Interobserver reliability of Coronal Plane Alignment of the Knee (CPAK) phenotype classification : external...

Coronal Plane Alignment of the Knee (CPAK) phenotyping is gaining momentum in research and clinical practice to understand individualized knee alignments and predict knee balance in total knee arthroplasty (TKA). The nine CPAK classes are based on joint line obliquity (JLO) and arithmetic hip-knee-ankle angle (aHKA), which are calculated using the medial proximal tibial angle (MPTA) and lateral distal femoral angle (LDFA). This study aims to assess CPAK classification reproducibility, and analyze what level of angular error is associated with CPAK misclassification.

Two readers labelled 75 long-leg radiographs (LLRs) from the Osteoarthritis Initiative database for analyses of CPAK inter-reader reproducibility. A single reader then labelled and classified phenotypes for an aggregate total of 1,128 LLRs. Finally, Monte Carlo simulations were run based on 1,128-patient phenotype distribution and the inter-reader reproducibility statistics to understand how CPAK agreement rates were affected by the reproducibility of MPTA and LDFA measurements.

There was excellent reproducibility in MPTA and LDFA measurements (mean absolute error: 0.41°/0.71°; and intraclass correlation coefficient: 0.96°/0.91°, respectively). These small angular deviations led to one-in-five disagreement in CPAK classification (20.0%; 95% CI 10.9% to 29.1%). An aHKA mean absolute error of < 0.1°, which is potentially unattainable, would be required to reduce inter-reader CPAK disagreement to below 95%.

CPAK phenotyping from long-leg radiographs may result in clinically significant rates of misclassification. CT imaging may improve reliability, particularly in cases where aHKA and JLO are near to discriminatory values.