Translation of biomarkers to clinical practice is hindered by the significant overlap in neurobiological measures between ADHD cases and controls. A risk-informed design can enhance the utility and validation of ADHD biomarkers by highlighting differences between individuals with ADHD and those without at differential risk. Participants were 2511 children and adolescents (aged 6 to 14 years) from the Brazilian High Risk Cohort for Mental Conditions. We calculated risk for ADHD among unaffected individuals using a multivariable clinical and sociodemographic risk model. We compared measures of three proposed ADHD biomarkers (polygenic scores, subcortical volumes, and executive function) between participants with vs. without ADHD, and ADHD vs. without ADHD with a high- vs. low-risk loading for ADHD. Compared to the unaffected group, children and adolescents with ADHD had higher ADHD polygenic scores (cohen's d = 0.17), smaller subcortical volumes (d = - 0.25), and poorer executive function (d = - 0.22). Separating the unaffected group into low- and high-risk subgroups revealed more pronounced differences (Cohen's d = 0.20 to 0.60) and nearly doubled the overlap-free area for these three neurobiological measures between the low-risk group and the other two groups. Upon adjustment for the number of ADHD symptoms, simple ADHD vs. without ADHD differences vanished, while the risk-informed analyses remained significant. Here, we demonstrate that a risk-based design increases effect sizes when comparing candidate biomarkers for ADHD. Our study provides a model that may hold promise for evaluating similar contrasts in other mental disorders and samples.
Keywords: Attention-deficit/hyperactivity disorder (ADHD); Machine learning; Multivariable; Neurobiological biomarkers.
© 2024. Springer-Verlag GmbH Germany, part of Springer Nature.