Wikipedia:Reference desk/Archives/Mathematics/2015 July 6

Welcome to the Wikipedia Mathematics Reference Desk Archives
The page you are currently viewing is a transcluded archive page. While you can leave answers for any questions shown below, please ask new questions on one of the current reference desk pages.

July 6

Margin of error in diabetes study ?

https://www.scientificamerican.com/article/bariatric-surgery-edges-out-lifestyle-changes-for-type-2-diabetes/

There are 61 patients, each randomly assigned one of 3 treatments. I'm thinking that's not nearly enough to draw any conclusions (and I'm irked at Scientific American for not making that clear). What's the margin of error on such a small study ? StuRat (talk) 21:51, 6 July 2015 (UTC)

The article did state at the end that more study was needed before drawing definite conclusions, so you could say that the what is reported comes under the heading of "preliminary results". If you compare the types of treatments though, the number 14 partial remissions for surgery vs. 0 for lifestyle does sound like it's statistically significant. But no conclusions should be drawn anyway until the experiment is repeated. It's a problem with the press in general that they report this kind of thing to the public who generally don't have the statistical knowledge or critical thinking skills to properly evaluate it, then a few years down the road when it's found to be an anomaly, the press has moved on. But imo the real problem with studies like this is that they only report on a single result. For example maybe your diabetes goes into remission but you die two years later from dietary issues. These surgeries are kind of a big deal, so it seems unrealistic to expect no unintended consequences. --RDBury (talk) 03:27, 7 July 2015 (UTC)

"After three years, 40% in the RYGB group, 29% in the LAGB group and no one in the lifestyle intervention group had at least a partial remission of diabetes". If the untold sizes of the three groups are 20:21:20 then the numbers 8:6:0 reproduce the data. (20+21+20=61, 8/20=40%, 6/21=29%, 0/20=0%). The probabilities of having "at least a partial remission of diabetes" have beta distributions with mean values ± standard deviations = (41±10)% for RYGB, (29±9)% for LAGB, and (5±4)% for lifestyle intervention. In this investigation RYGB is not significantly better than LAGB, but both are significantly better than lifestyle intervention. Bo Jacoby (talk) 06:07, 7 July 2015 (UTC).

Thanks. Other larger studies have shown a higher rate of remission with lifestyle changes, so I suspect there must also be something wrong with the methodology in this study. StuRat (talk) 14:08, 7 July 2015 (UTC)

Interesting. Do you know the numerical results of the studies in question? Bo Jacoby (talk) 20:18, 7 July 2015 (UTC).

I believe there were multiple studies. Here's one: [1]. StuRat (talk) 15:33, 8 July 2015 (UTC)