How AI helped NIH make literature reviews 6x faster

The National Institutes of Health needed to conduct a comprehensive literature review of new approach methods (NAMs), or research that doesn’t rely on animal testing.

They sought to more quickly determine which technologies are available and ready to implement to evaluate these innovative research alternatives.

But with thousands of scientific papers to assess—and more published every day—the manual review process was slow, resource-intensive, and unsustainable. NIH needed a faster, more scalable approach that also preserved scientific rigor.

Challenge

The NIH Complement Animal Research in Experimentation (Complement-ARIE) Common Fund Program, which supports research to advance understanding of human health and disease, provides funds to projects that deliver the maximum return on taxpayers’ investment. To do so, staff in the program often perform an extensive literature review of thousands, even millions, of scientific papers. Traditionally completed manually, this process is effective in identifying scientific research opportunities that could benefit from federal funding. Yet the process is cumbersome and can swallow up months—if not years—of staff time. That gets expensive, fast.

Solution

To accelerate its NAM literature review, NIH partnered with us to build a reproducible, scalable, and validated GenAI tool.

Our team of subject-matter experts and developers created a flexible workflow that integrates traditional machine learning (via our Litstream® product) and Anthropic’s Claude V2 Foundation Model. This tool is adaptable to meet the literature review needs of different agencies and takes advantage of new capabilities available in ICF Fathom, our suite of tailored AI solutions and services.

We conducted validation exercises on the tool and found it delivered a high degree of accuracy (91%). The tool reduced the overall literature review timeline from 6–12 months to mere weeks.

Results

By implementing a GenAI-driven workflow, ICF reduced the labor required for literature reviews by 95%, significantly accelerating project timelines. The solution is both scalable and reproducible, making it well-suited for scientific applications that demand consistency and reliability.

The success of this solution has led to its adoption by additional clients and earned public recognition. In 2024, ICF received the NIH Director’s Award for our role in informing strategic planning and development of the Complement-ARIE Common Fund Program.