This study was conducted from November 1999 to August 2000, at 17 medical centres across central Canada, following approval by a central ethics review board (Integrated Research Incorporated, Ethics Review Committee, Montreal, QC). Participants were recruited from the physician's private practice or the surrounding community. At the screening visit, after providing written, informed consent, each participant underwent a screening interview and was eligible to proceed to washout if all inclusion criteria and no exclusion criteria were met.

Inclusion criteria specified men and non-pregnant women, age 40–85 years, with primary OA of at least one knee, and a flare of pain after withdrawal of prior therapy with either an oral NSAID or acetaminophen (used at least 3 days per week during the previous month). Primary OA was defined by deterioration and abrasion of articular cartilage (joint space narrowing) or formation of new bone (osteophytes) at the joint surface of the knee (medial tibio-femoral, lateral tibio-femoral or patello-femoral), demonstrated on a radiological examination carried out within the previous 3 months 9 . Pain was measured by the Western Ontario and McMaster Universities LK3.1 OA Index (WOMAC) 5-item pain subscale, each item scored on a 5-point Likert scale (none = 0; mild = 1; moderate = 2; severe = 3; extreme = 4) 10 . Pain was scored at the screening visit, following which prior therapy was withdrawn. The patient scored the pain again at the baseline visit. A flare was defined as an increase in total pain subscale score of at least 2 and at least 25%, with a baseline total pain score of at least 6 (out of a possible 20), and a score of ≥2 (out of a possible 4) on at least one of the 5 items in the WOMAC pain subscale.

Participants were excluded if they had secondary arthritis related to systemic inflammatory arthritis (including rheumatoid arthritis, psoriatic arthritis, post-infectious arthritis and metabolic arthritis, traumatic arthritis or surgical joint replacement); corticosteroid use: (a) oral corticosteroid within the previous 14 days, or (b) intramuscular corticosteroid within 30 days, or (c) intra-articular corticosteroid into the study knee within 90 days, or (d) intra-articular corticosteroid into any other joint within 30 days, or (e) topical corticosteroid at the site of application within 14 days; intra-articular viscosupplementation (e.g., Synvisc^®) into the study knee in the preceding 90 days; ongoing use of prohibited medication including NSAID, other oral analgesic, muscle relaxant, or low-dose antidepressant for any chronic pain management; ongoing use of glucosamine or chondroitin (unless used continuously for 90 days prior to study entry); sensitivity to diclofenac, acetylsalicylic acid (ASA) or any other NSAID, acetaminophen, dimethyl sulphoxide, propylene glycol, glycerine or ethanol; clinically-active renal, hepatic or peptic ulcer disease; history of alcohol or drug abuse; lactation; concomitant skin disease at the application site; current application for disability benefits on the basis of knee osteoarthritis; fibromyalgia; other painful or disabling condition affecting the knee; or participation in another investigational drug trial in the previous 30 days.

At the baseline visit, all patients that met the final entry criterion of a flare of pain were randomly assigned to receive one of two treatments: (a) topical diclofenac solution (Pennsaid^®; Dimethaid Research Inc.), consisting of 1.5% (w/w) diclofenac sodium in a patented carrier containing dimethyl sulphoxide (45.5%, w/w), propylene glycol, glycerine, ethanol and water, or (b) vehicle control solution, consisting of the complete carrier (including dimethyl sulphoxide, 45.5% w/w) but no diclofenac. Participants applied a dose of 40 drops of study solution (about 1.3 mL) to the affected knee 4 times daily for up to 6 weeks. The participant was instructed to apply 10 drops of solution to each side of the knee (front, back, medial and lateral) either dripped directly onto the knee or first into the hand, and then spread over the site without massage. Compliance was verified by weighing the solution bottles at each visit. If the other knee was painful at any time during the study, it was treated and evaluated for safety, but efficacy analysis was performed on only the study knee – the one with the greater baseline pain score (or the dominant knee if both had the same score). Consumption of acetaminophen (up to four 325-mg tablets per day) was permitted for residual knee or other body pain throughout the treatment period, but not during the washout period prior to baseline assessment or during the week prior to final assessment at week 6. ASA (≤ 325 mg/day) was permitted for cardiovascular prophylaxis.

The primary outcome measures were defined as the change from baseline to final assessment of the study knee in the 3 core continuous variables 11 pain and physical function, assessed using the WOMAC subscales, and patient global assessment (PGA). There was no intermediate assessment of efficacy. The WOMAC is a validated questionnaire 12 consisting of 24 questions (5 on pain, 17 on physical function and 2 on stiffness), each scored on a 5-point Likert scale (see Participants). The PGA question asked: "How has the osteoarthritis in your study joint been over the last 48 hours?" and was scored on a Likert scale (very good = 0; good = 1; fair = 2; poor = 3; very poor = 4). This question focuses on the treated site, unlike a PGA in an oral NSAID trial that can probe the non-signal joints. Secondary measure was change in stiffness. Ancillary measures defined a posteriori were pain on walking – the first question of the WOMAC pain subscale (referred to as 'use-related pain' 13 ) – and the following dichotomous variables: 50% improvement in pain 3 ; final PGA score of "good" or "very good" 3 ; and response based on OMERACT-OARSI responder criteria 14 (a responder is defined as a participant with ≥ 50% improvement in pain or function that was ≥ 20% of the scale, or ≥ 20% improvement in at least two of pain, function or PGA that was ≥ 10% of the scale).

Safety was assessed during all clinic visits (weeks 3 and 6) and telephone 'visits' (weeks 1 and 5). Safety variables included adverse events, application-site dermatological reactions and vital signs. Adverse events were identified using open-ended questions and a checklist covering common oral NSAID side effects. Dermatological assessment of the knee was based on a standard scale 15 and any abnormality was recorded as an adverse event. All adverse events were categorised according to Coding Symbols for Thesaurus of Adverse Reaction Terms (COSTART) 16 . Laboratory assessment was not done.

Based on a power of 80% and a Type I error rate of α = 0.05_2-tailed, a sample size of 80 participants per group was required to detect an estimated important difference of 2 between the treatment arms, in the change in WOMAC pain dimension score from baseline to final (with standard deviation of 4.5). A total sample size of 200 participants (100 per treatment group) was specified in the protocol, which allowed for a non-evaluable rate of up to 20%.

The study kits were prepared, labelled and numbered according to a computer-generated randomisation schedule created by an outside consultant using a randomly chosen block size of 4 or 6. They were shipped to the sites in multiples of complete blocks to ensure that a balanced number of participants was assigned to the two treatment arms within each site. As a participant qualified for study entry at the baseline visit, the investigator assigned him/her the next randomisation number in a sequential manner. The randomisation schedule was concealed from the investigators, their support staff, study participants and the sponsor's clinical research personnel, until final data lock. Except for the individual participant identification number on the label, the two study solutions were identical clear, colourless liquids packaged in opaque bottles.

Safety analyses were performed on all randomised participants who received at least one dose of study solution. There was no imputation of missing safety data. Efficacy analyses were performed on an intent-to-treat (ITT) group, defined as a subset of all randomised participants who met critical inclusion criteria (primary OA by history, an abnormal radiological study, and any degree of knee pain), as per ICH guidelines 17 . For any missing efficacy data in the ITT analysis, the last observation was carried forward. A per-protocol group was defined based on stricter adherence to study conduct, including requirement for a moderate flare of knee pain (see Participants) and treatment continuing for at least 40 days.

Baseline demographic and clinical variables were analysed by Chi-square or Student's t-Test. Adverse event incidence was analysed by Chi-square or Fisher's Exact Test. Continuous variables (WOMAC dimensions, PGA and pain on walking) were analysed by ANCOVA with baseline score as the covariate without adjustment for testing secondary/alternative objectives. The dichotomous variables were analysed by Chi-square test. All statistical tests were two-sided and were performed at the 0.05 level of significance.

Two hundred and sixteen participants were randomised to treatment with either topical diclofenac (n = 107) or vehicle control (n = 109). All participants received their allocated intervention. More participants in the topical diclofenac group (86 [80%]) completed the entire 6-week treatment period compared to the vehicle control group (70 [64%]; p = 0.008). Discontinuation rate due to an adverse event was similar in both groups. Dropout due to lack of effect was lower for topical diclofenac (8 [7.5%]) compared to vehicle control (18 [16.5%]; p = 0.041). No participant was lost to follow-up (Fig. 1).

No significant difference was found between treatment groups in baseline demographic and clinical characteristics (Table 1). The mean (SD) screening and baseline pain scores were 8.2 (2.7) and 13.0 (3.2) in the topical diclofenac group versus 8.3 (3.0) and 12.8 (3.1) in the vehicle control group (12.9 [3.2] overall). Most participants treated both knees, either from baseline or by the end of the trial.

Mean (SD) duration of treatment in the topical diclofenac group was 38.2 (9.9) days versus 34.4 (12.5) days in the vehicle control group (p = 0.013). Compliance with the dosing regime was 83.1 % and 84.5% for the topical diclofenac and vehicle control groups, respectively. No significant difference was noted in the mean (SD) consumption of rescue acetaminophen tablets per day between the topical diclofenac (0.9 [0.9]) and vehicle control groups (1.1 [1.0]; p = 0.079).

Four of 216 randomized participants were not included in the ITT analysis group because they violated major entry criteria: 2 participants lacked radiological confirmation of OA (no radiological examination for one participant and a normal examination for the other), and 2 participants had secondary OA (related to osteochondroma). Inclusion of these participants yielded the same results in a subsequent re-analysis (data not shown).

The major adverse effect reported was dry skin at the application site, occurring in 42/107 (39.3%) and 23/109 (21.1%; p = 0.004) of topical diclofenac and vehicle control participants, respectively (Table 4). A skin-related adverse event led to discontinuation of only 5 participants in the topical diclofenac group. All skin reactions resolved promptly upon withdrawal of treatment. Abdominal pain and dyspepsia each were reported in 4 [3.7%] participants in the topical diclofenac group compared to 1 [0.9%] participant in the vehicle control group, but this difference was not significant (p = 0.21).