A risk score tells you who might leave, not why. The useful next step is to pair the score with the likely reason, because travel problems, side effects, scheduling conflicts, and fading motivation each call for a different response.
Two people, two responses
Two participants sit at the same risk level. One has been missing visits that require a long journey; the other has been reporting mild side effects in check-ins. A single reminder message would fit neither well. Offering a closer site addresses the first; arranging a quick call with a clinician addresses the second. The score flagged both, but the reason decided the action.
This is where AI stops and people start. The system can sort and group; it cannot have the conversation that actually keeps someone in the trial.