Predictors of length of stay in inpatient child and adolescent psychiatry:: failure to validate an evidence-based model

Objective To test if predictors of length of stay are based on empirical evidence. Method 100 1 regularly terminated treatment episodes from 13 child and adolescent psychiatric hospitals were analysed. In order to cross-validate the results, the sample was randomly divided into a definition sample (n = 500) and a validation sample (n = 501). The variables in the definition sample were screened statistically for their suitability as predictors of logarithmic length of stay (logLOS). Variables shown to be significant and uncorrelated were entered into multifactor analyses of variance in order to generate the model with the largest amount of explained variance of logLOS. Subsequently the results were tested against the validation sample. Results in the definition sample we found the three predictor variables admission as crisis intervention, out of home dispositions and psychoanalytic therapy which could explain 23.7% of the variance of logLOS. Unfortunately, this could not be replicated in the validation sample as a model. Conclusion Simple models of prediction of LOS in the field of child and adolescent psychiatry cannot be reliably based on empirical evidence. The main consequence is that fixed disorder-related reimbursement systems do not seem justified.