Selecting cases of major psychiatric and substance use disorders in Swedish national registries on the basis of clinical features to maximize the strength or specificity of the genetic risk

Abstract

We investigate how selection of psychiatric cases by phenotypic criteria can alter the strength and specificity of their genetic risk by examining samples from national Swedish registries for five disorders: major depression (MD, N = 158,557), drug use disorder (DUD, N = 69,841), bipolar disorder (BD, N = 13,530)) ADHD (N = 54,996) and schizophrenia (N = 11,227)). We maximized the family genetic risk score (FGRS) for each disorder and then the specificity of the FGRS in six disorder pairs by univariable and multivariable regression. We use split-half methods to divide our cases for each disorder into deciles for prediction of genetic risk magnitude and quintiles for prediction of specificity by FGRS differences between two disorders. We utilized seven predictor groups: demography/sex, # registrations, site of diagnosis, severity, comorbidity, treatment, and educational/social variables. The ratio of the FGRS in the upper vs two lower deciles from our multivariable prediction model was, in order, DUD - 12.6, MD - 4.9, BD - 4.5, ADHD - 3.3 and schizophrenia 1.4. From the lowest to highest quintile, our measures of genetic specificity increased more than five-fold for i) MD vs. Anxiety Disorders, ii) MD vs BD, iii) MD versus alcohol use disorder (AUD), iv) BD vs schizophrenia and v) DUD vs AUD. This increase was nearly two-fold for ADHD vs DUD. We conclude that the level of genetic liability for our psychiatric disorders could be substantially enriched by selection of cases with our predictors. Specificity of genetic risk could also be substantially impacted by these same predictors.

Fig. 1. The relationship between the decile of Family-Genetic Risk Score (FGRS) from our multivariable analyses (using a split-half training and test sample) predicting the strength of the genetic risk for, respectively (from upper left to lower right) major depression (MD), drug use disorder (DUD), bipolar disorder (BD), ADHD and schizophrenia (SZ).

The y-axis in each figure represents the mean FGRS (±95%CIs) for affected individuals in each decile. The x-axis represents the decile of the score from our multivariable regression predicting the strength of the genetic risk. The FGRS is a standardized measure of genetic risk obtained from extended pedigrees so that an FGRS score of 1.0 reflects a mean genetic risk of a set of affected individuals 1.0 SD above the population mean. Above each graph, we present three figures that capture, in different ways, the predictive power of our multivariable prediction formula: r², the linear slope of the mean FGRS estimates across the deciles, and the ratio of FGRS of the top decile divided by the two lower deciles.

Fig. 2. The relationship between the quintile of the Family-Genetic Risk Score (FGRS) difference from our multivariable analyses (using a split-half training and test sample) predicting the difference in genetic risk for, the following six pairs of disorders (from upper left to lower right): Major Depression vs. Anxiety Disorders (MD vs AD), Major Depression vs. Bipolar Disorder (MD vs BD), Major Depression vs. Alcohol Use Disorder (MD vs. AUD), Bipolar Disorder vs Schizophrenia (BD vs SZ), Drug Use Disorder vs. Alcohol Use Disorder (DUD vs AUD) and ADHD vs. Drug Use Disorder.

In these comparisons, we examine individuals affected with the first of the pair of disorders and predict clinical features that will maximize the difference in FGRS of the two disorders. The y-axis in each figure represents the mean difference in the FGRS scores (±95%CIs) for the two disorders. The x-axis represents the quintile of the score from our multivariable regression predicting the differences in the genetic risk. Above each graph, we present two figures that capture, in different ways, the predictive power of our multivariable prediction formula: r² and the linear slope of the mean FGRS estimates across the quintiles.