*3.2. Discriminative Capacity of the IGT-Learning Score*

Figure 2 contains the results of the ROC analysis obtained in the ED subsample. The optimal cut-off point in the IGT-Learning index to discriminate between good and bad CBT outcomes was 2, which achieved a sensitivity (Se), or true positive rate, of 64.2% and a specificity (SP), or true negative rate, equal to 54.8%.

Figure 3 shows the percentage of patients with a poor performance in the IGT in each group (based on the classification obtained for the cut-off = 2 in the global learning measure). The logistic regression (adjusted by age and education) valuing this cut-off's capacity for differentiating between the two groups achieved a significant parameter for differentiating between bad versus good groups (B = 0.754, SE = 0.301, OR = 2.12, *p* = 0.012). Goodness-of-fit was achieved (Hosmer and Lemeshow test: χ<sup>2</sup> = 5.95, df = 8, *p* = 0.653).

**Figure 2.** Valuation of the IGT-Learning raw score to predict the treatment outcome. Note. Results obtained for the ED subsample (*n* = 233).

**Figure 3.** Capacity of the IGT-Learning score to predict the treatment outcome. Each bar represents the percentage of participants with poor IGT-Learning in each group with a cut-off point equal to 2. Note. Results obtained for the ED subsample (*n* = 233).
