*3.3. Aim 2: Classification and Regression Tree Analysis Results*

The CART analysis revealed that the DLD SCS and Vineland-II community subdomain raw scores best-classified dementia (Figure 1). A DLD SCS less than 35 and a Vineland community score less than 34 are indicative of AD dementia.

*Brain Sci.* **2021**, *11*, 1128 6 of 10

**Figure 1.** Results of CART analysis. *Note.* AD = Alzheimer's disease; DLD SCS = Dementia Questionnaire for People with Learning Disabilities (DLD) sum of cognitive scores raw score; CMM Raw = Vineland-II community subdomain raw score. **Figure 1.** Results of CART analysis. *Note.* AD = Alzheimer's disease; DLD SCS = Dementia Questionnaire for People with Learning Disabilities (DLD) sum of cognitive scores raw score; CMM Raw = Vineland-II community subdomain raw score. **Figure 1.** Results of CART analysis. *Note.* AD = Alzheimer's disease; DLD SCS = Dementia Questionnaire for People with Learning Disabilities (DLD) sum of cognitive scores raw score; CMM Raw = Vineland-II community subdomain raw score.

### *3.4. Aim 3: Comparison of PCA and CART Model Classification Utility 3.4. Aim 3: Comparison of PCA and CART Model Classification Utility 3.4. Aim 3: Comparison of PCA and CART Model Classification Utility*

Comparing the PCA logistic regression and CART classification methods, there was no significant difference in AUC (D(65.196) = −0.57683; *p* = 0.57) (Figure 2). The PCA analysis resulted in an AUC of 0.87 while the CART model produced an AUC of 0.91. In terms of classification utility, the PCA model showed very good sensitivity (0.80) and good specificity (0.70), with high negative predictive value (0.824) and moderately high positive predictive value (0.667) at the combined sample base rate. The CART model demonstrated excellent sensitivity (1.00) and very good specificity (0.810), with excellent negative predictive value (1.00) and high positive predictive value (0.778) at the combined sample base rate. Comparing the PCA logistic regression and CART classification methods, there was no significant difference in AUC (D(65.196) = −0.57683; *p* = 0.57) (Figure 2). The PCA analysis resulted in an AUC of 0.87 while the CART model produced an AUC of 0.91. In terms of classification utility, the PCA model showed very good sensitivity (0.80) and good specificity (0.70), with high negative predictive value (0.824) and moderately high positive predictive value (0.667) at the combined sample base rate. The CART model demonstrated excellent sensitivity (1.00) and very good specificity (0.810), with excellent negative predictive value (1.00) and high positive predictive value (0.778) at the combined sample base rate. Comparing the PCA logistic regression and CART classification methods, there was no significant difference in AUC (D(65.196) = −0.57683; *p* = 0.57) (Figure 2). The PCA analysis resulted in an AUC of 0.87 while the CART model produced an AUC of 0.91. In terms of classification utility, the PCA model showed very good sensitivity (0.80) and good specificity (0.70), with high negative predictive value (0.824) and moderately high positive predictive value (0.667) at the combined sample base rate. The CART model demonstrated excellent sensitivity (1.00) and very good specificity (0.810), with excellent negative predictive value (1.00) and high positive predictive value (0.778) at the combined sample base rate.

(AUC) = 0.87, and CART model AUC = 0.91. **Figure 2.** ROC curve comparison for PCA versus CART derived models. PCA area under the curve (AUC) = 0.87, and CART model AUC = 0.91. **Figure 2.** ROC curve comparison for PCA versus CART derived models. PCA area under the curve (AUC) = 0.87, and CART model AUC = 0.91.
