The long-term clinical validity of the At Risk Mental State (ARMS) for the prediction of non-psychotic mental disorders is unknown.

Clinical register-based cohort study including all non-psychotic individuals assessed by the Outreach And Support in South London (OASIS) service (2002–2015). The primary outcome was risk of developing any mental disorder (psychotic or non-psychotic). Analyses included Cox proportional hazard models, Kaplan–Meier survival/failure function and C statistics.

A total of 710 subjects were included. A total of 411 subjects were at risk (ARMS+) and 299 not at risk (ARMS−). Relative to ARMS−, the ARMS+ was associated with an increased risk (HR = 4.825) of developing psychotic disorders, and a reduced risk (HR = 0.545) of developing non-psychotic disorders (mainly personality disorders). At 6-year, the ARMS designation retained high sensitivity (0.873) but only modest specificity (0.456) for the prediction of psychosis onset (AUC 0.68). The brief and limited intermittent psychotic symptoms (BLIPS) subgroup had a higher risk of developing psychosis, and a lower risk of developing non-psychotic disorders as compared to the attenuated psychotic symptoms (APS) subgroup (P < 0.001).

In the long-term, the ARMS specifically predicts the onset of psychotic disorders, with modest accuracy, but not of non-psychotic disorders. Individuals meeting BLIPS criteria have distinct clinical outcomes.

In the long-term, the ARMS designation is still significantly associated with an increased risk of developing psychotic disorders but its prognostic accuracy is only modest. There is no evidence that the ARMS is associated with an increased risk of developing non-psychotic mental disorders. The BLIPS subgroup at lower risk of developing non-psychotic disorders compared to the APS subgroup.

While incident diagnoses employed in this study are high in ecological validity they have not been subjected to formal validation with research-based criteria.