Does Meat Cause Diabetes?
Whenever I read about dietary guidelines and dietary advice, the same statement keeps turning up: meat (as everyone knows) causes diabetes - and yes, the word used is usually causes.
By way of a response, let me start with my own methodical understanding of the situation - which is what prompted me to look up the relevant studies that I will list and work through in detail afterwards.
My Understanding
Diabetes can - very crudely simplified - be described as a disease of carbohydrate intolerance. Whether it is type 1 or type 2: in both cases the body, for different reasons, cannot deal appropriately with the carbohydrates it is given - starch and sugar. Both are ultimately characterised by an elevated blood glucose level (the basis for the diagnosis) and by absent (type 1) or excessively high (type 2) insulin. When meat is said to cause diabetes, what is always meant is type 2 diabetes (T2D), so that is what I will be referring to for the rest of this article.
Methodically, the first thing that is entirely unclear to me is how anyone arrives at the idea that a completely carbohydrate-free food should produce a blood glucose spike so massive, and keep the level high for so long, that the three-month average crosses the diagnostic threshold for diabetes.
That goes double given that the dietary guidelines explicitly call for more than 50% of energy intake to come from carbohydrates. That alone is a guarantee of a significant rise in blood glucose and insulin (and you can try it out next door in the simulation), whereas meat - taken on its own - has only a negligible effect on insulin, which tends to lower blood glucose rather than raise it. That, too, is there to be tried out in the simulation.
And once you add reports(1,2) of short-term blood glucose normalisation and long-term remission of T2D, and the fact that with a suitable ketogenic or even carnivore diet T2D patients can reach normal blood glucose and insulin levels within days(3), my scepticism towards the claim that meat causes diabetes is high to begin with.
But I do not want to leave it at “I don’t believe it” - I want to analyse this with more substance. Perhaps there really is a mechanism that explains something I did not know before.
Reasonable Suspicion
In the article Is Meat Carcinogenic? I already devoted a few paragraphs to nutritional science in general, its fondness for epidemiological (observational) studies and its lax handling of how risks are phrased. My underlying suspicion in this case is therefore that things are similarly lax when it comes to meat and diabetes: I am going to look at the study type (remember: observational studies can show correlations, but not causal relationships, unless the correlation is exorbitantly large). On top of that, the risk classification - “relative or absolute risk” - becomes important again, and with a relative risk, so does the size of the factor found.
Essential Background on Relative Risk
Is Meat Carcinogenic?
And because this should be on the table right from the start: I also read every study asking who paid for it and what the researchers are convinced of - and that applies explicitly to the studies I like, too. Virta Health, for instance, sells precisely the intervention whose effectiveness the company researches. That does not invalidate the results, but it belongs in the open, and it cuts both ways(4).
But of course a mere suspicion is not enough - next we will also look at the proposed mechanisms: are there plausible reasons that have so far escaped me as an outside observer?
The State of the Evidence
The studies most readily cited for the link between meat consumption and diabetes are the two “Nurses’ Health Studies” NHS I + II, along with the “Health Professionals Follow-up Study” (HPFS, the male counterpart to NHS I+II). All of them are large cohort studies, started in 1976, 1989 and 1986, in which several hundred thousand participants were asked about their dietary habits every few years.
I have looked at four analyses of this data(5,6,7,8) - among them a meta-analysis that brought in further data sources, and one that looks specifically at cohorts in East Asia and China, where T2D is particularly common. The result up front, so we can spare ourselves the columns of figures: they all arrive at the same picture. Processed meat consistently comes off worse than unprocessed, the increases in risk that were found lie between 3% and 51%, and not a single one comes anywhere near a factor of 2. What is particularly remarkable is the Asian analysis(8): there the association for unprocessed red meat is no longer statistically detectable at all.
As a representative case I will take a closer look at the numbers from Pan et al. - they deliver the highest values of the four and are therefore the best possible case for the thesis that meat causes diabetes.
The Numbers: Relative Risk
A short recap: the general consensus in science is that observational studies can uncover correlations - but run the risk of being produced by confounding factors. The scientific disciplines outside nutritional science therefore demand at least a relative risk of 100% (a factor of 2) before pursuing the suspicion that a correlation is relevant.
The numbers Pan et al. give for the relative risk are:
- Unprocessed meat (100 g/day): 19%, a factor of 1.19 (the 95% confidence interval runs from 1.04-1.37).
- Processed meat (50 g/day): 51%, a factor of 1.51 (95% confidence interval 1.25-1.83)
If we squint really hard and look only at the very top end, the confidence interval for processed meat does get closer to 2 than the studies looked at so far - so we are gradually approaching the lower bound at which it can be worth investigating whether a correlation is relevant. But I also have to say: I cannot keep up this level of sarcasm, feigned neutrality and disbelief for long. What we apparently have here is yet again a correlation minimally different from 1, one that in no world outside the ivory tower of nutritional science would be granted so much as a second glance(9).
The Numbers: Absolute Risk
Relative risks have one decisive disadvantage: they say nothing about how large the risk is in the first place. A 20% increase in risk sounds threatening - but 20% of what?
Gu et al.(6) publish their raw data, so this can be recalculated. If you compare the people who eat the least red meat with those who eat the most - that is, the two most extreme groups in the study, not meat eaters against vegetarians:
That is the whole difference: around two people per 100 in ten years. For processed meat the gap is somewhat larger, for unprocessed somewhat smaller - the familiar pattern.
And this calculation is still generous: it puts the two extreme groups directly against each other without correcting for confounders. The study itself calculates more carefully and accordingly comes out lower. So even the most charitable reading yields an effect you can barely find with the naked eye.
Interim Conclusion: Significant Does Not Mean Relevant
The studies all demonstrate a relative risk well below 2 - a threshold that the studies in Is Meat Carcinogenic? also fell far short of. Since these are observational studies, the risk of confounding factors is too high, and an association this small - even when it is reported as “statistically significant” - is simply no basis for the cause/effect claim that meat causes diabetes - and certainly none for far-reaching consequences.
So how large would an effect have to be to actually stand out? Here is a comparison from the same condition: in a randomised trial(10) a change of diet put 36% of participants into remission from diabetes after two years - against 3% in the control group. That corresponds to a factor of roughly 26, not 1.19.
That is what an effect looks like when you do not have to go hunting for it with your eyes screwed shut. And with that, the question in the headline would actually be answered: there is no statistical reason to go looking for a cause at all.
Proposed Mechanisms
Having now established that there is really no reason to investigate any causality, my first instinct was to end this article right here: there are certainly a number of biological explanations put forward in support of causality, but if the statistics already deny them their relevance, why should I go to any particular trouble over the biology?
But then it turned out that this part became especially interesting - during the research I once again learned a great deal about the body.
Haem Iron: Less Iron, Same Insulin Resistance
Haem iron, a particularly bioavailable form of iron found in animal foods, is said to accumulate in the beta cells that matter for insulin production, and there to promote the production of “reactive oxygen species” (ROS) via the Fenton reaction(11).
In a randomised controlled trial (12) this mechanism was refuted: reducing iron stores did lead to an acute improvement in blood glucose and HbA1c, but had no significant effect on the HOMA index, which measures insulin sensitivity and is ultimately what matters for diabetes. That study decouples “iron stores” from “insulin activity” at exactly the point where the effect was originally described.
Nitrite/Nitrate: Bacon, Intravenously
Nitrite and nitrate - readily supplied by processed meat - are said, via the nitrosamines they form, to disrupt beta cells as well as insulin / Insulin-like Growth Factor 1 - a hormone structurally similar to insulin that promotes cell growth and division. Produced mainly in the liver, partly in response to insulin. Chronically elevated insulin levels typically correlate with elevated IGF-1. Relevant because IGF-1 as a growth factor can also promote the growth of cancer cells. β more and to lead to insulin resistance. The problem: the 2009 study(14) that prominently formulated this pathway (in the Alzheimer’s / type 3 diabetes context) did so in an animal model in rats, and with injected nitrosamine - this direct route into the body has nothing whatsoever to do with the route via food. All the more so since - sensitive souls may skip ahead or forgive me - some of the animals had it injected directly into the brain.
Incidentally: nobody would dream of trying to demonstrate the health properties of olive oil this way.
Furthermore, a 2024 meta-analysis(15) found no correlation (or, to be precise, a relative risk of (a factor of) 1.01, or 1%) for a link between nitrate intake and diabetes - this time via the relevant route of dietary intake.
AGEs: The Potato Catches Up With the Grill Crust
AGEs (advanced glycation end-products(16)) are substances that form when sugar molecules attach themselves non-enzymatically to proteins or fats. This reaction is also known as the “Maillard reaction” and is familiar, for example, from the formation of the characteristic surface of a grilled or fried steak. Here the reaction obviously happens within minutes - but it is confined to the surface, because the water content prevents deeper layers from getting hotter than 100 Β°C.
AGEs also form inside the body: “glycation” is a hint as to what is needed for that - sugar (or carbohydrates). They form particularly readily when blood glucose is elevated.
On this too there is a randomised controlled trial in humans, from 2023(17), in which no significant link could be established between the consumption of red meat and insulin resistance (or sensitivity, insulin levels, blood glucose, HbA1c or beta-cell function).
But I could not help myself and did a bit more research of my own - here are the results, this time without references:
A ketogenic diet, as discussed in many places on butterbei.de, changes the view of AGEs dramatically: under ketogenic conditions (or even just a generally low-carbohydrate diet) blood glucose stays far more stable than it does when following the current dietary guidelines, which call for more than 50% of daily energy intake to come from carbohydrates.
Continuous formation of new AGEs everywhere in the body at high glucose concentrations is a constant stress for the body. We can picture it as a latent inflammation spread everywhere across the body. True, body temperature is not as high as on the grill, but a blood glucose level that stays elevated over hours, days or weeks makes up for the missing temperature with more time. After 15 minutes on the grill a steak is only good as a shoe sole - but a diabetic walks around with elevated blood glucose 24 hours a day.
On top of that I found figures according to which only about 10% of the AGEs taken in through food are absorbed by the body at all(18), and are eliminated again very quickly by (healthy) kidneys.
It rather looks as though the satiety side dish eaten alongside the steak and every other meal catches up with the grill crust in AGE formation over time - and then overtakes it.
Under Suspicion
The Satiety Side Dish
What exactly makes us ‘sated’, and when can we eat again? Is it about volume? Hormones? Nutrients? Answers to rhetorical questions…
Read more βTMAO: Not the Arsonist, but the Smoke
This pathway is a little more convoluted than the previous ones: gut bacteria convert L-carnitine and choline - both plentiful in red meat - into trimethylamine (TMA). The liver oxidises this TMA further into trimethylamine N-oxide (TMAO)(19), and this TMAO is said to disrupt insulin action in the liver. And indeed: people with T2D have higher average TMAO levels in their blood. So far, so suspicious.
My first instinct was owed to the limited knowledge of TMAO I had until I did the research. Everything I knew about TMAO until recently was this: how much TMAO actually ends up in the blood after a given amount of meat depends heavily on the individual composition of the gut flora - I checked that again and looked out a reference(20) confirming what I already knew. That relationship is considerably stronger than any dependence on the amount of meat itself. So as a measure of “how much meat does this person eat”, TMAO is of limited use anyway.
In the course of further research it did get biochemical after all, and I will take the opportunity to describe an approach that I had not consciously come across before either:
Jia et al. investigated this in both directions in 2019(21). The result: genetically higher TMAO levels are not associated with a higher T2D risk. The reverse direction, however, is - diabetes and kidney disease drive TMAO levels up, not the other way round.
For that direction there is even a cleanly described mechanism: the enzyme FMO3, which oxidises TMA to TMAO in the liver, is inhibited by insulin(22). With insulin resistance that inhibition falls away, FMO3 ramps up - and TMAO levels rise. So the pattern is explained not just statistically but biochemically as well: on this reading, TMAO is not the arsonist, but the smoke.
And now comes the part I did not expect when I looked it up. I had counted on a recent review defending the TMAO hypothesis - but the review by Jaworska et al. from May 2025(23) is the fairest summary anyone could wish for. It does list the incriminating findings: in mice on a high-fat diet TMAO worsens glucose tolerance, disrupts hepatic insulin signalling there, and in cell cultures it lowers glucose-stimulated insulin secretion. But it discusses Jia et al. and the FMO3 mechanism just as thoroughly - and arrives in the end at the conclusion that TMAO is possibly no more than an indicator of hepatic insulin resistance, and that there is currently insufficient evidence to classify TMAO as a direct disease-causing factor in diabetes.
So the mechanism here is not invented - but the arrow of causality points quite clearly in the other direction. The method, Mendelian randomisation, is worth remembering though, because we will need it again in a moment.
BCAAs: The Arrow Never Pointed at the Plate
The branched-chain amino acids (BCAAs) - leucine, isoleucine and valine - are Essential nutrients are those that the body cannot synthesise itself and must obtain through food β vitamins, essential amino acids, and fatty acids.: the body cannot produce them itself, so they have to be eaten. Since a much-cited study from 2009(24) it has been known that people with insulin resistance have elevated BCAA levels in their blood. The proposed pathway: BCAAs over-activate a signalling pathway called mTORC1, which in turn disrupts insulin action.
Unlike with TMAO, I cannot deliver a reversal here, but I can shed light on other aspects: the evidence is, first of all, undecided.
In 2016 a study appeared(25) that likewise carried out a Mendelian randomisation. That study arrives at the result that genetically higher BCAA levels are indeed associated with a higher T2D risk - and markedly so: odds ratios from 1.44 (isoleucine) to 1.85 (leucine). Their conclusion, in the original: the findings are “consistent with a causal role of BCAA metabolism in the aetiology of type 2 diabetes”.
A considerably larger piece of work from 2025(26) - around 250,000 participants from the UK Biobank - arrives at the opposite picture: there, elevated BCAA levels appear primarily as a consequence of poor metabolic health, that is, of higher BMI, more triglycerides, lower HDL, stronger insulin resistance and higher T2D risk. In the other direction, BCAAs there do demonstrably cause a lipid metabolism disorder, but independently of T2D risk. The conclusion: BCAAs are above all “a biomarker reflecting metabolic health”.
Two Mendelian randomisations, two directions. Anyone now expecting me to pick whichever suits me will be disappointed - because for the question at issue here, that dispute is not decisive at all.
Because if you look up what Lotta et al. 2016 actually measured, you will find no steak. The genetic variant that carries most of the effect sits in the PPM1K gene and means that the body is less able to break BCAAs down. What is associated with diabetes here is therefore a defective breakdown pathway - not an increased intake. “My body does not break leucine down properly” and “I ate a steak” are two very different statements. The study demonstrates the first. About the second it says nothing.
And even if it did: BCAAs are the least specific candidate on this entire list. They are in every protein - in dairy products, in eggs, in pulses, in soya. Anyone who considers BCAAs a driver of diabetes would consistently have to warn against lentils and tofu just as much as against fillet steak. So as an argument against meat this pathway would not work either, even if it were biologically correct.
This section therefore ends differently from the previous one: not with a reversed arrow, but with an arrow that never pointed at the plate. And at least with the second application of Mendelian randomisation - and that in direct succession. A good haul, considering this is the first time I have consciously come across it.
Does Meat Cause Diabetes?
Researching the biochemical processes was fascinating and drawn-out, but in hindsight I am glad after all that I went down this long stony road - and, on the other hand, once again shocked at what gets waved through in nutritional science, and what then circulates unchallenged in public.
In summary, a statistically irrelevant result is being talked up because it points in the “right direction”, the one that was meant to be demonstrated. The biochemical mechanisms brought in for further explanation are equally poorly supported, refuted, or point in the exact opposite direction of causality.
So I come out of this article with a clear answer, for myself at least.
Does meat cause diabetes? I could not find anything that convinces me of it even remotely. On the contrary: I found plausible reports of successful treatment and remission of T2D through strict carbohydrate restriction and a considerable increase in meat consumption.
