As correctly pointed out by Dr. Goldberg and colleagues, our meta-analysis was hampered by the limited quality of included trials and the heterogeneity of their results. Therefore, the interpretation of the meta-analysis of all trials was difficult and the investigation of potential sources of heterogeneity mandatory (1). The restriction of the analysis to large-scale, high-quality trials covering 40% of patients was not based on subjective judgment, but on results from stratified analyses and corresponding interaction tests. All explored factors were prespecified before initiating our systematic review. The 3 factors associated with treatment effects—concealment of allocation, intention-to-treat analysis, and sample size—are known to be associated with bias (2, 3). The cutoff of 200 patients used to explore the influence of trial size was specified in a grant proposal submitted to and funded by the Swiss National Research Programme 53 on musculoskeletal health (http://www.nfp53.ch/e_module.cfm?kati=6), which was initiated in 2004 before 4 of the 5 large-scale trials became available. The only trial with more than 200 patients that was already available in 2004 (4) showed a large effect of chondroitin, which was incompatible with the effects found in any of the subsequent large-scale trials. It lacked adequate concealment of allocation, did not have a placebo control group, and failed to perform an intention-to-treat analysis.

It is joint pain that leads patients with osteoarthritis to seek medical help. We believe, therefore, that pain reduction should be the primary objective in osteoarthritis trials (5). Contrary to Dr. Goldberg and colleagues' notion, we addressed radiographic joint space narrowing in a secondary analysis and found clinically irrelevant effects of chondroitin.

We agree with Dr. Pelletier that the STOPP (Study on Osteoarthritis Progression Prevention) trial (6) and the study by Michel and colleagues (7) met the primary end point in terms of joint space narrowing. However, the difference in mean changes of joint space narrowing in favor of chondroitin of 0.14 and 0.12 mm are evanescent, corresponding to an effect size of merely 0.1 SD. We disagree with Dr. Pelletier that Clegg and colleagues (8) provided robust evidence for an association between severity of symptoms and treatment response in patients receiving a combination of chondroitin and glucosamine. The results Dr. Pelletier refers to come from 1 of a multitude of subgroup analyses and might be explained by chance alone. As addressed in our discussion, there might be a role of chondroitin for patients with low-grade osteoarthritis, but this needs to be evaluated in large-scale trials performed independently from manufacturers.

We agree with Mr. Levin that pharmaceutical quality is important in evaluations of food supplements. However, the 3 large-scale, high-quality trials, which showed no effect, had either selected the investigational product on the basis of analyses of the capsules for purity, potency, and quality (8) or used an established preparation manufactured by the trial sponsor, which is approved and monitored by the Swiss drug approval agency (6, 7).