The press release is here, and unsurprisingly I am not a fanhttps://www.sciencedaily.com/releases/2019/07/190710121607.htm …
-
Show this thread
-
The study is pretty simple - the researchers took a very large database of children (1.3 million) and identified every cancer diagnosis in that group 1.3 million children, about 2,000 cases of cancer (childhood cancers are rare)
1 reply 1 retweet 13 likesShow this thread -
They then compared the kids based on the characteristics of their mothers, in particular BMI, to see if that affected their risk of cancerpic.twitter.com/tiQQ5Wn9qz
1 reply 1 retweet 8 likesShow this thread -
They found that there was a very weak relationship between maternal BMI and the risk of childhood cancer (p=0.01), although this risk didn't become statistically significant until mothers were above 40Kg/m^2 BMI
1 reply 2 retweets 15 likesShow this thread -
They also found, in a piece of mind-blowing absurdity, that childhood leukemia was NOT statistically significantly related to maternal BMI (p-value highlighted)pic.twitter.com/r1r8sFEd5a
3 replies 1 retweet 15 likesShow this thread -
Replying to @GidMK
Minus the absolute/relative risks aside, why do you think statistical significance/lack of it matters in a study with no randomization? Perfect RCT + high power for the stat test, I might conclude that sig diff is because of exposure, but what about here 1/2
1 reply 0 retweets 1 like -
where other model assumptions are violated? like the assumed distributional form/random mechanism. What exactly would low P or high P mean here? 2/2
1 reply 0 retweets 0 likes -
Replying to @dailyzad
I reckon if you're going to use statistical significance based on p-values as your cutoff, you should report when your values are above them even if the whole model is a bit silly in the situation
1 reply 0 retweets 0 likes -
The really honest results would've been that there was an increased risk observed, but since there's no solid plausible mechanism and the CIs all crossed 1 except for very high BMIs, it's hard to make a solid conclusion based on this research
1 reply 0 retweets 0 likes -
Replying to @GidMK
If increase in risk is small and could easily be explained away by confounding, I think that may be reason to be skeptical, but the CIs crossing 1 just indicates nonsignificance, not lack of effect. Peto, 1988 all CIs cross 1, but pooled, there's an effect, studies were imprecisepic.twitter.com/sXoPiw7Gb4
1 reply 0 retweets 1 like
Right, but at the very least they should've addressed the fact that their results were not inconsistent with obesity being preventive of cancer up until BMI got above 40
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.