Vascular risk factors for depression and apathy | Part 1 66 for continuous variables. The independent variables age and neuroticism were checked for linearity in the logit by a Box-Tidwell approach. In addition, all independent variables were checked for multicollinearity by Pearson’s correlation coefficient. Missing data on covariates were treated by imputing the most reported value. Data were missing on: a history of depression (N=11; 0.8%), disability (N=12; 0.9%), education (N=10; 0.7%) or somatic comorbidity (N=28; 2.0%); the most reported value for all these variables was ‘0’. Differences in results for analyses with or without imputed data were checked as well as the effects of inclusion of dummy variables for missing data 40. Models for depression were tested using multivariate logistic regression. First, we wanted to test if stratification by gender was needed. To do this, the influence of gender on the interaction between neuroticism and vascular disease was studied using a three way interaction factor (gender*neuroticism*vascular disease), while correcting for lower order interactions. Since gender significantly affected the interaction between vascular disease and neuroticism (see results) subsequent analyses were stratified by gender. The robustness of the outcome was tested by leaving out outliers and influential cases in the solution and repeating the main analyses. In addition, the main analyses were repeated with different cut-off scores for the CES-D (15 and 17, respectively). Stepwise multivariate linear regression on the total CES-D score was performed using the same models. All analyses were carried out using the Statistical Package for the Social Sciences (SPSS) version 14.0. Results Twenty four study participants were excluded; 15 because of a diagnosis of dementia and 9 because of a history of bipolar disorder. Furthermore, 176 participants were excluded due to missing data on CES-D score (N=133; 8%), on vascular disease (N=27; 2%) or on neuroticism (N=16, 1%), leaving a sample of 1396 elderly participants. Missing data on CESD-score, vascular disease or neuroticism were related to female sex, marital status (living alone), disability, a low or medium level of education, ≥2 chronic comorbid diseases and a higher age (all at a P value <.05). The men and women in the remaining study sample did not differ in age and consisted of 799 men and 597 women, of whom 103 men (13%) and 72 women (12%) were 85 years of age and older. Baseline characteristics and group differences The study sample of 1396 elderly had a median age of 77.2 years (interquartile range) 73.3-81.5 years), 597 (42.8%) women participated in the study. In this population 291 (20.8%) had a CES-D score above cut-off. CRDS were significantly associated in univariate analyses with older age, female gender, marital status (living alone), a positive lifetime history of depression, lower education, presence of disability, higher levels of neuroticism, presence of cardiac disease and presence of stroke (Table 1). Correlations between all
RkJQdWJsaXNoZXIy MjY0ODMw