Results of the multinomial linear regression for each of the 4 scenariosa
| Model | Healthy vs Dementia | Healthy vs AD | Healthy vs VaD | AD vs VaD |
|---|---|---|---|---|
| Variables entered in the model | PVS-1 | PVS-1 | PVS-1 | PVS-1 |
| PVS-2BG | PVS-2BG | PVS-2BG | PVS-2BG | |
| Modified Scheltens | Modified Scheltens | Modified Scheltens | Modified Scheltens | |
| Variables included in the model | PVS-2BG | PVS-2BG | PVS-2BG | PVS-2BG |
| Modified Scheltens | Modified Scheltens | Modified Scheltens | Modified Scheltens | |
| PVS-2BG z score/β coefficient (significance) | −3.074/−1.094 (<.001) | −3.161/−1.002 (<.01) | −2.623/−1.956 (<.01) | 2.212/0.757 (<.05) |
| Modified Scheltens z score/β coefficient (significance) | −1.928/−1.221 (NS) | −1.82/−0.951 (NS) | −2.537/−2.727 (<.05) | 1.674/0.498 (NS) |
| Area under ROC curve | 0.855 | 0.774 | 0.928 | 0.7135 |
| Sensitivity (%) | 94.1 | 78.9 | 91.8 | 65.3 |
| Specificity (%) | 71.1 | 73.6 | 84.6 | 71.4 |
| Positive predictive value (%) | 83.1 | 75.0 | 83.8 | 80.0 |
| Negative predictive value (%) | 88.9 | 77.8 | 91.7 | 54.1 |
Note:—NS indicates not significant.
↵a Diagnoses were treated as categories (AD = 1, VaD = 2, and Healthy = 3). Patient age and sex were entered as covariates in each model. All imaging variables were standardized by calculation of z scores (β/standard error) and were entered into the model if they showed a baseline correlation with the diagnosis with significance < .05. Individual z score, β coefficients, and significance are given for each imaging variable in the final model.