ADHD consists of a count of symptoms that often presents heterogeneity due to overdispersion and excess of zeros. Statistical inference is usually based on a dichotomous outcome that is underpowered. The main goal of this study was to determine a suited probability distribution to analyze ADHD symptoms in Imaging Genetic studies.
We used two independent population samples of children to evaluate the consistency of the standard probability distributions based on count data for describing ADHD symptoms.
We showed that the zero-inflated negative binomial (ZINB) distribution provided the best power for modeling ADHD symptoms. ZINB reveals a genetic variant, rs273342 (Microtubule-Associated Protein [MAPRE2]), associated with ADHD (p value = 2.73E-05). This variant was also associated with perivascular volumes (Virchow-Robin spaces; p values < 1E-03). No associations were found when using dichotomous definition.
We suggest that an appropriate modeling of ADHD symptoms increases statistical power to establish significant risk factors.