A primary care physician refers a patient to a gastroenterologist in March.
In September, the patient comes back for something unrelated. The physician asks how the GI appointment went. The patient says they went, they had a procedure, and they were told everything was fine, they think.
The physician looks at the chart. There is nothing. No note, no result, no record of the visit at all. Six months of that patient's care has happened in a place she cannot see, and she is now expected to resume responsibility for a patient whose recent history is a rumor.
She does what everyone does. She has her staff call the GI office and request records, which takes two calls and eleven days. Or she orders the tests again.
Now consider the same event from the gastroenterologist's side. He saw the patient. He dictated a thorough note. He clicked the button that sends it to the referring physician. He has no idea whether it arrived. In interview research, a consultant described this exactly: "I hit the button and then I'm like, do they ever get it? I have no idea."
Two competent physicians. One patient. Both did their job. And the information did not move.
Everyone blames the fax machine. The fax machine is not the problem.
The scale of the leak
The numbers here are worse than most clinicians assume, because each of us only sees our own small slice.
Referrals that never complete. A study published in the Journal of General Internal Medicine examined 103,737 referral scheduling attempts in a large system. Only 34.8 percent resulted in a documented completed appointment. A further 38.9 percent lacked appointment dates entirely.
Roughly two thirds of referral attempts did not demonstrably result in the patient being seen.
Information that never returns. Research in the International Journal of Medical Informatics found that referrers reported receiving no information back after external consultations about 80 percent of the time. The same work found practices spending roughly 10 percent of clinic time on referral management, including 20 to 30 minute phone calls simply to locate missing results.
The cross-institutional cliff. A 2023 study in Applied Ergonomics identified 47 distinct barriers to closing the referral loop across institutions and found that consultant notes reached referrers in fewer than 35 percent of cross-institutional cases.
What the specialist receives. It fails in the other direction too. IHI and NPSF work found roughly 70 percent of specialists received no information at all before the visit, and about 25 percent of primary care physicians got nothing back within four weeks.
So the specialist sees a patient without knowing why they were sent, and the referrer never learns what happened. This is not a communication channel with a defect. It is the absence of a communication channel, with documentation on both ends creating the illusion of one.
The harm has a price tag
This is frequently treated as an administrative annoyance. The safety data says otherwise.
Research by Singh and colleagues, widely cited through IHI, found that referral breakdowns are implicated in roughly 20 percent of primary care diagnostic errors. Nearly half of ambulatory missed-diagnosis claims involve a failure to follow up.
The malpractice data puts a number on it. CRICO, analyzing claims across Harvard-affiliated organizations, identified 46 referral-related claims between 2006 and 2015, with $11 million in incurred losses, and 83 percent involving high-severity harm.
Eighty-three percent high severity. These are not billing disputes. When a referral loop fails, the failure mode is a cancer diagnosis that waited nine months, an incidental finding nobody followed, or a medication change nobody knew about.
And that figure comes from a few academically affiliated systems in one region. Scaled nationally, referral-loop failure is plausibly a nine-figure annual liability cost, sitting inside a problem everyone treats as a fax issue.
Twenty-five years, three EHR generations, no change
Here is the observation that should end the technology-first approach to this problem.
In 1998, a JAMA study of curbside consultations found that 80.2 percent of subspecialists reported receiving insufficient information to answer well.
In 2023, the Applied Ergonomics study of cross-institutional referral found substantively the same thing.
Between those two dates, American medicine implemented electronic health records essentially universally, spent tens of billions of dollars in federal incentives doing so, built health information exchanges, adopted interoperability standards, moved to FHIR APIs, and established a national exchange framework.
The insufficient-information problem is unchanged.
If twenty-five years and three generations of technology have not moved a number, the number is not measuring a technology problem.
There is corroborating detail inside the Applied Ergonomics study that makes the point precisely. Of the 47 identified barriers, 66 percent were categorized as tools and technology. But the ones rated highest in severity were relational: no confirmation of receipt, no accountable recipient, and no professional norm of replying across institutional boundaries. The study also found that organizational privacy policies actively limited direct clinician-to-clinician contact.
Most of the barriers are technical. The worst ones are not.
The actual diagnosis: nobody owes anybody a reply
Strip away the interfaces and look at the underlying transaction.
A referral is a request between two professionals. In every other professional context, a request creates a light social obligation: you acknowledge receipt, you say what you did, you close the loop. That norm exists because the two parties know each other and expect to interact again.
Medicine has systematically removed both conditions.
The referrer and consultant frequently have never met. In a fragmented, employed, multi-system landscape, the referral usually travels between strangers, often across competing organizations.
Neither will ever learn how the other behaved. There is no feedback of any kind. A consultant who never sends notes back faces no consequence and receives no signal. A referrer who sends 100-page faxes with no stated question faces none either.
Nobody is paid for the reply. No billing code compensates a consultant for ensuring an outside referrer received the note. It is pure unreimbursed effort at the end of a visit that is already running late.
Institutional interests actively discourage it. A consultant at a competing system has, if anything, a mild organizational interest in retaining the patient rather than efficiently returning them.
So the loop stays open because there is no relationship, no accountability, no compensation, and no norm. The fax machine is not the cause. The fax machine is what is left when you remove all four.
Why interoperability may make it worse
This is the counterintuitive claim in this article, and I want to state it carefully because it argues against the direction of enormous investment.
TEFCA, FHIR, and modern health information exchange genuinely solve the document movement problem. Records will increasingly flow between institutions automatically.
Consider what that produces at the receiving end. A primary care physician who currently receives too little will begin receiving everything: every note, every result, every encounter from every organization the patient touches, arriving continuously into an inbox that is already the leading complaint in American medicine.
Volume is not the same as closure. A loop closes when a specific person confirms that a specific question was answered. Automated document flow delivers material without ownership, and material without ownership is not information, it is work.
The plausible outcome is that referrers move from "I never hear anything" to "I receive so much that I cannot tell what requires action," which is a worse failure mode because it looks like success on every interoperability metric.
The relational gap does not close as the technical gap closes. It becomes the only gap, and it becomes more visible.
What would actually close a loop
If the failure is relational, the fix has to be relational, and it needs to be lightweight enough that busy people actually do it.
A receipt, not a document. The minimum viable close is one line from a named person: "Saw them, here is the answer to your question, back to you." That is thirty seconds of work and it is the entire substance of what the referrer needs. Everything else can follow at leisure.
An escalation when it does not happen. The current system fails silently, which is why it fails forever. An open loop that pings someone after a defined interval is the single highest-yield intervention available, and it requires no interoperability at all.
A stated question on the way out. The 80.2 percent insufficient-information finding is symmetric. Referrers frequently send records without a question. One sentence stating what you actually want to know transforms the consultation and takes ten seconds.
Reputation that travels. This is the piece that would genuinely change behavior. Every referring physician already ranks responsiveness informally, in memory, and routes accordingly. Making that visible across institutions, as a record of who closes loops and how fast, would create the accountability that currently exists nowhere.
That last point deserves emphasis. Responsiveness is already one of the top factors in referral choice. It is simply invisible beyond each individual's personal experience. A consultant with excellent turnaround gets no credit for it outside their immediate referral base, and one who never replies suffers no consequence beyond it.
A cross-institution record of who actually replies would be one of the most valuable and least technically difficult datasets in American medicine. And no institution can build it, because it would require an organization to publish data about clinicians who do not work for it.
Compensation, eventually. Interprofessional consultation codes exist. If closing a loop is professional work, and it is, then eventually it should be paid for. Until then, it depends on norms, which is exactly why it fails.
What you can do
If you refer
Put the question in the referral. One sentence: "Specifically, I want to know whether X." Both the specialist's answer and their willingness to reply improve dramatically.
Track your open loops. Most practices have no list of referrals awaiting a response. A simple dated list, reviewed weekly, catches the ones that vanish. This is the single most effective safety intervention available to a primary care practice, and it is a spreadsheet.
Tell consultants when they close well. Almost nobody does this. It costs nothing and it is the only positive feedback a responsive consultant ever receives.
Keep your own responsiveness list. Who replies, how fast. You are already doing this in your head. Writing it down makes it survive staff turnover and lets you route deliberately.
If you consult
Send the one-line receipt. Before the full note, before the dictation is transcribed, send one line to the referrer: seen, here is the headline, full note to follow. Thirty seconds. It is the difference between a colleague who trusts you with their patients and one who wonders whether you exist.
Ask what they wanted to know. If the referral has no question, ask. It is not a criticism, and it will make your note ten times more useful.
Check whether your notes arrive. Most consultants have never verified this. The Applied Ergonomics finding of under 35 percent delivery cross-institutionally means the odds are meaningful that your careful notes are not reaching anyone.
If you run a practice or system
Measure loop closure, including out-of-network. MIPS measures reward documenting receipt of information. That is not the same as obtaining it. Measure what share of your outbound referrals produced a documented answer, including the ones that went outside your system, which is where the failure concentrates.
Fix the outbound side first. Your consultants' reply rate to external referrers is a reputational asset you almost certainly are not measuring and your referrers absolutely are.
Stop treating this as an IT project. Sixty-six percent of identified barriers are technical and the highest-severity ones are relational. Software will not create a professional obligation that does not exist.
Frequently asked questions
How often do referrals actually get completed? Less often than assumed. A Journal of General Internal Medicine analysis of 103,737 referral scheduling attempts in a large system found only 34.8 percent resulted in a documented completed appointment, with a further 38.9 percent lacking appointment dates entirely.
How often do referring physicians hear back? Rarely for external referrals. Research in the International Journal of Medical Informatics found referrers reported receiving no information back after external consultations roughly 80 percent of the time, and an Applied Ergonomics study found consultant notes reached referrers in fewer than 35 percent of cross-institutional cases.
Is referral loop failure a safety problem? Yes, and a severe one. Referral breakdowns are implicated in roughly 20 percent of primary care diagnostic errors, and nearly half of ambulatory missed-diagnosis claims involve failure to follow up. CRICO identified 46 referral-related claims over a decade with $11 million incurred and 83 percent involving high-severity harm.
Why hasn't electronic health record technology fixed this? Because the binding constraint is not document transport. A 1998 JAMA study found 80.2 percent of subspecialists received insufficient information, and a 2023 study found substantively the same after universal EHR adoption, health information exchanges, and interoperability standards. Of 47 identified barriers, 66 percent were technical while the highest-severity ones were relational: no confirmation of receipt, no accountable recipient, and no norm of replying across institutions.
Will interoperability standards like TEFCA solve it? They will solve document movement, which is necessary and not sufficient. The likely near-term effect is that referrers receive far more material without any accompanying ownership or confirmation, shifting the failure from too little information to unmanageable volume, which is harder to detect because it appears successful on exchange metrics.
What is the single most effective fix a practice can make? Maintain a dated list of outstanding referrals and review it weekly. It requires no technology, catches the referrals that disappear silently, and directly addresses the failure mode implicated in a fifth of primary care diagnostic errors.
The bottom line
A referral is a request between two professionals. Medicine has industrialized it into a document transfer between institutions, and then spent twenty-five years and enormous sums trying to make the document transfer more reliable.
The document transfer got better. Two thirds of referral attempts still do not demonstrably result in a completed appointment. Four out of five referrers still hear nothing back from external consultations. And the same insufficient-information finding from 1998 reappears in 2023 essentially unchanged.
Somewhere a gastroenterologist is clicking send and genuinely not knowing whether anyone receives it. Somewhere a primary care physician is asking a patient how the appointment went because that is her most reliable source of information about her own patient's care.
Both of them are doing their jobs. Neither of them owes the other anything, neither will ever find out how the other behaved, and nobody is paid for the thirty seconds it would take to close the loop.
The fax machine has been blamed for a generation. It was never the fax machine. It was that we built an information system for a relationship that no longer exists.
Part of a series on the missing professional infrastructure of healthcare. Previously: Where Would You Send Your Own Mother?
Evidence note: sources include the Journal of General Internal Medicine (2018) analysis of 103,737 referral attempts; International Journal of Medical Informatics (2023) on referral management burden; Applied Ergonomics (2023) on cross-institutional loop closure barriers; IHI and NPSF work on referral management including CRICO claims analysis; Singh and colleagues on diagnostic error in primary care; and JAMA (1998) on information sufficiency in consultation. Single-system referral completion figures may not generalize across all practice environments.
