I work in the prevention field, in which the horizon on the outcome is very far away. How can I optimize under these circumstances? It seems like it would take decades.

It is always preferable to optimize based on the outcome(s) of ultimate interest.  However, in situations where the ultimate outcome is very far off in the future, you can choose to speed up the process by optimizing based on measures of mediators rather than the ultimate outcome.  This is less definitive than optimization based on the outcome of ultimate interest, and a follow-up evaluation of the optimized intervention with an RCT, using the outcome of ultimate interest, is critical.

This approach to optimization relies on a carefully constructed conceptual model.  It makes the assumption that the mediators have been properly specified, and that they are causally related to the outcome, so that changing the mediators will change the outcome.  It also requires reliable and valid measurement of the mediators.

Intervention optimization is challenging under these circumstances.  We suggest taking the long view.  In 15 years, will your field be further along if you use the classical treatment package approach, or if you use the MOST framework?