Objective To evaluate the extent to which the current designs of clinical trials in knee osteoarthritis (OA) permit detection of a therapeutic effect of disease-modifying OA drugs (DMOADs) on the incidence of knee replacement, and to provide estimates of the required sample sizes. Methods We selected distinct subcohorts of the Osteoarthritis Initiative (OAI), based on available information on eligibility criteria for clinical knee OA trials (ClinicalTrials.gov) and additional subcohorts stratified for age, sex, and the severity of radiographic OA. The observed incidence of knee replacement in these OAI subcohorts was used to estimate the expected incidence of knee replacement in the control group of a clinical trial. Based on this estimate, the sample sizes required to detect hypothetical treatment effects on the incidence of knee replacement were calculated, assuming observation periods of 2, 5, or 7 years. Results The cumulative knee replacement incidence rates in the OAI subcohorts ranged from 0.9% to 12.9%. The corresponding sample sizes required to detect 50% improvement by the DMOAD, with a power of 80% and 95% confidence, were 5,459 and 362, respectively. Including only women with advanced age and radiographic OA increased the incidence of knee replacement and decreased the required sample size. Conclusion The sample sizes that are commonly used in clinical trials do not enable the effects of a DMOAD on incident knee replacement to be detected with sufficient power and confidence. The estimated incidence rates of knee replacement and the corresponding sample sizes are important for informing the design of trials for disease course-modifying effects as well as for socioeconomic evaluation of a DMOAD in terms of preventing knee replacement.