*291. Measuring Aggression in Older Adults: A Latent Variable Modeling Approach
K Wristers, Houston Center for Quality of Care and Utilization Studies; 2. Department of Medicine, Baylor College of Medicine; CA Orengo, Mental Health Service, Houston VAMC; Department of Psychiatry and Behavioral Sciences, Baylor College of Medicine; ME Kunik, Houston Center for Quality of Care and Utilization Studies; Department of Psychiatry and Behavioral Sciences, Baylor College of Medicine; L Snow, Houston Center for Quality of Care and Utilization Studies; 2. Department of Medicine, Baylor College of Medicine; V Molinari, Mental Health Service, Houston VAMC; Department of Psychiatry and Behavioral Sciences, Baylor College of Medicine
Objectives: Physical aggression (i.e., hitting, kicking) is among the most dangerous and distressing behaviors in patients with dementia, and one of the most common precipitants of nursing home admission. Because aggression is both difficult to define and measure, many scales exist to measure it. These disparate measurements challenge validity within studies and make comparison across studies difficult. This paper demonstrates a technique for addressing these concerns by combining information from several aggression scales into one aggression score using latent variable modeling.
Methods: We analyzed data from a cross-sectional study conducted at the Veterans Affairs Medical Center Outpatient Geriatric Psychiatry Clinic. Male patients (n=49) with a DSM-IV diagnosis of dementia without abnormal liver enzymes or serum creatinine levels or unstable medical illnesses were included in the study. Data from seven aggression scales including the Overt Aggression Scale (OAS), the Overt Agitation Severity Scale (OASS), the Neuropsychiatric Inventory (NPI), the Behavioral Pathology in Alzheimer's disease Rating Scale (BEHAVE-AD), and the three subscales of the Cohen-Mansfield Agitation Inventory (CMAI) were used to indicate the latent aggression variable. We first evaluated whether the scales measured the same factor. We then compared the reliability of the latent variable with the reliability of each scale as estimated using Cronbach's alpha with the current sample.
Results: The unidimensional latent variable model of aggression adequately represented the data and latent variable aggression scores were estimated for all patients, even if they were missing data on a couple of scales. Reliability of the aggression latent variable was estimated as .91, whereas reliability of the separate scales estimated with this sample ranged from.55 to .84.
Conclusions: Our findings suggest that the multiple aggression scales are unidimensional and can therefore be combined into one aggression score using latent variable modeling. This technique can be used to estimate latent variable scores, even when patients do not have complete data. The latent variable scores were shown to be more reliable and comprehensive than any one of the seven scales used as indicators.
Impact: Using latent variable modeling to estimate patients' aggression is not the simplest technique; however results suggest it is the most accurate and comprehensive. Not only does latent variable modeling offer advantages over traditional techniques for measuring aggression, but it can be applied equally well to the measurement of other theoretical constructs such as anxiety, pain, depression, functioning, and quality of life.