Age and disease severity predict choice of atypical neuroleptic: a signal detection approach to physicians' prescribing decisions.

Objective: We used a novel application of a signal detection technique, receiver operator characteristics (ROC), to describe factors entering a physician's decision to switch a patient from a typical high potency neuroleptic to a particular atypical, olanzapine (OLA) or risperidone (RIS).
Methods: ROC analyses were performed on pharmacy records of 476 VA patients who had been treated on a high potency neuroleptic then changed to either OLA or RIS.
Results: Overall 68% patients switched to OLA and 32% to RIS. The best predictor of neuroleptic choice was age at switch, with 78% of patients aged less than 55 years receiving OLA and 51% of those aged greater than or equal to 55 years receiving OLA (chi(2)=38.2, P<0.001). Further analysis of the former group indicated that adding the predictor of one or more inpatient days to age increased the likelihood of an OLA switch from 78% to 85% (chi(2)=7.3, P<0.01) while further analysis of the latter group indicated that adding the predictor of less than 10 inpatients days to age decreased the likelihood of an OLA switch from 51% to 45% (chi(2)=7.0, P<0.01).
Conclusions: ROC analyses have the advantage over other analyses, such as regression techniques, insofar as their "cut-points" are readily interpretable, their sequential use forms an intuitive "decision tree" and allows the potential identification of clinically relevant "subgroups". The software used in this analysis is in the public domain (http://mirecc.stanford.edu).