Leveraging AI to transform medical research dissemination

Client

The National Institute for Health Research (NIHR) is the nation’s largest funder of health and care research. With a mission to improve the health and wealth of the nation through research, the NIHR works in partnership with the NHS, universities, local government, other research funders, patients and the public to deliver and enable world-class research that transforms people’s lives, promotes economic growth and advances science.

Background

The National Institute of Health Research (NIHR) faced significant challenges in their research dissemination process. Their manual approach to summarising research papers into alert formats was resource-intensive, creating a substantial bottleneck.

Despite approximately 1,500 new research papers being published monthly, resource constraints limited the production to only 10 alerts per month, significantly restricting the reach and impact of valuable research findings.

Solution

Audacia collaborated with NIHR to develop and implement an AI-powered solution to transform the research alert generation process. The project began with a focused 5-day proof of concept to validate the approach and demonstrate potential benefits, laying the groundwork for a comprehensive implementation roadmap.

At the core of the solution was the development of a custom NIHRbot, a specialised AI tool designed to generate alerts that seamlessly mirror the style and tone of manually written content. To achieve this level of consistency, the system was fine tuned using 500 previous alerts, ensuring the AI-generated content maintained NIHR’s established standards and communication style. Careful prompt engineering techniques were employed to structure model responses effectively, maintaining the precise format and depth required for research alerts.

The system's capabilities were extended to include direct extraction from various research paper formats, including PDFs and web-based articles. This automated approach eliminated the need for manual transcription and initial summarisation, significantly streamlining the alert generation process. The solution was further enhanced to handle multiple papers simultaneously, supporting the creation of themed research collections that provide comprehensive coverage of specific topics.

To ensure accurate interpretation of complex research materials, we integrated sophisticated extraction techniques that could identify specific sections, interpret figures, and analyse diagrams within research papers. This enhanced document analysis capability ensures that critical visual and technical information is accurately captured and conveyed in the generated alerts.

The solution also includes intelligent search functionality, leveraging the Dimensions API and natural language processing to efficiently identify relevant papers for alert generation. This feature helps staff quickly locate and process the most impactful research papers from the thousands published each month, ensuring optimal use of resources and maximum value for stakeholders.

Throughout the development process, we conducted thorough evaluations of various Large Language Models (LLMs), specifically comparing the performance of GPT-4 Turbo, GPT-3.5 Turbo, and Claude to determine the most effective technology for alert generation. This comparative analysis assessed each model's ability to accurately interpret and summarise complex medical research while maintaining the required tone and format. This careful assessment and selection process ensured that the final solution would deliver consistent, high-quality results while maintaining the scientific accuracy essential for health research communication.

Results

The AI-powered solution delivered significant improvements to NIHR’s research dissemination capabilities:

Leveraging AI enables NIHR to disseminate research findings effectively, significantly increasing the reach and impact of important health research.