The simulation-based optimization framework (Sim-Opt) uses a twin-loop computational architecture, which combines mathematical programming and discrete event simulation, to address this problem. This article extends our earlier work to present methods for integrating information from the inner loop (Sim-Opt time lines, reactive adjustment) and using it in the outer risk-control loop (Stochastic Optimization loop) to obtain statistically significant improvements in the solutions to the underlying stochastic optimization problem. Two classes of information can be obtained from the inner loop time lines: the first pertaining to portfolio selection and the second resource crowding associated with the chosen operation policy. Methods presented quantify the information on these two classes, and a three-step heuristic incorporates this information in the outer risk-control loop to drive the system toward improving solutions with respect to the mean net present value (NPV) of the portfolio and the probability of delivering a positive NPV. This method was used on a pharmaceutical product development case study, consisting of 11 projects, 154 activities, 14 resource types and a 20-year planning horizon with respect to patent expiration. Basic algorithm engineering efforts are also described to significantly improve the performance of formulation generation, the generation of a heuristic lower bound and the identification of cut families to effectively apply branch-and-cut methods.