A receiver operating characteristic (ROC) curve analysis of a model of mental health services use by Puerto Rican poor.

In this study, the contribution of four distinct domains of the Help Seeking-Decision Making model to predicting the use of mental health services is examined. Using a proposed methodology the authors assess the relevance of this model and its domains to mental services planning. The methodology combines logistic regression analysis and receiver operating characteristic (ROC) curves. Logistic regression analysis allows us to examine the individual variables of the model and generate predictions about use. ROC curves allow us to compare and interpret the relative contribution of a predisposing domain, a physical and mental health domain, an enabling-restrictive domain, and an organizational domain in correctly classifying users and nonusers of mental health services. The physical and mental health domain yielded a Somer's D-statistic of 0.7, which corresponds to an 85% correct classification of randomly selected pairs of users and nonusers. The study findings suggest that comparing ROC curves helps to describe and interpret the domains of the model that are relevant for making predictions about who will or will not use mental health services during a 1-year period.