Hacker News Reader

Comment by johnny313 | Hacker News Reader

The headline result (sedentary behavior leads to increased risk of mortality) is not completely supported by the results shown in the abstract.

They break the test subjects into four groups ranked by activity level, and compare the the highest activity group to the rest. But in the results, only the lowest activity group has a hazard ratio where the 95% confidence interval does not contain 1. Translation: the data only shows a meaningful difference in mortality risk between the highest and lowest activity groups. For the other two groups, the data does not support show an increased risk.

View on HN

Also they didn't randomize the activity level of the subjects by putting them on different exercise programs.

I would think that people being in an end of life state at the beginning of the study would tend to be more likely to be tired and avoid exercise.

It could well be that early undiagnosed terminal illness causes people to be sedentary and not the other way around.

Randomized trials are the gold standard, but I wouldn't blast a study straight for being prospective.

My main thingy with studies like these is that actuarial bookkeeping is hard. The difference in base mortality rate between a few age years in difference can be huge. But since the study runs 4, that difference in base rate multiplies. Properly accounting for that is hard.

Agreed. We're back in cause vs. correlation territory, no?

The most sedentary people may also be the most obese, the most likely to suffer from diabetes, the most burdened with respiratory and cardiovascular diseases, etc. Their expected lifespans will thus be quite shorter.

I'm willing to believe that appropriate exercise is good for you, almost no matter how robust or frail your health is. But this would be a much better study if it tracked cohorts with roughly parallel starting health. That way, the degree of active-vs.-sedentary living could be a study variable that wasn't tainted by undocumented variations in starting health status.

i imagine its a negative feedback loop. unhealthy people move less and become even more unhealthy.

that being said, obesity correlates positively with inactivity. being less sedentary is a huge goal of mine to improve my health, pvalues or not

This stood out to me also. I understand why they didn't do so - randomizing exercise for 8,000 people is all but impossible. (Unreliable for "do more", potentially unethical for "do less"). But it's pretty worrying because there are a stack of possible shared causes between mortality and activity.

Everything from obesity (cause of inactivity and shared causes with inactivity) to rarer mortality risks like cancer would simultaneously act to create this result even with no causative linkage.

It seems like a new version of that infamous NEJM study showing that low and healthy weights have especially high mortality - because terminally ill patients tend to lose a lot of weight.

I agree.

However the hazard ratio is monotonously increasing across activity groups in the direction one would expect if there is an effect. So the data does indeed show a (meaningful) rise in risk going lower on the activity ladder, only it is not statistically significant.

A continuous model might get more (statistical) traction here. They've got the data. Why use segments?

Reply on news.ycombinator.com