Predictors of admission in first-episode psychosis: developing a risk adjustment model for service comparisons

The aim of this study was to develop a risk adjustment model based on hospital admissions that would enable comparison between services for patients with a first episode of psychosis.

Candidate predictor variables for hospital admission were identified in a literature review, from which an expert panel selected 12 potential risk adjustment variables by using a structured process, the Template for Risk Adjustment Information Transfer. Multivariable logistic regression modeling with the 12 variables was used to develop models in one cohort of first-episode psychosis patients (N=297); these models were validated with data from a second cohort (N=309). The C statistic, a measure of model discrimination, was calculated to assess model performance.

In the data from the development sample, prior hospitalization was the only significant predictor of hospital admissions within one year of enrollment in the first-episode psychosis program (odds ratio [OR]=1.88, p=.05). Hospital admissions after two and three years from admission to the program were significantly associated with higher levels of initial positive symptoms (OR=1.07, p=.02; OR=1.06, p=.02, respectively), and prior hospitalizations (OR=2.72, p=.001; OR=3.34, p<.001, respectively). The logistic models performed well, with C statistics ranging from .72 to .74 for the three outcomes, where a value of 1.0 implies perfect model discrimination. In the validation data the C statistics were slightly lower, ranging from .67 to .72.

According to the C statistic estimates, the model developed provided good discrimination and was relatively robust in predicting hospitalization of first-episode psychosis patients.