Abstract:
Polygenic risk scores (PRS) are increasingly used to predict diseases (e.g., schizophrenia). However, the proper
application of the PRS in psychological studies is sparse. We aimed to evaluate the methodological aspects of
using PRS as a covariate in causal association studies in psychology. We conducted a simulation study using
realistic scenarios, such as gene and exposure or confounder interaction, gene and gene interaction, gene acts as
confounding and instrumental variable and replicated each scenario 1000 times. We found that when the genotype(s) that was used for calculating PRS interacted with the exposure and if the analyses considering this
interaction, the exposure effect similar with the true value. However, a significant amount of bias was present
even after adjustment for the measured confounders i) when the gene interacted with the unmeasured
confounder and this bias amplified when the interaction effect was two times stronger; and ii) when the gene acts
as a confounder or instrumental variable. Therefore, theoretical knowledge of psychology should be incorporated
when the PRS is utilized in psychological causal association studies in order to reduce systematic bias and
improve the precision of the exposure effect
