Academic ResearchCancerHeart DiseaseRed Meat

Red meat & mortality & the usual bad science

The media lit up on the evening of Monday March 12th as a press release was issued about an article in the Archives of Internal Medicine published that day.

The BBC were among the first to pick up the story and the story was featured extensively on BBC Breakfast TV and Radio 4 on Tuesday 13th March. Interestingly, John Humphries asked the pertinent question of science reporter Tom Feilden “We’re all going to die – let’s accept that. So what does this lower risk mean?” Tom couldn’t answer the question. He replied “It’s very difficult to unpick these statistics – these numbers are used as bald headlines.” Quite so!

So let us try to unpick the data and see what this article is all about:

At the outset we must highlight the error that this, and every similar study, makes. All that a study like this can even hope to achieve is to suggest a relationship between two things. To then leap from an observed association to causation or risk is ignorant and erroneous. This article makes this mistake – as has every other study I have reviewed demonising red or processed meat over the past year such as this or this.

The studies used in this article

There have been two large studies in America where people have been asked to record dietary intake, smoking, activity, weight and many other factors over a long period of time. The data from these two studies has been analysed retrospectively to look for patterns. This was not a study designed to review meat consumption over a period of time – some data just happens to be available and it has been reviewed to make headlines about meat consumption.

The two studies are the Health Professionals Follow-up Study (1986-2008) (abbreviated to HPFS) involving 49,934 men and the Nurses’ Health Study (1980-2008) (abbreviated to NHS) involving 92,468 women. A number of participants from these two studies were excluded in this meat review. After excluding people with cardiovascular disease (CVD) or cancer at the start of the study and excluding people whose dietary responses were incomplete, this article proceeded to review data from 37,698 men in the HPFS and 83,644 women in the NHS. Diet was assessed by validated food frequency questionnaires and updated every 4 years.

The dietary questionnaire offered 9 possible responses for meat consumption, ranging from “never or less than once per month” to “6 or more times per day.”

Unprocessed red meat was assumed to be “beef, pork, or lamb as main dish” (pork was queried separately beginning in 1990), “hamburger,” and “beef, pork, or lamb as a sandwich or mixed dish.” The standard serving size was 85 g (3 oz) for unprocessed red meat. As this was an American study, the great American hamburger has been included in unprocessed meat – it is of course as processed as meat can be. Hamburgers account for approximately half of American ‘beef’ consumption[i] and should be categorised as processed meat. If someone has had a beef sandwich or a pork kebab or a lamb curry – this has also been deemed unprocessed meat. Hardly what Paleo types would call real meat!

Processed red meat included “bacon” (2 slices, 13g), “hot dogs” (one, 45g), and “sausage, salami, bologna, and other processed red meats” (1 piece, 28g).

The Data – Table 1

Table 1 ( has the raw (baseline) data for the two studies separately categorised into quintiles for total red meat consumption (processed and unprocessed meat lumped together). The five quintiles take the lowest fifth consumption of red meat and then the next lowest and then the middle of the five groups then the second highest and then the highest. Table 1 is age standardised (to remove the impact of any age differentials between the different five groups of red & processed meat consumption) and it then lists other characteristics of the five groups.

Here is where the first problem emerges. As you can see for yourself in Table 1, Q1 is the lowest red & processed meat intake and Q5 is the highest. There are many other variables that correlate to the groups Q1 to Q5 – this is for the HPFS – the top part of Table 1:

– Physical activity, as measured by hours of metabolic equivalent tasks, falls from 27.5 in Q1 to 22.7 in Q2 to 20.2 in Q3 to 18.8 in Q4 to 17.2 in Q5. As red & processed meat consumption increases, so exercise falls. Could lack of exercise impact mortality?

– Body Mass Index – the average BMI for Q1 was 24.7; the average BMI for Q2 was 25.3; for Q3 it was 25.5; for Q4 it was 25.7 and for Q5 it was 26. As red & processed meat consumption increases, so does BMI. Could BMI impact mortality?

– Smoking – the percentage of people in Q1 who smoke was 5%; in Q2 it was 7.3%; in Q3 9.8%; in Q4 11.3% and 14.5% in Q5. As red & processed meat consumption increases, so does smoking – the top quintile virtually three times higher than the lowest. Could smoking impact mortality?

– Diabetes – the percentage of people in Q1 and Q2 with diabetes was 2%; in Q3 it was 2.2%; in Q4 2.4% and 3.5% in Q5. As red & processed meat consumption increases, so does diabetes. Could diabetes impact mortality?

– The interesting one was cholesterol. 14.8% of Q1 were recorded as having high cholesterol; 11.1% of Q2; 9.7% of Q3; 9% of Q4 and 7.9% of Q5. So, as red & processed meat consumption increases, cholesterol recorded as high fell. Could low cholesterol impact mortality? Given the protective nature of life vital cholesterol and the repair role that it plays in the body, it is highly likely that high cholesterol is protective against cancer and heart disease. Quite the opposite of what we have been led to believe in the interests of statin and plant-sterol-injected-low-fat spread profitability.

– Total calorie intake – the average daily calorie intake for Q1 was 1,659; the average daily calorie intake for Q2 was 1,752; for Q3 it was 1,886; for Q4 it was 2,091 and it was 2,396 for Q5. As red & processed meat consumption increases, so does calorie intake. Could calorie intake impact mortality?

– Alcohol intake – in Q1 an average 8.4 grams of alcohol were consumed daily; in Q2 this was 10.7; in Q3 it was 11.2; in Q4 it was 12.4 and 13.4 grams of alcohol were consumed daily in Q5. As red & processed meat consumption increases, so does alcohol intake. Could alcohol intake impact mortality?

The Nurses Health Study showed exactly the same correlations – the numbers were slightly different but the trends were the same. As red and processed meat consumption increased so exercise and high cholesterol fell; BMI, smoking, diabetes, calorie intake and alcohol intake all increased.

Table 2 looks at all mortality (I will stay at the all mortality level – the study does not stand up to scrutiny at this level so there is no point looking at cardiovascular (CVD) mortality vs. cancer mortality).

Table 2

Table 2 presents mortality data per quintile. The high level numbers are that:

– The HPFS covered 758,524 person years and there were 8,926 deaths in total: 2,716 attributed to CVD and 3,073 to cancer.

– The NHS covered 2,199,892 person years and there were 15,000 deaths in total: 3,194 attributed to CVD and 6,391 to cancer.

– The two studies combined, therefore, covered 2,958,416 person years and there were 23,926 deaths in total: 5,910 attributed to CVD and 9,464 to cancer.

The first point to make, therefore, is that the overall death rate was very small:

– In the HPFS, in 758,524 person years the overall death rate was 1.18% and the CVD death rate was 0.36% and the cancer death rate was 0.41%. Over a 22 year period, just over one in a hundred members of the study died.

– In the NHS, in 2,199,892 person years the overall death rate was 0.68% and the CVD death rate was 0.15% and the cancer death rate was 0.29%. Over a 28 year period, approximately one out of 150 members of the study died.

– In the two studies combined, in 2,958,416 person years the overall death rate was 0.81% and the CVD death rate was 0.2% and the cancer death rate was 0.32%. In the combined studies, fewer than one person in one hundred died in a 28 year period.

Table 2 is then supposed to have adjusted for all the other factors noted under the analysis of Table 1. The article says that the multivariate analysis adjusted for energy intake, age, BMI, race, smoking, alcohol intake and physical activity level. However, I don’t see how this can have been done – certainly not satisfactorily.

In Table 2 the raw data for deaths per person years for each quintile is presented. I have done a raw ratio (marked Z) on these numbers to show the following:

Health Professionals Follow up Study

  Q1 Q2 Q3 Q4 Q5 TOTAL
Total meat Deaths 1,713 1,610 1,679 1,794 2,130 8,926
person yrs 151,212 152,120 151,558 152,318 151,315 758,524
Death Rate (Z) 1.13 1.06 1.11 1.18 1.41
Multivariate (*) 1.00 1.12 1.21 1.25 1.37 1.14
Unprocessed Deaths 1,855 1,722 1,535 1,819 1,995 8,926
person yrs 150,676 149,097 154,352 150,925 153,474 758,524
Death Rate (Z) 1.23 1.15 0.99 1.21 1.30
Multivariate (*) 1.00 1.11 1.14 1.20 1.29 1.17
Processed Deaths 1,917 1,395 1,661 1,717 2,236 8,926
person yrs 171,619 131,069 152,481 152,128 151,227 758,524
Death Rate (Z) 1.12 1.06 1.09 1.13 1.48
Multivariate (*) 1.00 1.06 1.15 1.18 1.27 1.18

Above, I have simply taken the raw number of deaths for each quintile over person years and then calculated this as a ratio. The Multivariate line is the one presented in Table 2 of the article. It is the alleged comparison between the five quintiles – using quintile 1 as the base of 1.00 and relating the other quintiles to this base number. This multivariate line is supposed to have adjusted for the fact that exercise and cholesterol went down and BMI, smoking, diabetes, calorie intake and alcohol intake all increased alongside red and processed meat consumption. It is supposed to have removed all those correlations to isolate meat consumption alone.

My death rate line (Z) should therefore have all the other variables included and the multivariate line should have excluded all the other variables. The multivariate line should therefore be substantially below my death rate line (Z) for every quintile and it isn’t. Indeed the raw data for deaths per person years shows that the death rate was lower in Q2 and Q3 than Q1 for total meat, unprocessed meat and processed meat. Look at unprocessed meat (not withstanding that this includes hamburgers and other junk that it shouldn’t) – the death rate in quintile 3 (Q3) is 0.99 vs 1.23 for Q1. As meat consumption increases from Q1 to Q2 and Q1 to Q3, so the death rate falls. Only in Q4 and Q5 does this reverse and it is in these quintiles that we saw the highest levels of BMI, smoking, low activity, high calorie intake, high alcohol intake and so on and these have clearly not been adequately allowed for.

The nurses study shows exactly the same pattern. The death rate falls in Q2 and Q3 vs. Q1 in all cases. In fact even Q4 is lower than Q1 in all meat groups. Only Q5 is higher than Q1 on my ratio of raw data and this is with none of the smoking, exercise, weight, diabetes, alcohol having been allowed for.

Nurses Health Study

Q1 Q2 Q3 Q4 Q5
Total meat Deaths 2,946 2,759 2,658 2,872 3,765 15,000
person yrs 438,326 442,134 439,712 440,329 439,391 2,199,892
Death rate (Z) 0.67 0.62 0.60 0.65 0.86
Multivariate (*) 1.00 1.08 1.11 1.18 1.24 1.11
Unprocessed Deaths 3,079 2,885 2,545 2,709 3,782 15,000
person yrs 441,041 441,207 439,306 431,097 447,240 2,199,891
Death rate (Z) 0.70 0.65 0.58 0.63 0.85
Multivariate (*) 1.00 1.07 1.07 1.10 1.19 1.10
Processed Deaths 3,076 2,799 2,778 2,814 3,533 15,000
person yrs 442,594 420,403 455,365 441,369 440,161 2,199,892
Death rate (Z) 0.69 0.67 0.61 0.64 0.80
Multivariate (*) 1.00 1.04 1.08 1.14 1.20 1.21


The headline of the article

The key passage in the press release that attracted all the headlines was this:

“Unprocessed and processed red meat intakes were associated with an increased risk of total, CVD, and cancer mortality in men and women in the age-adjusted and fully adjusted models. When treating red meat intake as a continuous variable, the elevated risk of total mortality in the pooled analysis for a 1-serving-per-day increase was 12% for total red meat, 13% for unprocessed red meat, and 20% for processed red meat.”

This is what led to the big news story: “adding an extra portion of unprocessed red meat to someone’s daily diet would increase the risk of death by 13%. The figures for processed meat were higher, 20% for overall mortality…”

These numbers come from the bottom lines in Table 2 in the article. The bottom three lines in Table 2 come from the authors of the article combining all deaths in both studies from the multivariate model. They state that, using Q1 as the base line (1.0), the relative results for the other quintiles are as follows and they have added in a final column claimed to be the risk factor for increasing consumption of total, unprocessed or processed meat by one serving a day:


Q1 Q2 Q3 Q4 Q5 Risk factor
Total meat 1.00 1.10 1.15 1.21 1.30 1.12
Unprocessed 1.00 1.08 1.10 1.15 1.23 1.13
Processed 1.00 1.05 1.11 1.15 1.23 1.20


The 13% at the end of the unprocessed line is where the 13% headline comes from and the 20% at the end of the processed line is where the 20% comes from. I don’t know precisely how they have come up with these numbers. The corresponding consumption for each quintile was 0.25, 0.61, 0.95, 1.36 and 2.07 servings per day (for the HPFS). I suspect that their model allows them to look at the data for 1 serving vs 2 or half a serving vs one and a half and to compare ratios in this way.

None of this, however, reflects the facts from the raw data that Q2 and Q3 have lower death rates than Q1 in both studies and Q2, Q3 and Q4 are lower than Q1 in the Nurses study.

In summary

There are numerous key problems with this study – I’ll share seven:

1) This study can at best suggest an observed relationship, or association. To make allegations about causation and risk is ignorant and erroneous.

2) The numbers are very small. The overall risk of dying was not even one person in a hundred over a 28 year study. If the death rate is very small, a possible slightly higher death rate in certain circumstances is still very small. It does not warrant a scare-tactic, 13% greater risk of dying headline – this is ‘science’ at its worst.

3) Several other critical variables showed correlation with death rates – lack of activity, low cholesterol, BMI, smoking, diabetes, calorie intake and alcohol intake. These have not been excluded to isolate meat consumption alone. The raw data actually shows deaths rates falling with increased meat consumption up to the third or fourth quintile – and this is before all the other variables have been allowed for. This would suggest that meat consumption has a protective effect while weight, alcohol, calorie intake, lack of exercise and so on are all taking their toll.

4) Several other critical variables were not measured, which would logically correlate with certain meat consumption. Unprocessed meat inexplicably included sandwiches, curries, hamburgers (which come in buns) – has the correlation with bread, margarine, white rice, egg fried rice, poppadoms, burger buns, ketchup, relish or even fizzy drinks been correlated with the death rates? Indeed, Frank Hu, one of the authors of this meat study, is also quoted in today’s paper saying that one soft drink a day raises the risk of heart attacks.  It doesn’t of course – it is association at best, just as the meat article is – but one does wonder if that harmful soft drink was the one that just happened to be consumed with the hamburger or the bacon, lettuce and tomato sandwich ‘meal deal’?!

5) Hamburgers and pork sandwiches or lamb curries have been included as unprocessed meat. This is not a study of what real food devotees would consider unprocessed meat therefore. May I suggest that a study of consumers of grass fed ruminants would not deliver the desired headline? The lamb and beef grazing in the fields around me in Wales could not be further in health benefits from the hamburgers in buns and hot dogs in white rolls in fast food America.

6) We are all going to die. We have 100% risk of it in fact. We are not going to increase this risk by 13% or 20% if we have a hamburger and certainly not if we have a grass fed nutrient rich steak. This is headline grabbing egotistical academics doing their worst.

7) As I always consider conflict of interest, it would be remiss of me to end without noting that one of the authors (if not more) is known to be vegetarian and speaks at vegetarian conferences[ii] and the invited ‘peer’ review of the article has been done by none other than the man who claims the credit for having turned ex-President Clinton into a vegan – Dean Ornish.[iii]

All of this nonsense has given me an appetite, so I’m off to get my complete protein and essential fats plus the full range of B vitamins, ample fat soluble vitamins and lashings of iron, phosphorus, magnesium and zinc – also known as grass fed steak!




160 thoughts on “Red meat & mortality & the usual bad science

  • Pingback: PC Day 2 - Spaghetti, Paleo Style - Boulder CrossFit | Boulder CrossFit

  • Pingback: Workout Strategy: CrossFit Games 12.4 | CrossFit Ventura | West Coast Strength and Conditioning |

  • To Matt:

    World Cancer Research Fund is an advocacy group that made a presentation. That is in NO WAY a peer-reviewed study, because an ‘expert report’ isn’t the same thing as a peer-reviewed study. They didn’t even call the information they gathered ‘peer-reviewed’. They just called it ‘literature’. That doesn’t mean they went to REPUTABLE and EVIDENCED sources, you know. Are you familiar with the expression ‘Garbage In, Garbage Out”? Because that’s ALL that these moronic stats juggling studies give us. They don’t DO science. They nit-pick it.

    Give me the PEER-REVIEWED STUDY that isn’t a statistics juggle, but is based on OBSERVATION and EXPERIMENTATION. You know: SCIENCE.

    Because, so far, all we’ve seen is statistics juggling and very little SCIENCE.

    I’ll believe science. Cough some up, or STFU about what some group with an agenda is up to.

  • Pingback: Red meat & mortality & the usual bad science | Exigency In Specie

  • And also, what absolute rubbish from ‘Dr. James’. Go and look at the World Cancer Research Fund’s expert report, which was prepared by over 200 of the worlds leading scientists from 30 different countries, who summarized over 7000 peer-reviewed studies. These studies were not just correlational in nature, but included experimental and mechanistic studies.

    Your critiques do not outweigh over 7000 studies. This is pseudo-science as far as I am concerned.

  • Err…The recent Harvard study did control for all of the possible confounds that you have mentioned, and still found statistically significant correlation. I am not sure which study you are reading….

  • I think Dr. James Carison is a posing as a fraud over here. I have also practiced in my life, and I have seen the reverse. I have seen vegans and vegetarians living healthier and better than non veg people. So, the entire argument of “Dr. James Carison” is flushed out of the picture. I wonder how he got his degree??

  • Here is a link to the full study. I was able to get access to the full article for free, though it might be because I was on campus:

    It is a little down the page here (Red Meat Consumption and Mortality):

    I would like to hear the authors’ defense of their statistical methods. I have not seen a post of the authors’ response to Zoe’s questions to them.

    On the top of the right column on page 2 of the article they discuss (I use that word lightly) how they separated out meat consumption from the other independent variables (BMI, smoking, etc…. the dependent variable is mortality as determined by their scale). They used a multivariate technique. They don’t describe the specifics of the technique they used (there is not just one way to do a MANOVA). The best I can tell they standardized the independent variables (other than meat consumption) to meat consumption, making their influence in the model linear to that of the meat consumption. So, theoretically, they were calculating just the effect of meat consumption on mortality when they did their univariate analysis.

    However, there is a BIG problem. They fail to account for the multicollinearity of the independent variables. If there is a high degree of multicollinearity even among two variables, then the whole thing falls apart unless you exclude those variables. Gee, could hypertension be highly correlated with high cholesterol? How about diabetes with activity level? How about activity with hypertension? I could go on, but you get the picture. AND, they even ADMIT to this collinearity on the bottom right of the third page (first paragraph of Results).

    So, their conclusions are based off of INVALID statistics. If their research question was: “what factors/interactions of factors out of a,b,c,x,y,z are associated with mortality?” this study would be perfectly fine. But it was: “what is the quantitative relationship between red meat consumption and mortality?” It is impossible to get rid of the collinearity among the independent variables statistically. And even if you could, you still have other factors that relate to mortality that were not measured (mental health, for example). It was a very bad research question. This is what happens when you come up with a research question based on pre-existing data, instead of coming up the question first then designing an actual experimental study.

    They could have tried a factorial regression model that integrated all of the independent variables and the effect each one had individually, and all possible interactions of them interacting with each other, on mortality. But either they were too lazy, too dumb, or they knew if they did that, that one of the first IVs to drop out of the model (you eliminate variables/variable combinations from the model until only the ones that really effect the independent variable are left) would be red meat consumption. And there goes their headlines.

    They did actually do a regression (bottom of page 2, the Cox model). But they did it among red meat and what they considered risk factors (cardiovascular mortality and cancer mortality) of death. In other words, meat consumption was the independent variable. This excludes every other factor in the universe that could have influenced cancer and cardiovascular mortality in the subjects. And we’re supposed to believe this?

    At the end of the results they confess there were different numbers of people in each of the quintiles for total meat intake. With a non-even distribution of measurements along the x-axis (meat consumption) the data does not meet one of the requirements for doing a regression. They also said they calculated things (glycemic load, fiber, magnesium, different types of fats – middle right of page 2) that there is no way they could have known based on the description of their description of the questionaires. Did they just make this data up, or did they not do a good job describing the questionaires?

    This was a very, very, very questionable study in terms of how the data was analyzed (I didn’t even get into the problems with how it was collected). I’m half tempted to write a letter to the editor. This is normally a good journal (though they do seem to have a War on Meat in general). When I read the Methods of the study I wondered to myself if there could have been a conflict of interest here with the reviewers and/or editor(s) because the study is really awful. I then find this blog, and find out that it is very likely the case, and I am not at all surprised.

    • Hi Brian – thank you SO much for the link and your brilliant comments. As you say – for some reason this group has got it in for red meat and will try to find an association between red meat and anything they can conjure up no matter that this does not mean causation and ignoring (as you point out) interconnected variables.
      I hope you do write to the editor!
      Best wishes – Zoe

  • I have two (extra) problems with this study:

    Assuming that the authors did include all these other possible confounds in their model, why not just report the effects of each of them — why only report the effects of meat? A typical multiple regression (or it’s equivalent here — I am not familiar with the Cox analysis) could tell you the independent contribution of all these different factors to mortality, and allow the reader to compare them.

    Also, it is a shame, given that the main axis of the debate is whether fat or carbs are worse for you, that the authors don’t include carb intake as an additional factor / confound. It seems they included everything but. Surely the food frequency questionnaires have this information, so why didn’t they make use of it? Presumably someone else can re-analyse the same data in order to do this. Where’s Denise Minger when you need her?!

  • I’ve enjoyed John Miklavcic’s comments thus far, if admittedly not being able to understand most of it. But I get the impression he knows quite a bit about statistics, and that is the knowledge required to debate a paper such as this. My hunch is that if there is a major flaw in this paper, it most likely lies in the statistics. And due to the complexity of the methods that the authors have employed, I would be hesitant to trust a non-statistician to analyze this paper.

    Also John is right to question an MD’s grasp of stats. I am a US medical student at a “respected” institution, and we only get a smattering stats…at best. Sure, I can look for the major biases in a clinical trial, I can interpret a funnel plot for meta-analyses…but nothing NEAR the level of stats this paper employs. I HIGHLY doubt ANY MD program covers advanced, or even mid-level stats…there is just not enough time!

    Dr. Carlson, it is interesting to hear your experiences. Based on your observations, these really sick people do better on a high meat diet. But I wonder how “normal” (aka normal blood glucose, LDL, TG, etc.) fair on a low meat vs high meat diet.

  • Wow,

    Zoe got destroyed on her poor interp of this article. The only bits I’m seeing for defense are people who throw around tough language and try to point others as unethical to defend the thesis of their websites, diet program, books, and celeberty.

    It all comes down the money… indeed.

    How are book sales Zoe?

  • Hi! What you all have not mentioned is ,that all meat have high hormones levels, meat is washed with ammonia to kill bacteria therefore ammonia is also found in meat and inside our bodies and not to mention the preservatives some of which carcinogenic and antibiotics.

    Plus, butchers never throw away left overs of meat, so they put it together or I should say they glue it together with a with white powder called ;”Transglutaminase, aka Meat Glue ” a little dirty secret of the meat industry. I’m talking about meat glue, also known as transglutaminase, which restaurants and food producers use to create “steaks” out of “glued-together” stew meat, add body to dairy products, make imitation crab, improve processed meat mouth feel, to name a few. You should know that if you happened to breath this powder, you will end up in hospital imagine if you ingest this stuff. Anyway, in regards to protein intake your body will only process 30% of it and in the process stays inside your guts for several days if not week sitting there fermenting hence all the accumulation of toxins. Long term cause cancer heart problem, prostate etc.

    Beside, most people cannot cook a proper steak, they barbecue it until the outer skin is black. I must say that I have seen it so many time it is not funny. So, what does it do to your body when you ingest that black stuff around the meat??????

    If the meat is purely grass fed I think is ok otherwise do not eat meat at all.

    Take care of you health and you body will thank you.

  • First off, let me commend Zoe on an excellent article correctly analyzing that ridiculous study! As a Board Certified Family Physician and Clinical Biochemist with 20 years of experience treating actual, live, breathing, real human beings who suffer from obesity, heart disease, type 2 diabetes and who at times succumb to cancer; I knew as soon as I heard and read about that study that it should never have been published.


    Because the focus was on the consumption of red meat (processed and/or unprocessed)and the statement that even a small serving of red meat a day statistically increases(d) one’s risk for that ever present downer known as death was a ridiculous and yes ‘arrogant and erroneous’ comment made by researchers who just wanted to get ‘something’ published.

    (Yes, the cynical side of me upon learning of the results of this study immediately downgraded the ‘Internal Annals of Medicine’ to ‘The Internal Anals of Medicine.’)

    Well here’s the problem and it’s a BIG one. I am in the front lines of medicine and i treat patients on a daily basis with the disease as delineated above. The problem is that when my patients eat MORE red meat (or meat in general, even if it is processed, I can hear the ‘OMG’ sigh emanating from all the readers at this point) the disease process they are dealing with gets BETTER ladies & gentlmen NOT worse! The MORE my patients consume red meat, and meat in general, and the LESS fruits and whole grains and some veggies likes carrots and corn they eat——THEY GET BETTER, NOT SICKER!!!!!

    How do I define better? Let’s see, simply stated, the MORE meat and the less whole grains and fruits consumed equates with lowering of blood sugars to a normal range, lowering of elevated triglycerides to a normal range, raising of HDL (the protective cholesterol), loss of weight, normalization of blood pressures—ALL USUALLY WITHOUT THE USE OR NEED FOR MEDICATIONS!!! I’m using bold here people because we can play around all we want with multivariate statistical analysis lines/plots/curves/Z lines or whatever, but the fact of the matter is exactly what Zoe stated–to make the claims as stated in this article is indeed ‘ignorant and erroneous’ and downright deadly!

    What was that statement, ummmm “There are liars, there are damn liars…and then there’s statistics.” Mark Twain

    My major problem with this article (by the way I cannot even use the word research because I’m sure the ‘proud authors’ of this paper probably just plugged some numbers in their Excell Spreadsheets to come up with all this ‘crappy’ data) is that not only will most doctors, nutritionists and dieticians buy into the fallacious results; the real vicitms will be the vast multitudes of people who will be told that eating red meat is bad for them, wont eat it because they think that to be true, and then will go on to develop uncontrolled high blood pressure & blood sugars, high fats in the blood, low HDLs and the list goes on.

    Gee, what a shocker that Dean Ornish was one of the ‘reviewers’ of this article. He takes credit for converting the ex-president into a Vegan, does he also take credit for Clinton’s heart attacks and failing health?

    I’ve said it before and I’ll say it again—“Vegetarians are the sickest people in my practice and they require the most medications to treat their myriad disease processes.”

    …and they wonder why the main title of my book is GENOCIDE…

    Thank you for your time.

    Dr. James Carlson BS, DO, MBA, JD

    • Wow Dr James – love it! Someone who gets as passionate about this stuff as I do. Must check out your book
      Many thanks for this – Zoe

      • Zoe do you agree with above poster that Vegetarians are the sickest people and if so would you discourage people from becoming Vegetarian ?

  • Hi John,

    Thanks John.I know that. I thought the 13% came from the 1.13 HR. Where did that 13% come from?

  • Anoop, David didn’t get an HR. Hazard ratios are different from what David described: risk ratio (and odds ratios are something different as well).

  • Zoe writes, in reply to me:

    “Hi David – my reply to Matt from a few days ago may help – I was following the person years lead of the article, as that’s how the authors chose to present the data.”

    What you wrote, however, and I quoted, was:

    “Over a 22 year period, just over one in a hundred members of the study died.”

    Do you agree that that statement was wildly inaccurate? It would have been accurate if you had written “died per year.”

    You also wrote:

    “In the combined studies, fewer than one person in one hundred died in a 28 year period.”

    Again, do you agree that that statement was false–and, of course, inconsistent with the figures you had cited shortly before? Again, adding “per year” would have corrected the error.

    That may have been only careless writing–but correcting it eliminates what seemed to me, when I first read your piece, to be the strongest of the arguments you were offering–“The numbers are very small.” The per year numbers may be small–and the per month, per week, or per day figures smaller still–but none of those is the relevant figure. A reduction of 13% in a mortality rate of 1% per year may sound small, but it represents an increase in life expectancy of more than three and a half years, which is not so small.

  • 1. The study did not made such allegations, only the media.
    2. Others have pointed out your mistake here.
    Even if the death rate was low, your logic that “a small increase in the death rate
    is meaningless” is flawed.
    3. As I wrote above, your “raw” calculation is meaningless.
    To elaborate, consider the Health Nurses Health Study.
    While the average age in Q2 and Q3 is the same, the distribution of ages can be different.
    This explain why in Q3 the “death rate Z” is lower than Q2 (0.60 vs 0.62), while
    the “Age-adjusted model” ratio is higher (1.09 vs 1.07).
    To illustrate why this happens, consider the following example.
    Suppose we follows 2 groups of 1000 women each for 10 years.
    The average age at baseline in each group is 40.
    In group 1, 30 women died, while in group 2 only 20 died.
    Does this mean that the women in group 1 have higher risk of death?
    Now, suppose that in group 1, 20 women were at age 70 at baseline and 980 women were at age 39.4 at baseline.
    In group 2, all 1000 women were at age 40 at baseline.
    Out of the 30 deaths in group 1, 10 were women of age 70 at baseline, and the rest 20 deaths were women
    of age 39.4 at baseline.
    We can see that the women of age 39-40 have similar death rates in both groups, while the women of age 70
    have higher rate (which is expected).
    This means that the risk of death is the same in both groups.

    This also explain why the multivariate model can increases the risk ratio when you expect it to decrease
    the ratio (for example in Q3 of the HPFS): looking at the quintiles AVERAGE exercise level,smoking level,BMI,etc does not tell the whole story. The distribution is also important.
    4. Other variable were measured, and the researchers tried to do additional adjustment which did not
    change the result (see the paper).
    However, more importantly, your criticism here also applies to EVERY epidemiological study. There can always be
    an unadjusted factor.
    6. Again, the logic here is flawed.

  • These gross misinterpretations are the exact reason why Kaplan-Meier (step-wise) plots are used for survival analyses, not elementary linear Cartesian plots, as suggested above and performed by Dr. Kendrick. If I “replicate what Dr. Kendrick did,” I would be performing survival analysis incorrectly. It is poor practice to infer conclusions regarding survival analysis by extrapolating linear cohort data.

    I’m with Zoe’s graph for every step: “If life extension is not calculated in a paper it can be estimated … If there is adequate follow-up time to the death of the median patient, the difference in median lifetimes of placebo and treatment group can be quoted as the median life extension (Malkin. EJHF. 2005. 7:143-148),” using the step-by-step approach outlined appropriately by Zoe above.

    Consider first that the mean and the median can be very different statistics:
    While the median individual may have life extended (with treatment) by 1 year (over placebo), the average (mean) individual may have life extended by 5 years.

    Let us continue with “…limitations to this method. The first is that … studies do not continue follow-up to the death of the median patient. … trials without follow-up to the death of the median patient should be discouraged since a continually diverging curve suggests persistent and permanent benefit … (Malkin. EJHF. 2005. 7:143-148)” [Addressing this limitation would require following treatment and placebo groups until the (median individual or) cumulative survival was 50% (as the median is the 50th percentile) or less for both groups. It should be noted that following a cohort until an abstract number of participants are alive or dead is unethical and will not be approved by research ethics boards].

    For a continually diverging curve, see this example: [the empty (non-shaded) squares and empty circles illustrate continually diverging curves]. To understand “permanent benefit,” consider that the empty squares could represent the “statin” group in Zoe’s example and (for the sake of modesty,) the empty circles could represent the “placebo” group (because choosing any of the other data plots would exaggerate my point). Up to week 40, only a modest difference in survival exists; but at week 100 and beyond, there is a permanent benefit as placebo group survival continues to decline, while statin group survival is steady.

    Noting that the statin group survival is not changing beyond week 100 is key to addressing the flaw in Zoe’s method. Based on the data from weeks 0 to 40, Zoe would draw a (straight) line with negative slope for statin group. Over time (at about week 100), that line extends below 75% cumulative survival; but we can astutely note that at week 100 (75% cumulative survival), all individuals in statin group are still living for the next 2 years.

    Ergo, I would not counsel a patient as follows: “the expected increase in lifespan from taking statins is ~50 weeks.” I would instead, counsel as: “After 2 years of statin use, 75% of individuals are still alive. On the other hand, of individuals who elect to forego treatment, 85% perish within 2 years.” The responsibility of the physician is to advise based on evidence established from a population basis for improved mortality, not from extrapolation and conjecture of (incomplete) median survival data.

    Another “limitation is the right hand tail of Kaplan–Meier curves are based on fewer data (because of cumulative patient deaths and withdrawals) and are consequently less reliable. There is therefore increasing error in estimates further down (beyond) the curve [Pocock. 2002. Lancet. 359:1686-1689].”

    I do not deem Dr. Kendrick a credible source to manage, analyze and interpret data. After reviewing curricula for eleven MD training institutes (among, Canda, USA and UK), I see that, at most, only one course in basic statistics is taught to MD-trainees (throughout the course of their program).

    10,000,000 is not far-fetched, as approximately 20,000,000 Brits have high cholesterol. If 10,000,000 is far-fetched, then it’s unreasonable to believe that if all Brits with high cholesterol showed up to their doctor that even half would be treated. I certainly hope that’s not the case. If I’m developing symptoms of Alzheimer’s, I’m hoping my doctor wouldn’t flip a coin to decide whether or not he will prescribe treatment to me.

  • Your “Death Rate (Z)” numbers are meaningless. You need to account to the different ages of the persons and the different follow-up times. The real “raw” numbers are presented in the paper under the “Age-adjusted model” line.

  • Zoe, how do you suppose they calculated the hazard ratios (HR) in each quintile in figure two? How do you suppose these differ from your contrived calculation for mortality rate? ( simply Adding up the ‘man-years’ to substitute ‘unit of time’, really?).

    Finally, how do you consider your non-standardized mortality rate can be compared to their HR (per quintile or total) in any meaningful way?

    Suppose I had a velocity of 5 m/s and an acceleration of 10 m/s^2. The number 5 is less than 10. What could that comparison be used determine information about my displacement? Nothing.



  • The average death rate in USA is 0.78%. Consider that the death rate for men is higher than that of women (while these statistics are often presented in raw format, one will have to consider that there are more females than males in our population, and that standardization to proportion of sex will yield higher mortality for men than women). Consider that NHS looks mainly at women, and therefore, the mortality rate in USA of 0.78% (that averages men and women) is inapplicable, and affects the interpretation above.

    10 million is not “an awful lot of people to use as [your] denominator” with respect to statin use. 10 million is about 3% of the population of USA. A lot more than 3% of individuals residing in USA have high cholesterol (for which statins are used to treat). The number of individuals in USA with high cholesterol is probably closer to 30%. Therefore, 10 million is a very modest denominator considering the number of people in USA with high cholesterol is closer to 100 million.

    Dr. Kendrick requires survival analysis data to make adequate claims stated above. Heart Protection collaborative group studies did not collect survival analysis data. The difference: survival analysis follows a subject until “event” or death. Survival analysis is usually performed in terminally ill cancer patients for the comparison of new treatment vs. standard treatment. Analysis (until “event” or death) can then adequately determine how effective intervention was with respect to delaying time until death. I reiterate that The Heart Protection Study was NOT performed in this manner and therefore, Dr. Kendrick cannot claim that “if people take a statin for thirty years, this could lead to an average increase in lifespan of approximatelytwo months;” this is a gross misinterpretation.
    I guess that’s a great thing about writing one’s own book, statements made do not need to pass through credible peer review as they would for publication in scientific literature.

    • Hi John – Dr Kendrick and I are Brits! The Heart Protection Study is a British study. Hence – to make claims for 10 million people over the pond is to make claims for approximately a quarter of adults. That’s pretty high (further – to claim that there are 10 million high risk Brits is pretty far fetched).

      Any study can only report numbers of deaths in a defined period. No study can predict when each participant will finally die and, if they follow them for a period of time until they do die, they become a reported death within a time period.

      Here’s how you could replicate what Kendrick did – get a bit of paper and draw out a simple L shape (x and y axis). The y axis represents number of people alive (starts at 100% and 0% at the bottom). The x axis represents time….. say 5 years. Then draw two lines, starting at the top of the y axis, both at 100%. Over time, the lines separate from each other. After five years one line reaches 85%, the other 87% (these numbers are just as an example). if this were a statin study it would mean that 15% on placebo had died, and 13% on the statin had died. If you assume that you started the study with 10,000,000 people (as the Heart Protection Study press release wanted us to), then 1,500,000 million on the placebo would be dead, and 1,300,000 on the statin are dead. A difference of 200,000 (after five years). So, you can state that 40,000 people per year will have their ‘lives saved’ by statins? Actually, no you cannot. You can only say that after five years 200,000 more people are alive on the statin than on the placebo. What you do now, is draw a horizontal line along the 85% mark, then continue the statin line and see how long it takes to reach the 85% mark. That is the extra time that the 200,000 lived for. BUT, that figure is not correct either. Because half of those 200,000 will be dead after half that time. so half of the time it takes for the 87% line to reach 85% represents the median increase in survival time (for the 200,000)…

      If you think that is a gross misinterpretation, what do you make of the press release from the so called credible peer reviewed (statin company sponsored) journal article, which said: “If now, as a result, an extra 10 million high-risk people were to go onto statin treatment, this would save about 50,000 lives a year – that’s a thousand a week.” !?
      Best wishes – Zoe

  • You write:

    “Over a 22 year period, just over one in a hundred members of the study died.”

    I do not believe that is even close to correct. You appear to have confused the death rate per year, which is less than 1%, with the death rate over the entire period, which is about 20% (not identical for the two groups, of course). Note that, according to the study:

    “During the study follow-up period of more than two decades, almost 24,000 of the participants died” (from one news story). That’s out of somewhat over a hundred thousand participants.

    • Hi David – my reply to Matt from a few days ago may help – I was following the person years lead of the article, as that’s how the authors chose to present the data. The overall death rate is 23,926 deaths out of 121,342 = 19.7%

      Interestingly (The International Network of Cholesterol Skeptics) are also looking at this unfeasibly low presentation of the death rate. They have spotted that a number of people were excluded from the study and they are trying to understand the impact that this statement had on the study: “We stopped updating the dietary variables when the participants reported a diagnosis of diabetes mellitus, stroke, coronary heart disease, angina, or cancer because these conditions might lead to changes in diet.”
      Hope this helps – Zoe

      Hi Matt – great question!

      The USA death rate for all causes for 2006 was 0.78% ( I just happen to have this figure analysed in my book The Obesity Epidemic) (Centres for Disease Control and Prevention, (Using age adjusted data), (data page),, (detailed list of data available)

      This means that c. 777 people died per 100,000 USA citizens in 2006 i.e. for that year. You are right that there were 23,926 deaths from the 121,342 included in this study. The larger part of the study (the NHS) was over a 28 year period. Hence, using the 2006 death rate, we should have expected 777 people in 100,000 to die or 943 for 121,342. Over 28 years, that should be 26,399 people whereas in fact 23,926 died – less than the 2006 average, but then death rates have been falling since the 1950′s. CVD and cancer are the two biggies and they do account for about a third each.

      The – what are the chances of dying in any particular year is – as you say – the key question and the criticism of these studies is that they imply that lives will be saved. This one concludes: “We estimated that substitutions of 1 serving per day of other foods (including fish, poultry, nuts, legumes, low-fat dairy, and whole grains) for 1 serving per day of red meat were associated with a 7% to 19% lower mortality risk”, which is nuts! All they are effectively saying is – in a given period of time, X people should have died and if they do Y then some of them won’t have. However, they may die the very next day after the end of the study period. The best I’ve seen this presented is by Dr Malcolm Kendrick in The Great Cholesterol Con (p193) where he exposes the arguments used for statins…

      He quotes from the Heart Protection Study that claimed “If an extra 10 million high-risk people were to go onto statin treatment, this would save about 50,000 lives a year – that’s a thousand a week.” Kendrick puts it as follows: “Leaving aside the point that this 50,000 figure actiually equates to one life ‘saved’ for every 200 people taking the statin – 10 million is an awful lot of people to use as your denominator – the concept of saving lives is not best chosen. In reality, taking a statin can only delay death, not prevent it. By how much? Well, if one in two hundred people are alive after one year of taking statins, this means that if you wait another two hundreths of a year (plus another little bit) the statin group will have caught up on the placebo group in total number of deaths. This represents an increased life expectancy of slightly under two days. It would be considerably more accurate to state that, if ten million people (at very high risk of heart disease) took a statin for a year they would live, on average, two days longer. If people take a statin for thirty years, this could lead to an average increase in lifespan of approximatelytwo months.” Not quite the same as saving 50,000 lives a year eh?!

      Night night – Zoe

  • Let the unthinking and impressionable masses believe these studies. More meat for me and my family.

  • Seeing that Crossfit has decided to soothe their millions of savage-Paleo-eaters by linking to this page as proof that the Harvard study is wrong – and – seeing that Eric Wick above seems to have successfully refuted the statistics you’ve used to do so, I wonder if you might respond….

  • Good to see a lively discussion with reference to the experimental evidence.
    However a refereed journal that published an article with the substantive methodological errors that you allege should be actively pursuing their resolution. Have you made your concerns known to the journal and the authors–and how have they responded?

    Discussing statistical analysis in a lay forum is a good way to share perspectives, but is not a sensible way to advance understanding. I know enough statistics to know that I do not know enough statistics to participate in such discussions.

    Best wishes.

    • Hi Rick – I have just emailed the correspondent for the paper – Frank Hu – with the following query and the two main tables from my blog enclosed. I’ll let everyone know the reply:
      Best wishes – Zoe

      Dear Dr Hu
      Please can you help me understand how the multivariate analysis was done in your study (

      The raw data in table 1 shows the death rate falling – with Q2 and Q3 lower than Q1 in the HPFS and Q2, Q3 and Q4 lower than Q1 in the NHS. This is while exercise is falling and BMI, diabetes, smoking, calorie & alcohol intake were all increasing alongside red and processed meat consumption. I would have thought that once your multivariate analysis had allowed for these (and many other) variables to try to isolate meat consumption, the multivariate index would be substantially lower than my death rate line, which includes these risk factors. As a comparison, if I index my death rate for the HPFS is would be 1.00; 93.43; 97.70; 103.97 and 124.26 (doing the same 1.00 base line thing) vs. the multivariate index of 1.00; 1.12; 1.21; 1.25 and 1.37.

      Many thanks
      Kind regards – Zoe

  • It should be pointed out that the differences in baseline characteristics across the quintiles are exactly that – baseline characteristics, present at the beginning of the study. This includes a number of factors which could have changed between 1982 and 2006. Therefore, suggesting that increased likelihood of dying in the highest meat consumption quintile was due to a higher rate of smoking is not a valid argument, as it doesn’t take account of what happened to smoking in that group over the next 25 years.

    While this information wasn’t presented, it was available to the study authors (from 4-yearly questionnaires) who used updated information on all covariates in their multivariate analysis which clearly showed red meat consumption was associated with an increased risk of dying during the study period once other confounders were adjusted for.

  • Thank you for delving into this study. Like many readers above, I am most interested in evaluating the validity of the study, not supporting my personal views on the topic.

    I wonder if the corresponding increase in hazard ratio with a reduction in deaths in Q2-3/Q2-4 is due to the younger ages of these cohorts and specific adjustments made for these values?
    HPFS/NHS: Q1=53.8/47.3 Q2=52.6/46.0 Q3=52.5/45.8 Q4=52.5/45.3 Q5=52.2/46.0

    Otherwise, I agree that it is hard to explain how fewer deaths can result in higher hazard ratio when these groups also exercise less, drink and smoke more, etc. than Q1 (factors that following adjustment should reduce the hazard ratio for red meat). Age adjustments may carry more weight than other factors since there is an unquestionable relationship between increased age and increased chance of mortality.

    Also, perhaps no conclusions about the inverse relationship of cholesterol levels with red meat intake/mortality can be made without knowledge of statin/medication usage in these groups as drugs may artificially lower these values. It is hard to imagine increased red meat (no doubt this is not the grass-fed stuff) + increased smoking, drinking, lack of exercise actually reduce high cholesterol without any adjustments.

  • Andres – I was just reading the comments and wanted to to commend you for the appology related to your comments. Zoe – I was impressed that you responded to the facts not the comments.

    At any event, I am going to continue to try to eat high quality meat – including fully bunned hamburgers. I know but I really like them. -Thad

  • Sorry,I got this wrong by a factor of 10 !

    Just proves one should think before mailing.

  • As an aside, the alcohol consumption is quoted in grams per day.
    This gives the nurses an average consumption of between 5.8 and 6.6 g/d
    A bottle of wine 12 per cent in 75 cc is about 9 grams. Do american nurses
    drink on average about 2/3rd a bottle of wine a day ?

    For males the average goes from 8.4 to 13.4
    Do american doctors drink an average bottle to bottle and a half of wine every day ?

    Is anyone ever sober at the surgery ?

  • Zoe,
    I’m afraid that I have to agree with Andre here. I believe it it you who have made the math error. The reasons that the original authors may be right, despite your observations of Q2-4 is that *YOU* have not controlled for the mix of other variables within Q2-4.

    I’m generally familiar with regression techniques, and while I have not read the paper, it seems to me that the scientists in questions were actually … err, scientific. And I think you also lost some credibility with the misinterpretation of the death rate thing.

    But meat is still yummy, and I still feel much better when I go low carb, high-protein. So, rock on! Just fact check youself a bit when you do so.



  • It is erroneous to think that the researchers who published “Epigenetic Differences in Normal Colon Mucosa of Cancer Patients Suggest Altered Dietary Metabolic Pathways” were “looking at genetic differences between people… and … found differences in the genes…”

    Genetic differences refer to alterations in the sequence of base pairs in the genetic code, which subsequently affect the products of transcription and translation.

    Gene methylation was measured in the study. Methylation refers to an epigenetic modification of DNA, irrespective of DNA sequence. Methylation affects rate of expression of DNA, and may even reduce expression to zero; yet, no inference can be made regarding genetic differences between individuals.

  • Another new study just released links, supposedly, saturated fat to colon cancer. It is in Cancer Prevention Research and was conducted by the folks at Temple. “Epigenetic Differences in Normal Colon Mucosa of Cancer Patients Suggest Altered Dietary Metabolic Pathways” Any thoughts?

    • Hi Kimberly – found the abstract here. Not planning to pay $35 for 1 day’s access to the full article!

      The abstract doesn’t mention saturated fat and my first query would be – what do any researchers classify as saturated fat and how do they isolate it from unsaturated fat? The UK and USA governments think pies, pastries, cakes, biscuits, ice cream, confectionery, savory snacks and so on are saturated fat. They are not! They are 1) processed foods 2) carbohydrates and 3) most of these have more unsaturated than saturated fat (not that one is better or worse than the other, but just saying). On the isolation point, not many people seem to know that every food that contains fat contains all three fats – saturated, monounsaturated and polyunsaturated – there are no exceptions. Hence they say eat less meat (70-75% water, c. 20% protein and can be as little as 1% fat – mostly unsaturated) and they have reduced unsaturated fat consumption more than saturated fat consumption. Then they say eat more fish – oily fish has twice as much total fat as a sirloin steak and more saturated fat than the steak. Nutritional ignorance corrupts many studies.

      Back to this abstract – the abstract alone doesn’t seem to be claiming much. “We have compared DNA methylation in normal colon mucosa between patients with colon cancer and patients without cancer. We identified significant differences in methylation between the two groups at 114 to 874 genes. The majority of the differences are in pathways involved in the metabolism of carbohydrates, lipids, and amino acids. We also compared transcript levels of genes in the insulin signaling pathway. We found that the mucosa of patients with cancer had significantly higher transcript levels of several hormones regulating glucose metabolism and significantly lower transcript levels of a glycolytic enzyme and a key regulator of glucose and lipid homeostasis. These differences suggest that the normal colon mucosa of patients with cancer metabolizes dietary components differently than the colon mucosa of controls. Because the differences identified are present in morphologically normal tissue, they may be diagnostic of colon cancer and/or prognostic of colon cancer susceptibility.” So they’re looking at genetic differences between people with colon cancer and people without and have found differences in the genes for metabolism of the different food stuffs – carbs, fats and protein. They seem to be suggesting that the genetic stuff they have identified could indicate who may go on to suffer colon cancer. Doesn’t seem to have much to do with fat, let alone saturated fat, without seeing the whole article.

      Best wishes – Zoe

  • 1. The baseline statistics suggest that participants in different groups had (on average) significantly different behavior patterns besides red meat intake. A group with lower red meat intake on average was more likely to adopt behavior thought to be healthier than a group with higher red meat intake. The value of such an observational study is dubious unless the authors can claim that ALL behavior variables that potentially impact the mortality risk have been included or the effect size is so large that it is unlikely to be caused by variables that were not included.

    2. The discrepancy between the raw death rates and the mortality risks obtained seems to be attributable to age difference among groups assuming the age adjusted analysis was done correctly.

    3. For HPFS, the average hazard ratio associated with red meat intake goes down after other variables (besides age) were added to the model. If you look closely, however, you can see that except for Q5, hazard ratio associated with red meat intake goes up after controlling for other variables. This apparent paradox also exists in NHS, although to a lesser degree. The authors should have explained its cause.

    4. The discussions following “Several mechanisms may explain the adverse effect of red meat intake on mortality risk. “ show a certain amount of tunnel vision. At the very least, a discussion on covariates not included in the study but could have an impact should have been included.

  • *sigh*

    I could expressed my thought more constructively, and instead I just added to the incommensurable amount of disrespect on the web. Hopefully we can continue having a fruitful exchange of ideas.

  • I apologize for my speech. I should not have called “liar”. Rude on my part. Sorry. In the future I shall control my emotions. Let the discussion continue.

  • in my post above, the word “straight” should be implied before every use of “line”

  • It is important to distinguish a line from a curve. The authors of the paper published in the Archives of Internal Medicine do not fit a line to the data, they fit a multivariate curve to the data, which accounts for the hormetic (see radiation and hormesis for a good explanation) relationship between meat intake and mortality as Zoe astutely noted. Other parties can try and fit a line to non-linear data unsuccessfully until their hair falls out.

    This appears to be an open forum for discussion and as such, should not discourage thoughts. “you are a liar,” and “if you read … properly” are cheap shots, below-the-belt. We all want the truth and insulting ego won’t help us get there.

  • So you say “3) Several other critical variables showed correlation with death rates – lack of activity, low cholesterol, BMI, smoking, diabetes, calorie intake and alcohol intake. These have not been excluded to isolate meat consumption alone.”

    Well, you are a liar.

    As a person who actually read the study carefully and didn’t miss the obvious and huge statistical efforts made to control for these variables, why don’t we read to together this:

    “They conducted analyses separately for each cohort. In multivariate analysis, they simultaneously controlled for intakes of total energy, whole grains, fruits, and vegetables (all in quintiles) and for other potential nondietary confounding variables with updated information at each 2- or 4-year questionnaire cycle. These variables included age; body mass index (calculated as weight in kilograms divided by height in meters squared) (<23.0, 23.0-24.9, 25.0-29.9, 30.0-34.9, or 35.0); race (white or nonwhite); smoking status (never, past, or current [1-14, 15-24, or 25 cigarettes per day]); alcohol intake (0, 0.1-4.9, 5.0-14.9, or 15.0 g/d in women; 0, 0.1-4.9, 5.0-29.9, or 30.0 g/d in men); physical activity level (<3.0, 3.0-8.9, 9.0-17.9, 18.0-26.9, or 27.0 hours of metabolic equivalent tasks per week); multivitamin use (yes or no); aspirin use (yes or no); family history of diabetes mellitus, myocardial infarction, or cancer; and baseline history of diabetes mellitus, hypertension, or hypercholesterolemia. In women, they also adjusted for postmenopausal status and menopausal hormone use.

    To better represent long-term diet and to minimize within-person variation, they created cumulative averages of food intake from baseline to death from the repeated FFQs.12 They replaced missing values in each follow-up FFQ with the cumulative averages before the missing values. They stopped updating the dietary variables when the participants reported a diagnosis of diabetes mellitus, stroke, coronary heart disease, angina, or cancer because these conditions might lead to changes in diet.

    They conducted several sensitivity analyses to test the robustness of the results: (1) they further adjusted for intakes of other major dietary variables (fish, poultry, nuts, legumes, and dairy products, all in quintiles) or several nutrients or dietary components (glycemic load, cereal fiber, magnesium, and polyunsaturated and trans fatty acids, all in quintiles) instead of foods; (2) they corrected for measurement error13 in the assessment of red meat intake by using a regression calibration approach using data from validation studies conducted in the HPFS9 in 1986 and in the NHS10 in 1980 and 1986; (3) they repeated the analysis by using simply updated dietary methods (using the most recent dietary variables to predict mortality risk in the next 4 years)12 or continue to update a participant's diet even after he or she reported a diagnosis of major chronic disease or using only baseline dietary variables; and (4) they used the energy density of red meat intake (serving/1000 kcalx�d–1) as the exposure instead of the crude intake. In addition, they used restricted cubic spline regressions with 4 knots to examine a dose-response relation between red meat intake and risk of total mortality.

    They estimated the associations of substituting 1 serving of an alternative food for red meat with mortality by including both as continuous variables in the same multivariate model, which also contained nondietary covariates and total energy intake. The difference in their β coefficients and in their own variances and covariance were used to estimate the hazard ratios (HRs) and 95% CIs for the substitution associations.14 They calculated population-attributable risk (95% CI) to estimate the proportion of deaths in the 2 cohorts that would be prevented at the end of follow-up if all the participants were in the low-intake group.15 For these analyses, they compared participants in the low–red meat intake category (<0.5 servings daily, or 42 g/d) with the remaining participants in the cohorts.

    The HRs from the final multivariate-adjusted models in each cohort were pooled to obtain a summary risk estimate with the use of an inverse variance–weighted meta-analysis by the random-effects model, which allowed for between-study heterogeneity. Data were analyzed using a commercially available software program (SAS, version 9.2; SAS Institute, Inc), and statistical significance was set at a 2-tailed α�=�.05."

    The full list of contolled variabes is:

    Body mass index
    Ethnicity (though this was weak, white or non white were the only two choices)
    Smoking status (never, past, current with current being subdivided into 3 quantities)
    Alcohol intake (4 gradations)
    Physical activity level (5 gradations)
    Multivitamin use
    Aspirin use
    Family history of diabetes
    Family history of myocardial infarction
    Family history of cancer
    Baseline history of diabetes
    Baseline history of hypertension
    Baseline history of hypercholesterolemia
    Menopausal status
    Menopausal hormone use
    Total caloric intake
    Quintiles of fish consumption
    Quintiles of poultry consumption
    Quintiles of nut consumption
    Quintiles of legume consumption
    Quintiles of dairy consumption
    Quintiles of dietary glycemic load
    Quintiles of dietary cereal fiber consumption
    Quintiles of dietary magnesium consumption
    Quintiles of dietary polyunsatured fat consumption
    Quintiles of dietary trans-fat consumption

    They found that red meat is still dangerous after carefully controlling for all of these variables.

    Sorry, but if you wrote an article talking about an obvious "missing" elephant in the room, where in fact the elephant in the room is that you didn't read the article, I cannot have intellectual respect for you.

    • Hi Andrés – they say that they allowed for all those things. If you read my article properly you will read that what I am questioning is how they could have done this. The raw data shows the death rate falling from Q1 and Q2 to Q3 in the HPFS and from Q1, Q2 and Q3 to Q4 in the NHS. This is the raw data – before allowing for the fact that exercise fell and BMI/diabetes/smoking/calorie & alcohol intake increased. I am saying that their multivariate line – after allowing for all of these factors – should be substantially less than my line marked Z – assuming that exercise is a good thing and high BMI/diabetes/smoking/high calorie and alcohol intake are not good things.

      How can they have allowed for all these risk factors to isolate meat consumption and then managed to reverse Q1/2/3 being lower than Q3/4?
      I’m writing to them to ask and can then do a follow up blog
      Best wishes – Zoe

      • You should have wrote to them and established how they allowed for these factors other than meat before you wrote this blog entry using your broad brush technique. Why because plenty of people are running from this blog thinking, I told you so red meat is fine.

  • It’s a mistake to assume the only burgers eaten in the us are processed meat. Many of us eat only grass fed burgers.

  • Re: point 6.

    Mortality is a rate measure, meaning it is adjusted for time. In the case of this study, it is adjusted for a 22 (Health Professionals Follow-up Study)- or 28 (Nurses’ Health Study)- year periods. Yes, we all have 100% chance of death; but the researchers are measuring death with respect to that specific period of time for high and low meat consumers. In that time period (22- or 28-year), the low meat intake group had 13 – 20% fewer deaths relative to the high meat intake group. This is how to interpret the study; nothing more, nothing less. The study makes no inference of causality as Zoe might suggest in point 1 of the summary.

    If the study looked at red meat consumption over a 128-year period, both groups would have 100% mortality.

    What part of “Red Meat Consumption and Mortality” is a headline grabbing tactic by egotistical scientists? It doesn’t even state the findings in the title- No inference can be made whether there is a negative, neutral, positive correlation between red meat consumption and mortality. The title accurately captures what was measured in the study.

  • Re: point 3. Cited directly from the article:

    In multivariate analysis, we simultaneously controlled for intakes of total energy, whole grains, fruits, and vegetables (all in quintiles) and for other potential nondietary confounding variables with updated information at each 2- or 4-year questionnaire cycle. These variables included age; body mass index; race; smoking status; alcohol intake; physical activity level; multivitamin use; aspirin use; family history of diabetes mellitus, myocardial infarction, or cancer; and baseline history of diabetes mellitus, hypertension, or hypercholesterolemia.

    This means that confounders were adjusted for (or “excluded” in layman’s terms) in statistical analysis, contrary to what Zoe reports.

  • Re: point 2. Some drugs for treatment of neurological disorders associated with aging have benefit in <13% of treated individuals. These drugs are coincidentally the most efficacious at treating these diseases. By Zoe's logic, these drugs should not be used for treatment as the benefit achieved is only modest in magnitude and therefore, does not warrant appreciation.

  • What?
    “The overall risk of dying was not even one person in a hundred over a 28 year study”. I think you are wrong…
    The overall risk of dying is about one person in a hundred EACH YEAR, no over a 28 year study.

  • Great analysis Zoe. I would ask though if you noticed that the researchers did attempt to control for the confounding variables (smoking, ETOH, physical inactivity, etc.) using a time-dependent Cox proportional-hazard regression. I have little faith that this fancy math can account for the intricacies of the human body but I’d be interested to hear what you have to say about this “controlling” for the confounding variables you raised.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.