Ires Ghielen

25 BAI factor analysis in PD patients choosing “not applicable” for all five items, or male patients answering question 24 or 25. Therefore, we decided to exclude the sexual items of the SCOPA-AUT from the analyses. Statistical analyses We performed all analyses using IBM SPSS Statistics 20 for Windows. The significance level was set at p < 0.05 with two-sided testing. Acceptability of missing values on the BAI, BDI and SCOPA-AUT was determined as less than 16.67% of items. In the event of more missing data, we excluded the patient for the analysis by pair-wise exclusion. When less than 16.67% of data was missing, we filled in missing values by mean imputation. We performed no imputation of missing data on the UPDRS-III, since we considered this to be unreliable for this scale. In the first analysis, we assessed dimensionality of the BAI with a principal component analysis (PCA). To determine the number of extracted factors we combined the Cuttman-Kaiser Eigenvalue greater-than-one rule and the “scree plot” criterion. We used oblimin rotation because we expected the different factors to correlate with each other. The factors obtained in this analysis can be considered as subscales of the BAI or symptom dimensions of anxiety. Scores on the derived subscales of the BAI were used in further analyses. Second, we studied the relationship between the BAI, BDI, SCOPA-AUT and UPDRS-III, by conducting multiple linear regression analyses. Assumptions for regression analyses (normality and homoscedasticity of residuals) were checked. Multicolinearity was evaluated with a correlation matrix and calculation of the variance inflation factor (VIF). The total BAI score was the dependent variable in the first set of regression analyses. In the second set, it was the score on the subscale of the BAI, derived with PCA previously. The independent variables of interest were the total score on the BDI, SCOPA-AUT and UPDRS-III. We conducted all analyses first with only the independent variable of interest (unadjusted model). We then adjusted the model stepwise for age and gender (model 1), use of dopaminergic medication (model 2), and the other two independent variables of main interest, i.e. the BDI, SCOPA-AUT and/or UPDRS-III score (model 3). Finally, we examined confounding in all models. 2