We recruited participants from the community through advertisements in local clubs and the print and radio media in metropolitan Melbourne, Australia, between May 2005 and July 2008. Inclusion criteria were age 50 years or more, average knee pain on walking more than 3 on an 11 point scale (0=no pain; 10=worst pain possible) at telephone screening, pain located over the medial knee compartment, evidence of osteophytes in the medial compartment or medial joint space narrowing on an x ray film,14
and radiological knee alignment of 185 degrees or less (corresponding to a mechanical axis angle of ≤182 degrees and indicating neutral to varus (bow leg) knee alignment on an x ray film of the whole leg).15
All participants provided written informed consent.
Exclusion criteria were questionable or advanced radiographic knee osteoarthritis (Kellgren and Lawrence grades 1 and 4),16
predominant patellofemoral joint symptoms on clinical examination (location of pain, pain provoking activities, tenderness on palpation, and pain during mobilisation of the patellar),17
knee surgery or intra-articular corticosteroid injection within six months, current or past (within four weeks) use of oral corticosteroids, systemic arthritic conditions, history of knee arthroplasty or osteotomy, other musculoskeletal or neurological condition affecting leg function, disease of the ankle or foot precluding the use of insoles, use of foot orthotics within the past six months, usual footwear unable to accommodate insoles, contraindications to magnetic resonance imaging, planning to start other treatment for knee osteoarthritis, and regular use of a gait aid.
We carried out a double blind randomised controlled trial over 12 months, the methodology of which has been described previously.18
Potential participants underwent telephone screening followed by standardised semiflexed standing posteroanterior knee radiology, to assess the severity of knee osteoarthritis and knee joint alignment. A physiotherapist or medical practitioner and a podiatrist then carried out a screening clinical examination. Participants were stratified by disease severity (Kellgren and Lawrence grades 2 and 316
) and sex and randomly allocated in permuted blocks of 6 to 12 to either the lateral wedge insole or the control insole group. An independent investigator used a computer program to generate the randomisation sequence a priori. Allocation was sealed in opaque and consecutively numbered envelopes held centrally. Envelopes were opened sequentially by an independent person. Participants were informed that two types of insoles were being compared but the insoles and study hypotheses were not described.
Participants wore the insoles bilaterally in their own shoes every day. They were provided with two pairs of insoles, which were replaced every four months. The lateral wedge (5 degrees) insoles were made of high density ethyl vinyl acetate (similar to the midsole in a running shoe) and were wedged along the lateral border of the foot (Foot Function, New Zealand, see web extra). We chose a 5 degree wedge because greater wedging is less likely to be tolerated by the wearer6
and is difficult to accommodate within a normal shoe. The control insoles were made of easily compressible low density ethyl vinyl acetate but with no wedging (Foot Function, New Zealand).
A blinded examiner assessed the participants at baseline and 12 months. In participants with bilateral knee osteoarthritis the most symptomatic eligible knee was assessed. Baseline demographic information was collected and participants rated their expectation of a beneficial effect with insole treatment on an ordinal scale from 1 to 5 (0=no effect at all, 5=complete recovery), with higher scores indicating higher expectations.
The primary symptomatic measure was overall average knee pain (past week) using an 11 point numerical rating scale (0=no pain, 10=worst pain possible).19
This has well accepted clinimetric properties and is widely used and recommended for clinical trials on knee osteoarthritis.19 20 21 22
Secondary symptomatic measures included pain on walking (measured on the 11 point scale), pain, stiffness, and physical function subscales of the Western Ontario and McMaster Universities osteoarthritis index,23
assessment of quality of life instrument,24
and patient perceived global change in pain and in physical function (compared with baseline) measured on a 5 point ordinal scale and dichotomised into improvement (slightly better and much better) and no improvement (much worse, slightly worse, and no change). We also measured levels of physical activity using two methods: the physical activity scale for the elderly questionnaire, with higher scores indicating greater physical activity,25
and the average number of steps taken per day, as measured by a pedometer (KH-005; Omron Healthcare, Japan), worn for one week on two occasions. Participants recorded use of and discomfort with insoles daily in a log book (returned on a monthly basis) and using an 11 point numerical rating scale at follow-up. Adverse events and cointerventions were recorded in the log book and by open probe questioning at follow-up.
The primary structural outcome measure was the volume of cartilage in the medial tibial compartment on magnetic resonance imaging. Images of the knee in the sagittal plane were obtained on a T1 weighted whole body unit as previously described.26
We defined the volume of the medial tibial cartilage plate by manually drawing disarticulation contours around the cartilage boundary on each section. Data were resampled by bilinear and cubic interpolation (area of 312 and 312 μm and 1.5 mm thickness, continuous sections) for final three dimensional rendering. We determined the volume of medial tibial cartilage plate by summing the pertinent voxels within the resultant binary volume. Two trained blinded observers independently determined the measurements. The coefficient of variation for the cartilage volume measure was 3.4%.27
To control medial tibial cartilage volume for bone size, we determined the cross sectional area of the medial tibial plateau by creating an isotropic volume from the input image, which was reformatted in the axial plane, using the software program Osiris. The area was directly measured from this axial image as described previously.28
By subtracting the follow-up volume from baseline volume and dividing by time between scans we determined the annual change in tibial cartilage volume. This was also expressed as a percentage of initial volume.
Secondary structural outcomes included cartilage defects, graded with a classification system on a 0 to 4 scale, as described previously29 30
for medial tibial and femoral compartments: grade 0, normal cartilage; grade 1, focal blistering and intracartilaginous low signal intensity area with an intact surface and bottom; grade 2, irregularities on the surface or bottom and loss of thickness of less than 50%; grade 3, deep ulceration with loss of thickness of more than 50%; and grade 4, full thickness cartilage wear with exposure of subchondral bone. The measurement was carried out by one trained observer, who measured all images in duplicate on separate occasions. Progression of medial tibiofemoral cartilage defects was determined if the cartilage defect score increased by at least 1 from baseline to follow-up in the medial tibial or medial femoral compartment.
Bone marrow lesions were defined as areas of increased signal intensity adjacent to subcortical bone present in either the medial distal femur or the proximal tibia, and their size was graded on a 0 to 2 scale as described previously from coronal T2 fat saturated images31
: grade 0, absence of a lesion; grade 1, lesion encompassed up to 25% of the width of the tibial or femoral cartilage being examined from coronal images; and grade 2, lesion encompassed more than 25% of the width of the tibial or femoral cartilage being examined from coronal images. We also recorded the number of slices the bone marrow lesions encompassed. To provide the medial tibiofemoral bone marrow lesion we multiplied the bone marrow lesion grade (0-2) by the number of slices for the medial femoral and medial tibial compartment separately, which was then summed. Progression of bone marrow lesions was defined if there was an increase in bone marrow lesion score of 1 or more over the period—that is, follow-up tibiofemoral bone marrow lesion score minus the baseline lesion score of 1 or more. This scoring system is a valid measure of bone marrow lesions as it has been shown to be sensitive to change and to detect clinically important outcomes.32 33
Magnetic resonance imaging machines
We used two different magnetic resonance imaging machines; initially a Philips machine (Eindhoven, Netherlands) followed by a GE machine (Signa Advantage HiSpeed GE Medical Systems, Milwaukee, WI), owing to decommissioning of the Philips machine. In total, 117 (68%) participants were scanned on the same machine at baseline and follow-up (102 had both scans on the Philips machine and 15 had both scans on the GE machine), whereas 54 (32%) participants were scanned on the Philips machine at baseline and the GE machine at follow-up.
A validity study confirmed no systematic difference with change of machines. Fifteen participants underwent scans of one knee using both machines. The mean medial tibial cartilage volume as measured on the Philips and GE machines was 1706.3 (SD 361.2) mm3 and 1719.3 (SD 394.4) mm3, respectively (P>0.05). The intraclass correlation coefficient from a one way analysis of variance was 0.98 (95% confidence interval 0.95 to 0.99), showing excellent absolute agreement between measures. A Bland-Altman plot of the difference (Philips-GE machine) versus average measurements showed the mean to be −13 mm3, with 95% limits of agreement of −152 to 126 mm3. The standard error of measurement was calculated as 49.3 mm3. To compare between machine repeatability with within machine repeatability, 12 of these 15 participants underwent a second measurement on the Philips machine, yielding a standard error of measurement of 33.3 mm3, with 95% limits of agreement of −75.6 to 106.5 mm3. These limits indicate comparable but slightly less reproducibility between machines than within machines.
Overall, 126 participants were required to detect a minimal clinically important difference of 1.5 for change in pain between groups,34
assuming a standard deviation of 3 (based on previous data35 36
), with 80% power at a 5% significance level. We have shown that the mean rate of tibial cartilage loss is 5.3% (SD 5.2%) per year in knee osteoarthritis.27
Data suggest that a clinically beneficial outcome with lateral wedge insoles would be to reduce the rate of cartilage loss to less than 3% per annum, as this is associated with a reduced risk for arthroplasty within four years.37
Thus we required a minimum of 160 participants to detect a difference of 2.3% between groups, with 80% power at a 5% significance level. We increased the sample size to 200 to account for dropouts and to allow for at least 80% power for both the primary outcomes.
A blinded statistician carried out the analysis, which was by intention to treat. All analyses were done using Stata (Version 11), and we considered P values of less than 0.05 to be significant. For continuous outcome measures we used linear regression modelling adjusted for baseline values of the outcome to compare differences in mean change (baseline minus follow-up) between groups. Results are presented as estimated differences with 95% confidence intervals. The analysis of change in medial cartilage volume was repeated with further adjustment for age, sex, body mass index, medial tibial bone size, and magnetic resonance imaging machine. We summarised the total medial bone marrow lesion scores as the median change from baseline (interquartile range) and compared scores between groups using the difference in median change. Using log binomial regression we compared global change in pain and in physical function and progression of cartilage defects between groups. Results are presented as relative risks with 95% confidence intervals. We used bootstrap confidence intervals based on 5000 replications to compare changes in the size of bone marrow lesions.
To account for missing data (31/400 pain measures and 29/400 medial tibial cartilage volume measures, 7% of total baseline and follow-up dataset for each measure) we carried out a sensitivity analysis using single mean imputation of missing baseline measures38
and multiple imputation of missing follow-up measures, assuming data are missing at random and follow a multivariate39
normal distribution. As results were unchanged we present complete case analyses.
We undertook a second sensitivity analysis to estimate the between group difference that would occur if all participants adhered completely to their allocated treatment. For these analyses adherence was measured by the number of days the insoles were worn as reported in the log books. A t
test was used to compare adherence between the groups. For each of the two primary outcomes we used a two stage least squares instrumental variables approach.40
This involved a regression model of the outcome measure adjusted for the baseline value and adherence, and a second regression model of adherence was adjusted for randomised group. The two regression models were fitted simultaneously and we estimated the effect of the lateral wedged insoles under full adherence.