Strengths and limitations of new artificial intelligence tool for rare disease epidemiology

The recent paper by Kariampuzha et al. describes an exciting application of artificial intelligence to rare disease epidemiology. The authors’ AI model appears to offer a major leap over Orphanet, the resource which is often a “first stop” for basic epidemiological data on rare diseases. To ensure appropriate use of this exciting tool, it is important to consider its strengths and weaknesses in context. The tool currently incorporates only PubMed abstracts, so key information located in the full text of articles is absent. Such missing information may include incidence and prevalence values, as well as important elements of study design and context. Additionally, results from the public version of the tool differ from those described in the original article, including obsolete values for prevalence and the use of non-prevalence studies in place of those listed in the article. At present, it would be appropriate to utilize the AI tool much like Orphanet: a helpful “first stop” which should be manually checked for completeness and accuracy. Users should understand the benefits of this exciting technology, and that it is not yet a panacea for the challenges of analyzing rare disease epidemiology.