The use of machine learning in rare diseases: a scoping review

Background Emerging machine learning technologies are beginning to transform medicine and healthcare and could also improve the diagnosis and treatment of rare diseases. Currently, there are no systematic reviews that investigate, from a general perspective, how machine learning is used in a rare disease context. This scoping review aims to address this gap and explores the use of machine learning in rare diseases, investigating, for example, in which rare diseases machine learning is applied, which types of algorithms and input data are used or which medical applications (e.g., diagnosis, prognosis or treatment) are studied. Methods Using a complex search string including generic search terms and 381 individual disease names, studies from the past 10 years (2010-2019) that applied machine learning in a rare disease context were identified on PubMed. To systematically map the research activity, eligible studies were categorized along different dimensions (e.g., rare disease group, type of algorithm, input data), and the number of studies within these categories was analyzed. Results Two hundred eleven studies from 32 countries investigating 74 different rare diseases were identified. Diseases with a higher prevalence appeared more often in the studies than diseases with a lower prevalence. Moreover, some rare disease groups were investigated more frequently than to be expected (e.g., rare neurologic diseases and rare systemic or rheumatologic diseases), others less frequently (e.g., rare inborn errors of metabolism and rare skin diseases). Ensemble methods (36.0%), support vector machines (32.2%) and artificial neural networks (31.8%) were the algorithms most commonly applied in the studies. Only a small proportion of studies evaluated their algorithms on an external data set (11.8%) or against a human expert (2.4%). As input data, images (32.2%), demographic data (27.0%) and omics data (26.5%) were used most frequently. Most studies used machine learning for diagnosis (40.8%) or prognosis (38.4%) whereas studies aiming to improve treatment were relatively scarce (4.7%). Patient numbers in the studies were small, typically ranging from 20 to 99 (35.5%). Conclusion Our review provides an overview of the use of machine learning in rare diseases. Mapping the current research activity, it can guide future work and help to facilitate the successful application of machine learning in rare diseases.