Predicting rehospitalization and outpatient services from administration and clinical databases

Abstract

The study tests whether psychiatric services utilization may be predicted from administrative databases without clinical variables equally as well as from databases with clinical variables. Persons with a psychiatric hospitalization at an urban medical center were followed for 1 year postdischarge (N = 1384.) Dependent variables included statewide rehospitalization and the number of hours of outpatient services received. Three linear and logistic regression models were developed and cross-validated: a basic model with limited administrative independent variables, an intermediate model with diagnostic and limited clinical indicators, and a full model containing additional clinical predictors. For rehospitalization, the clinical cross-validated model accounted for twice the variance accounted by the basic model (adjusted R2 = .13 and .06, respectively). For outpatient hours, the basic cross-validated model performed as well as the clinical model (adjusted R2 = .36 and .34, respectively). Clinical indicators such as assessment of functioning and co-occurring substance use disorder should be considered for inclusion in predicting rehospitalization.