> We aimed to assess the relationship of objectively measured sleep regularity with risk for all-cause mortality, and mortality from cardiometabolic causes and cancer, in a large cohort (N = 60 977) who wore accelerometer devices for 1 week.
They look at one week of sleep data and then check mortality records about 10 to 15 years later. It's hard to argue a causative effect between one week of bad sleep and death potentially 10+ years out.
Obviously there's an implication that people with terrible sleep regularity in that one week snapshot had terrible sleep regularity chronically, which in turn had a causative effect on mortality, but we have to make a couple of deductive jumps to get to that conclusion. I'd really like to see the same study with longer term sleep data.
That’s because this is in UK Biobank, a cohort of >500K Brits and collecting actually in such a large cohort is a miracle, let alone for multiple days. All thanks to the people who volunteered into the study. Would it be nice to have even more? Sure. But at that scale, patterns start to emerge.
As you point out that's a multi day commitment, and if part of the volunteers either adjust the timing of the experiments to match specific weeks (e.g. parents choosing school vacations), or adjust their schedule accordingly, what is measured becomes fundamentally different in nature to what measuring longer periods would bring.
I'm with you on how we don't have a choice regarding to the quality of the study, it's just crazy hard to get any data at scale. But we can look at it as a very flawed "best of what we can do" and not take the patterns too seriously.
There’s a health research programme currently underway in the UK that’s looking to recruit up to 5 million people. I believe they’re currently at around the 1 million mark.
In statistical mechanics there's a concept of "ensemble average" and its provable that if you have a system, the average state of the system over say a 100 realisations ("ensemble average") run for 1 second each, is equal to the the average of one system run for a 100 seconds - under some assumptions of course.
I don't know enough a about human biology to make a statement about whether any of those assumptions will hold true, but maybe someone else will.
I would also say it's geospecific. Like if there is a nationwide trauma of sorts (politics, etc) it could influence a group to not be representative of the population at large.
That's not the scenario here. An ensemble average requires those 1 second periods to be randomly sampled from the overall runtime. You can't just measure the first second after starting for all samples, or even the 7th second after starting, and say anything statistically sound about the average of the whole runtime.
The weeks that were measured in this study were not random. How far from random is the big question.
I mean the data lacks any person's sleep history, but if there is a cohort of bad sleepers, as a group you get a collective pseudo-history (because altough study is done at the same time they are in different life stages) that is no-one's history but might track the average well for that cohort AND take into consideration more than a week.
In the similar way sampling the height of 1000 of 1M people will give you a good estimate of the average.
> It's hard to argue a causative effect between one week of bad sleep and death potentially 10+ years out.
Statistically—absolutely, I agree with you, but controls and sample sizes can always be improved.
Narratively—it's also not difficult to see: "gunk builds up in brain; gunk requires regular removal; sleep removes gunk; stable sleep removes gunk better than unstable sleep"
It's difficult to blame people for emotionally attaching more to the latter than the former.
It seems easier to see: "some people have lots of difficulties in their lives that makes them have an irregular sleep schedule; some people have lots of difficulties in their lives that makes them die early"
I agree - an obvious connection would be "Lower income jobs tend to have less control over their schedule", be it shift workers, hourly service jobs or similar. There also may be links to worse healthcare due to lack of insurance in those jobs.
It might just be another "Poorer people don't live as long" correlation.
The source data is from the UK so it has nothing to do with the health insurance attached to specific jobs. Healthcare is covered via general taxation in the UK.
Interestingly enough, even in countries with tax funded healthcare people with lower income and socioeconomic status tend to receive worse healthcare (fewer expensive interventions, etc).
I'm aware of the UK system, having been born and lived there for over 25 years :) I didn't note that the data was from UK persons, but the general point likely remains, if not as strongly.
While true health insurance isn't attached to jobs in the same way, there's still uneven access - the town I grew up in closed it's local surgery a decade or so ago, so it's about a 30 minute drive, or over 2 hours on a rather indirect bus. Not everyone has a car, or can afford to take pretty much an entire day off work to get there. Assuming you can actually get an appointment, too. Richer areas often are better served, and richer people have better access to transport and time flexibility.
And my job came with BUPA membership, which can also make some things easier, it's not the hard barrier to any care in the same way as the American system.
And while the paper said that it corrected for Socioeconomic status, knowing the /scale/ (and possible error) in that correction relative to the claimed meaningful difference would be useful in studies like this. It feels like the sort of correction should be detailed more than just saying they did it.
But I guess the paper isn't really claiming causation, merely correlation in that it's a predictor of mortality. Though many commenters here seem to assume.
Interesting, I initially thought this factor was probably negligible, because the cohort was aged "62.8 ± 7.8 years" (either before or after the 7-year observation period). I still think it's small, but perhaps not negligible.
I estimate that some 2-4% of the group could be heavily involved in caring for grandchildren, sorting themselves into the irregular category. It's well-known that school-aged children pass flu to grandparents [2], and then grandparents die, just in time for the 7-year post-birth observation window. The absolute death rate due to flu is 10^-4 to 10^-3 per year, which would be visible on the paper's mortality time course chart.
Estimation details:
* some 20% are grandparents (and very few are parents) of young children [0]
* 20-50% of grandparents care for grandchildren regularly [1]
So maybe 4-10% of the cohort as an upper bound. If the birth rate is 12 per 1000 per year, and babies cause sleep disruption for about two years to two persons, then that's about 4% also, but perhaps mostly not the same age group.
Also: "people in poor health don't sleep well". That's really hard to fully control for, because a lot of problems will affect mortality at sub-clinical levels that don't satisfy diagnostic criteria and won't appear on your medical records.
I would presume that they somehow controlled for this most obvious confounding factor. almost every experiment I find that they are naysayers (as important as they are) that assume someone just simply plotted a chi-squared distribution.
The issue is that almost all human behavior is correlated and, even if you have an easy-to-see [sic] method of action, e.g. “brain gunk”, that doesn’t automatically negate a nearly-infinite set of other possible causes from correlated behavior. Just a random example: those with high stress probably sleep poorly. You can think of a number of possible explanations that link high stress to shorter lifespans: more likely to commit suicide, more fat retention, less time for healthy activities, etc.
No disagreement here, but equally you can't be surprised when people can grapple with the narrative easier than statistics. If you want the statistical understanding to propagate you need a narrative.
Eh, you gotta be careful with One Weird Trick medicine.
Once you get past painfully obvious problems, everything not currently easily fixed by modern medicine tends to be "well it's really complicated" sets of problems; there's not one Cancer, your blood pressure can be elevated for many reasons, syndromes like chronic fatigue are almost certainly a mix of dozens of problems binned together by common symptoms but will have different causes and treatments.
Anyone saying they have one treatment to fix dozens of problems is a huckster, and trying to come up with a medical Theory of Everything to explain large swaths of disease is your origin story for how you become a huckster.
"Obviously there's an implication that people with terrible sleep regularity in that one week snapshot had terrible sleep regularity chronically, which in turn had a causative effect on mortality"
You got it.
"but we have to make a couple of deductive jumps to get to that conclusion"
This is always the case unless you assume reality is as big as what you can perceive.
"I'd really like to see the same study with longer term sleep data."
You can say this of literally 100% of the studies, it will never be enough. I understand it when authors put this at the end of a study because they want more funding and because their subject is all they think about. But for reasonable human beings you gotta make a common sense jump and allocate resources to other subjects. Yes, regular sleep has good effects on health, the burden of proof of that was already 0, this study is a nice added touch, there will be no double dessert, move on.
Every deductive leap comes with uncertainty in establishing a causal chain. I don't I'm being overly reductive about the nature of evidence when I say that.
There is a specific question that needs to be addressed: With a one week window into sleep habits, are we selecting for people who are chronically poor sleepers, with those poor sleep habits leading to disease?
OR
Are we selecting for people who are chronically ill for other reasons and those chronic illnesses prevent them from getting regular sleep?
For example, people with sleep apnea have terrible sleep, and they have lower life expectancy than the general population. However, the cause of that lower life expectancy is not the poor quality of sleep; it's the cardiac effects of abnormal breathing over a long period of time.
If a person with decades of excellent sleep habits developed sleep apnea in the last 5 to 10 years of their life, the accelerometer will capture their irregular sleep and the death registry will capture their early deaths. That doesn't mean poor sleep habits killed them.
There are many other such chronic illnesses that can be confounded this way. Heart failure. Obstructive lung disease. Dementia. All can lead to irregular sleep. Add them together and you've captured a large segment of the population with ~10 years left to live.
The authors address this relationship in the conclusion and supplementary materials, but they appear to approach it entirely from the framework of poor sleep being causative of cardiovascular disease. Well yes, there's evidence that poor sleep can cause cardiovascular disease, but it's also well established (as I explained) that it can happen the other way around. If you want to cement a full chain of causality, you need a longer time window. Capture a young population with a low burden of chronic disease, show that poor sleep habits came first (i.e. within a certain age window), then cardiovascular disease, and then shorter lifespan. That would be the ideal data, even if difficult to acquire.
It is easy to say that good quality sleep is good for a person. But what if I literally never get that type of sleep. What should I do? How concerned should I be? How much focus and effort should I put into this? Should I take the Ambien I'm prescribed or should I try to go for the best natural path - because I really, really don't have good experience with Ambien. And it's almost like it's the best modern medicine can offer.
FWIW it’s mentioned in the study explicitly as a limitation:
> There are several limitations in this study. First, the single week of data collected for each individual provides only a snapshot of their sleep–wake patterns, and future work should collect sleep–wake data over a longer timeframe and include multiple weekend-weekday transitions. It is nevertheless interesting that even a snapshot of sleep behaviors is predictive of mortality for a follow-up period of several years.
I would expect that P(bad sleep this week | bad sleep chronically) > P(bad sleep this week | not bad sleep chronically). It's a question of inferential power: what is the minimum detectable effect size with such an indirect measurement?
Another hypothesis that seems plausible to me is that what we’re really measuring when looking at sleep is mental health, and mental health is a strong predictor of mortality 10-15 years into the future.
You are already correct, but the actual factor here is trait Neuroticism (negative emotion personality dimension), which is very roughly speaking a combination of genetic factors and your ACE score. ACE score is shockingly good at predicting mortality, having one over 6 makes people die on average 20 years earlier.
Yeah that’s kind of what I’m after: ACEs mess up people’s sleep and cause early deaths. That makes sleep patters a strong predictor for all cause mortality, but the causal relationship is really ”ACEs cause both poor sleep and early death”.
They look at one week of sleep data and then check mortality records about 10 to 15 years later. It's hard to argue a causative effect between one week of bad sleep and death potentially 10+ years out.
Obviously there's an implication that people with terrible sleep regularity in that one week snapshot had terrible sleep regularity chronically, which in turn had a causative effect on mortality, but we have to make a couple of deductive jumps to get to that conclusion. I'd really like to see the same study with longer term sleep data.