Currently I am doing a systematic review of the predictors of clinically significant weight gain amongst those who take antipsychotics. This is usually defined as a 7% or more increase in weight. My inclusion criteria will allow for both RCTs and non-randomised studies of interventions. My question relates to distribution of predictor variables. My understanding is that randomisation will allow for even distribution of potential confounders and baseline characteristics of participants. In an observational study, this will likely not be the case. My question relates to whether I should be prioritising RCTs over non-randomised studies of interventions because of this. As regression analysis will just be conducted amongst those who are treated with the drug, will baseline data, some of which might function as a predictor variable e.g. sex, be more evenly distributed if the sample arises through a randomisation procedure or not?
I may be over-complicating this, as if I include for sex in the model - perhaps if accounts for this even if males vs. females are not evenly present within the sample?
Any help much appreciated!