High Protein Diets Cause Cancer, Study Says
Perhaps you’ve seen the headlines:
“High protein diets as bad as smoking for middle-aged people”
“High protein diet raises cancer risk as much as smoking”
“Study says high-protein diets in midlife liked to high risk of premature death”
All these media outlets are reporting on the same study recently published in the Journal of Cell Metabolism. Let’s take a look at the study.
The researchers administered a 24-hour dietary recall and measured nutrient intake (protein, carbohydrates, fats, and calories) to 6,000 participants. The participants were categorized into three groups based upon their protein consumption (low, med, high) on the first day. They then followed up with the participants 18 years later and measured cause of death (all-cause mortality, cardiovascular mortality, cancer mortality, and diabetes mortality).
As a potential mechanism of action the researchers hypothesized that eating protein increases insulin like growth factor (IGF-1) and examined if it would increase the rate of growth of already present tumors in mice. Additionally, 2,200 people were sampled for IGF-1 levels, although when and under what conditions were not reported.
Here were their reported findings:
- “high and moderate protein consumption were positively associated with diabetes-related mortality, but not associated with all-cause, CVD, or cancer mortality”
- “Among subjects with no diabetes at baseline, those in the high protein group had a 73-fold increase in risk, while those in the moderate protein category had an almost 23-fold increase in the risk of diabetes mortality.”
- “Among those ages 50–65, higher protein levels were linked to significantly increased risks of all-cause and cancer mortality. In this age range, subjects in the high protein group had a 74% increase in their relative risk of all-cause mortality and were more than four times as likely to die of cancer when compared to those in the low protein group.”
- “…when the percent calories from animal protein was controlled for, the association between total protein and all-cause or cancer mortality was eliminated or significantly reduced, respectively, suggesting animal proteins are responsible for a significant portion of these relationships.” This is to be expected, as the majority of protein consumed was from meat. If we remove all the protein we remove the relationship.
- “ Results showed that for every 10 ng/ml increase in IGF-1, the mortality risk of cancer among subjects ages 50–65 increases for the high protein versus the low protein group by an additional 9%”
- In mice “A 45% smaller mean tumor size was also observed in the low protein group compared to the high protein group at the end of the experiment at day 53”.
- “Although there was a [extremely low!] trend for an effect of substituting the same level of animal protein with plant protein on IGF-1 andIGFBP-1, the differences were not significant”
This was a study of observational epidemiology. What this means is we look for relationships between two or more phenomena. As researchers, we can then comment on the relationship; however, we cannot prove causation. As an example, we might look at the relationship between socioeconomic status and race. According to the American Psychological Society there are more African Americans living in poverty than Caucasians. It would be both irresponsible and ignorant of us to assume that there is something inherent to Caucasian genes that predisposes them to a greater percentage of success in the US when we know that not only is this not true, but that there are a myriad of other factors such as past and ongoing discrimination and segregation responsible for this.
Unfortunately, claiming causation based off of correlation is exactly what the researchers did. Let’s take a look at their conclusions:
- Overall, our human and animal studies indicate that a low protein diet during middle age is likely to be beneficial for the prevention of cancer, overall mortality, and possibly diabetes through a process that may involve, at least in part, regulation of circulating IGF-1 and possibly insulin levels.
- Our findings suggest that a diet in which plant-based nutrients represent the majority of the food intake is likely to maximize health benefits in all age groups.
- We also propose that at older ages, it may be important to avoid low protein intake and gradually adopt a moderate to high protein, preferably mostly plant-based consumption to allow the maintenance of a healthy weight and protection from frailty.
What we have here is the classic bias of compliance. In short, those who eat more meat are (unlike many of my readers) not living healthy life styles. They do not “listen to their doctors”: they are inactive, consume less fruits/veggies, smoke, rely heavily on caffeine, consume high amounts of added sugar, etc. – They live the typical western life. Despite the ability to control statistically for these factors in an equation, you cannot control physiologically for the interactions, and you certainly cannot pick one factor out of the myriad and claim it is the main causative factor. As another example, we could look at TV consumption and its relationship to metabolic disease, cardiovascular disease, and cancers. A number of studies has shown a positive result between TV consumption, obesity and diabetes. Is it the fact that people who watch a lot if TV are less active and living less healthy lifestyles? Or is it something about the TV device itself (radiation, frequency, etc.) that causes these diseases?
In addition, the authors claim a benefit of increasing the consumption of plant proteins; however, the results from the mice studies where animal protein was actually replaced with plant proteins do not support these statements. There was no effect on tumor growth regardless of type of protein consumption. Take a look at Graph J: high protein from casein or soy equally accelerated tumor growth. And, we should note that mice, unlike humans, do not consume very much of the natural diets from protein sources (they are predominantly herbivores, not omnivores).
Next, if we look at the raw facts provided in the supplemental data we see that the authors committed sensationalism of statistics, check out Table S1 and look at the absolute percentages:
- In the low protein cohort 9.8% died of cancer. In the moderate and high protein diet 10.1% and 9.0% died of cancer, respectively. Hardly a significant difference!
The authors then provided the rates of cancer for the 50-65 age group and 65+ age group following questions regarding the arrival at their conclusions and we see:
- Low: 18.07%
- Mod: 20.28%
- High: 26.15%
Age 50-65 Cancer Mortality
- Low: 2.58%
- Mod: 7.89%
- High 9.89%
Age 66+ All-cause Mortality
- Low: 70.97%
- Mod: 63.73%
- High: 64.04%
Age 66+ Cancer Mortality
- Low: 18.03%
- Mod: 12.94%
- High 7.96%
This may seem dramatic; however, if we dip further into the data we see that there was a large difference between the number of participants in each group such that for the 50-65 age group there were 219, 2,227, and 543, respectively. Taken a step further (and credit to Zoe Harcombe for doing so) we can calculate the total person years per group and then the cancer rate per 1,000 person years. If we do so the numbers are no longer dramatic: 2.18%, 5.95%, and 7.84%, respectively. In other words, the risk of eating a high protein diet with relation to cancer death is about 5.5% higher per 1,000 person years versus a low protein diet.
Another concern not discussed was if the 24 hour recall reflects the way people the majority of their adult lives, then why the sudden change around the age of 65? This was not an intervention trial – the older group did not eat a low protein diet early in life and then change to a higher protein diet upon retiring at 65. Quite the opposite, the older group ate a high protein diet their entire lives. Why then the discrepancy between cancer mortality at young vs. old age?
We may interpret this discrepancy as there are other factors involved, such as: dietary quality of food (increased consumption of processed foods in modern life), environmental decay, stress, lower activity levels, and everything else associated with the reduced health outcomes seen in the baby boomer generation. So is it protein that is actually causing the cancer (when no mechanism of cause has yet been identified), or just accelerating the growth (in rodents)? If the latter, then from the data we might actually conclude that protein intake plays a protective role in human health (alongside an active lifestyle rich in unprocessed foods). In which case, if Longo et al. are wrong regarding protein actually leading to the development of cancer (and not the acceleration of cancerous growths) then the .36 g./lb protein recommendation suggest by the author may lead many people into later life with a lower chance for survival. But this is all just speculation, we will never know either way until a controlled intervention trial is conducted. And until a controlled intervention trial is conducted, Longo et al. should not be making ridiculous claims based upon association.
Next the authors claim that protein consumption increases the release of insulin and IGF-1 which expedites the growth of [already present] cancer cells. It does not cause the development of cancer. If IGF-1 and insulin caused the development of cancer then we should all avoid exercising (as exercise is a powerful stimulator of IGF-1 release) and eating as every time we eat (especially carbohydrates) insulin is released to maintain normal blood glucose levels.
Despite these unanswered questions, the media firestorm ensued.
Now the real kicker. For those who are not involved in academia and research science, when we look at a list of the authors it is generally the first author who did most of the grunt work and the final author who supervised (and often designed) the study. The final and corresponding author of this study is Victor D. Longo (VDL). If we look at the related info we see that:
- VDL designed the study and obtained funding from the Nation Institutes of Health (NIH)
- The NIH had no role in study design, data collection and analysis, or the writing or and publishing of the manuscript
- VDL has an equity interest in L-Nutra, a company that develops medical food.
What exactly is L-Nutra and “medical foods”?
L-Nutra’s products are a “formulation of natural nutrients with the ability to provide nourishment and allow subjects to enjoy a combination of good and mostly organically grown and plant-based food.” L-Nutra’s major product ProLon is “an all-natural plant-based 5 five-day Fasting Mimicking & Enhancing™ Diet (FMED) program.”
Since none of us were present in the design or execution of the study we may only speculate that the interpretation/discussion of the results the suggestions to increase plant-based protein intake to prevent cancer despite any evidence in the study to back up this claim in conjunction with the lead authors affiliation (he is actually the founder!) are…well…suspicious to say the least!
Trackbacks & Pingbacks
- Is protein bad for you? Bad science says so! Don't believe everything!
- [Fan Club] LCHF Lifestyle - Part 3 - Page 11 - www.hardwarezone.com.sg
- Researchers: Animal protein-rich diets could be as harmful to health as smoking - Page 2 - www.hardwarezone.com.sg
- Er kød nu farligere end rygning?
- Animal protein causes cancer « Peter Risdon
- Does Meat Really Cause Cancer? | Modern Health Monk
- 03.10.14 – Know All The Facts » CrossFit Sunnyvale
- Animal Protein Posed to be As Bad as Smoking. The research does not add up.
- High Protein Diets | Golf Fitness Los Angeles Blog
- The Most Anabolic Androgenic Exercises | Jason Cholewa - Assistant Professor of Exercise Science at Coastal Carolina University
Jason Cholewa - Assistant Professor of Exercise Science at Coastal Carolina University 
Hi Jason,
I’ve been seeing this study posted all around so thank you for providing a great breakdown of the issues with it.
The one thing I haven’t been able to find out – All these issues are reported in the 50 – 65 age group for high protein. In the study, the 65+ group does not seem to have the same issues for high protein intake.
Does the study, in any way say if the 65+ group only started eating high protein once they were over 65 or were they also once in the 50 – 65 group and still consuming a high amount of protein?
Regards,
Jon
Thanks Jon! I did not see anything detailing the differences between the two such as when sampling started and when one categorized into 50 vs. 65. We can only speculate. My guess though, is that it would have been very, very irresponsible to promote a lower protein diet in an age group that requires more protein to maintain lean and bone tissue, and also requires optimal growth hormone and IGF-1 release to maintain said tissue.
Hi Jason,
I think you misunderstood table S2. Look at the average age at the top of each column. Around 65 years old. That means each column includes everyone 50+, not just the 50-65 year old group that the authors claim they see significantly higher deaths from cancer in.
Looking at the full 50+ group, protein does appear to be mildly protective for cancer. But that presumably results from the 50-65 year old group showing a higher death rate from cancer + and the 65+ year old group showing a lower rate. Together they cancel out.
I see what you are saying, but how do we discern between the two since it was not reported nor graphically illustrated?
When we look at the hazard ratio’s for moderate and high protein we also see that they are not statistically significant with p values of .38 and .8, respectively. The only significant hazard ratio was high protein x IGF-1. Or am I missing something?
Do you mean how do we discern between the 50-65 year old group and the 65+ year old group? Table S2 doesn’t discern those 2 groups. It looks to me like S2 is meant to be a description of the full 50+ data set.
I think if you wanted to dig any deeper into the raw data set used by the authors you would either need to obtain it from them or reconstruct it. I presume NHANES III is publicly available.
NKSL55, yes, that is what I was referring to. I guess the deepest we can look at it without obtaining and analyzing the NHANES III data is to look at table S4, where none of the hazard ratio’s for protein intake alone reached (or even came close to reaching) significance.
My guess is this (if it’s true) is because it’s correlated with the normal onset period of diabetes and associated disorders for people who are (epi)genetically susceptible. This should have been a sugar study, not a protein study. I think it’s disgusting that ‘researchers’ can get such an unquestioned public audience about a serious issue which also gives them visibility to flog some cr@ppy supplement/blocker/test that fits with their moral world view.
Troy, those are some very valid comments, thanks for also taking a close look at the study.
As a researcher I am appalled at both the heavily biased conclusions based on a lack of data, and in some cases making conclusions that are actually refuted by the data. As far as we can tell, there was a direct monetary/corporate motive for publishing these conclusions, which is reprehensible.
Although we have to hold some of the media outlets and journalists some what accountable for their propagation of this non-sense, they are not trained to critically review scientific studies. You know who is though…the journal of cellular metabolism and it’s peer reviewers. In my opinion they committed an atrocity by not critically reviewing the manuscript and by publishing this misrepresentation of data.
>In the low protein cohort 10 people died of cancer. In the high protein diet 17 people died of cancer. This is around .1% and .17%. That is a difference of .7% increase in cancer mortality, not the 70% suggested by the authors.
Did you actually think that “a 74% increase in their relative risk of all-cause mortality” meant that most of the high-protein participants died? 🙂
Where did you find the 10 people and 17 people? That is a 74 % increase (or so) in absolute risk but it is a 0.07 % increase in absolute risk. That’s why it is unacceptable to only report relative risk. Of course, in nutrition, unacceptable is the order of the day. But where did you find these numbers?
Just wanted to point out that I clicked the link referring to L-Nutra’s site and there was no Victor D. Longo working for them, but rather a Valter Longo.
The study itself says Valter D. Longo so you should change your text so people won’t think you’re deliberately fooling them 🙂
Reblogged this on mindovermonkey and commented:
Brilliant rebuttal to the sensationalist headlines surrounding a suspicious “scientific study”
“In the low protein cohort 10 people died of cancer. In the high protein diet 17 people died of cancer. This is around .1% and .17%. That is a difference of .7% increase in cancer mortality, not the 70% suggested by the authors”
.17-.1 = .07, not .7
.07 also happens to be 70% of the .1% in the low protein group. That means there was a 70% higher instance of cancer mortality in the high protein group.
There are some issues with this study, but there are also some issues with your math.
Joe, you are right regarding the .17 – .10 = .07%.
And if you want to look at the math on the 70% increase, then sure, its correct. However, to say there was a 70% higher instance of mortality in the high protein group (at least without presenting the absolute numbers) insinuates to many (and was interpreted by the media without clarification from the authors) that there was a huge difference. Also, many people interpreted this as 70% of the high protein participants died of cancer. Now, had the numbers been 10% vs. 17% cancer mortality rate, that would have been a different story.
Joe what does the “.1% in the low protein group” refer to?
Did 1.7 times as many people on the high protein diet die of cancer compared to a low protein diet?
I’m just wondering if I should stop eating all meat and fish and cheese. I don’t eat much red meat but I do eat a lot of organic chicken and fresh fish and cheese.
Joe, look at the raw data in table S1 and you will see that the all-cause mortality was the exact same rate between the low and high protein groups, and just slightly higher in the moderate protein group. It was about 40% for all groups.
If you want to look at the absolute number of people who died of cancer between the low and high protein groups (43 vs. 103) then the risk of cancer with high protein consumption would appear to increase substantially. However, when we consider that there were only 437 participants in the low protein group compared to 1,146 in the high protein group we see that the actual rate of cancer deaths were nearly the exact same (9.8 vs. 9.0%).
Unfortunately, what the authors did was take the number of people who died in each group and divide it by the total number in the study to arrive at the conclusion that the rate of cancer deaths was only .007% vs. .017% (which they claimed nearly doubled the risk!). This conclusion would be synonymous to me comparing cancer in Hawaii to California.
There are 1,360,000 people living in Hawaii. The cancer rate is .38% which would equate to 5,168 people having cancer.
There are 37,000,000 people living in California. The cancer rate is also .38% which would equate to 140,600 people having cancer.
What the authors did was take the total amount of people between the two states (38,360,000) and calculate the cancer rate based upon this number.
So for Hawaii 5,168/38,360,000 comes to .0001%
For California 104,600/38,360,000 comes to .0036%
Clearly living in California causes cancer.
And therefore we can conclude that living in California increases the risk of cancer up to 36X and we should all move to Hawaii.
There were 2578 deaths in this study. 631 were cancer. You can find all this info in the supplement where the number 113 doesn’t even appear. http://www.cell.com/cell-metabolism/abstract/S1550-4131%2814%2900062-X#suppinfo
I believe you’ve mistaken Stuart Phillips made up numbers on twitter for being real.
That being said. This is a pretty bad study with a horrible press coverage
Gudiol, you are correct. I interpreted that at the time as raw numbers (if you add up the all-cause mortality in the three groups), not a %.
Ok, so if we look at the numbers then there really isn’t very much of a difference in the % of people who died of cancer between the three groups.
Low, moderate, and high were: 9.8, 10.1, 9.0, respectively.
The problem with the data in the supplement is that it’s for the entire groups. And then there were no differences. They haven’t given us the splits for when they did split the groups in younger-older. And that’s fishy …
Considering most that died of cancer probably were older the number of deaths in the low protein <65 and high protein <65 still could be low. We just don't know since this data isn't reported.
That’s also a good observation Gudiol. I updated the blog to better reflect the data presented in S1.
Again, I don’t see how you got the raw numbers. The mortality in figure S1 has no units.
RDFeinman, I aksed the authors to disclose their raw numbers on the comments section of the original article of cell metabolism. They disclosed the frequencies per group (low/med/high for 50-65 and 65+). From there we were able to see the n of the 6 groups and calculate the raw numbers of cancer deaths per each.
Jason, can you send me off-list email at feinman@mac.com
Graph J is comparing casein and soy – whey does not appear to be present.
This is why you work at Coastal Caroline.
Valter D. Longo is an idiot and destroyed his career with that fake and ridiculous study.
Thanks for a great article,and for the links and research on conflict. I read the actual study after this article and wanted to punch something. How on earth can a paper like this get through review?
There are a hundred things wrong with the paper, but the worst I could find are as follows – for (L)ow, (M)edium and (H)igh groups:
1) Their hazard ratio for mortality (S2) shows that HR is HIGHEST for the Low protein group for All Cause, Cancer and CHD. It is only lower for Diabetes as a subset of All Cause. Opposite result to what media headlines are saying about this study.
2) The incidence of cancer death (L 9.8, M 10.1, H 9.0) shows (if anything) that a HIGH protein diet is PROTECTIVE against cancer death overall.
3) Their central ‘conclusion’ is related to IGF-1&Protein in the 50-65 age-group. However, in the table S4 only 1 out of 15 published hazard ratios has a p-value of >0.1 (yes 10%, not 1%!), and some are over 0.8. This should not even be reportable when linked to a conclusion.
Also there are a couple of population data inconsistencies which they don’t (publicly) deal with:
4) History of diabetes is ~4x in the M group and ~6x in the H group. It seems a stretch that protein intake is causing a huge uptick in diabetes, which is causing a tiny uptick in cancers in a specific age group. Is it not more likely to be related to some undescribed dietary or metabolic component of the sample groups relating to L,M,H & sugar/carbs/insulin sensitivity. Surely if they really believed this protein>>diabetes was causative with HR of 73 (4-1200) with a p-value of 2x) than the H group. Doesn’t this indicate that the group would have more cancer survivors (who’d naturally be more careful about relapse detection, and therefore be less likely to die of cancer) than the other group?
Something feels off about the baseline L group – it’s less than 7% of the sample and I would be really interested to know if there’s other info about this group that’s not published (e.g. vegetarian/lacto-veg, running club etc.) …. it might explain why they’re so much worse at remembering what they’ve eaten 🙂 Also interesting to see they eat the most, are the leanest and weigh the least – surprising that there’s no exercise data given.
It would be great to get the raw data from this study. I suspect a conclusion set that might be reached with an open mind would look like this:
1) Diabetes makes you get sick and die young of a number of things, including cancer
2) If you are insulin resistant (genetic, fat, lazy, bad diet,whatever) and your blood sugar and insulin are high, you will get diabetes/heart disease/cancer … and more likely do so younger (50-65)
3) High protein intake, combined with extended daily periods of low insulin/blood glucose and associated low IGF-1 will protect you from cancer (and it looks like heart disease) – lots of other studies say this
Just seems bizzare to me that they’d publish data with which it seems easier to draw opposite conclusions to those they’ve come to. I guess they (rightly) think that no-one will bother to read anything but the abstract – so thanks for making a difference!
Sorry 4 & 5 about got merged:
4) History of diabetes is ~4x in the M group and ~6x in the H group. It seems a stretch that protein intake is causing a huge uptick in diabetes, which is causing a tiny uptick in cancers in a specific age group. Is it not more likely to be related to some undescribed dietary or metabolic component of the sample groups relating to L,M,H & sugar/carbs/insulin sensitivity. Surely if they really believed this protein>>diabetes was causative with HR of 73 (4-1200) with a p-value of 0.001 they would trumpet PROTEIN GIVES YOU UP TO 1200x DIABETES RISK (but no more risk of death)
5) Double cancer risk in L than the H group. Doesn’t this indicate that the group would have more cancer survivors (who’d naturally be more careful about relapse detection, and therefore be less likely to die of cancer) than the other group?
Saw your comment on the Cell page about “funky math” and as a professor of Epi, I felt compelled to read your blog post since you say you wrote a critique yet it is apparent you have no understanding of basic statistics. Do any of you on here even understand how these models are run, or how research of this kind is conducted?…apparently the answer is no. I agree with Bob Hope’s comment; this blog makes it glaringly obvious why you work at coastal carolina. It sounds like you don’t even understand this simple paper. The authors aren’t doing anything brilliant or extraordinary, it is all stuff you might learn in a slightly advanced PhD statistics class (maybe you missed that class). First, you cannot calculate a hazard ratio the way you talk about on the Cell comments (by looking at the raw numbers). You need to generate a baseline hazard computationally (using a computer) based on person-years for each event. Also, my guess is they used a competing risk model which means that you have to censor deaths from other diseases. Look it up; you have a lot of learning to do if you want to “evaluate” studies. Also the authors said they adjusted for other variables. Please tell me you have heard of confounding factors????
As for Troy, you are making the same blatant mistake. The values (L 9.8, M 10.1, H 9.0) don’t “show anything”. Again, look up confounds and proportional hazard ratios on Wikipedia…learn something, it will be good for you. Also the table you are talking about (S2) has no significant values and therefore does not “show HR is HIGHEST for the Low protein group”. Finally, table S4 is testing interactions, not main effects. Again, look it up before attacking it.
I don’t know if the findings from the study are true. I think they call for more research. However, make sure when you attack something that you know what you are talking about. It’s just embarrassing.
Amelia,
You have no right to insinuate my institution is less than.
As for your comments direct toward me, my statistical training reflects my line of work, which is intervention studies. Criticism taken. So I read around a little bit and edited the blog accordingly.
Within the context of the crux of this blog, the concluding statements of this (and other epi studies) often overlook the difference between statistically controlling for cofounders and the physiological effects of said cofounders. While you may control for other variables in a model, an equation (such as the ones used in this study) cannot “remove” or physiologically explain the biochemical interactions of several variables within the organism itself. This is the major issue with forming causative conclusions based on correlative studies, that happens all too often in our fields. You can observe a relationship, you can conclude that there is indeed a relationship, you can form a hypothesis based upon said relationship, but you cannot form causative conclusions until you have performed a controlled trial. Basic scientific method.
Hi guys – great conversation – just adding my humble piece below, this kind of stuff irks me something awful:
http://www.slideshare.net/ivorcummins/20140309-protein-debacle-simplified-version
Best Regards
Ivor
I enjoyed this quite a bit, and will share. Thank you.
Amelia – thanks for the response. When I quoted the direct cancer death rates, I did so in a flippant way to make a point: which is that there is little in this study which links the data they show to the conclusions, so one might as well look at what the data (summary) actually says. But you are right – one’s got to analyse the data and draw conclusions to be convincing: so here’s a bit more detail on the thought process I went through before replying.
It’s instructive to see that less % High protein people died of cancer in the all age group, and Low and High protein all cause rates were equal. I suppose this could be because the High protein death rates are “confounded” by group members being fatter, much more diabetic, less educated and probably doing less exercise (looking at calorie/waist/BMI balance) than the low protein group. Hey – call me a cynic: I’d expect that to kill them earlier, not later, but maybe there’s some other “proportional hazard” we haven’t seen.
Actually if you look at the study’s own published hazard ratios – they show that all the calculated HRs are highest in the Low protein group except for Diabetes death. Ok granted – they bracket 1.00, so not significant except for Diabetes but
a) the trend is uninterrupted
b) while this in isolation might not be a reason to say “High Protein protects against cancer”, it is DEFINITELY not a reason for the lead author to go on record saying “Eating animal protein as bad as smoking”, especially while the company he founded is plugging “plant based fasting supplements”
A bit of background – my interest in these things is mainly practical/personal. I had some really bad inflammatory/general health issues which I finally sorted out (with unbelievable increase in my own quality of life, and lipid labs etc.) by doing my own research and experimentation with a diet/exercise profile that at the time was contrary to most medical/scientific advice (that’s changing a bit now) – which also said at the time I could not expect any improvement, and should just treat symptoms and take statins.
My issue with this study – particularly as it’s been marketed so hard outside of academia, on every news channel, with the message “PROTEIN KILLS” – is that it has the potential to damage many lives, without any clear basis. A percentage of people will read the news headline, or maybe the abstract, and make life changes because “a scientist told me to”, where it’s really hard to see any demonstrable benefit that’s generally applicable, and actually there may be harm.
The only calculated human data in this study that purports to show negative associations between protein and health (except for diabetes) is not all ages or 66+, but for the 50-65 age group. However, if you look at the study’s own calculated HRs – you will see that of the 15 p-Values (relating to HR’s modestly over 1) 14 are in the range 0.1->0.8. If that doesn’t make you uncomfortable with the conclusions drawn, then maybe you need to look up p-Value in Wikipedia. You’ll can also look at the much lower p-values relating to protein protectiveness in later life and draw your own conclusions.
The only thing in this study that’s really statistically significant is the observed high-protein association with diabetes and diabetes death, which is interesting, but the study makes no effort to explain the data. While high protein in pre-existing diabetics might make more AGEs and *might* affect atheromatous plaques – who knows – I think that few people would consider protein directly causative of diabetes based on other evidence out there; maybe it increases diabetes death (but probably interaction of protein & uncontrolled sugar). It’s not clear the factors here are properly controlled, though:
1 – while they note in the results that diabetics are told to reduce fat and carbs (so naturally have higher protein % in diet), and this may explain the top level association, they talk no further about it and it’s not clear that they factor into their models that pre-existing diabetics are more likely to end up in the high-protein group (although I think that’s pretty obvious). And guess what – a lot of diabetics who die of “diabetes” die young (50-65) because their sugar and insulin is out of control.
2 – what is Diabetes mortality? In medicine this is a well known issue of reporting – while some causes like acido-ketosis are clear, others like “MI as as result of …”, not so much. Diagnosed diabetics (of which there are 4x an 6x in the M & H groups) more likely to have it recorded as a cause / contributor. CVD/Diabetes death a grey area of reporting everywhere else, so why not here? How is it dealt with?
3 – lastly – no control for exercise. Odd in this kind of study, given all the other questions asked. Was the data collected and not reported? If so, why?
You say in their defence that “the authors said they adjusted for other variables”, but they don’t say how, and as you must know, a study is the result of many factors. Raw data – Yes, Models, – Yes, but also researcher interpretation and intent (including previously held views). Everyone has their own motivations which colour whatever they do; and given the lead author of this paper has a public commercial interest in a company that sells supplements under the headline “Animal Protein is bad”, his “results” need to be scrutinised. If you don’t think this is a factor, look up Funding Bias on Wikipedia. And if you genuinely don’t think that studies get manipulated to reach pre-determined conclusions all the time, it probably means you haven’t graduated yet, and you’re in for a rude shock when you hit the real world.
The real issue I have with this study is that it’s twisting science into marketing in incremental steps:
1) First, the data looks odd. There are things like exercise that probably were in the questionnaire and are not published. Raw BMI is not published, and the L group is <7% of the population. The L group appear to have a different "health profile", but OK
2) Second, the analysis with respect to the age bracketing looks like a classic example of data torture – the overall data says that protein is protective over life (not much signif). If you adjusted properly for uncontrolled diabetes, it would probably be significant. Enter data subset showing protein~harm associations published with p-values 0.1-0.8, but OK
3) Then, the result: "These results suggest that low protein intake during middle age followed by moderate to high protein consumption in old adults may optimize healthspan and longevity." There is NOTHING in terms of causal demonstration, particularly of part 1. Or of any evidence supporting dietary variation over time. Where does this conclusion come from? Off their own top level calcs, they'd be better off saying "Eat high protein to live longer", and if they adjusted for diabetes control it would probably be significant, but OK
4) Then Cell Metabolism publish. Bearing in mind that part of the result recommendation is based on Hazard Rates which at pval 0.1-0.8 shouldn't drive any conclusions, but OK.
5) Then the story is marketed hard and read by tens of million of people under some variant of the headline "A big boy said protein will kill you". If you think that's hyperbole, do a Google News search. And if you think they are just being misunderstood or mis-quoted, here's a direct interview quote from Longo :
. Seriously, worse than smoking? NOT OK!
6) People don’t read the detail, and choices follow which are not based on science, or even facts, but support a position which is of commercial benefit to the researcher. That’s wrong on a lot of different levels.
Science is of no use in isolation unless you have a real closed system. Even a detailed study (which this isn’t) needs to be taken and reported in the context of reality if it’s to be useful universally as knowledge, and not marketing of a vested interest. I think that eventually “science” will have cried wolf once too often, and that loss of trust will make it even harder to progress “real” discoveries.
I feel passionately that ordinary people all over the world are being misinformed on a daily basis about many aspects relating to their well-being: diet, medicine, money, whatever – and the options they have. So when I seen an article on a forum like this, I try to read the background, think about it, draw some conclusions and leave a quick response. It wasn’t intended to be a full rebuttal, or for a science-y audience (it’s a blog reply) – but your point about accuracy of criticism is well made. I’ll certainly be more precise about my thoughts in the future! But my personal conclusions about the very flawed nature of this study and it’s results (and subsequent marketing): I stand by them. And while internet criticism can be ill-informed or conspiratorial, I think it’s always useful on balance.
While criticising the replies is genuinely useful, I think you’d be doing more good for more people if you cast your eye over the published study in question.
Read the whole study. In the context of how it’s then being used to tell ordinary people that steak is the new Lucky Strike, it’s bad. I think the real ball-droppers here are Cell Metabolism (and the tabloids) – they shouldn’t have published with the recommendations as is, and they certainly should have made some noise when the authors went out on their worldwide meat hatchet-job, using the cloak of a “publication” for credibility.
Thanks,
Troy
PS: For my two cents worth of what I think is going on here (without the raw data): I think this study is made unintelligible by the failure to control diabetes (incidence and control). If you are awash in insulin and glucose, you are going to get sick and die young, probably from target organ damage accelerated or amplified by diabetes. For people who get this early and don’t control properly, they’re going to a) probably end up in the H group because of diet recommendations, b) be more likely to die age 50-65. Probably once you control for that unfortunate sub-group, you’ll find that moderate to “high” protein – in proportion to your activity / exercise level – will keep you healthy and alive longer. Even in these results you can see that once the “unhealthy” people have died out, protein’s protective (and statistically significant). Also bear in mind on a 2000kcal diet, you’re only talking 2 chicken breasts (assuming 0 other protein daily) to get into the H group – we’re not talking about eating the hind leg of a cow for lunch & dinner.
Sorry this was the Longo quote, which I bracketed so it disappeared!
“High levels of protein can be can be as bad for you as smoking. People should understand the distinction and be able to make the decision about what they eat,” said Dr Longo.
“Some proteins are better for you than others, for example plant-based proteins like beans. Vegans seem to do better in studies than those who eat animal-based proteins.
Belfast Telegraph
Reblogged this on A zero-bullshit approach to Supplements, Sports Nutrition, and Training. and commented:
10/10 article!
The funky math is doing modelling on a stratified dataset. They should have included age in the modelling but they didn’t (as far as I understand).
That’s why the graphs show no effect of protein on cancer (fits with raw data on suppl table that also shows nothing), only on diabetes (but sample size is ridiculously low for diabetes). They don’t report the raw death rates in both age groups after their “stratification”. I would really like to see those IN ADDITION TO their modelling (which has stats purposes only). It’s very bad practice to leave out the raw data.
Very good commentary but I am not sure how one calculates the primary data, the number of deaths in each condition. I suppose on can try dividing by each N but if your calculation is right: 3 death out of 437 participants (9.8%) was calculated as 43/6381 or .007% (which was rounded to .01%)
480 deaths out of 4,798 participants (10.1%) was calculated as 480/6381 or .075% is a 70 % increase in risk but is meaningless because the absolute difference in risk is 0.07% which can’t be predictive of anything.
On the general case, I think as I pointed out in my blogpost http://wp.me/p16vK0-gz, it is that observational studies do not necessarily imply causality. The association of smokers and cancer does because it is a strong association and meets other criteria that I cited.
My comment on Cell Metabolism – “awaiting moderation” (March 6. 2014.). I think – the censorship works. Here is my comment:
This article – without sodium (content and intakes) – is only pseudo-science.
http://www.science20.com/entropy_and_sodium_intakes_wicked_problems_health_sciences-120016
I would also like to point to another epidemiological study done in Southern Europe. In this study the greatest contributor to breast cancer was the consumption of starch. The greatest protectors of breast cancer were unsaturated fatty acids consumed in the form of raw vegetables. Protein and saturated fat in this study did not effect the risk for cancer.
http://www.sciencedirect.com/science/article/pii/S0140673696910089
Can a statistically stronger (and even more absurd) case be made that consuming a single incidence of ‘high’ level of animal protein within a 24-hour period can increase your risk of cancer if you are age 50 or less but lowers cancer risk if you are age 65 or more?
Amelia,
The tendency to insult rather than instruct seems to be common in statistics. I have some experience with scientific data and the statistical principle that I hold to was voiced by Norman & Streiner in their book PDQ Statistics
“The important point…is that the onus is on the author to convey to the reader an accurate impression of what the data look like, using graphs or standard measures, before beginning the statistical shenanigans. Any paper that doesn’t do this should be viewed from the outset with considerable suspicion.”
Longo’s paper is not clear on the primary data. All we want to see is how many people in each group died and at what time. Easy enough to show but I may be at fault but I cannot find it.
I good way to do it might be my graphic version of the Common Language Effect Size Measure: (Feinman RD, Fine EJ, Volek JS: Analysis of dietary interventions. A simple payoff matrix for display of comparative dietary trials. Nutr J 2008, 7(1):24).
This is the data in Table S1. No units are indicated. What do these numbers mean? How many people died and what is the average mortality rate. Refrain from insulting me.
Full-Samples
N=(6,381)
Low
(437)
Moderate
(4,798)
High Protein
(1,146)
Died (All-Cause) 40.4 42.9 39.6 42.9
Died (CVD) 18.7 21.3 18.0 20.7
Died (Cancer) 9.9 9.8 10.1 9.0
Died (Diabetes) 1.0 0.2 0.9 2.0
Why are these broken into groups at all. If there is a dose-response curve, we should see all the data.