Users of mobile devices in eHealth systems are often novices, and the learning process for them may be very time consuming. In order for systems to be attractive to potential adopters, it is important that the user interface should be very convenient and easy to learn. However, the community of potential users of a mobile eHealth system may be quite varied in their requirements, so the system must be able to adapt easily to suit user preferences. One way to accomplish this is to have the user interface driven by intelligent policies. These policies can be refined gradually, using inputs from potential users, through agents supported by artificial intelligence approaches such as neural networks. These policies may then be used to support adaptation to interfaces that are suitable for users who may have different requirements. This paper develops a framework for policy refinement for eHealth mobile interfaces, based on dynamic learning from user interactions.