Integration of symptom ratings from multiple informants in ADHD diagnosis: a psychometric model with clinical utility

The Diagnostic and Statistical Manual of Mental Disorder-Fifth Edition explicitly requires that attention-deficit/hyperactivity disorder (ADHD) symptoms should be apparent across settings, taking into account reports from multiple informants. Yet, it provides no guidelines how information from different raters should be combined in ADHD diagnosis. We examined the validity of different approaches using structural equation modeling (SEM) for multiple-informant data. Participants were 725 children, 6 to 17 years old, and their primary caregivers and teachers, recruited from the community and completing a thorough research-based diagnostic assessment, including a clinician-administered diagnostic interview, parent and teacher standardized rating scales, and cognitive testing. A best-estimate ADHD diagnosis was generated by a diagnostic team. An SEM model demonstrated convergent validity among raters. We found relatively weak symptom-specific agreement among raters, suggesting that a general average scoring algorithm is preferable to symptom-specific scoring algorithms such as the "or" and "and" algorithms. Finally, to illustrate the validity of this approach, we show that averaging makes it possible to reduce the number of items from 18 items to 8 items without a significant decrease in validity. In conclusion, information from multiple raters increases the validity of ADHD diagnosis, and averaging appears to be the optimal way to integrate information from multiple raters.