Mortality Rates in Adventist Health Study-2
I have updated the article Disease Rates of Vegetarians and Vegans with the results from a report on vegetarian mortality rates that was released this week from the Adventist Health Study-2. I have reproduced the highlights below.
There was also an article on this study published in the Wall Street Journal, Vegetarians Live Longer Than Meat-Eaters, Study Finds.
In 2013, death rates for the first 5.8 years of Adventist Health Study-2 (AHS2) were released (1). When combining vegans, lacto-ovo-vegetarians, pesco-vegetarians, and semi-vegetarians into one group, vegetarians had a 12% lower risk of mortality. Vegans had a 15% lower risk of death, but it was not quite statistically significant.
The difference in mortality rates can mostly be explained by a lower incidence of cardiovascular disease among vegetarian men. Vegetarian women had about the same rates as non-vegetarian women. This is similar to the findings from the first Adventist Health Study. There was also a benefit for all vegetarians for death from renal and endocrine (mostly diabetes) disease.
The researchers said that having only 5.8 years of follow-up would bias the results towards not finding differences.
In comparing their findings to British vegetarians, they said:
“The lack of similar findings in British vegetarians remains interesting, and this difference deserves careful study. In both cohorts, the non-vegetarians are a relatively healthy reference group. In both studies, the nutrient profiles of vegetarians differ in important ways from those of non-vegetarians, with vegetarians (especially vegans) consuming less saturated fat and more fiber. It appears that British vegetarians and US Adventist vegetarians eat somewhat differently. For instance, the vegetarians in our study consume more fiber and vitamin C than those of the EPIC-Oxford cohort: mean dietary fiber in EPIC-Oxford vegans was 27.7 g/d in men and 26.4 g/d in women compared with 45.6 g/d in men and 47.3 g/d in women in AHS-2 vegans; mean vitamin C in EPIC-Oxford vegans was 125 mg/d in men and 143 mg/d in women compared with 224 mg/d in men and 250 mg/d in women in AHS-2 vegans. Individuals electing vegetarian diets for ethical or environmental reasons may eat differently from those who choose vegetarian diets primarily for reasons of perceived superiority for health promotion. We believe that perceived healthfulness of vegetarian diets may be a major motivator of Adventist vegetarians.”
Make sure you eat your fiber!
1. Orlich MJ, Singh P, Sabaté J, et al. Vegetarian Dietary Patterns and Mortality in Adventist Health Study 2. JAMA Intern Med. 2013;():1-8. doi:10.1001/jamainternmed.2013.6473. | link
June 4th, 2013 at 8:53 pm
The difference between Adventists and the British I would guess is:
Health and nutrition is a huge part of Adventist culture. I’d bet they know way more about nutrition than the average person.
Secondly, it could be better access to produce. I haven’t looked at the data on where most of the Adventists in the AHS2 live, but it could tilt toward Southern California. Sure, living in So Cal doesn’t automatically mean one eats healthier, but when you combine with solid health educations it would be much easier to act on the knowledge than say Northern England!
June 4th, 2013 at 9:28 pm
Hi Jack
Did the paper listed the unadjusted hazard ratio? If so, can you list these ratios for the different groups (vegetarian, lacto-v., pesco-v., vegan)?
Thanks
June 4th, 2013 at 9:37 pm
Dave,
They didn’t list the unadjusted ratios.
June 5th, 2013 at 2:03 am
Vegan males had 0.72 mortality of omnivores (HR). I find this pretty impressive, especially considering the short follow-up. Moreover, only one dietary survey at the baseline. Vegan diet harder to stick with, hence some people may changed their diet; participants in other sub-groups may have switched to a vegan diet at the face of illness. All these could have potentially drove the findings towards null.
June 5th, 2013 at 5:24 am
Matt makes a great point — the Adventists are being raised in a very veg-friendly and veg-nutrition-informed community. If I had to bet, I’d bet that is the key.
And Dave asks the same question I had. It could well be that fish-eaters had the lowest mortality.
Pretty telling that every vegetarian in the US is touting this Adventist study, when they all studiously ignored the EPIC study.
June 5th, 2013 at 6:14 am
This was a fascinating study but I hate to be a wet blanket. Just a glance at Table 1 and the first paragraph of the results tells you how much *measured* confounding there is. All the statistical measurement in the world is not going to adjust for the unmeasured ‘care and caution’ and health-consciousness that goes with the vegetarian lifestyle – buckling your seat belt, driving slowly, not gambling, not having illicit dangerous unprotected sex, not snorting cocaine, etc. Ok, so 7th day adventists don’t do any of those latter things (right?), but you get my point. I would say it is extremely optimistic to believe that even 10% of total confounding is measured and appropriately adjusted for (look how they dichotomized their MEASURED variables! ugh! the information bias!). Again observational epidemiology is highly suspect.
Question for the webmaster. I am trying to figure out through your lovely application peacounter why I am getting angular stomatitis (cheilitis). I had this once before on a very poor diet but am getting it now and am supplementing my lacto-vegan diet with B12, omega-3, iodine and D. I am a low carb lacto vegan and do not consume much fruit (other than one small tomato and two avocados per day), and eat two low carb wraps with dinner (so perhaps not much riboflavin?). In my review of the literature this could be zinc, B12, riboflavin or something else I may be missing (vitamin C? iron?).
Pretty much eat the same thing every day:
For breakfast – 2 avocados, 250 mL yogurt, 7 types of nuts and seeds mixed into the yogurt, 1 tablespoon of macadamia nut butter, 20 lupin beans
For lunch – 5-vegetable salad (that small tomato plus green pepper, radishes, green onions, small ‘dill type’ cucumbers drenched in tahini and lemon juice) plus either a veggie burger or 2 tofurkey sausages and a honey-mustard aioli (olive oil and egg-based)
For dinner – 2 tortillas (low carb wraps) with sauteed tofu, orange pepper, broccoli, greek yogurt, often with a ramikin of the nuts/seeds mix on the side
I drink only water. Something missing here is starting to give me angular stomatitis (perleche; cheilitis), and I can’t seem to figure out from peacounter what the missing nutrient is. I wonder if taking 2 multivitamins a day on top of all the rest (DHA 300 mg, EPA 600 mg, iodine 150-650 mcg, D3 1000 IU, B12 100 mcg) would fix it. Or add a B vitamin complex? Or more fruit (maybe some berries)?
It is surprising to me also because the tofu products at lunch are said to be enriched with B vitamins and iron. Could the issue be Zinc?
I am totally confused. Please help! Thanks.
June 5th, 2013 at 10:28 am
Dan,
> This was a fascinating study but I hate to be a wet blanket. Just a glance at Table 1 and the first paragraph of the results tells you how much *measured* confounding there is.
Do you mean because of all the various foods that have been associated with reduced mortality?
> buckling your seat belt, driving slowly, not gambling, not having illicit dangerous unprotected sex, not snorting cocaine, etc.
Unnatural deaths were not included in the study.
> I would say it is extremely optimistic to believe that even 10% of total confounding is measured and appropriately adjusted for (look how they dichotomized their MEASURED variables! ugh! the information bias!).
“even 10%”?! That’s a pretty bold statement.
> Again observational epidemiology is highly suspect.
I’ve been hearing lately that observational studies should be considered to be 10% off. Not sure what that means exactly, but I think it means the 95% confidence interval needs to fall outside of .90 to 1.1 before a finding is considered true. I haven’t seen any author of any observational study mention this, but my experience is that it seems like a good rule of thumb. I tend not to “write home” about a finding until the CIs are out of that range.
Hmm. Getting above 1.1 is a lot easier than getting below .90, since you only have 90 points between 0 and 90 but an infinite amount above 1.1. Am I right about that? Maybe all the findings above 1.1 need to be flipped to see if they are below .90?
I hope I’m making sense – I understand what I’m saying. 🙂 And I realize there could be exceptions to this 10% rule.
June 5th, 2013 at 10:29 am
> before a finding is considered true.
Oops. I didn’t mean before a correlation could be considered causal, I meant before a correlation can even be considered a correlation.
June 5th, 2013 at 6:46 am
Just went back to your lovely peacounter application.
I think it is either choline or vitamin A deficiency. The latter would make sense to me, because I have had an odd foreign body sensation in my right eye for a couple days, to go along with the cheilitis.
I guess I will be chomping down on some carrots with my high fat lunch (it seems 2 large guys per day would do the trick).
I have never heard of vitamin A deficiency being associated with angular stomatitis/cheilitis/perleche.
June 5th, 2013 at 8:13 am
Dan,
It would be hard for me to say, and it might not be nutrition related. I have also suffered from angular stomatitis, off and on my entire life (before I was vegan). I haven’t had it much lately. It is one of the reasons I take a vitamin B complex rather than just a B12-only supplement. But I really haven’t been able to clearly determine that the B-complex helps; I’m skeptical that it does. I also have been taking more zinc and more vitamin A (since you mentioned it) in the last few months. But the weather has been warmer, too, and I can’t say I’ve done a good job tracking the angular stomatitis to really pinpoint when I have it and when I don’t. In any case, I don’t have any sign of it now.
I’m actually more concerned about your eye. I wouldn’t let that go for very long without seeing a doctor.
June 5th, 2013 at 8:25 am
Thanks Jack. That is helpful.
In my case, I wonder if sun exposure is actually a contributing rather than preventive factor – i.e. an acutely photosensitizing effect. I’ve been sitting outside in the sun ALOT lately, which has just come to southern Canada where I live.
A search through MEDLINE did not find any reports on an association between vitamin A deficiency and cheilitis, but I am asking the library to pull one possible article (unlikely possibility) from Bulgaria, circa 1976. I do not seem to have diminished night vision.
Nutritional deficiencies often produce nonspecific or odd harbingers in the body. We vegans ignore them at our own peril. I think I am going to pick up some vitamin B-complex. I’ve been reluctant because a large trial of folic acid supplementation (I believe for the prevention of colon cancer) showed a significant increase in prostate cancer, which I have a strong family history of. The B-complex typically contains folic acid. But given that I eat very little grain, it probably wouldn’t give me too much. I wonder if I should be adding zinc too. Definitely 2 large carrots, as the peacounter indicates I am very deficient on vitamin A (getting only 7% of RDA. (It also says I am getting no B12 or vitamin D – but I supplement with those). I will get some B-complex, eat more carrots and if it does not disappear, consider adding zinc as a supplement.
The last time I had cheilitis (perleche) was 17 years ago when I was eating only spaghetti every day while living in a dorm out of town during a summer studentship in a research lab. The fridge was communally shared and I did not trust the facilities for cooking much else; plus I was a dumb student (still am, in many respects).
Interestingly, the peacounter does not indicate any deficiency in B vitamins for me (other than B12, which I’m somewhat certain I can absorb well). It indicates a paucity of choline, B12, vitamin A, and vitamin D. Odd.
June 5th, 2013 at 8:28 am
There are riboflavin-only supplements available the last I checked, if you want to avoid the folic acid. Good luck! Keep us posted.
June 5th, 2013 at 9:22 am
Jack, thanks. I will look into that.
Dave, they didn’t list unadjusted risk ratios, but they can be calculated by simple division from Table 3 (specifically the column of age-sex-race adjusted mortality rates towards the right-hand-side). Simply divide each group’s mortality rate by the non-vegetarian mortality rate (6.61). This will give you each group’s all-cause mortality risk ratio. Unless it’s cause-specific mortality risk ratios that you are after, in which case you’d have to contact the authors for these.
June 5th, 2013 at 9:37 am
I’m proud to be part of this ongoing study! 🙂 (Even though I’m not Adventist, but I did attend a local church for many years.)
June 5th, 2013 at 3:47 pm
Jack, the conventional rule of thumb is that a factor is an important prognosticator (confounder) in a model if it moves the point estimate for the exposure variable by at least 10% (in relative terms). So, for example, let’s say an observational study shows vegetarian diets reduce mortality with an unadjusted hazard ratio of 0.80. The authors appropriately adjusted for, say, smoking (which tends to be imbalanced between veg and non-veg), and the hazard ratio moved up or down by 10% (so in absolute terms by 0.08). In this case, we can say that smoking was an important confounder and should be included in the model. I think that is the 10% rule you are referring to, but I may well be wrong.
I did a PhD in clinical epidemiology and I am medically trained. I have seen so many disconnects between observational studies and definitive randomized trials that I have become extremely skeptical of all of these data. (just look at homocysteine, or vitamin E, or HRT) To take one example, the Nurses’ Health Study and multiple other studies showed that people who took vitamin E had lower rates of all sorts of bad things, and then when the trials were done, there was actually increased rates of death (at doses >400 IU od), hemorrhagic stroke (at lower doses) and heart failure (at lower doses). The problem is that of healthy user bias. I don’t think you can get more healthy user bias than a vegetarian diet. A vegetarian diet goes with a whole package of things that could conceivably be measured, but almost never are, for logistical reasons.
And regarding your point that they excluded accidental death (to my point about seatbelts), establishing the precise cause of death is always a mug’s game (I got this from someone who sits on the critical events adjudication committee for a large randomized trial). An example would be a car crash that leads to a heart attack that leads to death. I’ve actually seen this scenario clinically (though it’s usually the other way around – first the heart attack while driving, then the crash). So which is the cause of death here? Depends on the coder. It could be car crash, it could be heart disease, or it could be neither. I don’t tend to put alot of weight on cause-specific mortality as these are typically secondary endpoints, capture fewer events, and adjudication as to cause is fraught with issues.
Then there is the issue of how many analyses they have put into the manuscript and which ones they decided to leave out. 1 in 20 will be positive due to chance alone (actually the risk is more like 1 in 11 because of correlated data – ie the endpoints tend to cluster with each other and are not truly independent observations). Multiple hypothesis testing, reporting bias, data dredging absolutely hound scientific research.
Don’t get me wrong. I think vegetarian diets can be quite healthy. It is reassuring that the hazard ratios are not in excess of 1.00. But studies like this, with potential residual/unmeasured confounding (and in particular the healthy user bias) cannot possibly establish causal associations.
As to the risk estimates needing to be above 1.1 or below 0.9, it is not actually easier to get above 1.1 than to get below 0.9. As you said, if you flip the inverse the hazard rate ratio, you will get a number below 1.0 when the original estimate is above 1.0. I think this is more of a power/precision issue (the 95% CI is relatively tight around the hazard estimate and excludes estimates close to 1.0), rather than a truth/validity issue.
June 5th, 2013 at 4:05 pm
Dan,
Thanks for the explanation. Very interesting.
I don’t think that the 10% rule I was talking about is the 10% rule you’re referring to, but more what I had said about the CI not overlapping with .90 to 1.1, but I’m probably not exactly right about the numbers or if it’s the actual ratio rather than the CI.
> As to the risk estimates needing to be above 1.1 or below 0.9, it is not actually easier to get above 1.1 than to get below 0.9. As you said, if you flip the inverse the hazard rate ratio, you will get a number below 1.0 when the original estimate is above 1.0. I think this is more of a power/precision issue (the 95% CI is relatively tight around the hazard estimate and excludes estimates close to 1.0), rather than a truth/validity issue.
Hmm. What I meant was that a finding of, say, .50 is only 50 points below 100. But if you flip it, it’s 2.00, a full 100 points above 100. However, after doing the math, I see that .90 is the flip of 1.11, so it’s almost the same at that distance from 1.00.
June 6th, 2013 at 4:19 am
I don’t understand why the authors didn’t add vitamin C/fiber consumption to the model, to see whether adjusting to these variables explain (some of) the difference in mortality.
June 6th, 2013 at 4:45 am
Dan,
Why in this study (and other studies), some variables in the regression model are grouped into categories and not treated as continuous variables? (e.g. the exercise have categories <= 20, 21-60 etc).
June 6th, 2013 at 7:07 am
I posted a long message but the system would not accept it.
The gist of it was that for most hypothesis-directed scientific experiments, it is not at all the case that there is an equal likelihood of landing on a risk ratio of 0.00 to 1.00 versus 1.00 to 2.00. If that was the case, then yes I would agree with you that numbers above 1.0 would be far more likely. Many phenomena seem to follow a bell curve distribution centered around 1.00, which means that the most common findings in science cluster around the null finding. If it WAS the case that all statistically significant potential risk ratios in science are equally common, then yes I would agree with you that estimates above 1.0 would be more likely to occur (due to their greater numeracy) than estimates below 1.0. But that would assume that random variation is the norm, not the exception.
(picked up B100-complex and 10 carrots. If this doesn’t work, I’ll need to add some vitamin A as a loading dose and maybe some zinc, but it’s already getting better. Odd that I am not night-blind after depriving myself of vitamin A for months. Spoke to my brother, a pediatric general surgeon, who said that nutrition-induced angular stomatitis would be a marker for poor mucosal healing throughout the body (respiratory, gastrointestinal, genitourinary, etc), but I am only ‘seeing’ it on the outside – ie at the mouth. Something to keep in mind, although not all cases of angular stomatitis are nutrition-induced. My 2 cents. For what it’s worth….)
June 6th, 2013 at 10:33 am
Dan,
Glad to hear you’re getting better! Did you mean your eye, your stomatitis, or both?
I think we’ve had a big failure to communicate on what I’m saying regarding the risk ratios. I’ll try one more time (and perhaps I’m wrong and that’s why there is a communication problem). Assuming I’m correct, you can look at any risk ratio two ways. For example, this:
non-vegetarians 1.00
vegetarians .50
Is the same as this:
vegetarians 1.00
non-vegetarians 2.00
And all I was saying is that when you look at it the second way, the distance from 1.00 is greater than when you look at it the first way. So that needs to be considered when one is thinking about the distance from 1.00 that is being achieved with various measurements and how important that might be.
June 6th, 2013 at 8:57 am
I wonder if the amount of adjustment the researchers did to factor out confounders is being overlooked here. Their least adjusted (age, sex, race) comparisons gave a death rate of 5.40 per 1000 for vegans and 6.61 for nonvegetarians. That’s a relative difference of 0.817. After adding smoking, exercise, education, income, marital status, region where they lived and how much sleep they got (!), they arrived at the final figure of 0.85 for all vegans (and 0.72 for male vegans; female vegan data was also adjusted for menopausal status).
That’s a wide net to factor out confounding factors! Should we really be that suspicious, considering that the vegan advantage was on the cusp of significance at the 95% level (0.73-1.01) and the male vegan effect was well within it (0.56-0.92)? Does multivariate adjustment have a big enough bias (10%?) to wipe out a lot of those differences? And let’s not forget that confounding factors can skew the results away from a vegan health effect too. Maybe there’s an angry misfit effect (but probably not among SDAs).
June 6th, 2013 at 12:27 pm
Dave, some studies using categorization for ease of interpretation. It is much easier to interpret the effect of a continuous variable when it is split into ordinal categories. On the other hand, this study isn’t reporting the risk ratios for confounders; it has only used them in the model but not reported the covariates. Therefore, there is some information bias by doing these categorizations instead of preserving the continuous nature of these variables. Maybe they’ve done it the way they did because of a holdover from a previous study, where they actually reported their full model.
Jack, now I’m totally lost. I thought I had understood you previously, and my thinking was that if you took a frequency plot of all scientific studies (or all nutrition studies, etc), then more of the risk ratios would be above 1.00 then below 1.00, because there are only 100 points below 1.00 (to 2 decimal places) but infinite risk ratios above 1.00. In your last example, I agree that 2.00 is the inverse (reciprocal) of 0.5. However, you will notice that most studies do not report symmetric 95% confidence intervals (one side is usually a little bit larger than the other). It IS possible to compute symmetric 95% CI but for some reason (lost on me), authors never do this with their models. In absolute terms 2.00 is double the distance from 1.00 that 0.5 is from 1.00 (2-1 = 2x (1-0.5)). However, in relative terms, they are identical distances apart from unity (1.00).
(I think the eye was just seasonal allergic conjunctivitis resulting in scratching resulting in an irritating corneal abrasion)
Syd, I’m not sure I understand your point either. The fact that the risk ratio went from 0.82 to 0.85 after adjusting for measured confounding means what exactly? Couldn’t there still be a mountain of unmeasured or hidden confounding? Are veg*ns psychologically identical to non-veg*ns? Are they more health conscious in general? Do they check in with their doctors for mammograms, PSA tests, colonoscopies; take more vitamins; drive more carefully; suffer less from depression and anxiety and the sequelae thereof? Yes I believe they do! I’m willing to bet that the veg*ns in this study were much more health-conscious, or perhaps even more ethically-minded, than their carnivorous ‘cousins’. And while you can measure and adjust for some of these things using static baseline measurements on questionnaires and such, there are just far too many prognostically relevant factors that are below the tip of the iceberg. IMO.
June 6th, 2013 at 12:37 pm
Dan,
> I thought I had understood you previously, and my thinking was that if you took a frequency plot of all scientific studies (or all nutrition studies, etc), then more of the risk ratios would be above 1.00 then below 1.00,
Not at all. All findings could be represented as below 1.00 (except for those that are exactly 1.00). There’s no sense in trying to discuss it further.
June 6th, 2013 at 1:14 pm
Dan,
A colleague of mine saw your mention of chelitis and sent me this:
I was just reading your blog and saw someone posted a question about their
chelitis. I had it a couple years ago, figured out something that helped,
and I’ve suggested it to at least 3 other people and it helped within 24
hours. Not sure if you’ want to pass this on to the commentator on your
blog, but I always try to let people know about this (especially veg
people so they don’t blame their diet). I followed these suggestions:
http://healthyliving.msn.com/health-wellness/skin/preventing-cracks-at-the-
corners-of-the-mouth-1
June 6th, 2013 at 1:29 pm
This is extremely helpful! Thank you.
I must say that the last time I had anything resembling cheilitis was 17-18 years ago when I was a young summer student eating nothing but spaghetti and living in a dorm in another city (not my home town). There are a lot of case reports on nutritional deficiency-associated cheilitis and I think it must be somewhat common at the population level. As you know, my diet is a bit odd.
The cheilitis itself was only severe for about 2 days, and is now definitely improving. I still feel it was a mucosal healing defect, due to my weird diet (low carb vegan). But I’m going to keep this advice in mind.
June 6th, 2013 at 2:16 pm
“The fact that the risk ratio went from 0.82 to 0.85 after adjusting for measured confounding means what exactly?”
Dan, it means that even after what are assumed (based on some combination of research findings and conjecture) to be materially significant confounders (not counting age, race and gender, which were already factored out of the first ratio – Table 3 in the study) have been adjusted for, the vegan advantage (surprisingly, to me) only dropped from 18% to 15% (i.e., from 0.82 to 0.85). How much more of a drop can we reasonably expect if, say, frequency of visits to healthcare providers, is included? Some of the other potential confounders you mentioned could, in light of recent research, detract from the vegan advantage (e.g. taking megadoses of some nutrients where studies have counterintuitively found an increased mortality effect). And then there are other possible differences – like the putative “angry vegan misfit” – that would intuitively be expected to reduce the vegan mortality advantage. Others are a question mark. Are vegans – even SDA vegans – less likely to take medically prescribed drugs or undergo medically recommended procedures and more likely to turn to complementary and alternative therapies, and does this increase or decrease their mortality? Are they less prone to mental illness and suicide, as you suspect, or more (vitamin B12 and long chain omega-3 deficiencies both are correlated with mental illness that can lead to death; very low cholesterol is associated with increased suicide risk, and it may be commonplace among fat-phobic vegans).
So, there will always be uncertainty in epidemiologic studies like this, but how reasonable is it to expect that more diligent pursuit of confounding variables will significantly move the hazard ratios in this one?
June 6th, 2013 at 2:22 pm
I think I’m going to have to side with Syd on this one.
June 6th, 2013 at 4:42 pm
There are two possible explanations for the fact that the hazard ratio only shifted from 0.82 to 0.85 after adjustment for the non-age/sex/race confounders:
1) First, it may be that they did a good job of measuring potential confounding variables, and there really weren’t any — the association between vegetarianism and improved survival is truly causal.
2) Second, it may be that there are important potential confounding variables that remain residual in the model, due to either under-adjustment (e.g. lack of modeling interaction effects, or inappropriately dichotomizing continuous variables), mismeasurement of certain variables that may not be static after baseline, or true hidden confounding (unmeasured prognostically important variables that are asymmetrically distributed between groups).
Given the fact there are so many statistically significant differences in baseline characteristics between the groups in such things as smoking status, it is very likely that there are a lot of other variables that exist, and are asymmetrically assorted between groups, but weren’t measured. See the following text:
“Vegetarian groups tended to be older, more highly educated, and more likely to be married, to drink less alcohol, to smoke less, to exercise more, and to be thinner. The proportion of blacks was highest among pesco-vegetarians and lowest in lacto-ovo–vegetarians. Of postmenopausal women, far fewer vegans were receiving hormone therapy.”
YES, they adjusted for these differences. YES, it made the hazard ratio move only from 0.82 to 0.85. But the fact of the matter is that no two persons are even remotely similar to each other. One person could be incredibly slothful, never going to an MD or RD or RN in their life until they get sick with a terminal illness, never undergoing any preventive screening or self-care, never putting on sunscreen when going out in the sun, fat, diabetic (undiagnosed at that!), smoking, alcoholic, eats only potato chips and meat, etc. Another person could be incredibly health-conscious, attend regular MD or RD or RN visits, undergo multiple preventive screening maneuvers that have been shown to reduce cancer mortality, always puts on sunscreen when going out in the sun, thin, non-diabetic, never smoked, drinks green tea, avoids all meat and potato chips.
How many factors in life are associated with the thousand ways that people can die? And within each condition (say Alzheimer’s), which we now know are very heterogeneous entities, how many factors are associated with that condition? Was salt intake measured? Use of sunscreen? Type of exercise (resistance vs aerobic, intermittent vs sustained frequency)? Attendance for an annual physical examination? Use of appropriate screening maneuvers? Use of multivitamins (which we know from PHS II reduces cancer and fatal MI)? Binge drinking versus a daily tipple? Wearing seatbelts? Driving Volvos? Family history of cardiovascular disease or cancer? Spousal violence? Psychological comorbidity?
There is simply no way to distinguish between possible explanation #1 and possible explanation #2 above. I do not find observational studies in most instances to be probative. Natural experiments – maybe. Randomized trials with high quality and precision – much more probative.
I can’t draw much inference from this study. My belief remains that health conscious behavior causes people to select vegetarianism as a lifestyle, and that vegetarianism is a marker but not a cause for improved health. In my experience, the innate qualities and characteristics of the vegan and the carnivore are like ‘night’ and ‘day’.
June 6th, 2013 at 8:16 pm
Thus study was done on the Adventists. The Adventist church encourage its congregation to avoid animal products as it claims they are unhealthy. Therefore, most of the vegetarian/vegan in this study are due to perceived health reason. This may be another explanation why all 3 British studies showed different results.
Dan, I have another question.
When you do multiple regression, is it correct that the computation of the HR and 95% confidence intervals assume the following:
1) The variables (the exposure and confounders) are independent.
2) The variables are measured correctly (namely, if someone reports he does 60 min/week of physical activity, this is the exact amount he actually does).
In other words, since 1+2 are not true in real life, is it correct that the results of the multiple regression can be inaccurate?
June 7th, 2013 at 9:15 am
Dave, I am not a statistician but I do believe the following:
Regarding point #1, if the variables (e.g. exposure and confounder) are highly collinear (overlapping), then adjusting for both will lead to a multicollinearity problem, with artificial inflation of the standard errors and very imprecise effect estimates in the model. Is this what you mean by “independent”? If the relationship between vegetarianism and survival is causal, then it will be independent of all known and unknown confounders (something that is impossible to know without a randomized trial – or a heckuva lot of inferences).
Regarding point #2, yes, information bias can be introduced by systematic measurement errors. The question would be whether the measurement error differs substantially by exposure status – i.e. do vegans report their exercise more accurately than non-vegans? I don’t see why that would be the case. On the other hand, even non-differential measurement error can make a potential confounder much less potent in the model.
June 7th, 2013 at 11:00 pm
Jack – you said this:
“Assuming I’m correct, you can look at any risk ratio two ways. For example, this: non-vegetarians 1.00; vegetarians .50
Is the same as this: vegetarians 1.00; non-vegetarians 2.00
And all I was saying is that when you look at it the second way, the distance from 1.00 is greater than when you look at it the first way. So that needs to be considered when one is thinking about the distance from 1.00 that is being achieved with various measurements and how important that might be.”
You are correct – this is the problem with the fact that the hazard ratio (or odds ratio) scale is not symmetrical – bounded by 0 at the lower end but unbounded in the other direction. So although a HR for mortality in vegetarians of .50 would be mathematically equivalent to a HR for non vegetarians of 2.00 (with vegetarians being the ref group with HR of 1.00), they do not seem like equivalent effects – the latter “seems” larger. For this reason statisticians advise researchers to not report HR/ORs less than 1 – they should flip the categories so as to report the effect in terms of HRs over 1 (as per your (and my) example). But no one heeds this advice! Also – as HRs approach 0, things get seriously skewed. Have a look at the second half of the following article if you want to get it clear – not that there is too much risk of HRs approaching 0 comparing vegetarians and omnivores (unfortunately). http://pareonline.net/pdf/v11n7.pdf
June 16th, 2013 at 12:31 am
Jack, as far as I understand, ‘mortality rates’ is not the same as ‘longevity’. So why media are saying ‘Vegetarians Live Longer Than Meat-Eaters’ when reporting about the results of the study? Is it due to the science news cycle (www.phdcomics.com/comics.php?f=1174) ?
June 16th, 2013 at 9:28 am
Dima,
I don’t think mortality rates are the same as longevity, but longevity can possibly be calculated for the Adventist Health Study 2. I’d have to see the exact article you’re talking about.
June 16th, 2013 at 9:32 am
Dima,
> http://www.phdcomics.com/comics.php?f=1174
I forgot: great comic!
June 16th, 2013 at 10:45 am
Jack, I am talking about the “Vegetarian Dietary Patterns and Mortality in Adventist Health Study 2” study and “Vegetarians Live Longer Than Meat-Eaters, Study Finds” article by the Wallstreet Journal, both mentioned in your post, which we are commenting 🙂
June 16th, 2013 at 10:52 am
Dima,
Oh, I thought you had seen an article giving a specific length of time that vegetarians lived (or something similar). The article says vegetarians live longer which is pretty much the same as saying their death rates are lower (after a certain age; it wouldn’t take into account deaths before the age of recruitment). I’m not sure about AHS2, but most of these studies determine the mortality rate before age 90, so it also wouldn’t take into account longevity past 90. But I think it’s safe to say, relatively, that one group’s adulthood life expectancy is higher if their mortality rates are lower.
June 17th, 2013 at 10:01 pm
Hello All,
As part of your ongoing discussion you may appreciate this study from the EPIC o Oxford Study. Vegetarians and esp vegans were particularly deficient in vitamin B12 and D. D is known influence cancer rates. And I am not sure about studies in people who have low B12, but wonder if there is data showing health effects for those with subclinical low b12 over the long term. Could there be an impact on DNA synthesis etc?
Lifestyle factors and nutrient intake in EPIC-Oxford
http://www.epic-oxford.org/publications/1278/davey-et-al-2003-05-public-health-nutr
June 18th, 2013 at 3:01 pm
Tushar,
The nutrient intakes in EPIC-Oxford included fortified foods, but it didn’t include supplements. And, as you probably know, vitamin D can be obtained through sunshine. Vitamin D levels were later measured in EPIC-Oxford and the vegans’ average vitamin D levels were lower than the non-vegans and just above the healthy range:
http://veganhealth.org/articles/bones#veg_levels
Low vitamin B12 levels increase homocysteine levels which has been associated with stroke, dementia, and heart disease:
http://veganhealth.org/b12/hcy
http://veganhealth.org/b12/depression
So, yes, it is possible that the low B12 status of the vegans in EPIC led to a higher mortality rate.
June 18th, 2013 at 7:59 pm
Thank you Jack!
Here are a couple of more questions if you could help me on this
1) were the vegetarians in the Epic oxford study “health conscious”. It is written that the meat eaters were health conscious, and if they were compared to the average vegetarian/vegan in the UK then it may not be a fair comparison.
2) regarding homocysteine, you list the trials that show deficiency is related to vascular disease, dementia, and depression. however, are there any studies showing that supplementing B12 and lowering levels of homocysteine reduces disease?
3) do you think that there are cancers prevented from a “good” vegan diet.
4) All cause mortality and vegan diet – are there other good studies besides adventist and epic?
5) do you think that the adventist study was fair. the respondants above feel that seem to feel that he veg group was more health conscious. or is it more likely that the adventists are health conscious as a whole, and we are comparing health conscious meat eaters to equally conscious veg*ns ?
I just started reading your web articles … thanks for your diligent work.
Best!
Tushar
June 18th, 2013 at 9:53 pm
Tushar,
I’m going to answer these one at a time.
> 1) were the vegetarians in the Epic oxford study “health conscious”.
I think it is a safe assumption that the vegetarians in EPIC were as health conscious as the meat-eaters. I would say this mainly because the vegetarians had slightly lower smoking rates than the non-vegetarians.
So far, EPIC-Oxford mortality rates have been somewhat preliminary due to a lack of deaths in the cohort. Another thing to consider is that results are usually adjusted for lifestyle factors (although in the case of Key et al.’s 2009 paper, death rates were only adjusted for age, genders, smoking, and alcohol). The meat-eaters in the study do not eat as much meat as the greater population at about 2.5 servings per day:
http://veganhealth.org/articles/dxrates#epic_heartdx_2009
June 18th, 2013 at 10:01 pm
2) are there any studies showing that supplementing B12 and lowering levels of homocysteine reduces disease?
There are small amounts of evidence. It appears to have an effect on stroke for both secondary and primary prevention:
http://veganhealth.org/b12/hcy#sec
http://veganhealth.org/b12/hcy#primary
http://veganhealth.org/b12/depression
“A 2008 meta-analysis of vitamin supplementation and cognitive function found little benefit for people already diagnosed with dementia, but did improve cognition in elderly people with elevated homocysteine but who were not diagnosed with dementia (6).”
June 18th, 2013 at 10:02 pm
Tushar,
Another thing to consider is that vegans who don’t supplement with B12 usually have much higher homocysteine levels than the general population with elevated homocysteine. So the chances for damage/prevention are greater.
June 19th, 2013 at 5:07 am
Tushar Mehta:
There is strong evidence that homocysteine-lowering with B vitamin therapy does not prevent strokes or heart attacks. In fact, the latest meta-analyses of trials have all been neutral. HOWEVER, none of these trials were conducted in vegans, who tend to develop B12 deficiency and very high homocysteines over time in the absence of oral B12 supplementation. Therefore, in my view, the safest thing is to continue to supplement with low doses of B12 (although some people may need higher doses, especially people with malabsorption or other defects in B12 metabolic pathways). It is remarkable how common B12 deficiency is, even in the general population. Jack has an excellent website on these issues at veganhealth.org.
June 23rd, 2013 at 9:10 am
Thank you Jack and Dan
I read through and also went through the references you provide in the links. It makes sense that Vegans that are particularly low in B12 from a dietary point of view may benefit.
Sorry for more questions – but do you think there is a harm in excess B12?
Loving your website and reading a lot there!
June 24th, 2013 at 6:11 am
I think excess B12 could be harmful in people with diabetic nephropathy (shown in DIVINe, in JAMA) or coronary artery disease (NORVIT in JAMA). In vegans, who tend to be prone to deficiency of B12 and high homocysteines, it is unlikely that excess B12 would result in adverse consequences. However, I would prefer to keep my B12 levels in the normal range, so I do not take more than 50-100 micrograms per day of this constituent. Some people cannot absorb B12 efficiency in its oral form without using very high doses – the only way to know is to get a blood level done on a steady state of intake.
(Interestingly (at least to me), I had normal homocysteine (9.4) when my B12 level was extremely low (100), and high homocysteines (17.5, 14) when my B12 level was high-normal (400-600). I may have another defect in the homocysteine pathway involving pyridoxine or folate. I am not too worried about it, as I view homocysteine to be an epiphenomenon and not a causal risk factor for bad outcomes. Also, there is great variability in B12 levels even in the same person, so if one is truly worried about deficiency, I would suggest getting a methylmalonic acid level – MMA – as well as homocysteine done. Jack has a section on his website about this)
June 25th, 2013 at 5:58 pm
Tushar,
> 3) do you think that there are cancers prevented from a “good” vegan diet.
These questions always hinge on what you’re comparing it to. Will a vegan diet prevent cancer compared to an almost-vegan diet in people who have other healthy lifestyles? If you go by the AHS2’s paper on cancer, vegans had a lower risk of female specific cancers:
http://www.ncbi.nlm.nih.gov/pubmed/23169929
I think it’s really too soon to know much about how much a vegan diet can and cannot prevent cancer, whether that diet is “good” or not.
June 25th, 2013 at 5:59 pm
Tushar,
> 4) All cause mortality and vegan diet – are there other good studies besides adventist and epic?
No.
June 25th, 2013 at 6:00 pm
Tushar,
> 5) do you think that the adventist study was fair the respondants above feel that seem to feel that he veg group was more health conscious. or is it more likely that the adventists are health conscious as a whole, and we are comparing health conscious meat eaters to equally conscious veg*ns ?
We are comparing health conscious meat eaters to equally conscious vegans. As I mentioned above, the vegans might have been slightly more health conscious.
June 25th, 2013 at 6:05 pm
Tushar,
> do you think there is a harm in excess B12?
Rare cases of acne are the only harms that have been detected so far. The few other correlations I’ve seen can easily be explained by higher intakes of animal products, so I wouldn’t implicate B12 until there is some real direct evidence that it does harm.
Here are the studies I’ve seen:
http://veganhealth.org/b12/inc
It sounds like Dan has seen others, but again, but I haven’t heard any respectable public health professionals blow raise concern that high B12 levels are harming people. That said, I see no reason to take more than one needs to keep relatively normal levels.
June 26th, 2013 at 4:50 am
>It sounds like Dan has seen others,
Re: B12 supplementation and adverse events:
In diabetic nephropathy (Effect of B-vitamin therapy on progression of diabetic nephropathy: a randomized controlled trial. House AA, Eliasziw M, Cattran DC, Churchill DN, Oliver MJ, Fine A, Dresser GK, Spence JD. JAMA. 2010 Apr 28;303(16):1603-9. doi: 10.1001/jama.2010.490. PMID: 20424250 [PubMed – indexed for MEDLINE]) — http://www.ncbi.nlm.nih.gov/pubmed/20424250
In patients with coronary disease (Cancer incidence and mortality after treatment with folic acid and vitamin B12. Ebbing M, Bønaa KH, Nygård O, Arnesen E, Ueland PM, Nordrehaug JE, Rasmussen K, Njølstad I, Refsum H, Nilsen DW, Tverdal A, Meyer K, Vollset SE. JAMA. 2009 Nov 18;302(19):2119-26. doi: 10.1001/jama.2009.1622.) — http://www.ncbi.nlm.nih.gov/pubmed/19920236
>>We are comparing health conscious meat eaters to equally conscious vegans.
Although you are referring only to the Adventist Health Study-2, I would like to quote from the EPIC authors in their recent study in BMC Medicine on meat intake and mortality:
“subjects with very moderate meat consumption may be the group with the highest proportion of health-conscious subjects who also try to optimize their diet (as part of a healthy lifestyle).”
That is exactly the point I am trying to get across regarding all observational studies confounded by healthy user bias, particularly those which try to dissect out the effects of vegetarianism enmeshed within a smorgasboard of unmeasured health-conscious behaviors (e.g. smoking, alcohol abuse, seatbelt wearing, looking both ways before crossing the street, vitamin-taking, supplement-taking, yoga-practicing, meditation-practicing, biofeedback-practicing, acupuncture-practicing, exercising, vacationing, classical music-listening, charity-supporting, non-tax-evading, non-recreational drug-abusing, etc, etc). While not all of these will be directly related to mortality, some will, and most go unmeasured in even the best observational studies. That is why we do randomized trials. That is why we do not make important decisions about licensing new therapies on the basis of anecdotal data (and observational data). This is not a heretical opinion. Read any epidemiology textbook (e.g. Fletcher and Fletcher) and this caveat will be made obvious in one of the earliest chapters.
My guess is that any diet which eliminates processed food, junk food, fast food, sugary/salty/fatty foods will be healthy and reduce cancer and mortality, regardless of its “brand label” designation as ‘vegetarian’ or ‘weight watchers” or “whatever”. It is the western food crap that is killing us. I see this every day in my clinic in patients with abdominal obesity and perturbed indices of glucose-insulin homeostasis, triglycerides/HDL levels, systemic blood pressure and inflammation. The western diet leads to metabolic syndrome, which about half the population now has (46.5% – the prevalence of elevated blood sugar in the latest report from NHANES, as reported in the annual – 2013 – AHA statistical report). People who still eat a native cuisine – e.g. the kitivan islander horticulturalists – have undetectable leptin levels. Even lean and athletic American marathoners don’t have undetectable leptin levels.
“Food for thought”!
June 26th, 2013 at 12:54 pm
Dan,
Not great news on B12, but as you like to point out, there is quite a bit of confounding there and assuming it was the B12 is a stretch.
June 26th, 2013 at 1:12 pm
Jack,
These were randomized trials. The design of the Norwegian studies specifically vindicates pyridoxine as the culprit, since there was no increased risk in the B6-only group.
See also: http://www.ncbi.nlm.nih.gov/pubmed/20698927
I think the confounder could be folic acid intake, but it’s hard to tell, as the intervention came as a package (B12 + folate vs B12 + folate + B6 vs B6 alone).
I have not seen a B12-only randomized trial in the literature containing hard outcomes like cancer or survival.
Interestingly, in the pubmed link above, the authors note that it is the homocysteine portion that does NOT respond to B vitamin therapy that remains strongly predictive of major cardiovascular events in follow-up (ie not the portion that gets lowered by the B vitamins).
None of these studies were conducted in vegans, by the way. Nor in B12-deficient people, for obvious ethical reasons (placebo being harmful there).
I am interested in the research you cite on B12 and acne, since the latter is an inflammatory condition. Is it available on your site?
June 26th, 2013 at 1:32 pm
> These were randomized trials.
I realize that, but there were still many variables that prevent it from applying to apparently healthy vegans taking only vitamin B12.
June 26th, 2013 at 2:09 pm
Agreed. But that’s not confounding. The definition of a confounder is a variable which is asymmetrically distributed between groups (say, higher in the B-vitamin group than the placebo group) AND prognostically related to the measured study outcome. What you are getting at is generalizability rather than validity. The trials are large, randomized, blinded and internally valid with minimization of any confounding influence between groups (unlike observational studies where people get to pick their own interventions – e.g. vegan diet). But because of the sample studied (likely carnivores with CAD or diabetes and definitely baseline normal B12 levels), the results cannot be generalized to potentially B12-deficient, healthy vegans.
Other terminology for generalizability of the study findings includes “applicability” and “external validity”.
All we can say is that in diabetic nephropathy or CAD, it appears the combination of B vitamins worsens outcomes.