Oral Papers: Collaborative Care & Community C-L II
Background: After medical hospitalization, people with serious mental illness are at high risk of psychiatric destabilization and emergence of suicidal ideation and behavior. For example, in the week following discharge from medical hospitalization, individuals are 10 times more likely to die by suicide compared with matched controls(1). Although suicide research traditionally focused on psychiatric hospitalizations, half of patients who suicide within 3 days of discharge were hospitalized for non-psychiatric reasons(2). Despite this extreme vulnerability, identification of patients at high risk of suicide following medical hospitalization remains unclear. To address this issue, we developed an actuarial risk algorithm predicting readmission for suicidality after medical hospitalization.
Methods: In this longitudinal cohort study, electronic health record data from adult ( >18-years-old) patients with serious mental illness (major depressive disorder, bipolar disorder, or psychotic disorder) were used to develop an algorithm to predict readmission for suicidality after medical hospitalization. We extracted codified data on diagnosis (ICD-9 and 10 codes), medications, care utilization (ambulatory, emergency department, and hospital encounters), demographics, and financial charges. Classification Trees were implemented to hierarchically structure linear, nonlinear, and interactive predictors. The gold-standard outcome was readmission for suicide attempt and/or ideation in the 12 months following medical hospitalization. As the outcome of suicidality was infrequent, balanced trees were derived and 10-fold cross-validation was used to internally validate the models. The sensitivity, specificity, accuracy and area under the curve (AUC) were compared. Analyses were conducted in R 3.4.0 and Python 3.7.0, SciPy v0.19.1. The UCLA IRB approved this study.
Results: There were 16,552 medical hospitalizations (N=5,255) of patients with serious mental illness (major depressive disorder, bipolar disorder, or primary psychotic disorder) from 2006-2016 across two study sites. 287 patients were rehospitalized (5.5% of all patients) for suicide attempt (N=83) and/or suicidal ideation (N=220) following medical hospitalization. The model accurately identified 107/108 rehospitalizations for suicide attempt and 312/338 rehospitalizations for suicidal ideation. The model identified three pathways of increased risk of suicide attempt [Sensitivity:99.1%, Specificity:97.3%, Accuracy:97.3, AUC:0.69], suicidal ideation [Sensitivity:92.3%, Specificity:84.6%, Accuracy:84.7, AUC0.93], and any suicidality [Sensitivity:92.2%, Specificity:84.9%, Accuracy:85.1, AUC:0.71]. Highest risk phenotypes were characterized by prior suicidality, moderate Elixhauser Comorbidity Index, and psychiatric comorbidity. Low risk phenotypes were characterized by frequent contact with medical providers with antecedent medical hospitalizations and ambulatory visits.
Conclusions: The high concentrated risk of suicidality among certain phenotypes of patients following medical hospitalization may justify increased referral to psychiatric services or aftercare interventions. Identification of modifiable non-linear and interactive predictors may inform hospital-based interventions to mitigate suicide risk after hospitalization.
1.Qin,P.,et al."Hospitalization for physical illness and risk of subsequent suicide: a population study." J.of internal medicine 273.1(2013):48-58.
2.Drake,S.A.,et al."Suicide within 72 hours after discharge from health care settings: Decedent characteristics." Am J.of forensic medicine and pathology 37.1(2016):32-34.