Though it is definitely a good idea having a SAP in every kind of biomedical research, it may not be compulsitory in merely exploratory studies. However, in clinical trials or hypothesis testing research it can be regarded mandatory to have it even before any kind of generating or colecting data.
The advantage of preparing a SAP is to reduce the number of possible comparaisons between the different experimental groups, as you have predicted the direction of the expected changes. Therefore, the protection factor, like Bonferoni, against multiple comparaison biases is less heavy due to fewer comparaisons. In other words, you may reach valuable conclusions with fewer subjects , less work and lower costs in your experimental plan.
In industry-sponsored trials the SAP is mandatory and should be finalized preferably soon after protocol finalization but in any case before database lock.
My guess is that in general any clinical trial should have a SAP in place for many reasons among which (aside primary and secondary end point analysis and the multiple comparison issue already mentioned in a previous answer) pre-planned subgroup analysis and/or stratifications an/or exploratory, hypothesis generating tests.
Useful guidabce can be found in the attached CHMP guideline
Everything will be well defined if you have SAP with protocol. I would like to again remember the quote from Sir R. A. Fisher that "To consult the statistician after an experiment is finished is often merely to ask him to conduct a post-mortem examination. He can perhaps say what the experiment died of."