In another article I wrote how ketosis can help you sleep less. A general idea that I mention on this website is that improved sleep quality can lower your sleep need.
Sleep quality is often measured by how much deep sleep (slow-wave sleep, or SWS) you can cram into a short amount of time. For several reasons, it seems like SWS and REM are the two main “need” stages. It’s not exactly clear how much we exactly need of each or how they work in conjunction with each other.
But once we look at very poor sleepers (say, those with neurological disorders) we ubiquitously find less SWS. That has helped lead to the belief that SWS is the “power” stage of sleep — the one that gives the most bang for your buck.
Anyway, back to ketosis.
A 2008 study from the University of Sydney looked at the effects of very low carbohydrate diets on sleep quality and confirmed many of the comments I make on this website.
The study looked at 2 groups:
Group 1 ate a 72% carbohydrate diet for 2 days.
Group 2 ate a <1% carbohydrate diet for 2 days (38% protein, 61% fat).
The subjects had their sleep quality measured during the study. Two measures I found interesting. First was sleep efficiency, which is found by taking total time asleep and dividing it by total time in bed (e.g. 90% sleep efficiency means you spend 10% of your time in bed tossing and turning). The second was SWS amount.
Results:
72% carbohydrate group
Sleep Efficiency: 89.1%
SWS amount: 66 minutes (13.9% of total)
<1% carbohydrate group
Sleep Efficiency: 92.4%
SWS amount: 88 minutes (17.7% of total)
There are a couple things worth taking away from this study:
- The subjects were all normal sleepers and perfectly healthy. So even in “normal” people, improved sleep quality is possible.
- The study was only 2 days (2 nights). So even on very short time scales, improved sleep quality is possible. But this begs the question of how much we can improve sleep over the course of weeks or months.
Ketosis has been used to treat epilepsy, and is believed to improve Alzheimer’s and schizophrenia.
Ketosis has also been linked to euphoria.
Ultimately, I think there’s something about the insulin roller coasters we put ourselves on by constantly consuming sugary or starchy foods — and it seems to be impairing how our brains function. The idea is that by fueling our brain cells with an “alternative” source of energy (ketones) we allow them to operate more efficiently, perhaps despite any damage done from too much glucose and insulin.
Now I don’t necessarily recommend a <1% carbohydrate diet. That’s not the idea. The idea is that there seems to be clear benefits to reducing sugar and refined starches from our diets.
My statistics are a little rusty, but even I can see that this study doesn’t show anything at all. Not even talking about showing anything “clearly”.
First this study was conducted on 14 men. An n=14 is a puny amount of data. In order to say anything, the observed effects would have to be very big.
Which they are not.
“The percentage of slow wave sleep (SWS) significantly increased for both the VLC acute (17.7 +/- 6.7) and ketosis (17.8 +/- 6.1) phases compared to control (13.9 +/- 6.3), P = 0.02 for both phases. ”
VLC acute and ketosis are basically the same, so the only question that remains is, if there is a difference between those two and the control.
17.8 +/- 6.1
13.9 +/- 6.3
If you plot those values out, you will see that, from the given sample, one can’t conclude that there is a difference, since both values are well within each other’s (huge!) confidence interval.
Even though the values for the sleep efficiency aren’t given in the summary, it’s the same: You just can’t conclude from a sample of 14 that there is a difference in two values, if the average of your datapoints show a difference of 3% (one could, if they were very sharply clustered, but I doubt they are). From a statistical perspective they are the probably also equal.
Hi wollff,
Most people in the medical field suffer from sample size obsession. They were taught in med school that large n values make a study more meaningful, in some magically linearly scaling fashion.
This has led to massive-scale studies like The Nurse’s Health Study or “The China Study”, etc — both of whose results I think are grossly overstated.
From a purely statistical standpoint, large n is good. From looking at true correlations of intractably complicated systems — that is, what studies are actually trying to accomplish — then the sample size is just one factor among many.
Read this, for example: http://www.blog.sethroberts.net/2008/08/22/citizen-science-whats-your-sushi/
If a study randomly picked 4 sushi restaurants in NYC and showed that all 4 contained cyanide poisoning, then would you be willing to risk going out for sushi in NYC? Or would you stick to what statistics 101 taught and cast it off as a statistical fluke?
In neuroscience, one of the most fascinating studies has been from an n=1 case: http://en.wikipedia.org/wiki/HM_(patient) — the only person in history to live without a hippocampus. We’ve learned a great deal from studying him, and no one is complaining about it being a “n=1″ study.
Additionally, the p values from the ketosis study above were <0.01. So they weren’t statistical flukes. The confidence intervals overlap, but that doesn’t affect the statistical significance.
Thank you very much for your answer.
Much of my confusion was indeed self-made.
To your example:
“would you be willing to risk going out for sushi in NYC?”
No, I wouldn’t eat it. After all, from what’s stuck from statistics in my head, with a probability of 95% about half or more of NYC’s sushi joints are poisoned.
It’s just that with an n this small you can’t say if half of NYC’s sushi is poisoned, all of it, or anything in between…
It’s for saying: “The data indicates that about 75% of NYC’s sushi is poisoned”, that statistics teachers whack you.
“The confidence intervals overlap…”
That was my mistake: For some reason in my mind I took the confidence interval for the variance.
“Additionally, the p values from the ketosis study above were <0.01."
Not really. One of their p values is <0.01, but only in the case when they suddenly reduced their n to 11 for some reason…
Anyway, their p values are between 0.006 (really doesn't sound like a value they set before the experiment though…) and 0.05, which still is totally legit.
Thank you for clearing up the main things, for anything else I would most love a look at the data. But that desire is not so strong that I would pay for it, so I'll let it be.
My understanding of ketosis (and experience with ketogenic diets) is that it doesn’t occur as soon as carbs are removed from the diet. Did they test these folks to see if they were in ketosis? Or were they actually burning thru their glycogen during the study period?
If the latter, your conclusion that it may be the insulin roller coaster more than the ketones is likely the bigger factor.
How do we know if the high carb diet just needed less sleep BECAUSE of their diet?
Beth,
I decided to upload the pdf of the article to rapidshare: http://rapidshare.com/files/369399255/ketosis-sleep.pdf.html
They took measurements to confirm that the 2nd group was in ketosis.