Digital Cousins in Arthroplasty: An Interpretable Clustering Framework

Digital twins are increasingly proposed as a route toward personalized medicine, but in routine clinical data the more realistic near-term goal may be the identification of digital cousins, that is, clinically similar patients who can support outcome interpretation. This study explored whether interpretable clustering of baseline arthroplasty data can provide a useful methodological foundation for outcome analysis. Hip arthroplasty-related records were extracted from MIMIC-IV and separated into primary and revision representations. Because MIMIC-IV is an acute-care, ICU-enriched hospital dataset rather than a dedicated arthroplasty registry, this cohort should be interpreted as a medically complex arthroplasty subset rather than the standard elective population. This non-registry selection is also clinically relevant because revision-related hospital episodes are high-risk and burdensome. Patients were clustered using standardized continuous variables (age, body mass index, and Charlson Comorbidity Index), while grouped diagnosis categories were reserved for post hoc interpretation rather than Euclidean distance calculations. The resulting clusters produced clinically readable patient profiles and highlighted differences between primary and revision surgery patterns. The findings support an interpretable digital cousin framework rather than a full digital twin claim.