Sample size was based on a one-sided alternative hy-pothesis that physicians would be less likely to recommend TKA to the female patient compared with the male patient. By convention, the type I error rate ora was set at 0.05 and the type II error rate orb was selected to be 0.20. We esti-mated the magnitude of the hypothesized effect of patients’ gender on the physicians’ treatment recommendations based on findings from a population-based study, which showed a greater than threefold gender disparity in access to total joint arthroplasty (5.3 per 1,000 vs. 1.6 per 1,000 for women and men, respectively) [17]. To obtain an odds ratio of 3.3, we assumed that 30.7% of physicians would recommend TKA to the man but not the woman and that 9.3% would recommend the procedure to the woman but not the man. For the remaining physicians, we assumed that 30% would recommend TKA to both the man and the woman and that 30% would recommend neither patient for TKA. We based our sample size on the exact McNemar test for paired proportions; each pair consisted of the fe-male and the male patient visiting each physician. We determined that a sample size, h, of 58 physicians was required and assuming a 15% dropout rate, 71 physicians needed to be enrolled (58/(1 0.15)5 71).