Self-controlled designs to control confounding

The best available evidence for decision-making on interventions will often come from observational studies. Given the vulnerability of observational studies to confounding, complete and transparent reporting on confounding is necessary to enable readers to assess the validity of study findings. In his thesis Koen Pouwels assessed whether the reporting of confounding improved after the publication of the STROBE guideline. The reporting of confounding was similar before and after the publication of this reporting guideline and was not better in journals that more actively endorsed the guideline. Hence, research into the development and evaluation of strategies to improve the quality of reporting and adherence to reporting guidelines is needed rather than adding another guideline to the steadily increasing pile.

One of the more novel developments with regard to the control of confounding in observational studies is the application of a self-controlled design. Pouwels applied and empirically compared various available self-controlled designs. Although self-controlled methods seemed to be able to control confounding better in situations where strong unmeasured time-invariant confounding was present, such designs are often applied when it is likely that assumptions underlying these designs are violated. Discrepancies between parallel group designs and self-controlled designs could be predicted by failure to meet assumptions of the self-controlled designs. Therefore, Pouwels recommends to apply self-controlled designs when there are concerns about unmeasured confounders, but also to verify whether assumptions are likely met and in case of uncertainty to apply sensitivity analyses.

Dissertation: http://hdl.handle.net/11370/1b5ed2b5-50a9-4d30-a50f-b772d23fce48