Large Language Models for Rapid Instrument Prototyping: Design and Structural Optimization of the Dry Eye Disease in Pregnancy Questionnaire (DED-PREG).

INTRODUCTION

Gestational Dry Eye Disease (DED) affects up to 50% of expectant mothers, yet current diagnostic tools are generic and fail to capture pregnancy-specific symptom patterns. Developing and validating new instruments in this population is logistically and ethically challenging due to recruitment barriers. This study describes the development and computational prototyping of the Dry Eye Disease in Pregnancy Questionnaire (DED-PREG) using a Generative Artificial Intelligence (GenAI) framework.

METHODS

We utilized a multi-stage in silico framework involving two independent synthetic cohorts. First, a qualitative focus group cohort was generated to simulate clinical dialogues for content derivation, followed by semantic vectorization for algorithmic item reduction. Subsequently, an independent validation cohort of 500 pregnant personas was instantiated. We evaluated the resulting 20-item instrument for internal consistency, structural validity, and test-retest reliability via a longitudinal simulation engine utilizing temporal context injection to model gestational progression across five distinct timepoints (T1-T5).

RESULTS

The DED-PREG mapped to three distinct domains: Ocular Symptoms, Functional Impact, and Lifestyle & Environmental Modulators. The instrument demonstrated satisfactory internal consistency (Cronbach's alpha = 0.89) and excellent temporal stability in a strictly stable subsample (ICC = 0.99). Confirmatory Factor Analysis indicated acceptable model fit for synthetic high-dimensional data (CFI = 0.82; RMSEA = 0.11). Longitudinal analysis confirmed the instrument's responsiveness to gestational change (Global Cohen's d = 0.44), with Linear Mixed Models (LMM) revealing a significant interaction between low socioeconomic status and symptom exacerbation (β=0.053, p < 0.001).

CONCLUSION

This study presents the first pregnancy-specific DED instrument structurally optimized via AI simulation. While human validation remains the gold standard, this computational approach demonstrates that GenAI can serve as a rigorous "stress-test" for instrument design, enabling the rapid prototyping of robust clinical tools prior to in vivo deployment.