Sunday, January 31, 2010

Who pays the piper: Disturbed earth

I hope this is going to be the last post on Bradley and Hunter. Once again Table 5, but now the HbA1c line.




An HbA1c of 5.2, 5.3, 5.4 or 5.5 is normal. Any variation within these limits is virtually irrelevant clinically. Moving a tenth of a percentage point in this region is utterly insignificant. The changes in HbA1c don't even get a mention in the discussion. But perhaps they should do.

The low fat group dropped their HbA1c from 5.4 to 5.3, shrug. The low carb group dropped their HbA1c from 5.3 to 5.2. Mega shrug. Sound just about identical, don't they?

There are two very strange things about this. The first is that the drop for the low fat group does not reach statistical significance. The p value is 0.4 for within LF group change.

But the drop from 5.3 to 5.2 in the low carbohydrate group is statistically very significant, p=0.01. How can this be?

The second strange thing is that, in the whole of Table 5, there is only one value expressed to three decimal places. That is the standard deviation of the HbA1c of the post diet low carbohydrate group. This sd has to use three decimal places because it is so small. Having zero would not look good.

What does an sd of +/- 0.003 mean? Because three standard deviations around a mean include 99% of the population we know that 99% of the values lie within the 5.2 +/- 0.009 range. But HbA1c is only reported to one decimal place and you can only ever deal with a whole number of people. So, by the end of the eight week study, 11/12 subjects on the low carbohydrate diet will have had an HbA1c of 5.2 and the 12th person will either have been 5.3 or 5.1, there is no other way of looking at the data.

This is what happens when you take a group of people eating anywhere between 190g/d and 275g/d of carbs and drop them all to close on 104g/d. Everyone's post prandial glucose drops and the range of HbA1c both drops and tightens up.

Not so the low fat diet people.

Before the study they started with carbohydrate intakes of about 120-240g/d and tightened in to a minimal spread round 226g/d on the low fat diet. So some individuals increased their carbs, some decreased them, some didn't change them that much.

The changes in glycaemia will have been all over the place following changes in individual carb intake. Some will have risen, some dropped. There was a spread to begin with and a spread at the end. The overall effect is a small and insignificant drop because the carb intake variation is blunted by a 500kcal energy deficit being made up of zero-carb butt fat!

Had this diet not included weight loss the low fat people would have really come out badly on an HbA1c basis as they would have been eating glucose instead of burning butt fat!

In terms of glycaemia, everyone benefited in the LC group. Some benefited and some were injured, ever so slightly, in the low fat group. This is hidden in the average but shows through the standard deviations!

Overall this aspect is more of interest for looking at how information gets buried, rather than any genuine clinical effect. Information gets buried all right. Luckily you can see the disturbed earth in the results tables.

That's it. I've had enough of Hunter and will give it a rest.

Peter

9 comments:

Matt Stone said...

Eating more carbohydrates and less fat does not magically make your
A1C test results go up. All kinds of people are using a very high-carbohdydrate diet to lower A1C levels, just as there are people using a low-carb diet to lower A1C levels. No global study could ever link A1C levels to % of calories derived from carbohydrates because there is no such correlation.

Jim Purdy said...

Matt, an extended period of eating carbs sends my A1c very high, and a subsequent low-carb diet gets my A1c into the normal range.

That's not a "global study," but it's all the study I need to convince me to follow a low-carb diet to treat my Type 2 diabetes.

And eliminating grains and cereals are another huge improvement for me.


The 50 Best Health Blogs

Lauren said...

Thanks Peter, very helpful and interesting as always. And bigger thanks also for the info on magnesium/potassium last year, muscle cramps did finally go away as I upped the leafy greens. Funny how everyone goes on and on about bananas for potassium when greens have so much more per calorie.
Lauren

Markus said...

Suppose I make a huge, well designed study looking at A1C levels and carbohydrate intake and find no correlation. What I find is that one third of the subjects lowers their A1C levels, one third raises their levels and one third keeps their levels.

Will I then conclude A1C levels and carbohydrate intake do not correlate? I think so. Is that helpful? No.

To me the only reasonable next step seems to find out what differentiates the three groups. What is the difference between someone who raises A1C levels with carbohydrate and someone who lowers or keeps them.

Same with weight loss. If you get awesome results for your brand new diet but 10 % of the subjects gain incredibly much weight, you blame them and not your diet. Who asks why the 10 % gain weight?

One day we might have to admit that people are different. I had this strange feeling when I happened to see people looking different than I do.

We need to find physiological laws that apply to everyone. Correlations might only lead the way.

Peter said...

Matt (and Marcus),

Yes, I would agree completely with this. But this is not a global study. This is a VERY tightly controlled intervention study using 24 mildly broken livers in an attempt to normalise their insulin function. At a given degree of broken-ness (and these people are all mildly broken and are all acting as their own controls as well as being compared between groups) I would expect HbA1c to correlate very tightly with carbohydrate intake, as it does in intervention studies with type two diabetics placed on LC diets. When someone acts as their own control for sequential measurements of HbA1c this then correlates very tightly with carbohydrate intake, particularly when it is not confounded by having to reduce exogenous insulin or oral meds. The more broken the liver and, eventually, the more broken the pancreas (and hypothalamus?) the better the correlation should become...

So globally for associations, I agree. For tight interventions, no.....

Jim, in the kindest possible way, you've been damaged by the system more than luckier people!!!

Lauren, nice to hear. Interestingly I noticed you now get the sodium/potassium amounts for meat on Fitday. For those to whom it seems to matter, meat obviously has an excellent potassium to sodium ratio. It should have as it is made of muscle cells and muscle cells are full of potassium. We add the sodium to bacon to upset this ratio. Or I add salt to get that lovely cracking full of AGEs on roast pork. I doubt it matters for most of us but perhaps for a few it does....

Peter

Ned Kock said...

> But the drop from 5.3 to 5.2 in the low carbohydrate group is statistically very significant, p=0.01. How can this be?

That might have been a result of the dramatic drop in SD to .003 (everybody has almost the same A1C after, which is unlikely). But I think that is either a calculation mistake or a typo.

Judging by the variations in SD before and after, I would expect the SD to be .3 instead of .003, and the difference in mean A1C's to be statistically insignificant.

Peter said...

Hi Ned,

You are very kind to Hunter. The big snag with this kindness is that people do not work out the mean and sd with a pencil then compare the two populations after this calculation has been done. What is actually done is that all of the raw data are put in to a software program which then prints out a table of mean, sd, se and whatever other stats you specify with the p value ready calculated. So to follow your path there would have to have been the typo of 0.003 for 0.3 and a separate typo in the same line for the p value, giving p=0.01 rather than the more likely p=0.1. Where as these numbers will actually have been copy-pasted from stats software.

So either a double typo during manual transfer or a real situation of everyone having the same HbA1c and a near zero sd to make the p significant.

I am less kind than you. They copy pasted the data from Statsview and either didn't understand the significance or did understand and are bent. The later is more likely and the real bent-ness is in failing to report that no one increased their HbA1c on low carb where as at least some of the people in the low fat will have done. The numbers are correct and this group is bent.

Real researchers give you a line plot of all individuals, joining their pre and post values, so you can see exactly what is happening. There are tons of these plots in Nutrition and Metabolism papers. They're not difficult to do unless you have something to hide..

Peter

Carissa Doherty ND said...

My name is Carissa Doherty and I am a Naturopath in Burlington Ontario. I got a question I am having difficulty answering. Is coconut safe to use in pan frying? I had thought that when you hear the snap, crackle of oils in a frying pan, that is bad. I know coconut oil has a high smoke point and it is a saturated fat which can't turn into a trans fat because it has no double bonds. I can’t seem to find any information about when oils hydrogenated due to heat. I know that at smoke point oils completely denature. I know in my heart you can not fry in oil, but I would like to find exactly why. I believe the reason lies in the oils contained in the foods being cooked, and the other damage done by high heat such as oxidation and glycosilation of the sugars in the food, but I am still working our finding the proof. I don't like to go on a feeling. Am working on the research.
Thank-you for your time

Peter said...

Hi Carissa,

That's an interesting question. I don't think that anyone has a decent set of answers, so I'm a bit loathe to comment really. It really comes down to lipid breakdown/oxidation porducts and, while there are papers suggesting dietary ALEs and AGEs accumulate in humans from dietary sources none of them are particularly convincing. By pure chance I came across this report

http://www.sciencemag.org/cgi/content/abstract/326/5960/1677

http://www.sciencedaily.com/releases/2009/12/091222105051.htm

I don't think we started eating meat by boiling it. We started eating it at body temperature or cooler. But we were living grouped around a hearth 750,000 years ago. Holding a leg of meat over a fire is the logical way cooking started. That gives us a 750,000 year head start in dietary AGE and ALE tolerance over any other species.

Here in the UK the general naturopathic approach is to base the diet on carbohydrate and then add sufficient vitamins and minerals to convert rice to a food. There is a phobia about meat and lipid peroxidation is one explanation. I can't really see it.

BTW coconut oil is not 100% saturated, there is about 2g/100g PUFA and 6g/100g MUFA... Not a lot but some...

Sorry not to be any more use, I certainly fry in beef dripping. It, and butter, are my preferred fats.

All the best

Peter