A New Perspective on Clinical Scale Validation: Leveraging Sentence Embeddings for Pre-Validation Semantic Analysis

Résumé fourni par la source

Traditional clinical scale development is a lengthy and resource-intensive process, particularly during the empirical validation phase. This paper proposes a paradigm shift by introducing a computational pre-validation framework that leverages sentence embedding models to assess semantic properties of scale items prior to data collection. By transforming items into high-dimensional vector representations and analyzing their relationships within an MTMM-like structure, our method identifies potential item ambiguity, evaluates construct coherence, and tests discriminant validity at an early stage. We demonstrate its utility using widely used clinical instruments—including the SNAP-IV, CBCL Aggressive Behavior scale, and DSM-5-TR Oppositional Defiant Disorder (ODD) criteria—revealing semantic patterns that align with established psychometric properties. While not intended to replace traditional validation, this approach offers a novel, cost-effective complement that can inform early revisions and reduce pilot testing cycles. We develop an open-source R package “embedScaleValid” to facilitate practical implementation. This work introduces a new perspective in clinical assessment science, integrating advances in NLP to enhance the efficiency and validity of psychological measurement tools.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.