Abstract
Knee osteoarthritis (OA) is a disease that can cause substantial pain and disability in patients. The progression of OA has been linked to inflammatory, mechanical, genetic, and metabolic factors, yet patterns of symptoms and structural damage vary considerably between knee OA patients. The heterogeneity of the disease presents a need for identifying and understanding patient subgroups to inform more personalized treatments. Identifying anatomical morphotypes, a type of classification defined by anatomical and morphological attributes, is critical for identifying subgroups of patients who are most likely to benefit from particular treatments. Cluster analysis is an unsupervised learning method that can be used to uncover subgroups in datasets without labeled outcomes to guide the analysis. In this perspective, we will review analytic challenges in identifying anatomical morphotypes using clustering methods, including finding patterns that are not clinically relevant, navigating the unique correlation structure of anatomical data, and working with high dimensional data. With the exciting applications of clustering methods to improve personalized medicine in knee OA, it is essential to consider these analytic challenges to ensure that analyses yield clinically actionable insights.