Ranstam Review DataBase (RR.db)

Templates for phenomenon "confounding"

  1. Propensity score matching is used as a tool to adjust for confounding bias. It should perhaps be clarified that in contrast to randomisation, which prevents bias from all confounders, adjustments can only be made for known and measured confounders. Furthermore, the cause-effect relations between the variables included in the analysis needs to be considered. The problem of inducing adjustment bias by adjusting for mediators and colliders (see Shrier I, Platt RW. Reducing bias through directed acyclic graphs. BMC Med Res Methodol 2008;8:70) needs to be considered in ANCOVA, and the effect of "M-structures" (see Pearl J. Remarks on the method of propensity score. Stat Med 2009;28:1415–1424) needs to be considered when using propensity scores. With both ANCOVA adjustments and propensity score matching a rationale for the adjusted factors, in terms of cause and effect, should thus be provided.

  2. The authors analysed the influence of potential confounders and did not find any "significant difference". Please specify if the word "significant" here refers to practical importance (clinical significance) or to inferential uncertainty (statistical significance). In the former case, what is the minimal clincally significant difference? and was this included in or excluded from the parameter estimate's confidence interval? In the latter case, why would this be relevant? How is the tested null hypothesis related to the estimated effect size?