We teach medical students and residents to sit at the bedside, or even to kneel. Get to eye level. Don’t tower over someone lying in a hospital bed. Ask open-ended questions. Meet patients where they are. Sometimes I hear a trainee call a patient ‘friend’ and it makes me cringe a little, but I understand what they are trying to do. They are trying to make an inherently unequal interaction feel less unequal. They are trying to communicate warmth, respect, a message: I see you as a person. I believe in the impulse, even if I do not always agree with the execution.

But you can kneel at the bedside and still look down on someone.

Years ago, I took care of a Spanish-speaking roofer who was admitted to the ICU with diabetic ketoacidosis. He also reported weakness in both legs. By the time he was transferred to my care, the weakness had generated an enormous diagnostic evaluation – consultations, imaging, tests for increasingly unusual things. He had become a diagnostic mystery. Nothing had revealed an answer.

Except there was a basic fact about this patient that I couldn’t reconcile with the narrative: he was a roofer.

How long had this weakness been going on? If he had profound bilateral lower extremity weakness, how exactly had he been making his living on a roof?

I reviewed the chart. I went to the bedside. I walked through the history with him again. I examined him. His knees were swollen with effusions. Then, with an interpreter, I finally asked him what we probably should have asked much earlier:

When you say your legs are weak, what exactly do you mean?

Do you mean they don’t have strength? That you feel unstable when you stand? That you feel like you might fall? Or that you can’t stand because it hurts?

The answer was the latter: he couldn’t stand because his knees hurt. The problem had started only a few days before his hospitalization. A few basic blood tests and a joint aspiration later, we had the diagnosis: gout.

The problem was not that he had a mysterious, exotic cause of weakness. The problem was that we had never clearly established what he meant by ‘weak.’

What bothered me even more was what happened afterward. When the missed diagnosis became apparent, the language barrier was offered as an explanation: he didn’t speak English. Of course it had been difficult to understand what he meant.

But that explanation never sat well with me. A language barrier can make it harder to understand a patient. It does not explain why we stop trying to understand what they are telling us.

I think about this patient whenever we talk about language and cultural concordance. Both matter. Caring for someone in their preferred language can remove an enormous barrier to understanding. Shared culture and lived experience can make some things easier to recognize and some questions easier to ask.

But in this case, a physician who spoke perfect Spanish could have made exactly the same mistake. Language and cultural concordance can help us understand someone. Neither can substitute for wanting to understand someone, nor can either tell us whether we have actually understood someone.

I once cared for another patient who had been bouncing between local emergency departments for several weeks reporting recurrent fevers. He was a young man, a little odd, slightly disheveled, and flighty, almost manic in his behavior. No one had documented a fever during his previous visits. By the time he arrived in our ED, malingering and drug-seeking had crept their way into his chart. He was about to be discharged when his temperature spiked to 102, so he was instead admitted to the hospital.

I had no idea what he had, but I remember deciding to start over with one assumption: What if the fevers are real?

Not: What if everything he tells me is true?

Just: What history would I take if I believed this patient had actually been having recurrent fevers for weeks?

So I took a fever-of-unknown-origin history. Eventually, I got to exposures.

Had he been around animals?

Yes, actually. He had recently been present for the birth of a baby goat. And then he had taken the baby goat home as a pet.

Baby goat. Fever. Mildly elevated liver function tests.

The history and findings raised the possibility of Q fever.

It was unusual enough that not everyone thought it belonged on the differential. And to be honest, I wasn’t sure either – but I didn’t have a stronger alternate explanation. So we sent the serologies.

A few weeks later, they came back positive.

Yes, this case remains in my catalog of greatest hits in terms of making an unusual diagnosis. But what is even more memorable to me is how easily we might never have asked the question that made the diagnosis. Once malingering is accepted as an adequate explanation for why someone keeps showing up in the ED reporting fevers no one else has seen, why take a meticulous exposure history for fever of unknown origin?

Once we have an explanation for why a patient’s account does not need to be taken seriously, we stop doing the clinical work we would have done if we had taken it seriously. Those explanations are not always overtly pejorative. Sometimes they sound like well-meant attempts to meet patients where they are. A patient who describes non-specific, vague symptoms becomes an unreliable historian. Once they are ‘unreliable,’ how hard do we work to figure out what they are trying to tell us? A patient who doesn’t understand our explanation has low health literacy. Once we have localized the failure of understanding to the patient, how often do we ask whether the failure might actually be ours?

And yet, patients really do sometimes give unreliable histories, conceal things from us, or outright lie. Sometimes a patient truly does not have the capacity or motivation to fully understand an illness. Taking patients seriously cannot mean believing everything they tell us or taking every answer at face value.

Taking someone seriously is different than assuming they are always right. It means trying to understand precisely what they are describing before deciding how to interpret what they have told us. We can start by assuming that the patient’s words correspond to something real from their perspective, even if we have not yet figured out exactly what that something is. It is our job to ask enough questions to figure it out.

Taking someone seriously is not merely a matter of style or bedside manner.

Taking someone seriously changes the medicine.

I have spent most of my career caring for patients in safety-net hospitals, and trainees sometimes ask me how I do it. Their questions usually contain some version of the same concern: How do you keep from despairing when so much of what makes our patients sick is beyond anything we can fix? When so many systems, and sometimes the odds themselves, seem stacked against them?

Physicians can’t prescribe someone a living wage. We cannot undo years of trauma during a hospitalization. We cannot make housing magically appear, undo racism, or repair every system that has failed someone before they arrived in our hospital. There is so much about this work that can feel futile.

But my answer has always been that there is something I can almost always give a patient.

Dignity. Respect. The experience of being seen.

The experience of knowing that someone heard what they were worried about. That someone believed their concerns were real. That someone took them seriously enough to try to understand.

Sometimes that changes absolutely nothing about the material conditions of a person’s life. But it still matters.

And sometimes, as with a roofer with swollen knees or a young man with a baby goat, taking someone seriously changes the medicine and leads to a diagnosis.

I think we have made clinical excellence and the work of health equity too separate in medicine. We put taking a careful history, examining the patient, and making the right diagnosis in one bucket; and measuring disparities, understanding structural determinants, improving language access, and bridging cultural differences in another.

We need all of those things. But there is no health equity without clinical excellence.

Both of these patients had treatable illnesses. Seeing them mattered not only because every patient deserves to feel heard and respected, but also because whether we took them seriously affected whether they received the right diagnosis and treatment. That, too, is health equity.

Neither of the patients in these scenarios needed me to share their cultural ancestry. I did not possess some special knowledge of their lives. I did not need a new framework for understanding people from backgrounds different from my own.

I needed to take them seriously.

For the roofer, it meant noticing that the medical story we had constructed did not make sense in context of the life he was actually living.

For the young man with fevers, it meant asking what followed if I took his fevers seriously enough to investigate them.

Being seen may seem like a gift of humanism, but in these cases, being seen was inseparable from getting the diagnosis right.

Taking a patient seriously means not letting what we know about their language, culture, or circumstances answer questions about them that we have not actually asked.

We can kneel at the bedside and feel aware of the ways a patient has been marginalized. We can understand the structural forces that shape their lives. We can know how important it is to treat them with dignity and respect. None of that guarantees that we actually see the particular person in front of us.

And when we don’t, the consequence isn’t only that a patient feels unseen. We may also misunderstand what they are telling us. We may stop asking questions too soon or even miss the diagnosis entirely.

I started with the idea that you can kneel at the bedside and still look down on someone.

But I think that gives us too much credit.

You don’t actually have to look down on someone to fail to see them. You can care about them. You can desperately want to help them. You can order $100,000 worth of tests trying to solve their problem. You can speak the same language. You can know all the right things about structural racism and cultural humility and social determinants of health.

You can do all of those things and still never ask the roofer what he means by weak.

You can genuinely want to help the strange young man who keeps coming to the ED and still never ask the questions you would have asked if you believed his fevers might be real.

I wonder why we don’t talk about this more in medicine. We teach people how to communicate respect. We teach the language of inclusion. We teach cultural humility and patient-centeredness. We tell trainees to meet people where they are.

All of these things matter, but none of them can do the actual seeing for us.

Seeing someone does not require fully understanding them. It does not require agreement, shared ancestry, language, or lived experience. It starts with the assumption that there is more to know. With taking someone seriously enough to ask, and then taking their answer seriously enough to let it change what we thought we knew.

Yes, we should keep teaching trainees to sit down. We should keep telling them to get to eye level. Kneel if it helps.

But kneeling is not the same thing as seeing the person in front of you.

Practicing medicine is the systole to the diastole of research and evidence appraisal. Hospital medicine, in particular, is a fast-paced, intellectually demanding exercise in solving complex, deeply human problems—often with near-immediate feedback.

Every stretch on service teaches me something new (usually several somethings new) despite nearly two decades in practice caring for thousands of patients. Clinical presentations are endlessly variable. Knowledge evolves. There are thousands of ways to be sick enough to require hospitalization. Few careers offer that kind of breadth and pace.

Tales from the Wards is a recurring series where I share what I learned with others.

The theme is whether AI can help

Below is a rundown of the medical inaccuracies, controversies, and errors I encountered in the wild during my last stretch on the consult service. I was planning to include cases from from my own direct care service, including two of my own errors, but the list became unwieldy. I’ll save those for a Tales from the Wards 2.5.

The throughline of these cases is whether today’s AI, or a version that exists in the next 3 years, could help.

For clinicians, please play along, and chime in the comments yes/no for each of the 8 scenarios represented by the 6 cases below.

1

I staffed a new consult for a guy disengaged with medical care with a gnarly gangrenous toe. I was told by team members who reviewed his chart that his hemoglobin A1c was 6.7% and his foot X-ray showed acute osteomyelitis. Neither of those were true. His A1c has been consistently over 15% and his X-ray said no such thing. Although after discussion with the surgeon, he did have strong clinical evidence of osteomyelitis.

Could AI be more factually accurate? Yes. This is low hanging fruit.

2

A middle-aged man presented with subacute left hand numbness and found to have cervical myelopathy with cord compression. I consulted on him postoperatively after decompression surgery. Unrelated to the reason for consult, I learned that he also developed subacute dysarthria at the same time as his hand numbness began. His brain MRI upon admission showed a “chronic” right thalamic stroke. At that time, the patient was unable to provide an accurate history and his closest family members who he lived with were not present. His history morphed into chronic dysarthria, and he was considered ‘at baseline’—perhaps the two most dangerous words in the hospital.

After a multidisciplinary discussion, I re-diagnosed his presentation as a subacute right thalamic stroke one month prior to his admission with incidental cord compression. Whether some component of his hand numbness was attributable to the cord compression cannot be teased out at the bedside. But, probabilistically, developing slurred speech from a stroke and hand numbness from spinal cord compression simultaneously would be more rare than getting struck by lightning twice.

Could AI have made the diagnosis? No. A correct diagnosis required careful history-taking skills to recognize when a patient’s history is incomplete or inaccurate and the need to corroborate symptom onset with his family who saw him everyday. An AI would be able to sift through the MRI images or the report itself to identify a possible lesion. But the radiographic chronicity combined with documented ‘chronic dysarthria’ would have stumped its compute. Perhaps a future version with ambient sight and sound capabilities would recognize the need for better history taking and bring to bear the clinical pearl that I learned that strokes on MRI can evolve from the acute to the ‘chronic’ phase as early as a few weeks out.

3

An older gentleman with mild COPD, compensated cirrhosis, and recurrent episodes of self-limited hemoptysis (typically while receiving perioperative heparin infusions) underwent salvage bypass surgery for severe peripheral arterial disease. Unsurprisingly, he developed another bout of hemoptysis while on heparin. His bleeding gradually improved, and he was discharged with stable, persistent small-volume hemoptysis after tolerating rechallenge with his oral blood thinner. The thought was that his hemoptysis was a combination of friable airways and mild coagulopathy from cirrhosis that was unmasked by heparin, but ultimately no one was certain.

The dilemma in this case wasn’t the diagnostic uncertainty. Nothing was amiss. Practicing medicine is hard. The controversy was the interpretation of his CT scan findings of his lungs. It showed new large patchy groundglass opacities (GGOs) and confluent consolidation in dependent portions of his right lower lobe (RLL), “consistent with aspiration pneumonia.” It certainly looked impressive. But, he never had shortness of breath, chest pain, chills, sweats, headaches, trouble swallowing, delirium, or loss of consciousness. He never spiked a temperature nor had leukocytosis during his hospital stay. My colleague, who first saw him at night on our consult service after the hemoptysis began and his CT resulted, showed remarkable restraint by not starting antibiotics. I saw the patient the following morning, and concurred. The patient’s pre-test probability for pneumonia was incredibly low. We transferred his care to a medicine service now that hemoptysis, and not his peripheral vascular disease, was his most pressing concern.

The new medicine attending disagreed and prescribed 5 days of broad spectrum antibiotics. Differences in management are common. Out of curiosity I reached out wondering if the patient’s clinical course evolved. It hadn’t. The decision to treat rested entirely on how “nasty” the CT appeared.

Would this patient’s clinical course have been similar without antibiotics? I would wager so, but I have no crystal ball to ascertain the counterfactual. While he had no adverse effects from antibiotics, these decisions matter because you don’t know ex ante who gets kidney injury, a drug rash, or C. difficle.

Could an AI reconcile differences in Bayesian clinical decision making? No. The patient’s pre-test probability for aspiration pneumonia was undoubtedly very low. I am doubtful that the three attendings involved would disagree. The difference in clinical management rested on how much weight to lend to the CT findings. Does ‘nasty’ GGOs in the RLL have such a strong positive likelihood ratio to push the post-test probability past the treatment threshold (say 50% perhaps)? No study tells us the likelihood ratio of increasingly “nasty” GGOs. In my experience, infiltrates without corresponding signs or symptoms are rarely diagnostic of pneumonia, and these CT findings alone were insufficient to cross my treatment threshold. No AI in the foreseeable future can practice the art and science of medicine.

4

An older man with a liver transplant for hepatocellular cancer from now cured hepatitis C was hospitalized for surgical repair of his fractured leg after he was pinned by a car against a wall. I was consulted the following day for new kidney injury after his tibia was repaired. When my team was presenting his case to me, his hemodynamics during the operating room were unknown. His alcohol history, relevant given his liver transplant, was also underreported. We reviewed his chart together and learned that he was hypotensive during the operating room and was on a vasopressor for several hours. At the bedside, I also learned that he drank so often bartenders knew him by name and from a later chart review, that his PeTH levels from another hospital (which are stored in a separate section of our EHR) were consistently through the roof, meaning that he drank like a fish. Consistent with an aphorism I learned as a wards resident (thanks Dr. Ellwood Jones!), people lie about sex, drugs, and taking their iron pills (really, all pills) With more complete information gathering, we diagnosed him with prerenal acute kidney injury from transient post-operative hypotension due to sedation and surfaced his very severe AUD, which makes liver transplant failure much higher.

Could an AI diagnose his acute kidney injury? Yes. An AI would have no difficulty reviewing hour-by-hour intraoperative vitals and identifying vasopressor use, and could reason sufficiently well enough to connect his intraoperative hypotension to his AKI.

Could an AI recognize severe alcohol use disorder and its relevance? No. An AI could locate relevant laboratory results buried elsewhere in the EHR that showed excessive alcohol consumption. My hesitation is whether an AI, without prompting, can identify the unknown knowns. Would it independently recognize severe alcohol use disorder from chart review and appreciate its relevance in a liver transplant recipient? I am doubtful.

5

An older gentleman with Parkinson’s disease was hospitalized for a spinal vertebral fracture after falling. He was nonoperatively managed with a brace, pain control, and physical therapy. He was euvolemic. Eating and drinking, albeit less than normal. With physical therapy the day prior he was noted to have asymptomatic orthostatic hypotension. This was attributed to volume depletion, and to a lesser extent, his amlodipine and finasteride. He was given intravenous fluid boluses which transiently improved his orthostasis, but was short lived. I saw him the following day, again with asymptomatic orthostasis. However, I noted that while his systolic blood pressure dropped > 20 points after standing up, his heart rate did not change with position, which was consistent with his prior orthostatic vitals. Given his Parkinson’s disease, I re-diagnosed him as neurogenic orthostatic hypotension and recommended harm reduction measures (eg. having something sturdy to grab) in the advent he ever became dizzy when standing up too quickly).

Could an AI diagnose his orthostatic hypotension correctly? Yes. A chart review recognized the the lack of heart rate change despite a sizeable drop in blood pressure (∆HR /∆SBP < 0.5 has a negative LR of 0.1 for hypovolemia) should indicate a neurogenic etiology given his euvolemic state and his Parkinson’s disease which can cause autonomic dysfunction. This requires manipulation of his vital sign changes into a ratio, an understanding of the diagnostic test characteristic from literature, and chart review of his physical exam and comorbidities. This seems solvable.

6

A young woman with uncontrolled diabetes (A1c 13%) was admitted for left upper thigh cellulitis with a phlegmon from folliculitis that evolved over several days in the hospital into an abscess requiring incision & drainage in the operating room. We were asked by the surgical service to see her for hyperglycemia the following day. She was billed to me by my team as “straightforward” with a correspondingly pithy presentation of standard hyperglycemia inpatient management – tweak her insulin, continue sliding scale, and connect to her primary care doctor. Reviewing her labs on my phone, I immediately noticed a pattern in her basic metabolic panel that I’m highly attune to in patients with poorly controlled diabetes: a high blood glucose, low bicarbonate, and an elevated anion gap. She also had ketones on her urine analysis. After brief hallway teaching on diagnosis and management of early diabetic ketoacidosis (no infusions needed, just dose subcutaneous regular insulin semi- frequently and ask your patient to drink more water), we entered her room.

During my history, I learned this was not her first rodeo. She described prior inguinal abscesses that began as “pimples”. She was obese, had scarring in her opposite groin, and suddently the diagnosis shifted from uncomplicated soft skin tissue infection to suspected hidradenitis suppurativa (HS), a chronic inflammatory skin disorder that can lead to disabling deep-seated inflammatory nodules, abscesses, and tunneling in the intertriginous regions like the groin. I am not particularly savvy in skin. Dermatologists confirmed stage 1 disease after identifying open comedones of photographs of her dressed wound that I had overlooked. In addition to the planned course of oral antibiotics, she was discharged with topical therapies.

Could AI recognize early DKA? Yes. The laboratory pattern is classic

Could an AI differentiate early hidradenitis from garden variety SSTI? No. Even perfect chart review cannot recover history that was never elicitied. Nor am I convinced that an AI with ambient vision and audio could independently synthesize recurrent inguinal infections, groin scarring, subtle comedones, and the broader clinical context than the internists and surgeons who saw her.

AI can help in 50% of these situations

I surmised that AI could help in half of the 8 discrete issues represented by 6 cases I saw in my practice.

Now it’s your turn. For each of the 8 scenarios, what’s your vote—yes or no? Which ones did I get wrong? Which cases do you think AI could solve, and which still fundamentally require human judgment? Leave your scorecard in the comments. I’ll be interested to see where the consensus emerges.

If you grew up in the 1990s, is your brain already singing the next line?

If you work in medicine, the phrase takes on a different meaning. Spend enough time in medicine and you’ll hear those three words in some iteration almost every day.

“Our jobs are so hard.”

“It’s so hard to keep up with documentation.”

“It’s so hard to recruit faculty.”

“It’s so hard to meet RVU expectations.”

“It’s so hard to avoid burnout.”

“It’s so hard to practice medicine the way we know it should be practiced.”

None of those statements are wrong, In fact, many of them are painfully true. But lately, I have been wondering if those words are crowding out something else that is essential to the practice of medicine. Focusing on what is ‘so hard’ naturally creates a survival mindset. When that becomes the dominant story we tell ourselves, the practice of medicine starts to feel like something to endure rather than a craft to pursue, and it becomes difficult to say other important things out loud:

“Medicine is a craft worth mastering.”

“Becoming a clinically exceptional physician is still one of the highest aspirations of our profession.”

“Serving patients with extraordinary clinical skill is foundational to equity.”

But is it even possible to inhabit a difficult profession without slipping into survival mode? Then I found an unexpected example: Boyz II Men.

I have a confession that would surprise most people I work with: over the past few months, I’ve spent an embarrassing number of hours watching Boyz II Men clips and interviews, catching up on what became of them after the year 2000. I expected mostly adolescent fangirl nostalgia. What surprised me was finding three professionals more than thirty years into their careers who no longer had anything left to prove but, even long after the spotlight had moved on, continued to approach every performance with the same sense of purpose they had when the world was watching.

There was no irony, no distancing themselves from their own success, no pretending this was ‘just a job’, no cynicism about singing the same song in an unglamorous venue for the 10,000^th time. Instead, they spoke with an unapologetic sincerity about the responsibility they felt towards the people who chose to spend an evening with them. They could have focused on everything that had become so hard: decades spent living out of buses and hotel rooms, singing the same songs thousands of times, aging in an industry obsessed with youth, and watching popular music move on without them. None of those realities would have been untrue. But they refused to let those things become the defining narrative of their profession. Instead of letting all the things that had become so hard also harden them, they chose instead to organize their professional identity around the privilege of what they had been trusted with, some of the most precious memories of other people’s lives.

They understood that every performance carried emotional weight. Their music had become woven into weddings, first dances, heartbreaks, funerals and countless private moments they would never witness. Recognition of that trust compelled them to continue honing their craft rather than becoming jaded by reality. They believed that being entrusted with something so precious called them to bring their very best, night after night.

Watching them made me realize that what sustained them wasn’t simply fame or talent. It was a particular understanding of professional responsibility.

My profession is obviously quite different from theirs. The stakes are different, the responsibilities are different. But at their core, both professions depend on people choosing to place something precious in our hands. And across our professions, we face the same choice: whether to let the realities of the work harden us, or to remain anchored in the purpose and craft that allow us to honor what has been entrusted to us.

In medicine, what is entrusted to us is very different. Patients invite us into some of the most vulnerable moments of their lives: the day they hear the word ‘cancer’, the night they cannot breathe, the morning someone has to decide whether to let go of the person they love. As physicians, we are entrusted with experiences we did not create but are privileged to help carry. The only adequate response to that kind of trust is a lifelong commitment to becoming better at our craft. Not because perfection is possible, but because our patients entrust us with what is most precious: their bodies, their futures, and sometimes even their lives. They deserve the very best we are capable of bringing to them.

With this realization came a nagging question: if entrustment calls us to continually refine our craft, why do we as a profession spend so little time talking about the craft itself? Why has clinical craft become oddly absent from center stage in our professional conversations?

And by craft, I do not mean technical skill in isolation. I mean the lifelong discipline of thoughtfully bringing together scientific knowledge, clinical judgement, communication and compassion in service of the person in front of us.

I worry that our focus has slowly drifted without us noticing. The realities of practicing medicine, especially in resource-constrained settings, are so demanding that we naturally gravitate toward talking about the systems surrounding clinical care. Burnout. Staffing. Policy. Structural inequities. Those conversations are essential because they address real barriers to excellent care. But they are also problems that no individual physician can solve alone.

Over time, I worry they have crowded out conversations about the one thing over which every physician still has meaningful agency: continually refining our craft. And when that happens, we lose more than opportunities to improve – we lose one of the primary ways physicians stay connected to the purpose of our profession.

Where are the conversations about how we approach becoming better diagnosticians? Better at understanding and applying evidence at the bedside? Better communicators with our patients? Where are the conversations that leave you thinking, ‘Wow, I can’t wait to try that on rounds tomorrow?’

The conversations that occupy center stage eventually become the culture of our profession.

In resource-constrained settings, where many of the greatest barriers to good care are structural, conversations about clinical craft can feel like they miss the point, or even distract from the work of advancing equity. But no matter how profound the structural barriers, every patient still entrusts us with the immediate problem that brought them to us.

That is why I wonder if we have it backwards. Perhaps excellence in our clinical craft is one of the ways equity becomes tangible for our patients. Patients do not experience equity as an abstract principle. They experience it when the right diagnosis is made despite an atypical presentation, when a medication choice reflects careful consideration of their priorities and circumstances, when a frightening diagnosis is explained with clarity and compassion. Equity requires many things – fair systems, access to care, adequate resources and sound public policy. But at the bedside, equity becomes tangible through excellent clinical care. Bringing the very best of our craft to the bedside, day after day, in unglamorous venues whether or not anyone is watching, is one of the ways equity becomes real for our patients.

Watching Boyz II Men reminded me that reality and purpose are not mutually exclusive. Difficult professions do not inevitably become cynical ones. Medicine is hard. Systems are strained. Burnout is real. None of those things should be minimized. But neither should they become the defining story of our profession. ‘It’s so hard’ cannot be the only story we tell about ourselves about our profession. Continually refining our craft is not elitist, nostalgic, or somehow at odds with equity. It is how we honor the trust our patients place in us.

If conversations shape culture, then all of us have some responsibility for the culture of our profession. For those of us in leadership, that responsibility includes deciding what conversations deserve center stage. As a division chief, I don’t want our faculty meetings to ignore operational realities. Those realities deserve our attention. They just don’t deserve our entire identity. I do not want those realities to become the only story we tell about ourselves. I want people leaving our meetings talking about a diagnosis that changed how they think, a colleague who inspired them, an evidence-based innovation that made care better, a patient who reminded them why this work matters.

I want us to celebrate our greatest hits, not because we’re nostalgic for some golden age but because our patients still come to us for the same reason audiences still come to hear Boyz II Men. They are placing something precious in our hands, hoping we will bring them our very best.

Boyz II Men have spent more than thirty years refusing to let some the realities of their profession eclipse its purpose. They continue to refine their craft because they believe the people who entrust them with life’s most meaningful moments deserve nothing less than their very best. I wonder what it would look like if medicine intentionally embraced that same ethic, one in which bringing our very best to the people who have entrusted us with what matters most to them never goes out of style.

Medicine has always been hard. Maybe what has become even harder is remembering why it was worth committing ourselves to this work in the first place, and refusing to let ‘it’s so hard’ become the only story we tell ourselves about our work.

I’ve long recognized the value of GLP1s, dating back to 2016 when the first large outcome trial was published, which I discussed in a State-of-the-Art Circulation review on applying an evidence-based framework to treating hyperglycemia in diabetes. I was an early adopter, and more recently expanded my research lab’s portfolio of work studying how Medicaid formulary policies shape GLP1 drug use, a potential policy lever to increase the dismally low use of these drugs.

With AI, the writing is also on the wall. Unlike with GLP1s, there are no large definitive trials confirming its benefits. But with AI, I don’t need an RCT to appreciate its value. I live it. I see how it makes my own ideas stronger. I see it compete with master diagnosticians. And I can already see AI outperforming humans in the physical world through autonomous vehicles, which serve as a potential solution to road traffic injuries, one of our most underappreciated public health problems.

I love living in San Francisco, despite its shortcomings. My one genuine day-to-day worry is my family getting hit by a car while walking. So when Waymos first arrived on the scene in 2022, I was suspicious. Anytime a Waymo was stopped at an intersection, I would jog across the street to avoid getting smushed if the sensors went haywire. Now, they’re so commonplace that I don’t think twice. It’s the human drivers I’m worried about. On more than one occasion a human driver has attempted to turn into the intersection while I was still crossing. These near misses are more than hypotheticals. Nationwide there are over 40,000 deaths each year from car crashes.

In 2014, San Francisco committed to Vision Zero, a 10-year policy initiative aimed at eliminating traffic deaths by 2024. While the city has implemented a range of interventions, such as street redesign, speed management, and enforcement, fatal and severe injury rates have proven stubborn, with more traffic deaths in 2024 than in 2014. On average, someone is brought to the trauma hospital where I work every 15 hours because of a car crash, accounting for roughly 1 in 5 hospitalizations requiring trauma surgery care. More recently, a car blew through a red light near my home, an intersection I’ve crossed countless times, crashing into a GoCar, ejecting its occupant and crushing the pedestrian, breaking his hip. Brutal.

So when I saw the Waymo Safety Report, I was both impressed and curious. Compared to human drivers, Waymo vehicles reportedly reduced crashes by nearly 90%. But company safety reports, while encouraging, aren’t enough. The question that mattered to me wasn’t whether Waymos crash less, it was whether autonomous vehicles reduce injuries. If Waymo’s report was true, trauma centers should see fewer patients. Then it hit me. Our UCSF team based at Zuckerberg San Francisco General Hospital and Trauma Center – The Hospital Outcomes and Policy Evaluation (HOPE) Lab – is uniquely positioned to answer this exact question. San Francisco is one of the few places where autonomous vehicle use is substantial, and our hospital is the city’s only Level I trauma center.

That’s why, in partnership with Waymo, our multidisciplinary team of investigators in hospital medicine, trauma surgery, emergency medicine, epidemiology, and biostatistics is launching the AVERT Study: Autonomous Vehicles’ Effect on Reducing Trauma. Rather than asking whether Waymo vehicles crash less, we are asking a more important question: are traumatic injuries from car crashes lower for autonomous vehicles than human drivers?

One obvious question is whether a study funded by Waymo can be trusted. I’ve spent enough of my career thinking about biases from industry-funded research to take that concern seriously. To strengthen confidence in our findings, we will have exclusive access to the data, retain complete independence over analyses and publication decisions, preregister our protocol, and establish an independent study advisory board.

I’m bullish on AI. But I’m also a scientist. Evidence matters. If autonomous vehicles truly reduce traumatic injuries, they could become one of the most important public health interventions of our generation. If they don’t, we should know that too. That’s the question we’re setting out to answer. Stay tuned to a publication near you.

Eric Topol, a key opinion leader, called it the “best proof” of the concept to date.

STAT News cautiously amplified the findings.

Patients seized on the hope, calling their clinics to reschedule infusions to the morning.

The academy moved quickly, too. In just three months, the study was cited 22 times. That’s remarkable given that most papers take many months to move from conception to publication.

But there was just one problem. The trial wasn’t true. After an editorial investigation, Nature Medicine retracted the study on June 24^th.

Celebrate the retraction?

Some celebrated the retraction as proof that science will self-correct. Don’t get me wrong. I do agree that, over time, science bends toward the truth. Sadly, untruths can last years, sometimes decades or even centuries. But eventually we move past hype, fads, and politics.

Others celebrated the editors for their swift action, retracting the study only months after publication. Again, I believe we should incentivize this behavior, including encouraging authors to come forward and admit mistakes that might otherwise be buried—even when the errors are genuine—for fear of reputational harm.

Is this an example of science self-correcting, led by heroic editors who did the right thing?

No, we should all feel profoundly sad about the state of science

I first learned of the retraction from an email by Rebecca Robbins, a New York Times journalist, reaching out for comment based on my earlier AAOx2 post in February, apropos of the situation, titled, “Retract Until Verified.”

I wish I could say I got lucky. Or that I have some paranormal psychic power. Or that I possess prodigy-level critical appraisal skills.

The truth is much simpler, which makes me profoundly sad for the entire research enterprise: from funders to authors to editors to peer reviewers to the scientific public to experts to the media. Countless responsible entities were asleep at the wheel. Or maybe they were never awake to begin with.

Either way, it’s just another tale in a very large book breaking at the spine about the sad state of science.

So how did I pen a post four months before an official retraction?

I read the study.

Seriously, I don’t think many make it past the tweet, let alone the abstract.

The top-line findings were extraordinary—turning the knob on the clock had the same effect as the drug itself compared with placebo. As Reverend Bayes would say, extraordinary claims require extraordinary evidence.

Yet, the more I read, the more suspicious I became. No treatment-related adverse effects, no loss to follow-up, and a large, clean separation in infusion times that defies the messiness of routine clinical practice.

These red flags were compounded by many discrepancies surfaced by other unsung heroes, namely @houndcl (why is their Twitter account suspended?!?!).

The only plausible conclusion I could muster was some major unseen clerical error or, far more likely, plain fraud.

Retraction cannot undo the damage

Among the 22 citations accrued in just 3 months is a meta-analysis published in another high-profile journal, JAMA Network Open, on time-of-day immunotherapy administration for advanced cancers.

The authors included 29 studies, but just one was a randomized controlled trial rated by two independent reviewers as having a “low risk of bias.” They conclude:

Although evidence from RCTs is currently limited to [lung cancer], our findings highlight the potential relevance of administration timing as a modifiable factor in cancer immunotherapy.

Will these 22 studies issue notices of error? Or, in the case of the meta-analysis, a retraction, given that the retracted study played an outsized role as the sole RCT included?

I wouldn’t hold my breath.

And if there were already 22 citations in just 3 months, I can assure you there are many more studies in peer review or production just waiting to add to the citation count. Some may scrub the reference, but I doubt they’ll abandon their fervor for circadian immunobiology in cancer therapy. More enthusiasm will yield more grants, more papers, and more press releases.

The retraction is not a death knell for the idea that circadian rhythms matter to immunobiology. But we should reserve our enthusiasm for time-of-day immunotherapy administration when the field rests largely on animal studies, confounded observational data, and implausible pharmacology for drugs whose effects unfold over weeks, not minutes or hours.

There is always room for the dreaded phrase, “future research is needed.” Maybe when we arrive at that future, things will be different.

But that seems hopeful. Sorry, Biggie Smalls, it’s the Same Song. Things have not Done Changed.

A 2023 JAMA review on the diagnosis and management of prediabetes repeats a familiar claim: metformin is an effective therapy for prediabetes.

On the surface, the logic is compelling. The use of metformin lowers the chance that people with prediabetes cross the diagnostic threshold into diabetes. Therefore, metformin should prevent diabetes complications. Open-and-shut case, right?

The problem is that prediabetes is not a distinct clinical syndrome.

It is not a disease with a recognizable constellation of symptoms, signs, pathology, and prognosis. It is a category created by a lab threshold. The difference between prediabetes and diabetes is not a biologic cliff. It is a glycemic cutpoint.

That does not make the cutpoint entirely meaningless. Risk rises as glucose rises. Labels can crudely help identify people at higher future risk. But a risk label is not the same thing as an illness, and lowering the chance of acquiring a more serious label is not the same thing as making people live better or longer. This is the central problem with prescribing metformin for prediabetes.

The strongest argument for metformin comes from the Diabetes Prevention Program and its long-term follow-up. In the 15-year outcomes study, metformin reduced the absolute incidence of diabetes by about 6 percentage points. That is real. But the key question is not whether metformin delays crossing a hemoglobin A1c or glucose threshold. The key question is whether delaying that threshold prevents the outcomes patients actually care about: kidney failure, blindness, neuropathy, amputations, heart attacks, strokes, death, or even meaningful symptoms.

On that question, the evidence is disappointing. In the same long-term follow-up, metformin did not reduce the composite microvascular outcome (which also were largely surrogate outcomes). If anything, the point estimate went in the wrong direction: 13.0% in the metformin group versus 12.4% in the placebo group, not statistically significant. That certainly does not mean it was harmful, and is more than likely just noise. But it does undercut the causal story that preventing the diabetes label necessarily prevents diabetes complications.

The second problem is treatment burden.

One possible benefit of preventing diabetes is that patients may avoid pharmacologic treatment later. But if the prevention strategy is to give metformin now, then the prevention strategy simply moves the treatment burden earlier.

This is an odd bargain. To prevent some fraction of people from eventually needing metformin, we prescribe metformin to everyone now.

That may still be reasonable if metformin clearly improves patient-important outcomes. But per above, that is wishful thinking. Without that, we are not preventing treatment burden. We are frontloading it.

This is also why the argument resembles a longstanding but debunked proposal to lower the treatment threshold for diabetes. Committing patients with prediabetes to metformin is akin to further intensify glycemic targets for initiation of metformin at an Hgb A1c of 6%, despite no improvement in outcomes even with a higher HbA1c target of 7%.

In other words, if metformin should be prescribed at an A1c of 6.0% to prevent progression to diabetes at 6.5%, then why not intensify therapy in diabetes to lower thresholds? Why is first-line therapy for diabetes with an A1c of 7.0% lifestyle change, and not metformin initiation?

As and I argued in our review in Circulation on antihyperglycemic therapy and overtreatment, the point of diabetes treatment is not to normalize laboratory values. It is to reduce complications while minimizing treatment burden and harm. The same principle should apply before diabetes is diagnosed.

To be clear, this is not an anti-metformin argument. I prescribe it regularly for type 2 diabetes. Metformin is inexpensive, familiar, and generally safe.

It is also not an argument against identifying patients at high risk and supporting intensive lifestyle interventions. Diet, physical activity, weight loss, sleep, and cardiometabolic risk reduction matter.

But the standard for medicating a risk state should not be “does the drug move the arbitrary glycemic surrogate?” It should be “does the drug improve outcomes enough to justify treating people who do not yet have the disease?”

Metformin is usually well tolerated, but “usually” is not always.

Gastrointestinal side effects like nausea, upset stomach, and diarrhea are common. Vitamin B12 deficiency is more infrequent but real. There is also the daily burden of taking a pill, the downstream monitoring, the medicalization of risk, and the subtle behavioral tradeoff that can occur when pharmacologic prevention creates false reassurance. A similar phenomenon has been described with statins, where users appeared to increase caloric and fat intake over time in a study memorably titled “Gluttony in the Time of Statins?”

None of these harms are catastrophic for most patients. But when there is no apparent benefit whatsoever, small harms matter.

That is the asymmetry of preventive medicine. When we treat symptomatic disease, patients may accept substantial downsides for a chance at relief. But when we treat people who feel well, the evidentiary bar should be higher. We should be especially cautious when the main demonstrated benefit is delaying a diagnostic label.

Summary

Prediabetes is useful as a risk marker for cardiovascular disease. It is a signal to take metabolic health seriously. It is a reason to discuss weight, diet, physical activity, blood pressure, lipids, smoking, sleep, and the other conditions that make healthy living harder.

But metformin for prediabetes is a different claim. It says that changing the timing of a diagnosis is itself enough to justify medication. That is where the logic breaks.

The clearest effect of metformin is not to help people with prediabetes live better or longer, but to alter when the label “pre” is affixed and when it is dropped, an arbitrary cutoff that lacks inherent value.

Every stretch on service teaches me something new (usually several somethings new) despite nearly two decades in practice caring for thousands of patients. Clinical presentations are endlessly variable. Knowledge evolves. General medicine is vast. There are thousands of ways to be sick enough to require hospitalization. Few careers offer that kind of breadth and pace.

This post is a potpourri of lessons from the wards: evidence-based pearls and diagnostic reflections with a few consequential misses. The goal is simple, to share what I learned with others. Tales from the Wards will be a recurring series.

Continuing aspirin (with a PPI) is safer than switching to clopidogrel in patients with vascular disease and prior peptic ulcer.

I long believed aspirin both causes peptic ulcers and increases bleeding risk, whereas clopidogrel only does the latter, and is thus safer with respect to lower the risk of an upper GI bleed. My intuition was wrong. A colleague recently pointed me back to a 2005 NEJM trial showing that continuation of aspirin (with a proton pump inhibitor) was safer than switching to clopidogrel. (see the Kaplan Meier curve below)

The key caveat is that the aspirin group received a PPI. Whether clopidogrel plus a PPI would perform similarly remains less clear. But the practical takeaway is straightforward: if a patient has a strong indication for aspirin, add a PPI rather than reflexively switching agents.

Bevacizumab can be effective for steroid-refractory radiation necrosis.

Radiation necrosis after treatment for brain metastases is something I’ve encountered infrequently. Most patients improve with time and corticosteroids. This patient did not. After a week of maximal steroid dosing, my patient remained functionally hemiplegic.

We gave a single dose of bevacizumab. The next day, he could grip my fingers, flex his elbow, and lift his leg slightly off the bed. While not a home run by any means, his progress was at least a single. And it’s far too early to judge. Time will tell.

Mechanistically, the rationale is strong. VEGF drives vascular permeability, leading to capillary leak and vasogenic edema. Bevacizumab, a VEGF inhibitor, targets that pathway. Observational data support benefit, though these study designs are too often biased. However, in this case the bias could cut both ways: either selecting patients least likely to respond since patients treated in the wild have the most severe disease, or capturing natural regression toward improvement since they are so severely affected.

Randomized data are limited but directionally supportive. The largest RCT is an open-label, multicenter trial of 112 patients that showed reduced edema on imaging and improved clinical symptoms with bevacizumab vs steroids (62% vs 43%), though recurrence rates at six months were similar (~20%). The open-label design introduces potential bias in symptom reporting, although radiographic assessments were blinded.

While the evidence base is not certain, it is strongly suggestive of benefit and one worth trying in patients who are refractory to steroids.

Diagnostic errors are stubbornly common, even at elite medical centers

My practice has been largely within safety-net hospitals affiliated with major academic institutions. These are thought to be of the highest-quality health care systems by most conventional metrics of reputation, training, and scholarship. And yet diagnostic errors are routine.

Roughly 20–30% of my patients carry some degree of diagnostic inaccuracy. Most are low stakes without direct harm. Yet, even these still lead to waste, unnecessary treatment, and near misses. For example, the diagnostic error I most commonly encounter is treating asymptomatic bacteriuria as a urinary infection. This can lead to a non-trivial risk of adverse drug effects from unnecessary antibiotics (like a rash or kidney injury), and a driver of antibiotic resistance.

My recent experience was no exception. Two cases stood out.

Not everything red is infected

A man presented with a red, painful, swollen leg which was diagnosed as cellulitis. He did not respond to oral antibiotics at home. He then did not improve after nearly two weeks of broad-spectrum intravenous antibiotics. Meanwhile, the swelling worsened, extending proximally and involving the scrotum, ultimately leading to a broader diagnostic search for causes of venous insufficiency.

A CT scan eventually revealed iliac vein compression between the artery and spine (known as May Thurner Syndrome). The interventional radiologist confirmed the compression and found an extensive thrombus with near-complete obstruction. After thrombectomy and stenting, the swelling improved

Hindsight is dangerous in diagnostic error reviews. But the clues were there early. He never had systemic signs of infection such as fevers or chills. He did not have the expected inflammatory response for what was considered severe, worsening non-responsive infection (a normal WBC). His edema extended more proximally than the upper demarcation of erythema, even at the outset of treatment. So even if cellulitis explained the erythematous portion of his leg, it did not explain why the rest of his leg was edematous. He did not respond to appropriate oral antibiotics or expectedly to broad spectrum IV antibiotics. Antibiotic failure without abscess should force reconsideration. Bacteria do not particularly care whether antibiotics are oral or IV if bioavailability is adequate. When the clinical course does not fit, the diagnosis is usually the problem.

After we ultimately treated the actual disease, his leg improved quickly.

Not all neurologic symptoms can be explained anatomically

A patient with a known L5-S1 disc herniation presented after a fall with worsening back pain. Imaging showed progression of the disc bulge. That part fit. Everything else did not.

The patient reported inability to move their left leg, bilateral numbness, and toe contractures. But across repeated exams by different clinicians from different disciplines, the examination findings were inconsistent. Different clinicians noted effort-variable strength with the ability to stand and bear weight intermittently; myoclonus that diminished with distraction; and sensory deficits that did not map to the L5-S1 neuroanatomic distribution. The patient had a positive Hoover’s sign. (Coincidentally, I was able to compare this maneuver with a patient with hemiplegia and it clearly distinguished the two!)

This is the pattern of functional neurologic disorder (FND), not structural disease due to disc herniation. I made that diagnosis, while acknowledging much uncertainty around potential secondary gain. In practice, the distinction often matters less than clinicians think since management often converges to physical therapy and psychiatric care, often with treatment of underlying mental illness and cognitive behavioral therapy.

Shorty after disclosing this diagnosis, we deprescribed an extensive medication list including topiramate, baclofen, gabapentin, benzodiazepines, hydromorphone, and morphine. The patient’s symptoms improved, although they did not abate. The most important benefit was diagnostic clarity.

A well-designed study that is underpowered, narrowly sampled, or contextually limited can meaningfully generate hypotheses. A study that is biased cannot. That distinction matters.

Hypothesis-generating work should be defined by validity with imprecision or lack of generalizability. The estimate may be noisy, but the signal is directionally credible. A small, well-phenotyped prospective observational cohort with appropriate measurement and outcome ascertainment fits this mold: limited generalizability, wide confidence intervals, but internally coherent.

By contrast, a study with major, unresolvable sources of bias produces estimates that may be extremely precise, however, uninterpretable. When the direction and magnitude of bias are unknown or unstable, the results are not “hypothesis generating”. They are noise.

This is why a massive retrospective analysis of electronic health records data is not inherently more informative than a small prospective study. In fact, it is often less so.

Scale does not rescue bias. It amplifies it. Confounding by indication, measurement error, misclassification, unclear temporality, and differential outcome ascertainment do not attenuate with sample size. They risk becoming more precisely wrong.

An Illustrative Example: Thick vs Thin Liquids in Dementia

One of the most problematic complications of end-stage dementia is trouble swallowing, known as dysphagia. These patients are at high risk of aspirating food into their airway, leading to respiratory distress, pneumonia, and death.

A common strategy to mitigate this risk is to thicken liquids. While this approach improves surrogate measures of aspiration, it has not translated into fewer clinically meaningful outcomes, such as pneumonia, in smaller trials. It also comes with downsides of poor palatability and reduced intake, raising concerns about dehydration. (For the brave soul, give the GeriPal #ThickenedLiquidChallenge a go.)

So there is a clear need for better evidence on whether thickened liquids actually help. Enter a 2024 JAMA Internal Medicine study comparing thick versus thin liquid diets in hospitalized older adults with Alzheimer’s dementia and dysphagia.

The authors executed a technically sophisticated retrospective cohort using multi-hospital electronic health record data of nearly 9,000 patients, including advanced adjustment methods. This is careful work. But the limiting issue is not effort or technique. It is the nature of the data itself.

I served as a peer reviewer for this study. At that time, I wrote:

Unfortunately, these limitations are too great to consider these findings hypothesis generating as the direction of the biases are large and unpredictable.

Here is a summary of my core critiques:

• Critical confounders omitted: Severity of dementia and dysphagia, arguably the most dominant determinants of both exposure and outcomes, are not measured.

• Exposure misclassification: “Thick liquids” was defined heterogeneously (>75% of days), with unknown adherence and no accounting for co-interventions, such as chin tuck maneuver that speech therapists teach caregivers.

• Outcome and covariate validity concerns: Reliance on unreliable administrative billing codes for key constructs like delirium, choking, and admission diagnoses.

• Unclear temporality: Due to lack of timestamps of exposure and outcome, diet changes may merely follow aspiration events, making reverse causation a highly plausible explanation.

• Internal inconsistency: Increased respiratory complications but decreased intubation in sensitivity analyses using an alternate modeling approach.

Taken together, these are not minor threats to inference. They fundamentally destabilize it. The relationship of cause and effect may not even hold. If it does, the direction of bias is potentially large and unpredictable. In the context of multiple large biases, the observed association cannot be mapped back to a plausible causal effect.

The Overreach

To their credit, the authors are unusually transparent about these limitations and explicitly caution against influencing clinical practice. That level of restraint is rare and commendable.

But there is still a conceptual overreach. The study was funded by an NIH R03 grant with the stated goal:

The findings from this proposal will be used to support the application for a large-scale clinical trial to prospectively evaluate the effects of dietary restrictions…

From this study, the authors conclude:

This cohort study emphasizes the need for prospective studies that evaluate whether thick liquids are associated with improved clinical outcomes in hospitalized patients with ADRD and dysphagia.

This framing and conclusion position the work as foundational, as laying the groundwork for future trials. That is reasonable in intent. But it hinges on the assumption that the findings provide clarity. They do not.

A study whose estimates are dominated by unmeasured confounding and ambiguous temporality does not meaningfully inform the design of a future trial. At best, it identifies that the question is important, which we already knew in this case. At worst, it suggests misallocation of scarce research funding by prioritizing misguided research directions.

Where Hypothesis Generation Actually Lives

If the goal is to generate hypotheses about thick versus thin liquids in this population, a better design is obvious: a dataset with careful phenotyping of dementia and dysphagia severity, standardized exposures, well-measured confounders, and clear temporality between diet order and outcomes. That dataset likely does not exist and would need to be prospectively assembled.

It would be small. It would be underpowered. But it would produce more interpretable findings. That is the tradeoff: imprecisely valid versus precisely invalid.

In this case, though, the hypothesis has long been established. What is needed now is not another attempt at hypothesis generation, but a more definitive and generalizable clinical trial.

When I was a medical student, someone once said to me ‘I don’t really understand the difference. Medical student, intern, resident — you’re all doctors to me.’

I bristled at the comment. At that time, I had not yet earned the title of ‘doctor.’ The distinctions mattered. They reflected a progression of responsibility, authority, and accountability. They marked the difference between observing and deciding, between apprentice and expert. For me, the title of ‘doctor’ signaled a serious transition, a weight of professional responsibility I had not yet taken on, and one I did not feel he fully understood.

Years later, I have started to wonder whether we sometimes risk making the opposite mistake — flattening that same distinction but in the opposite direction, and in doing so, naming less of the professional weight that is already present.

Across academic medicine, residents and medical students are collectively referred to as ‘learners.’ The term appears in accreditation language, institutional emails, faculty meetings and mission statements. It is meant to demonstrate a thoughtful approach to medical education, a broader effort to ensure that training is psychologically safe and free from the cruelty that once characterized parts of medical training culture.

Those efforts are necessary. Residents deserve supervision, mentorship, and protection from abuse. Medicine is better for having moved away from humiliation as pedagogy.

But residents are also entrusted with lives, the lives of their patients, the lives of real people. No patient says they are coming to the hospital to see their ‘learner.’ Patients come to see doctors.

Residents write orders. They interpret data. They explain uncertainty. They make decisions overnight. They serve as the primary team for people in their most vulnerable periods of life, when there is little margin for error. Whether or not we choose to name it explicitly, patients confer authority on them the moment care begins.

Residents have authority at the first ‘hello.’

Language shapes professional identity. What we repeatedly name becomes what feels central. Calling residents ‘learners’ is not wrong; it is simply incomplete. It names their development but not their entrusted responsibility nor their vested authority, and in doing so, it can obscure the professional weight they already carry. What we leave implicit, what we choose to not name explicitly, can shift our sense of what is primary.

Apprenticeship in medicine has never been an abstract nor classroom-based educational exercise. It has always, necessarily, occurred in service of patient care. Learning is essential but it is not the organizing principle of the hospital: patient care is. Training exists within that duty.

In rebuilding residency training to be more humane, we have clearly and rightfully reduced cruelty. The next challenge is ensuring that our commitment to excellence in patient care remains explicit and nonnegotiable. Humane training and excellence in patient care are not opposites nor mutually exclusive priorities. Psychological safety and high standards are not competing values. Residents deserve and need both support and nonnegotiable expectations.

Responsibility dignifies people. When it is explicit, clear, culturally reinforced, and visibly consequential, it stretches physicians to their potential. When it becomes secondary — buffered quietly by faculty or softened in our language — stretch can give way to comfort. Not because residents lack ability but because systems shape what is expected and what is tolerated.

The culture we are trying to build should be both humane and demanding — both protective of residents’ growth and uncompromising about the seriousness of patient care. Outstanding patient care should not be a slogan buried in orientation slides — it should be the first organizing principle of training, reinforced in expectations, role modeling, and feedback.

Residents are physicians in formation. They are supervised, developing and growing. And they are also the primary physicians — doctors — responsible for patients. Naming both truths is essential.

This clarity may be especially important for those whose authority is already questioned. Women, physicians of color, international graduates are disproportionately mistaken for non-physicians. Language that affirms professional identity does not inflate power; it reflects reality.

When I was a medical student, I resisted being called a ‘doctor’ because I had not yet earned the responsibility the title implied. The distinction mattered.

Now being on the other side of training, I see the opposite risk: residents have the title and the responsibility, but we sometimes speak in ways that make both feel provisional, labeling them as ‘learners.’ Residents are still learning, as they should be, but they are also already entrusted with real authority, authority that patients assume and rarely question. That responsibility is not something they grow into later, nor is it secondary; it is present from the beginning.

While we can choose how we describe residents within our institutions, patients cannot. For them, the resident at the bedside is their doctor.

I do not worry that residents are learners. I worry about anything that makes their responsibility to patients feel secondary, whether it is in our systems or in the language we choose. Because for patients, the responsibility is never secondary. Their expectation is simple and absolute: that their doctors will place their care first.

But her sodium was chronically low at 128, low enough to portend poor prognosis in cirrhosis. With diuresis, it fell further into the mid-120s. She felt fine. She urinated more, which over time, would decrease her abdominal swelling. But her hepatologist worried we were trading short-term relief for long-term harm, potentially increasing her risk of death by worsening her hyponatremia.

So we stopped the diuretics and focused on optimizing her sodium instead. I would normally do the opposite. But since I was not going to care for her after she left the hospital, I deferred to her hepatologist who will be calling the shots after her discharge. I’ve long wondered whether that was the right call.

Hyponatremia is a classic hospital medicine conundrum. When it’s acute and symptomatic, such as nausea, confusion, and seizures, the decision is clear: correct it. But most cases aren’t like that. They’re chronic, asymptomatic, and often stem from complex, advanced disease.

Still, the reflex is powerful. You see a sodium of 124 and immediately ask: what’s the cause, how fast should I correct it? We treat it like a problem to fix. But what if it isn’t the problem at all?

Hyponatremia is Prognostically Bad

Chronic hyponatremia is consistently associated with worse outcomes, including higher mortality, more readmissions, and functional decline.

The association is strongest in conditions where hyponatremia emerges late, such as advanced heart failure, cirrhosis, malignancy. In these settings, low sodium reflects a common physiology of systemic vasodilation, reduced effective arterial blood volume, and/or maladaptive neurohormonal activation driving water retention. In other words, hyponatremia is typically downstream of physiologic stress, not upstream of it.

That distinction matters. Our treatments, including fluid status manipulation, diuretics, and salt, primarily adjust water and sodium balance. They do not reverse the underlying disease processes driving the hyponatremia.

So the key question remains unsettled: Is hyponatremia causing harm, or simply a marker of it? Observational data cannot answer that.

A Pragmatic Randomized Controlled Trial: Methods Rundown

Against this backdrop, a new pragmatic, international, multicenter RCT was published in NEJM Evidence. The trial enrolled 2,173 hospitalized adults across nine European centers with asymptomatic hyponatremia <130 mEq/L (I doth protest mmol/L).

They excluded patients with symptoms requiring urgent correction and those near the end of life, including the patients with the utmost advanced cirrhosis. Hyponatremia could develop at any point during hospitalization. So strictly speaking, “chronic” may not be in the physiologic sense, but consistent with the greater than 48-hour convention often used clinically.

Patients were randomized to either targeted sodium correction or usual care, the latter nominally guided by European guideline recommendations but ultimately left to the treating physician.

The intervention arm followed a detailed, consensus-based diagnostic and treatment protocol as captured in a dense, two-page figure (see below merely for its complexity). Despite standardization, it remains complex and highly subjective, mirroring real-world practice where diagnosis, treatment choice, and adjustments vary across clinicians.

This is notable for two reasons. First, it highlights an inherent limitation: unlike oncology or cardiology trials, there is no clean, reproducible intervention algorithm. Second, it’s a reminder of how cognitively demanding hospital medicine is (see Oanh’s AAOx2 post on hospitalists’ expertise). Hyponatremia is rarely the only problem, and often not the main one. yet it demands considerable cognitive bandwidth.

The primary outcome was a composite of death or readmission, questionably equivalent endpoints, but not unusual. The secondary outcomes were appropriately exhaustive and included a few notable ones: 30-day recurrence or persistence of hyponatremia, quality of life, and neurocognitive function at discharge. The former assesses the sustainability of the more intensive protocolized intervention group. The latter two aim to capture subtler, but still meaningful outcomes that mortality and readmission miss.

The Findings

The enrolled patients look like the ones I actually treat: older, multimorbid, with high rates of cancer, heart failure, and liver disease. Nearly half were euvolemic, and a quarter developed hyponatremia in the hospital—so not “chronic” in the way I often consider it.

As expected, the intervention group received more treatment: more fluid, less fluid, and various adjuncts. In other words, more effort directed at improving the sodium.

It worked, modestly. Sodium rose in both groups, but about 1.3 mEq/L more in the intervention arm, with 14% more patients reaching eunatremia (60% vs 46%). The intervention did what it was designed to do: it moved the number. Though notably, usual care also moved the number because usual care already targets hyponatremia.

The separation persisted at discharge (56% vs 37% eunatremic), but it didn’t last. By 30 days, hyponatremia recurrence or persistence was nearly identical (42% vs 41%).

So what did we get for this transient ~3-week improvement in sodium?

Nothing. No difference in death, readmission, length of stay, falls, fractures, quality of life, or neurocognitive function.

A modestly improved number, without a better outcome.

Caveats (There Are Always Caveats)

You don’t boil the ocean in a single study. Before overcorrecting our interpretation, a few caveats matter.

First, a quarter of patients had hospital-acquired hyponatremia, so were not truly chronic. This likely dilutes relevance to the patients we worry about most, but also probably doesn’t change the message.

Second, the intervention is complex, because medicine is. The protocol depends on subjective assessments and clinical judgment, so fidelity inevitably varies. Eunatremia rates ranged from 30–70% across centers. One can reasonably assume similar variability within centers. This kind of heterogeneity biases toward the null, potentially masking a real effect.

Third, the treatment effect was modest and transient, on the order of weeks. By 30 days, sodium levels converged. If sustained correction is required to improve outcomes, this design may underestimate benefit. But there’s two fundamental issues. First, the underlying physiology driving hyponatremia remains unchanged. Second, the reflexive act of correcting sodium levels is methylated in a hospitalist’s DNA epigenetically from years of training and practice. So, the usual care arm, with these deeply ingrained habits, drifts closer to the protocolized intervention arm over time, and again, biases findings towards the null.

Take Home

Despite these challenges, this trial is a breath of fresh air. It directly challenges a deeply held assumption: that correcting hyponatremia improves outcomes.

The next, more uncomfortable step would be to randomize against unusual care of doing less, or frankly, even nothing, in asymptomatic patients. If there was a Polymarket on this, I would bet on conservative management.

For my patient with cirrhosis, this trial shifts the needle slightly to corroborate my worry: that prioritizing sodium over symptom relief was ill advised.

For my own practice, I’ll still tepidly correct chronic hyponatremia, especially when the physiology is reversible.

But when it’s not, I’ll be more comfortable leaving the number alone—treating it as a prognostic marker, not a therapeutic target—and redirecting effort toward interventions that actually help people liver longer or feel better.

That question stuck with me, but transfigured into a satirical version about the way modern American healthcare measures value: If Jesus practiced in a modern, insured American healthcare system, what would his coding and billing look like? What follows is a facetious but methodologically earnest attempt to answer that question.

The thought experiment

If you read the four Gospels clinically, setting theology aside for a moment, you find a surprisingly concrete case mix: paralysis, blindness, chronic hemorrhage, seizures, psychosis, edema, musculoskeletal deformity, and a few extremely high-acuity cases involving apparent death.

So, I asked: what is the collective value of his narrated healing episodes? In other words, if you treat each episode as a discrete clinical encounter, mapping the phenomenology (not the supernatural claim) to modern ICD-10 syndromes, assigning contemporary Evaluation and Management (E/M) CPT codes as a proxy for time and cognitive work, and applying 2026 CMS work RVUs (wRVU), what would Jesus’ total come out to?

Methods

Study Design: This is a retrospective narrative review of the four canonical Gospels (Gospel of Mark, Gospel of Matthew, Gospel of Luke, and Gospel of John) integrating original-language nuance (Hebrew/Aramaic terms as mediated through Greek), Second Temple Jewish context, modern medical pattern recognition, and narrative reliability (multiple attestation vs symbolic storytelling). This analysis was conducted with the assistance of ChatGPT, which was used to systematically identify and map Gospel narratives to modern ICD-10 and CPT codes and calculate aggregate wRVUs. All assumptions and framing choices were reviewed and curated by the author.

Definition of healing: Included healings, exorcisms, and resuscitations, but excluded nature miracles (water to wine, walking on water) and provisioning miracles (loaves and fishes).

Unit of analysis: Each narrated healing episode counted once. Parallel accounts across different Gospels were collapsed into a single encounter. Vague summaries like “he healed many” were treated as one clinic session, not one bill per patient.

Diagnosis mapping: ICD-10 codes reflect best-fit modern syndromes based on described signs and symptoms, not claims about etiology. “Demon possession” is mapped to contemporary neurologic or psychiatric categories where appropriate.

• CPT selection: Office/outpatient E/M codes (99212–99215, 99204–99205) were used as time/complexity proxies, based on narrative detail and interaction length. Apparent deaths were mapped to critical care (99291) as the closest modern analogue.

• RVUs: 2026 CMS work RVUs were used. No procedures. No modifiers. No facility fees. No inflation adjustment for first-century Judea.

This is satire, but it is methodologically robust satire.

Results

Across 28 distinct narrated healing encounters (Table), Jesus totaled 68.27 wRVUs with a total estimated clinician time of ~16–23 hours.

What does this say about the value of medicine?

If even the most consequential healer in Western history barely exceeds 68 wRVUs, that may say more about the metric than the medicine. Three lessons follow. First, our system rewards billable inputs more reliably than meaningful outcomes. As long as the diagnoses and codes are equivalent, miraculous healing and snake oil salesmanship are not distinguishable in the logic of reimbursement. Second, much of the real work in these encounters—listening, judgment, reassurance, authority conveyed through presence—is precisely the kind of cognitive and relational effort that modern payment systems devalue relative to procedural labor. Third, our inability to measure what matters most is not a reason to overvalue what is merely easiest to count. The worth of clinicians is not captured by what can be coded and billed, but by what actually heals, restores dignity, and changes lives.

Post-Stack Peer Review Edit

The first two commenters raised important critiques. Commenter 1 appropriately questioned the use of “established” CPT codes. Commenter 2 astutely recognized that in real world practice, time-based billing would be used. To address these thoughtful critiques, I have now added two sensitivity analyses.

In the first sensitivity analysis, I assumed that every narrated healing encounter was for a new patient rather than an established one. This remapped office visit codes from 99212–99215 to their new-patient analogues 99202–99205, while leaving encounters already coded as 99204, 99205, or 99291 unchanged. Under that assumption, the total rises from 68.27 to 79.27 work RVUs, a 16.1% increase, and the implied total clinician time rises from roughly 16–23 hours to 21–29 hours. In other words, even granting Jesus a practice full of exclusively new patients only modestly improves the billing picture.

In the second sensitivity analysis, I assumed all encounters were billed strictly by time and that Jesus documented a reasonable high-end reimbursable time for each visit. In this sensitivity analysis, I assumed every healing encounter that could legally take a prolonged-time add-on received three additional add-on time codes. That means coding all qualifying office visits as either 99215 + G2212 x3 or 99205 + G2212 x3, and all resuscitation-like encounters as 99291 + 99292 x3. Under that assumption, the total rises from 68.27 to 104.99 work RVUs, with an implied 34–38 hours of documented clinician time depending on whether one uses the minimum or upper end of the qualifying time bands. Even in this deliberately generous scenario, the billing yield remains modest relative to the cultural and moral weight of the underlying acts.

The liver community was quick to embrace Resmetirom. In October 2024, the American Association for the Study of Liver Diseases (AASLD) issued expert practice guidance on its use. In 2025, a team of liver specialists (hepatologists) published their early experience of treating 110 patients.

But even as resmetirom entered practice and was warmly embraced by the hepatology community, a different story was already unfolding.

The figure below shows findings on the same surrogate endpoint of MASH resolution from three randomized trials for three different drugs.

Which drug would you choose before you even know the names?

Resmetirom: novel, but already obsolete

When the MAESTRO-NASH trial was published in 2024, it felt transformative. About 30% of patients experienced resolution of the active inflammation from MASH without worsening of liver scarring versus just 10% of the placebo group (Drug A in the figure above).

Furthermore, about 11% more participants in the resmetirom group showed less liver scarring by at least 1 stage on a 5-point scale (25% vs 14% of the placebo group).

Even without showing improvement in patient-centered outcomes, like symptoms from end-stage liver disease (cirrhosis), liver cancer (hepatocellular carcinoma), or living longer, FDA approval made sense. MASH had no therapies beyond the elusive lifestyle modification (aka diet & exercise).

Any credible benefit mattered. But even then, the writing was already on the wall.

Triple win for GLP1s, which wasn’t new

Weight loss is critical for preventing and treating MASH given that morbid obesity is the main culprit for the ensuing inflammation and scarring. Diet and exercise have long been the cornerstone of therapy. Unsurprisingly to anyone who has tried, or even contemplated a New Year’s resolution to become healthier, behavior change is hard, and meaningful weight loss is uncommon. Enter GLP-1 receptor agonists, or GLP1s for short.

Over a decade of research has convincingly shown enormous weight loss (pun intended) with GLP1s (including for the newer dual and now triple agonists). This is of course largely if people stay on the medication. Several smaller trials dating back to 2020 (see this NEJM trial) also showed compellingly high rates of resolution of MASH and signals for fibrosis improvement.

So when two large confirmatory trials arrived just after MAESTRO-NASH, they didn’t surprise anyone paying attention.

• SYNERGY-NASH in 2024: Tirzepatide (Drug C above) showed an absolute 53% greater MASH resolution vs placebo with fibrosis improvement in 20% more participants.

• ESSENCE published in 2025: Semaglutide (Drug B above) showed an absolute 30% improvement in MASH resolution vs placebo, with fibrosis improvement in 14% more participants.

These weren’t incremental gains. Resmetirom may have secured FDA approval first, but GLP1s won the game on the same surrogate outcome, by quite a large margin.

Outcompeting on surrogate endpoints isn’t even the most compelling advantage

Yes, improvement in inflammation and fibrosis are surrogate benefits which do not necessarily improve human flourishing. We’re still waiting for evidence of improvement in clinically meaningful liver outcomes that translates to living better and longer. This is why resmetirom received accelerated approval instead of full approval. But here’s what matters even more.

GLP1s not only outcompete on the same surrogate endpoint but they also substantially reduce weight, save lives, reduce heart attacks & strokes, treat type 2 diabetes, slow kidney disease progression, lessen arthritic pain, & treat sleep apnea.

Resmetirom, to a lesser extent, improves a liver biopsy endpoint and does none of the above.

In a disease defined by multisystem metabolic risk, choosing an inferior liver-specific drug over a more effective MASH drug with widespread cardiometabolic proven clinically meaningful outcome benefits requires extraordinary justification.

And GLP1s are cheaper

Resmetirom isn’t just less effective, it’s far more expensive.

The list price for Resmetirom is $4,000/month list price, with monthly copays of $1,000–1,400.

Yes, GLP-1s are still expensive, but a fraction of the cost, with copays often in the hundreds or less, and falling.

Bottom line

Resmetirom is inferior on histologic improvement in inflammation and fibrosis, inferior on outcomes that actually matter, and more expensive. Game, set, match.

This post unpacks my disconnect with that well-intentioned advice and makes the case for a different way of conferencing, based on my own experience (or “lived experience,” as the kids say, as if there’s another kind).

What Conferences Are Supposed to Do

Conferences are designed to accelerate scholarship and careers by cultivating conversation, feedback, collaborations, and knowledge. In the ideal version, a conference is a high-intensity environment for critique, community building and learning. You present early work to solicit constructive critiques before they harden into Reviewer #2 comments (IYKYK). More than anything else, the main purpose of conferencing is the conversation they stimulate, the relationships nurtured, and the career opportunities created.

The Reality of the Modern Academic Conference

The typical conference experience often falls short of that ideal. Content is siloed into a single specialty and often subdivided further into tracks that mirror existing disciplinary boundaries. Talks are constrained by time and format, and often do not permit methodological clarity or rigorous debate. The Q&A session is short, frequently deferential, performative, and shaped by hierarchy and social dynamics. Poster sessions include some truly excellent work, but also include too many overly ambitious, infeasible, or poorly conducted projects that will never travel beyond the hall. Even when the science is strong, apart from a select few podium presentations at major specialty conferences (hello cardiology), the reach is limited to the few audience members who are often your or your co-presenters’ mentors, advisors, collaborators, and friends.

The costs are not trivial. Registration, flights, hotels, and missed time add up quickly. So do the less visible costs, including childcare, recovery time, and the opportunity cost of being away from the work that actually moves ideas forward.

After nearly a decade of conferencing two to three times per year, when I ask myself, “Did this conference change my research or career?” the answer has too often been a resounding No.

Modern Conferencing Lives in the Public Square

If the purpose of conferencing is exchange, then the public square is where exchange happens at scale.

A curated social media experience is continuous. It is not limited to one annual meeting. It is not restricted to one society or discipline. It is not constrained by the same relational power dynamics. Ideas can breathe on their own and are judged less on the credentials of those espousing them. You can participate when you want to, follow the evolution of an idea over time, and engage in dialogue with people across the globe within your field as well as those outside who can offer fresh perspectives.

The public square is an incredible way to solicit feedback. Conference feedback is important, but it is limited due to constraints of time, format, social norms, and expertise in the room. It is unlikely to solicit honest input from the range of experts that are often needed, ranging from content, theory, methods, translation, and policy. In contrast, the public square offers the opportunity to elicit feedback across a wider spectrum of expertise. The signal-to-noise ratio depends on curation of who you follow and engage with, but when it works, it is far closer to robust pre- and post-publication peer review than casual commentary. It is specific and timely.

The economics matter too. The public square has dramatically lower costs. No travel. No institutional gatekeeping. That matters for equity, but it also matters for efficiency. If your goal for attending a conference is knowledge exchange, why pay the overhead of hotels and convention centers to do what can be done better with an internet connection?

The Public Square has Tangible, Meaningful Benefits

For me, the benefits are not abstract. I have found the public square indispensable for learning new methods, tracking emerging evidence, and picking up practical insights I would not have encountered in my usual lanes. The back-and-forth is often richer than conference Q&A because it is not constrained by the clock or by the social norms of not upsetting the wrong person. It also draws from a broader audience: subspecialists, methodologists, policy leaders, trainees, and members of the public. That mix is hard to replicate at conferences. Good ideas are not concentrated by geography, discipline, or even within academia.

The public square is where preliminary work in abstracts and preprints gets vetted, and where post-publication debate happens in real time. A recent example was the vigorous discussion on X of a recent flashy Nature Medicine study on time-of-day for immunotherapy. That conversation inspired my own AAOx2 post contributing to the critique which collectively led to an editorial investigation covered here (where I am quoted).

Another vivid example was a study our team published on the limited evidence base supporting the American Thoracic Society (ATS) clinical practice guidelines that effectively define the standard of care for lung disorders and many critical care illnesses. We had shared our findings by email in advance of publication, but the then-Chair of ATS still launched a diatribe on X (since deleted). Within 48 hours of publication, there was a substantive back and forth exchange that amounted to the equivalent of a letter to the editor and a letter in reply. On the usual publication timeline, that would have taken months!

The public square has also amplified my teaching. Through tweetorials, videos, and Substack posts, I interact with far more trainees and colleagues than I ever could at an in-person meeting. Instead of speaking to a room of ten or twenty, the same content can reach thousands, persist, and be revisited. Just as important, the public square flattens the usual gatekeeping. The activation energy is lower, especially for trainees at institutions without deep benches of scholarship and mentorship. This has led to back-and-forth exchanges, email messages, and invitations to trainee-run podcasts (Diagnostic Excellence on The Good Omen and Evidence-Based Medicine on Baseline Observations).

These interactions have produced tangible rewards. Through conferencing in the public square, I have secured two contracts, received two real job offers, and started several collaborations. They have also expanded my peer network. Several of these relationships have become real-life collaborators, mentors, mentees, and friends. That is the spirit of conferencing, achieved more consistently and with far greater reach than the typical meeting.

None of this requires pretending social media is perfect. It is not. It is noisy, performative, and yes, occasionally toxic. But those are design problems, not fatal flaws. With careful curation of who I follow, mute, and block, the public square can approximate what conferences are supposed to do, and in my experience, far exceeds them.

What We’re Really Letting Go Of

If we allow that the public square can accomplish or even exceed the goals of conferencing, then defending the primacy of the conference model becomes harder to justify. What often gets defended is not learning or critique, but the infrastructure of academic signaling. Conferences remain a reliable way to add to CVs and build the case for national visibility, promotion and leadership selection. The problem is treating the conferences as the sole, or even the best venue for discovery, feedback, networking, and opportunities.

The public square is not anti–human connection. Many things are better face-to-face, like relationships and collaborations. The public square is a vital first layer for discussion, critique, and dissemination, and developing connections that would have otherwise never been possible in the conference model that often requires social capital and networking skills to succeed. The relationships and ideas generated through conversation can then spiral into further offline conversations, remote collaborations, site visits, mentee-mentor relationships, contracts, and job offers. These are real, tangible, and meaningful benefits.

If the goal in academia is exchange, networking, and collaborations, then we should stop pretending the hotel ballroom is the only or even the best way to achieve that in modern times. Until then, the best conferencing will keep happening where the conversation is already taking place, in the public square.

If you had told me when I was a resident that I would one day become not only a hospitalist but in fact, the Chief of Hospital Medicine, I would have laughed. I would have also been insulted. I would have assumed you didn’t think I was smart enough, capable enough, or ambitious enough to do something ‘better.’

Back then, I thought hospitalists were the doctors who couldn’t cut it anywhere else: the ones who took the overflow patients from the ‘real’ medicine service, the teaching service. In the hierarchy of academic medicine, hospitalists seemed like an afterthought: a necessary patch to prevent resident work-hour violations, perhaps even occasionally helpful, but certainly not aspirational as a field. I saw them as physicians who had never quite moved beyond residency: still doing ward medicine, still managing admissions, but without the trajectory toward a clearly defined specialty. In my mind, they occupied a kind of in-between space: more responsibility than trainees, but not clearly different work. No one ever said this outright, but it was implied everywhere, from how residents directly compared their own workload and responsibilities to those of the hospitalists, and how prestige accrued to some roles but not others. In fact, late in residency when I told an admired attending physician I was considering a general internal medicine fellowship — then, the closest thing to a hospital medicine fellowships — he was quick to tell me that I would never be respected as an academic without subspecialty training, preferably in cardiology or critical care.

That reaction says as much about how medicine has historically viewed hospitalists as it does about my own assumptions. At the time, I didn’t question that hierarchy — I accepted it.

When the term ‘hospitalist’ was first introduced three decades ago, it described physicians defined primarily by where they practiced – the hospital – and by their potential to improve efficiency and coordination. What has been less clearly articulated since then is the distinctive clinical expertise that has emerged from that role.

Understanding that expertise – and claiming it – is essential to the future of the field.

In retrospect, perhaps it was not terribly surprising where I ended up. I trained in a residency environment defined by its intensity – high-acuity inpatient rotations, enormous clinical volume, and patients with layers of medical and social complexity. The more complicated the management or discharge plan, the greater the sense of accomplishment. Residents wore these cases like merit badges. We informally collected them: the most abnormal A1c, the lowest hemoglobin, the largest anion gap, the most improbable successful discharge. The program produced a disproportionate number of intensivists, which made sense. You either learned to tolerate that level of acuity, volume, and complexity, or you didn’t survive.

At the time, I didn’t recognize how profoundly that training was attuning me to the kind of clinical reasoning hospital medicine requires: comfort with unstable patients, rapid prioritization across multiple problems, decision-making before certainty. One of the first moments I recognized this explicitly came later, after residency, when I started precepting in primary care. I noticed something surprising: I felt most comfortable managing patients with long lists of conditions – COPD, obstructive sleep apnea, diabetes, pulmonary hypertension, heart failure – all interacting with each other. When someone presented with a single focused problem, like recurrent bacterial vaginosis, I felt adrift – disconnected from the complexity that had always oriented my thinking. It was the first time I understood that complexity itself could be a domain of expertise, and that it might define where I belonged in medicine.

Hospital medicine sits at the intersection of complexity, acuity, uncertainty, and volume. Hospitalists care for patients when illness is most unstable and least understood. Decisions often must be made before diagnoses are clear, before trajectories declare themselves, and before ideal information exists. We do this at high volume, across nearly every domain of adult medicine. While pulmonologists may see more rare lung disease and cardiologists may see more advanced heart failure, hospitalists see more acute pneumonia, more acute CHF exacerbations, more undifferentiated instability than anyone else. Volume builds a library of illness scripts, the cognitive database that allows experienced physicians to recognize danger early and intervene before deterioration becomes obvious.

But volume alone is not the expertise. The expertise lies in integrating acuity, multimorbidity, context, and uncertainty, deciding what matters most right now, what can wait, and what is safe for this particular person in front of you – in other words, applying evidence-based medicine to patients with multiple competing complex conditions and real-world constraints. Clinical guidelines are written for isolated diseases in controlled circumstances. They tell us optimal hemoglobin A1c targets. They do not tell us what to do when a patient has type 1 diabetes but no refrigerator to store insulin, inconsistent access to food, needle aversion, and limited health literacy. They do not resolve the tradeoffs when treating heart failure worsens kidney function or when blood pressure targets conflict with fall risk. Guidelines describe ideal physiology and circumstances. Hospitalists manage humans in the real world.

This work often becomes most visible after specialists have defined the boundaries of their domains. When a cardiologist evaluates chest pain and concludes it is non-cardiac, the problem is not solved. It has simply changed owners. Someone still has to determine what is causing the symptoms, how dangerous it might be, and what to do next. That someone is usually the hospitalist.

Specialists narrow the question within their domain: “Non-cardiac.” “Non-pulmonary.” “Non-neurologic in etiology.” “Correlate clinically.” Hospitalists reassemble the whole – determining what explains the patient in front of them and how best to move forward.

Years ago I encountered a framing that has stayed with me: generalists as comprehensivists and specialists as partialists. The intent was not hierarchy but orientation. Specialists focus deeply on particular physiologic domains. Generalists focus on the whole patient across systems. Modern medicine depends on both forms of expertise. Comprehensiveness, however, is cognitively demanding in ways that are less visible. Integration rarely looks as dramatic as depth. Yet that expertise is often flattened into something else: hospitalists are perceived less as experts and more as work horses – the solid, reliable physicians who keep hospitals running.

I have experienced this perception directly. More than once, I have been approached by leaders recruiting dual-physician couples with the assumption that the partner seeking a hospital medicine position would naturally find a place in my division. After all, don’t hospitalist programs always need more bodies? The question is rarely malicious. But it reveals an underlying assumption: hospital medicine as interchangeable staffing rather than specialized practice. No field matures academically when it is viewed primarily as labor and rather than respected as an area of expertise.

And if we are honest, we have reinforced that perception ourselves. We talk about census, flow, and staffing ratios more readily than we talk about diagnostic reasoning, integration, and judgment. We have defined our value operationally – and then been surprised when others do the same. That framing does not only obscure our actual expertise, it reinforces the perception that hospitalists do not possess a distinct expertise. That perception shows up most clearly in how hospitalists are compared to trainees. From the outside, resident teaching services and hospitalist services can look similar — teams rounding on wards, discussing plans, writing orders. But the difference between training and expertise is the difference between a cadet and a career officer. The environment overlaps. The role does not. Residents are learning foundational skills under supervision. Hospitalists are practicing independently, making decisions with compressed timelines and full accountability

To be clear, the daily reality of hospital medicine is rarely elegant. Census is high. Pages (or secure chats) are constant. Unremitting pressure for high turnover and improving flow is real. The work can feel grinding and relentless. But the cognitive labor does not disappear simply because it is embedded in volume. If anything, making sound decisions under those conditions requires even greater discipline, clarity and skill.

The problem is not the grind. It is that we have allowed the grind to define how we describe – and even understand – our work.

The intellectual foundation of hospital medicine is general medicine — a discipline often mistaken for the absence of specialization. In reality, it is a specialization in integration, prioritization, and uncertainty management. Hospital medicine is the discipline of general medicine practiced in its most acute setting. The expertise is already there. The question is whether we are willing to articulate it clearly.

Hospitalists are experts in comprehensive, multi-system acute care. That is not a consolidation prize. It is a distinct clinical domain.

The invisibility of that expertise is reinforced even with the name of the field, which doesn’t help. Cardiologists treat the heart. Neurologists treat the nervous system. Hospitalists, in contrast, sound like doctors who treat hospitals. The term describes location rather than knowledge. It obscures the reality that hospital medicine is not about buildings or throughput. It is about acute clinical decision-making across multiple organ systems under uncertainty.

But the field will struggle to grow – academically, intellectually, and culturally – until hospitalists themselves fully claim our own expertise. When we describe ourselves — or accept other’s descriptions of us primarily as coordinators, triage doctors or throughput managers, we unintentionally reinforce the idea that expertise lives elsewhere, in the subspecialties we consult, rather than within our own practice.

If my resident self could see my work now, the surprise would not be that I chose hospital medicine. It would be how profoundly I had misunderstood the field, how much expertise I once failed to see. Hospitalists are not the doctors who know less. We are the doctors who see more: more illness, more uncertainty, more complexity, more context – and we assume responsibility for distilling all of that into decisions that are evidence-based, contextually grounded, and right for the patient in front of us.

Doing that well is its own form of mastery. It is time we own it and practice like we believe it.

When Eric Topol lashed out at Peter Attia in a viral tweet, the broad claim—setting aside the personal material—was familiar: that longevity proponents overreach with unbridled enthusiasm, conflicts of interest, and a knack for persuasive storytelling over evidence. If you’re inclined to be skeptical of health optimization culture, it was catnip. A respected academic voice warning the public about a huckster.

But the episode is more revealing as a mirror than as a takedown. The interesting question isn’t whether Attia oversells. It’s why medicine keeps anointing a different kind of overreach—one dressed up in credentials, delivered in the language of “The Science,” and granted institutional authority that no longevity promoter could ever buy.

Wellness overreach is optional. Expert overreach distorts reality

Medicine loves to criticize the obvious hustlers: supplement sellers, wellness influencers, celebrity doctors turned brands. Their incentives are transparent. Their claims are transactional. You can opt in, opt out, and any downside is usually personal and contained.

The more dangerous failure mode is harder to see because it sounds responsible. It’s the academic elite expert whose authority outruns the evidence while retaining the moral weight of “settled science.”

Finance has a vocabulary for this. Warren Buffett waits for durable signal inside his circle of competence. He acts rarely, but decisively. Jim Cramer performs the opposite skill: high-velocity interpretation of noise, rewarded for clarity, conviction, and airtime. Then there’s the third character: the active fund manager selling a thesis to consumers—often persuasive, sometimes thoughtful, yet historically underperforms the market.

In medicine, Peter Attia is functionally that third figure: an active manager of health. He sells a worldview and a framework to a voluntary audience. You may think the evidence is thin. You may find the confidence excessive. You may question the sincerity of his motives. But the transaction is explicit.

Academic elites are supposed to be Buffett, but too often are Cramer

Not uncommonly, prominent academics operate like Cramer—rapid interpreters of incompletely vetted evidence, broadcasting conviction before certainty, while wearing the clothes of authority. That mismatch, Buffett’s prestige with Cramer’s tempo, is deeply problematic.

Eric Topol is an illustrative model of this archetype. Deeply credentialed, widely respected, routinely quoted by journalists and consulted by policymakers, Topol’s statements are rarely received as one clinician’s view. They’re received as credibility itself. Fittingly, he even names his brand accordingly—“Ground Truths.”

But look at how that credibility is manufactured. Much of it is not careful appraisal: expert reading that cuts through study design flaws, and the endless ways “statistically significant” becomes “true.” Much of it is rapid translation of what has cleared peer review, delivered with the tone of “this is what the science shows.”

The problem is not reading the literature. It’s what happens when reading becomes parroting, and parroting becomes narrative.

Because the literature—especially in hard-to-study topics like nutrition, lifestyle, and digital medicine—is often fragile. Observational studies by convenience. Unmeasured confounded by default. Key data measured poorly. Publications optimized for attention. Early signals shrink. Some are just flat out wrong. And for many others, we may truly never know the answer (is one drink of alcohol bad for you?).

Peer review in that ecosystem doesn’t certify truth. It certifies that a few people didn’t object.

And that is where danger lies: in the respectable shortcut that equates “published” with “reliable,” and “peer-reviewed” with “settled.” Treat publication as a stamp of approval instead of a starting point and you can do real damage amplifying a study’s errors at scale while appearing maximally responsible.

Topol provides an illustration. He was an early enthusiast for Theranos despite no public validation. He amplified long-COVID narratives implying common, severe multi-system injury compounds with repeat infections, claims that have proven untrue in the lived experience with rising life expectancy (a horribly misleading statistic) and labor participation. He routinely boosts observational lifestyle findings (coffee, exercise, nutrition) where healthy-user bias remains undefeated.

Most recently, with over 500,000 impressions on X, he celebrated a flashy Nature Medicine “time-of-immunotherapy” trial as “probably the best proof” of circadian biology despite glaring red flags that have since triggered immediate criticism (I’m quoted & see my post here) and editorial investigation.

This isn’t about any single miss. It’s the repeatable mechanism—take provisional papers, convert them into confident public narrative, and let prestige do the rest.

None of this requires fraud. None of it requires malice. It requires one epistemic habit: substituting peer-reviewed publication for independent appraisal, then broadcasting the result with scientific authority.

The wellness entrepreneur sells products. The academic elite sells beliefs

When a wellness proponent overreaches, the consumer bears the cost. When an academic elite overreaches, the system internalizes the distortion. It becomes how clinicians talk, how institutions message, how funders prioritize research, how journalists center truths, and how public digests science.

That’s the asymmetry.

This is why Topol-versus-Attia moment is revealing. Topol positions himself as the defender of evidence against hype. That critique of longevity enthusiasts is more than deserved. But the deeper issue is that the renowned academic who transmits deeply flawed studies with high conviction, is the one that turns questionable literature into authoritative narrative.

In finance, nobody mistakes Cramer’s velocity for Buffett’s discipline just because Cramer can cite earnings calls. Track record matters. Expertise matters. Humility under uncertainty matters.

Medicine needs that distinction badly. Not fewer experts—fewer Jim Cramers wearing academic robes, and more Buffets: people who wait for signal, resist narrative pressure, and speak carefully when the data are immature or wrong.

The most dangerous wolf doesn’t peddle supplements. They cite indiscriminately.

Medicine runs on an assumption so deeply embedded that we rarely notice it: we assume doctors are competent. Not perfect, not omniscient, but fundamentally capable. Safe enough to trust with our bodies and our lives. But how would you actually know if a doctor was competent? Good? Great? How can you tell the difference between a good doctor and a bad one? What does it even mean to be a ‘great doctor’?

Patients have very few signals to rely on. A diploma on the wall, board certification, a white coat, maybe a recommendation from a friend. Online ratings that often reflect parking logistics and wait times more than clinical skill.

Inside the profession, the signals are not much stronger. To maintain a medical license in the United States, physicians must complete continuing medical education credits and avoid serious disciplinary action. Board certification requires passing periodic multiple-choice exams. These are not meaningless standards, but they are also not particularly good measures of what most people intuitively mean by a ‘good doctor,’ raising a problem hiding in plain sight:

We do not have a clear, shared definition of what makes a doctor good.

The House of God, Samuel Shem’s 1978 darkly comic novel that generations of medical trainees eventually encounter, describes several familiar physician archetypes. The doctor with terrible bedside manner but dazzling clinical brilliance. The doctor beloved by patients but medically mediocre. But the trope that warmth and intelligence exist on opposite ends of a spectrum is mostly fiction. In real life, the physicians most admired by their peers tend to integrate technical skill, judgment, and human connection in ways that are difficult to disentangle.

Ask physicians privately who they would trust to care for their own family members, and there are some patterns that might emerge. Physicians often assume they can recognize excellence in their peers. But that recognition is shaped by context and perspective. We see one another in specific settings – on rounds, in conferences, in meetings – but not necessarily at the bedside when conversations are hard or trust is fragile. A physician admired by colleagues may be dismissive with patients. A doctor who projects confidence may not consistently exercise sound judgment. And a physician struggling with their own limitations may not reliably recognize excellence in others.

Recognition of what makes a good doctor exists – but it is imperfect and difficult to articulate.

In 1964, Supreme Court Justice Potter Stewart famously wrote about pornography:

‘I shall not today attempt further to define the kinds of material I understand to be embraced…But I know it when I see it.’

Defining clinical excellence in medicine is uncomfortably similar.

We may struggle to define it precisely but experienced physicians – and often patients – recognize clinical excellence when we see it: the doctor who makes patients feel heard and at ease, who takes a careful history that reveals what others missed, who can generate possibilities others wouldn’t consider when a problem doesn’t fit the usual patterns, who senses when something isn’t right despite reassuring tests, who explains uncertainty without eroding trust, who makes decisions that are both technically sound and emotionally attuned. But translating that recognition into measurable criteria has proven elusive.

Competence in medicine lives across multiple domains:

· Technical competence: Knowledge, diagnostic reasoning, procedural skill;

· Relational competence: Communication, empathy, the ability to understand what matters to a patient;

· Personalized, evidence-based decision-making: Applying current scientific knowledge to the specific person in front of you while aligning recommendations with their values, circumstances, and goals, under conditions of uncertainty.

We can test knowledge. We can sometimes observe procedures and outcomes. But relational skill and individualized clinical judgment – the parts of medicine patients experience most directly – are extraordinarily difficult to measure. And they are also where clinical excellence most often lives. Excellence is not just knowing the right answer. It is knowing what the right answer is for the specific person in front of you.

And yet, we have almost no systematic way to measure any of that.

When we can’t measure excellence directly, we substitute signals that feel reassuring: credentials, prestigious institutions, job titles, time in clinical care. I have seen this play out in my own career. I have spent much of my career trying to become an outstanding physician – not just competent, but excellent. The kind of doctor that I hope colleagues might trust with someone they love. At the same time, I built a research career. I study health systems, evidence, and how medical decisions actually play out in the real world. More than once, someone has casually assumed that because I am a researcher, I must not enjoy taking care of patients very much -- or that I am probably not very good at it. I remember the first time I heard those sentiments, as well as the offhand certainty of the assumptions. I felt genuinely stunned.

Over time, I came to realize that those reactions were not really about me specifically. They reflected the proxies we rely on to identify clinical excellence, visible signals that are easy to see but are not necessarily accurate. In this case, the proxy was the label ‘researcher’ itself, a visible identity that people used to infer something about clinical ability. The assumption that physician-researchers are somehow not good doctors reflects how easily proxies distort perception. It treats bedside excellence and analytical rigor as separate domains, when in reality they can often reinforce each other.

Even institutional structures sometimes reinforce the same fallacy. At my own university, physicians can be recognized as ‘master clinicians,’ but eligibility depends partly on the current proportion of one’s professional time spent in clinical care (and not necessarily cumulative experience or skill), a crude proxy for clinical excellence. Time spent doing something is not the same as doing it well. What makes this especially striking in my case is that my research training has cultivated skills relevant to clinical decision-making — clinical epidemiology, critical appraisal, and applying evidence to individual patients — skills that improved my ability to deliver personalized, evidence-based care at the bedside. Yet those competencies are largely invisible to the criteria being used. Under current criteria, I would not qualify – not because I don’t practice medicine, but because the metric being used prioritizes visible time over less visible expertise.

All of this points to a central problem: the qualities that make someone potentially a great doctor are often invisible and not measurable in the current systems we use to recognize excellence.

Why does any of this matter? Because modern health care increasingly depends on measurement.

We measure hospital quality. We measure outcomes. We measure productivity. We measure patient satisfaction. We measure costs. We compare physicians to algorithms. We debate whether artificial intelligence might outperform human clinicians. But if we cannot clearly articulate what makes a doctor good, we risk optimizing for proxies that miss the point.

We risk confusing board scores with sound judgment; efficiency with wisdom; friendliness with competence; confidence with accuracy. And, perhaps most importantly, we risk misunderstanding what kind of doctor we actually want.

This is especially visible in the safety-net, where I spend my clinical time. There is, appropriately, enormous focus on measuring disparities: measuring differences in outcomes across race, income, language, and access. Those measurements matter; you cannot improve upon what is unmeasured. But one of the most immediate levers for advancing equity is the quality of the individual care patients receive once they are in front of us – in order words, how good the doctor is – and that is something we measure far less well.

Medicine does track certain aspects of ‘quality.’ We measure adherence to guidelines, complication rates, readmissions, mortality, and patient satisfaction. These metrics are useful for understanding systems, but they are not the same as measuring the quality of how a doctor actually thinks, communicates, and practices. They tell us whether recommended steps happened, but they rarely tell us how well decisions were made – or whether the right decisions were made for each patient. They do not capture whether a physician recognized a subtle diagnosis early, avoided an unnecessary test, adapted treatment to a patient’s life circumstances, or noticed the one detail that changed everything.

In other words, we measure whether boxes were checked — not whether the doctoring was excellent.

That distinction matters enormously for closing disparities. Two patients can have identical insurance, identical diagnoses, and identical demographics, and still receive very different care depending on the judgment, attention, and decisions of the physician in front of them. The most immediate lever that physicians control to improve disparities is not policy or funding. It is the quality of the care we deliver. And yet that is the dimension we have the hardest time defining — and almost no reliable way to measure.

When physicians talk among themselves about colleagues they admire, the conversation rarely centers on exam scores or publication counts. Instead, it sounds more like this:

“She’s incredibly thoughtful.”
“He never misses things.”
“She’s so good with families.”
“He knows when to worry.”
“She’s calm when things get chaotic.”
“I’d trust her with my mother.”

These statements are less about credentials and more about the qualities that are the hardest to measure: judgment, attentiveness, and how someone shows up when decisions are uncertain. Not just what they know, but how they apply their knowledge and skills in real situations to individual people.

That’s what people are sensing when they recognize a great doctor.

And it raises a more personal question: if you needed to choose a doctor for someone you love, what would you look for?

Credentials help. Experience helps. Reputation helps. But much of what matters are qualities that are difficult to observe directly and nearly impossible to measure.

Medicine assumes competence. Patients depend on it. Physicians strive for it. Yet we still lack a clear way to define – let alone quantify – what makes someone excellent.

We know it when we see it.

In Part 1, I set the stage on why guidelines are not synonymous with evidence-based medicine because, (1) they don’t guide, and (2) the evidence supporting many recommendations is often poor.

In Part 2, Brad picks up the thread about the need for humility of uncertainty and proposes WikiGuidelines as a way forward.

In Part 1, I set the table by reviewing the two fundamental reasons why following guidelines is not necessarily evidence-based medicine.

Stay tuned for Part 2 by Brad on the humility of evidence and his efforts leading WikiGuidelines as a better way forward.

Subscribe now

This is a must watch on why Shorter is Better for antibiotics.

I sped it up to 1.5x speed since he doesn’t drink coffee.

Imagine a caffeinated Brad ...

When an effect that large emerges from something as mundane as clock time, the responsible posture is verification. Extraordinary effects demand extraordinary level of evidence. But when that extraordinary effect is paired with multiple red flags, the Time-of-Day Immunotherapy trial raises not just skepticism, but the possibility that the findings may be unreliable or worse, fabricated.

Until independently audited, this study should be retracted until verified.

The Effect Size Strains Credibility

The Nature Medicine immunotherapy trial reports hazard ratios for progression-free and overall survival of roughly 0.40 for patients receiving infusions before 3 pm compared with those treated later in the day. Median overall survival differs by more than eleven months.

Those hazard ratios live in the same neighborhood as the landmark trials that introduced immunotherapy into first-line treatment for metastatic NSCLC. In KEYNOTE-189, adding immunotherapy (pembrolizumab) to chemotherapy fundamentally altered survival curves. That was the effect of a drug.

Here, the claim is that timing alone produces a comparable magnitude of benefit.

Circadian biology is real. Immune function varies by time of day. But when a scheduling variable appears to generate a survival benefit on par with drug innovation, the prior probability that bias, artifact, imbalance, analytic flexibility, and yes, potentially more serious data integrity problems, explains the result is high.

Independent scrutiny, particularly by the Twitter/X account @houndcl, has surfaced multiple red flags that warrant careful examination. These concerns extend beyond interpretation and into the credibility of the data itself.

🚩 Zero Treatment Discontinuation Due to Adverse Events

The trial reported no adverse events leading to treatment discontinuation.

That is not a small discrepancy. It defies our expectations of immunotherapy across modern phase 3 trials.

In KEYNOTE-189 and KEYNOTE-407, roughly 14% of patients receiving chemo-immunotherapy discontinued treatment because of toxicity. In IMpower150, the discontinuation rate was 25%. Dual checkpoint blockade regimens such as CheckMate-227 report discontinuation rates closer to 18%. Across different PD-1/PD-L1 combinations, chemotherapy backbones, and global trial networks, double-digit toxicity-driven discontinuation is the norm.

Against that backdrop, observing zero treatment discontinuations due to adverse events in 210 advanced NSCLC patients would be an extreme statistical outlier, to put it lightly.

Conservatively assuming that the underlying discontinuation rate was 13.8%, as seen in KEYNOTE-189, the probability of observing zero events in 210 patients is one in a 37 trillion [(1–0.138)²¹⁰ ≈ 2.7 × 10⁻¹⁴].

Even if you assume this trial’s population was uniquely fit or adherent with therapy, and cut expected treatment discontinuation rates in half, the probability of seeing zero events in 210 patients remains about one in five million, statistically closer to impossibility than to a conceivable finding.

🚩 Zero Censoring in the First Twelve Months

A second concern is the absence of censoring during the first year of follow-up.

Trials are inherently messy. Patients transfer care. They miss scans. They withdraw consent. Sites close. Administrative data cutoffs occur. Even in tightly run randomized trials, early censoring is common.

To observe no censoring across an entire year in advanced NSCLC stretches plausibility. Survival datasets are not this clean. The absence of expected noise is not reassuring. It is anomalous.

When hazard ratios are dramatic, anomalous follow up is concerning.

🚩 Registry and Protocol Discrepancies

Public review of the cited ClinicalTrials.gov record reveals meaningful and unusual changes in how the study was described over time, including consequential changes to the study design itself.

On March 19, 2024, the registration was modified from an interventional randomized trial to an observational case-control study. Eleven days later, it was changed back to an interventional trial. Elsewhere, the described intervention shifted from immunotherapy to immunochemotherapy, and critical elements, including the definition of exposure timing and the primary outcomes, were revised.

Separately, the uploaded version of the Standard Analysis Plan dated January 2022 cited multiple publications from 2023 and 2024. That discrepancy does not invalidate the trial, but it does suggest either post hoc document assembly or inaccurate record keeping.

Registry updates are not inherently problematic. Trials evolve. Clerical errors occur. But, pre-registration is not bureaucratic theater. Major changes in study design, exposure definitions, and primary endpoints, particularly when occurring in combination, weaken confidence that the analytic plan was prespecified and locked before data were known.

When the reported treatment effect is this large, methodological stability is not optional. It is essential. If the design evolved in material ways, those changes must be transparent and reconciled with the reported findings.

🚩 Infusion Timing Inconsistent with Workflow or Random Assignment

The trial also did not show an expected distribution of infusion times across clinic hours. Instead, infusion times cluster heavily between 9-12:00pm, then resume between approximately 3-6 pm with a striking three-hour gap in between. There are essentially incredibly few infusions between 12:00-3pm.

That pattern is not what one expects from routine infusion workflow. In most oncology centers, infusion times are spread out more throughout operating hours. If the exposure cutoff is 3pm, one would expect infusions on both sides of the boundary, including patients at 2:50, 2:55, 3:05pm, and so forth. Instead, the figure depicts two structurally distinct clinic blocks separated by a nearly three-hour void.

More concerning, the distribution of infusion times in this single center trial, conducted between 2022-2024, appears to differ from infusions time distributions previously documented from the same exact hospital between 2018 and 2023. While workflows can evolve, a dramatic restructuring of infusion scheduling, especially one that produces such a clean mid-day gap, would be unusual and warrants explanation.

Additional forensic observations suggest that infusion timing for participants in the morning group appears tightly clustered across cycles, often within roughly an hour of prior infusions, whereas the afternoon group displays much more variation, with some patients following predictable schedules and others spread broadly from 3-8pm.

If corroborated, that asymmetry raises the possibility that the early and late groups reflect distinct operational sessions rather than random variation around a biologic threshold.

If early and late correspond to different clinic blocks, potentially involving different staff, workflows, patient flow, or scheduling constraints, then the exposure may in part reflect clinic operations rather than circadian immunobiology.

More worrisome, the wider variation in the afternoon group suggests the possibility of non-random assignment driven by patient availability, reflecting differences in work schedules, caregiving demands, transportation access, and other social and structural factors. If those factors differ systematically between morning and afternoon patients, then timing may simply be standing in for differences in adherence, social support, or other determinants of outcomes.

That distinction is not cosmetic. It fundamentally alters causal interpretation. If timing reflects operational differences or socioeconomic status rather than circadian biology, the observed survival difference may reflect confounding rather than chronotherapy.

The Responsible Interpretation

If this trial is valid, it is transformative. A no-cost scheduling change that doubles survival would reshape oncology practice overnight!

If it is not valid, it risks misleading clinicians, distorting research agendas (like the amyloid hypothesis for Alzheimer’s dementia), and eroding trust.

The solution is verification. Release de-identified patient-level time-to-event data with exact infusion timestamps and censoring indicators. Publish the locked statistical analysis plan and version history. Provide the code used to generate the hazard ratios and survival curves. Clarify definitions of discontinuation and censoring.

When the only knob you turn is the clock and survival doubles, extraordinary claims don’t earn applause, they earn scrutiny.