Transdiagnostic early warning score for psychiatric hospitalisation: development and evaluation of a prediction model

The lack of an early warning score for psychiatric hospitalisation means that the decision to initiate preventative interventions is based solely on clinical judgement, which is prone to bias.

The objective is to develop and externally validate a transdiagnostic score that predicts psychiatric hospitalisation.

In this retrospective cohort study using deidentified electronic health records from 20 healthcare organisations in the NeuroBlu Data, we identified all patients with any of seven major psychiatric disorders with at least five Clinical Global Impressions of Severity and five Global Assessment of Functioning measured over a period of 6 consecutive months before any hospitalisation. From these measurements, metrics of clinical severity and instability and functional severity and instability were derived and incorporated into a score predicting the 6-month risk of incident psychiatric hospitalisation. Discrimination and calibration of this score were validated in an external sample. The transdiagnostic validity of the score was evaluated and its performance was compared between white and non-white people.

Altogether, 37 049 individuals (531 incident hospitalisations) were included. The predictive model showed good discrimination in the training (optimism-adjusted c-index: 0.74, 95% CI 0.72 to 0.76) and external validation (c-index: 0.80, 95% CI 0.78 to 0.82) samples, with adequate calibration. Discrimination improved with adjustment for organisation-level hospitalisation rates (c-index: 0.80, 95% CI 0.78 to 0.82 and 0.84, 95% CI 0.82 to 0.86 in the derivation and validation samples). Good discrimination was also achieved for each diagnostic category (c-index: 0.71-0.82 and 0.64-0.75 with/without adjustment for organisation-level hospitalisation rates, respectively). There was no significant difference in model performance between white and non-white people.

A transdiagnostic early warning system based on simple longitudinal measurements can reliably and robustly predict psychiatric hospitalisation. It will help target preventative interventions to individuals most at risk.