Systematic Isolation of Key Parameters for Estimating Skeletal Maturity on Knee Radiographs

Abstract

Background:The ability to estimate skeletal maturity using a knee radiograph would be useful in anterior cruciate ligament (ACL) injuries and limb-length discrepancy in immature patients. Currently, a quick, accurate, and reproducible method is lacking.Methods: Serial knee radiographs made 3 years before to 2 years following the chronologic age associated with 90% of final height (an enhanced skeletal maturity gold standard compared with peak height velocity) were analyzed in 78 children. The Pyle and Hoerr (PH) knee method was simplified by developing discrete stages for the distal part of the femur, the proximal part of the tibia, the proximal part of the fibula, and the patella. The Roche-Wainer-Thissen (RWT) knee method was simplified from the 36 original parameters to 14 parameters by removing parameters that were poorly defined, were not relevant to the peripubertal age range, were poorly correlated with 90% final height, or were poorly reliable on a 20-radiograph pilot analysis. We also compared the recently described central peak value (CPV) of the distal part of the femur. The Greulich and Pyle (GP) left-hand bone age was included for comparison.Results: In this study, 326 left knee radiographs from 41 girls (age range, 7 to 15 years) and 37 boys (age range, 9 to 17 years) were included. Stepwise linear regression showed higher correlation in predicting years from 90% final height using the modified RWT and demographic characteristics (R 2 = 0.921) compared with demographic characteristics alone (R 2 = 0.840), CPV and demographic characteristics (R 2 = 0.866), GP and demographic characteristics (R 2 = 0.899), and PH and demographic characteristics (R 2 = 0.902). Seven parameters were excluded from the RWT and demographic characteristics model using stepwise linear regression and generalized estimating equations analysis, leaving 7 parameters (2 femoral, 4 tibial, and 1 fibular) in the final model. Compared with RWT and demographic characteristics (R 2 = 0.921), there were minimal incremental increases by adding CPV (R 2 = 0.921), GP (R 2 = 0.925), or PH (R 2 = 0.931).Conclusions: This large analysis of knee skeletal maturity systems isolated 7 discrete radiographic knee parameters that theoretically outperform the GP bone age in estimating skeletal maturity.Clinical Relevance: We present a modified knee skeletal maturity system that can potentially preclude the need for additional imaging of the hand and wrist in reliably estimating skeletal maturity.A ccurate estimation of skeletal maturity is critical in the treatment of certain pediatric orthopaedic conditions, including limb-length discrepancy and anterior cruciate ligament (ACL) injury, where surgical timing and technique depend on remaining growth 1,2 . The Greulich and Pyle (GP) bone age atlas 3 and the Tanner-Whitehouse staging system 4,5 remain 2 of the most commonly used systems to estimate remaining growth. However, these systems are based on subjective interpretation of radiographs and thus frequently produce inconsistent results 1 . Addition...

Estimating Skeletal Maturity by Segmented Linear Modeling of Key AP Knee Radiographic Parameters

“…9 Interrater and intrarater reliability values were > 0.6 (range, 0.679 to 0.927) for all 7 parameters (intraclass correlation coefficient of > 0.6 for continuous variables, φ coefficient of > 0.6 for dichotomous categorical variables, and Spearman ρ of > 0.6 for categorical variables). 9 Segmented linear regression analysis was performed on these remaining 7 parameters to create models for estimating skeletal age. Four models were created by partitioning data based on different variables: (1) sex, (2) chronological age, (3) a combination of sex and chronological age, and (4) a combination of sex and capping of the lateral distal femoral epiphysis over the metaphysis (FemK).…”

“…9,15 Patients with at least 1 radiograph of the left knee obtained during peripubertal years (ages 9 to 17 in males, ages 7 to 15 in females) and available final height measurements were included, with knee skeletal maturity grading available in our preexisting and previously reported dataset. 9 Forty-two girls (194 radiographs) and 38 boys (190 radiographs) met these criteria. Fifty-four radiographs were excluded due to out-of-plane rotation on the AP view, and 4 radiographs were excluded due to poor quality, leaving 78 patients (326 radiographs), including 41 girls (172 radiographs) and 37 boys (154 radiographs), remaining for analysis.…”

“…FemK is a trichotomous parameter representing the degree of FemK and was chosen based on its high correlative strength with skeletal age among the 7 parameters. 9 To account for multiple visits from the same subject, parameters identified as significant by stepwise linear regression were included in a generalized estimating equation (GEE) analysis (Appendix I, Supplemental Digital Content 1, http://links.lww.com/BPO/A419). Cases with ≥ 1 missing data points (n = 42) were handled using listwise deletion so that each model was created using only radiographs with complete parameter data (n = 284).…”

Systematic Isolation of Key Parameters for Estimating Skeletal Maturity on Anteroposterior Wrist Radiographs

“…Importantly, the timing of 90% FH can be easily calculated using a longitudinal data set with yearly data. In recent studies, skeletal maturity systems developed with the 90% FH standard have had promising results 8,11 .…”

“…Development of a new skeletal maturity system requires the developer to choose a skeletal maturity standard to calibrate the system's predictions 7,8 . Peak height velocity (PHV) has been the most widely used skeletal maturity standard to date 7,9,10 .…”