Ethics approval
This study was conducted in accordance with the declaration of Helsinki and was approved by the Clinical Research Ethics Board of the University of British Columbia. All participants gave written informed consent at all three time points.
Data collection
Source data came from a longitudinal study conducted in Vancouver, Canada, [11] a population-based cohort of individuals aged 40 to 79 with knee pain “on most days of the month at any time in the past and any pain in the past 12 months.” Data collection has been previously described [12, 13]. The clinical examination was performed by an experienced rheumatologist (JC). We have previously reported in this cohort that, based on MRI cartilage damage and X-ray findings, 13% had no OA (KL < 2 and no cartilage damage), 49% had pre-radiographic OA (cartilage damage but KL < 2), and 38% had radiographic OA (KL ≥ 2) [13]. This cohort enrolled 255 individuals, stratified by age decade and sex in roughly equal group sizes to ensure adequate sample size across the age-sex spectrum [14]. Baseline visits occurred between 2002 and 2005. In addition to the baseline cycle, two follow-up cycles were undertaken, at stratum-sampling-weighted mean 3.3 (SD 0.6) and 7.5 (SD 0.6) years. The present study uses the baseline sample (N = 255) for the cross-sectional modelling, as well as the intersection of the first and second follow-up cycles (N = 108 × 2 = 216) for longitudinal modelling.
The study knee was the more painful knee at baseline. X-rays were obtained using a weight-bearing fixed-flexion posteroanterior view with the SynaFlexer (BioClinica Inc., Newark, CA, USA) positioning frame, and a skyline view in the supine position [15]. Radiographs were read blinded to clinical information by two independent readers for KL 0–4 grading [5]. Previous studies using these data have demonstrated good interrater reliability (ICC = 0.79) [12]. Differences in readings were adjudicated by consensus readings with both readers. MRIs were acquired on a GE 1.5 T magnet at a single centre using a transmitter–receiver extremity knee coil. The imaging protocol included four MRI sequences, as previously described [13, 14]. MRIs were scored by a board-certified musculoskeletal radiologist (AG) who was blinded to clinical, radiographic, and time sequence information. Cartilage was scored in 6 subregions: lateral and medial femur, lateral and medial tibia, patella and trochlear groove. The trochlear groove was delineated from the weight bearing surfaces of the femur by oblique lines, tangent to the anterior tips of the anterior horns of the medial and lateral menisci [14]. Cartilage was graded on a 0–4 semi-quantitative scale based on the following definitions, previously described by Disler et al.: [16] 0: normal, 1: abnormal signal without cartilage contour defect, 2: contour defect of < 50% cartilage thickness, 3: contour defect of 50–99% cartilage thickness, 4: 100% cartilage contour defect with subjacent bone signal abnormality. 0 and 1 were collapsed since 1 represents signal hyperintensity on T2-weighted images of indeterminate significance, hence the analysis variables ranged from 0–3. Osteophytes (defined as osteo-cartilaginous protrusions growing at the margins of osteoarthritic joints from a process that involves endochondral ossification) were scored using criteria described in Hunter et al [17]. Osteophytes (0: absent, 1: small, 2: moderate, 3: large) were scored in 8 regions: lateral and medial femur, lateral and medial tibia, and lateral, medial, superior and inferior patella. Meniscal damage was scored as: 0: normal, 1: intra-substance signal, 2: tear. 0 and 1 were collapsed, hence the analysis variables ranged from 0–1. Meniscal damage was scored in the following 6 regions: lateral anterior, lateral body, lateral posterior, medial anterior, medial body and medial posterior. Intra-rater reliability analyses were previously performed on the scoring of each surface within each feature. The ranges of intraclass correlation coefficients (ICCs) across regions were: cartilage 0.84–1.00, osteophytes 0.77–0.89, meniscus 0.60–0.83 [8].
MTF tenderness (ICC = 0.94) and LTF tenderness (ICC = 0.85) were assessed in examination by palpating the medial or lateral tibiofemoral joint line while the patient sits with legs hanging over the edge of the examination bed. PF grind (ICC = 0.94) was assessed with the subject lying on the examination bed with their legs extended, then asked to contract their quadriceps muscle while the examiner applies downward and inferior pressure on the patella. Pain was the positive signal in both tests (not grinding).
Statistical methods
To obtain population-representative results, a baseline sample weight was developed as the ratio of knee-pain population age-sex distribution over the baseline knee-pain sample distribution, and was used in the cross-sectional models. Cross-sectional (prevalence) models were weighted with the baseline sample weight. A sample weight was developed for the longitudinal sample as the baseline sample weight multiplied by the ratio of baseline sample proportion in a given age-sex cell over the longitudinal sample proportion in that cell. Longitudinal (incidence) models were weighted with the longitudinal sample weight.
For our first objective (modelling cross-sectional association), we used baseline data to fit age-sex-BMI adjusted logistic models predicting prevalent MTF and LTF knee joint tenderness as well as PF grind versus each relevant cartilage/osteophyte/meniscus (COM) predictor in separate models. Relevant COM predictors for MTF tenderness included medial femoral cartilage (MFC), medial tibial cartilage (MTC), medial femoral osteophytes (MFO), medial tibial osteophytes (MTO), medial anterior meniscus (MAM), medial meniscal body (MBM), and medial posterior meniscus (MPM). The relevant COM predictors for LTF tenderness included the lateral equivalents of those medial predictors. The relevant COM predictors for PF grind included patellar cartilage (PC), trochlear groove cartilage (TC), medial patellar osteophytes (MPO), lateral patellar osteophytes (LPO), superior patellar osteophytes (SPO) and inferior patellar osteophytes (IPO). In addition to the age-sex-BMI adjusted single COM predictor models, we also fit fully adjusted multivariable models including as predictors all relevant COM predictors together, plus age, sex and BMI. For our second objective (modelling 3-year incidence), we combined baseline to 3-year follow-up with 3-year to 7-year follow-up (two records per subject), and fit binary generalized estimating equations (GEE) models predicting 3-year incident MTF and LTF knee joint tenderness as well as 3-year incident PF grind versus the relevant COM predictors at “baseline” for each cycle (i.e., either actual baseline or 3 years depending on the cycle represented on a given record). Incidence models were also fit as age-sex-BMI adjusted single COM predictor models, as well as fully adjusted models including as predictors all relevant COM predictors together, plus age, sex and BMI. All longitudinal models were also adjusted for individual follow-up time between cycles. Model fit was assessed via the Hosmer and Lemeshow goodness of fit test [18]. The predictive utility of each model was assessed via the area under the receiver operating characteristic (ROC) curve (AUC).
Due to theoretical considerations (lack of a plausible biological pathway), in the primary models we did not regress PF grind versus medial or lateral tibiofemoral MRI scores. However, in a sensitivity analysis, we fit models predicting PF grind versus each of the medial and lateral predictor sets listed above.
Analyses were performed using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA).