OBJECTIVES Individual simulation is increasingly used in economic models, partly because of its capability of predicting event risks based on individual patient characteristics. However, due to lack of individual patient level data, models often use means and standard deviations to create patient profiles. The objective of this study is to evaluate different simulation approaches of creating patient profile at baseline and their impact on model outcomes. METHODS Patient level data (N=8,857) from National Health and Nutrition Examination Survey (NHANES) was used to evaluate three approaches of creating baseline patient profiles for simulation models.  10 samples of 1000 patients each were created through 1) random sampling from patient level data; 2) using means and standard deviations of the profile variables without correlating the characteristics; 3) using means and variance-covariance matrix among the continuous variable characteristics with cholesky decomposition approach. 10-year cardiovascular diseases (CVDs) rates are estimated using the created patient profiles from these 3 different approaches. RESULTS The predicted CVD rates based on random sampling are 18.2% for males and 9.7% for females using the random sampling approach, 14.5% for males and 7.9% for females using the mean and standard deviation approach and 16.0% for males and 9.2% for females using the cholesky decomposition approach. The CVD rates using the NHANES entire population are 18.5% for males and 9.8% for females. CONCLUSIONS Random sampling from patient level data provided the best approximation of actual NHANES population predicted CVD rates.  The cholesky decomposition approach was slightly limited since only continuous variables could be utilized which could explain the deviation from the population predicted CVD rates.  Independent sampling underestimated the mean risk by ~20%, an interesting finding as many individual simulation models created patients with this approach.  Researchers should be cautious in their use of summary statistics when populating individual simulation models.