging.
- On scheduling: slot length by clinical complexity, because new problems, abnormal-result conversations, and patients needing interpreters or trauma-informed care require more time, and pretending otherwise does not create any.
Two of these have been tested. A 2025 cohort study found that reserving one appointment slot per half-day for record work reduced after-hours use at small cost in RVUs, which challenges the premise that every open slot must become another visit. Virtual scribes have been associated with less total record time and less work at night. An organization can absorb documentation labor rather than externalizing it onto its physicians’ evenings. That is a choice, and it is made on a spreadsheet.
And regulators should scrutinize consolidation for site-of-service shifting, referral steering, and labor-and-delivery access, rather than headline price alone.
● ● ●
Go back to the woman on the examination table.
She sees a physician in a white coat, and she assumes, reasonably, that the recommendation she is about to receive reflects that physician’s independent medical judgment. That assumption is the foundation of the entire relationship. Without it there is no reason to disclose anything and no reason to comply with anything.
Behind the encounter now stand a schedule template, a productivity formula, a compliance algorithm, a satisfaction target, a referral network, and a billing architecture. Most physicians resist these pressures most of the time, at real personal cost. But the pressures are constant, and they run one direction.
The ethical problem is not that money is present in medicine. Money was always present.
The problem is that these structures now shape the advice while remaining wholly invisible to the person who has to trust it. She cannot see the dashboard. She was never shown the compensation formula. She has no way to know whether the twenty minutes she received was a clinical judgment or a scheduling constant.
The charts in the first essay were accurate. Every number on them was real. The war was lost anyway, because the institution had confused the precision of its instruments with knowledge of the ground.
A health system should measure whether its business supports good medicine.
It should not redefine good medicine as whatever its business can most easily measure.
One essay remains, and it has to answer the question the first five have been circling. Every instrument in this series was built by a serious person to correct a real disorder, and every one of them was promoted. I built one myself. What is left to a republic that cannot stop doing this, and what does it owe the things it has no way to count?
* * * * *
A note on this essay’s provenance. I wrote it before the series existed and published it on its own, and it is the reason the series exists — I wrote it about medicine and understood while writing it that I was describing something a great deal older and larger than medicine. Then I went back and did the other five. It appears here revised, with the Pentagon material cut back to what a physician needs, because the first essay now carries that weight.
* * * * *
Charles C. Jett is an author, civic educator, and Professional Certified Coach based in Chicago. A graduate of the U.S. Naval Academy (Class of 1964) and Harvard Business School, he served during the Cold War aboard the nuclear submarine USS Ray (SSN 653), where he developed the Geographic Plot, a contact-tracking method for submarine operations. He is the author of six books, including Super Nuke!, hosts four podcasts, and writes across his Critical Skills Blog platform on history, leadership, and the health of the American republic. In his writing he employs AI tools in a limited, supporting role for research, occasional image creation, and editing, while the prose and judgment remain entirely his own. He and his wife, Dr. Nancy Church, live and co-host the Chicago Salons at Water Tower Residences.
Leave a Reply