Artificial intelligence in reproductive endocrinology: an in-depth longitudinal analysis of ChatGPTv4's month-by-month interpretation and adherence to clinical guidelines for diminished ovarian reserve.

Abstract

Objective: To quantitatively assess the performance of ChatGPTv4, an Artificial Intelligence Language Model, in adhering to clinical guidelines for Diminished Ovarian Reserve (DOR) over two months, evaluating the model's consistency in providing guideline-based responses.

Design: A longitudinal study design was employed to evaluate ChatGPTv4's response accuracy and completeness using a structured questionnaire at baseline and at a two-month follow-up.

Setting: ChatGPTv4 was tasked with interpreting DOR questionnaires based on standardized clinical guidelines.

Participants: The study did not involve human participants; the questionnaire was exclusively administered to the ChatGPT model to generate responses about DOR.

Methods: A guideline-based questionnaire with 176 open-ended, 166 multiple-choice, and 153 true/false questions were deployed to rigorously assess ChatGPTv4's ability to provide accurate medical advice aligned with current DOR clinical guidelines. AI-generated responses were rated on a 6-point Likert scale for accuracy and a 3-point scale for completeness. The two-phase design assessed the stability and consistency of AI-generated answers over two months.

Results: ChatGPTv4 achieved near-perfect scores across all question types, with true/false questions consistently answered with 100% accuracy. In multiple-choice queries, accuracy improved from 98.2 to 100% at the two-month follow-up. Open-ended question responses exhibited significant positive enhancements, with accuracy scores increasing from an average of 5.38 ± 0.71 to 5.74 ± 0.51 (max: 6.0) and completeness scores from 2.57 ± 0.52 to 2.85 ± 0.36 (max: 3.0). It underscored the improvements as significant (p < 0.001), with positive correlations between initial and follow-up accuracy (r = 0.597) and completeness (r = 0.381) scores.

Limitations: The study was limited by the reliance on a controlled, albeit simulated, setting that may not perfectly mirror real-world clinical interactions.

Conclusion: ChatGPTv4 demonstrated exceptional and improving accuracy and completeness in handling DOR-related guideline queries over the studied period. These findings highlight ChatGPTv4's potential as a reliable, adaptable AI tool in reproductive endocrinology, capable of augmenting clinical decision-making and guideline development.