An evening news digest brought two lines that I read one after the other and froze. The first was about strikes on Ukrainian operators’ data centers: “The Ministry of Digital Transformation counts not destruction, but households without connectivity.” The second was about insuring tanker voyages: “the war risk premium on oil costs less than a missile.” Both are about the same thing: when a system fails, the winner is not the one who counts better, but the one who chose the right metric. Households instead of concrete. An insurance premium instead of a hit.
So I stepped away from infrastructure—and ran into a story that turns the same logic inside out. For eighty years, doctors measured starvation in calories and kilograms. And all that time, what killed people was not what they were measuring. What killed them began after they had been saved.
In the spring of 1945, British soldiers entering Bergen-Belsen saw walking skeletons—and did what any normal person would have done: gave them their rations, chocolate, canned meat. One witness later put it this way: “the internees simply could not take this rich food, and eating it invariably worsened their condition or led to death.”
Rescue killed faster than hunger. That is the story I started digging into—because behind it lies the most unpleasant thing one can say about any system: what is measurable and what matters are not the same thing.
No one prepared the liberation of the camps in 1945 as a medical operation. The British entered Bergen-Belsen on April 15—there were about 60,000 prisoners there, most of them starving, with typhus and tuberculosis. There are no precise records whatsoever of causes of death in the first days. Estimates diverge: according to Yaakov Lozowick, former director of the Yad Vashem archives, more than 30% of the prisoners died in the first days after liberation; in other camps, even more. The physician Mark Dworzecki, who had himself passed through ten camps, observed deaths from cardiopulmonary and immune failure and considered them inexplicable: in his words, many of those people should have survived.
The interesting part is not the number. The interesting part is that they were looking for the mechanism in the wrong place. The first published reports—the British journal The Lancet on the Dutch winter, Collis’s article on Bergen-Belsen, the work of Mollison, Mitchell and Black, Walters—are still cited as pioneering. But none of them openly identified rapid refeeding as the determining cause of death. The blame went to concurrent illnesses, parasites, vitamin deficiency. The electrolyte hypothesis was formulated years later, and it took decades more to be accepted.
One of those details that lands more precisely than any generalization. Janet Vaughan, who worked at Bergen-Belsen, admitted many years later: “small amounts of food proved the best treatment for starving patients.” How she arrived at that conclusion is not stated in the report. The right answer was found; the rationale was absent. The doctor guessed the dose before medicine understood that the dose mattered at all.
Running in parallel, and in the same direction, was the most famous experiment in the history of nutrition science. From November 1944 to December 1945, Ancel Keys and his laboratory in Minnesota conducted the Minnesota Starvation Experiment: 36 volunteers, recruited from pacifist sectarians who refused military service and agreed to a year of deprivation instead of prison and the front. Twelve weeks of a baseline phase at 3200 kcal, then 24 weeks of semi-starvation—1560 kcal from potatoes, rutabagas, turnips, bread, and pasta, precisely in the spirit of the European diet during the war’s final years. Average weight loss: about 25%.
Then comes the part I came here for. Recovery was split into phases: 12 weeks of restricted rehabilitation—four groups of eight men, with different calorie levels and different protein and vitamin regimens—and only then 8 weeks unrestricted. That is: even in the most controlled starvation study in history, the first three months of refeeding proceeded under strict restriction. Not because they understood the mechanism. Because they knew empirically: give it all at once, and things will get worse.
The Warsaw Ghetto gave this knowledge its most terrible confirmation. Doctors trapped in the ghetto, including Emil Apfelbaum-Kowalski and those working under Israel Milejkowski, selected one hundred cases of “pure” starvation—patients without concurrent disease—and conducted glucose tolerance and basal metabolic rate tests, meaning they gave glucose, proteins, and carbohydrates abruptly and in large quantities. The result: as soon as the metabolic changes induced by starvation rapidly reversed, the circulatory system could not adapt in time—the already weakened heart received an additional load, and the person died of heart failure.
Apfelbaum-Kowalski’s work remains the only cross-sectional study of prolonged “pure” starvation: “never before had starvation lasted long enough to lead to cachexia and death, and never before had the clinical material been so numerous.” The term they gave the phenomenon was hunger disease.
Consider the irony: the only complete dataset on how starvation kills was obtained by people it was killing too, and published in 1946, when the infirmary corridors had already emptied.
On May 29, 1945, the Royal Society held a special meeting, “The Physiology and Treatment of Starvation.” The bet was on protein hydrolysates—extracts that had performed well during the Bengal famine of 1942–1944. The disappointment was nearly total.
One detail stood out, and I like it more than all the others: “Bengal mixture”—flour enriched with sugar, salt, yeast, and fish oil—was mass-produced for starving people and was rejected by most of them for its taste. Too sweet for European mouths. It was not mentioned at the meeting at all.
But Mollison, examining the children and adults of Bergen-Belsen, recorded an observation pointing the other way: survivors refused liquid diets and demanded solid food, while some were afraid to eat—because of immediate, severe diarrhea. Bodies no one had taught physiology knew more about refeeding than the doctors did. And one more observation of his, short and horrifying: the thinnest died first.
If you had asked me which indicator describes refeeding, I would honestly have bet on sugar. The logic is obvious: during starvation there is no glucose; during refeeding, there is plenty. That is the metric everyone looked for: calories, glucose, insulin. And that is exactly what turned out to be wrong in the details.
The mechanics look like this. During starvation, in the first 24–72 hours glucose falls, liver glycogen stores are depleted over roughly the same 72 hours, and the body switches to a “fat economy”: gluconeogenesis from muscle protein via alanine, ketone bodies from fatty acids. Along with fat and protein go potassium, phosphate, and magnesium. But here is the trick: homeostasis maintains serum ion concentrations at the expense of intracellular stores. Serum can look normal when the body’s total deficit is already profound.
Then food appears. Glucose enters the blood, insulin spikes, and insulin does to electrolytes what it always does—drags potassium, phosphate, and magnesium into cells. The gradient is large, intracellular stores are empty, and extracellular concentrations plummet. On top of that, the body draws sodium and water inward—to preserve osmotic neutrality.
Notice where the metric “calories” breaks down. A calorie is an input. But the blow to the body comes from what happens before the calorie and after it: the depleted intracellular compartment and insulin, which drains it to the end. Serum phosphate is not about bones; it is a sensor on the line between two states, and it shows what a calorie physically cannot show.
This is what the real cascade looks like, the one clinicians describe as severe. Paresthesia, muscle weakness, delirium, disorientation, encephalopathy, areflexic paralysis, seizures, coma, tetany. On the cardiac side—hypotension, shock, decreased stroke volume and mean arterial pressure, increased wedge pressure, arrhythmias. On the respiratory side—diaphragm weakness, respiratory failure, shortness of breath. In the blood—hemolysis, thrombocytopenia, leukocyte dysfunction. Rhabdomyolysis, muscle necrosis.
And on a line of its own—thiamine.
Thiamine is the most eloquent character in this story, because its failure is not about deficiency. Its failure is about reserves lasting exactly until they were needed.
In a starving body, carbohydrate pathways are almost shut down, and thiamine, as a cofactor, is barely used. Bring carbohydrates back—pyruvate dehydrogenase and transketolase reactivate, and the entire scant reserve is exhausted within days. The result: lactic acidosis, Wernicke encephalopathy, Korsakoff syndrome, wet and dry beriberi. A vitamin that was not “absent,” but burned up at the entry point.
There are many similar stories, because thiamine is generally poorly designed for the modern world. A study of post-bariatric beriberi published in Obesity Facts in 2014 counted 255 published cases in the literature by December 2013 of thiamine deficiency and Wernicke-Korsakoff syndrome after weight-loss surgery. More than 90% were women. Symptoms do not rise immediately, but 4–12 weeks after surgery.
There is also one line that is not medical at all: in Germany, post-bariatric follow-up is not covered by health insurance, and most doctors have little knowledge of nutritional complications. Thousands of people undergo surgery every year after which their brains can fail, and the system sees no need to monitor them.
And in 2025, BMC Neurology described a case that carries this logic from the operating room into the pharmacy. A thirty-seven-year-old woman began taking semaglutide, 0.5 mg a week. Recurrent vomiting, loss of appetite, a loss of 45 kilograms in ten weeks from a starting weight of 158.8 kg. Glycated hemoglobin dropped seven points in ten weeks. Four months later—progressive weakness, numbness, paresis, and then confusion. Vitamin B1 level: 61 nmol/L, against a normal range of 74–222. Diagnosis: non-alcoholic Wernicke encephalopathy plus diabetes-induced neuropathy. Six months later, the patient was still mostly bedridden.
This is an important turn, and it is specifically about the metric. The drug did exactly what it was expected to do, and did it brilliantly: weight fell, blood sugar normalized, the indicators improved.
And the person was dying in her legs and brain from the speed at which those indicators improved. A review in Nutrients (February 2026) adds the broader scale: semaglutide 2.4 mg produces an average weight loss of about 14.9% over 68 weeks, tirzepatide up to 20.9%, with one person in three losing a quarter of their body mass or more. And a direct quote about the trials: “systematic assessment of micronutrient intake and status is rarely included” in the pivotal randomized trials, because priority is given to weight, glycemia, and cardiovascular endpoints.
We have brought back the world of 1945, except now the camp ration of 400 calories has been replaced by a once-weekly injection, and the people themselves do not realize they are taking part in an experiment.
Then comes the most interesting part, because this is where medicine finally tried to turn observation into protocol—and produced a document that describes the correct physiology while, at key points, saying precisely the opposite of what is happening in the body.
The term appeared in 1981: Weinsier and Krumdieck published an article in the American Journal of Clinical Nutrition titled “Death resulting from overzealous total parenteral nutrition: the refeeding syndrome revisited.” Notice the word revisited in a 1981 title. By the time the phenomenon officially received a name, it had already demanded a reconsideration.
Then came the definitions. Crook and colleagues in 2001: a syndrome of severe electrolyte and fluid shifts together with metabolic disturbances in depleted patients undergoing refeeding—oral, enteral, or parenteral. Britain’s NICE issued guideline CG32 in 2006, where risk is calculated from a list: BMI below 16, loss of more than 15% over three to six months, near-total absence of food for more than ten days, low potassium, phosphate, or magnesium before feeding begins. One criterion is enough for high risk; two for moderate risk—BMI below 18.5, loss of more than 10%, food refusal for more than five days, alcohol, or drugs such as insulin, diuretics, antacids, chemotherapy.
Treatment according to NICE: begin with a maximum of 10 kcal/kg per day, in extreme cases with 5, and reach full requirements over four to seven days. Thiamine, 200–300 mg per day—immediately before feeding begins and during the first ten days. Potassium, phosphate, and magnesium supplements—with a caveat deserving its own paragraph: “unless pre-feeding plasma levels are high.” And immediately after: “pre-feeding correction of low levels is unnecessary.”
Parse what is written here. In people with low phosphate before feeding, correction is not mandatory, because the threshold value is considered unreliable. At the same time, potassium, phosphate, and magnesium are considered risk criteria—meaning low values signify something. Which means they signify nothing. The protocol simultaneously uses the metric as a risk marker and acknowledges its unreliability to the point that treatment based on it is optional.
Fourteen years later ASPEN tried to fix this—and the result was more honest, but stranger. The 2020 consensus defines refeeding syndrome as a drop in any one, two, or three measures—phosphorus, potassium, magnesium—of 10–20% (mild), 20–30% (moderate), more than 30% and/or organ dysfunction (severe), within five days of restarting or substantially increasing calories. The risk criteria were rewritten, and a physical component added: loss of subcutaneous fat, loss of muscle mass, concurrent conditions. And two points make this document the most candid text in the entire field.
First: the authors state outright that the criteria for severity stratification are arbitrary. Not “require refinement.” Arbitrary.
Second: ASPEN explains why phosphate became the syndrome’s calling card. Quote: “hypophosphatemia is often considered a hallmark... however, this may be the result of definition bias, or the relatively fewer causes of hypophosphatemia compared with hypokalemia.” In other words: phosphate became the main marker not because it is central to pathogenesis, but because it has fewer competitors on the list of possible causes. The metric was chosen for convenience of measurement, not proximity to the mechanism. This is literally the same choice as “households without connectivity” instead of “destruction”: count what can be counted.
And now the part that made me reread everything from scratch. I came here for a story about a life-saving metric—and discovered that the metric is disputed from two sides, both sides armed with data.
Let us begin with the block of numbers where everything falls apart. A meta-analysis in Nature Scientific Reports—Schneider and colleagues, 28 studies, 10,412 patients, registered in PROSPERO—found a reported incidence of refeeding syndrome among critically ill patients ranging from 0% to 88%. The pooled rate was 23%, heterogeneity I² = 99%. That is a level at which studies should not be combined at all. By age: adults 29%, children 5%, newborns 25%. And most importantly: no association with mortality was found—odds ratio 1.51, confidence interval 0.79–2.88. The authors write plainly: “in this meta-analysis, refeeding syndrome was not statistically associated with mortality; however, the certainty of evidence was low and heterogeneity high.”
A separate line on treatment: “a recent systematic review found insufficient evidence to reliably support the effectiveness of these interventions for prevention and treatment of refeeding syndrome in critically ill patients.”
Now the figures from which the ones above were selected. A 1996 study defined the syndrome as hypophosphatemia in the first 72 hours—and found 34% among critically ill patients. Another study with the same definition found 8% in the at-risk group. A prospective cohort with a strict three-component diagnosis—severe electrolyte shifts plus fluid overload plus organ dysfunction—found 2% of 243. In anorexia, one series found hypophosphatemia in 5.8% of 69 patients; another found 38% moderate hypophosphatemia, without a single severe case among 46. Pediatric intensive care: 27% shifts in the entire group and 8 out of 15 in the at-risk group. And for calibration—hypophosphatemia in an ordinary hospital population: 2% of all phosphate-test requests over 18 months at one institution.
The same phosphate, the same patients, a seventeen-fold range. This is not about physiology. It is about definition.
Next comes what I have to set beside the Bergen-Belsen story, because otherwise this entire text becomes an apologia. An Australian study published in the European Journal of Clinical Nutrition in 2018, titled “Throw caution to the wind: is refeeding syndrome really a cause of death in acute care?” Eighteen years, from 1997 to 2015, about 260,000 hospital deaths in Queensland hospitals. Refeeding syndrome as a cause of death—in five people. None had a diagnosis placing them in a high-risk group: no anorexia, no history of alcoholism. Not once was refeeding syndrome listed as the primary cause of death. No patient consumed more than 3400 kilojoules a day—roughly 810 kilocalories. In 80% of those who died, electrolytes improved before death.
Eighty percent. People died with their biochemistry normalizing.
The follow-up appeared in 2021 under the title “Further evidence to throw caution to the wind”: a retrospective cohort, 70 at-risk patients, 94% malnourished, three quarters received thiamine before feeding began, 79% received a standard, not cautious, plan. Cases of refeeding syndrome: zero. There was no difference whatsoever in electrolyte decline among cautious, standard, and liberal plans (p = 0.912 and p = 0.688). The authors’ conclusion: implementing more liberal guidelines appears safe.
And third, the part that does not fit through any gate. An English prospective study in BMJ Open—Rio and colleagues, 243 adults, 2007–2009, King’s College Hospital complex—begins with a hypothesis formulated against existing guidance: refeeding syndrome risk factors are weak and cause unnecessary delays to nutrition. The result: 133 people had risk factors, the syndrome developed in three (2%), and there were no related deaths. Of the 243 patients, 13 died during feeding and 68 during hospitalization, but from stroke, trauma, respiratory failure, and terminal conditions.
And the most valuable observation concerns the other side: intravenous carbohydrate infusion before artificial nutrition began may have triggered the onset of the syndrome. In other words, the culprit is not refeeding as such, but the carbohydrate load before it—and that is exactly what we already know from the physiology of the insulin hit.
There is also a very short study that I like best of all because it contains not one unnecessary word. “Low serum phosphate is not diagnostic for refeeding syndrome in critical care” (Intensive Care, Liverpool, 2015): 104 patients with hypophosphatemia, 39 diagnosed with refeeding syndrome by NICE criteria. Area under the ROC curve for phosphate at baseline: 0.51. For the phosphate drop after feeding began: 0.56. Zero point fifty-one is a coin toss. The variable on which all the guidelines are built literally cannot distinguish those who will develop the syndrome from those who will not.
It is separately interesting to see what those who deliberately went against caution say. The Matthews-Rensch systematic review (2021): out of 2803 abstracts, 24 studies, 15 of them “assertive,” starting at more than 20 kcal/kg per day. Hypophosphatemia—up to 74% in assertive regimens and up to 72% in conservative ones. Diagnosed refeeding syndrome—up to 20% in both types. Time to reach target calories had no effect on outcomes. The conclusion: there is no consensus on the optimal method, starting energy, or pace.
Rapid refeeding in anorexia in a 2022 Canadian study—326 patients, BMI below 18—shortened hospital stay by 21 days, increased weekly gain by 0.21 kg, electrolyte disturbances were mild to moderate and managed orally, and there were no deaths and no cases of the syndrome under either protocol.
So now I have two mutually exclusive sets of facts in my hands, and I will not pretend I have chosen a side.
On one side: 30% mortality in the first days after the liberation of Bergen-Belsen, the recorded “bad response” to the glucose challenge in the Warsaw Ghetto, a witness’s direct quote that rich food killed, Stalingrad soldiers dying from small portions of meat pâté, zero deaths from refeeding in 260,000 reviewed cases, and a coin toss instead of a predictor in Liverpool.
On the other: in critically ill patients, calorie restriction in the only randomized study—Doig, 2015, thirteen Australian clinics, 339 patients—improved mortality on day 60. Plus a meta-analysis of 23 randomized studies and 11,444 patients, in which “permissive underfeeding” reduced ICU mortality—risk ratio 0.90, interval 0.81–0.99—and reduced gastrointestinal adverse events (0.79), without improving overall mortality. Meaning caution works, but judging by these data, not because it prevents refeeding syndrome—the mechanism of benefit is different and still unknown.
There is an ASPEN sentence I cannot get out of my head. In a hospital where electrolyte monitoring is part of standard care, refeeding complications “may in fact be rare.” And immediately after: routine use of nonspecific screening increases length of stay and mortality.
Read that as it stands. We do not know what refeeding syndrome is because we have not agreed on the threshold; we do not know how many people get it because we have not agreed on the threshold, and we get a spread from zero to 88%; we do not know whether cautious refeeding helps because the evidence is insufficient; and yet we know for certain that screening by imprecise criteria does harm—delaying nutrition for people who need it. The best thing one can say about this area of medicine is that the only reliable data here were obtained in 1945, from bodies that had no protocol.
It seems a stretch to explain all the deaths at Bergen-Belsen through refeeding. Typhus and tuberculosis were present in the majority, and according to Mollison, tuberculosis was found in 40% of those examined, while medics at the time attributed deaths precisely to it. The nine-volume Der Ort des Terror rejected the “death from feeding” version altogether, citing pre-existing pathologies. Who is right cannot be established: medical records of the circumstances of death in those first days do not physically exist. That 30% is the upper bound of the unknown. It is not proven refeeding statistics.
And still. Whoever these people were, one fact remains: a significant share died after the war had ended and help arrived. Mollison recorded that the thinnest died first. Eighty years of operating protocols, consensus statements, and guidelines later, we still do not have a definition that describes this.
This has now stopped being a historical story, which is precisely why I did not write only about 1945.
California’s Department of Corrections and Rehabilitation has a working document, “Hunger Strike, Fasting, and Refeeding”—a practical guide, updated in August 2018, for people refusing food in protest. Everything is laid out by condition. Low risk—fewer than fifteen days without food and no medical factors: the person eats and drinks freely, no specific monitoring is required. Moderate risk—BMI from 16 to 18.5, loss from 10 to 15%, food refusal for fifteen to twenty-eight days: for the first 48 hours, they are fed a “half” diet.
High risk—BMI no higher than 16, loss of more than 15%, low potassium, magnesium, or phosphate before feeding resumes, refusal for more than twenty-eight days: refeeding generally takes place in a licensed medical facility, escalating from 10 to 30 kcal/kg per day over a week with fluid, electrolyte, and heart-rhythm monitoring.
Notice where the document draws its boundaries. It counts not only physiology, but time: fourteen days—a BMI measurement for everyone, including those not at risk; after the twenty-first day—written information on advance directives and orders in case of incapacity. The refeeding protocol is as much a derivative of legal procedure as of physiology. The prison has to have a paper trail for handling a person, and the thresholds in this document are placed where responsibility begins.
And in May 2024, PLOS Global Public Health published a short piece by Pooja Yerramilli on the situation in Gaza, one worth rereading in full. The Integrated Food Security Phase Classification projected in March 2024 that 1.1 million people—half the population of the Strip—would be in catastrophic food insecurity. Acute malnutrition in northern Gaza rose from 1–6.8% in January 2024 to 12.5–16.5% in February.
Then comes the claim I consider central to this entire text: in the first one to two weeks of refeeding, a depleted person faces a window of risk for electrolyte and fluid shifts with seizures, respiratory failure, cardiac arrest. From October 2023 to January 2024, 84% of Gaza’s medical facilities were destroyed or damaged; those remaining operated without electricity or water.
The author’s conclusion, which in a vacuum sounds like bureaucratic caution: food cannot be delivered without medical infrastructure capable of monitoring refeeding. In context, that means aid can have a phase in which excess kills. And that a humanitarian system whose metric is tons delivered is structurally unprepared for this phase.
The same text contains one detail that travels from 1945 into today’s news without a single edit. Some released Israeli hostages were monitored in hospitals in part for suspected refeeding syndrome. The 2018 protocol for California prisons and the 2024 medical program for hostages are the same document, because the situations are the same: a person lives for years under an uncertain ration, then is brought out, and after that everything depends on whether the receiving side is clever enough not to feed them immediately.
Eighty years. Not one question resolved.