17 Study Meta-analysis suggests low-carb is healthier than low-fat diets


#1

http://news.meta.com/2015/11/05/lowcarb/

The study was funded by Atkins Nutritionals,
a corporation founded by Dr. Robert Atkins for the promotion of
low-carbohydrate diets. Atkins Nutritionals had “no role in study
design, data collection and analysis, decision to publish, or
preparation of the manuscript”.
The primary limitation of the study was
the authors’ lack of access to patient level data. Secondary to this,
low completion rates were problematic because half of the studies
had completion rates of less than 73%.


#2

Slightly less sensational coverage:

CardioBrief: Low-Carb Diet a Little Better in Meta-Analysis

This study only found a very “modest” difference in weight loss for this particular set of studies, and the analysis of cardiovascular risk was highly, highly speculative.


#3

Because the study was so short term?


#4

Ok… So what about those of us who are not overweight?

“The low-fat group had a larger beneficial effect on LDL cholesterol, while the low-carb group had a better effect on HDL cholesterol, triglycerides, and systolic blood pressure:”

So what’s more important for the general population here?

As I’ve increases the unsaturated fat in my diet I’ve seen all of the above (little change in LDL, increased HDL, and decreased tryglycerides). I think exercise also boosts HDL though. My HDL is higher than my LDL (103 vs 85) and my triglycerides have gone down from 54 to 41 in the last two years.


#5

Aren’t those already pretty good numbers?


#6

Aside from the over all high total cholesterol (just under 200), yes I believe they are good numbers. Not sure how much of that I can attribute to my 40% fat diet.


#7

@MentalNomad is probably referring to:

One expert, who did not wish to be quoted, agreed that the small difference in weight loss was quite plausible. But, he warned, the analysis of cardiovascular risk “is a whole other story,” since the use of summary trial data to predict cardiovascular risk “is highly suspect and may not be valid” for the prediction of 10-year effects.

It seems an extra superlative was slipped in, when presented to you here. To answer your question, yes, some unidentified expert thought it unwise to extrapolate in time. There is always an argument for more research funding amongst the research community. Of course, far more damage has be done by fraudulent diet research, than be extrapolating data sets. Thank you for bringing this to my attention.


#8

Interesting ketogenic diet study.


#9

Actually, no - I wasn’t quoting anyone when I said,

That was me, talking.

No, because of the statistical methods. Summary trial data has been used to to feed statistical classification and regression analysis to find relationships between current conditions and future disease risk. Several models have been built which attempt to identify future risk. This is a speculative and new area, and one which is very useful for finding associations… but not necessarily casual relationships. More importantly, it’s generally a mistake (though a tempting one to make) to use a risk score as a diagnostic.

Your goal is to reduce your risk of cardiovascular disease, not to improve your placement on the scoring and risk assessment scale. The scoring and risk assessments are useful tools in research into underlying causal relationships, but most of the elements being scored are, themselves, coinciding symptoms with cardiovascular disease, not causes of it, so it’s a mistake to address them as if they were the goal.

It’s like taking a backwards inference, and then using to it make a forward assumption - each has a large margin of error because they have enormous error bounds, and they compound each other because they multiply.

On top of this, the models only have validity on the populations on which they were built. A particular level of cholesterol may be a risk factor for CVD among people eating a common American diet for the past 30 years. Does that mean that particular cholesterol level threshold is the correct predictor of CVD in someone eating a dramatically different diet? Not likely. So you’ve got error estimating back and then error estimating forward, AND it’s being applied to a different population in each step.

The simplest real-world analogy that leaps to mind is this:

“Aha! Where’s there’s smoke, there’s usually fire!”

This is very useful for research - example:
look for fire where you find smoke <- good move

But it’s not useful as a target to address - example:
reduce smoke via better ventilation <- bad move if you’re actually trying to stop fires