Constitutional morphological features and risk of hip osteoarthritis: a case–control study using standard radiographs

Objectives To evaluate the risk of association with hip osteoarthritis (OA) of 14 morphological features measured on standard antero-posterior pelvis radiographs. Methods A case–control study of 566 symptomatic unilateral hip OA cases and 1108 controls without hip OA, using the Genetics of OA and Lifestyle database. Unaffected hips of cases were assumed to reflect pre-OA morphology of the contralateral affected hip. ORs with 95% CI adjusted for confounding factors were calculated using logistic regression. Hierarchical clustering on principal component method was used to identify clusters of morphological features. Proportional risk contribution (PRC) of these morphological features in the context of other risk factors of hip OA was estimated using receiver operating characteristic analysis. Results All morphological features showed right–left symmetry in controls. Each feature was associated with hip OA after adjusting for age, gender and body mass index. Increased sourcil angle had the strongest association (OR: 6.93, 95% CI 5.16 to 9.32). Three clusters were identified. The PRC varied between individual features, as well as between clusters. It was 35% (95% CI 31% to 40%) for all 14 morphological features, compared to 21% (95% CI 19% to 24%) for all other well-established risk factors. Conclusions Constitutional morphological variation strongly associates with hip OA development and may explain much of its heritability. Relevant morphological measures can be assessed readily on standard radiographs to help predict risk of hip OA. Prospective studies are required to provide further support for causality.


INTRODUCTION
Osteoarthritis (OA) is a common complex disorder with multiple interactions between genetic, constitutional and environmental risk factors. 1 Strong genetic contribution to hip OA is supported by 60% heritability in a classic twin study in women, 2 and a fivefold increased prevalence of radiographic hip OA in siblings of people with hip OA requiring total hip replacement. 3 Morphological variation of the hip and pelvis is also emphasised as a potentially important constitutional risk factor for hip OA. [4][5][6][7][8][9] It is recognised that rare monogenic abnormalities of bone shape such as severe acetabular dysplasia can cause young-onset hip OA. 10 However, it is possible that more subtle variations in joint and bone morphology, resulting from multiple common gene polymorphisms, may impose biomechanical insult and partially explain genetic predisposition in common hip OA. This is supported by studies showing that mild hip dysplasia, 5 non-spherical femoral head ('pistol grip' deformity) 4 11 and high or low neck shaft angle (NSA) 4 10 are relatively common and associate with increased risk of hip OA. Studies using statistical shape modelling also report associations between variations in proximal femoral shape and risk of hip OA. [12][13][14] It is also noteworthy that three genetic associations with large joint OA confirmed with genome-wide significance (GDF5, 15 16 FRZB 17 18 and MCF2L 19 ) are involved in early skeletal growth. Furthermore, hip OA frequently occurs without OA at other sites, 20 21 supporting the importance of local factors in its development.

Key messages
What is already known about this subject? ► Several constitutional variants of hip joint shape associate with increased risk of hip osteoarthritis (OA). However, whether these relate to each other, and the overall contribution of morphological variants to risk of hip OA are unknown.
What does this study add? ► Fourteen morphological features of the hip and pelvis, ten of which had not been studied before, were shown to associate with hip OA after adjusting for age, gender and body mass index. The strongest association was with more vertical sourcil angle (SA). Three clusters of features were identified, and the proportional risk contribution to hip OA was 35% for the combined variants, compared with 21% for other recognised risk factors combined.
How might this impact on clinical practice or future developments? ► Although prospective studies are required to provide further support for causality, morphological variation is a strong risk factor for hip OA and may partially explain its heritability. SA measured on standard radiographs may be used as a single surrogate marker to assess morphological risk of hip OA.

Osteoarthritis
Previously we used the Genetics of OA and Lifestyle (GOAL) database to demonstrate that mild acetabular dysplasia (assessed by acetabular depth (AD), centre edge angle (CEA)), 5 nonspherical femoral head shape (assessed by femoral head to femoral neck ratio (FHNR)) 4 and both high and low NSA 4 associate with hip OA. Because morphological features can be secondary to hip OA, we undertook measures of the unaffected hip of people with unilateral hip OA under the assumption that this reflects the constitutional morphology of the affected hip prior to hip OA development. This assumption was supported by right-left symmetry in normal controls without hip OA. 4 5 However, these and other morphological features may relate to, or interact with each other to increase risk of hip OA. In addition, the proportional risk contribution (PRC) of local morphological features in the context of overall risk of developing hip OA is unknown. The objectives of this study were to: (1) examine 10 additional morphological features of the hip and pelvis that can be measured readily on plain radiographs, for right-left symmetry and age variation; and (2) measure their risk contributions, both individually and in combination with others, and in the context of other recognised risk factors for hip OA. The new features we assessed were: femoral head diameter (FHD) 22 ; femoral neck length (FNL) 23 and femoral neck width (FNW) 6 23 24 ; femoral head offset (FHO) 25 ; femoral outer shaft diameter (OSD) and inner shaft diameter (ISD); sourcil angle (SA) 26 27 ; mid-centre distance (MCD); and pelvic width (PW) and pelvic height (PH).

Cases and controls
All participants (566 unilateral hip OA cases and 1108 non-OA controls) were selected from the Nottingham GOAL database, which was a hospital-based case-control study to investigate genetic associations and gene-environmental interaction in people with knee or hip OA. Fifty-nine per cent of unilateral hip OA individuals had right hip OA and 41% had left hip OA. The laterality of unaffected hips was matched in the same ratio to controls. All participants were Caucasian and aged between 45 and 80 years. Details of recruitment, exclusion criteria, questionnaire and clinical and radiographic assessments of participants have been published previously. 4 5 28 29 Radiographic assessment of hips A standard protocol was used to obtain antero-posterior (AP) non-weight-bearing radiographs of the pelvis with the participants supine and feet internally rotated 10°. 4 All radiographs were scored previously by a single observer for radiographic features of hip OA, which included minimum joint space width (JSW). 4 5 Radiographic hip OA was defined as JSW ≤2.5 mm. 30 Those participants with unilateral hip OA, that is no symptoms and normal radiographic appearance (JSW >2.5 mm and no other OA features) in the contralateral hip, were included for morphological assessment of the unaffected hip. The asymptomatic control group (all with JSW >2.5 mm and no radiographic features of OA in either hip) underwent morphological assessment of both hips. These controls also had no symptoms or radiographic evidence (Kellgren Lawrence grade <2) of knee OA. The anatomical indices that were measured are described in table 1 and figure 1. Data for four (AD, CEA, FHNR and NSA) of these features had previously been scored by a single observer with good reproducibility, 4 5 and were reused in the current study. The 10 other new features were measured both in normal controls and participants with unilateral hip OA by a different single trained reader (HA) using HIPAX software Table 1 Descriptions of the morphological landmarks and measurements of the hip joint and pelvic bones examined in this study

Morphological measurements Descriptions
Centre of femoral head The equatorial centre of the head was determined by fitting it is geometry within a concentric circle on the Perspex template of the Lequesne arthrometer. 48 Femoral shaft axis Two points in the centre of the femoral shaft were measured to be equidistant from the medial and lateral borders, one at the lowest part of the femoral shaft and the other one below the lesser trochanter. The line connecting these two points described the axis of the femoral shaft.

Femoral neck axis
The midpoint of the shortest segment of the femoral neck was measured to be equidistant from the superior and inferior borders. A line passing through the centre of the femoral head and the midpoint of the femoral neck described this axis.

Acetabular depth
The distance between the deepest point of the acetabular roof to a line drawn between the edge of the articular surface of the acetabulum and the upper corner of the symphysis pubis on the same side. 7 Centre edge angle The angle between the line from the femoral head centre to the lateral aspect of the acetabulum, and a vertical line drawn from the centre of the femoral head at right angles to the line joining the two femoral head centres. 49 Femoral head to femoral neck ratio The ratio of femoral head diameter divided by femoral neck width. 4 Neck shaft angle The angle between the femoral shaft axis and femoral neck axis.
Femoral head diameter The maximum diameter was described by drawing a line through the central point of the femoral head and at a right angle to the femoral neck axis line.
Femoral neck width This was the minimum femoral neck diameter, determined by drawing a line at the narrowest point of the femoral neck and at a right angle to the femoral neck axis.

Femoral neck length
The distance from the defined centre of the femoral head to the intersection of the femoral neck axis and femoral shaft axis.
Outer shaft diameter of the femur This was defined as the full diameter of the femoral shaft, which was made at the level of half of the femoral head diameter, distal to the lesser trochanter.
Inner shaft diameter of the femur This was measured at the level of half of the head diameter distal to the lesser trochanter. This measurement represents the thickness of the medullary canal of the femoral bone.
Mid-centre distance The distance from the centre of the femoral head to the midline of the pelvic X-ray and perpendicular to this midline point.

Sourcil angle
The angle formed between a line extending from the medial to the lateral edge of the sourcil and a horizontal line. 27 Femoral head offset The distance from the centre of the femoral head to the axis of the femoral shaft in a right angle.

Pelvic width
The widest diameter of the pelvic bone on the radiograph.

Pelvic height
The greatest height of the pelvic bone at the centre of the pelvis on the radiograph.
(Hipax, Vorstetten, Germany). As in our previous studies, this reader was blind to participant identifiers, demographic and clinical information.

Patient and public involvement
There was no patient and public involvement for this study.

Statistical analysis
The intraobserver reproducibility of measuring the 10 new morphological features was assessed using a random sample of 30 pelvis radiographs on three occasions (beginning, middle and end of study). Interobserver reproducibility was assessed by measuring 30 pelvis radiographs for two previously assessed measures (NSA and FHNR) and comparing results to those of the previous readers. 4 Intraclass correlation coefficient (ICC) was used to determine reproducibility. Symmetry of the morphological measurements was determined using paired t-test and minimal detectable change (MDC) in the control group. 31 The difference between groups was determined using t-test (continuous data) or χ 2 test (categorical data). Correlations between the measurements were examined using Pearson correlation coefficient. The dose-response relationship of individual morphological measurements in tertiles and risk of OA was examined. Logistic regression model was used to calculate OR and 95% CIs adjusting for confounding factors such as age, gender and body mass index (BMI).
Cluster analysis was undertaken using the hierarchical clustering on principal component (HCPC) method to examine clusters of morphological measurements. HCPC was done using 'factoextra' and 'FactoMineR' packages in R. 32 Distribution of clusters was plotted in the factor map.
The PRC was estimated using receiver operating characteristic (ROC) curves where areas under the curve (AUC) were proportionalised according to risk factors. 33 First, we built the full risk model with all risk factors available in an ROC curve (AUC f ). The full risk model included established risk factors such as age, gender, weight, height, BMI, calcaneal bone mineral density (BMD), finger nodes in at least two rays of each hand, type 3 pattern of index to ring finger (2D:4D) ratio, history of hip injury, manual occupation, 4 29 34 and all 14 morphological features (ie, both the newly assessed and previously measured features in GOAL). Second, we removed the risk factor(s) of interest to examine the contribution of the risk factor(s) removed through the reduction of the ROC curve, that is, the partial AUC (AUC p ). Third, we calculated the PRC using the following formula: PRC=(AUC f −AUC p )/(AUC f −0.5), where 0.5 is the AUC under the diagonal line of the ROC curve indicating no discrimination at all by all included risk factors. 33 Data were analysed using STATA V.15 and R V.3.5. A significance level of p<0.05 was set for all analyses.

Characteristics of the study participants
Characteristics of study participants are shown in table 2. Of 1674 participants, 566 had unilateral hip OA (cases) and 1108 had no hip OA (normal controls). Gender, height and manual occupation were similar between groups, but cases were older and had higher weight, BMI and BMD than controls. Prevalence of nodal hand OA, type 3 pattern 2D:4D finger ratio and frequency of self-reported hip injury were also higher in the OA group.

Repeatability of measurements
In addition to the excellent reproducibility of the four features reported previously, 4 5 the 10 new features had good intraobserver agreement across the three time points, the ICCs ranging from 0.84 to 0.97 for all features (p<0.05). There was also good agreement between the two readers for NSA and FHNR with ICCs of 0.87 and 0.85, respectively (p<0.05).

Symmetry and age association in non-OA controls
In the non-OA control group the paired t-test showed that mean differences between left and right sides for most measurements were not statistically significant except for AD, CEA, ISD and MCD. However, the magnitude of these differences was less than MDC 90 (online supplemental table S1). While age was associated with most morphological features on the left and right, it was not associated with symmetry, that is, the difference between left and right (online supplemental table S2).

Proportional risk contribution
The AUC for the full model including all risk factors was 0.81 (95% CI 0.79 to 0.83), of which 34

DISCUSSION
This is the first large study to assess 14 hip and pelvis morphological features, individually and in composite, and their contribution to the risk of hip OA. The right-left symmetry of all measures demonstrated in the normal controls supports the assumption that the unaffected hip of unilateral hip OA cases represents the pre-OA morphology of the affected hip. 4 5 Although age associated with some morphological features, it was not associated with the symmetry, that is, the difference between left and right.   Osteoarthritis associated with increased risk of hip OA independent of age, gender and BMI, with larger SA being the strongest risk factor; (2) two patterns of associations were observed-dose response and U-shaped curve response (both higher and lower values associating with increased risk); (3) three clusters were identified (figure 2); and (4) the total contribution of the 14 morphological features to risk of hip OA was greater (35%) than the sum of other recognised risk factors (21%).
Our findings of small FHD, wide FNW and short FNL as risk factors for hip OA concur with the previous studies. 6 11 14 22-24 Biomechanically many of these features have a plausible aetiological mechanism. For example, small FHD and/or wide FNW may both encourage 'cam type' impingement of the proximal femur on the acetabulum, 25 as does a non-spherical femoral head. 35 Furthermore, a small femoral head has a smaller surface area for load transmission, thus the force per unit area may be higher and cause increased joint tissue stress. On the other hand, a wide FNW may encourage 'pincer-type' impingement of the femoral head-neck junction against the acetabular rim. 25 The explanation for smaller measurements of both OSD and ISD could relate to the inverse relationship between osteoporosis and OA. 36 Low FHO and wide MCD necessitates a greater abductor muscles force to maintain body balance 37 and the resultant greater stress on the hip may predispose to OA. The association of AD, CEA, FHNR and NSA with hip OA were reported and discussed in our previous studies. 4 5 Importantly, our findings indicated that of the 14 features studied, increased SA was the strongest individual risk factor for hip OA and showed the highest PRC. Departure of the acetabular sourcil orientation from the horizontal plane will negatively affect the equilibrium of forces across the hip joint, 26 and with bigger SA the femoral head is less covered by the acetabulum, which is consistent with the negative correlation between SA and CEA, so the unit force per surface area is increased. In previous studies, SA related more than other indices with development of OA 38 39 and it is considered a more precise measure for mild dysplasia than CEA. 40 Therefore overall, more vertical SA is a major morphological risk factor and may be used as a single surrogate marker in clinical practice to assess morphological risk of hip OA.
The 14 morphological features were assigned into three clusters. Cluster analysis may uncover relationships between measures. For example, in a case with high NSA (coxa valga), the increased inclination of the weight-bearing surface of the acetabulum (assessed by SA) can increase the compressive forces on the joint and lower the threshold for the onset of OA. 41 The coexistence of less acetabular coverage and shorter femoral neck were reported in one hip shape mode (HSM) derived by statistical shape modelling which positively associated with incident hip OA. 14 But in another HSM, more coverage of the femoral head and wider PW were found to associate with OA, 14 which is inconsistent with our findings. The higher proportion of women and the different definition of PW in that study 14 should be considered when comparing the results with ours. However, the possible explanation for the associations observed for PW and PH are open to speculation. Further prospective study for causality is still required.
The risk contribution of the 14 morphological features (PRC=35%, 95% CI 31% to 40%) was significantly larger than other established risk factors including age, gender, BMI, history of hip injury, physical occupation, nodal OA and 2D:4D finger ratio (PRC=21%, 95% CI 19% to 24%). This suggests that local morphological risk factors may contribute more than systemic factors to development of hip OA. The results align with the literature for incidence and progression of hip OA 42 43 and may be explained by shared single nucleotide polymorphisms between OA and hip shape. 44 45 There are several caveats to this study. First, this was a crosssectional case-control study. Whether these morphological features cause hip OA requires a prospective population-based study. Although we used the unaffected hips of people with unilateral hip OA to determine constitutional pre-OA shape, it is possible that the morphology in the unaffected hip had adapted to altered gait pattern and abnormal loading caused by hip OA on the other side, 46 in accord with Wolff 's law which states that bones adapt their mass and shape in response to loading. 47 In addition, the apparently normal hips could have undergone bone remodelling due to early OA before other features such as cartilage loss were evident. 23 Furthermore, we did not account for presence of symptoms or structural OA in other lower limb joints (knees, ankles, feet) of cases which may have affected biomechanical stress on the unaffected hip. Also radiographic assessment is less sensitive to early OA changes than other imaging modalities, such as MRI. We also found that some morphological features changed with age in the control group. Although symmetry was unaffected by age, we cannot be certain that the current features measured in unaffected hips of cases would fully reflect the pre-OA morphology on the affected side before it developed OA many years ago. Second, although we observed symmetry of morphological features in the nondisease control group, this does not exclude the possibility of asymmetry in the cases before they developed unilateral hip OA, or the presence of additional unidentified risk factors on the affected side, or protective factors on the unaffected side. This again requires a prospective cohort study to confirm whether the predisease morphological features are truly symmetrical between the left and right sides, and to determine how many people with the features of interest subsequently go on to develop bilateral hip OA. Third, the GOAL database includes only Caucasian participants so the generalisability of the findings is limited and requires study in other populations. Fourth, we undertook measurements on a single two-dimensional standard AP pelvis radiograph without other views. Although this is conventional and readily applicable to large-scale population studies, it has major limitations for identifying true morphological variations in three-dimensions. A further caveat is that measurement of morphological features was not undertaken blind of hip OA status, since pelvic images were saved on software (HIPAX) that prevents image cropping. Furthermore, despite the use of a standardised protocol, variations in positioning may have affected some assessments, for example, due to anteversion or rotation secondary to pain or deformity in the affected hip.
In conclusion, we have confirmed 14 morphological features that associate with increased risk of hip OA. The risk contribution of these features is more than that of other conventional risk factors combined. SA is the strongest risk factor and could be used as a single surrogate measure of morphological risk in large epidemiological studies or in clinical settings. Future prospective studies are required to provide further support for causality between these features and OA.