Models used to inform decisions in health are often complex, data- and computation-intensive, and difficult to interpret. For this reason, many models developed by operations researchers to support clinical, operational, and public health decision making are never used by decision makers. This talk focuses on the use of metamodels – that is, a statistical approximation of an original model – to replace complex models for health-related decision making. We describe three metamodels focused, respectively, on personalization of drug treatment for schizophrenia, control of hepatitis C in prisons, and use of mass prophylaxis for controlling disease outbreaks. We show how machine learning methods can be used to develop approximate models that can perform nearly as well as the original model, but with fewer data requirements, less computational burden, and often greater interpretability. Although each metamodel is specific to the original model, these examples illustrate general principles for creating simple yet useful models to support health decision making. More broadly, these examples suggest ways in which complex models used to support decisions relating to the production and provision of manufactured products and services can in some cases be replaced with simpler metamodels.