OBJECTIVES: Heterogeneity can distort traditional indirect comparisons of treatments. Simulated treatment comparisons (STC) can overcome this with regression equations to balance differences in populations. Equations are derived using patient-level data from one trial (drug A, index); however, only mean values of predictors are typically known for the comparator (B). Thus, adjusted results must be generated by plugging these means in the equation, which can be biased for non-linear outcomes (e.g., time-to-event) since it yields the geometric rather than the required arithmetic mean. We describe a solution and illustrate its application in an STC of treatments of non-valvular atrial fibrillation (NVAF). METHODS: Data from the trial of drug A were used to derive an equation for the rate of major bleeds (MB) using Poisson regression. Predictors included gender, age, region, history of stroke/transient ischemic attack, hypertension, diabetes, renal dysfunction, prior use of various treatments. To avoid non-linearity bias, patient profiles were simulated by sampling predictor values from a multivariate-normal distribution with means set to drug B’s population and covariance matrix derived from the index trial. The average predicted rate for simulated patients represents the adjusted MB rate. To demonstrate that the approach works, we also apply it to the index trial.  RESULTS: A rate of 21 MBs/1000 person-years were observed with drug A. The predicted rate at the means of predictors of patients on drug A produced an estimate of 19 (16.4-21.0), whereas the mean of predicted rates with actual profiles was 22 (15.1-31.9). Repeating the calculations with simulated patients yielded 22.5 (15.3-33.0). The simulated MB rate in patients matching of the population of drug B was 30 (20.5-45.1), which contrasted with its observed rate (36.0) yielded a rate ratio of 0.84 (0.56-1.27). CONCLUSIONS: Predicting outcomes with a simulated comparator population produces accurate adjusted results for use in STCs.