Use of a Generative Pretrained Transformer to Answer Questions and Facilitate a Large Randomized Controlled Trial

Area of Science:

• Clinical Research
• Artificial Intelligence
• Medical Informatics

Background:

• Generative artificial intelligence (AI) shows potential for enhancing clinical trial operations.
• This study investigated the utility of a customized generative pretrained transformer (GPT) in supporting clinical trial sites.

Purpose of the Study:

• To determine if a customized GPT could provide rapid responses to protocol and procedure-related questions for clinical trial sites.
• To assess the effectiveness and user satisfaction of a GPT-based support system in a real-world clinical trial setting.

Main Methods:

• A customized GPT was developed and implemented within the Sleep SMART (Sleep for Stroke Management and Recovery Trial).
• The GPT provided real-time answers to procedure-related questions for active trial sites.
• Central study teams monitored questions and responses, and primary study coordinators completed an anonymous survey on their experience.

Main Results:

• The GPT answered 75% of the 785 questions received over a 10-month period.
• Manual review indicated that 98% of GPT-answered questions were helpful and complete.
• Of surveyed coordinators who used the GPT, 89% found it very helpful or helpful, and 79% deemed responses accurate.

Conclusions:

• A customized GPT shows promise as a valuable tool for supporting clinical trial sites.
• Further research is recommended to validate the GPT's accuracy and safety against human support.
• This technology could streamline information dissemination and improve efficiency in large-scale clinical trials.