Readmissions for mental health reasons are common.  The consequences of rapid readmission (30-day) to inpatient psychiatric units are significant from not only a clinical and public health perspective, but also specific to the disruption to the patient’s and their families’ lives (Vigod et al., 2015).  Vigod et al. (2015) note, to apply appropriate interventions, clinicians must be able to identify those who might be at greater risk for readmission and who may benefit from interventions to reduce readmission risk.

Due to limited research regarding methods to quantify risk for readmission, the authors focused on deriving and validating a clinical risk index to predict individual risk for rapid readmission.  The goal was to use this risk index to help identify high-risk individuals and offer interventions and support during the hospitalization and after discharge.  The authors focused on four categories of predictive factors:  socio-demographic, prior health care utilization, basic clinical and administrative information from hospital admissions, and data from psychiatric rating scales administered during an inpatient psychiatric admission.  Overall, the authors considered the feasibility of collecting the data to be used in the index as well as the predictive capabilities.

Vigod et al. (2015) collected data from individuals discharged from an acute psychiatric facility in Ontario, Canada between 2008 and 2011.  Population-level socio-demographic and health administrative data were collected from different databases.  The cohort was aged 18 or older and consisted of individuals hospitalized for greater than 72 hours.  The outcome variable was psychiatric readmission to any hospital in Ontario within 30 days where a mental health condition was the most responsible diagnosis.  Predictor variables were socio-demographic (age, education, martial status), prior health service use (lifetime mental health admissions, outpatient mental health visits, medical comorbidity), index hospitalization variables (threat to self, threat to others, unable to care for self, primary diagnosis, personality disorder, planned and regular discharge, length of stay), and clinical assessment characteristics (stressful life events, positive symptoms, negative symptoms, anxiety symptoms, aggressive behavior, self care levels, interpersonal conflict, and control interventions).

The authors used a split-sample method to build and validate the predictive models.   Thus, multivariable logistic regressions were used to determine the best-fit model, as well as assess for internal validity.  Four models were assessed in order to identify the best model to predict 30-day readmission.  The rational for assessing four models is centered on the goal of creating a model that not only best predicted readmission, but also was feasible to obtain data related to the predictor variables.  Model 1 was the most basic, only assessing socio-demographic variables.  Model 2 was more comprehensive assessing socio-demographic variables and prior health service variables.  Model 3 added basic clinical and administrative data related to the index admission. And Model 4, the most complex, added specific scores from clinical scales and assessment protocols during a psychiatric admission.  Model 4 was considered difficult to standardize as no uniform clinical information system was utilized within the sample.   Once the model was established, it was converted into a risk index system.

Overall, there were approximately 65,000 index admissions and the 30-day readmission rate was 91.9% (N = 6044) for the sample.  The highest risk period for readmission was within the first year after discharge, with minimal variation between the years assessed.  The average age for the sample was 44, with about half being female and most participants living in an urban area (89%).  Almost 2% were homeless at discharge and about one-quarter was employed. The most common diagnoses were Major Depressive Disorder and psychotic disorders, followed by Bipolar Disorder.

Results indicated that predictive capacity was poor with only socio-demographic variables (Model 1), and improved slightly when prior health service variables were included (Model 2).  With Model 3, adding clinical and administrative variables improved the discriminative capacity.  Adding more specific clinical variables resulted in significant improvement, yet discriminative capacity only increased slightly (Model 4).  Thus, the authors selected Model 3 to create the risk index, since the variables are easily collected and computed in most clinical settings.

The authors converted Model 3 into a risk index, and created an acronym (READMIT) to organize and identify the predictive variables.  More specifically, the variables within the model are:  R, history of admissions; E, emergent nature of index admission, specifically, harm to self, harm to others, inability to care for self; A, age; D, diagnoses of psychosis, bipolar, and personality disorder, and unplanned discharge; M, medical comorbidity; I, intensity of outpatient and emergency department use prior to admission; and T, time in hospital (Vigod et al., 2015, p. 208).  The range of scores for READMIT is 0 to 41.   Notably, a READMIT score increase of 1 point increased the odds of 30-day readmission by 11%.  Expected probability for the READMIT index ranged from 2% with a score of 0 to 49% for a score of 41.  See the Table 1 below (reproduced from Vigod et al., 2015, p. 211) for a summary of points allocated per risk factor.

The READMIT score quantifies 30-day readmission risk after hospitalization.  Consistent with existing literature, one or more previous admission is the most consistent predictive variable for readmission (Durbin, Lin, Layne, & Teed, 2007; Vigod et al., 2015).  Other risk factors identified by Durbin et al. (2007) related to rapid readmissions are younger age, forensic background, low family support, severe mental illness, and leaving the hospital against medical advice.  Here, READMIT also considers medical comorbidity, thus highlighting patient complexity.  Though the READMIT index only reported moderate predictive capacity, such is indicative of the unpredictability of events that occur after discharge that could mitigate or exacerbate risk.  Thus, it is not perfectly discriminative.

Overall, this study utilized population based socio-demographic and health service variables to identify risk factors related to rapid readmission.  Further, the authors indicated that the data are easily collected, which makes this index a feasible tool to identify those individuals at higher risk.  The goal of READMIT was to assess risk for readmission while individuals are in the hospital.  As a result, interventions are intended to be given prior to discharge in order to reduce the risk of readmission after discharge.  The authors note that READMIT could be used to indicate individuals in need of further assessment or specific intervention directed at modifiable risk variables either in the hospital or post-discharge.  It could also be used in research to identify post-discharge interventions designed to reduce readmission risk or to align resources for prevention efforts to areas or targeted populations of greater risk.  Overall, the authors created a risk index that moderately predicts readmission rates; such efforts can be useful specific to inpatient and post-discharge interventions.  The goal is to reduce readmission risk, yet it is difficult to impact risk if we are not able to identify those at greater risk.  In short, the READMIT score is a step in the direction of identification, with a focus on feasibility and predictive function.

References
Durbin, J., Lin, E., Layne, C., & Teed, M. (2007). Is readmission a valid indicator of the quality of inpatient psychiatric care?  Journal of Behavioral Health Services & Research, 34(2), 137-150.

Vigod, S. N., Kurdyak, P. A., Seitz, D., Herrman, N., Fung, K., Lin, E.,…Gruneir, A. (2015). READMIT: A clinical risk index to predict 30-day readmission after discharge from acute psychiatric units. Journal of Psychiatric Research, 61, 205-213.

Table 1

* See Vigod et al. (2015) for coding of specific medical conditions and diseases.

Dannie S. Harris, M.A., M.A., M.A.Ed., Ed.S.
WKPIC Practicum Trainee