Carolien Zeetsen

48 the remaining results were sufficient to validly classify each patient. Mean NPA domain scores and patient characteristics were compared between patients with and without NCD by using independent t –tests, chi–square tests or Mann–Whitney U –tests (non–normal variables). Criterion validity Level of education was classified on a seven–point scale ranging from 1 = less than primary school to 7 = university degree or higher (Duits et al., 2014), a classification system comparable to the ISCED (UNESCO, 2012). As it was found that years of education affects performance on the MoCA (Nasreddine et al., 2005; Chertkow et al., 2011), Spearman’s rho correlations were used to relate the unadjusted MoCA–TS at baseline and follow–up to this level of education. Based on the studies by Chertkow et al. (2011) and van der Elst et al. (2005), the MoCA–TS was then adjusted for education (low level of education, classifications 1, 2 and 3: two additional points; average level of education, classifications 4 and 5: one additional point; and highly educated patients, classifications 6 and 7: no additional points), with the maximum MoCA–TS remaining 30 in all cases. MoCA results were then explored and differences between patients with and without NCD were computed using independent t –tests. Furthermore, MoCA–DS were correlated with mean z –scores on the corresponding NPA domain, and systematic differences between MoCA–DS and –TS at baseline and at follow–up were assessed with paired t –tests. The predictive validity was assessed by computing a receiver operating characteristic (ROC) curve with the corresponding area under the curve (AUC) for the MoCA–TS at baseline, with NPA classification (NCD or no–NCD) at follow–up as a criterion. The cut–off point was determined by the optimal sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). After applying the cut–off to the data, overall agreement and chance–adjusted agreement (Cohen’s kappa) were determined. The concurrent validity was assessed in the same way as the predictive validity by using MoCA and NPA results at follow–up. Substance type and abstinence duration The influence of substance type and abstinence duration on MoCA performance at follow– up was estimated using logistic regression with abstinence duration, substance type (alcohol versus other drugs), MoCA–TS, and interactions between each as predictors, with NCD classification (NCD versus no–NCD) as the dependent variable. The Outlier Labelling Rule (Hoaglin & Iglewicz, 1987) was used to exclude outliers, leading to the exclusion of one outlier for abstinence duration. All data were computed and analysed with IBM SPSS version 24.0.