Predicting rehospitalization and outpatient services from administration and clinical databases

Authors:

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.

This publication has 16 references indexed in Scilit: