Statistical analysis was performed with IBM SPSS version 19 (Chicago, IL, USA) and the statistical software package R (R Foundation, Vienna, Austria). Correlations between the different methods for MTL assessment Selleck PD-1/PD-L1 inhibitor 2 and between MTL and other biomarkes (CSF, APOE-e4 allele) were analyzed by the Pearson correlation coefficient for continuous data, and the Spearman rank test for correlations including the MTA-score. Intraclass correlation coefficient (ICC) and paired t test were performed to investigate the agreement between MTL measurements. The main outcome measure was the area under the curve (AUC) for AD-type dementia after 2 years of follow-up, calculated using a time-dependent receiver operating characteristics curve in R (Heagerty et al., 2000). Differences in AUC between methods were tested as described elsewhere (Hanley and McNeil, 1983). We calculated the sensitivity, specificity, positive predictive value, negative predictive value, odds ratio (OR), and hazard ratio (HR) for AD-type dementia at the 2-year follow-up using data-driven cut points based on a time-dependent receiver operating characteristics analysis. First we calculated the cut point that maximized the Youden Index (sensitivity + specificity − 1) for predicting AD-type dementia after 2 years. Second, we selected cut points that predicted AD-type dementia with a sensitivity of 85%. Cut points were calculated in the whole sample and in each cohort separately. Spline analyses were performed to determine AD-free survival after 2 years as a function of each MRI measurement in the total cohort. Slope analyses with mixed models were performed to investigate whether MTL atrophy was associated with change on the MMSE and a cognitive composite score at follow-up. The analyses included the baseline score and available follow-up scores up to 5 years after baseline and were corrected for age and education. An unstructured covariance structure with center as a random effect was used because this model provided the best −2 log likelihood compared with models with simpler covariance structures (Visser et al., 2009). From the total group of MCI patients included in this study 37% were diagnosed as nonamnestic MCI (73% single-domain and 27% multi-domain) and 63% amnestic MCI (44% single-domain and 56% multi-domain). Baseline and follow-up characteristics are shown in Table 1. Both cohorts were comparable for age and APOE-e4 status. The DESCRIPA cohort included more female participants and education was lower compared with the VUmc cohort. Scores on the MMSE and delayed recall task were lower in the VUmc cohort. At follow-up, 91 subjects were diagnosed with probable Alzheimer-type dementia (28%). Conversion rate was higher in the VUmc cohort (35.5%) than in the DESCRIPA cohort (19.2%), p < 0.001.