Seven hours, and what that number is made of
75 min
Two hosts talk the lesson through. The voices are synthetic; the script was written from this lesson and checked against it, and asserts nothing the lesson does not.
- State the consensus recommendation and identify the verb it uses about health outcomes
- Compare an experimental finding with an observational one on the same subject and say what each licenses
- Explain what the long-sleep result is and why it matters to reading the short-sleep one
Lesson 2 was one experiment on forty-eight people. This lesson is a recommendation made by two professional societies and a study of one million three hundred and eighty-two thousand nine hundred and ninety-nine people.
The bigger evidence is the weaker evidence about cause, and understanding why is the most useful thing in this course.
The recommendation
In 2015 the American Academy of Sleep Medicine and the Sleep Research Society published a joint consensus statement.1 This course read part of it: the recommendation and four surrounding statements, quoted below. The accompanying methodology paper was not opened.
The recommendation, entire:
"Adults should sleep 7 or more hours per night on a regular basis to promote optimal health."1
Seven or more. Not eight, and the recommendation itself names no upper number.
And what it says about what short sleep does:
"Sleeping less than 7 hours per night on a regular basis is associated with adverse health outcomes, including weight gain and obesity, diabetes, hypertension, heart disease and stroke, depression, and increased risk of death. Sleeping less than 7 hours per night is also associated with impaired immune function, increased pain, impaired performance, increased errors, and greater risk of accidents."1
That's a long list of serious things and it's the sentence most retellings compress. Read the verb.
Before you read on. The panel wrote "is associated with" twice, in a document whose whole purpose is to tell people how much to sleep. Write down what you think that choice of words was doing, and what a panel that meant something stronger would have written instead.
Show the answer
A panel that meant something stronger would have written "causes", and they had the option.
These are the field's own peak bodies, writing the sentence the whole document exists to deliver, to an audience of clinicians and journalists. Every word in it was chosen. "Associated with" is what you write when the evidence behind a list is observational: nobody assigned anybody anything, and what was recorded was what people already did and what later happened to them.
What that leaves open is the direction. An association between short sleep and diabetes is equally consistent with short sleep contributing to diabetes, with early diabetes disturbing sleep, and with some third thing, shift work or poverty or pain, producing both.
And the panel isn't being coy. The statement is a recommendation and it makes one. What it declines to do is claim a mechanism it doesn't have, and noticing that is the difference between reading the document and reading a headline about it.
The study underneath
The largest single piece of evidence behind that list is a meta-analysis of prospective cohort studies published in 2010.2 A prospective cohort is a group recruited before anything has happened to them, measured once, and then followed: nobody is assigned to anything, which is exactly what makes it observational. This course read fragments of its abstract, from a proof copy and from the journal's abstract page, and did not open the paper.
The scale, in the authors' own words: "the 16 studies analyzed provided 27 independent cohort samples", and "They included 1,382,999 male and female participants".2
And the two results:
"short duration of sleep was associated with a greater risk of death (RR: 1.12; 95% CI 1.06 to 1.18; P < 0. 01) with no evidence of publication bias (P = 0.74) but heterogeneity between studies (P = 0.02)."2
"Long duration of sleep was also associated with a greater risk of death (1.30; [1.22 to 1.38]; P < 0.0001)".2
Their conclusion: "Both short and long duration of sleep are significant predictors of death in prospective population studies."2
Two words in that first result are worth unpacking, and they cut in opposite directions. Focus and Deep Work lesson 4 taught both. No evidence of publication bias means no sign that studies finding nothing went unpublished, which is a reason to take 1.12 seriously at all. Heterogeneity between studies means the pooled studies disagreed with each other more than chance would explain, which is a reason to hold 1.12 loosely. The authors report both, in the same sentence, and most retellings keep neither.
Two things are in that picture and most retellings keep one.
What the panel would say back
This lesson is about to spend several pages on what these numbers don't support, so here is the other side first, at its strongest. A member of that consensus panel would make three points, and none of them is silly.
Cohorts are designed against reverse causation. A well-run prospective study adjusts for illness measured at the start and often throws away deaths in the first years of follow-up, precisely so that "they were already ill" cannot explain the result. Whether the studies pooled here did that is not something this course can tell you, because it read fragments of an abstract and not the methods, and that cuts both ways rather than in this lesson's favour.
The two kinds of evidence point the same way. Lesson 2's randomised experiment found a dose-dependent cost to short sleep, measured under control. These cohorts find short sleepers dying sooner. Neither proves the other, and a panel weighing both together is doing something reasonable rather than something sloppy.
And a recommendation is a decision made under uncertainty. On that view, sleeping seven hours costs most people little, though it costs more to somebody on nights or working two jobs. Advising it on an association, while declining to claim the mechanism, is what public health does with the evidence it has rather than the evidence it wants. The panel's own wording is the proof that it knew the difference.
How big is 1.12?
A relative risk multiplies a baseline. It isn't a number of anything on its own, and without the baseline it can't be turned into a number of anything.
The following baseline is invented for the arithmetic and is not from any source read for this course.3 Suppose a group in which eight people per thousand die in a year.
A relative risk of 1.12 takes that to 8.96 per thousand. The excess is about one death per thousand people per year.
Now do the long arm yourself, before you read it. The relative risk is 1.30, on the same invented baseline of eight per thousand. What is the excess, and is it bigger or smaller than the short-sleep excess you just saw?
Show the answer
Eight times 1.30 is 10.4, so the excess is 2.4 per thousand per year. Two and a half times the short-sleep excess.
And that ratio is the one figure here that doesn't depend on the invented baseline, because both excesses scale with whatever the baseline is. Pick any starting rate you like and the long-sleep excess stays two and a half times the short-sleep one: 0.30 divided by 0.12 is 2.5 whatever you multiply them by.
Which is worth sitting with. The arm nobody worries about carries the bigger number, and the next section is what that does to the arm everybody worries about.
Logic and Argument taught this conversion and this course is using it rather than teaching it again. What it adds here is that both numbers come out small, and that the larger of the two belongs to the arm nobody worries about.
The long arm, and what it does to the short one
Here's the wrinkle, and it's the reason this lesson exists.
Nobody believes that spending longer in bed kills people. The standard explanation for the long-sleep arm is reverse causation: illness lengthens sleep, so people who are already unwell sleep longer and also die sooner, and the sleep is a symptom rather than a cause.
That explanation is reasonable. It's also, as far as this course can tell, the field's usual reading rather than a thing anybody has demonstrated in these cohorts, and nothing read for this course tests it.
Now do the thing this lesson is for. If reverse causation is the explanation for the long-sleep arm, what happens when you apply the same explanation to the short-sleep arm?
Show the answer
It works just as well, and that's the uncomfortable part.
Four ways that could happen, none of them measured in anything this course read.3 Pain could wake somebody who is already ill. A mood disorder could shorten sleep in one person and lengthen it in another. Poverty could shorten sleep and bring everything else that comes with poverty. Working two jobs could do the same. These are possibilities rather than findings, and the one thing that would settle any of them is in the last paragraph of this block.
So the same move that rescues the long arm from absurdity undermines the short arm's causal reading, and a lesson that used reverse causation only where it was convenient would be doing what lesson 7 is about.
What survives. The association is real, in an analysis pooling 27 cohort samples and 1,382,999 people, with an interval that keeps it clear of nothing-at-all. People the studies classified as short sleepers died sooner than the studies' reference group. That's a finding.
What does not survive. The sentence "sleeping less will shorten your life", which needs a direction the design can't see. And that is this course's reading of the pair rather than a statement either source makes,3 so hold it as reasoning you can check rather than as a result.
What would settle it is the experiment nobody can run: randomise a hundred thousand people to sleep six or eight hours for thirty years. The first exercise in lesson 2 was about why that study doesn't exist, and this is the bill for its absence.
What the recommendation does not say
Two sentences from the consensus statement are almost never quoted, and both matter more than the seven.
On long sleep: "Sleeping more than 9 hours per night on a regular basis may be appropriate for young adults, individuals recovering from sleep debt, and individuals with illnesses. For others, it is uncertain whether sleeping more than 9 hours per night is associated with health risk."1
Uncertain. Not risky, not fine. A peak body saying it doesn't know.
And on people: "Individual variability in sleep need is influenced by genetic, behavioral, medical, and environmental factors."1
That's a population recommendation admitting, in its own document, that the population contains people it doesn't describe. Lesson 8 is where that sentence comes back, because it is the whole of the course's closing refusal in fourteen words.
Two sentences from one outcome
Take one item from the list: increased risk of death, which is the one the meta-analysis measures.
The sentence the sources license: people the pooled studies classified as short sleepers died sooner than the studies' own reference group, in an analysis pooling twenty-seven cohort samples, by a relative risk of 1.12 with an interval from 1.06 to 1.18.2 Where those studies drew the line between short and long is not something this course knows, because nothing read here says.
The sentence they do not: sleeping less than seven hours a night will shorten your life.
The second one is shorter, more useful, easier to remember and unsupported. It's also the one you've heard, and the gap between those two sentences is a fair description of what this course is for.
Three things people get wrong about the seven hours
"Everybody needs eight hours." The recommendation is seven or more, from the societies that make it, and it names no upper bound for healthy adults.1
"Sleeping too long is dangerous." The long-sleep association is the larger of the two and, as far as this course can tell, nobody reads it as causal. If you're willing to explain that one away, you have to say why the same explanation isn't available for the other.3
"A million people is enough to prove it." Size fixes noise, not direction. Twenty-seven cohort samples with the same blind spot have the same blind spot twenty-seven times.
"It's only an association, so you can ignore it." An association in 1,382,999 people, with an interval well clear of nothing, is one of the larger pieces of evidence about human health that exists. What it cannot carry is a direction, and refusing the direction is not the same as refusing the finding.
Practice
Take twenty minutes.
Find a health claim in the news with a relative risk, a percentage increase or a "doubles your risk of" in it. Any subject; it doesn't have to be sleep.
Then three things.
- Write down the relative figure exactly as the article gives it.
- Find the baseline rate. Look in the article first, then in the study if it links one. Give up after fifteen minutes and record that you could not find it, which is a result.
- If you found it, do the arithmetic. If you didn't, write down what the claim is worth without it.
Most articles do not give the baseline, which is this course's expectation rather than something it has counted,3 and the reason the exercise is worth doing is that you'll remember the feeling of looking for it.
Allow fifteen minutes.
Take one article about sleep and health, printed or on a screen where you can annotate.
Mark every place where the article states a relationship between sleep and an outcome. For each one, write A if the sentence says associated, linked, tied to, or found in people who; and C if it says causes, leads to, damages, raises your risk of, or drives.
Then find the study it is reporting, if it names one, and mark the same for the study's own wording.
The comparison is the exercise. An article whose Cs all correspond to As in the source has done something specific and nameable to its evidence, and lesson 7 is where you watch that happen to a literature you now know.
Connections
Back. Lesson 2 is the other kind of evidence, and the pair is the point: a randomised fortnight with an outcome you can measure in an afternoon, against cohorts followed for years with death as the outcome. Logic and Argument is where relative and absolute risk were taught, and Memory lesson 7 is the sample question, which here comes out the other way round than usual: the sample is enormous and that is not the problem.
Forward. Lesson 4 is the quantity nobody in this lesson measured, which is when people slept rather than how long. Lesson 7 is a documented case of exactly the compression this lesson's second exercise looks for.
Go deeper
- Recommended Amount of Sleep for a Healthy Adult, Watson and colleagues, 2015. Open access and two pages long. This course read the recommendation and four surrounding statements, which is most of it, and reading the whole thing takes five minutes and is the best five minutes available in this subject.
- Sleep duration and all-cause mortality, Cappuccio and colleagues, 2010. This course read fragments of the abstract and did not open the paper. If you can reach it, the part worth looking for is how the studies defined short and long, because where those boundaries fall decides what the two numbers mean. That is this course's judgement rather than something it has read.
Sources
- Nathaniel F. Watson and colleagues, "Recommended Amount of Sleep for a Healthy Adult: A Joint Consensus Statement of the American Academy of Sleep Medicine and Sleep Research Society", Sleep 38(6), 2015, pp. 843 to 844; published simultaneously in the Journal of Clinical Sleep Medicine 11(6), 2015, pp. 591 to 592. Read in part: the recommendation and four surrounding statements, from the PubMed Central copy. The accompanying methodology and discussion paper was not opened, which the body says where the statement is introduced. Supports: the recommendation, the two lists of associated outcomes, the sentence about more than nine hours, and the sentence about individual variability.
- Francesco P. Cappuccio, Lanfranco D'Elia, Pasquale Strazzullo and Michelle A. Miller, "Sleep duration and all-cause mortality: a systematic review and meta-analysis of prospective studies", SLEEP 33(5), 2010, pp. 585 to 592. Read in part, in fragments: the Objectives, the scale, both results and the Conclusion, from an author-hosted proof copy and the journal's abstract page. The paper was not opened, which the body says. Supports: 16 studies, 27 independent cohort samples, 1,382,999 participants, and both relative risks with their intervals. The lesson prints nothing about how the studies defined short or long sleep, because nothing read here says.
- The course's own constructions, each labelled where it appears in the body. The baseline of eight deaths per thousand is invented for the arithmetic and the body says so at the point of use; every figure worked from it is arithmetic on an invented number. The move of turning the reverse-causation explanation from the long arm onto the short arm is this course's reasoning, said as such in the checkpoint, and nothing read for this course tests reverse causation in either arm, which the body also says. The expectation that most articles do not give a baseline rate is an expectation, and the exercise says so.
Check your understanding
This lesson has a 6-question quiz. Pass it and the questions come back on a schedule in Review, so what you learned stays learned. Your progress is saved in your browser; no account needed.