Abstract
Psychotic disorders are one of the leading causes of disability worldwide. A substantial proportion of psychiatric inpatients are re-admitted after discharge. Readmissions are disruptive for patients and families and are a key driver of rising healthcare costs. Reducing readmission risk is therefore a major unmet need of psychiatric care. Electronic health records (EHRs) contain detailed descriptions about a patient’s illness presentation, prior course, and treatment plans – all vital information for identifying readmission risk. Developing clinically implementable machine learning tools to enable accurate prediction of risk factors associated with readmission offers opportunities to inform the selection of treatment interventions and implement appropriate preventive measures.
We have previously identified seven clinical important risk factor domains and clinical sentiments as regard to these seven readmission risk factor domains in electronic health records (EHR) associated with readmission (Holderness et al 2019). In this study we develop and test a classification model that predicts the risk of early readmission (readmitted within 30 days from the discharge date) for psychotic patients using prior identified risk factors and additional extracted features from EHR. The classifier establishes whether the most recent admission of a patient will be followed by an early readmission or not.