Presentation (CTI)

OBJECTIVES: Rare diseases impact over 300 million people globally and many have no approved treatments. Orphan drug designations are offered by regulatory agencies if it can be demonstrated that the prevalence of disease falls within a specific threshold. Rapid reviews can quickly identify evidence necessary to support regulatory applications. Artificial intelligence (AI) has been shown to expedite literature review processes, but the capacity in which AI can support rapid reviews without sacrificing quality or robustness has yet to be established. We aimed to investigate whether a rapid review approach with AI integration was able to replicate the findings of published epidemiology systematic literature reviews (SLR) with meta-analysis (MA).
METHODS: Two recently published SLR/MAs evaluating the epidemiology of rare diseases were replicated as rapid reviews using Nested Knowledge AI capabilities. Smart search developed an algorithm to search PubMed for literature relevant to the research questions of the selected reviews. CORE smart tags were applied to the search yield and used to inform PICO-based screening decisions. The AI screening model was trained and advancement probabilities were leveraged for inclusion/exclusion decisions. For included records, data were manually extracted and used as inputs to automatically generate pooled incidence and prevalence estimates.
RESULTS: AI tools identified and selected 54% to 73% of the articles included in the published SLR/MAs while also capturing additional relevant articles not previously included in a fraction of the time needed for a systematic approach. Despite some differences in study inclusion, pooled prevalence and incidence estimates were similar to those previously published.
CONCLUSIONS: AI and automation tools are crucial for quick evidence generation, and rapid reviews can arrive at similar findings to MAs where the evidence base was identified by more robust SLRs. This approach may be beneficial for supporting orphan drug applications which require evidence of low prevalence in the geographic areas of interest.