Detecting comorbidity patterns in rare disease patients with machine learning.

Introduction: Whilst individually rare, affecting only a small percentage of the population, rare diseases as a whole impact around 6% of the global population (with this number likely an underestimate). Rare diseases are often complex, with specific challenges in diagnosis, management, and treatment due to limited knowledge and research. Rare disease patients have been shown to have more comorbidities compared to those without a rare disease diagnosis. Studying comorbidities in patients with rare diseases is particularly important as these patients may exhibit unique patterns of multiple diseases which are not well understood. Understanding these comorbidity patterns can lead to insights into the etiology and progression of rare diseases, potentially identifying new therapeutic targets and improving clinical management strategies. Additionally, studying comorbidities can help in predicting complications, improving the quality of life of patients, and offering a more comprehensive approach to health care for those affected by rare diseases. Methods: A machine learning based method known as hierarchical clustering was applied to diagnosis data from the UK Biobank to study comorbidity patterns in patients with rare diseases. The results were then compared with patterns detected for the general population. Results: Twelve clusters were identified for the rare disease group, and 14 for the no rare disease group. Discussion: Unique comorbidity patterns were observed for individuals with and without a rare disease diagnosis, highlighting potential priorities for intervention to improve both disease management and patient care.