Distinct Recovery Phenotypes Identified Following Anterior Cruciate Ligament Reconstruction Through Multivariate Clustering: Implications for Post-Traumatic Osteoarthritis Risk Stratification

Objectives: Anterior cruciate ligament injuries are associated with a risk of developing post-traumatic osteoarthritis (PTOA). Knee PTOA disease incidence and progression has been linked to lower knee extensor strength, which may be a consequence of muscle atrophy following injury. Furthermore, intra-muscular fat infiltration may occur in the injured leg, which is concerning given a higher intramuscular fat tissue content is associated with accelerated joint degeneration in idiopathic OA. In parallel to potential soft tissue maladaptations after injury, bone tissue may be affected. Although the role of bone health in PTOA development is still unclear, if altered densities and geometries exist, there are potential implications of altered bone metabolism and mechanics influencing OA pathogenesis. To date, no studies have simultaneously evaluated soft tissue, bone composition, and mechanical properties of bone across multiple scan sites in both the thigh and shank. Consequently, the aim of this study was to determine whether multivariate clustering could identify biologically distinct subgroups that may reflect differing levels of tissue integrity which may serve as a foundation for future longitudinal research in PTOA risk stratification.

Methods: Thirty-two participants (18 female, 14 male; 26 ± 6 years old; 25.12 ± 3.95 kg/m2) 1-6 years after ACL reconstruction (ACLR) underwent three peripheral quantitative computed tomography (pQCT) scans (XCT 3000, Stratec Medizintechnik GmbH, Pforzheim, Germany) at the tibia (66% from distal growth plate) and femur (20% and 50% from distal growth plate) in the ACLR leg and the uninjured contralateral leg. pQCT derived bone parameters were cortical density (CoD), cortical area (CoA), trabecular density (TraD), and trabecular area (TraA). Mechanical bone properties were evaluated using strength stress index (SSI), polar second moment of area (iPo), and compressive strength index (BSId). Soft tissue variables measured were muscle cross sectional area (MCSA) and muscle density (MD). Limb symmetry indices were calculated as (ACLR leg / contralateral uninjured leg) × 100, for each pQCT variable. pQCT derived variables were z-transformed; 0.5% missing data were imputed via multivariate iterative imputation. Principal component analysis (PCA) reduced dimensionality and addressed multicollinearity, explaining 91.2% of the variance. Hierarchical agglomerative clustering (Ward’s linkage, Euclidean distance) was conducted. Internal clustering validity was evaluated using Silhouette Score, Davies-Bouldin Index (DBI), and Calinski-Harabasz Index (CHI).

Results: Hierarchical clustering produced a Silhouette Score of 0.20, a DBI of 1.70 indicating acceptable separation, and CHI of 8.17 reflecting moderate between-cluster dispersion. Nine participants (28%) were classified as high integrity recovery (HIR) and 23 (72%) as low integrity recovery (LIR). Cluster visualisations in PCA space demonstrated clear qualitative separation.

Conclusion: Cluster analysis identified two distinct recovery phenotypes. The LIR group presented lower LSI values for MCSA, MD, CoA, TraA, TraD, SSI, iPo, and BSId, indicating widespread deficits in muscle and bone geometries and densities, and bone mechanical properties. Early identification of this LIR phenotype may enable targeted strategies to restore musculoskeletal integrity and reduce future PTOA risk. Longitudinal research is needed to determine whether individuals in this group experience accelerated structural decline or symptom progression over time.