Fig. 2

Comparison of adjusted annual mean HCRU per patient between patients with and without diagnosed behavioral symptoms. Adjusted annual means were obtained from one-part (outpatient visits, other medical claims, and pharmacy visits) or two-part (inpatient visits, inpatients days, ER visits, and physician office visits) multivariate generalized linear models. In the two-part models, the odds ratios for a positive value were obtained in the first part of the model, and for the second part, the counts were modelled within the subgroup of patients with a positive outcome. With the one-part model, the counts were modelled in all patients. The models used a distribution selected between Poisson, negative binomial, and gamma based on the Akaike information criteria. In each model, the baseline covariates included were age group, index year, gender, insurance plan type, region of residence, urbanicity, use of antidepressants, use of antipsychotics, use of ASH/benzodiazepines, use of ASH not elsewhere classified, the Elixhauser Index Score, and institutionalization, as well as the each of the following comorbidities: hypertension, diabetes, cancer, anemia, congestive heart failure, cardiac arrhythmia, chronic pulmonary disease, renal failure, fluid and electrolyte disorders, and depression. The average marginal effects represent, in absolute terms, the incremental counts associated with behavioral symptoms when all other covariates are kept constant, using a recycled prediction technique. The mean number of inpatient days of stay per patient is calculated for patients with at least one inpatient visit. ASH: Anxiolytic, sedative, and hypnotic; BS: Behavioral Symptoms; CI: confidence interval; ER: the emergency room