In the past two weeks I've published two blogs about stories that arose around blood type: the one piece of biology we all happen to know about ourselves. This got me thinking about why humans love stories so much, and led to this post. During the V-1 bombing campaign of 1944, Londoners became convinced the Germans were sparing certain neighborhoods. The bombs seemed to cluster, hitting some districts again and again while leaving others conspicuously untouched. The explanation people arrived at was that German sympathizers lived in the spared areas, feeding targeting information back across the Channel. Wealthy neighborhoods, poor neighborhoods, everyone had a theory about who was being protected and why. An actuary named R. D. Clarke ran the numbers. He divided south London into a grid of quarter-square-kilometer cells and counted how many bombs landed in each. If the strikes really were clustering, non-random, guided by something, the distribution should have looked lumpy: some cells hit repeatedly, others suspiciously empty. What he found instead was a close match to a Poisson distribution, the statistical signature of pure chance events scattered randomly across a fixed area. No clustering. No sparing. The bombs were falling exactly where randomness alone would put them, and the human brains living underneath that randomness were doing something brains cannot apparently help but do: finding a story in it anyway. That's the whole essay in miniature. Here's the slower version. The brain you have is a prediction engine, not a camera It's tempting to think of the brain as something like a recording device, taking in the world and producing an accurate internal copy of it. That's not what's happening. Your retina, skin, and cochlea are together delivering something on the order of eleven million bits of sensory information per second. Conscious awareness processes somewhere around forty to fifty. The gap between those two numbers isn't a rounding error; it's nearly the entire signal. Almost everything reaching your senses gets thrown away before you're ever aware of it. What survives isn't a copy of the world. It's a prediction, generated by an internal model that gets updated only when incoming data disagrees with it strongly enough to be worth correction. Consider your blind spot: there's a patch on each retina with no photoreceptors at all, roughly the width of your fist at arm's length, and you have never once seen a hole there, because the visual cortex fills the gap with a prediction drawn from the surrounding pattern rather than showing you nothing. Or consider phonemic restoration: erase a single sound from a recorded word, replace it with a burst of noise, and play it back. Listeners report hearing the missing sound clearly, correctly, with no sense that anything was ever removed. In both cases the brain isn't a passive receiver waiting for complete information. It's actively manufacturing the missing piece and presenting the manufactured version to you as though it were simply what's there. Researchers working in predictive processing, Karl Friston, Andy Clark, and Jakob Hohwy among the most influential, have spent the last couple of decades formalizing this: perception is mostly inference, corrected only at the margins by the world. The reason for this architecture is not mysterious. A brain that waited to react until it had fully perceived a threat would be a brain that reacted too late. Prediction is faster than perception. So evolution built brains that guess first and check the guess against reality only as needed, because the guessing is what keeps the animal alive long enough to guess again. Pattern-detection with no off switch Here's where it gets interesting. The machinery built to predict predators, weather, and the location of food doesn't come with a governor that shuts it off once survival needs are met. It's a general-purpose pattern detector, tuned by evolution for a specific set of adaptive problems, but with no way of distinguishing “a pattern that matters for staying alive” from “a pattern,” full stop. Once you have a system that scans constantly for structure and rewards you, neurochemically, for finding it, that system does not stay confined to its original job description. It runs on everything you point it at, including things with zero survival stakes. You can watch this same circuitry misfire on something with nothing at stake at all. Sit at a roulette table and watch black come up five times running, and something in you will insist red is “due.” The wheel has no memory. Each spin is independent of the last, governed by exactly the kind of process that produces Poisson-style clumping over the long run, the same statistical signature Clarke found in his bomb data. But a brain built to assume that streaks mean something, a predator is near, a food source is reliable, does not easily accept that five blacks in a row are simply what randomness looks like some of the time. A pattern-hungry brain looks at a clump, at a card table or anywhere else, and assigns it a cause, because assigning causes is the job, and the job doesn't come with a built-in check for whether a cause is actually there. Run that same overshoot through a slightly different circuit, one built for modeling other minds rather than physical events, and you get agency-detection: the reflex to assume a “who” behind an occurrence rather than a “what.” It's a useful reflex when the rustle in the grass really is a predator. It's a much less useful reflex when it's applied to a losing streak at a card table, a rustle that's just wind, or a universe that doesn't owe you a reason. But the circuit fires the same way regardless, because the circuit was never built to check whether firing was warranted. It was built to fire fast. What happens when the machinery overshoots Most of the time this overshoot is just noise, a superstition here, a false pattern there. But four of the more notable things humans do look a lot like this same machinery, run deliberately and at scale, on inputs where the stakes were never survival to begin with. Art is expectation, set up and then either satisfied or productively violated. A melody creates a prediction about the next note; the pleasure of hearing it resolve, or the deeper pleasure of hearing it almost resolve and then swerve, is the prediction engine registering its own model being tested. This is true even of art that refuses to resolve at all. A plotless novel, an atonal piece of music, a painting with no clear focal point, these aren't failures of pattern-making. They're a different kind of pattern, one where the recognizable shape is the absence of a shape, and the brain does the same sense-making work either way. It just makes the unresolved feeling the point. Comedy runs on the same clock, just tuned differently. A setup builds a prediction about where a sentence, or a situation, is headed. A punchline breaks that prediction and replaces it with a second interpretation that also, on half a second's reflection, makes sense of everything that came before. The laugh is what a violated-but-resolved prediction feels like from the inside, which is why the same joke told too slowly falls flat: the first model gets too much time to settle in, and the swerve reads as merely wrong rather than delightfully wrong, and why a joke told too fast doesn't land either, since there was never enough of a prediction built to violate in the first place. Timing isn't an aesthetic flourish sitting on top of comedy. Timing is the mechanism, precisely dosed. Conspiracy theories are the same causal hunger, unsupervised. This is, in miniature, exactly what happened over London: a real pattern of scattered, random events, and a population of brains reflexively assigning it a hidden intentional cause rather than accepting chance. Conspiracy theories run on that same logic at any scale, a cluster of events that feels too coordinated to be accidental, an agent posited to explain the coordination, a community that reinforces the pattern by trading confirming details back and forth until the story feels load-bearing. Agency-detection, the reflex to assume a “who” behind an occurrence, does a lot of the work here, and it costs the system almost nothing to overuse: assuming agency where there is none is cheap, while failing to notice agency where it actually exists can be fatal, so evolution built us to err generously toward “someone did this.” The reasoning isn't stupid. It's the ordinary causal-inference machinery, running exactly as designed, with no external check on whether the inferred cause is real. There's no separate gullible circuit sitting apart from the sensible one. There's one circuit, and it doesn't come pre-labeled with which of its outputs to trust. Science is the same hunting instinct, except with a leash on it. A hypothesis is a prediction like any other, generated by the same overactive pattern-finder that saw spies in a bombing grid. What makes science different isn't that scientists have quieter versions of this instinct. It's that the discipline built an external check the brain doesn't supply on its own: test the prediction against reality, and be willing to let reality win. Clarke's own paper is an act of this discipline turned back on the very phenomenon that produced it, using statistics to catch the brain in the act of pattern-hunting and correct it. Science is, in that sense, the prediction engine's most successful attempt to police itself. The part that doesn't resolve Here's the uncomfortable symmetry. The same system that produces a Beethoven symphony produces a conspiracy theory. The same system that let Clarke catch the brain in the act of pattern-hunting was doing the pattern-hunting in the first place. A cluster appears, therefore something must be causing the cluster, therefore let's find out what, or who. Sometimes that search turns up something real: a virus's mode of transmission, the actual cause of a bridge collapse, a genuine regularity in orbital motion. Sometimes it turns up a spy network that was never there. Most of the time, this machinery has no internal way of telling you in advance which search you're running. I don't think there's a tidy place to land on this. The pattern-detection that makes you capable of art, of asking unanswerable questions, of doing science at all, is the exact same machinery that will hand you a false spy network with total conviction if you don't build something external to check it against. We didn't get the beautiful, meaning-saturated version of human cognition and the paranoid, error-prone version as two separate features. We got one system, running exactly as designed, and it was never going to be able to tell the two apart on its own.

In Japan, it is entirely normal to be asked your blood type on a first date. Dating profiles list it next to height and occupation. Some employers have, at various points, sorted job candidates by it, on the theory that Type O makes a better leader and Type A makes a better accountant. There is a specific term for this: ketsuekigata, blood type personality theory. It has no biological mechanism, no supporting evidence at scale, and roughly fifty years of cultural staying power [1]. Where it started The theory traces to 1927, when Japanese psychologist Takeji Furukawa published a paper linking ABO type to temperament, describing Type A individuals as gentle and anxious, Type B as cheerful and self-centered, and so on [2,3]. Furukawa's data were thin, and his own academic community pushed back at the time [3]. The theory should have died there. It got a second wind in 1971, when journalist Masahiko Nomi, who had no medical or psychological training, published a book reviving and expanding it, adding claims that blood type predicted not just personality but disease susceptibility [1,3]. Nomi's work drew the same methodological criticism Furukawa's had. It didn't matter. By the 1980s, blood type personality theory was a fixture of Japanese popular culture, and it has stayed one since, with related belief systems taking hold in South Korea and Taiwan [4]. What the evidence shows Unlike the blood type diet, this one doesn't even offer a mechanism worth debunking. Nobody claims to know how an antigen expressed on the surface of a red blood cell would shape whether you're anxious or gregarious. There's no lectin story here, no agglutination metaphor. It's association without a proposed pathway, which makes it a cleaner test case for pattern-seeking on its own terms. And the pattern isn't there. Large-scale surveys using validated personality instruments, run in both Japan and the United States, have found no meaningful correlation between ABO type and standard personality traits [1]. This isn't a case of a single small study failing to find an effect. It's the specific finding, repeated across populations that supposedly believe in the theory most strongly, that belief and reality have parted ways. Why it took hold where it did The interesting question isn't whether ketsuekigata is true. It's why a country with excellent scientific infrastructure and health literacy adopted a folk taxonomy this thoroughly, while most of the world shrugged. Part of the answer is historical: Furukawa's early work was entangled with period ideas about national and ethnic character, giving the theory an early institutional foothold it didn't earn on data [3]. Part of it is more mundane. Once a categorization system is culturally legible, media reinforces it, and confirmation bias does the rest. If you're told Type B people are unpredictable, you notice the unpredictable things your Type B friend does and forget the rest. The Barnum effect, the same mechanism that makes horoscopes feel accurate, doesn't need a real correlation to feel like one. The diet's cousin If you read last week's post on the blood type diet, you'll notice the throughline: ABO type is one of the only pieces of biology most people can name about themselves, and that legibility makes it an appealing anchor for a story, regardless of whether the story holds up. The diet claims a mechanism and fails on evidence. The personality theory doesn't even bother with a mechanism, and fails just as completely. Both persist anyway. Maybe that's the actual finding here: mechanism was never the load-bearing part of either belief. The load-bearing part is wanting a story. References 1. Nawata K. No relationship between blood type and personality: evidence from large-scale surveys in Japan and the US. Jpn Psychol Res. 2014;56(2):163-173. 2. Furukawa T. A study of temperament and blood-groups. Psychol Res. 1927;2:612-634. 3. Cha J. Blood type and personality. Hektoen International. 2020. 4. Relationship between ABO blood type and personality in a large-scale survey. Int J Psychol Behav Sci. 2021;11(1).

A friend at a dinner party once turned down the bread basket because she was "type A, and grains don't agree with her blood." I nodded, passed the butter, and spent the rest of the meal thinking about lectins instead of listening to the conversation. Occupational hazard. The blood type diet has been around since 1996, when naturopath Peter D'Adamo published Eat Right 4 Your Type [1]. Its premise: your ABO blood type reflects your ancestors' diet, so eating in accordance with your type optimizes digestion and health. Type O, the "hunter," should eat like a caveman: high protein, minimal grains. Type A, the "agrarian," should go largely vegetarian. Type B gets dairy. Type AB gets a little of everything, marketed as the "enigma" for reasons that say more about branding than biology. The mechanism, as advertised The theory rests on lectins: proteins found throughout the plant kingdom, concentrated in legumes and grains, that bind to carbohydrate molecules. D'Adamo's claim is that dietary lectins interact with the antigens on your red blood cells the way incompatible blood interacts in a transfusion reaction, causing agglutination, clumping, and a cascade of downstream health problems specific to your type [2]. If you work in transfusion medicine, you already know where this goes wrong. What agglutination actually requires Agglutination in a clinically meaningful sense requires antibody binding to a compatible antigen, at a concentration and in an environment that supports crosslinking. It is not a generic hazard that any given protein produces on contact with blood. The lectin literature bears this out: a systematic review of blood type diet claims found no controlled evidence that dietary lectins clump blood cells inside the body in a type-specific way [3]. Most food lectins denature with cooking. Of the ones that survive digestion intact, the review found the overwhelming majority react with all ABO types indiscriminately, not selectively with one [3]. There is a small, real exception: raw legumes contain lectins with some blood-type-preferential agglutinating activity in vitro [4]. But "some lectin behaves selectively in a test tube" is a long way from "your dinner is attacking your blood type." What the trials actually show A few groups have tested the diet's health claims directly rather than the mechanism. The most-cited systematic review, Cusack et al. in the American Journal of Clinical Nutrition, found that while several of the individual "type" diets were associated with favorable changes in metabolic risk markers, the effect held regardless of the subject's actual blood type [3]. In other words: the Type O diet made people healthier whether or not they were Type O. The diets are, structurally, healthier eating patterns. Less processed food, more vegetables, controlled portions. People who follow any of them tend to do better than people who don't, and the ABO matching adds nothing measurable on top of that [3]. A separate analysis, Wang et al. in PLoS ONE, looked directly at ABO genotype against the diet's proposed cardiometabolic outcomes and found no interaction: type didn't predict which diet worked better for whom [5]. Why it persists anyway None of this has hurt sales. The obvious answer is that personalization sells better than generic advice, and blood type is one of the few pieces of biology most people can actually name about themselves. "Eat less processed food" is true and unsatisfying. "Eat less processed food because you are Type A" is a story, and stories are what people remember at dinner parties. There's a less cynical read too. We want our biology to explain us: why we feel good after some meals and sluggish after others, why one relative thrives on a diet that wrecks another. ABO type is legible in a way gut microbiome composition or polygenic metabolic risk isn't. It's one letter, drawn on a card in your wallet, and it feels like it should mean more than it does. It doesn't. Eat the vegetables. Skip the antigen mapping. References 1. D'Adamo P, Whitney C. Eat Right 4 Your Type. G.P. Putnam's Sons; 1996. 2. Nachbar MS, Oppenheim JD. Lectins in the United States diet: a survey of lectins in commonly consumed foods and a review of the literature. Am J Clin Nutr. 1980;33(11):2338-2345. 3. Cusack L, De Buck E, Compernolle V, Vandekerckhove P. Blood type diets lack supporting evidence: a systematic review. Am J Clin Nutr. 2013;98(1):99-104. 4. Sharon N, Lis H. History of lectins: from hemagglutinins to biological recognition molecules. Glycobiology. 2004;14(11):53R-62R. 5. Wang J, García-Bailo B, Nielsen DE, El-Sohemy A. ABO genotype, 'blood-type' diet and cardiometabolic risk factors. PLoS ONE. 2014;9(1):e84749.

The patient is a composite, built from enough cases to be nobody in particular: an elderly man admitted with a colonic obstruction, febrile, blood cultures eventually growing a gram-negative rod. His chart says A positive. It has said A positive for decades, through two prior admissions and one prior transfusion. The type-and-screen that comes back this time says otherwise. The forward type shows both A and a weak but unmistakable B reactivity. The reverse type, his own serum against reagent cells, still shows anti-B, exactly as it should for a lifelong group A patient. Forward and reverse disagree. Someone on the bench who has seen this before doesn't reach for a rare subgroup workup. They reach for the chart, note the fever and the bowel pathology, and write "acquired B" before the second tube has finished spinning. He is still, genetically, group A. Nothing about his ABO gene has changed. What's changed is what a bacterial enzyme has done to the surface of his red cells, and that distinction, between what your genes say and what your cells are currently showing, is the whole subject of this post. Genotype and phenotype, briefly Your ABO genotype is fixed at conception: the specific alleles you inherited, encoding glycosyltransferase enzymes that add either N-acetylgalactosamine (making A antigen) or galactose (making B antigen) onto a common precursor structure on the red cell surface. Your ABO phenotype is what a lab actually detects: the antigens expressed on your cells, tested by forward typing, and the antibodies circulating in your plasma against the antigens you lack, tested by reverse typing. In the overwhelming majority of people, genotype and phenotype agree completely, which is why we treat "blood type" as a fixed, inherited fact roughly on par with eye color. Most of the time it is. It's just not guaranteed to be, and the exceptions are where transfusion medicine gets interesting. How the mismatch happens Acquired B is the cleanest example because the mechanism is fully worked out. Certain gram-negative bacteria, often colonic flora that have gained access through a compromised gut wall from a tumor, obstruction, or infection, carry a deacetylase enzyme. That enzyme strips the acetyl group off the terminal sugar of the A antigen, N-acetylgalactosamine, converting it to galactosamine [1,2]. Galactosamine happens to be structurally close enough to galactose, the terminal sugar that defines the B antigen, that commercial anti-B reagent cross-reacts with it [1]. The cell hasn't started making B antigen. It's wearing a chemically modified version of its own A antigen that a reagent mistakes for B. Pull the source of bacterial exposure, whether by treating the sepsis or resecting the tumor, and the discrepancy resolves on its own, typically within weeks [3]. Genotype never moved. Phenotype took a temporary detour. Post-transplant chimerism works differently but lands in the same place. A patient who receives an ABO-mismatched hematopoietic stem cell transplant will, over the course of engraftment, gradually stop typing as their native blood type and start typing as their donor's. Eventually the genotype itself has changed, at least in the hematopoietic compartment: the red cells being produced are now genetically the donor's. This one isn't a transient artifact to be explained away. It's a real, durable shift, and it means "blood type" for a post-transplant patient has to be tracked as a moving target rather than looked up once and filed away [4]. A third category lives entirely on the genotype side: weak ABO subgroups, where an inherited variant allele produces a glycosyltransferase with reduced enzymatic activity. The antigen is present, just sparse enough that forward typing looks weak, ambiguous, or occasionally falls out as an apparent O in a person who is not, in fact, group O. No bacteria involved, no transplant involved. Just an inherited enzyme that's technically functional and practically underpowered. Why it matters beyond the interesting case report Every one of these scenarios is a reminder that "check the type" and "trust the historical type" are not the same instruction, and blood banks build entire policies around knowing when to prefer one over the other. A patient with a documented history of acquired B doesn't need a lifelong flag reclassifying their type; they need the discrepancy recognized as transient and resolved on the next clean sample. A post-transplant patient needs active tracking through engraftment, because giving blood matched to their pre-transplant genotype can become the wrong call partway through their course. A weak subgroup needs to be distinguished from acquired B and from genuine group O, because the transfusion consequences of getting that wrong are not symmetric. None of this is exotic. It's the ordinary, unglamorous discipline of not assuming a chart from three years ago is still telling you the truth. Where sequencing fits Serology answers what the cells are doing right now. It doesn't always answer why, and in ambiguous cases, that gap matters. Genotyping resolves the categories serology can't cleanly separate: distinguishing a weak subgroup from an acquired phenomenon from early mixed-field chimerism, in a single pass, without waiting for an infection to clear or a repeat sample to confirm a trend. That's the appeal of ABO genotyping as a clinical tool, and it's the part of this space I've been spending the most time in lately. More on that as the work develops. The identity question People treat blood type the way they treat a birthday: fixed, inherited, a fact about you that predates memory. Mostly that's fair. But "mostly" is doing real work in that sentence. Somewhere in a transfusion service right now, a phenotype is quietly disagreeing with a genotype, and the discrepancy is not a lab error to be explained away so much as a reminder that even the facts we consider most fixed about our own biology are, on some timescale, conditional. References 1. Blood Bank Guy Glossary. Acquired B Antigen. bbguy.org. 2. Judd WJ, Friedman BA. The acquired B antigen phenomenon. ASCP Check Sample Program, Immunohematology No. 1-82; 1975. 3. Campbell TA et al. Acquired B antigen: an ABO typing discrepancy successfully reversed by transfusion with type A red blood cells. Transfusion. 1980;20(3):345-348. 4. Resolution of an unexpected ABO typing discrepancy in a 9-month-old patient with juvenile myelomonocytic leukemia. Clin Case Rep. 2020.

Families who request directed donation from an “unvaccinated” donor believe they are choosing the safer option. The literature says the opposite. Directed donations, particularly from first time donors recruited specifically for the occasion, carry higher rates of infectious disease marker reactivity than units from repeat community donors. The donor pool people trust the most, because they know the person, is measurably less safe than the anonymous pool they're trying to avoid. That paradox sat quietly in the guidelines for years, cited in position statements and discouraged in policy language, without much data on what actually happens when a request like this proceeds anyway. A two year single center series out of Vanderbilt, published this year in Transfusion, gives us that data. It's worth sitting with. What They Found Between January 2024 and December 2025, the VUMC blood bank received 144,856 total blood product units. Of those, 48, or 0.03 percent, were directed donor units collected specifically because a patient or family refused standard inventory over concerns about vaccinated donors. Every single directed donation in the study period, with no exceptions for rare blood types or other medically recognized indications, was motivated by this concern. Those 48 units covered 15 patients. Median age was 17, ranging from 4 months to 73 years, and 60 percent were pediatric. The requests weren't rare and holding steady either. They climbed from 4 patients in 2024 to 11 in 2025. Thirteen of the 15 patients were transfused at least one directed unit. And 7 of the 15, nearly half, had at least one unit collected on their behalf that was never actually transfused to them. Those units didn't vanish. Most were released back into general inventory. Someone drew blood from a specific person, for a specific patient, under a specific belief about safety, and then that blood went to a stranger anyway. That's not a rare edge case in this series. It's closer to the norm. Where the System Broke Here is the finding that matters most to me as a transfusion medicine physician: of the 15 cases, only 1 had a documented ethics consultation. Only 1 had transfusion medicine notified before the directed unit arrived at the blood bank, and even that notification came after the perioperative service had already approved the request going forward. This isn't a story about families making an uninformed choice in a vacuum. It's a story about a workflow. Requests here were routed directly from a family or a primary clinical team to an external blood donor center's online form, entirely outside the consultation structure that exists precisely to handle requests this ethically and medically complicated. By the time anyone with transfusion medicine expertise heard about it, the decision had usually already been made. It didn't have to go this way. Mayo Clinic's Bloodless Medicine and Surgery Program uses structured shared decision making for exactly this kind of request, and most families end up accepting standard blood products through that process. Mayo has since restricted directed donation absent a genuine medical indication. Seattle Children's Hospital built a similar structured consultation model, pairing transfusion medicine with ethics, for pediatric cardiac surgery cases with vaccine related concerns, and it worked there too. The difference between those institutions and this series isn't the families. It's whether anyone with the right expertise was in the room before the blood was drawn. Harm in Both Directions The clinical consequences in this series are not abstract. One patient's hemoglobin fell to 5.9 g/dL with symptomatic anemia while transfusion was delayed awaiting arrival of directed units. That same patient later received a transfusion at a hemoglobin of 9.2, a clear deviation from institutional guidelines, because the clinical team didn't want the directed unit to go to waste after all that effort to obtain it. Sit with that sequence for a second. A patient was harmed by the delay. Then a guideline appropriate threshold was overridden to avoid wasting a unit that should never have been the deciding factor in the first place. The instinct to avoid waste, once the unit exists, quietly overrides the standard of care that exists to protect the patient. A second patient developed hemodynamic shock with a hemoglobin nadir of 3.6 while awaiting directed blood. Two additional patients had surgery delayed or cancelled entirely because of directed component logistics. Four of fifteen patients, more than a quarter of this small cohort, experienced a documented adverse clinical or operational event tied directly to this workflow. Outcome Patients Received at least one directed unit 15 Transfused at least one directed unit 13 (87%) Had at least one unit collected but never transfused to them 7 (47%) Clinical deterioration while awaiting directed units 2 (13%) Transfusion deviating from institutional guidelines 1 (7%) Surgical delay or cancellation 2 (13%) Ethics consultation documented 1 (7%) Transfusion medicine notified before unit arrival 1 (7%) The Weight Falls on Children Nine of the fifteen patients in this series were minors. Among pediatric patients, surrogate decision making applied in 100 percent of cases, compared to 17 percent of adult cases. These are children absorbing the downstream consequences, clinical and logistical, of a belief about vaccination status that they had no part in forming and no ability to consent around. The ethical weight of that imbalance is hard to overstate, and it's the piece of this paper I keep returning to. A Familiar Failure Mode I've written before about laboratory medicine as a kind of governance layer, the expert checkpoint that's supposed to sit between a high stakes request and its execution, whether that request involves an unvalidated algorithm or a unit of blood. This series is that same failure mode wearing a different face. An ethically loaded, medically consequential request bypassed the expert consultation layer almost entirely, not because the layer didn't exist, but because the workflow routed around it. Governance failures rarely look like a single bad decision. They look like a form that lets you skip the conversation. Where This Leaves Us The authors propose a fix that sounds almost too simple: mandatory transfusion medicine consultation for every directed donation request, before collection proceeds, replacing what is currently an optional and easily bypassed step. Mayo and Seattle Children's suggest that when this consultation happens, most families accept standard products anyway. That's an encouraging signal, but it doesn't fully resolve the harder tension underneath this paper, the one between respecting a family's autonomy to make decisions about their own care and preventing exactly the kind of harm this series documents. A mandatory consult can close the routing gap. It can't, by itself, tell us how to balance those two obligations when a family still says no after hearing everything transfusion medicine has to say.

I knew Wisconsin had hard water before we moved here — everyone warns you about the white crust that builds up on faucets, the way soap won't quite lather right, the kettle that needs descaling every few weeks whether you want to deal with it or not. What I didn't expect was how much I'd find myself thinking about it on a quiet Saturday afternoon, half-bored, half-curious, turning over a question I hadn't really asked before: does any of this actually do anything to me? Hard water, by definition, is just water carrying more dissolved calcium and magnesium than usual — picked up as it filters through limestone and rock on its way to your tap. And calcium is, of course, the mineral I spend a fair amount of my professional life thinking about, just usually in the context of a patient's blood draw rather than my own kitchen sink. So the question followed naturally: if I'm drinking measurably more calcium every day than I was in my old soft-water house, is more of it ending up in my blood? The Body Doesn't Work That Way The honest, slightly anticlimactic answer is no — and the reason is one of the more elegant pieces of human physiology. Serum calcium isn't a passive reflection of however much calcium you happen to eat or drink on a given day. It's tightly, almost stubbornly regulated by a feedback loop involving parathyroid hormone, vitamin D, and the kidneys, all working continuously to hold your blood calcium inside a narrow range regardless of what's coming in from your diet. Drink more, and your body simply absorbs and excretes accordingly, defending the same set point it always defends. This isn't just theoretical. A Swedish study comparing people living in hard-water and soft-water regions looked directly for a correlation between calcium and magnesium levels in drinking water and the corresponding levels in serum — and found none. The water mineral content moved. The blood mineral content didn't follow. Homeostasis, doing exactly what it's supposed to do. But "no change in serum calcium" doesn't mean nothing happens at all — it just means the action shifts somewhere else. A small trial gave healthy young men a single glass of calcium-rich mineral water and tracked their bloodwork over the next few hours. Serum calcium, predictably, barely budged. But parathyroid hormone dropped significantly, and a marker of bone resorption fell right along with it. The body didn't need to raise blood calcium because it had just gotten a supply from the gut — so it dialed back how much it was pulling out of bone to maintain the same number. One glass of water, one afternoon, and you can already see the lever moving. So if hard water isn't raising anyone's blood calcium, the question becomes: does it matter for anything else? Where the Real Signal Shows Up It turns out the more interesting story isn't in serum calcium at all — it's in bone, and specifically in fracture risk, measured over years rather than in a single blood draw. A 2026 ecological study out of England found that neighborhoods with harder water had meaningfully fewer hospitalizations for childhood fractures than neighborhoods with soft water, even after accounting for the usual sociodemographic confounders. Norwegian researchers, looking at hip fractures in older adults, found a related pattern: lower calcium and magnesium in municipal water tracked with lower bone mass density and higher fracture incidence. That's a real, if quiet, signal — and it makes some biological sense. If chronic, low-grade mineral intake nudges PTH down just slightly over years and decades, the downstream effect is less ongoing resorption from bone to maintain serum levels. Nobody's blood calcium moves. The skeleton just quietly gets to keep slightly more of what it already has. But It's Never Just Calcium Here's where the story gets messier, in the way real epidemiology usually does. Hard water isn't a single-ingredient exposure — it's calcium and magnesium, almost always together, in ratios that vary by region and geology. The same Norwegian research group that found the calcium-fracture association also found that it depended heavily on what else was in the water; once they accounted for other minerals, the calcium signal alone wasn't clean or independent. You can't pull calcium out of "hard water" and credit it alone for what the studies are finding. Whatever benefit exists is probably shared, tangled, and not easily assigned to one element over the other. Which raises a natural next question: if it's not clearly calcium doing the work, what happens when you study calcium supplementation directly? The Calcium Supplementation Letdown Not much, as it turns out — at least not in the way most people assume. Systematic reviews of calcium supplementation in adults have found, at best, modest gains in bone mineral density: low single-digit percentage improvements at the hip, spine, and forearm, even at fairly substantial daily doses. One major review concluded there was no clear evidence that increasing dietary calcium intake actually lowers fracture risk at all. The mineral that seemed like the obvious hero of bone health, taken as a standalone pill in adulthood, just doesn't move the needle the way the supplement aisle would have you believe. That's a genuinely useful correction to a very widespread assumption — but it also means the hard-water fracture data probably isn't a calcium story either. Magnesium's More Promising, Less Finished Story So I went looking at magnesium on its own, and found a more encouraging — if still unfinished — picture. A 2021 systematic review found that higher magnesium intake was associated with increased bone mineral density specifically at the hip and femoral neck. Small clinical trials back up a plausible mechanism: magnesium supplementation in postmenopausal women has been shown to lower parathyroid hormone, raise markers of bone formation, and lower markers of bone resorption — the same hormonal lever that chronic mineral exposure from water seems to be quietly pulling. What's missing is the big, definitive trial: a large, long-term, fracture-outcome study in older adults that could say, with real confidence, "magnesium supplementation prevents fractures." That trial doesn't exist yet. What exists is a coherent mechanism and some encouraging early biomarker data — promising, but not proven. Two Different Stories, One Word So "hard water" turns out to be two separate stories wearing the same name. The calcium half looks like it matters most early — in childhood and adolescence, when bone is actively being built and every bit of mineral exposure has more to work with. That window is the one with the clearest evidence, and it's also the one I've already missed; I moved to Wisconsin at 41 and a half, well past peak bone accrual, with whatever skeleton I'd already built. The magnesium half tells a different, more open-ended story — one that may still have something to offer people my age and older, as the supplementation evidence slowly matures. I won't get the childhood benefit of this water. I might still get something from what's dissolved in it now, if the early signals hold up. Mostly, though, what stays with me is the bigger and slightly strange realization underneath all of it: something as ordinary as where you happen to live — the geology under your house, the minerals leaching into your tap water — has been quietly shaping human skeletons for as long as people have been drinking from the ground. I only thought to ask about it because I moved somewhere new and noticed the kettle scale. Most people never think to ask at all.

Microplastics are having a moment. They've turned up in blood, in placentas, in breast milk, and, most alarmingly, in a 2024 NEJM paper that's done more to shape the public conversation than almost anything else in this space, in carotid artery plaque, where their presence correlated with a higher rate of cardiovascular events. Here's what's actually established: microplastic particles are measurably present in human tissue and blood, and we absorb them through what we eat, what we breathe, and what touches our skin. That part is real. Here's what isn't established: whether any of it matters. There's no dose-response curve. No outcome data tying a given blood concentration to a given health effect. No consensus on whether the microplastics circulating in your blood on a Tuesday afternoon bear any meaningful relationship to the total burden sitting in your tissues. We have exposure. We do not have consequence, not yet, not with the kind of evidence that lets a clinician make a recommendation. That gap is exactly the space a wellness industry moves into fastest. Uncertainty reads as opportunity. So when a paper crossed my desk this year claiming that therapeutic plasma exchange can lower your circulating microplastic burden, I wanted to like it. I run an apheresis clinic. I would love a legitimate new indication. I read the methods first, the way I always do. What They Did Weinstein and colleagues, publishing in the Journal of Clinical Apheresis in 2026, tested 114 patients undergoing 174 single-plasma-volume TPE procedures on a Spectra Optia system. Blood was drawn immediately before and after each procedure and tested for microplastic particles using PlasticTox, a proprietary assay that involves drying a blood sample on a card, mailing it to a central lab, staining the isolated particles with Nile Red, and counting them under fluorescence microscopy. The procedures took place in functional medicine outpatient clinics. The indications were longevity support, postural orthostatic tachycardia syndrome, myalgic encephalomyelitis, and long COVID, none of which are established indications for TPE by any apheresis society guideline I'm aware of. No IRB approval was sought, because the authors classified this as normal-course-of-care data collection rather than research; consent was a checkbox on the standard TPE consent form, agreeing that results could be used anonymously in publications. The Number That Should Stop You The topline result, stated plainly: TPE lowered circulating microplastic counts, but only in patients who started with a lot of them. Below a certain threshold, TPE made things worse. Starting MP (per 100μL) Pre-TPE mean Post-TPE mean p-value 0–9 4.4 14.4 <0.001 (increase) 10–19 13.8 11.7 0.062 (no change) 20–29 23.6 16.1 0.040 (decrease) ≥30 52.2 21.1 <0.001 (decrease) Read the top row again. Patients who started with the lowest microplastic burden had over three times as many circulating particles after the procedure meant to remove them. That's not noise; the p-value is under 0.001. The explanation is almost certainly mechanical rather than biological: the apheresis tubing set and fluid bags themselves shed microplastic particles into the circuit as it runs. The paper's own tubing measurements bear this out: meaningful particle counts in the priming saline before it ever touched a patient. At low starting burdens, the plastic you're being infused through outpaces whatever the procedure removes. Only above roughly 30 particles per 100μL does removal clearly win. Who's Selling This Seven of the paper's thirteen authors are affiliated with Circulate Health, a company that provides contract TPE services to private clinics. The remaining four are officers of the clinics where the data were collected. I'm not raising this to imply fraud; the data appear to be honestly reported, reversal and all. But when the people measuring a therapy's effect are also the people selling it, that's a fact the reader needs before they get to the conclusion, not after. An Assay Grading Its Own Homework PlasticTox is described in the paper as validated by a CLIA-certified reference laboratory. The validation data itself is not disclosed; it's held as proprietary by the company that sells the test. So the entire quantitative backbone of this study rests on an assay whose performance characteristics you're asked to take on faith. This is precisely the kind of thing laboratory medicine exists to prevent. We don't let a diagnostic test dictate a treatment decision until someone outside the company selling it has verified that the test measures what it claims to measure, reliably, across the range of values that matter clinically. An unvalidated assay with undisclosed methods, sitting downstream of a commercial incentive, isn't a minor limitation buried in a discussion section. It's the whole foundation, and it's built on trust rather than evidence. What's Missing Entirely Even if you grant every number in this paper at face value, even if TPE reliably lowers circulating microplastic counts above some threshold, nobody has shown that doing so changes anything for the patient. There is no clinical outcome data here. No symptom scores, no follow-up, no signal that a lower particle count translates to less disease of any kind. The paper answers whether a number can be moved. It does not, and cannot, answer whether moving it helps anyone. Where I Land I don't think the underlying question is silly. Whether microplastics move freely between tissue depots and the bloodstream, whether the blood compartment is a meaningful proxy for total body burden, or just a transit lane, is a real and interesting mechanistic question, and one worth studying properly. If it turns out blood truly is in dynamic equilibrium with tissue stores, apheresis might someday be a legitimate tool for something we don't yet have tools for. But that is not what this paper demonstrates. What it demonstrates is a commercial TPE provider running a therapy already being sold to patients, measured with an assay the company can't independently verify, with no outcome data attached, and a result that reverses at exactly the exposure level most of their patients probably start at. The biology might be worth chasing. The product being sold on the back of it is not the same thing as evidence that it works.

The call comes in from oncology. Their patient — a fifty-something with metastatic melanoma, two months into pembrolizumab — has been confused for three days. Low-grade fever. Can’t tell you the year. Can’t tell you where they are. Infectious workup negative. LP unremarkable. MRI with some FLAIR signal in the mesial temporal lobes. They started high-dose IV methylprednisolone 48 hours ago and the patient isn’t better. Neurology thinks it’s ICI encephalitis. Oncology wants to know if you can do something. You can. Here’s what you need to know. What Is ICI Encephalitis Immune checkpoint inhibitors are monoclonal antibodies. They work by blocking co-inhibitory receptors — CTLA-4, PD-1, PD-L1 — that normally keep T and B cells in check. Releasing those brakes is the whole point: you want the immune system to attack the tumor. The problem is that the same mechanism that kills cancer cells also breaks peripheral self-tolerance, and the collateral damage can affect virtually any organ system. These off-target autoimmune and inflammatory complications are collectively called immune-related adverse events, or irAEs. The nervous system is a particularly vulnerable target. Neurological irAEs occur in roughly 3–12% of patients on ICI therapy, depending on the agent and whether it’s given as monotherapy or in combination. The spectrum is broad: encephalitis, meningitis, Guillain-Barré syndrome, myasthenia gravis, transverse myelitis, cranial neuropathies. ICI-associated encephalitis specifically — inflammation of the brain parenchyma — is rare but can be severe and difficult to treat, particularly once it has failed first-line steroids. Why the Immune System Attacks the Brain The pathophysiology of ICI encephalitis is not a single pathway. It is a convergence of mechanisms, and understanding them matters because the mechanism predicts who will respond to TPE. The most common mechanism is amplification of pre-existing subclinical autoimmunity. Many cancer patients harbor latent B cells or low-titer neural autoantibodies before they ever start ICI therapy — a consequence of the immune system’s exposure to tumor antigens that cross-react with neuronal proteins. These pre-formed autoantibodies are usually held in check by co-inhibitory signaling. When checkpoint blockade removes those inhibitory signals, a subclinical autoimmune process that was already present gets amplified to clinical disease. This is supported by retrospective data showing that anti-Ma2 and anti-acetylcholine receptor antibodies were detectable in pre-treatment sera of patients who later developed ICI-related neurological syndromes. The drug didn’t create the autoimmunity from scratch. It unmasked it. A second mechanism is tumor-driven antigen presentation. Many tumors ectopically express neuronal proteins that are normally sequestered behind the blood-brain barrier. When tumor cells die and release these antigens, dendritic cells take them up and present them to the immune system. Under normal conditions, co-inhibitory signals prevent a full response. Block those signals with a checkpoint inhibitor, and the immune system is primed against neuronal targets shared between the tumor and the brain. The clinical distinction that follows from this is the one that determines your treatment strategy. Encephalitis associated with cell-surface antibodies — anti-NMDAR, anti-LGI1, anti-GABA-B, anti-GAD65, anti-AMPA — is directly antibody-mediated. These antibodies internalize receptors, block synaptic signaling, and activate complement. Remove the antibodies, and the pathogenic process is interrupted. This is why IVIG, TPE, and rituximab work in these cases. Encephalitis associated with intracellular or onconeuronal antibodies — anti-Hu, anti-Ma2, anti-Yo — is driven primarily by a cytotoxic T-cell response. The antibodies are biomarkers, not effectors, and removing them with PLEX is unlikely to change the course of disease. These cases carry substantially worse prognosis — mortality around 23–35% — and respond poorly to antibody-depleting therapies. Why TPE — and Why It’s Different Here When you’re called about a patient with steroid-refractory ICI encephalitis, or other irAE, TPE offers something that other immunosuppression doesn’t: two mechanisms operating simultaneously. The first is the one you’d expect — removing the pathogenic autoantibodies driving the disease process. The second is less obvious and easy to overlook: you are also removing the checkpoint inhibitor itself. Remember that ICIs are monoclonal antibodies, and monoclonal antibodies have long half-lives. Pembrolizumab’s half-life is approximately 27 days. Nivolumab’s is around 27 days as well. Ipilimumab’s is about 14 days. Discontinuing the drug, which is always the first step, does not mean the drug is gone. The patient in your opening scenario stopped pembrolizumab when their symptoms started, but they still have weeks of circulating drug maintaining receptor occupancy on T cells and continuing to drive the inflammatory process injuring their brain. TPE accelerates clearance of the ICI in a way that waiting simply cannot. This dual mechanism — antibody removal plus drug clearance — is what makes TPE uniquely well-suited to the ICI setting. TPE Beyond Encephalitis ICI encephalitis is one indication for TPE. It is not the strongest one. The table below summarizes the irAEs for which PLEX appears in guideline-based management, the role it plays, and key clinical notes. The ICI half-life argument applies across all of them: regardless of the specific irAE, you are treating the immune injury and simultaneously clearing the drug that is sustaining it. Disorder Role of TPE Notes Myasthenia gravis / MG-like syndrome First-line Initiate PLEX or IVIG alongside IV methylprednisolone 1–2 mg/kg/day at grade 3–4; do not wait for steroid failure Guillain-Barré syndrome First-line Start PLEX or IVIG at any grade above mild; steroids added in ICI-related GBS unlike idiopathic GBS Encephalitis Steroid-refractory escalation Add PLEX or IVIG if severe or progressing after 24–48 hours on high-dose methylprednisolone; cell-surface antibody profile predicts better response than intracellular antibody profile Demyelinating disease (optic neuritis, transverse myelitis, ADEM) Steroid-refractory escalation Consider PLEX or IVIG if no response or worsening after 48 hours of high-dose IV methylprednisolone Myocarditis Steroid-refractory escalation Among additional options for hemodynamically unstable patients not improving within 24–48 hours on steroids Triple M syndrome (myocarditis + myositis + MG) First-line for MG component IVIG and/or PLEX specifically indicated for MG-like presentations within the syndrome ICI-induced TTP First-line Standard TTP protocol; single procedure simultaneously removes anti-ADAMTS13 antibody, the ICI itself, and replaces ADAMTS13 from donor plasma What to Tell Oncology When They Call Back to your patient, still confused on day two of methylprednisolone. You are not going to wait for an antibody panel — those are send-outs, and in most centers they aren’t available at all outside a research context. The clinical picture is enough: steroid-refractory ICI encephalitis in a patient who has been off pembrolizumab for days and still isn’t improving. That’s a patient with weeks of circulating drug still driving the process. TPE is a reasonable next step. When oncology asks why TPE and not just more immunosuppression: you are doing two things at once. You are removing whatever antibodies are driving the encephalitis. And you are clearing weeks of circulating pembrolizumab that discontinuation alone cannot touch. That dual mechanism is what makes this worth doing — and what makes it a transfusion medicine problem, not just a neurology one. Checkpoint inhibitors have transformed oncology. They have also created a new category of patient that transfusion medicine is increasingly being asked to manage — patients whose immune systems have been deliberately unleashed and are now doing damage that can’t be walked back with steroids alone. The irAE landscape is broad and still evolving, and the role of TPE within it is broader than most people outside the field realize. Understanding why it works here, and when to reach for it, is increasingly part of what it means to practice transfusion medicine.

There was an attending I worked with as a fellow who had a habit. Patient looks fluid overloaded, hemoglobin is borderline-low-ish, reach for a unit of red cells. The logic, stated out loud more than once: it'll help pull some of that fluid back into the vessels. Oncotic pressure. It made physiologic sense in the moment, the way a lot of things in medicine make sense until you actually look up the numbers. I looked up the numbers. What oncotic pressure actually is Oncotic pressure — colloid osmotic pressure, if you want the precise term — is the pressure exerted by large proteins suspended in plasma that can't easily cross the capillary wall. It's the force that keeps fluid inside your blood vessels instead of leaking into the interstitium. Normal human plasma runs around 25–28 mmHg of oncotic pressure, and the overwhelming majority of that comes from one protein: albumin. Not hemoglobin, not clotting factors, not globulins in any meaningful way. Albumin. This matters because a unit of packed red blood cells is not albumin-rich plasma. It's red cells suspended in a small volume of additive solution with very little protein left in it. Whatever oncotic punch it has isn't coming from the cells themselves — red cells are too large to meaningfully contribute to a colloid osmotic gradient — and there isn't much plasma left to carry albumin along for the ride. The numbers A 2020 study directly measured colloid osmotic pressure across blood products and found packed red cells sit at about 1.9 mmHg. For comparison, fresh frozen plasma measures around 20.1 mmHg, and normal human plasma is roughly 25.4 mmHg. Platelets land somewhere in between, around 7.5 mmHg. Storage didn't change any of this — old units and fresh units had essentially the same low oncotic pressure. The authors' own conclusion is worth sitting with: because RBC oncotic pressure is so low, pulling extra fluid into the vasculature ("third-spacing" fluid into the blood) is an unlikely mechanism behind transfusion-associated circulatory overload, one of the most feared complications of transfusion. In other words, the product my attending reached for to manage fluid overload doesn't have much oncotic pressure to offer. They're just not doing much osmotically, full stop. Okay, but albumin must work, right? This is where it gets more interesting than "RBCs don't work, use albumin instead." Because albumin — the actual oncotic heavyweight, the protein doing 75–80% of the work in normal plasma — doesn't have a clean track record either. The 2026 Surviving Sepsis Campaign guidelines suggest using crystalloids alone over crystalloids with supplemental albumin for fluid resuscitation in adults with sepsis or septic shock, a conditional recommendation based on moderate-certainty evidence. The ALBIOS trial, the largest sepsis-specific study of its kind, found no difference in 28-day mortality between albumin-plus-crystalloid and crystalloid alone. The SAFE trial, comparing albumin to saline across a broad ICU population, found no overall mortality difference either — though subgroup analyses have repeatedly hinted at a possible benefit in septic shock and a signal of harm in traumatic brain injury. Cochrane's own position on albumin has flipped at least once over the decades as new trial data accumulated. The guidelines do carve out two situations where supplemental albumin may still be reasonable: patients who've already received large volumes of crystalloid, and patients with cirrhosis. Outside of those, the oncotic theory and the clinical outcomes data aren't telling the same story. Part of the disconnect may be mechanistic. Albumin's advantage is theoretically largest in a vessel wall that's behaving normally. In sepsis and other inflammatory states — exactly the conditions where clinicians are most tempted to reach for it — capillary permeability increases, and infused albumin can leak into the interstitium right along with crystalloid. Once albumin is outside the vessel, it pulls fluid outside the vessel with it, paradoxically worsening fluid overload. Sitting with the gap So here's where I land, and I want to be honest that it's not a clean place to land: neither RBCs nor albumin reliably produce the specific physiologic outcome — durable intravascular fluid retention — that you would predict. RBCs because there's barely any oncotic pressure to speak of. Albumin because the theoretical advantage gets diluted by capillary leak in exactly the patients where it's most often considered. I don't think this means "never give albumin" or "never transfuse RBCs in fluid overload." There may be other legitimate reasons to transfuse a fluid-overloaded patient — symptomatic anemia doesn't go away just because someone's also volume overloaded, and the clinical picture is rarely just one variable. What I think this does mean is that "it'll help pull fluid back in" is doing a lot of work that the data doesn't actually support, for either product. The harder version of this "It physiologically makes sense" and "it has been measured to do that in patients" are two different claims, and a fair amount of practice in medicine quietly substitutes the first for the second. Oncotic pressure is real, measurable, and important. It's also not a license to assume that giving a product with theoretically favorable properties produces the clinical effect we're hoping for. Sometimes the most rigorous thing you can do with a comfortable physiologic story is go check whether it survived contact with a colloid osmometer. References Klanderman RB, et al. Colloid osmotic pressure of contemporary and novel transfusion products. Vox Sanguinis. 2020.

A patient calls the clinic the morning of their appointment. Their ride canceled. Again. The scheduler offers to rebook, but the patient hesitates — they've already canceled twice this month for the same reason, and they're starting to wonder if the clinic thinks they just don't want to come in. They do. They just can't get there. I've heard versions of this story enough times recently that it stopped feeling like a string of coincidences and started feeling like a pattern. Transport companies are declining Medicaid and Medicare rides because the reimbursement doesn't cover the cost of fuel to get there. No-shows and cancellations are climbing. And in a separate but oddly parallel thread, research coordinators are reporting that study participants — people who once reliably showed up for their visits — are skipping appointments because the incentive payment no longer covers what it costs to drive there. These are two different systems, two different funding mechanisms, two different sets of patients. But they're failing for the same underlying reason, and I don't think that's a coincidence. How a flat rate breaks under a variable cost Non-emergency medical transportation, or NEMT, is supposed to be the safety net that gets Medicaid and Medicare beneficiaries to dialysis, infusion, oncology follow-up, and the dozens of other appointments that can't happen by telehealth. The way these trips are usually priced is a base rate plus a per-mile mileage fee, and that mileage fee is explicitly built to reflect local fuel prices, vehicle maintenance, and regional economic conditions. Reimbursement rates themselves vary enormously by state — the same wheelchair-accessible trip might pay around $100 in one state and roughly a third of that in another, because federal law requires states to provide NEMT but leaves the actual payment rate entirely up to them. That structure works fine as long as the underlying cost of driving stays roughly where it was when the rate was set. It does not work when fuel prices climb faster than the rate gets revised. A transport company running on Medicaid mileage reimbursement doesn't have the option of just absorbing the loss trip after trip — they stop taking the trips. Which is, anecdotally, exactly what's happening. The data we already had Here's the part that surprised me a little: we didn't need a fuel crisis to know transportation barriers cause missed appointments. That literature already exists, and it's not small. A frequently cited estimate puts the number at roughly 5.8 million Americans missing or delaying medical care annually because of transportation barriers, concentrated in rural and underserved urban areas where transportation options are already limited. In one study of caregivers in Houston, an inability to find a ride caused at least one missed appointment in a quarter of the sample. A systematic review and meta-analysis of interventions aimed at exactly this problem — vans, bus vouchers, rideshare — found they meaningfully reduced missed appointments, though the evidence on whether that translated into better health outcomes or lower costs was too thin to say for sure. So the mechanism by which "can't get a ride" becomes "missed dialysis session" was already well established. What's new isn't the mechanism. It's the scale and speed at which fuel prices are stressing a system that was already running close to the edge. The same problem, wearing a different badge The research side of this is structurally different but rhymes uncomfortably well. Clinical trial and study compensation has its own literature on travel reimbursement, and it's clear on one point: covering travel costs isn't a perk, it's often the thing standing between "this person can participate" and "this person can't afford to." One review of payment practices noted that travel costs remain one of the most significant barriers to clinical trial participation, particularly for low-income participants. Separately, researchers studying recruitment and retention have argued that travel reimbursement is an appropriate and valuable incentive precisely because, without it, participation becomes a luxury good — available to people who can absorb the cost of getting there, and closed to everyone else. If incentive payments were calibrated to cover a $15 round trip in gas and now the actual cost is closer to $25, that calibration has quietly become a barrier, even though the dollar amount on paper hasn't changed. The people most likely to drop out under those conditions are, predictably, the people for whom that gap matters most — which is its own quiet threat to the diversity and generalizability of the data we're collecting. What I can't tell you yet I want to be honest about the limits of what I'm describing. There is, as far as I can find, no published literature yet on this specific moment — on fuel prices rising fast enough to push NEMT providers out of Medicaid and Medicare contracts, or on research incentive payments failing to keep pace with gas prices in real time. What I have is a well-documented mechanism (transportation barriers cause missed appointments and lower trial retention) colliding with an acute, recent stressor (fuel costs outpacing reimbursement) that hasn't been studied yet because it's still happening. It would be tidier to end this with a clear causal claim and a clean policy fix. I don't think I'm entitled to either yet. What I can say is that two systems I don't normally think about together — clinical transportation logistics and research recruitment economics — are both showing the same symptom right now, and that symptom is patients and participants disappearing from the schedule not because they don't want to be there, but because the math of getting there no longer works. A structural irony, if you're looking for one The patients most likely to need frequent transportation-dependent care — dialysis, transfusion, infusion therapy, complex follow-up — are, by definition, the ones who can least afford for this particular gap to widen. We built a system where access to care depends on a per-mile rate someone set years ago, in a different fuel market, and we're now finding out what happens when that assumption quietly stops holding.

In 2008, an FDA advisory panel sat down with a meta-analysis that pooled thirteen randomized trials of cell-free hemoglobin-based oxygen carriers — HBOCs, for short — and found that, as a class, these products increased the risk of myocardial infarction and death compared to controls. Within the year, the FDA had effectively frozen HBOC development in the United States. Almost two decades later, the freeze hasn’t really lifted. One of the products caught in it, Hemopure, has spent that entire time legally available in South Africa, used there since 2001 for acute surgical anemia, with no comparable reckoning. That gap is the interesting part. Not whether Hemopure works — it does, in the narrow sense of carrying oxygen — but why a product can be standard of care in Johannesburg and investigational-only, accessible solely through expanded access protocols, in Boston. The Pitch Hemopure (HBOC-201) is purified, glutaraldehyde-polymerized bovine hemoglobin, suspended in a balanced electrolyte solution and packaged in a 250 mL bag. It solves, on paper, two of transfusion medicine’s oldest structural problems at once. First, compatibility: there’s no antigen to react to, so no type and screen, no crossmatch, no antibody workup — a feature that matters enormously for a patient with a complex alloantibody history, or a Jehovah’s Witness declining allogeneic blood, or a combat medic with no time and no lab. Second, supply: it’s shelf-stable at room temperature for years, not the 42 days we get out of refrigerated red cells. No cold chain, no expiration anxiety, no donor recruitment problem. It is, in other words, exactly the product blood banking has wanted since the first synthetic oxygen carrier was proposed. Which is part of why its failure to gain US approval stings more than a typical drug rejection — this isn’t a marginal improvement on an existing therapy. It’s a different category of solution to a problem we still haven’t solved. What the Meta-Analysis Actually Said The 2008 Natanson analysis, published in JAMA, didn’t study Hemopure alone. It pooled data across five distinct molecules — HemAssist, PolyHeme, Hemolink, Hemopure, and Hemospan — spanning surgical, trauma, and stroke populations treated between 1980 and 2008. The conclusion was stark: roughly a 30% increase in risk of death and nearly a threefold increase in risk of myocardial infarction across the pooled trials. The FDA responded by putting HBOC research as a class on clinical hold, and pharmaceutical interest in the space mostly evaporated. The proposed mechanism made biological sense and still does: free hemoglobin outside the protective confines of a red cell membrane scavenges nitric oxide, the molecule responsible for vasodilation. Scavenge enough of it and you get vasoconstriction, hypertension, and — plausibly — myocardial ischemia. This isn’t a manufacturing defect specific to one company. It’s closer to a property of cell-free hemoglobin itself, which is a much harder problem to engineer around. The Harder Question Here’s where I think the story gets genuinely uncomfortable, and where I have a hard time being charitable to the paper that started all of it. Natanson and colleagues pooled thirteen trials of five chemically distinct molecules — different polymerization strategies, different patient populations, different routes and doses, trauma and elective surgery and stroke trials run across nearly three decades — into a single composite risk estimate, and reported finding no significant statistical heterogeneity across that grab-bag. I find that more suspicious than reassuring. Getting a clean, homogeneous-looking signal out of five drugs that don’t share a structure, in populations that don’t share a baseline ischemic risk, is exactly the kind of result that should prompt a second look at the methodology rather than a press release. Several independent groups thought so too: JAMA ran six separate rebuttal letters in the same issue — from South African clinicians with the largest real-world experience with Hemopure, from the manufacturers, from trauma surgeons, from bioethicists — which is not a normal amount of pushback for one meta-analysis to generate. Natanson’s own paper acknowledged that the authors had struggled to obtain complete trial data directly from the companies, meaning the headline number was built partly on data the authors themselves described as incomplete. It also doesn’t help that the senior author, Sidney Wolfe of Public Citizen’s Health Research Group, had already petitioned the FDA over HBOC trial safety back in 2006 — two years before he co-authored the analysis that became the FDA’s rationale for freezing the entire class. None of that automatically makes the conclusion wrong. But a paper with this many independent critics, this much acknowledged missing data, and an author who’d staked out the answer in advance is not the kind of evidence I’d want sitting alone at the foundation of a two-decade regulatory freeze — and yet here we are, two decades later, and it still is. And then there’s the South Africa and Russia question, which nobody seems eager to sit with for very long. If the safety signal were straightforwardly damning, you’d expect those approvals to have been revisited over twenty-plus years of real-world use. They haven’t been. Either the signal doesn’t replicate cleanly outside the specific trial populations that generated it, or post-marketing surveillance in those countries simply isn’t rigorous enough to have caught it — and I genuinely don’t know which of those is true. Both possibilities should make a transfusion medicine physician uneasy, just in different directions. Meanwhile, expanded access use in the US has quietly continued for patients with life-threatening anemia and no other option — mostly Jehovah’s Witnesses and patients with antibody profiles that make compatible blood functionally unobtainable. Case series from these programs report real patients surviving severe anemia they likely wouldn’t have survived otherwise, alongside the same cardiovascular signal the trials raised. The regulatory caution and the individual patient calculus are not measuring the same thing, and I don’t think they’re supposed to converge. A population-level hold protecting against a class-wide signal can be correct and still be the wrong answer for the specific patient in front of you with no other option. Sitting with that tension honestly is harder than resolving it in either direction. Where It Sits Now Hemopure remains investigational in the US, available only through expanded access or clinical trial. HbO2 Therapeutics, the company that now holds the product after Biopure’s bankruptcy and a subsequent ownership chain, has kept it alive primarily through that compassionate-use pathway and continued approval in South Africa and Russia. The broader HBOC field never really recovered momentum after 2008; most of the other products named in the Natanson analysis are gone entirely. Hemopure is something closer to a survivor than a success — still infused, still studied in scattered case reports, still without a clear path to a US indication. There’s a newer thread worth watching: small case literature on HBOC-201 for ischemic rescue in cardiology and vascular contexts, distinct from its original blood-substitute framing. Whether that becomes a real niche or stays anecdotal is an open question, and I’m not going to pretend I know which. I don’t think this is a story with a villain. The FDA did what regulatory agencies are supposed to do when a meta-analysis raises a mortality signal across a drug class. But twenty years on, with the same product still quietly saving the occasional patient who has no other option, and still in routine use on two other continents, I find myself less sure than I’d like to be about whether the caution and the evidence are still pointing in the same direction — or whether we’re applying a 2008 verdict to a 2026 question.

The system is collapsing. Before we try to save it, we should ask whether it was working. Something Has Changed Something has changed in academic publishing. Papers I submit take longer to get reviewed than they used to. Desk rejections — the kind where a paper doesn’t make it out to reviewers at all — feel more common. When a review does come back, it sometimes arrives months after submission, accompanied by an apology from an editor who clearly struggled to find anyone willing to assess the manuscript. I’ve wondered if I’m imagining it, or if my experience is just narrowly my own. It isn’t. The Infrastructure Is Fraying Simberloff and colleagues recently published 21 years of editorial data from Biological Invasions — a granular, longitudinal dataset that makes the pattern hard to argue with. In 2003, more than 60% of invited reviewers accepted. By 2023, that number had fallen to just below 40%. Decline rates rose to match. The lines have now converged: for every scientist who says yes, one says no. If the trend holds, declines will soon outpace acceptances. This is one journal, one field. But a 2018 Publons survey found something consistent across all scientific disciplines: 10% of reviewers complete more than half of all reviews. The system is not failing uniformly. It is being held together by a small, overloaded minority while everyone else declines — and, increasingly, doesn’t bother explaining why. In the Biological Invasions data, the most common reason given for declining is being too busy, a response that has grown more frequent over time. Lack of expertise is also frequently cited. But roughly half of all decliners give no reason at all. There are no consequences for saying no, so scientists have stopped feeling the need to justify it. But Was It Ever Working? Before we treat this as an unambiguous crisis, it’s worth asking what exactly we’re losing. Peer review has long been treated as the quality-control mechanism of science — the filter that keeps bad research out of the record. That assumption deserves scrutiny. The psychologist Adam Mastroianni has written compellingly about peer review as a failed experiment. The evidence he marshals is uncomfortable. Studies in which researchers deliberately inserted major errors into manuscripts — things like misrepresented study designs, unsupported conclusions, obvious discrepancies between data and graphs — found that reviewers caught somewhere between 25 and 30% of them. Not 25 to 30% of minor quibbles. Major methodological flaws. Most of what reviewers are supposed to catch, they miss. The fraud data tell the same story. If peer review were functioning as a rigorous filter, we would hear about fraud attempts stopped at the gate. We don’t. Almost every high-profile case of scientific fraud begins with a paper that passed review and was published. The detection comes later — from a lab member, a methodologist, someone on the internet who noticed something odd about the error bars. Review did not catch it. Post-publication scrutiny did. None of this means peer review does nothing. It probably catches some errors, improves some papers, and deters some bad actors who would otherwise have no barriers at all. But the gap between what peer review promises and what it delivers is substantial. We have been running on faith more than evidence. The Bargain We Made The deeper problem is what got built on top of peer review’s assumed reliability. Hiring committees treat publication in peer-reviewed journals as a proxy for scientific quality. Grant agencies use it as evidence of track record. Clinicians — and I count myself here — use peer-reviewed literature to make decisions about patient care. The peer-reviewed label became a kind of certification, and institutions downstream of the scientific record built their practices around it. That certification was always shakier than it looked. But the response, broadly, has been to defend peer review rather than examine it — to argue that more of it, or better-resourced versions of it, would fix the problem. The collapse now underway is forcing a different question: not how do we sustain peer review, but what do we actually need from it, and is there a better way to get there. Why Nothing Will Change Here is the detail from the Simberloff paper that has stayed with me. The editors-in-chief of Biological Invasions — the people running the journal, watching decline rates climb year after year, doing the actual work of recruiting reviewers into an increasingly reluctant pool — asked Springer Nature, their own publisher, for reviewer incentives. They asked multiple times. Springer Nature declined. This is not surprising. It is clarifying. Springer Nature collects subscription fees, article processing charges, and the commercial value of a prestigious catalog, all sustained by the unpaid labor of reviewers and the prestige conferred by the peer-review label. There is no version of that business model that benefits from fundamental reform. The current system, however dysfunctional, is profitable. Incentives to change it would have to come from somewhere else. This is also part of a larger pattern. Park and colleagues’ 2023 analysis of 45 million papers spanning six decades found a steady decline in disruptive science — work that challenges existing frameworks rather than incrementally extending them. The same incentive structure that rewards volume over depth is now degrading the mechanism that was supposed to ensure quality. More submissions, fewer willing reviewers, and the institutions profiting from the system declining to invest in its sustainability. We Need a New Model There are alternatives being tried. Preprint servers like bioRxiv and medRxiv allow rapid dissemination before formal review, with post-publication scrutiny doing some of the work that pre-publication review was supposed to do. Open peer review, where reviewer identities and comments are made public, attempts to introduce accountability into a process that currently operates without it. Some journals are experimenting with paying reviewers. These are not nothing. But none of them have yet accumulated the institutional weight that peer-reviewed publication carries. Hiring committees still count papers. Grant agencies still look at journals. Clinicians still defer to the peer-reviewed label, even knowing what we know about its limitations. The alternative models exist at the margins while the incumbent system, imperfect and increasingly unsustainable, holds the center. I don’t know what the right model looks like. I don’t think anyone does with confidence. What I do know is that we need one, that the timeline is shorter than it probably feels, and that the people with the resources and infrastructure to build it have spent decades demonstrating they have no intention of doing so. That is the peer review bargain in 2025: a system that over-promised on quality, under-delivered on rigor, is now running out of the volunteers who kept it going, and has no obvious succession plan. Referenced works: Simberloff D et al. (2025). Quantifying reviewer declines in scientific publishing: twenty-one years of data from Biological Invasions 2002–2024. Biological Invasions, 27, 223. https://doi.org/10.1007/s10530-025-03679-1 Mastroianni A. (2022). The rise and fall of peer review. Experimental History. https://www.experimental-history.com/p/the-rise-and-fall-of-peer-review Park M et al. (2023). Papers and patents are becoming less disruptive over time. Nature, 613, 138–144. https://doi.org/10.1038/s41586-022-05543-x