Yeah, their design paper is very non-descriptive about their interim analysis. Just says that they'll do one every two years, and that they have "targeted median follow up of 5 years" with required >=1 year follow up per participant. No specific stopping rules given.
-
-
Replying to @ADAlthousePhD @Ricky_Turgeon and
And statistical analysis plan doesn't help much more with respect to interim analyses, alpha-spending, etc.
1 reply 0 retweets 3 likes -
Replying to @ADAlthousePhD @Ricky_Turgeon and
Before plunging into further questions about their reporting, I also have to note, that Table 1 looks...MAYBE a little suspicious. I think you know I'm not much of a fan of the "test for baseline balance" thing or taking this too seriously, but it looks almost suspiciously good.
3 replies 0 retweets 7 likes -
Replying to @ADAlthousePhD @Ricky_Turgeon and
I was about to ask you that. We shouldn’t look for imbalances obviously, but inversely, when looking at that many variables, and so few have low p values, can you say that that is unlikely and a bit suspicious? Anything’s possible obviously.
1 reply 0 retweets 1 like -
Replying to @load_dependent @Ricky_Turgeon and
Not even that so few have low p-values, but the numbers themselves look too good. And the first couple of variables that I've tried (admitting that there could be some rounding issues since it's reporting to 1 decimal place) the p-values do not seem to match what I would expect.
1 reply 2 retweets 6 likes -
Replying to @ADAlthousePhD @load_dependent and
Now it's possible that there is some missing data, or that there are rounding issues since I am inferring based on the summary stats to 1 decimal place rather than using the exact data.
1 reply 0 retweets 3 likes -
Replying to @ADAlthousePhD @load_dependent and
I'll need more time to unpack that Table 1. For now, let's proceed as though everything is on the up and up.
1 reply 0 retweets 3 likes -
Replying to @ADAlthousePhD @load_dependent and
Ricky's second question is a good one - but they do list the number of events in the paper "During the median follow-up period of 6.3 years (inter-quartile range 4.1–8.3 years), 3246 participants had a listed registered event; of them, 1752 experienced the main CVD outcome..."
2 replies 0 retweets 3 likes -
Replying to @ADAlthousePhD @load_dependent and
...though admittedly they do not separately list the numbers in the two treatment arms. Again, though, that's forgivable if a bit sloppy.
1 reply 0 retweets 1 like -
Replying to @ADAlthousePhD @load_dependent and
I, personally, don't mind all *that* much that they would choose to report only the adjusted HR. I've actually seen conflicting opinions on this. Some say to report unadjusted and adjusted; others say that you should only report the adjusted results (since those are primary)
2 replies 0 retweets 1 like
Unadjusted vs adjusted HRs I'm ambivalent about, but I think not reporting the crude figures by group is poor form. Not uncommon, but it irritates me no end
-
-
Replying to @GidMK @ADAlthousePhD and
Given how model/predictor selection affects estimates, is it unreasonable to look for large differences between adjusted & unadjusted estimates as possible red flags for gaming? Obviously a sensitivity analysis to assess robustness of results would be the ideal.
1 reply 0 retweets 3 likes -
Replying to @raj_mehta @GidMK and
On a side note, study says end-point assessment was blinded, but I can't find anything to suggest data-analysis was also blinded.
1 reply 0 retweets 3 likes - Show replies
New conversation -
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.