Development and validation of a prediction rule for psychiatric hospital readmissions of patients with a diagnosis of psychosis
Vazquez Montes M., Maruri-Aguilar H., Rugkasa J., Yeeles K., Morandi S., Burns T.
Multiple studies of psychosis have demonstrated an association between a variety of clinical and demographic patient characteristics and psychiatric hospital readmission. Our aim is to investigate whether these potential predictors retain their predictive value when put together to create a prediction tool for hospital readmission that clinicians could use to guide best care practice for patients with psychosis when discharged from hospital. A prediction model for psychiatric hospital readmission will be developed using data from an existing single-outcome, parallel arm, non-blinded randomised trial (OCTET). The trial recruited patients aged between 18-65 years, diagnosed with psychosis, registered in one of 60 NHS Trusts providing mental health services in England. Recruitment took place from 10 November, 2008 to 22 February, 2011. A total of 119/336 (35%) patients were readmitted to hospital in the 12-month follow-up after discharge at baseline (60/167 patients were re-admitted in the control group only). A set of 17 potential predictors has been identified from those found in the literature. Univariate and multivariate logistic regression models will be fitted using a smooth supersaturated polynomial technique which allows for correlations between the predictors. No variable selection will be done and multiple imputation will be used to account for missing values. The final multivariate prediction model will be internally and externally validated, using bootstrap and applying the model to an independent dataset to evaluate its performance, respectively. We will present the final model and describe any challenges we encounter to obtain it, both clinical and methodological. We will report the independent and adjusted prediction value for each of the pre-selected predictors; the multivariate model’s discrimination and calibration, the latter adjusted for optimism; and the model’s sensitivity, specificity and area under the receiving operating curve. Findings from the external validation analysis will also be presented.