Background ADHD is classically seen as a childhood disease, although it persists in one out of two cases in adults. The diagnosis is based on a long and multidisciplinary process, involving different health professionals, leading to an under-diagnosis of adult ADHD individuals. We therefore present a psychometric screening scale for the identification of adult ADHD which could be used both in clinical and experimental settings. Method We designed the scale from the DSM-5 and administered it to n = 110 control individuals and n = 110 ADHD individuals. The number of items was reduced using multiple regression procedures. We then performed factorial analyses and a machine learning assessment of the predictive power of the scale in comparison with other clinical scales measuring common ADHD comorbidities. Results Internal consistency coefficients were calculated satisfactorily for TRAQ10, with Cronbach’s alpha measured at .9. The 2-factor model tested was confirmed, a high correlation between the items and their belonging factor. Finally, a machine-learning analysis showed that classification algorithms could identify subjects’ group membership with high accuracy, statistically superior to the performances obtained using comorbidity scales. Conclusions The scale showed sufficient performance for its use in clinical and experimental settings for hypothesis testing or screening purpose, although its generalizability is limited by the age and gender biases present in the data analyzed.

Background : ADHD is classically seen as a disease of children, although it persists in one out of two cases in adults. The diagnosis is based on a long and multidisciplinary process, involving different health professionals, leading to an under-diagnosis of adult ADHD individuals. We therefore present a psychometric screening scale for the identification of adult ADHD, in order to serve as an aid in the decision whether or not to engage in a diagnostic process.Method : We designed the scale from the DSM-5 and administered it to n=110 control individuals and n=110 ADHD individuals. The number of items was reduced using regression techniques. We then performed factor analyses and a machine-learning assessment of the predictive power of the scale.Results : Internal consistency coefficients were calculated satisfactorily for TRAQ10, with Cronbach's alpha measured at .9. The 3-factor model tested was confirmed, with standardized factor loadings greater than .53 for all items. Finally, analysis by machine learning showed that a GNB-type classification algorithm could identify the subject's group appartenance with an high average precision of .88 based only on the participant’s responses on the scale.Conclusions : The scale showed sufficient performance for its use in clinical routine. It could thus help to reduce the time required to diagnose ADHD in adults. Similarly, it could be used in research for screening purposes.