7/n I don't agree that preprints would necessarily reduce the risk of publication bias in this context - people may be less likely to preprint null results because no one is going to care very much
-
Show this thread
-
8/n For example, there's at least one study registered on clinicaltrials dot gov that I can't find results posted for (and does not appear to have been included in this analysis)pic.twitter.com/9AvT7HynPq
2 replies 3 retweets 36 likesShow this thread -
9/n Another issue is that I'm not sure I agree with the risk of bias assessment Now, this is inherently subjective, so take my opinions with a grain of salt, but some of this looks a bit oddpic.twitter.com/FdafgFov2x
1 reply 1 retweet 29 likesShow this thread -
10/n For example, the Elgazzar study. Low or uncertain risk for every element To me, this study is at a pretty high risk of bias!pic.twitter.com/9vfUArgcfE
1 reply 3 retweets 31 likesShow this thread -
11/n Randomization sequence generation is rated at low risk, but here's the information we have on randomization. It contradicts the protocol, which isn't great (they should've had 6 groups, not 2)pic.twitter.com/6C1BKGFvwd
2 replies 3 retweets 27 likesShow this thread -
12/n Allocation concealment is rated as low risk, but as far as I can tell from the study there's no information on how the allocation sequence was kept concealed except for those two sentences
1 reply 2 retweets 26 likesShow this thread -
13/n Technically there's no attrition bias because the authors report 100% follow-up data, but that in and of itself is weird. Usually you get at least a few dropouts!
1 reply 5 retweets 34 likesShow this thread -
14/n I could go on. Suffice to say that my assessment of this study would place it at moderate to high risk of bias, especially considering that it's really quite short and there's a fair bit of important information missing
2 replies 5 retweets 38 likesShow this thread -
15/n Worth noting that I don't disagree with the assessment for every study - I had a detailed read of Mahmud (2020) and it looks like a very well-done piece of research at a low risk of biaspic.twitter.com/Esr9s7DnC9
3 replies 2 retweets 34 likesShow this thread -
Replying to @GidMK
I agree. Being a suspicious SOB I looked at Mahmud closely given the result. For the 19 symptoms and comorbidities at baseline I calculated the p value from a chi square test and plotted the distribution. Spread evenly between 0 and 1. Reassuring.pic.twitter.com/QG93IzAtGV
3 replies 0 retweets 2 likes
The ones that really set off my potential fraud radar are the ones with 100% follow-up from baseline. You just don't see that in real life, there's always at least one of your 400 people who drops out
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.