Tuesday, May 22, 2007

The Devil in the Details - Low-Carb High Protein Diet Kills

A study - Low-carbohydrate–high-protein diet and long-term survival in a general population cohort - published in the May edition of the European Journal of Clinical Nutrition (originally available ahead of print, November 2006), concluded that "Prolonged consumption of diets low in carbohydrates and high in protein is associated with an increase in total mortality."

When I read the first publication of the study, back in November, I decided not to write about it since I was focused on penning articles about diabetes throughout National Diabetes Month and this study wasn't about diabetes. It remained on my pile of studies to consider, but as time passed, it seemed less and less important to write about. Now, with it hitting the print edition, and a renewed interest in the findings, it's time to take a look at what the researchers found and how they reached their conclusions.

Jake Young, who writes at Pure Pedantry, thought it worthwhile to take a look too. In summing up the design of the study, he wrote "[j]ust to cross the t's and dot the i's, they also control sex, age, years of schooling, smoking, BMI, physical activity, ethanol intake, and (in the data I am going to talk about) energy intake;" thinking, I believe, that adjusting for these confounding variables leads to higher quality conclusions.

Except in this case, adjusting for some of these variables led to a leap-of-faith extrapolation from the data because without making adjustments there was no significant finding to speak of, "[i]n model 2, the LC/HP score (absolute values) was positively associated with mortality, although the association did not reach statistical significance (P=0.14)." [emphasis mine]

In other words the association was statistically due to chance and therefore a null finding.

But, let's not let that get in the way of a good chance to find statistic significance!

Rather than take the data at face value for absolute intake (what people were eating), the researchers decided to adjust energy intake and considered this adjustment "as they should be, isocaloric..." to show how, after adjusting for energy intake variables things change.

And change they did.

With this one adjustment, statistical significance magically appears and now "mortality was significantly associated with reduction of energy-adjusted carbohydrate intake and nonsignificantly with increasing protein intake."

But still, how exactly to find protein problematic?

Ah, one more adjustment just might do it.

The researchers noted that their "model 3" "does not specify the complementary changes that have to be introduced for the preservation of total energy intake, when carbohydrates and proteins change." So, they created "model 4" and said it was "the most appropriate" since it was both isocaloric (adjusted for energy intake) and now adjusted to reflect changes from less carbohydrate and more protein.

Viola!

"In this model, increasing LC/HP score was significantly associated with mortality (P=0.001)."

Except, no one in the cohort ate like the adjusted diet, the statistically significant findings are not from real people eating real food (absolute intake), but are only found upon adjustment of energy and intake from protein and carbohydrate. The researchers justified this adjustment as necessary with "[i]ndeed, many of contemporary public health policies rely on extrapolations, so that if something is detrimental at a certain exposure level, its effect is likely to be more detrimental at a more extreme level."

Perhaps this is part of the problem with our public health policy? We're basing recommendations on extremes, extrapolated from adjusted models, rather than real world eating habits.

This is how the American Heart Association comes up with recommending less than 7% saturated fat to all Americans, even those at low risk for heart disease; they take data from studies that found no benefit from reducing saturated fat to 10% or less of calories and do the mental gymnastics to leap to an idea that benefit will be seen when saturated fat is less than 7% of calories. No data, just pure extrapolation and wishful thinking.

But, I digress...

There'd be nothing to talk about with this study if the researchers only went with their cohort and their habitual diet. Remember, when the absolute intake data was crunched, there was no statistically significant finding.

It's only after the researchers play with the data and make adjustments - adjustments outside the "norm" of their population cohort - into the extreme models, that they find significance.

In the real world you'll be hard pressed to find someone eating a diet within the extreme models they created. That's because the fundamental flaw in their model adjustment was the belief that decreasing carbohydrate means increasing protein. As the researchers noted in their paper, their adjustment model "relies on opposite changes of two nutrients with equivalent energy values and tends to be unrelated to total energy intake."

Simply put, the assumption was that reduction of carbohydrate translates to an isocaloric increase in protein.

The problem with this is that in the real world, increasing protein beyond what the body needs is extremely difficult due to satiety hormones that effectively shut down hunger when protein intake is adequate and carbohydrate is restricted. So the idea that one is effectively replacing carbohydrate with protein has a limit. A limit that seems beyond the grasp of this research group, who didn't seem to ask the important question - why did those with higher LC/HP (low-carb/high-protein) scores consume less calories than those eating a higher carbohydrate diet?

Rather than delve into this, they adjusted protein as if this group could and would consume more calories and more protein in a linear progression, adjusted their habitual intake to reach an isocaloric level comparable with their higher carboydrate counter-parts. But, because protein is such a sating macronutrient, this adjustment is not based on reality, but is purely hypothetical; and a hypothetical situation that one is likely to find impossible in a real world eating situation.

My take - the researchers would have been better off publishing their findings from the absolute intake data and letting the data stand as it is rather than make various adjustments to convoluted extremes and try to make headlines. Nice try though.

6 comments:

  1. Anonymous1:49 PM

    In this study
    The effects of diet differing in fat, carbohydrate, and fiber on carbohydrate and lipid metabolism in type II diabetes.
    http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=PubMed&list_uids=2547860&dopt=Abstract

    As I recall, the people in the high-protein, low-fat group had difficulty eating all the protein they were prescribed.

    Unfortunately, the full text isn't available online. I have it somewhere, but I can't find it.

    ReplyDelete
  2. Regina, I don't quite understand the models, maybe you can explain. How did they come up with these models and how do they know they actually apply to real people in real life?

    ReplyDelete
  3. Anonymous12:12 AM

    I think it makes sense. Our food pyramid suggests that we eat carbohydrates most and Low carb high protein diet is clearly low on carbohydrates.
    It's not surprising that unhealthy eating increases mortality.

    ReplyDelete
  4. Gotta love those critical thinking skills, Alex.

    Anyway, while I haven't read this study, I suspect the conclusions are bogus on purely mathematical grounds. You can't change the results of a hypothesis test by extrapolation or any other mathematical transformation. The information content of the data is what it is, and unless you collect more data (or otherwise change your information).

    Extreme example: suppose you were testing if some value is above zero. Your find the result to be not significant. To get the answer you want, you divide all of your data values by zero, making them infinite. Now (following the flawed approach) your result is infintely significant.

    Of course that's wrong, because any transformation you apply must be applied to the full posterior distribution to assess significance. You can't get something for nothing. This sort of nonsense is what you get when the truth of a hypothesis is assumed before you actually test it.

    Dave

    ReplyDelete
  5. Anonymous8:45 PM

    I agree that those kinds of diets don't have to be permanent. It's a risk for our health to continue with them. If we want to stay healthy, we should eat a balanced diet.

    ReplyDelete
  6. Define "balanced diet".

    ReplyDelete