Unsupervised machine learning models reveal two distinct post-operative physical activity profiles among joint...

To identify and characterise distinct post-operative physical activity profiles in joint arthroplasty patients.

This cohort study utilised wrist-worn accelerometer data from the UK Biobank, linked to hospital records, to identify patients who underwent primary unilateral hip or knee arthroplasty. Daily step counts from 4 to 12 months post-operatively were extracted using validated algorithms. Principal component analysis (PCA) was applied to demographic and clinical variables to reduce dimensionality, followed by clustering using k-means and Partitioning Around Medoids (PAM). The optimal number of clusters was determined using the elbow method and silhouette score. Clustering validity was assessed using the Rand Index and Adjusted Rand Index.

237 patients were included, the majority of whom were female, with a mean age of 66 years. Based on the elbow plot and the highest average silhouette width, a two-cluster solution was deemed optimal, consistently emerging across both clustering methods as distinct high-and low-performing physical activity profiles. High performers had significantly higher daily step counts (mean > 10,000 vs. < 6,000, P < 0.001), were younger, had a lower body mass index, fewer comorbidities, and were more likely to have undergone total hip replacement. Sociodemographic factors such as higher educational attainment and lower deprivation index were also associated with the high-performing group. The clustering methods demonstrated a weak-but-positive agreement (ARI = 0.224).

Unsupervised learning of accelerometer-derived physical activity data revealed two clinically meaningful recovery profiles following joint arthroplasty. These findings underscore the multifactorial nature of post-operative recovery and support the development of personalised rehabilitation strategies to improve outcomes in lower limb arthroplasty patients. Video Abstract.