When a GeoEvent Service processes an event record, the processing is generally atomic. In other words, a filter or processor considering an event record's attributes and geometry has no information on other event records previously processed and will not cache or save the current event record's attributes or geometry for later consideration by an event record not yet received.
There are a few exceptions - monitor processors such as the Incident Detector or Track Gap Detector necessarily cache some information in order to monitor ongoing conditions. And filters configured with ENTER or EXIT criteria need to know something about the position of the last reported event with a given TRACK_ID.
So how do you configure real-time analytics to compare an event's geometry against some other geometry? You use geofences. Christopher Dufault has collected some best practices for importing, synchronizing, and using geofences in GeoEvent Server. Check out his blog Geofence Best Practices and comment with tips and tricks with geofences you've found useful in analytics you've designed.