Study design and setting
The Tasmanian Older Adult Cohort (TASOAC) study is a prospective, population-based cohort study, which aimed to identify factors associated with development and progression of OA and osteoporosis in older adults. Men and women aged 50-80 years in 2002 were selected from the electoral roll, which is the most complete population listing for adult Australians, in Southern Tasmania (population 229,000) using sex-stratified random sampling (response rate 57%). Participants were excluded if they lived in an aged care facility, or had contraindications to magnetic resonance imaging. The Southern Tasmanian Health and Medical Human Research Ethics Committee approved the study, and we obtained written informed consent from all participants.
Baseline data (Phase 1) were collected from February 2002 to September 2004 in 1099 participants. Follow up data (Phases 2, 3 and 4) were collected on average 2.6 (n=875), and 5 years (n=769) later. Participants who had a hip replacement prior to Phase 1 were excluded from analyses in this manuscript (n=13).
Outcome: Total Hip Replacement
Incidence of primary THR was determined by data linkage to the Australian Orthopaedic Association National Joint Replacement Registry (AOANJRR), and includes data from both public and private hospitals. Data validation against State and Territory Health Department data is done using a sequential multi-level matching process.(20) Matched data were then obtained; this included the date, side of joint replacement, primary or revision joint replacement and the reason for the procedure (e.g., OA, fracture of neck of femur, osteonecrosis, inflammatory arthritis, tumour). In this study, we only considered primary THRs that were due to OA. We include data from the AOANJRR between 1 March 2002 and 21 September 2016. These data excluded participants who died, collected from the Tasmanian Death Registry and who left Australia, which was collected from TASOAC questionnaires.
Body mass index (BMI) was calculated (weight (in kilograms)/height (in metres)2) using weight measured to the nearest 0.1 kg (with shoes, socks, bulky clothing and headwear removed) using a single pair of calibrated electronic scales (Seca Delta Model 707), and height measured to the nearest 0.1 cm (with shoes and socks removed) using a stadiometer.
Self-reported hip pain over the past 30 days was assessed by questionnaire at Phase 2 and 3 using the Western Ontario and McMaster Universities Osteoarthritis (WOMAC) index.(8, 21) Briefly, the WOMAC pain scale has five items, each rated on a 10-point numeric rating scale from 0 (no pain) to 9 (most severe pain). Each pain item was summed to create a total pain score (0–45).
Hip radiographs and assessment of hip radiographic OA (ROA) and cam morphology.
Anteroposterior radiographs of the pelvis were obtained at Phase 1, with the individual standing with both feet internally rotated by 10 degrees. Radiographs were read by two trained readers using the OARSI (Osteoarthritis Research Society International) grading system.(22) Radiographic features of joint space narrowing (JSN) (axial and superior) and osteophytes (superior, acetabular and femoral) of both hips were graded separately on a 4-point scale (range 0–3 where 0 is no disease and 3 is severe disease. Data from these four features were summed (range 0-12). Any score other than 0 for either JSN or osteophytes was regarded as evidence of radiographic hip OA. Thus, after combining the JSN and osteophytes scores, the presence of radiographic hip OA was defined as a total score of 1 or greater.
The α angle measures the extent to which the femoral head deviates from spherical and is used to quantify cam morphology. It is measured by first drawing the best fitting circle around the femoral head, and then a line through the centre of the neck and the centre of the head. From the centre of the femoral head, a second line is drawn to the point where the superior surface of the head-neck junction first departs from the circle. The angle between these two lines is the α angle. We defined cam morphology by using a previously published standardised cut off point of 60° either in one or both hips.(23) The α angle was calculated by drawing a circle of best fit based on the statistical shape modeling (SSM) points around the femoral head using custom code in MatLab (v 9.0). This method has good reliability as was shown previously with intraclass correlation coefficient (ICC) for inter-observer reliability of 0.73 and intra-observer reliability of 0.85-0.99.(9)
DXA Imaging and Statistical Shape Modelling (SSM)
Participants had dual-energy X-ray absorptiometry (DXA) images taken of the left hip, unless contra-indicated, using a Hologic Delphi densitometer (Hologic Inc., Waltham, MA, USA) as part of the Phase 1 assessment. Participants were excluded from DXA scanning if their weight exceeded 130 kg (n=3). Left hip images were used to assess bone mass; examined as areal BMD at neck of femur (g/cm2). This is calculated by dividing the bone mineral content (BMC) by the area measured. Precision was estimated to be 2% in vivo.
Statistical shape modelling (SSM) was used to describe hip shape variation within the study population. Briefly the proximal femur and acetabulum were modelled for each image using a template of 85 points placed on defined anatomical landmarks using the Active Shape Modelling toolkit (University of Manchester, UK).(24, 25) The images and points were transferred to the Shape software (University of Aberdeen, UK), where they were rotated and scaled using the Procrustes transform and then subjected to Principal Component Analysis to generate independent, orthogonal modes of variation. The modes of variation were then normalized to a mean of 0 and expressed as standard deviations from the mean. The modes of variation described decreasing amounts of variation within the model with the first 6 modes describing 68% of the total model variation. To test reproducibility of the measures, two observers (HGA and FRS) assessed joint shape on ten images randomly selected from the TASOAC dataset. Point‐to‐point variability (the distance between equivalent points placed by each observer) was calculated. The distribution was not normal and the median was 1.6 pixels, which is a small difference given the image dimensions for all images are 252 x 258 pixels.
Magnetic resonance imaging (MRI).
A subgroup (n=250) had MRI. The right hip was imaged in the sagittal plane during visits at phases 2 and 3 using a 1.5 Tesla GE Signa whole-body magnetic resonance scanner, as previously described.(8) Subchondral BMLs and effusion-synovitis were assessed on the short T1 inversion recovery (STIR)–weighted, fat saturation, 2-dimensional fast spin-echo sequence using OsiriX software (Mac version, University of Geneva, Geneva, Switzerland). BMLs were identified as areas of increased signal intensity adjacent to the subchondral bone on the femoral head and/or the acetabulum.(8) Intraobserver repeatability was assessed in 25 subjects (at both time points), with a 2-week gap between the measures. The intra-class correlation coefficient for hip BMLs was 0.98, similar to the reproducibility of our knee quantitative BML measure.(26) Hip effusion-synovitis was identified and assessed in STIR images from phases 2 and 3. The observer (HGA) manually selected the MRI slice with the largest effusion-synovitis and determined the maximum cross-sectional area (CSA) of the bright region by manually drawing contours around the outer edges, as previously described. Inter-rater reliability was excellent (0.84).(8) BMLs and effusions were dichotomised as present (CSA >0) or absent (CSA=0).
Differences between participants who did and did not have hip replacements were assessed using Students' t-tests and chi-squared tests.
Risk of THR in addition to the ‘base model’ (WOMAC hip pain score, and radiographic hip OA score) was assessed using mixed-effect Poisson regression, in which each potential risk factor was designated as a fixed effect and participant identification as a random effect. Models were run for each hip separately using the xt function, with side-specific WOMAC pain score and data from radiographs (ROA and alpha angle) used for risk of THR of each hip, while data from DXA (BMD and SSM) and MRI (BML, effusion) had data from one hip only (left hip for DXA, right hip for MRI) and was used to predict risk of THR in either hip. Standard errors were adjusted using the sandwich (robust) estimator of variance. We used WOMAC hip pain as continuous data (range 0-35), but collapsed radiographic hip OA scores into categories as effect sizes were similar within groups. The relationship between each of the risk factors and the incidence of THR during follow-up was assessed using Cox proportional hazards regression models. Hazard ratios (HRs) and 95% confidence intervals (CIs) were calculated. Model assumption was checked and confirmed using the proportional hazards test. We performed a sensitivity analysis, using a competing risk regression model to account for competing risks, which occurred within the study time frame (death, left Australia).
We used Stata 15.0 (StataCorp LP) for all statistical analyses. Statistical significance was defined as a p value ≤0.05 (two tailed).