Calibration of large, complex simulation models is difficult, as there are several parameters that are involved in state transitions, and many of these parameters are estimated only within the literature. The Phase I model was calibrated using a Bayesian method that constructed a neural-network meta-model of the agent-based model that allowed testing of hundreds of thousands of parameter sets to find the sets that best fit the outcomes data across multiple counties. The Phase II model is far more complex and has well over 100 parameters which must be calibrated to fit overdose and fatal overdose data over time. Furthermore, many of these transition probabilities are dependent upon specific characteristics (i.e., covariate values) of the individuals in the model (e.g., age, gender, race, income, etc.). We have developed a Discrete Time Markov-chain-based method to effectively calibrate these large, complex simulation models with many parameters. For more details see: