Associations between radiographic hip OA assessed using Kellgren–Lawrence grading and hip shape characterised using statistical shape modelling in a British birth cohort

Purpose: To investigate the associations between OA defined using Kellgren–Lawrence grading and hip shape characterised by statistical shape modelling of hip DXA scans.

Methods: The MRC National Survey of Health and Development (NSHD) is a UK birth cohort from 1946. Between the ages of 60 and 64, eligible cohort members (n = 2856) were invited to attend a comprehensive health assessment at one of six clinical research facilities in England, Scotland and Wales, of which 1690 attended (59% of the surviving cohort). N = 1636 individuals had a DXA scan of their hips, although 3 scans were subsequently excluded from analysis due to poor image quality leaving 1633 scans for analysis. Scans of the right hip were modelled using Shape software (University of Aberdeen). The model template consisted of 59 points and modelled the femoral head and neck, greater trochanter, femoral shaft, lesser trochanter and part of the acetabulum. Principal component analysis generated orthogonal modes of variation (HM) and in this study we included 10 modes of variation (80.7% of total variance). Images were graded for radiographic OA (rOA) using the Kellgren–Lawrence grading scale (KLG) by FS and a random selection compared by IF with grades previously assigned by FS. Inter-rater repeatability was determined using quadratic-weighted Kappa scores in MedCalc v17.2. Correlations between age, sex, BMI, KLG and HM1-10 were performed using Spearman's Rank correlations. One-way ANOVAs with Tukey post-hoc analysis were performed to assess the relationships between HM and KLG in those HM identified as being significantly correlated with KLG. All tests were performed in SPSS v23.0.

Results: Quadratic-weighted kappa scores indicated that the reliability of KL grading of images between two separate observers was very good (kappa = 0.967). There were few cases of radiographic OA (rOA) identified in this cohort, the majority of participants had a KLG of 0. Greater numbers of rOA cases (KLG ≥ 2) were identified in men.
Spearman’s rank correlations identified 5 modes of variation that were significantly correlated with KLG (HM1 and HM7-10). One way ANOVA analysis of HM1 (23.0% of total variation) demonstrated an increase in score with increasing KLG (P ≤ 0.001; Fig. 1a). The shape changes described by this mode included migration of the femoral head, flattening of the femoral neck and increased incidence of osteophytes (Fig. 1b). HM7 (4.1% of total variation), associated with flattening of the femoral neck and increased numbers of osteophytes (Fig. 1c) showed a negative association with KLG (Fig. 1d) as did HM8 (3.2% of total variation) describing an increased acetabular coverage of the femoral head and larger superior osteophytes.

Conclusions: Several modes of variation were identified as being associated with radiographic features of OA in a relatively young cohort. Statistical shape modelling demonstrated a clear relationship between hip shape and OA.