067_ Collider Bias Affects Genetic Prediction in Psychotic Disorders
Research Question and Aims
Polygenic risk scores show promise as biomarkers of psychiatric illness. However, clinical utility of psychiatric PRS has been limited by inconsistency in associations between the bipolar PRS and clinical outcomes. In data from the Suffolk County Project, we show that because both the schizophrenia and bipolar PRS increase odds of hospital admission, and therefore selection into the sample, the correlation between the schizophrenia and bipolar PRS is attenuated among cases, a phenomenon called collider bias (for details see the poster accompanying this proposal). This skews associations between the bipolar PRS and clinical outcomes. Our aim is to replicate these findings—specifically, the attenuation of correlations between the schizophrenia and bipolar PRS in sub-sets of cases of higher severity—in the PsyCourse Study.
We hypothesize that correlation between the schizophrenia and bipolar PRS will be highest in the full PsyCourse sample, and decrease as the sample is restricted to cases, cases who have received outpatient mental health treatment, and inpatient mental health treatment.
Data from all samples in PsyCourse who have genotype data available will be included in this study.
Polygenic risk scores
Polygenic scores for PsyCourse participants will be calculated in Munich, using summary stats from the most recent schizophrenia (Trubeskoy et al 2022) and bipolar (Mullins 2021) GWASs. Available imputed GSA chip data will be used. Analyses will include the first 10 principal components of population stratification.
Statistical analysis Primary statistical analysis will be the correlation between the schizophrenia and bipolar PRS, corrected for 10 PCs.
Genetic data (imputed).