Two cardiologists appear in your referral directory. Identical entries. Same specialty, same board certification, same hospital affiliation, same training pedigree in the same decade.
One of them runs a device clinic. Interrogations, adjustments, generator changes. She has not read a stress echo in eleven years and would not want to.
The other reads imaging all day. She has not managed an active device patient since her fellowship.
Your patient needs one of these things specifically. The directory presents them as interchangeable, because in the only language a directory speaks, they are.
This is not a data quality problem, and that distinction is the entire point of this article. The entries are not wrong. Both cardiologists really are board-certified cardiologists at that hospital. Every field is accurate.
The directory is answering a question nobody asked.
The three findings that should have changed how medicine records expertise
There is a small body of research that is well known individually and almost never assembled into a single argument. Put together, it demolishes the idea that a credential describes current capability.
Finding one: performance and years in practice frequently move in opposite directions
A systematic review by Choudhry, Fletcher, and Soumerai published in Annals of Internal Medicine examined the relationship between physician experience and quality of care. Across 62 evaluations, 32 (52 percent) showed decreasing performance with increasing years in practice.
Roughly half of measured relationships pointed the wrong way relative to the intuitive assumption that more years means better care.
Finding two: the effect is real but it is not about age
Here is where you have to be careful, and where most commentary on this topic gets sloppy.
Tsugawa and colleagues, publishing in the BMJ, examined 30-day patient mortality by physician age among hospitalists. They found mortality of 10.8 percent for physicians under 40 versus 12.1 percent for those aged 60 and over.
That is the headline everyone remembers. Here is the part almost nobody quotes, and it is by far the most important sentence in the paper:
There was no association among high-volume physicians.
Older physicians who maintained high patient volumes performed just as well as their younger colleagues. The mortality difference existed only among those whose volume had fallen.
And the picture gets more interesting still. A 2026 systematic review of 17 studies examining surgeons found that older surgeon age was generally associated with lower patient mortality, the opposite direction from the hospitalist finding.
Put those three results side by side and a much more precise conclusion emerges than "older doctors are worse," which is both wrong and unhelpful.
Age is a lousy proxy. The real variables are volume and recency, and they behave differently across domains.
In fields where the core skill is procedural and improves with cumulative repetition, more years plus sustained volume produces better outcomes. In fields where the core skill is keeping pace with rapidly changing evidence and practice patterns, sustained volume and active engagement are what protect performance, and their absence is what degrades it.
In every case, the operative variable is what the physician is actually doing now, and how much of it.
Which is precisely the variable that no directory, no credential, no board certification, and no employment record captures anywhere.
Finding three: you cannot ask people, because they do not know
The obvious fix is self-report. Let physicians describe their own current focus and expertise. Most directories already work this way.
Davis and colleagues, publishing in JAMA, systematically compared physician self-assessment against external measures of competence. Across 20 comparisons, 13 showed little, no, or an inverse relationship between how physicians rated themselves and how they actually performed.
And the failure was not random. It was worst among the least skilled and the most confident.
This is the Dunning-Kruger pattern, documented in physicians, on clinical competence. Self-assessment fails most severely exactly where accurate information matters most: the person whose skills have quietly drifted is the person least equipped to notice.
So the three findings compose into a genuinely hard problem:
- Expertise changes materially over a career, in ways that affect patient outcomes.
- The variable that matters is current volume and recency, not age or credential.
- The individual cannot reliably report it, especially when it has declined.
Meanwhile, the directory is also just wrong
Layer on top of that conceptual failure the more mundane one.
A CMS Online Provider Directory Review found that 48.74 percent of Medicare Advantage directory locations contained at least one inaccuracy. Follow-up work found that more than 40 percent of flagged listings were still wrong 500 days later, with only about 13 percent fully corrected.
So the referring physician is working with a resource that is factually wrong about basic details roughly half the time, and conceptually incapable of describing current expertise even when the facts are right.
Which raises an obvious question: what do referring physicians actually do?
They ignore it.
Research by Kinchen and colleagues in Annals of Family Medicine asked primary care physicians what factors matter when choosing a specialist:
- Medical skill: rated of major importance by 87.5 percent.
- Previous experience with that specialist: 59.2 percent.
- Board certification: 33.9 percent.
Board certification, the central organizing fact of every directory in American medicine, ranks a distant third behind two things that no directory records.
And the behavioral consequence is exactly what you would predict. Referral data indicates that roughly 72 percent of physicians usually refer to the same provider for a given specialty rather than searching for more specific expertise.
That is not laziness. That is a rational agent routing around a broken instrument. The referrer has one reliable source of information about medical skill and previous experience, and it is their own memory. So they use it, over and over, for the same handful of people, regardless of whether someone better suited exists.
What "expertise decay" actually means
Let us name the phenomenon precisely, because vague versions of this argument do real harm.
Expertise decay is not "doctors get worse as they age." The surgical data directly contradicts that, and the framing is both insulting and inaccurate.
Expertise decay is this: clinical expertise is a live, changing property of a person, while every registry that describes it updates only on discrete administrative events.
A license renews. A board certification recurs. An employer changes. Those events are what registries record, and none of them has any relationship to what a clinician actually does day to day.
Consider what happens invisibly between those events:
- A generalist becomes the regional expert in a niche condition because a patient population found her. Recorded nowhere.
- A surgeon stops doing a procedure they trained in, because volume moved to a colleague. Recorded nowhere.
- An internist takes on a new device or technique that did not exist when they certified. Recorded nowhere.
- A physician shifts to administration and now practises one day a week. Recorded nowhere.
- Someone develops deep expertise in a rare condition after managing eleven cases. Recorded nowhere, and this is the one that costs patients years of their lives.
Drift happens in both directions and neither direction is captured. The system is as blind to a clinician who has developed valuable new expertise as it is to one whose practice has narrowed.
The growth case matters at least as much as the decline case, and it is the one that makes this a positive project rather than a punitive one. Most of the professional expertise in medicine is invisible upward: people know things nobody has any way to discover.
Why nobody records it
Every party who could fix this has a specific reason not to.
Boards certify knowledge, not focus. Maintenance of certification tests general specialty knowledge. This is a deliberate design choice and defensible on its own terms. A few boards have created focused-practice designations, hospital medicine being the notable example, and they remain rare.
Employers will not publish decline. A health system that formally documented that one of its surgeons had a reduced current volume in a given procedure would be creating a discoverable record with obvious litigation implications. No general counsel will approve it.
Individuals fear devaluation. Voluntarily recording that you no longer do something is professionally costly with no offsetting benefit under current arrangements.
Payers cannot obtain it. They can see claims, which is a partial and lagging proxy, and they are contractually and practically unable to collect verified practice-focus data at scale. They struggle to maintain addresses.
Directories are compliance artifacts. They exist to satisfy network adequacy filings. Nothing about that purpose requires or rewards describing what a clinician actually does well.
So the most consequential professional information in medicine goes unrecorded, not because it is hard to write down, but because every party with the ability to record it has a reason not to.
What would work: peers, dates, and decay
If self-report fails and institutions will not record it, only one mechanism remains.
Peer corroboration, with timestamps, that fades.
Consider what a colleague can honestly attest that neither a board nor an employer nor the physician themselves can:
"I referred three complex valve patients to her in the last year and the outcomes were good." "He was in the room for four of these with me this year." "She is the person our group calls for this, and has been for two years."
That is observed, recent, specific, and verifiable by someone with no incentive to inflate it. It is the same information that already circulates through medicine informally and constantly, and it is the actual basis on which the 87.5 percent who rank medical skill first are making their judgments. It has simply never been written down.
Three design properties make this work rather than becoming another endorsement system:
Everything carries a date. An expertise claim without a date is data about an unspecified past. "Managed nine cases, most recent 2026" and "managed nine cases, most recent 2011" are entirely different facts about a person, and the second one is still useful, just differently useful.
Claims decay unless renewed. This is what makes the record honest without anyone having to declare decline. Nobody has to say "I stopped doing that." The claim simply ages, exactly as the underlying expertise does. Decay is not a punishment; it is an accurate model of reality.
"Trained in" and "currently does" are separate fields, permanently. These are different facts about a person and collapsing them is the original sin of every medical directory ever built.
The obvious objection is LinkedIn endorsements: uncorroborated social proof, gamed instantly, worth nothing. The difference is that a LinkedIn endorsement asks "does this person seem to know about X," which anyone can answer, while a useful clinical attestation asks "did you personally observe this person do X, when, and how many times," which only a genuine peer can answer and which is checkable against the person they named.
There is also a genuinely elegant consequence for late-career physicians. Legacy exposure, explicitly marked as historical, becomes a feature rather than an embarrassment. The retired surgeon who managed forty cases of a rare complication between 1995 and 2015 is exactly who you want to reach when that complication appears, and current systems make them completely invisible. A dated model of expertise makes them findable and honest about recency at the same time.
Why this is becoming urgent
Three forces are converging.
Regulation is tightening on the wrong variable. CMS accuracy requirements and No Surprises Act provisions create real penalties for directory errors, driving substantial spending on verifying addresses and phone numbers. That work will not improve a single referral decision, because address accuracy was never the binding constraint.
AI systems are training on this data. Answer engines are being asked "who should I see for this" and are drawing on directory data that is roughly half wrong and conceptually incapable of describing expertise. Errors that used to affect one referral now propagate at scale, with fluent confidence.
Practice focus is drifting faster. Subspecialization, employment changes, rapid device turnover, and new procedures mean the gap between "what my certification says" and "what I actually do" is widening for almost everyone.
What to do
If you are a clinician
Write your own dated expertise inventory. What you personally do now, roughly how often, and since when. Then a separate list: what you trained in and no longer do. Most physicians have never separated these two lists and find the exercise genuinely clarifying.
Tell your referrers what you actually want. Most specialists have never explicitly told their referring physicians what they currently focus on. Your directory entry is doing that job badly on your behalf.
Attest for others, specifically. When you observe a colleague's real expertise, say so concretely, with the number and the date. "She has done a lot of these" is worth little. "She managed four of these with me this year and I would send my own family" is worth a great deal.
If you refer
Ask the two questions the directory cannot answer. "How many of these do you do now?" and "Is this still your focus?" Awkward to ask once, extraordinarily useful thereafter, and specialists generally appreciate being asked because it means they get appropriate referrals.
Notice your own 72 percent. Most referrers send to the same person by habit. That habit was formed at some past point on some past information. Audit it once a year: is this still the right person for this specific thing?
If you build or regulate directories
Stop optimizing for addresses. Address accuracy is measurable and nearly irrelevant to referral quality. The consequential inaccuracy is expertise recency and no attestation campaign will fix it.
Require dates on every expertise field. This single change would improve directory usefulness more than a decade of accuracy enforcement.
Design for corroboration. Self-report is documented to fail worst where it matters most. Any system relying solely on it is reproducing a known error.
Frequently asked questions
Does physician performance decline with years in practice? Sometimes, and the picture is more nuanced than headlines suggest. A systematic review in Annals of Internal Medicine found 32 of 62 evaluations showed decreasing performance with increasing years in practice. But Tsugawa's BMJ study found the mortality difference by physician age disappeared entirely among high-volume physicians, and a 2026 systematic review of 17 surgical studies found older surgeon age generally associated with lower mortality. The operative variables appear to be current volume and recency, not age.
How accurate is physician self-assessment? Poor, particularly where it matters. A JAMA systematic review by Davis and colleagues found that in 13 of 20 comparisons, physician self-assessment bore little, no, or an inverse relationship to external measures, with the worst accuracy among the least skilled and most confident.
Do provider directories tell you what a doctor actually does? No. Directories record credentials, affiliations, and specialties, which are facts about training and employment rather than current practice focus. Two physicians with identical entries can have entirely non-overlapping current practices. Separately, roughly 48.74 percent of Medicare Advantage directory locations were found to contain at least one factual inaccuracy, with over 40 percent of flagged errors still uncorrected 500 days later.
What do doctors actually use to choose a specialist? Not certification. Research in Annals of Family Medicine found 87.5 percent of primary care physicians rated medical skill as of major importance and 59.2 percent rated previous experience with the specialist, while only 33.9 percent rated board certification. Since neither of the top two factors appears in any directory, roughly 72 percent of physicians default to referring to the same provider habitually.
Does maintenance of certification measure current expertise? It measures general specialty knowledge, which is a different thing from current practice focus. A physician who has not performed a procedure in a decade can remain fully certified. A few boards have created focused-practice designations, but they remain uncommon.
What would fix this? Dated, peer-corroborated attestations of current practice focus that decay unless renewed, kept separate from records of training. Self-report alone is documented to fail, institutions will not record declines for liability reasons, and only peers can honestly attest what they have actually observed a colleague do recently.
The bottom line
Medicine's directories answer the question "what is this person entitled to do?" That question has a paying customer: the institution that would otherwise carry legal and financial risk. It is answered exhaustively and enforced rigorously.
The question every referring physician is actually asking is "what does this person actually do, well, right now?" That question has no paying customer, so nobody answers it, so referrers fall back on memory, so 72 percent of referrals go to the same handful of people, and the generalist who quietly became the regional expert in a rare condition remains invisible to everyone who needs her.
Half of the measured relationships between experience and performance point in the counterintuitive direction. Self-assessment fails worst among those who need it most. Almost half of directory locations contain errors. And board certification, the organizing fact of the entire system, ranks third behind two things nobody records.
The directory is not lying. It is faithfully describing a doctor who was accurate on the day the form was filled out, and has been drifting quietly ever since.
Part of a series on the missing professional infrastructure of healthcare. Previously: Year Zero
Evidence note: sources include Choudhry, Fletcher and Soumerai in Annals of Internal Medicine (2005); Tsugawa et al. in the BMJ (2017); a 2026 systematic review of surgeon age and outcomes in the Journal of General Internal Medicine; Davis et al. in JAMA (2006); the CMS Online Provider Directory Review (2018); Kinchen et al. in Annals of Family Medicine (2004); and industry referral behavior data from Kyruus, identified as a vendor source. The 500-day error persistence figure comes from secondary reporting of a directory secret-shopper study and should be verified against the primary source before republication.
