Analysis of Predictive Factors for Aggression within Inpatient Psychiatric Hospitalization
Faculty Advisor Name
Bernice Marcopulos
Department
Department of Graduate Psychology
Description
Aggression is a heavy concern within inpatient psychiatric hospitalization. Previous research has turned to various methods of assessing and predicting aggression risk, including comprehensive psychometric risk assessment tools (e.g., DASA, VRAG-R, PCL:SV) as well as the analysis of specific dynamic and static risk factors (e.g., Patient Demographics, Situational Factors). Previous studies have focused on predicting whether aggression will or will not occur during a patient’s admission, with an ultimate goal of creating interventions to prevent acts of aggression from happening in the first place. However, little research has been done focusing on a solely aggressive sample. The goal of the current study was to determine if various patient and hospital variables cited in the literature to aid in prediction of the occurrence of aggression would maintain predictive utility in assessing whether a patient would be aggressive once (Dependent Variable = 0) or more than once (Dependent Variable = 1) during their admission. Three logistic regression models were created to assess this outcome. Data on 246 aggressive incidents involving patients (MAge = 36.97;SDAge = 13.37) were obtained from a state psychiatric facility on the east coast. The first model contained patient variables, including sex, age, presence of substance abuse disorder, presence of non-mood psychosis disorder, presence of mood disorder, presence of personality disorder, marital status, and legal status. The second model contained hospital/event related variables, including client living area, event date, event time. The third model contained all predictors. It was hypothesized that Model 1 and Model 3 would be significant overall; specifically, it was hypothesized that all patient variables would be significant, and that hospital/event related variables (pertaining to Model 2) would not be significant. Logistic regression analyses showed that no model significantly accounted for the binary outcome variable (Aggressive Once, Aggressive More Than Once). No model did significantly better than an intercept-only model, and there was a significant amount of deviance left unaccounted for by each model. Despite this lack of significance, a series of nested and non-nested model comparisons were utilized to compare model-data fit amongst the three models. Model 3 was shown to not work statistically better than Model 1 or Model 2. Model 1 and Model 2 were compared via their AIC values. While Model 1 had a lower AIC value than Model 2, the difference in AIC value was not substantial. It was concluded that in the context of the current psychiatric hospital from which data were obtained, empirically supported variables did not maintain predictive utility. Limitations and implications are discussed, as well as directions for future study.
Analysis of Predictive Factors for Aggression within Inpatient Psychiatric Hospitalization
Aggression is a heavy concern within inpatient psychiatric hospitalization. Previous research has turned to various methods of assessing and predicting aggression risk, including comprehensive psychometric risk assessment tools (e.g., DASA, VRAG-R, PCL:SV) as well as the analysis of specific dynamic and static risk factors (e.g., Patient Demographics, Situational Factors). Previous studies have focused on predicting whether aggression will or will not occur during a patient’s admission, with an ultimate goal of creating interventions to prevent acts of aggression from happening in the first place. However, little research has been done focusing on a solely aggressive sample. The goal of the current study was to determine if various patient and hospital variables cited in the literature to aid in prediction of the occurrence of aggression would maintain predictive utility in assessing whether a patient would be aggressive once (Dependent Variable = 0) or more than once (Dependent Variable = 1) during their admission. Three logistic regression models were created to assess this outcome. Data on 246 aggressive incidents involving patients (MAge = 36.97;SDAge = 13.37) were obtained from a state psychiatric facility on the east coast. The first model contained patient variables, including sex, age, presence of substance abuse disorder, presence of non-mood psychosis disorder, presence of mood disorder, presence of personality disorder, marital status, and legal status. The second model contained hospital/event related variables, including client living area, event date, event time. The third model contained all predictors. It was hypothesized that Model 1 and Model 3 would be significant overall; specifically, it was hypothesized that all patient variables would be significant, and that hospital/event related variables (pertaining to Model 2) would not be significant. Logistic regression analyses showed that no model significantly accounted for the binary outcome variable (Aggressive Once, Aggressive More Than Once). No model did significantly better than an intercept-only model, and there was a significant amount of deviance left unaccounted for by each model. Despite this lack of significance, a series of nested and non-nested model comparisons were utilized to compare model-data fit amongst the three models. Model 3 was shown to not work statistically better than Model 1 or Model 2. Model 1 and Model 2 were compared via their AIC values. While Model 1 had a lower AIC value than Model 2, the difference in AIC value was not substantial. It was concluded that in the context of the current psychiatric hospital from which data were obtained, empirically supported variables did not maintain predictive utility. Limitations and implications are discussed, as well as directions for future study.
