Prediction without intervention: why fifteen years of no-show models have not closed the loop on patient adherence.

Fifty-two published studies since 2010 have modeled which patients will miss which appointments. Ambulatory no-show rates have moved only modestly. This Perspective asks why fifteen years of prediction research has produced so little change in care, and argues that the answer is architectural, not algorithmic. In most health systems, the no-show model, the patient outreach platform, and the clinical risk tool are separate tools run by separate teams. Prediction is now routine, and prediction-guided interventions have reduced no-shows in trials. None of the systems in the studies and reviews examined here sense the full behavioral signal of their own outreach, fuse behavioral prediction into clinical risk, or learn in routine practice from the interventions they deliver. I describe three failures behind this gap: outreach platforms that discard the behavioral data they generate, behavioral and clinical risk models that never exchange a feature, and interventions that are never measured. I then propose three design principles for a closed-loop alternative, and a research agenda to test them. Maternal health, where a missed prenatal visit can mean an unmonitored pregnancy complication, is the motivating case; the principles apply wherever outcomes depend on sustained engagement with care.