Good evening,

In reply to the peer review of a paper up for publication, I've been asked to consider multivariate regression analysis. In analysing patient data we looked to show the safety of preoperative regional blocks. the data is non-parametric, observational, retrospective, and is missing some variables that weren't the direct variables being analysed such as smoking status, anesthetist etc.

It was my understanding that in this instance the commonly used was not appropriate in this scenario as the assumption of complete data was violated. This is my argument for not performing multivariate analysis. As a substitute I calculated the effect size of 1) receiving a block, 2) surgeon performing the procedure and, 3) anesthetist performing the anesthetic relative to the length of stay (LOS) and opiates on discharge (OOD). I then used these effect sizes to demonstrate that variations in LOS and OOD between the cohorts likely due to factors 2 & 3 (high effect sizes - Cramer's V) rather than the block itself (low effect size).

In summary, my argument is that further multivariate analysis is not appropriate in this instance as the study primarily looks to show that blocks are safe and not identify the causes of increased LOS and OOD. Second, our data violates the assumptions required to perform a Multivariate Kruskal-Wallis Test. Although there are complex alternatives to enable multivariate regression analysis in this instance it would be expensive, time-consuming, and offer little in achieving our primary objective - that of the efficacy of regional blocks.

Any thoughts would be greatly appreciated.

Kind regards,

Nick