Title: The Limits of Rating Scales: Using LLMs to Uncover ADHD Indicators in Turkish Teacher Accounts

Abstract:

Attention Deficit Hyperactivity Disorder (ADHD) stands as one of the most prevalent neurodevelopmental conditions in children. Its diagnosis typically involves a multifaceted approach, integrating professional clinical judgment with standardized rating scales alongside reports from parents and educators. While structured tools like the Conners' Teacher Rating Scale-Revised Short Form (CTRS-R:S) effectively quantify behaviors associated with ADHD, educators also supply open-ended narratives. These qualitative accounts may hold supplementary insights that structured metrics often miss. Yet, the degree to which these teacher narratives contain signals ignored by conventional rating scales remains an open question.

This research examines de-identified Turkish teacher evaluation forms gathered during clinical ADHD assessments, analyzing both CTRS-R:S scores and accompanying open-ended narratives. By comparing predictive indicators derived from structured scores against those from narrative text, we pinpoint instances where standardized assessments struggle to differentiate between students with and without ADHD, whereas models based on narrative text successfully identify distinct behavioral trends. Interestingly, the cases identified by the narrative model show very little overlap with those overlooked by structured tools, indicating that structured data and narrative content provide complementary information.

To make sense of these discrepancies, we employ an LLM-assisted theme discovery pipeline. This approach uncovers specific patterns related to attention, behavior, and family dynamics. The findings underscore the potential of natural language processing (NLP) to extract clinically significant insights from teacher narratives, thereby offering a valuable complement to traditional ADHD screening methods.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC