Identifying risk factors for early psychiatric rehospitalization (EPR, rehospitalization within 90 days) can inform strategies to reduce rehospitalization rates. Random forest (RF), a tree-based classification algorithm, can be useful for identifying potential risk factors for EPR from a large number of patient factors. Patient characteristics were collected from 519 psychiatric inpatients at eight New York City hospitals. RF was used to identify potential risk factors for EPR. Multiple logistic regression was performed to assess the association between the identified risk factors and rehospitalization. Top risk factors identified by RF were previous psychiatric hospitalizations, number of post-discharge needs, social isolation, and sense of belonging in one's community. Follow-up analyses confirmed the significant association between EPR and number of previous psychiatric hospitalizations, number of endorsed post-discharge needs, and social isolation after adjusting for demographic variables. Understanding the contributors to EPR can better inform mental health service planning, policies, and programs that promote recovery.
Keywords: Machine learning; Psychiatric hospitalization; Random forest; Readmission; Rehospitalization; Serious mental illness.
Published by Elsevier B.V.