Between December 2001 and December 2003, we recruited 194 participants aged 45-80 years with osteoarthritis of the hip or knee from five rural general practices in Mid Devon (see bmj.com
). Osteoarthritis was diagnosed by a consultant (orthopaedic surgeon or rheumatologist) or a general practitioner, and we sought confirmatory radiological evidence for participants who had none recorded in their general practice notes. Participants had to score 8-20 points on the Western Ontario and McMaster Universities osteoarthritis index (WOMAC A) on entry.13,14
We excluded people with a cardiac pacemaker, current magnetic bracelet, surgery to the index joint (excluding arthroscopy), or haemophilia and women who were pregnant or breast feeding.
Recruitment was by referral from doctors, advertising, or invitation after a search of practice records. Trial nurses arranged radiological confirmation of diagnosis if needed, and they collected data in surgery based clinics at 0, 4, and 12 weeks. Participants were given a full strength bracelet at the end of the trial.
Intervention and randomisation
The participants, trial nurse, and healthcare providers, were blinded to treatment allocation. Treatments consisted of identical looking bracelets containing three different components. The manufacturer's specifications were:
Group A—Standard neodymium magnets set in a steel backing cup, with the open side facing the ventral wrist, creating a fluctuating magnetic pattern across the bracelet (fig 1). The field strength at the wrist contact surface was 170-200 mTesla.
—Weak magnets with no backing plate. The field was strong enough to seem magnetic on testing (21-30 mTesla), but previous research suggests this is insufficient to be therapeutic.15
This was intended to provide an undetectable placebo.
Group C—Non-magnetic steel washers.
The National Physical Laboratory tested five bracelets of each type before the study, confirming the manufacturer's specification.
An independent researcher randomised participants in blocks of 15 (five of each bracelet type per block), using random numbers generated in Microsoft Excel. A decode sheet was sealed and locked away. A second researcher checked the procedure. On enrolment, participants were told that they would receive either an active or an inactive bracelet.
The predefined primary outcome measure was change in WOMAC A score after 12 weeks' follow up.13,14
Secondary outcomes were a visual analogue scale asking, “How bad was the pain from your arthritis in the last week when it was at its worst?” with verbal and numerical anchors from none (0) to worst imaginable (100) 16
; WOMAC B and C scores, measuring leg stiffness and functioning13,14
; the number of days participants had used analgesics in the past week; and perceived monetary value of the bracelet.
We assessed compliance with wearing the bracelet at 4 and 12 weeks using a visual analogue scale. Blinding was assessed at 12 weeks by asking whether participants thought they had an active bracelet and the reason for such belief.
The estimated effect size was based on a 20% differential reduction in WOMAC A score, which was considered commensurate with effect sizes in studies of analgesics and osteoarthritis.13,17
A sample size of 52 in each of the groups would have 80% power to detect a difference in one way analysis of variance of change scores, assuming mean changes of 3, 1.5, and 1 and a common standard deviation of 3.4. Assuming 15% dropout, we planned to recruit 64 subjects to each group.18
We checked the suitability of these numbers for an analysis of variance across the three groups by using a range of estimated small average changes for the weak magnet group.
The analysis was specified in advance of the study as follows. Last value carried forward was used to impute missing values for subsequent visits. The blinded statistician conducted analysis of variance on all three groups using SPSS version 11.5, with change in WOMAC A score at 12 weeks as the response. The robustness of the results was checked with analysis of covariance on the WOMAC A score at 12 weeks with baseline WOMAC A as covariate, and checking sensitivity to baseline imbalances. Dunnett's test was then used to compare the means for the dummy and weak magnet group separately with the mean for the standard magnet group. The protocol specified that the primary comparison was dummy versus standard magnets, the other comparison being secondary, unless a high degree of unblinding was observed. Models were checked by examination of residuals and sensitivity to imputed values.
Subsequent analyses were unblinded. We used general linear models on all subjects to explore the association between outcomes and magnetic strength of individual bracelets. Similar analyses were then carried out, where appropriate, for WOMAC B and C and the global pain score.