Field notes on the coordination failures of the health professions: why medicine can verify anyone and find nobody, what that costs, and what would have to exist to fix it.
92 field notes · RSS
Medicine records what clinicians are certified in and never what they have actually done a lot of, and recently. Every other failure in this series inherits from that one gap.
9 field notes
The distance between the specialist a directory lists and the one a physician would actually send their own family to, and what that distance costs patients.
11 field notes
The moments with the highest stakes and the least infrastructure: the rare presentation, the complication, the summons, the investigation, the diagnosis in your own family.
11 field notes
Medical research treated as a search and coordination problem: sites that never enrol, collaborators never found, reviewers who cannot be recruited, results nobody hears about.
9 field notes
A highly credentialed workforce whose jobs, coverage and credentials still move through agencies and boards, because the relationship graph that would replace them is unowned.
9 field notes
Career outcomes in medicine are decided by access to particular relationships at particular transitions, and that access is distributed by accident of institution and geography.
10 field notes
Thousands of organisations independently solving identical problems, while continuously deleting the human context that explains their own documents.
8 field notes
Clinical expertise and innovation capital cannot find each other except through brokers who are paid for the opacity between them.
11 field notes
The largest untouched surface in healthcare coordination: the gaps between professional tribes, and the places where the needed expertise is simply not present.
8 field notes
The substrate: portable verified identity, a record of contribution that survives employers, and what human expertise is for once answers are free.
6 field notes
A systematic review of 37 studies found a median 94% of adverse drug reactions never reach the formal reporting system. Even serious reactions go unreported 85% of the time. The barrier is confidence, not paperwork, and nothing fixes that except a trusted peer to ask first.
95% of surveyed family physicians report requesting informal medical advice from colleagues, yet only 11% feel satisfied doing so. None of it appears in a CV, a promotion packet, or an RVU report. The record captures billing. It skips the work that actually holds medicine together.
Roughly 5 billion people lack access to safe, affordable surgical and anesthesia care when needed. Project ECHO moves knowledge on a weekly schedule. The clinician with the question tonight cannot wait for the next hub session.
Team Rubicon says it has grown to 180,000-plus volunteers, and every one of them was vetted independently by every other organization they deploy with. There is no portable credential across humanitarian medicine, and nobody has built one.
72.5% of surveyed pharmacists report familiarity with pharmacogenomics, but only 8.8% received formal education in it. When a rare CYP variant shows up, there is no verified peer network, only static guidelines and a listserv.
Hospital CEO turnover held at 16% for three straight years, and exits climbed to 111 in 2025. Every departure takes the reasoning behind years of standing decisions with it. Board minutes record the vote. Nobody records why.
More than 80% of 17,543 German patients who received dental care abroad reported no information exchange between their treating physician overseas and their doctor at home. Only 1 in 3 got structured follow-up care.
Pelvic floor, vestibular, and lymphedema therapy each require hundreds of hours of advanced certification beyond a general license, and clinical outcomes in at least one worsen directly with delay. No cross-institution directory distinguishes these practitioners from generalists, so referral still runs on word of mouth.
Black women physicians are 55% less likely than white men to be promoted to associate professor, across 673,573 graduates. The real promotion bar is never published, it lives in the committee room, and faculty without proximity to that room prepare their case blind.
80% of active US physician-investigators run just 1 to 5 trials in three years, on razor-thin margins, while sponsor payment timelines and site solvency stay invisible to each other until after the contract is signed. Nobody discloses first, so both sides discover the truth only after it is too late to matter.
The sickest 10% of patients drive 65% of US health spending, and roughly 75% of their readmissions are potentially preventable. Even where the chart is now visible across hospitals, no system tells a clinician who else is actively treating this patient right now, or lets them reach that person today.
41% of rural physicians have no ethics committee at all, and of those who do, only 43% can reach it around the clock. 23.9% of clinicians already carry moral-injury symptoms with at least moderate functional impairment. Nothing routes the physician facing an allocation decision to anyone who has faced the same one.
45% of 14,764 studied guideline authors had a financial conflict of interest, and among studies that checked, 32% had industry payments they never disclosed. The tool that finds the expert and the database that discloses the payment have never been joined.
A median of only 11% of cardiology guideline recommendations rest on strong trial evidence; 48% rest on expert opinion alone. For that entire zone, medicine's only tools are anonymous crowd polls or a guideline process too slow to help this week.
Only 11.3% of surgical residents call their case logs highly accurate, and the record is orphaned entirely at graduation. A pilot's logbook is owned by the pilot. A physician's procedural history belongs to whichever employer happened to be watching.
Medical applicants with food or housing insecurity match at 73.3% versus 83.4% for their peers. First-generation students name limited access to advisors and mentors as their core barrier, and no map of the hidden curriculum survives past graduation.
Retained search firms charge 25-35% of first-year pay to fill medical director and CMO roles, while 46% of healthcare executives plan to leave within a year. The candidates exist inside the profession's own network. Nobody can see them.
Software explains only 19.8% of clinician EHR satisfaction variance. Two hospitals on identical Epic builds score wildly differently because the analyst who made the local configuration choices left, and nobody documented why.
In 2026, 9,696 active applicants went unmatched into a US residency. Reapplication success varies by up to 33 points depending on which gap-year route someone takes, and access to the good route is almost entirely a matter of who happens to know them.
Ontario's eConsult network avoids a formal referral in up to 81% of cases. In the largest US community health center rollout, only 10% of 13,769 providers ever submitted one in three years. The tool is proven and funded. It sits unused because nobody knows who answers well.
Doximity can show a physician has 4,200 followers. Nobody can show who that physician actually calls when a case doesn't fit the textbook. The two numbers are not correlated, and medicine has never built the second one.
Medical Reserve Corps units fell from 957 to 839 as its federal budget was cut to $3.9M during the exact fiscal years COVID demanded surge capacity. The registry built to verify volunteer clinicians in a crisis was never built to stay continuously current.
44% of physicians say they are somewhat or very likely to leave clinical practice within two years, and early-retirement consideration jumped from 34% to 46% in one year, with no verified way to find someone who already made the move.
81% of physicians now use AI in practice and 92% want more education, yet peer recommendations remain weak because nobody verifies or routes them. The physician who already solved your exact operational problem is out there and unreachable.
General surgery program director tenure fell 63% since the 1990s, from 13.6 years to 5.0. A national survey found 14% of departments produced zero written guidelines from two full years of M&M conference discussion.
US nursing schools turned away 92,672 qualified applications in 2025, mostly for lack of preceptors, not lack of applicants. The same shortage follows graduates into their first year, where pooled turnover-intention among new nurses runs at 36%.
Community-based surgeons have 90% lower odds of having a research mentor than academic peers, yet every ACGME program requires scholarly activity. Nearly half of surveyed physician-scientists are considering leaving research within two years.
55% of physicians know a colleague who has considered, attempted or died by suicide, and a UK review found that no models exist to guide support for a department afterward. One survey of anaesthetists found 22 felt supported against 179 who did not.
The physician gender pay gap held at 26% in 2025, a $122,276 average annual gap, unchanged from 2024. Employers buy MGMA's enterprise benchmarking data; the physician across the table gets a self-reported salary site and a friend's guess.
Only 6.5% of general medical journal papers have a coauthor from the study population's country, and one in five African COVID papers had no African author at all. The verified local co-investigator exists everywhere. Nobody can find her.
Surgeons hold just 3.3% of surgery-related patents issued since 1993, concentrated in 2,164 people, 97% of them men. The bottleneck is not ideas. It is a conversation with an engineer, a regulatory veteran, and a licensing officer nobody can arrange.
47% of clinical trial results are never made public, and an estimated 85% of research spending is avoidably wasted. Nobody indexes who already tried an idea and quietly abandoned it, so the same failed study gets repeated in hospital after hospital.
93% of US adults trust their own doctor, yet a two-hour deadline hands the microphone to whoever answers the email, not whoever knows. HARO, the crowdsourced fix, was discontinued and revived within five months, with no verification layer either way.
Practice membership alone explains 60.5% of how many industry payments a clinician receives. Medical affairs can find the academic KOL with a following and never the community clinician who runs the workflow. The selection signal is broken.
Visa-dependent international medical graduates matched at 54.4% in 2026 versus 67.9% for IMGs with permanent work authorization, a five-year low. The gap is not a score gap. It is a relationship gap, and nobody is paid to close it.
New physician hires saw 1.72 fewer patient encounters per clinic day than tenured colleagues in their first quarter, a gap that took two years to close. It is not a skill gap. It is a referral graph that does not exist yet, and nothing rebuilds it faster than it takes to build it the first time.
Among 270 physician authors of 2020 clinical practice guidelines, 73.7% had received industry payments and only 1.5% disclosed them accurately. An open-call panel produced a 45% women, 79% historically excluded roster. The selection process, not the disclosure form, is the fixable problem.
Clients pay $1,000 to $1,400 an hour for a physician expert call. The physician is paid $200 to $500. The gap is not the intermediary's margin problem, it is a selection mechanism: expert networks book whoever answers the screener fastest, not whoever actually knows.
87.5% of subspecialists fielded a curbside consult in the past week. Solo physicians and the 47% in practices of ten or fewer have no hallway to stand in. This is the everyday case review gap, not the emergency one, and nothing in medicine routes it.
Two-thirds of 90 surveyed hospitals had no protocol for benzodiazepine-refractory alcohol withdrawal. Roughly 34,000 protocol rebuilds happen across US hospitals every year, an estimated $200 million in labor, because the document can be shared but the person who built it cannot be found.
88% of surgery residents use group texting for case discussion. In one clinical WhatsApp group, consent was documented for only 52% of posted images and 32% of lab results. For 166 doctors serving 7 million people, the group chat is not a workaround. It is the specialist network.
66.7% of Medicare specialist visits fall outside a beneficiary's assigned ACO. Health systems call this leakage and buy software to plug it. A one-SD rise in referral concentration cuts spending 7.4% with no quality loss. The system is fighting the wrong graph.
Roughly 300,000 people in the US have Chagas disease, and fewer than 1% were treated over a seven-year period. 77.9% of surveyed obstetricians had never considered the diagnosis in patients from endemic countries. Imported malaria was misdiagnosed at first visit in 46.1% of one migrant cohort. The expertise exists in the country already.
Peer references are required at residency application, fellowship, every job, every hospital appointment, every reappointment, every payer panel, and every locum assignment. Accreditation standards require them. They go to the same few over-solicited attendings, are written from memory years later, and evaporate after use.
7.8% of physicians and trainees report a disability, with lower fulfillment, higher burnout, and reduced belonging. Those with accommodations do better. But finding out what accommodations are even possible in your specialty requires finding someone who did it, and there is no way to look.
About 43% of NCI K awardees ever obtain an R01-equivalent, at a median of 5.5 years. Physician-scientists are 1.3% of US physicians, and 49% of those surveyed had considered leaving. First R01s now arrive in investigators' forties rather than their thirties. The transition is won by assembling a team and lost by inheriting one.
General surgery program director tenure fell from an average of 13.6 years in the 1990s to 5.0 years in the 2010s. Programs with four or more director changes posted 76% board pass rates against 83% in low-turnover programs. Nothing requires or enables the outgoing director to hand over what they actually knew.
44% of 224 venture-backed digital health companies had a clinical robustness score of zero, and robustness was essentially uncorrelated with funding raised. Two companies with FDA-cleared products, each valued at $1.5 billion, went bankrupt on adoption rather than science. Clinical background alignment is now described as mandatory for diligence, and there is no way to verify it.
An MIT analysis of 300 enterprise deployments found roughly 95% of generative AI pilots produced no measurable financial impact. Implementation research has a name for the healthcare version, pilotitis, and identifies the unnecessary repetition of pilots as the core waste. The deciding variable is usually one clinician nobody can find.
More than 80% of rare conditions received no research investment over a five-year period despite more than a billion pounds spent. Patient organizations report that only serendipitous pathways exist to reach an interested researcher. Meanwhile the clinician quietly following nine patients with the condition has no idea anyone is looking.
Editors now invite six or seven reviewers per manuscript, up from two or three a decade ago, and sometimes twenty or more to secure two. At one journal, 58% of invitees simply ignored the request and 14% who accepted never delivered. The scarce resource is not reviewers. It is knowing which qualified reviewer is free this month.
In otolaryngology, 37% of applicants who sent a program signal received an interview, versus 0% of those who did not. Ten percent of anesthesiology programs received 23% of all gold signals. Away rotations average $958 each and more than half of students declined at least one on cost. The application system now runs on insider knowledge.
Hospitals and post-acute providers spend an estimated $38.6 billion a year on regulatory compliance administration across 629 discrete federal requirements, with 59 full-time equivalents at the average hospital. An audit of 2,000 hospitals found prices for identical procedures varying 468 times over within the same state, which is what independent interpretation of one rule looks like.
Fewer than 40% of healthcare improvement initiatives spread beyond one area of one organization. Michigan's collaborative quality initiatives have saved an estimated $4.08 billion, entirely within Michigan. A pediatric collaborative reports 35,000 children spared serious harm. The model works and was never built as a general capability.
There are 606,000 foreign-trained doctors in OECD countries, up 62% since 2010. Africa has roughly 1,974 neurosurgeons for 1.3 billion people. In Kenya, 51% of physician telemedicine use is physician-to-physician. The busiest cross-border clinical consultation channel on earth runs on WhatsApp between former classmates.
Among surveyed surgeons, 65% had a mentor, 28% had a sponsor, and only 48% could tell the difference. Female general surgery residents leave at 25% versus 15% for men. Black surgical residents face 2.59 times the risk of unintended attrition. Sponsorship predicts career success and nobody records who does it.
79% of physicians refer patients out of network. About a third of those referrals are avoidable with better information about colleagues' subspecialty focus. 72% refer to the same person by habit and 45% cannot readily tell who is in network. The person with the right expertise is frequently in the same building.
99% of physicians receive requests from relatives for advice, diagnosis, or treatment. In 1993, 96% of physicians extended professional courtesy to colleagues and their families, and 79% said it solidified the bonds of the profession. That reciprocity still runs medicine, and it depends entirely on who you happened to train with.
267 rural hospitals ended obstetric services between 2011 and 2021, with 63 closures in the 2020 to 2021 window alone, eliminating roughly a quarter of rural OB units. The rural generalist left managing those cases has no reliable, verified way to reach subspecialty backup when a complication exceeds their comfort zone.
Average new-patient specialist waits reached 31 days across 15 large US metros in 2025, with individual waits recorded at up to 231 days. Appointment timeliness is a top-three referral selection factor and the only one with no available data. Referring physicians are steering patients into queues nobody can see.
Of 1,792 manuscripts stating data were available on reasonable request, 93% of authors did not share when asked. Yet among researchers who shared privately, 74% said it produced authorship or a future collaboration. A decade of data-sharing policy has been solving the wrong problem.
Of 59 complaints to one neurosurgical society's conduct committee, 40 ended in sanction, and 57 of the 59 targeted plaintiff-side experts. Among 306 orthopedic malpractice experts, plaintiff experts averaged 36 years of experience versus 31 for defense. Experts are, in one journal's words, recruited and compensated without established standards.
Community oncologists treat about 80% of US cancer patients. The multidisciplinary tumor boards that set the standard of care sit at roughly 70 NCI-designated centers, and their deliberations, in the words of one analysis, exist in silos. Virtual boards work: over 96% of participants found them useful. The constraint is not technology. It is the invitation.
Independent rural physicians fell 43% between 2019 and 2024, from 21,956 to 12,467. Roughly 3,300 rural practices closed. Only 21% of freestanding US hospitals routinely succession plan. There are early-career physicians who would take over an established panel, and no mechanism connects them to the physicians retiring out of one.
51% of early-career physicians search through referrals and 40% get their first job that way. Yet median time to fill is 118 days, offer acceptance fell from 83% to 71% in a year, and employers pay search firms 20 to 35% of first-year compensation to rediscover relationships the profession already has.
A physician who has signed a contract waits an average of 112 days before seeing a patient, and one in seven waits up to six months. The top causes are not clinical. They are awaiting physician items, state licensing, and waiting on references. Every institution re-verifies everything from scratch, because nothing portable exists.
Expert networks charge clients $1,000 to $1,400 an hour and pay the physician $200 to $500. Survey work pays $1 to $6 a minute. Expert witness rates cluster at $300 to $800 an hour. The same hour of the same brain sells for a 10 to 20 times spread, and the difference is not expertise. It is channel access.
More than a third of active US physicians are within a decade of retirement. Roughly 160,000 are already inactive, 78.9% are open to returning, and the step that stops most of them is finding a colleague willing to serve as proctor. Meanwhile the day someone retires, every verified credential that would make returning easy is switched off.
There are roughly 1.5 million US adults with congenital heart disease and 509 board-certified adult congenital cardiologists, fewer than half practising it full time. Single-specialty adult hematology programs fell from 74 in 1995 to two in 2018. Medicine's greatest survival success has produced a population its directories cannot describe.
RN turnover runs 17.6% at $60,090 per departure. The national shortage is estimated at 158,600. Travel nurses cost about $91 an hour. Meanwhile the wound care specialist who could answer tonight's question works six miles away, and the only cross-institution venue for nurses is an anonymous ad-supported forum.
57.1% of US physicians received industry payments over a decade, with a median of $48. The top 0.1% averaged $1.99 million. Advisory roles typically involve a few hours a quarter for equity, nobody records what was delivered, and discovery runs on the two signals most easily gamed: payments and publications.
The US locum tenens market runs about $9.6 billion, with agencies retaining roughly 30 to 50 percent of the bill rate. Yet 58% of locum physicians work in their own community and 67% of placements are gap-fills. Physicians who would happily cover for each other cannot find, verify, privilege, or insure one another. That gap is the entire business model.
Since 2023, 22 states plus two territories have created pathways to license internationally trained physicians without US residency. Each has different rules. The map is public. The precedent is not. An IMG deciding between Tennessee, Florida and Texas has no way to find the physician who already made that exact move.
Roughly half of physicians who file an FDA Form 1572 conduct exactly one trial. The industry calls them one-and-done and blames regulatory burden. But 44.4% of them said they wanted to run another trial and no opportunity ever came. The US principal investigator pool shrank from 28,292 to 26,115 while trial volume rose.
Physicians grade their EHRs at 45.9 on the System Usability Scale, an F, in the bottom 9% of more than 1,300 usability studies across all industries. 44% of venture-backed digital health companies have no clinical evidence at all. And the primary evidence buyers use is a list of three customers the vendor selected.
At one academic center, only 61.7% of 26,020 transfer requests were accepted, and rural requests fared worse. Acceptance correlates with the receiving hospital's boarding, not the patient's need. Every 40-point rise in relational coordination between the two physicians cut time to acceptance by 25%. The bed is not the bottleneck. The relationship is.
82.3% of surgeons want a respected peer to discuss an adverse event with. Only 43.1% would use their own institution's program. Doctors under investigation show 2.08 times the rate of suicidal ideation. The support that exists is run by the institutions that are also investigating you, and the trust boundary does not match the employer boundary.
After an intraoperative complication, 52% of surgeons turn to peers, 37% get no support at all, and only 7% access formal help. Failure to rescue, not complication rate, separates good hospitals from bad. And nothing in medicine indexes the one thing that matters in the first six hours: who has managed this exact complication before.
In one large system, only 34.8% of 103,737 referral attempts ended in a documented completed appointment. Referrers report getting nothing back after external consults about 80% of the time. Three generations of electronic health records have not fixed it, because the problem was never the fax.
Peer-rated surgical skill from video review predicted complication rates of 14.5% versus 5.2% and mortality of 0.26% versus 0.05% between bottom and top quartile surgeons. Physicians know who is excellent. They act on it for their own families. And they cannot write it down.
Low-volume surgeons carried an adjusted odds ratio of 3.61 for operative death after pancreatic resection. Even among surgeons who all met Leapfrog volume standards, complication rates varied two-fold. Learning curves for robotic procedures run 15 to 55 cases. And no referring physician, patient, or credentialing committee can look up how many a given operator did last year.
Board certification is a fact about an examination, often decades old. Half of measured evaluations show performance declining with years since training, self-assessment is worst among the least skilled, and nearly half of Medicare Advantage directory locations contain an error. The consequential inaccuracy is not the address. It is the expertise.
Physicians who finished training in the last six years average under two years in their first job, against about six for previous generations. Nearly 60% leave the first post-training job within three years. Physicians now exit clinical practice at a mean age of 48.1, nine years earlier than the 2008 cohort. On July 1, the cohort that made training survivable is dissolved, and nothing replaces it.
Roughly one in nine activated trial sites never enrolls a single patient while Phase III burns about $55,716 a day. Half of investigators who run one trial never run another, and 44% of them wanted to. Ninety percent of patients want their own doctor to tell them about trials; under 1% are referred by one. This is a discovery problem wearing a recruitment costume.
81% of physicians now use AI in practice, up from 38% in 2023. One AI answer engine reports roughly 18 million clinical consultations a month. Roughly one in five LLM answers to oncology questions contains inaccurate information, and leading models elaborate on planted false clinical details in up to 83% of vignettes. Nothing routes from a questionable AI answer to an accountable human.
Fewer than 40% of healthcare improvement initiatives spread beyond one area of one organization. All 43 systems in one survey were piloting ambient AI at once, each evaluating separately with non-comparable metrics. The knowledge that separates the successful third from everyone else is held by named people and indexed nowhere.
28.7% of US physicians have been sued. The average physician spends 50.7 months of their career under an open claim, most of it on claims that end with no payment, under legal instruction not to discuss it. Medical malpractice stress syndrome is now described in the literature. The peer support infrastructure is almost entirely absent.
Senate investigators booked an appointment in 18% of calls to listed mental health providers. An Oregon study found 58.2% of directory listings were phantom providers seeing zero patients. A JAMA analysis of 449,282 physicians found 81% had inconsistent entries. The directory is medicine's only public expertise map, and it is fiction.
When a patient's primary care physician and specialist trained together, patient ratings of that specialist jump 9 percentage points, moving from the median to the 91st percentile. In a study of 40,495 referrals, 92% of eligible co-trained pairs never fired. Medicine has a proven trust graph and no way to query it.
Medicine measures door-to-balloon, door-to-needle, and time to antibiotics. It has never measured the interval that governs every non-protocolized decision: the time from needing specific expertise to getting a qualified human answer. Here is why that number matters, what we know about its range, and how to measure it.
87.5% of subspecialists were curbsided last week. Half of that information was incomplete. The curbside consult is the highest-volume clinical decision channel in medicine, and it has no routing, no record, no quality layer, and no compensation. Here is the anatomy of the problem, and a working protocol for doing it better.
Medicine has the most verified workforce on earth and no way to find the one clinician who has actually seen your case. This is the credential-expertise gap, why it costs billions, and what expertise routing would look like if anyone built it.