top of page

EPIC-Oxford and the risk of bone fracture in vegans

Updated: May 21, 2023

When I set up U-Turn Health I deleted all my previous websites (in total there were six) and most of my social media accounts. In the aftermath of Covid and a few other things that were going on for me at the same time I felt like I needed a fresh online start, and that I had left behind me over the course of about a decade a trail of internet litter it was my responsibility to clean up. There were some posts it was worth saving however, and this was one of them. In the 'Memory Lane' category I will be revisiting some of my old posts and finding out whether they still carry weight or whether the 'new me' has learnt a few things since then. I originally posted this one on November 29th 2020, and it is one I think is worth re-posting exactly as it was...

Okay this is a long read, and not for the faint-hearted, but I promise you, if you get to the end, and understand what I'm saying, it could change your life.

The EPIC-Oxford paper (Tong et al., 2020) on vegetarian and vegan diets and the risks of total and site-specific fractures is all over Instagram right now, with various plant-based doctors posting advice on bone health and the importance of exercise, vitamin D supplementation, etc. It was produced by members of the Cancer Epidemiology Unit in the Nuffield Department of Population Health at the University of Oxford, and the Centre for Exercise, Nutrition and Health Sciences in the School for Policy Studies at the University of Bristol.

Yes, bone health is a complex and not that well understood phenomenon: there are lots of different factors at play. And yes, vegans, like everyone else, can take steps to safeguard their bone health. But I’m not sure we all need to start walking around wearing weighted jackets to make up for the fact that we are not overweight or obese. A vitamin D supplement is probably prudent (I take one), if you have dark skin and/or live in more temperate latitudes, but it’s certainly not essential for the entire population of the world. Human beings have been getting by reasonably successfully for a long time without them.

This article was published on 23rd November 2020 and it presents data from the EPIC-Oxford cohort study which collected dietary information from participants resident in England, Scotland or Wales at baseline between 1993 and 2001, and then again in/around 2010.

EPIC-Oxford is a large prospective observational epidemiological study looking at the relationships between diet and later disease. This paper looked specifically at the relationship between diet-group (meat-eaters, fish-eaters, vegetarians and vegans) and total and site-specific fractures (arm, wrist, hip, leg, ankle, other: clavicle, rib or vertebra) over a follow-up period of 17.6 years.

What were the results? Well, the study reported that the vegans had a statistically significant increase in the risk of hip (HR 2.31; 1.66-3.22) and total (HR 1.43; 1.20-1.70) fractures compared to the meat-eaters, and The Times (Blakely, 2020) reported that vegans are ‘40% more likely to suffer a bone fracture’.

Is this something that as a vegan/raw vegan community we need to take seriously? Well it might be, but I always read papers like this these days with my critical hat on. This was a big, well-designed study, but already there are a number of problems leaping out at me.

Firstly, the relative numbers of vegans compared to the other three groups. The study followed 29,380 meat-eaters, 8037 fish-eaters, 15,499 vegetarians, and only 1982 vegans. That means that it had relatively less power to detect a statistically significant difference for the vegan group, and this is reflected in the wider confidence intervals apparent in Figure 1. Although there is certainly a trend towards a greater risk of fracture among vegetarians and vegans, it’s worth noting that authors always highlight what is significant, and not what isn’t. They didn’t say in their abstract that vegans had a non-statistically-significant increase in fractures of the arm, wrist, leg, and other main sites, and no appreciable difference in the rate of ankle fractures, and although I understand that this does not constitute headline news, an unbiased report should recount the facts as they are. Any other rendition constitutes not objective scientific truth, but an author’s narrative that tells the story the author wants to tell.

It’s also worth noting up-front that the vegans constituted only 3.6% of the total number of participants. Anyone who is willing to voluntarily place themselves in a 3.6% minority group because they object to the mass slaughter of animals for food (this is the definition of a vegan, as opposed to someone who eats a plant-based diet) has personality characteristics that set them apart from the majority that may also (and in a variety of ways) impact on their propensity to break bones, and those may be difficult to control for: they may be more independent, more defiant, more risk-taking, and more angry. The study did not record the cause of the fractures (for example whether they were considered pathological or traumatic), or the propensity of each group to engage in dangerous sports.

Secondly, everyone broke bones. It’s important to note that none of the diets studied prevented fractures. Total fractures were 2468 (8.4%) in the meat-eaters, 464 (5.8%) in the fish-eaters, 862 (5.6%) in the vegetarians, and 147 (7.4%) in the vegans.

Hang on a minute. 8.4% of the meat eaters sustained fractures, while 7.4% of the vegans sustained fractures. But the hazard ratio (HR) for total fractures for vegans compared to meat-eaters is given as 1.43 (95% confidence interval 1.20, 1.70, p<0.001). How does that work?

The hazard ratio is the ratio of the chance of an event occurring in one group divided by the chance of that event occurring in another group over a specified period of time (Toledo, 2018). It’s the ratio of the risk of an event occurring in one group compared to the risk of that event occurring in another group.

The range 1.20-1.70 is known as a confidence interval. It’s the range within which we can be 95% confident that the true value lies. If the confidence interval for vegans does not overlap with the confidence interval for meat eaters, we can be reasonably, though not absolutely, confident that there is a true difference: if we did this study 100 times in the same population we would find a difference of at least this magnitude 95% of the time.

Unfortunately we can’t find out if the confidence intervals overlap because since the meat eaters have been taken as a reference group, no confidence intervals have been given for them. Their hazard ratios are given as 1.00 (i.e. no difference). This reminds me a bit of Einstein’s theory of relativity. Movement is always measured in relation to another, usually static, body. We understand that things fall downwards in relation to the earth, but in fact movement is always relative: determining whether movement is occurring or not depends on the presence of a static (or more slowly moving) reference object. If you take away this object, there is no movement.

In relation to this study, the correct reference object is the hypothetical healthy population with zero fractures. We should be measuring all diet groups against this yardstick, and only then can we make a valid comparison between groups, none of which is given the privileged position of normality. We need to begin with the null hypothesis of no difference between groups, and the size of each group is irrelevant here except to the extent that it affects the size of the confidence intervals. This is a major design flaw in this study that introduces bias from the very beginning, and demonstrates that the authors do not understand the basics of the scientific method.

So now we have to so some maths. Actually the vegans had a lower percentage of any kind of fracture than meat-eaters, but the study reported in its abstract that ‘the vegans also had higher risks of total (1.43; 1.20-1.70) fractures […] than meat eaters’. So what we need to understand here is how we got from a lower absolute rate of total fractures to The Times’ statement of a 40% increase in risk.

Okay first of all let’s do the calculations again. Just to be sure. The number cases of any kind of fracture in meat eaters was 2468, and the number of meat eaters was 29,380. 2468/29,380 = 0.084 = 8.4%. The number of cases of any kind of fracture in vegans was 147, and the number of vegans was 1,982. 147/1,982 = 0.074 = 7.4%.

The absolute unadjusted risk of any fracture in vegans compared to the risk of any fracture in meat eaters for this study group was therefore 7.4/8.4 = 0.88.

Let’s just repeat that very clearly: the absolute unadjusted risk ratio for any fracture for vegans compared to meat eaters was 0.88 (where <1.0 indicates a lower risk). But the authors reported an adjusted HR of 1.43. How did they do this?

Well first we have to do these calculations again with person-years rather than number of persons as the denominator. Then we have to calculate something called a hazard ratio (HR), which is basically a risk ratio taking into account a particular period of time. The number of person-years will be different depending on the length of follow-up for each person. Loss to follow-up will reduce person-years, and differences in rates of loss to follow-up between groups introduces another source of potential bias into our calculations because there may be reasons why one group did not engage with the study as much as another group. For example if vegans did not feel respected or represented by the researchers, or if they were very ill, or if they had a greater propensity to emigrate to the tropics, they might not come back.

Years of follow-up per participant were 33,510/1,982 = 16.9 for vegans and 511,459/29,380 = 17.4 for meat eaters. This difference therefore accounts for the fact that the hazard ratio is slightly higher than the absolute risk ratio, but we have no way of knowing whether the vegans who were lost to follow-up suffered fractures during that time or not:

Vegans = 147/33,510 = 0.00438

Meat eaters = 2468/511,459 = 0.00482

HR = 0.00438/0.00482 = 0.91

This is interesting because nowhere in the article is either a risk ratio of 0.88, or a hazard ratio of 0.91, reported. What we are given is the number of participants in each group, the number of ‘cases’ of fracture, and the number of person-years. Then next to these in Figure 1 we are given the adjusted hazard ratio of 1.43, and underneath Figure 1 we are given the statement:

Compared with meat eaters […] vegans (1.50; 1.26,1.78) had higher risks of total fractures after adjustment for confounders (Table 2 model 1). The associations attenuated with additional adjustment of BMI […] but remained clearly significant in vegans.

Why were we not given the unadjusted ratios? Were the authors hoping no-one would do the calculations for themselves and would simply take their word for it? Because if they had reported those they would have had to account for why their adjustment process made the risk so much higher.

Well, to be fair, adjusting for age could potentially account for a lot of this difference, because on average the vegans were younger than the meat eaters. The average age of the vegans was 38.9, and the average age of the meat eaters was 50.1. So next let’s look at the adjustments they did and the effects those adjustments had on the results. Table 2 gives the adjusted risks for total and site-specific fractures by diet group. Helpfully, it does not give an unadjusted hazard ratio for comparison (can you detect a small note of sarcasm here…??). To be fair again, they have adjusted for a lot of known confounders. Confounders are things that also affect the outcome of interest, other than the thing being studied, that if unrecognised can make it appear that an association exists when in fact one does not, because the confounders are also correlated with the thing being studied. Known confounders can be adjusted for: unknown confounders can’t. To be confident that the adjusted ratios accurately reflect reality we need to be confident in the adjustment process.

The authors used five different models for adjustment. Model 1 adjusted for a wide range of variables: sex, ethnicity, region, method and year of recruitment, socio-economic deprivation, education, physical activity, smoking, alcohol consumption, dietary supplement use, height, menopausal status, hormone replacement therapy, and parity (for women, how many babies you’ve had). Model 2 added in body mass index (BMI) (vegans tend to be lighter than vegetarians, who are lighter than meat eaters, and a lower BMI increases fracture risk whatever kind of diet you are on), model 3 added in dietary calcium intake, model 4 added in dietary protein intake but excluded dietary calcium intake, and model 5 added in both dietary calcium and dietary protein.

The HRs for total fractures in vegans were 0.91 (unadjusted), 1.50 (model 1), 1.43 (model 2), 1.31 (model 3) 1.39 (model 4), and 1.30 (model 5). This is an interesting pattern, and also interesting is the fact that the authors chose to report model 2 as their headline news. Why did they choose to discount their own adjustment for calcium and protein intake, after they had done it? The correct scientific method is to design your study in advance, then to do it, then to report your findings exactly as they are. You do not do an analysis and then not report it (or only report it in the small print).

Table 1. shows the baseline characteristics of the different groups. This is interesting: compared to the meat eaters, the vegans had a higher % who had received higher education (42.8% compared to 30.8%), but compared to all groups had the lowest % in the top socio-economic quartile (17.6%, compared to 27.4% of the meat eaters, 21.9% of the fish eaters, and 21.7% of the vegetarians). The vegans also had the highest % of alcohol consumption (13.5%, compared to 12.6%, 12.2%, and 12.6%).

What does this tell us? It tells us that for whatever reason, this group of vegans wasn’t doing so well. Why might intelligent vegans be doing badly in the socio-economic stakes? Perhaps they were choosing to work for charities or not-for-profit organisations because of their ethics? That’s a possibility. Also possible is that vegans at the time of this study were systematically excluded from, and discriminated against by, mainstream society.

If I’m right, I suggest that the effects of this are far-reaching, and impossible to fully adjust for. Adjusting for socio-economic status, BMI and alcohol consumption can only go so far.

The table of baseline characteristics also does not include an analysis of pre-enrolment medical diagnoses, but it is worth noting here that one of the reasons people become vegan is ‘for their health’. That means they have pre-existing medical diagnoses they are trying to resolve. This was not a population of lifelong vegans: the article reports that 39% of vegans had been vegan for less than five years, and 18% for less than three years. Someone who has eaten a terrible diet for their whole life and become sick, and then been vegan for the past year in an attempt to rectify the situation is not at all the same kind of animal as someone who has eaten a calorically adequate, supplemented, whole food vegan diet high in fresh fruits and vegetables since they were weaned.

Okay let’s do the same analysis for hip fractures. For the risk ratio (or relative risk) (RR):

Meat eaters = 227/29,380 = 0.0077 = 0.8%

Vegans = 41/1,982 = 0.0206 = 2.1%

RR = 2.1/0.8 = 2.63

And for the HR:

Meat eaters = 227/525,706 = 0.04%

Vegans = 41/34,489 = 0.12%

HR = 0.12/0.04 = 3.00

There does appear to be a difference here that must be accounted for. But bear in mind that this is based on absolute values for the groups overall. These percentages are before any adjustment for body mass index, socio-economic variables, or any other possible confounding factor. Is a 1.2% absolute difference between groups important? Is it a real difference, or could it have been due to chance? If you repeated this study a hundred times, would you get the same result each time, or would sometimes the vegans do better and sometimes the meat eaters, with no difference when you averaged things out overall?

Here the authors reported a lower adjusted HR of 2.31 (1.66-3.22) (model 2). Since the confidence interval does not include one (and without getting into the mathematics of how you calculate confidence intervals or p-values), this is likely to be a real difference, assuming all significant confounders have been adjusted for. But it’s not quite clear to me how adjusting the absolute ratio for the same set of confounders might result in a higher HR for total fractures but a lower HR for hip fractures (you will remember that for total fractures the HR went up from 0.91 unadjusted to 1.43 adjusted). The same adjustment should have the same effect for both.

I also note that the number of person-years given for total fractures in Figure 1 (33,510) is lower than that given for all other types of fracture (for hip fracture, 34,489), and I’m not clear either why following up a certain group of individuals for a set period of time, and recording the number of fractures of any sort that they have during that time, and then pooling the results to give an overall total, should result in a difference in the number of person-years of follow-up unless certain persons and/or certain periods have been excluded from the total fractures analysis.

I also can’t work out whether the number of ‘cases’ reported in Figure 1 represents the number of fractures, or the number of people who suffered fractures. This is an important distinction. Someone involved in a horrible road traffic accident might suffer multiple fractures and none of them would be related to diet, but these would skew the results quite dramatically. Equally one person with osteogenesis imperfecta could account for multiple fractures, but again, none of them would be related to diet and again these would skew the results. If you know the answer to this then please do let me know.

One of the really big problems I see with this study is that it correlates fracture risk with dietary category, not with diet. It’s a thing I’ve talked about before and will undoubtedly talk about again, but there are many different kinds of vegan diet, and a raw vegan diet that’s comprised of between 20 and 30 portions of raw fruits and vegetables per day, including plenty of leafy greens, is not at all the same diet as the chip butty diet, the vegan cheese on white toast with a serving of spaghetti hoops on the side diet, the vegan doughnut diet or the vegan bangers and mash diet.

The EPIC-Oxford study didn’t report on the specific diets being consumed, and it’s a mistake to think that the label ‘vegan’ tells you very much about dietary quality. It’s quite possible for both a vegan diet and an omnivorous diet to be either very high or very low in fruits and vegetables for example. Which leaves me (as a raw vegan) not at all clear whether these results apply to me at all.

Another unacknowledged problem, which is a hard thing to talk about as individuals, and a hard thing for the vegan community to acknowledge overall, is the prevalence of eating disorders within the vegan community. But the fact is that as long as veganism is seen as a restrictive diet, there will always be a proportion of vegans who are vegan not for ethical or health reasons, but for the opportunity it presents to further restrict their dietary intake. Particularly in places where there is generally no vegan food to be had, those who identify as vegan will be left nibbling on a few pieces of wilted lettuce while everyone else in their party chows down on steak and chips deep-fried in beef dripping.

Can you tell I’ve been in this situation? It still applies to large sections of the country even now. Fortunately I’ve never suffered from an eating disorder, but I have worked with young people who do and identifying as vegetarian or vegan is not uncommon in this group.

What this means is that unless you ask about lifetime restrictive eating patterns specifically (because bone health is affected not only by what we’ve eaten in the past few years, but by what we’ve eaten throughout our lives), the closest proxy you have is BMI at enrolment, which only tells part of the story, particularly for those with a tendency towards bulimia and significant fluctuations in weight over time.

It’s worth asking who the authors are (as people: I don’t mean their job title), and why they might have decided to carry out this analysis. Did they have a particular reason for asking this question of the data? Did they have a particular point they wanted to prove? Why, in Figure 1, did the authors only report p-values for the difference between vegans and meat eaters, and not for fish eaters or vegetarians? The only reason I can think of is that they were trying to make a point about vegans, specifically.

I’m going to end by saying something about my own bias. Of course, as a vegan, I want these results to be wrong (as much as every meat-eater seeking justification for his or her continued participation in mass slaughter wants them to be right). But they might be right, and we might just have to swallow that if so.

But there is something else going on here for me. These results fundamentally contradict my own experience. I have never (yet) broken a bone in my life. Last year I fell onto my outstretched hand (what is known in the A&E trade as a ‘FOOSH’) on the ice rink and ended up in the walk-in-centre having my wrist x-rayed because I was absolutely certain I had broken it. I hadn’t, but it was very bruised and swollen and so unbelievably painful when I went down that I almost passed out. A very helpful nurse skated over and said I’d gone white as a sheet and made me lie down while she held my skates up over my head!

This is an anecdote but since I became a raw vegan I have definitely noticed a change in my nails. I know nails aren’t the same as bones but I can’t quite believe that a diet that makes my nails as hard and sharp as razor blades will make my bones fragile and weak. Every health problem I had has gone away on a raw vegan diet. My skin is clear, my mind is clear: I can run for miles and at 44 I can still do the splits.

So what this tells me is that there is something wrong with this study. The data is what it is, but the way it’s being interpreted and reported is a serious problem.

Let’s look for a moment at how this study is being reported in the mainstream media. A New Scientist (Wilson, 2020) headline says ‘Meat-free diets linked with breaking bones’. Although later in the article it does acknowledge that ‘the overall level of risk to vegans was relatively small, equating to about an extra 20 bones broken per 1000 people over 10 years’.

The Independent (Singh, 2020) says that ‘those who follow a vegan diet are at a 43 per cent higher risk of bone fracture compared to those who eat meat’, but Singh does at least include in her title the words ‘study suggests’. On the other hand The Times (Blakely, 2020) reports that ‘vegans who forgo all foods derived from animals have a far higher risk of broken bones than people who eat meat and fish’.

The attack on vegans continues. The authors of EPIC-Oxford declare they have no competing interests, but as is usual for authors of this kind of study, they do not disclose their own dietary habits or the biases that always come along with any dietary style (if you eat food, you are biased, period). One can assume they identify with their reference group (meat eaters), otherwise they wouldn’t have made this their reference group. If I was doing this study from my perspective as a vegan, to learn something useful for vegans rather than to wage yet another war on an already vulnerable minority, I would have made the vegans the reference group. No I wouldn’t actually, because I would have started with the null hypothesis of no difference, but this alone speaks volumes about bias: the majority is not always right.


BLAKELY, R., 2020. Vegans 40% more likely to suffer a bone fracture [online]. 23 November. Available from: [Accessed 28 November 2020].

SINGH, N., 2020. Vegans 43% more likely to suffer bone fractures, study suggests [online]. Available from: [Accessed 28 November 2020].

TOLEDO, E., 2018. Hazard Ratio [online]. Available from: [Accessed 28 November 2020].

TONG, T.Y.N., APPLEBY, P.N., and ARMSTRONG, M.E.G. et al.. Vegetarian and vegan diets and risks of total and site-specific fractures: results from the prospective EPIC-Oxford study. BMC Med. 18, 353. Available from: [Accessed 28 November 2020].

WILSON, C., 2020. Meat-free diets linked with greater risk of breaking bones [online]. 23 November. Available from: [Accessed 28 November 2020].

19 views0 comments


bottom of page