OBJECTIVES : Every year, HEOR researchers expand the depth and breadth of research available for the use of artificial intelligence (AI) and machine learning (ML). Our ability to examine terabytes of data is greatly expanding the way we can tackle previous intractable problems. One area of recent interest for the application of AI/ML is the ability to predict patients that may eventually be diagnosed with rare diseases as well as our ability to aid physicians with more accurate and/or earlier confirmation of a rare disease diagnosis. Both of these research tracts ultimately afford the medical community better tools to help hasten the delivery of treatments to these patients.

METHODS : Our landscape assessment evaluated two questions: what research is being conducted (A) to enhance identification of variables that may predict rare disease diagnosis in the future and (B) to evaluate uses of AI/ML based algorithms in medical practice to enable easier diagnosis of rare disease patients (hereinafter called identify and diagnosis tracts, respectively). Both research tracts used PubMed to identify research conducted in these areas over the past five years that was published in English.

RESULTS : Identify tract found 61 articles. Research was conducted primary in areas associated with CNS, oncological, and genetic disorders. In this identification research, genomic data are used most when predicting rare disease; computational imaging data is second-most common. Although many computation algorithms are represented in the research, the most common are Deep Learning, Neural Networks, and Random Forest. For diagnosis tract, there are far fewer studies. Beyond six studies reviewed by Brasil et al. (2019), two additional papers were found. The diseases, AI/ML methods, and research exhibited no fixed pattern.

CONCLUSIONS : The breadth of research in these fields is far greater than the depth in any one facet. Eventually, we will need both to establish strong methods.