ICH is not my topic, but as we were preparing for the ESO Summerschool I explored the for me as yet untouched areas of stroke research. That brought me to this paper by Shoamanesh et al in JAMA Neurology which investigates a potential interaction between CMB and the treatment at hand in relation to outcome in patients with ICH. Their conclusion: no interaction. The paper is easy to read and has some at first glance convincing data, but then I realized some elements are just not right:
- the outcome is not rare, still a logistic model is used to assess relative risk
- interaction is assessed based on multiplicative interaction even while adding variables could lead to other estimates of interaction due to the non-collapsibility of the OR
- the underlying clinical question of interaction is arguable better answered with an analyses of additive interaction.
I decided to write a letter to the editor. Why? Well, additionally to the methodological issues mentioned above, the power of the analyses was quite low and the conclusion of “no effect” based on a p value >0.05 with low power is in itself a problem. Do I expect that there is a massive shift in how I would interpret the data when they would have analysed the data differently? I don’t think so, especially as the precision of any quantification of additive interaction will be quite low. But that is not the main issue here: the way the data were presented does not allow the reader to assess additive interaction. So my letter was focused on that: suggesting to present the data in a slight different way, and then we can discuss whether the conclusions as drawn by the authors still holds. Then, and only then we get the full picture of the value of CMB in treatment decision. The thing is that we will then realize that the full picture is actually not the full picture, as the data are quite limited and imprecise and more research is required before strong conclusions can be drawn.
But the letter was rejected by JAMA Neurology because of space limitations and priority. I didn’t appeal. The same happened when I submitted an edited version of the paper to Neuro-epidemiology. I didn’t appeal. In the meantime, I’ve contacted the corresponding author, but he did not get back to me. So now what? Pubmed commons died. Pubpeer is, to my taste, too much focused on catching image frauds, even though they do welcome other types of contributions. I know my comments are only interesting for the methodologically inclined, and in the greater scheme of things, their value is limited. I also do understand space limitation when it comes to print, but how about online?Anyway, a lot of reasons why things happened why they happened. But somebody told me that if it was important enough to write a letter, it is important enough to publish it somewhere. So here I am, posting my initial letter on my own website, which almost certainly means that no single reader of the original paper will find out about these comments.
Post publication peer review ideas anybody?
The original paper can be found here, on the website of JAMA Neurology.
My letter can be found here: CMB and intense blood pressure lowering in ICH_ is there an additive effect
bob siegerink 