Background: The exclusion of patients from the analysis of randomized trials may result in biased estimates of treatment effects and increase heterogeneity. Objectives: We determined whether exclusions from the analysis are associated with biased estimates of treatment effects in osteoarthritis research. Methods: We searched for meta-analyses of at least two randomized controlled trials comparing an intervention with placebo or no intervention in patients with osteoarthritis. We calculated effect sizes (ES) from between-group differences in means of pain intensity divided by the pooled standard deviation. Within each meta-analysis, we calculated the difference in ES between trials with adequate analysis (inclusion of all patients in the analysis according to the intention-to-treat (ITT) principle) and trials with inadequate or unclear approaches and calculated a weighted average across all meta-analyses using a randomeffects model. Then, we determined the impact of restricting randomeffects meta-analyses to trials with adequate analysis according to ITT. Results: Fourteen meta-analyses including 172 trials and 39,298 patients were identified. Forty-eight trials had performed an ITT (28%); in 113 trials (66%) were exclusions from the analysis; and in 11 trials, the approach was unclear (6%). Trials with inadequate or unclear analyses tended to be more beneficial than trials with adequate analyses (difference in ES, -0.09, 95%-CI, -0.23 to 0.05). Attrition bias was evident in metaanalyses with a large degree of between-trial heterogeneity (difference in ES, -0.22, 95%-CI -0.43 to -0.02), but not in meta-analyses with small to moderate heterogeneity (0.05, 95%-CI -0.06 to 0.15, p-value for interaction, 0.042). Treatment effects became less beneficial in ten metaanalyses and more beneficial in four after restriction to trials with ITT (range 0.49 ES less to 0.21 ES more beneficial). Results became less heterogeneous in 11 meta-analyses and more heterogeneous in 3. In five meta-analyses, results lost statistical significance at the 5% level. Conclusions: In osteoarthritis research, there is evidence for attrition bias, particularly in meta-analyses with a high degree of heterogeneity between trials.