Hacker News Reader

Comment by webprofusion | Hacker News Reader

What is it they say about correlation?

I'm not a scientist but in a group of 60 year olds doom-scrolling in bed vs those who just read a book (or even just go to sleep) is it outside the realm of possibly that the former group may be more likely have other aspects to their life that are unhealthy general?

View on HN

That's a very good point. There are lots of potential confounders here. But I also want to point out that this study isn't a shot in the dark for random correlations -- there are theoretical reasons to believe that poor sleep hygiene would increase the risk of diabetes. Cortisol is tightly coupled to the sleep-wake cycle and to blood sugar levels. It is known to be causative of diabetes in certain rare disease (Cushing disease, for example) and strongly associated in garden variety type 2 diabetes. The article passingly references this (but it uses the more general term glucocorticoid).

aHR and RR are broadly comparable.

Key findings from the diabetes study (this one):

    Night light exposure in the 90-100th percentile (brightest nights): aHR = 1.53 [1.32-1.77]
    Night light exposure in the 70-90th percentile: aHR = 1.39 [1.24-1.57]
    Night light exposure in the 50-70th percentile: aHR = 1.29 [1.14-1.46]

Key findings from N=1 million birth control use impact on depression study (https://scholar.google.com/scholar_lookup?journal=JAMA+Psych...) :

    Combined oral contraceptives: RR = 1.23 [1.22-1.25] for first use of antidepressants
    Progestin-only pills: RR = 1.34 [1.27-1.40] for first use of antidepressants
    Patch (norgestrolmin): RR = 2.0 [1.76-2.18] for first use of antidepressants
    Vaginal ring (etonogestrel): RR = 1.6 [1.55-1.69] for first use of antidepressants

In my opinion, if you are okay with birth control, you are probably okay with using light at night.

Light exposure was measured by a "wristwatch" worn on the dominant hand, so reading a book is going to get coded as more light than smartphone use probably.

It also seems the sample size was so large that it could easily convert these differences into a significant p-value. Is there a statistical compensation to make confounding factor adjustment more stringent with larger sample size?

We know that sleep is important. We know that light can mess up your sleep. It stands to reason that if it can mess up your important sleep, it can have negative consequences down the line.

Reply on news.ycombinator.com