Unsupervised machine learning identifies two distinct posterior tibial slope clusters largely independent of Coronal Plane Alignment of the Knee in osteoarthritic knees.

PURPOSE: Coronal-plane classifications such as the Coronal Plane Alignment of the Knee (CPAK) classification support individualized total knee arthroplasty (TKA) planning but do not account for sagittal-plane morphology, particularly the posterior tibial slope (PTS). This study aimed to determine whether unsupervised machine learning (k-means clustering) could identify naturally occurring PTS phenotypes in osteoarthritic knees and to evaluate their association with CPAK classification and joint line obliquity (JLO). METHODS: This retrospective radiographic cohort study included 3337 osteoarthritic knees from 1746 patients. PTS was measured on lateral knee radiographs in all 3337 knees, using the posterior tibial cortex and the medial tibial plateau as reference lines. Full-length standing radiographs were available for 464 knees; in these knees, the medial proximal tibial angle (MPTA) and the lateral distal femoral angle (LDFA) were measured to calculate the arithmetic hip-knee-ankle angle (aHKA) and JLO, and each knee was classified according to CPAK. K-means clustering was used to identify PTS subgroups without predefined thresholds, and cluster validity was assessed using the elbow method and silhouette analysis. Associations between PTS clusters, CPAK phenotypes and radiographic parameters were analysed using the χ 2 test, adjusted standardized residuals, Fisher's exact test and correlation analysis. RESULTS: The median PTS was 5.6° (interquartile range 3.4°-7.7°). K-means clustering identified two PTS clusters: a shallow-slope cluster centred at 3.56°, comprising 55.91% of the knees and a steep-slope cluster centred at 8.52°, comprising 44.09%. No individual CPAK phenotype was significantly overrepresented within either cluster. When phenotypes were grouped according to joint line orientation, however, apex-distal phenotypes (I-III) were more frequently assigned to the steep-slope cluster, whereas apex-neutral phenotypes (IV-VI) were more frequently assigned to the shallow-slope cluster. JLO was significantly higher in the shallow-slope cluster and PTS showed a modest inverse correlation with JLO. CONCLUSIONS: Unsupervised k-means clustering identified two distinct sagittal PTS phenotypes in osteoarthritic knees. PTS varied largely independently of the individual CPAK phenotypes, although an association with JLO was observed. Sagittal morphology may therefore provide complementary information beyond coronal alignment classifications for the individualized planning of knee surgery. LEVEL OF EVIDENCE: Level IV.