Comparison of the tic disorder phenotype risk score and polygenic score in a large clinical biobank

The phenotype risk score (PheRS) is a tool for evaluating comorbidity patterns within a disease of interest based on a patient's prior clinical diagnoses. Previously, we generated a tic disorder PheRS based on 69 medical phenotypes enriched in tic disorder patients by training within a dataset of 3.6 million individuals and deploying the PheRS in an independent set of 90,051 genotyped individuals from the Vanderbilt University Medical Center biobank, BioVU. The phenotypic features used in the PheRS calculation included known comorbidities (OCD, ADHD, ASD), as well as several additional neuropsychiatric and neurological phenotypes. The tic disorder PheRS was significantly higher in clinically validated tic disorder patients, strengthening the rationale for using this method to identify individuals with tic disorders from large electronic health record databases. Here we evaluate the tic disorder PheRS compared to the tic disorder polygenic score (PGS) in the same patient population, BioVU. The tic disorder PGS was calculated using PRS-CS with summary statistics from the most recently published tic disorder genome-wide association study (Yu et al. Am J Psychiatry, 2019) and represents the genetic liability of tic disorders across the BioVU population. The tic disorder PheRS and PGS were compared within a subset of clinically validated tic disorder cases (n = 319) and controls (n = 1,585) using linear regression models, with covariates for current age, median age of medical record, genetic principal components 1-10, sex, diagnosis code density, and genotype batch. Both the tic disorder PheRS and PGS are significantly higher in the subset of clinically validated tic disorder cases versus controls (PheRS p= < 2e-16, beta+SE=0.13+0.004, PGS p= 0.001, beta+SE=0.03+0.008); however, the tic disorder PheRS and PGS were only modestly associated with one another (p=0.03, beta+SE=0.09+0.04). When tested independently, the prediction performance for the tic disorder PheRS in the clinically validated cases/controls (Nagelkerke's Pseudo r2=0.32) is markedly higher than the tic disorder PGS performance in this same population (r2=0.04). When we combine the tic disorder PheRS and PGS using a multivariable linear regression model, we find that the combined model modestly improves the prediction value (r2=0.33), suggesting that clinical history in conjunction with genetic information may provide the strongest method at identifying individuals at risk for a tic disorder diagnosis. Future studies will evaluate additional modeling strategies to combine these measures in search of the best performing metric. The PheRS approach (both independently and in combination with the tic disorder PGS) may provide a way to identify additional individuals with tic disorders that were previously undiagnosed – increasing sample sizes for genetic analyses and expanding our current understanding of tic disorder epidemiology and biology.