OBJECTIVES: When asking broad research questions that cover a large evidence base, with differences in keywords and terminology between studies, a systematic literature review (SLR) can take more time and resource than are available. There is little guidance on semi-reproducible non-systematic alternatives and no guidance, to our knowledge, on supplementing targeted Population, Intervention, Comparison, Outcomes and Study-based searches with artificial intelligence (AI)-driven searching. We formalized an agile literature review methodology as an alternative to an SLR and applied this to the complex evidence landscape for how diversity is defined in genomics research.

METHODS: Our protocol applied search strings for several sub-questions within PubMed and Google. References were screened and prioritized pragmatically, based on relevance to sub-questions and informal assessment of the strength of the evidence. This allowed iterative searching, with new themes informing additional searches. A subject matter expert reviewed themes identified, noting areas requiring more evidence, for which we designed additional searches. AI-based literature mapping expanded from key studies, forming citation maps to identify similar studies. Policy documents and third-sector initiatives were identified using Google. Results were presented as narrative synthesis.

RESULTS: Our search strings identified >37,000 references in PubMed with low specificity; agile reviewing led to approximately 200 references being reviewed in detail. The combination of PubMed, Google and AI-based mapping had greater sensitivity and specificity than any of the techniques alone. Citation mapping was able to deepen evidence on particular subjects by providing example case studies, but was less useful for uncovering new topics and, used alone, would not have covered the breadth of studies.

CONCLUSIONS: Agile literature reviewing efficiently orientated us around a complex evidence landscape. We prioritized understanding the breadth of information, but maintained an element of reproducibility from the initial search terms. Findings from our agile approach could help design SLRs into specific themes.