Text mining of outpatient narrative notes to predict the risk of psychiatric hospitalization

The primary purpose of this paper is to investigate whether text mining of the outpatient narrative notes for patients with severe and persistent mental illness (SPMI) can strengthen the predictions concerning the probability of an upcoming hospital readmission. A five-year study of all clinical notes for SPMI patients at the outpatient clinic of a tertiary hospital was conducted. The clinical notes were studied using ensemble classification i.e., entity recognition. Confounding variables pertaining to the patient's health status were extracted by text mining. A mixed effects logistic regression model was used for estimating the re-hospitalization risk during a clinic visit. The factors included frequency and continuity of outpatient visits, alterations in medication prescriptions, the usage of long-acting anti-psychotic injections (LAIs), the presence or absence of a legal compulsory treatment order (CTO) and the hospitalizations. The appearance of certain words in the outpatient clinical notes has a statistically significant impact on the risk of an upcoming hospitalization. This study also reconfirms that the risk of a re-hospitalization of an SPMI patient is reduced by the presence of a CTO and the utilization of LAIs, whereas it is increased by the patient dropping out of outpatient care. Our findings pertaining to the risk of re-hospitalization could facilitate preventive interventions for SPMI patients with higher risk.