Methods/Statistics
A graphical user interface tool to facilitate assessment of bias from differential pre-baseline selection Kelvin Pengyuan Zhang* Kelvin Pengyuan Zhang Sneha Mani Lindsay Kobayashi Alden Gross Jennifer Weuve Ryan Andrews
Pre-baseline selection encompasses the processes–e.g., death, illness, inconvenience–that determine who enrolls in a study. These processes can potentially bias a study’s estimated effect of an exposure on an outcome if they occur differentially and jointly with respect to the exposure and outcome (or their correlates). For example, when the exposure of interest affects the risk of mortality and there is an unmeasured common influence on the outcome and mortality, the pool of eligible participants for a given study may have susceptibility to the exposure’s effects on the outcome that differs from that of the referent population, thereby potentially leading to collider-induced selection bias. Because this selection bias arises from processes occurring prior to baseline, standard analytic tools like inverse probability weighting cannot be applied to address the potential bias. Tools that can help quantify the magnitude of the potential bias (e.g., simulation studies) have steep learning curves.
To address the need for easy-to-use tools to evaluate the potential for and magnitude of pre-baseline selection bias, we developed a graphical user interface via the Shiny package in R. Users can input key statistics, like cohort sample sizes, age distributions of their cohort, and the probability of mortality (unstratified or stratified by participants’ sex or country of residence), along with structural assumptions about the types and strengths of pre-baseline selection processes that could be present. Users are also able to choose pre-specified values for these inputs. These input parameters are then processed by our Shiny application to produce summary statistics that quantify the potential influence of selection bias on the estimated effect of interest. Because this tool removes many technical barriers to implementing pre-baseline selection bias analyses, we hope that it assists researchers in incorporating bias analyses into their own work.