Lots on missing
<img src=“https://img1.sendscraps.com/se/005/123.jpg", width = “200”, height = “400” style=“float: left;margin-right: 20px;margin-top: 0px;margin-bottom:0px;">
The last couple of weeks, there’s been lots of output from our work on Bayesian models for missing data coming out. Firstly, I participated in the workshop we did at ISPOR in Barcelona, as part of the Special Interest Group. Both that session and the other one on using R went very well — we got large participation (despite the fact that the latter was the second-last on the very last day!) — so much so that I couldn’t go to the next workshop after we finished the session on R because I stopped to talk to quite a few people!
Then Andrea’s main paper (from his PhD, which he’s due to defend shortly) has finally come out in Statistics in Medicine. I have to say I’m quite pleased with the final version of the paper — it’s a nice bit of modelling in which we cover lots of different issues, from non-normality, to correlation across costs and benefits to missingness in the main outcome as well as in the baseline characteristics. It also paves the way for a (probably even more) extensive paper, which we’ve submitted to JRSS/A, in which we tackle the issues of the underlying longitudinal data observed in HTA.
We had to fight one reviewer in particular, who had some reasonable comments, but also basically wanted us to re-write the paper the way they would have written in the first place (which, I think, we have very good reasons not to). In the end, I think the editor did quite a good job in keeping us to take into account and respond to many of the points raised, while allowing us to argue our way out of doing some of the proposed changes (when we had a rational argument) — for example, we didn’t want to do simulations, because we didn’t feel it would have made the point clearer. Instead, we offered to analyse a different dataset to highlight some additional issues (eg in terms of dealing with a wider range of distributional assumptions), which the editor has appreciated, thankfully.
“All” there’s left for us to do is to finalise our R package missingHE, in which we have implemented much of the modelling framework we describe in the Stats in Medicine paper.