Leveraging artificial intelligence to summarize abstracts in lay language for increasing research accessibility and...

Area of Science:

• Clinical Research
• Artificial Intelligence
• Health Communication

Background:

• Returning aggregate study results to the public is an ethical imperative but faces barriers like cost and infrastructure.
• Developing lay summaries for a general audience is time-consuming and resource-intensive.
• Limited public accessibility hinders trust and informed decision-making regarding clinical studies.

Purpose of the Study:

• To generate, evaluate, and implement AI-generated lay summaries using ChatGPT 4.
• To assess the accuracy, relevance, accessibility, and transparency of AI-generated lay summaries.
• To integrate AI lay summaries into ResearchMatch, a national clinical study recruitment platform, for scalable results dissemination.

Main Methods:

• AI prompts were engineered to create summaries at an accessible literacy level, focusing on clarity and relevance.
• Researchers and volunteers evaluated AI-generated lay summaries on accuracy, relevance, accessibility, transparency, and harmfulness.
• Precision analysis and adaptive random sampling determined the optimal number of summaries for evaluation.

Main Results:

• AI-generated lay summaries demonstrated high accuracy (95.9%) and relevance (96.2%) based on researcher review.
• A majority of volunteers found AI summaries more accessible (85.3%) and transparent (73.5%) than original abstracts.
• No harmful content was identified in the AI-generated summaries, and over 750 studies on ResearchMatch were updated with lay summaries.

Conclusions:

• AI-generated lay summaries offer a scalable and cost-effective solution for disseminating research findings.
• Implementing AI summaries on ResearchMatch enhances research accessibility and transparency for the public.
• This framework is generalizable to other platforms, improving public engagement with scientific research.