While the value of information (VoI) methods have been developed for model based cost-effectiveness analyses (CEA), there is a gap in applying such methods to CEAs conducted alongside clinical trials. In the present it is shown that by treating the probability distribution of net benefits within each trial arm as a random quantity, calculations for the Expected Value of Sample Information (EVSI) for model-based CEAs can be extended to the setting of a trial-based CEA where individual net benefits from subjects are available. In this case EVSI calculates the expected return of investment from conducting a future trial with the similar design as a function of its sample size. Based on such analogy, a nonparametric method for EVSI calculation based on two-level bootstrap is introduced. At the first level, a Bayesian bootstrap (Rubin, 1981) from the vector of individual-level net benefits within each arm of the trial is performed. The empirical distribution of such a bootstrap sample amounts to a random draw from the ‘posterior distribution of the distribution’ of net benefits given the observed data. A second bootstrap from this sample then models the distribution of net benefits in the future study. The data of the future and current trials will be combined to identify the maximum net benefit and the cycle is repeated over several iterations. We also extend this framework to address parametric analysis (e.g. net benefit regression), missing values, and incorporation of external evidence. We use data from a randomized clinical trial of combination therapy in COPD as an application. Since the two-level bootstrap directly generates samples of the individual-level data for the future trial, it allows modeling realistic scenarios (e.g. missing values, complex statistical analysis). This, combined with the nonparametric nature of the method, should provide a robust framework for VoI analysis for trial-based CEAs.