In 1999, a cardiologist named Elizabeth Bradley and colleagues started publishing on something that seems obvious in hindsight and was not obvious at all at the time: the interval between a heart attack patient arriving at a hospital and the balloon inflating in their coronary artery.
Door-to-balloon time. Before it was measured, it varied enormously and nobody could say by how much. Institutions believed they were fast. Some were. Many were not, and had no way of knowing. Once it was measured, benchmarked, and published, something remarkable happened: the interval collapsed. National median door-to-balloon time fell dramatically over the following decade, not because anyone invented a new catheter, but because a number existed and people could be held to it.
That is what measurement does. It converts an invisible process into a manageable one.
Medicine has since become superb at this. Door-to-needle for stroke thrombolysis. Time to first antibiotic in sepsis. Referral-to-appointment. Time to surgery for hip fracture. When the profession decides an interval matters, it measures it obsessively, benchmarks it publicly, and drives it down relentlessly. This is one of the genuine triumphs of modern quality improvement.
Now here is a question, and I would like you to sit with it before reading on, because the answer is not "we do not know precisely." The answer is that nobody has ever asked.
How long does it take a clinician to find another clinician who actually knows the answer?
Not to get an appointment. Not to complete a referral. The interval from the moment a clinician thinks "I need someone who knows about this" to the moment a qualified human being gives them a substantive answer.
That interval governs the quality of every clinical decision that is not covered by a protocol. It is the rate-limiting step in the diagnostic odyssey, in complex case management, in rural care, in anything rare. And it has never been measured, benchmarked, published, or managed anywhere in the world.
Let us give it a name so we can talk about it. Expertise routing latency.
The range is absurd, and that is the finding
You would expect an unmeasured interval to be poorly characterized. What is startling here is not the imprecision but the spread. The same clinical question, asked by the same clinician on the same day, can resolve in two hours or four years depending entirely on which channel it happens to fall into.
Here is what the published evidence tells us about each channel.
Minutes to hours: the curbside. Ask someone you know. Research in JAMA found 87.5 percent of subspecialists fielded at least one such request in the prior week. Speed: excellent. Quality: a Journal of Hospital Medicine comparison found the information conveyed was inaccurate or incomplete 51 percent of the time and management advice changed in 60 percent of cases after formal consultation. Coverage: strictly limited to people you already know.
One to two hours: informal digital consult, where it exists. A Japanese app-based physician consultation service reported that 52 percent of questions were resolved within the chat itself, typically within about two hours. This is a genuinely important data point, because it demonstrates that the theoretical floor for a routed human answer is measured in hours, not weeks, when the routing layer exists.
1.2 to 2.6 days: the formal e-consult. Well-studied programs including Champlain BASE in Ontario and the Veterans Health Administration consistently report turnaround in this range. A transgender-care e-consult program reported a median response of 1.2 days requiring 18 minutes of specialist time. These programs avoid somewhere between 32 and 70 percent of face-to-face specialist referrals, with one Ontario analysis of specialist-to-specialist e-consults finding a referral avoided 69 percent of the time. Caveat: this channel exists only if your employer or payer bought it, and reaches only the specialties in that contract.
26 days: the specialist appointment. That is the average new-patient wait in large US metropolitan areas, rising to 34.5 days in dermatology, with survey series suggesting waits are up roughly 19 percent since 2022. This is the channel most clinicians default to when the informal ones fail, and it delivers, weeks later, an encounter that an eighteen-minute structured exchange would frequently have resolved.
Six to eight weeks, after months of preparation: the formal expert program. The NIH Undiagnosed Diseases Network accepts roughly 30 to 42 percent of applicants and takes six to eight weeks simply to reach an acceptance decision, following months of record assembly.
Four to eight years: the diagnostic odyssey. When routing fails completely, the interval is absorbed by the patient. The rare disease odyssey averages roughly 4.7 years in Europe and 7.6 years in the United States, with patients seeing up to eight physicians and accumulating two to three misdiagnoses.
Now look at that list as a single distribution. Two hours at one end. Four years at the other. Same question. Same clinician. Same underlying knowledge sitting in someone's head somewhere.
The variable is not knowledge, difficulty, or effort. It is which channel the question happened to enter.
And no clinician anywhere has a tool that tells them, for the question in front of them right now, which channel is fastest and most likely to reach someone who genuinely knows.
The European experiment that tells you how hard this is
Before proposing that this is fixable, it is worth studying a serious, well-funded, government-backed attempt to fix it, because the results are sobering and instructive.
The European Reference Networks were built to route rare disease questions to expert centers across member states. For rare endocrine conditions, the system connected 111 reference centers. Over four years, it logged 144 expert panels.
One hundred forty-four. Across 111 expert centers. Over four years. That is roughly one third of a panel per center per year.
The system was not badly designed, and the demand certainly existed. Evaluators identified the barriers plainly: login difficulty, lack of time, and lack of awareness. The response was to pay EUR 200 per completed panel to stimulate participation.
Three lessons are sitting in that story, and they generalize.
First, friction beats good intentions, every time. A clinician with a hard case and eleven minutes will not navigate an unfamiliar portal. They will text someone. The competing product is not another portal; it is WhatsApp.
Second, awareness is not a marketing problem, it is a routing problem. "Lack of awareness" means clinicians did not know the system existed at the moment they needed it. Any solution that requires the asker to remember an institution, log in, and choose a category has already lost to the person who remembers a name.
Third, and most importantly: expert time had to be paid for. The EUR 200 is the honest part of this story. Routing does not fail only because askers cannot find answerers. It fails because answerers have no reason to answer. Supply in this market is currently either free (and therefore rationed by personal relationship) or institutional (and therefore bounded by contract). Neither scales.
Why has nobody measured this?
If the interval is this consequential, its absence from every quality dashboard in medicine demands an explanation. There are four, and they are structural rather than accidental.
It spans institutions, and nothing spans institutions. Every measured interval in healthcare lives inside one organization's four walls. Door-to-balloon happens in one hospital. Referral-to-appointment happens within one network. Expertise routing latency begins in one clinician's head and ends, if it ends, in a different organization entirely. There is no entity that sits above institutions and could measure it end to end.
Each channel measures only itself. E-consult vendors report their own turnaround times, which are genuinely good, and have no reason to report the questions their panel could not cover. Payers measure appointment availability for network adequacy compliance. Nobody measures the cross-channel interval, which is the only one the clinician actually experiences.
The failures are invisible. This is the deepest reason. When a clinician needs an expert and never finds one, nothing happens. No ticket is opened. No metric degrades. No incident report is filed. The patient is managed reasonably by a generalist and the counterfactual is never observed. A failure that produces no artifact cannot be counted, and healthcare measurement is built almost entirely on artifacts.
Supply is unpriced. Specialist attention is the input, and outside formal programs it has no price, so it cannot be allocated. You cannot manage the latency of a system whose critical resource is distributed by friendship.
The reframe: access is partly a routing problem
The dominant frame for all of this in health policy is workforce shortage. Not enough specialists, in the wrong places, so waits are long. The AAMC projects a shortage of between 13,500 and 86,000 physicians by 2036. That analysis is serious and the shortage is real.
But hold the latency data next to it and something uncomfortable emerges.
An e-consult resolves a question in a median of 18 minutes of specialist time. The alternative pathway consumes a 26-day wait plus a full clinic appointment, and in 32 to 70 percent of cases produces the same answer.
The specialist hours needed to answer most questions already exist. They are being spent on the wrong channel: multi-week appointments delivering answers that structured minutes would have provided. The system is not only short of specialists. It is spectacularly inefficient in how it consumes the specialist time it has.
That reframe matters because the two problems have completely different solutions. If access is purely a workforce problem, the answer is a decade of training pipeline expansion. If a meaningful share is a routing problem, the answer is measurable in months and costs a rounding error by comparison.
Both are true. Only one of them is currently being worked on.
There is a vendor estimate, widely quoted and worth treating with appropriate caution because it comes from a company selling referral software, that roughly 19.7 million US referrals per year are misdirected. Take the specific number with skepticism. The direction is hard to argue with: e-consult programs report roughly 20 percent reductions in specialty care costs where implemented, which is only possible if a substantial share of what they replaced was misrouted in the first place.
How you would actually measure it
Here is the part that turns an observation into a research program. Expertise routing latency is measurable, and the method is neither expensive nor novel. It borrows directly from time-and-motion research and from the secret-shopper methodology that reshaped behavioral health policy when investigators demonstrated that provider directories were largely fictional.
Method one: the prospective diary study
Recruit 300 to 500 clinicians across specialties, practice settings, and geographies. Every time a participant experiences the trigger thought, they log a short structured entry.
The trigger has to be defined precisely, because this is where such studies usually fail. The trigger is: "I need someone who knows more than I do about this specific thing, and I do not currently know who that is."
That last clause does the essential work. It excludes routine referral to a known colleague, which is not a routing failure. It captures the actual phenomenon: the moment the internal directory comes up empty.
For each entry, capture:
- Timestamp of the trigger, and a one-line description of what was needed.
- Question type: diagnostic, management, procedural, operational, or navigational.
- Every channel attempted, in order, with timestamps.
- Timestamp of first substantive human answer, if any.
- Whether the answerer had genuine relevant exposure, judged by the asker.
- The asker's confidence before and after.
- Whether the question was abandoned, and at what point.
Then report the distribution. Not the mean, which will be useless because the distribution is heavily skewed. Report the median, the 90th percentile, and, crucially, the abandonment rate, which is the share of needs that never receive any qualified human answer at all. My strong prior, based on the Belgian finding that 73 percent of general practitioners could not name a single rare-disease information source, is that abandonment is the headline number and that it will surprise people.
Segment by specialty, practice setting, rural versus urban, career stage, and question type. The segmentation is where the policy relevance lives, because it will identify who is structurally isolated.
Method two: the standardized ask
The diary study measures lived experience and therefore inherits the biases of who participates. Pair it with a controlled protocol.
Construct 100 standardized, de-identified, realistic questions across specialties, calibrated for difficulty and vetted by clinical panels. Then have trained clinician researchers attempt to route each one using only publicly available means, following a documented protocol with a hard time cap.
Measure: how many reach a named, contactable, currently practising person with genuine relevant exposure, within what interval, and by what path. The output is a findability rate and a latency distribution per specialty and per question type, plus a league table of which question types are effectively unroutable.
This is the same logic as the secret-shopper studies that demonstrated only 18 percent of Senate Finance Committee investigators' calls to listed mental health providers produced an appointment. That methodology moved policy because it was concrete, replicable, and impossible to argue with.
Method three: instrumented routing
If a routing system exists, latency becomes telemetry rather than research: ask created, matched, accepted, answered, rated. Every event timestamped. This is the cheapest and most accurate measurement, and it has an interesting property that the other two lack.
It measures the failures. When an ask receives no qualified answerer anywhere, that is a recorded event rather than an invisible non-event. Aggregate those and you get a map of expertise deserts: the topics and geographies where the needed knowledge is not reachable by anyone.
Negative space turns out to be the most valuable data a routing system can produce, and it is completely inaccessible to any other method. A workforce map counts bodies. An expertise desert map answers a different and better question: can a question about this actually be answered by anyone reachable?
What the number would change
Suppose the work gets done and the number exists. What follows?
Benchmarking becomes possible. Once there is a median time-to-qualified-answer by specialty, every health system can ask where it sits. Some will be shocked. That was exactly the door-to-balloon dynamic, where institutions confident in their performance discovered otherwise.
The abandonment rate becomes a quality indicator. The share of clinical questions that never reach a qualified human is a genuine safety metric, and there is currently no substitute for it anywhere in patient safety measurement.
Rural and isolated practice gets a number. Every discussion of rural health disadvantage currently runs on workforce ratios. A latency and abandonment differential between rural and academic practice would be a far more direct measure of the thing that actually harms patients.
Expertise deserts become visible to policy. Knowing which conditions have no reachable expert in an entire region is directly actionable in a way that headcount projections are not.
The value of routing becomes calculable. If moving 100,000 questions a year from 26-day referrals to 48-hour answers avoids 30 percent of those referrals, the savings are straightforwardly estimable at roughly $150 to $500 per avoided referral, before counting earlier diagnosis and reduced patient time.
And the AI conversation gets a missing axis. Machine answers have driven the latency of a generically correct answer to roughly zero, which is a genuine achievement. More than 40 percent of US physicians now use one such tool daily. But nobody has measured the latency of an accountable answer from a human with relevant exposure, which is the thing you need when the machine's answer is plausible and you are not sure. Right now we are optimizing the interval we can see and ignoring the one that determines what happens with hard cases.
What you can do this week
Time your own. Next time you think "I need someone who knows about this," note the time. Note when you got a real answer, or when you gave up. Do it for a month. Almost every clinician who tries this discovers two uncomfortable things: how often the honest ending is "I gave up and managed it myself," and how completely their routing depends on three or four people they trained with.
Count the abandonments specifically. They are the invisible part, and they will not stick in memory unless you write them down at the moment they happen.
Ask your trainees. They have the least developed personal networks and therefore experience the highest latency in your department. Their answer tells you what the system does for someone without twenty years of accumulated contacts.
If you run a service, pick your ten hardest recurring questions and time them honestly. Not the protocolized ones. The ones where somebody has to know someone. Publish the result internally. It will be the first time anyone in your organization has seen that number.
If you do research, this is unusually available ground. The diary study is inexpensive, the standardized-ask protocol is straightforward, and nothing comparable has been published. The first credible measurement of expertise routing latency will be cited for a decade, because it will be the only one.
Frequently asked questions
What is expertise routing latency? It is the interval between a clinician recognizing they need specific expertise they do not have, and receiving a substantive answer from a qualified human who has relevant experience. It differs from referral wait time, which measures only the formal appointment pathway, and from e-consult turnaround, which measures only one funded channel within a contracted panel.
Why does it matter more than referral wait times? Because referral wait time measures one channel and assumes the referral was correctly routed in the first place. Expertise routing latency measures the interval the clinician actually experiences across all channels, including the failures. Its most important component is the abandonment rate: needs that never reach any qualified human, which no existing metric captures.
How long does it currently take to get a specialist answer? It depends almost entirely on channel. Roughly two hours through informal digital consultation services where they exist, with 52 percent of questions resolved in chat. Between 1.2 and 2.6 days through a formal e-consult program. About 26 days for a new-patient specialist appointment in large US metros, and 34.5 days in dermatology. Six to eight weeks for an undiagnosed disease program decision after months of preparation. Years, when routing fails entirely.
Do e-consults solve this? They solve a large piece of it, well: 1.2 to 2.6 day turnaround, 32 to 70 percent of face-to-face referrals avoided, roughly 18 minutes of specialist time per question. Their limitation is structural rather than technical. E-consults are purchased by an employer or payer and route only within a contracted panel, so they cannot reach the person outside your system who has actually managed your rare case.
Is this just a doctor shortage? Partly, and the projected shortage of 13,500 to 86,000 US physicians by 2036 is real. But the latency data suggests a large share is misallocation rather than absolute scarcity. An 18-minute structured answer and a 26-day appointment frequently produce the same clinical result. The hours often exist and are being spent on the wrong channel.
Has anyone measured this before? Individual channels have been measured extensively: e-consult turnaround, appointment wait times, diagnostic delay in rare disease. The cross-channel interval from need to qualified human answer, including abandonment, has not been published. That absence is the opportunity.
The bottom line
Door-to-balloon time did not fall because someone invented a better catheter. It fell because someone decided to measure an interval that everyone had previously experienced and nobody had ever counted.
Expertise routing latency is in exactly that pre-measurement state today. Every clinician has lived it. Every clinician has a story about the case where they finally found the right person, and a quieter story about the case where they never did. Nobody has a number.
The range appears to run from two hours to several years for the same underlying question. The rate-limiting factor is not knowledge or difficulty but which channel the question happened to enter, which in practice means whether the clinician happened to know somebody.
That is a measurable quantity. It is a manageable quantity. And right now it is nobody's key performance indicator anywhere on earth.
The first person to publish it credibly will have defined how the field discusses expertise access for the next decade.
Part of a series on the missing professional infrastructure of healthcare. Previously: Medicine's Largest Consult Channel Has No Router and Healthcare Knows Where You Work. It Does Not Know What You Know.
Evidence note: channel intervals are drawn from published e-consult program evaluations (Champlain BASE, Ontario eConsult, Veterans Health Administration), JAMA and Journal of Hospital Medicine studies of curbside consultation, published specialist wait-time survey series, European Reference Network evaluation data, NIH Undiagnosed Diseases Network published criteria, and rare disease diagnostic delay literature. The 19.7 million misdirected referrals figure is a vendor estimate and is identified as such. Specialist wait-time figures come from commercial survey series and vary by methodology.
