Published guidelines have incorporated topical NSAIDs as recommended treatment for OA of the knee [4–6]. However, there has been controversy surrounding the adequacy of data supporting their benefit beyond 2 weeks [2, 3, 8, 18]. Moreover, the studies identified in these meta-analyses generally did not conform to current standards for OA trial design [11, 19]. In contrast, the present trial utilized standardized radiological and clinical entry criteria and measured efficacy with validated outcome measures. Baseline pain score was substantial; mean (SD) score was 12.9 (3.2) out of a maximum of 20, indicating a flare of pain following withdrawal of prior therapy. Analysis of all of the primary and secondary measures demonstrated that treatment with this topical diclofenac solution relieved the symptoms of primary knee OA at 6 weeks in this study population. Two other recently published trials using this topical diclofenac solution showed it to be superior to vehicle control and/or placebo; a 4-week, non-flare trial of 248 participants  and a 12-week, flare trial of 326 participants . As with most NSAID trials, the subject population in this study was selected by the inclusion criterion of a flare of pain, which demonstrates the potential to respond to NSAID/analgesic. In clinical practice, an individual not taking an analgesic may have considered previous NSAID therapy ineffective, in which case s/he would not be expected to respond to topical diclofenac. However, where an individual is intolerant to oral NSAID, one may consider topical diclofenac as a treatment option.
Comparison of efficacy results from independent trials with various treatments is facilitated by the introduction of benchmark determinants that are mathematically derived from the experimental raw data, such as effect size  for improvement of a continuous variable (e.g. "How much did the patient's pain improve, relative to placebo?"). We calculated an effect size (95% CI) of 0.41 (0.14 to 0.68) for pain relief, 0.44 (0.16 to 0.71) for improved physical function and 0.34–0.47 for improved measures of PGA, stiffness and pain on walking (Table 2). In contrast, Lin et al.  calculated a pooled effect size for pain relief of 0.04 (essentially no effect) in 3 placebo-controlled topical NSAID trials of 4 weeks duration. A meta-analysis of 23 oral NSAID trials for OA knee, lasting 2–13 weeks, reported a pooled effect size of 0.32 for pain reduction and 0.29 for improving physical function . Another meta-analysis of 14 OA trials found a pooled effect size of 0.37 for pain reduction with oral NSAIDs and 0.44 for coxibs . Zhang et al. , using data from 2 oral NSAID studies of 6–12 weeks duration, calculated a pooled effect size for OA pain reduction of 0.34.
Efficacy of a treatment is being expressed increasingly as a dichotomous result, e.g. "Did the patient's pain improve by 50%; yes or no?". We derived the response rate for each dichotomous variable from our raw data, and demonstrated the superiority of topical diclofenac over vehicle control for 50% reduction in pain, achieving a good or very good final PGA response, and 'response' by OMERACT-OARSI criteria (Table 3). The benchmark determinant for comparing dichotomous efficacy results of various treatments is the number-needed-to-treat (NNT) . We calculated a NNT between 4 and 6, depending upon the variable (Table 3). In their meta-analysis of topical NSAIDs, Mason et al.  cited 5 placebo-controlled trials of short duration for OA knee pain – 8 days (1 trial), 14 days (3 trials), and 28 days (1 trial). Their definition of clinical success, representing approximately a 50% reduction of pain, was estimated using patient or physician global assessment as the outcome measure (4 trials and 1 trial, respectively). They calculated a NNT of 5.3.
Few oral NSAID studies have reported dichotomous data. Osiri et al.  reported a NNT for pain improvement of 4.4 with etodolac and 3.8 with tenoxicam. Defining improvement as an increase of at least 2 grades (on a 0–5 scale) in the patient's global rating of arthritis, Edwards et al.  reported a NNT of 11–13 for valdecoxib treatment of OA.
The OMERACT-OARSI initiative used a consensus approach to derive dichotomous 'responder' criteria . Through their vast meta-analysis of suitable trials, the authors found that for trials of oral NSAIDs vs. placebo the responder rates were 65.4% and 45.9% respectively. Responder rates of 60–65% have been reported for 13-week treatment of OA with celecoxib and lumiracoxib, with placebo responder rates of 49–53% [28, 29]. The OMERACT-OARSI initiative did not look at topicals but we applied its criteria to this study and found a responder rate for topical diclofenac of 65.7% with a placebo responder rate of 49.5%, similar to their oral NSAID data.
A caveat in the application of the mathematical benchmarks, effect size and NNT, is the influence of trial design, outcome measures and patient population on the apparent magnitude of response to a given treatment. Because the trials with topical diclofenac were designed according to the OARSI guidelines, like most recent NSAID and cyclooxygenase-2 (COX-2) inhibitor studies, such comparison of results is reasonable . Although the data observed for topical diclofenac in this trial are comparable to other NSAID trials, a direct head-to-head comparison trial is required to prove equivalency of two treatments. A previously published 12-week comparative trial of 622 participants with OA knee confirmed the clinical equivalence between topical diclofenac solution and oral diclofenac .
Safety analysis revealed no serious clinical adverse effects and only minor application-site skin reactions, mostly skin dryness, following treatment with topical diclofenac. While dimethyl sulphoxide in the carrier acts as a penetrant , it also dissolves normal surface oils and leaves the skin dry. Common skin lubricants may prevent most application site reactions and any related discontinued therapy, but such products were not permitted in this trial in order to detect the maximum potential side effect profile of the study solutions. The low dropout rate due to skin reactions (5/107 [4.7%] for topical diclofenac) suggests patient acceptance of the overall topical treatment regime.
The use of a checklist to prompt the patient about possible adverse events likely yielded a high estimate of the true incidence of gastrointestinal adverse reactions caused by topical diclofenac. The report of abdominal pain and dyspepsia each in 3.7% of patients is consistent with what was seen in other published trials of this topical diclofenac [20, 21] and much lower than commonly experienced with oral NSAIDs or COX-2s . Those other trials included results of laboratory testing and found minor abnormality of liver enzymes in 2–5%, creatinine in 1% and haemoglobin in 2% of patients, significantly lower than with oral diclofenac . This safety profile can be predicted from the low systemic availability of topically applied diclofenac. Although the patient applies a daily dose (40 drops, 4 times a day) of 86 mg of diclofenac to the knee, the blood level is only 12 ng/mL . The level reported after oral administration of 50 mg Voltaren® is 1500 ng/mL . Similar improved safety with topical NSAIDs has been reported previously .