A novel semi-automated classifier of hip osteoarthritis on DXA images shows expected relationships with clinical outcomes in UK Biobank

Benjamin G Faber* (Corresponding Author), Raja Ebsim, Fiona Saunders, Monika Frysz, Claudia Lindner, Jenny S. Gregory, Richard M Aspden, Nicholas C Harvey, George Davey Smith, Timothy Cootes, Jonathan H. Tobias

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

9 Citations (Scopus)
14 Downloads (Pure)


Conventional scoring methods for radiographic hip osteoarthritis (rHOA) are subjective and show inconsistent relationships with clinical outcomes. To provide a more objective rHOA scoring method, we aimed to develop a semi-automated classifier based on dual-energy X-ray absorptiometry (DXA) images, and confirm its relationships with clinical outcomes.

Hip DXAs in UK Biobank (UKB) were marked up for osteophyte area from which acetabular, superior and inferior femoral head osteophyte grades were derived. Joint space narrowing (JSN) grade was obtained automatically from minimum joint space width (mJSW) measures.
Clinical outcomes related to rHOA comprised hip pain, hospital diagnosed OA (HES OA) and total hip replacement (THR). Logistic regression and Cox proportional hazard modelling were used to examine associations between overall rHOA grade (0-4; derived from combining osteophyte and JSN grades), and the clinical outcomes.

40,340 individuals were included in the study (mean age 63.7), of whom 81.2% had no evidence of rHOA, while 18.8% had grade ≥1 rHOA. Grade ≥1 osteophytes at each location and JSN were associated with hip pain, HES OA and THR. Associations with all three clinical outcomes increased progressively according to rHOA grade, with grade 4 rHOA and THR showing the strongest association [57.70 (38.08-87.44)].

Our novel semi-automated tool provides a useful means for classifying rHOA on hip DXAs, given its strong and progressive relationships with clinical outcomes. These findings suggest DXA scanning can be used to classify rHOA in large DXA-based cohort studies supporting further research, with the future potential for population-based screening.
Original languageEnglish
Pages (from-to)3586–3595
Number of pages10
Issue number9
Early online date17 Dec 2021
Publication statusPublished - 1 Sept 2022

Bibliographical note

The authors would like to thank the MRU patient and public involvement group at the University of Bristol for their input into planning our research and Dr Martin Williams, Consultant Musculoskeletal Radiologist North Bristol NHS Trust, who provided substantial training and expertise for this study. This work has been conducted using the UK Biobank resource (application number 17295).

Funding and grant award information:

BGF is supported by a Medical Research Council (MRC) clinical research training fellowship (MR/S021280/1). RE, MF, FS are supported, and this work is funded by a Wellcome Trust collaborative award (reference number 209233). CL was funded by the MRC, UK (MR/S00405X/1) as well as a Sir Henry Dale Fellowship jointly funded by the Wellcome Trust and the Royal Society (223267/Z/21/Z). NCH acknowledges support from the MRC and NIHR Southampton Biomedical Research Centre, University of Southampton and University Hospital Southampton. BGF, MF, GDS, JHT work in the MRC Integrative Epidemiology Unit
at the University of Bristol, which is supported by the MRC (MC_UU_00011/1). This research was funded in whole, or in part, by the Wellcome Trust [Grant number 223267/Z/21/Z]. For the purpose of open access, the author has applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission.

Data Availability Statement

Supplementary data are available at Rheumatology online.


  • OA
  • dual-energy x ray absorptiometry
  • total joint replacement
  • hip pain


Dive into the research topics of 'A novel semi-automated classifier of hip osteoarthritis on DXA images shows expected relationships with clinical outcomes in UK Biobank'. Together they form a unique fingerprint.

Cite this