Lies, damned lies and statistics, part n+1. Riera-Crichton et al.

In Macronutrients and obesity: Revisiting the calories in, calories out framework, the conclusion is:-
"Our structural VAR results suggest that, on the margin, a 1% increase in carbohydrates intake yields a 1.01 point increase in obesity prevalence over 5 years while an equal percent increase in fat intake decreases obesity prevalence by 0.24 points."

So, carbohydrates are fattening but fat is slimming, eh? I declare shenanigans! Two can play at that game.

In Effect of Dietary Protein Content on Weight Gain, Energy Expenditure, and Body Composition During Overeating, Bray et al increased kcals by 40% by adding Fat grams. Carb grams didn't change. Protein grams changed a bit. ∴ Protein %E & Carb %E decreased by ~29%. %E means "as a percentage of total Energy".

Weight (lean body mass + body fat) increased as Fat kcals increased ± some interpersonal variation.

From Fig. 6.

_
 _Decreased P %E & C %E result in increased weight.
∴ Increased P %E & C %E result in decreased weight.

∴ Fat is fattening, but Protein & Carbohydrate is slimming! Q.E.D.

Do you see what's going on? Here's a summary:-

Diet contains A, B and C.
The amount of A increases, but the amounts of B and C remain constant.
A%E increases, but B%E and C%E decrease.  

In Riera-Crichton et al, A = Carbohydrate, B = Fat and C = Protein.
In Bray et al, A = Fat, B = Carbohydrate and C = Protein.

Some fascinating healthcare statistics: Article by Ezra Klein.

Spotted on Twitter. From Is America better at treating cancer than Europe? , by Ezra Klein.

Healthcare rankings.

I highlighted Great Britain in green.

Cancer mortality.

Siri-Tarino et al, Forests & Trees and "Eureka!" moments.

Here's Fig. 2 from Meta-analysis of prospective cohort studies evaluating the association of saturated fat with cardiovascular disease:-

Risk ratios and 95% CIs for fully adjusted random-effects models examining associations between saturated fat intake in relation to coronary heart disease and stroke.

The above "Forest" plot has a subtotal RR of 1.07 (95% CI 0.96 1.19). The overall conclusion is that there's no association between saturated fat intake and the RR for CHD. Hmmm.

I looked at the data in Table 3. Of the 16 studies contributing to the CHD results, only 3 of them specify high sat fat intakes over a wide range. The results from these 3 studies are as follows:-

Pietinen et al: RR=0.93 (95% CI 0.6, 1.44).
Mann et al: RR=2.77 (95% CI 1.25, 6.13).
Boniface et al: Pooled RR = 1.37 (95% CI 1.17, 1.65).

The results from Pietinen et al are statistically-insignificant (95% CI values are way above & below 1) with an overall slight protective effect. The results from Mann et al have a RR >> 1 with both 95% CI's >1 and the results from Boniface et al have a RR >1 with both 95% CI's >1.

Other studies either have sat fat intakes varying from very low to low, or specify mean/median sat fat intakes without values for highest & lowest tertiles/quartiles/quintiles etc. Other studies have results that are statistically-insignificant.

However, there are some studies that show a slight protective effect of small amounts of sat fats. How come?

Thanks to George Henderson, I had a "Eureka!" moment. He posted a link to Dietary intake of saturated fat by food source and incident cardiovascular disease: the Multi-Ethnic Study of Atherosclerosis.

Here's Fig. 1 from that study.

HRs and 95% CI's of CVD risk according to quintiles of energy-adjusted SF from different sources (n = 5209).

The Meat SF plot has a net positive slope (bad news, but the range of intake is so small that the results are probably due to chance), the Butter & Plant SF plots are random, but the Dairy SF plot has a net negative slope (good news). Dairy saturated fats in amounts of up to 10g/day are protective against CHD. As the Dairy sat fat intake is too small to have a significant effect on lipids, what's the mechanism? I think that it's Vitamin K2. See Chowdhury et al, More forests & more trees and more "Eureka!" moments with cheese.

When you average out the results from all studies, the result is null. This is data dilution statistics.

EDIT: See also Study: Saturated Fat as Bad as Sugar!

Elvis lives!

This bloke said he saw Elvis down the chip shop. What? You mean he was mistaken or lying? Surely not!

Regularly, you read or hear:- "Aspartame gives you cancer!" "Meat gives you cancer!" "Crisps give you cancer!" "Coffee gives you cancer!" "Coffee doesn't give you cancer!" "Coffee gives you cancer!" and so on and so forth...... What are we to believe?

In terms of Diet & Nutritional information, there's a hierarchy of credibility. Here's a rough list of credibility, starting with the least credible and ending with the most credible.

Inexpert Opinion. Basically, anyone can say anything (especially on the internet) but that doesn't make it true. This includes anything that I say, which is why I try to back up what I say with evidence from higher up the pile.

Anecdote/Testimonial. Just because it worked for Joe or Josephine doesn't mean that it'll work for anyone else. There's also the placebo effect e.g. Nothing acts faster than Anadin (so I use nothing!)

Expert opinion. Even "experts" with lots of letters after their names get things wrong and have personal biases, hence the controversies over MMR, HIV/AIDS, Cholesterol etc.

In Vitro (in glass) studies. What happens in Petri dishes/test-tubes etc doesn't necessarily happen in humans.

In Vivo (in life) Animal studies. What happens in rats/monkeys etc doesn't necessarily happen in humans.

Epidemiological (population) Human studies. A being associated with B doesn't necessarily mean that A caused B, as the association may have been due to random chance or due to both being caused by C, D, E........Z etc.

Case studies. The number of subjects is usually quite small and often just 1.

Small and/or non-randomised and/or unblinded and/or non-placebo-controlled and/or non-crossover trials.

Large, Randomised, Double-blinded, Placebo-controlled Crossover trials with a decent Washout.

Meta-studies of large, randomised, double-blinded, placebo-controlled crossover trials.

Even the last two in the list aren't perfect. Study outcomes can be manipulated by tweaking the methodology (e.g. using pre-trial screening) or statistical jiggery-pokery. If a trial is funded by (a) drugs company(ies), the methods used & the results obtained should be scrutinised very carefully.

EDIT: Systematic Reviews & Meta-studies can be manipulated by including studies with completely different input & output ranges, so as to dilute the data and force a null result. See Meta-analysis of prospective cohort studies evaluating the association of saturated fat with cardiovascular disease and Low Carbohydrate versus Isoenergetic Balanced Diets for Reducing Weight and Cardiovascular Risk: A Systematic Review and Meta-Analysis.

Abstracts can misrepresent the data. See Intensive lipid lowering with atorvastatin in patients with coronary heart disease and chronic kidney disease: the TNT (Treating to New Targets) study.
"Conclusion: Aggressive lipid lowering with atorvastatin 80 mg was both safe and effective in reducing the excess of cardiovascular events in a high-risk population with CKD and CHD."

There was a 15-32% reduction in major cardiovascular events. Wow, that's impressive! What the abstract failed to mention is the fact that there were 26 more deaths in the 80mg group, than in the 10mg group. What's worse? Having a major cardiovascular event or being dead?

In conclusion, when you read something in the media, do some research of your own before accepting it as fact. Use Google, Wikipedia, PubMed and the Cochrane Library.