Osteophyte size and location on hip DXA scans are associated with hip pain: findings from a cross sectional study in UK Biobank

Objective It remains unclear how the different features of radiographic hip osteoarthritis (rHOA) contribute to hip pain. We examined the relationship between rHOA, including its individual components, and hip pain using a novel dual-energy x-ray absorptiometry (DXA)-based method. Methods Hip DXAs were obtained from UK Biobank. An automated method was developed to obtain minimum joint space width (mJSW) from points placed around the femoral head and acetabulum. Osteophyte areas at the lateral acetabulum, superior and inferior femoral head were derived manually. Semi-quantitative measures of osteophytes and joint space narrowing (JSN) were combined to provide a measure of rHOA. Logistic regression was used to examine the relationships between these variables and hip pain, obtained via questionnaires. Results 6,807 hip DXAs were examined. rHOA was present in 353 [5.2%] individuals and was associated with hip pain [OR 2.07 (95% CI 1.54-2.80)] and hospital diagnosed OA [5.73 (2.89-11.36)]. Total osteophyte area and mJSW were associated with hip pain [1.29 (1.21-1.36), 0.84 (0.77-0.92) respectively] in unadjusted models. After mutually adjusting and adding demographic covariates, total osteophyte area continued to have strong evidence of association with hip pain [1.31 (1.23-1.39)] but mJSW did not [0.95 (0.87-1.04)]. Acetabular, superior and inferior femoral osteophyte areas were all independently associated with hip pain [1.19 (1.13-1.26), 1.22 (1.15-1.29), 1.21 (1.14-1.28) respectively]. Conclusion The relationship between DXA-derived rHOA and prevalent hip pain is explained by osteophyte area rather than mJSW. Osteophytes at different locations showed important, potentially independent, associations with hip pain, possibly reflecting the contribution of distinct biomechanical pathways.


3
The relationship between DXA-derived rHOA and prevalent hip pain is explained by osteophyte area rather than mJSW. Osteophytes at different locations showed important, potentially independent, associations with hip pain, possibly reflecting the contribution of distinct biomechanical pathways.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
KL classification of rHOA has been shown to have a poor sensitivity and predictive value for symptoms (9). That said, severity of radiographic changes is associated with likelihood of symptoms and total hip replacement, a proxy for end-stage disease (10, 11). Previous studies have also examined the relationship between individual features of rHOA and hip pain, for example JSW was found to be only weakly associated with symptomatic measures of HOA (12). Another study examined the relationship between individual semi-quantitively graded components of rHOA and hip pain in women, observing that femoral head osteophytes were related to hip pain more strongly than JSN (10). A recent small study found that inferior medial femoral head osteophytes seen on computed tomography (CT) scans were associated with hip pain more strongly than other (superolateral, intra-articular, anterior and posterior) osteophytes, indicating that the relationship between osteophytes and hip pain may differ according to osteophyte location (13). With improving technology, it is now possible to measure features of rHOA in greater detail, for example measuring osteophyte size quantitively although this has not previously been applied to large population-based studies (14)(15)(16). By studying individual features of rHOA in greater detail this may help to better understand their . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 26, 2021. ; https://doi.org/10.1101/2021.04.26.21255905 doi: medRxiv preprint contribution to the development of hip pain, providing a basis for more accurate diagnostic/prognostic imaging biomarkers, and greater understanding of the biomechanical pathways underpinning OA development.
To date, large epidemiological studies of rHOA have almost exclusively been based on radiographs using well recognised atlases (17). In contrast, dual-energy X-ray absorptiometry (DXA) hip scans, widely used to evaluate patients for osteoporosis, and obtained in many large cohort studies, have previously had insufficient resolution to evaluate features related to osteoarthritis such as osteophytes (6). However, a new generation of DXA machines is now available with resolution comparable with that of radiographs, which have been validated for KL grading (18). This opens up the possibility of using cohort studies, in which large numbers of individuals have undergone newer generation hip DXA scans, to study rHOA; such as the UK Biobank (UKB) extended imaging study due to comprise 100,000 individuals (19, 20).
Here, we aimed to evaluate the feasibility of this approach, by deriving a measure of rHOA in a subset of 7000 hip DXA scans from UKB and, relating this to previously diagnosed HOA and hip pain. Further, we examined the relationship between hip pain and the different elements of rHOA in this substantial sample, and hip pain, including the contribution of osteophyte size and location.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) internal rotation using a standardised protocol (22). This study is based on a random subsample of 7000 individuals, selected from the overall sample of 13,496 individuals with DXA scans available at the time (February 2020). The first 20% of the subsample were selected randomly from those with a self-reported diagnosis of OA (the question did not ask at which joints) with the aim of increasing the number of pathological scans for our automated model training as part of a wider research programme. The remainder of the sample (80%) was selected randomly, throughout randomisation was achieved using a random number generator whilst we ensured the sexes were split equally.
Across all UKB participants 8.6% have self-reported a diagnosis of OA. All demographic information was taken from questionnaires completed on the same day as the DXA scan.
Ethnicity was self-reported, and individuals were categorised into white, Asian, black, mixed-. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 26, 2021. ; https://doi.org/10.1101/2021.04.26.21255905 doi: medRxiv preprint heritage, Chinese and other. The participants were asked via electronic questionnaire; "Have you had hip pains for more than 3 months?" They could answer "yes", "no", "don't know", "prefer not to say" or leave the answer blank, for this study only those who answered "yes" were categorised to have hip pain and the rest were not. Of note the hip pain question was not side specific. Hospital episode statistics linked with UKB were reviewed for ICD-9 & -10 codes related to HOA and if any were present then the individual was categorised to have hospital diagnosed HOA, as a binary variable.

DXA and osteophyte mark up
The left hip DXA was examined from each participant, 85 outline points were placed around the outline of the superior acetabulum, femoral head and metaphysis, lesser and greater trochanters by an automated Random Forest-based machine-learning algorithm before being reviewed and corrected where necessary by 4 manual annotators (23). 18 key points were anatomically guided, and the remaining points were equally spaced between these (Supplementary Figure S1).
A DXA-based atlas was created by BF, FS and MW (see acknowledgements) describing osteophytes at the lateral acetabulum, superolateral femoral head and inferomedial femoral head, based on the OARSI radiographic atlas (17). Femoral head osteophytes are referred to as superior and inferior femoral head osteophytes for simplicity. Two annotators (BF & FS) examined all the images to mark-up osteophytes, using a custom tool (The University of Manchester) to mark each osteophyte area and move the outline points inside of the osteophyte margin ( Figure 1). All osteophytes and adjoining points were agreed between these two annotators. The area of each osteophyte in millimetres squared (mm 2 ) was then derived for each image to be used as a continuous variable describing osteophyte size. The osteophytes . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 26, 2021. ; https://doi.org/10.1101/2021.04.26.21255905 doi: medRxiv preprint from the first 1930 DXAs were semi-quantitatively graded (grade 1-3) based on the aforementioned DXA-based atlas. Receiver operating characteristic curves (ROC) were used to define a threshold using osteophyte area scores for grade ³1 and grade ³2 osteophytes at each location to automate semi-quantitative grading of the remaining images (the presence of a grade 1 osteophyte was set at a threshold of osteophyte area ³1mm 2 at all locations, area under the curve (AUC) 1; acetabular grade ³2 osteophyte: threshold ³10mm 2 , AUC 0.96; superior femoral grade ³2 osteophyte: threshold ³17mm 2 , AUC 0.98; inferior femoral grade ³2 osteophyte: threshold ³19mm 2 , AUC 1). It was necessary to combine manually graded 2 and 3 osteophytes due to low numbers of grade 3 osteophytes (grade 3 osteophytes by location: acetabular n = 11, superior femoral head n = 6, inferior femoral head n = 4).

Joint Space Width
An automated method for measuring the width of the superior joint space, which is well demarcated on DXA (Figure 1), was subsequently developed. A custom Python script calculated mJSW between the acetabulum (points 78-84) and superior femoral head (points 22-31) as follows: A segment is created by drawing a straight line between two neighbouring points, for example, two points on the acetabulum. Then the shortest distance is calculated between this line and an opposing point, in this example on the femoral head. The automated method repeats this process for all segments and points selected, and the shortest distance representing mJSW (in mm) is saved. Additionally, the first 1930 DXAs were semiquantitatively graded for JSN, blinded to mJSW, using a DXA-based JSN atlas created by BF, FS & MW, based on the OARSI atlas (17). Height-adjusted ROC curves were used to define thresholds for JSN automatically on the remaining images, as these thresholds were found to be more accurate at defining JSN than from mJSW alone, giving AUC 0.92 for JSN grade ³1 and 0.97 for grade ³2. Grades 2 & 3 were merged due to the low numbers of grade 3 JSN (n=9).
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 26, 2021. ; https://doi.org/10.1101/2021.04.26.21255905 doi: medRxiv preprint Radiographic hip osteoarthritis rHOA was defined as grade ≥1 JSN combined with a grade ≥1 osteophyte(s), as this was felt to be most equivalent to Kellgren-Lawrence and Croft definitions based on JSN combined with a definite osteophyte(s) (3,4). Subchondral sclerosis and cysts were not examined as part of this study due to their relative infrequency (5). A more stringent definition of rHOA termed grade ≥2 rHOA, was defined as grade ≥2 osteophyte(s) combined with grade ≥ 2 JSN.

Statistical analysis
The demographic data are given as a mean and range for continuous variables and binary variables are given as counts and frequency. The initial analyses investigated categorical measures of rHOA, osteophytes, JSN and hip pain using logistic regression with results presented as odds ratios (OR) with 95% confidence intervals (CI). Later analyses examined continuous measures of osteophyte area and mJSW against hip pain again using logistic regression. Use of directed acyclic graphs informed the a priori selection of covariates, which included age, sex, height, weight and ethnicity to be added into an adjusted model. Logistic regression was also used to examine the independent relationships between rHOA features and hip pain through mutually adjusted models. Graphical representations of logistic regression models were created by deriving the probability of hip pain from the regression model at specific intervals of osteophyte area or mJSW, and plotting these. We refer to this as the likelihood of hip pain rather than probability to avoid confusion with P-values. All statistical analysis was performed using Stata version 15 (StataCorp, College Station, TX, USA).
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Descriptives: Population characteristics
Of the initial sample of 7000 participants with a left hip DXA, 193 were excluded (72 had a significant artefact, 39 were missing the greater trochanter, 32 were missing the lesser trochanter, 29 were missing part of the femoral head or femur, 3 were missing part of the ilium or acetabulum, 16 were poor quality, and 2 individuals withdrew consent for the study). This left a total of 6,807 individuals (mean age 62.7 years old, standard deviation (SD) 7.5 years) with left hip DXAs available for analysis ( In terms of continuous measures of osteophytes in those individuals with osteophytes, mean total area of all osteophytes present was 25 mm 2 with a range from 2 mm 2 to 268 mm 2 . Mean . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)  Table S2).

Osteophytes and joint space width (CATEGORICAL measures) versus hip pain
The presence of a grade ≥1 osteophyte at any site was associated with hip pain [OR 1.64 (1.35-2.01)] in unadjusted analyses, which were unaffected by adjustment as above (  Table S2). Sex-stratified results showed similar associations between features of rHOA and hip pain in males and females (Supplementary Tables S3 & S4). . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 26, 2021. inferior femoral osteophyte area OR 1.21 (1.14-1.28) (unadjusted analyses)] (Figure 3). When In a large (n = 6,807) cross-sectional study of both men and women, we have developed and applied a method for performing detailed phenotyping of rHOA based on high resolution DXA scans. As expected, those with rHOA as defined by DXA were associated with a higher prevalence of self-reported and hospital-diagnosed OA. We then went on to explore the relationship between rHOA and its individual features, and prevalent hip pain. We found that DXA-derived rHOA is associated with prevalent hip pain and that this association is predominately driven by the presence of osteophytes, rather than joint space narrowing.
Subsequently, we examined the relationship between osteophytes and hip pain based on quantitative evaluations of osteophyte size and osteophyte location. We found a positive relationship between osteophyte area and the likelihood of hip pain, such that the latter exceeded 50% when total osteophyte area reached 150 mm 2 , implying florid osteophytes are most reliably associated with hip pain. In addition, we found that osteophytes at all three sites examined, namely acetabular, superior femoral and inferior femoral, all showed potentially independent relationships with hip pain, consistent with roles in partially-independent biomechanical pathways. Inferior femoral osteophytes showed the strongest association with hip pain, and acetabular osteophytes the weakest.
Previous studies have shown that rHOA is poorly predictive of hip pain but these have focused on semi-quantitative composite measures of rHOA which may have limited accuracy in the assessment of joint pathology (9). Semi-quantitative measures of rHOA generally group together different osteophyte locations and sizes and use broad definitions of JSN, which may partly explain the weak associations observed with symptoms at both hip and knee joints (9,(24)(25)(26). We observed similar findings in our analysis, as even though individuals who had either DXA-derived rHOA or a single osteophyte (grade ≥1) were at an elevated risk of hip pain, it . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 26, 2021. ; https://doi.org/10.1101/2021.04.26.21255905 doi: medRxiv preprint was still the case that the majority of them did not have any hip pain (84% and 88% respectively). We are not aware of any previous studies to have examined clinical outcomes in relation to quantitative measures of hip osteophyte size as presented here. However there have been two previous studies analysing the relationship between osteophyte location and hip pain, with which our results are consistent. One previous study (n = 5,839) found that femoral osteophytes have a greater association with hip pain compared to acetabular osteophytes in women (10). A small CT-based study (n = 29) found that inferior osteophytes had a stronger association with hip pain compared with anterior, posterior and intra-articular osteophytes (13).
Osteophytes are a key component of OA although little is known about if or how they might induce pain, with many patients who have osteophytes not suffering from pain (27). Kijima et al. suggest that inferior femoral head osteophytes are a proxy for hip instability which might be causing hip pain through impingement of the femoral head and acetabulum (13). It is known that osteophytes are a poor prognostic sign for arthroscopic interventions for hip pain potentially due to a stabilising effect they have on a joint which is lost if they are removed (24, 28). Others have shown that osteophytes contain sensory fibres suggesting pain could be derived from the osteophyte itself (29,30), although arthroscopic removal of osteophytes is ineffective in the treatment of knee pain and no longer recommended (31,32). In addition, pain might be associated with osteophytes due to periostitis or inflammation which leads to their development rather than the osteophyte itself causing pain (33).
Our analysis, showing independent relationships between osteophytes at different sites and hip pain suggests location-specific mediators are a possibility, such as a role of distinct biomechanical pathways. Along similar lines, associations between hip morphology and rHOA and risk of hip replacement are presumed to be mediated through aberrant biomechanical . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 26, 2021. ; https://doi.org/10.1101/2021.04.26.21255905 doi: medRxiv preprint pathways (6,34,35). How such variations in morphology are related to specific constituents of rHOA remains unclear. Studies from high bone mass individuals show a global predisposition to osteophyte formation (hypertrophic OA), suggesting a strong genetic influence on osteophyte formation (5,36), which might point against specific local biomechanical factors in the development of osteophytes. On the other hand, it could still be the case that osteophytes lead to pain through local mechanisms as suggested by the independent relationships seen in this study. Understanding if and how different osteophytes contribute to pain is of clear clinical interest and requires further investigation.
Superior mJSW was associated with hip pain in our unadjusted model, but the relationship attenuated after adjustment for total osteophyte area and demographic covariates. These findings are consistent with a previous systematic review which only found weak associations between JSW and hip pain (12). A previous study on incident knee OA in a high bone mass population found that change in Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) pain score over time was attenuated to a greater extent by adjustment for osteophyte score, compared with joint space narrowing (36), further suggesting that osteophytes are the main contributing factor to the relationship between rHOA and joint pain.
To the extent that JSW contributes a limited amount to the evolution of hip pain in rHOA, this would seemingly undermine its use as an endpoint in clinical trials of disease modifying osteoarthritis drugs (DMOAD) (37).
A major strength of this study was the use of a novel method for characterising different components of rHOA on DXA scans, developed as part of our investigation This enabled us to examine relationships between detailed measures of rHOA and hip pain in a large sample of participants from UKB. Although there are limited data available on the validity of using hip . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 26, 2021. ; https://doi.org/10.1101/2021.04.26.21255905 doi: medRxiv preprint DXA scans to ascertain rHOA, the measures we obtained showed expected relationships with hospital-diagnosed and self-reported OA. Whilst DXA scan images appear suitable for deriving characteristics such as osteophytes and superior joint space width, including the potential for automation, they have several inherent limitations in evaluating rHOA. A potential limitation in the use of DXA scans to measure joint space width is that scans are obtained with the patient supine, rather than weight bearing as is the norm for radiographs (38). However, a previous study found little difference in JSW between weight bearing and non-weight bearing hip radiographs (39). Limitations in DXA imaging prevented us from evaluating other radiographic features associated with rHOA, such as subchondral sclerosis and cysts which were difficult to visualise. In addition, in contrast to the superior joint space, we were unable to visualise or evaluate the medial or inferior joint space as is often possible on x-rays.
The limitations of this study include, the observational and cross-sectional study-design which makes it not a suitable basis for drawing causal conclusions. In particular we can only comment on relationships with prevalent rather than incident hip pain. The hip pain information is limited in that it is not side-specific, although it does cover a prolonged duration (³3 months) which makes it pertinent to HOA (33). Further, this study used a weighted sample to include a greater proportion of individuals with self-reported OA which means we cannot use this data to comment on the prevalence of rHOA in UKB.
To conclude, we have developed and applied a method for large scale phenotyping of rHOA on DXA scans in UKB. The measures of rHOA obtained showed expected relationships with clinical outcomes such as hip pain. Focusing on individual semi-quantitatively graded features, JSN and osteophytes at different sites, these showed associations with hip pain. On examining these relationships in more detail, based on quantitative measures derived for osteophyte area . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 26, 2021. ; https://doi.org/10.1101/2021.04.26.21255905 doi: medRxiv preprint 18 and mJSW, we found that mJSW had no independent association with hip pain, in contrast to osteophytes which showed potentially independent relationships at all three sites. Further studies are justified to characterise site-specific biomechanical alterations that result in or from the formation of osteophytes, to further understand if and how these changes might be causally related to symptoms of pain in HOA, and to explore whether their correction might provide a means of slowing OA progression.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 26, 2021. ; https://doi.org/10.1101/2021.04.26.21255905 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Top left graph shows the unadjusted likelihood of hip pain by total osteophyte area. Top right graph shows the unadjusted likelihood of hip pain by mJSW, the x-axis is reversed. Bottom left graph shows likelihood of hip pain by total osteophyte area, adjusted for mJSW, age, sex, height, weight and ethnicity. Bottom right graph shows likelihood of hip pain by mJSW, adjusted for total osteophyte area, age, sex, height, weight and ethnicity.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Top right graph shows the unadjusted likelihood of hip pain by inferior femoral osteophyte area. The corresponding graphs below represent the respective models adjusted for area of osteophytes at the other sites, age, sex, height, weight and ethnicity.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 26, 2021. ; https://doi.org/10.1101/2021.04.26.21255905 doi: medRxiv preprint Table 1. Demographics of the sample studied with grade ≥1 abnormalities included.

Males
Females Combined