OBJECTIVES Multivariate functions can be used to predict individual risk for cardiovascular (CVD) events and also to estimate baseline risk in economic models. We present a comparison of deterministic versus stochastic risk predictions using Framingham’s [D’Agostino 2008] and REACH’s [Wilson 2012] functions. Stochastic risk prediction accounts for patient-level heterogeneity, but involves a number of issues including increased complexity, data requirements, need for assumptions and computational burden. To our knowledge, this topic has not been studied in the CVD setting. METHODS D’Agostino 2008 and Wilson 2012 modeled primary (PE) and recurrent event (RE) risks, respectively. Both studies considered fatal and non-fatal aggregate CVD events and estimated a Cox Proportional Hazards (CPH) multivariate risk function. In the deterministic prediction, the means of the risk factors were used to predict the population’s risk directly from the functions. In the stochastic prediction, individual patient profiles (n=10,000) were generated using Monte Carlo simulation. Individual risks were then estimated from the functions and averaged to compute the population’s risk. Multinomomial distributions were assumed for discrete variables (e.g. diabetes, number of vascular beds) and normal or log-normal distributions were assumed for continuous variables depending on skewness (e.g. age, total cholesterol). Probability distributions were parameterized based on the risk factors descriptives reported in the original references. Simulations were performed with and without considering dependence of risk factors. RESULTS Due to the non-linearity of the CPH function, the stochastic prediction yielded 23% (PE) and 17% (RE) higher risks than the deterministic approach (14% and 10%, respectively, if age was kept constant). Differences between prediction approaches are even higher if the estimated correlation structure of risk factors is accounted for. CONCLUSIONS When compared to the stochastic prediction, the deterministic approach leads to lower estimates of CVD risks. Therefore, economic models using this approach might underestimate treatment effect.