Table 4.
Sensitivity and specificity for One Card Learning accuracy performance based on derived, optimal cut-offs and conventional cut-offs for subtle objective cognitive impairment (sOBJ) at baseline and longitudinal decline (∆OBJ) defined using three different methods.
| A−T− and A+T+ comparisons | ||||
|---|---|---|---|---|
|
| ||||
| Index | Threshold type | Threshold | Sensitivity | Specificity |
| sOBJ | Optimal | <−0.215 z | 0.50 (0.34, 0.66) | 0.82 (0.76, 0.87) |
| Conventional | ≤ −1 z | 0.09 (0.00, 0.22) | 0.97 (0.94, 0.99) | |
| ΔOBJ-WSD | Optimal | <−0.691 z | 0.19 (0.06, 0.31) | 0.97 (0.95, 0.99) |
| Conventional | ≤ −1 z | 0.12 (0.03, 0.25) | 0.98 (0.95, 1.00) | |
| ∆OBJ-LR | Optimal | < 29.6%ile (0.0047) | 0.50 (0.31, 0.69) | 0.69 (0.63, 0.75) |
| Conventional | < 10%ile (−0.0199) | 0.22 (0.09, 0.38) | 0.89 (0.84, 0.93) | |
| ∆OBJ-LME | Optimal | < 16.5%ile (0.0173) | 0.62 (0.47, 0.78) | 0.75 (0.70, 0.81) |
| Conventional | < 10%ile (0.0162) | 0.38 (0.22, 0.56) | 0.85 (0.80, 0.90) | |
| A−T− and A+T− comparisons | ||||
|
| ||||
| Index | Threshold type | Threshold | Sensitivity | Specificity |
|
| ||||
| sOBJ | Optimal | < 0.293 z | 0.56 (0.44, 0.68) | 0.57 (0.50, 0.64) |
| Conventional | ≤−1 z | 0.06 (0.01, 0.12) | 0.97 (0.94, 0.99) | |
| ΔOBJ-WSD | Optimal | <−0.636 z | 0.12 (0.04, 0.21) | 0.96 (0.93, 0.98) |
| Conventional | ≤ −1 z | 0.06 (0.01, 0.12) | 0.98 (0.95, 1.00) | |
| ∆OBJ-LR | Optimal | < 6.4%ile (−0.0257) | 0.16 (0.07, 0.25) | 0.92 (0.89, 0.96) |
| Conventional | < 10%ile (−0.0199) | 0.16 (0.09, 0.25) | 0.89 (0.84, 0.93) | |
| ∆OBJ-LME | Optimal | < 17%ile (0.0172) | 0.41 (0.29, 0.53) | 0.79 (0.73, 0.84) |
| Conventional | < 10%ile (0.0162) | 0.31 (0.21, 0.43) | 0.85 (0.80, 0.90) | |
Note: Conventional cut-offs for sOBJ and ∆OBJ-WSD are ≤ −1 SD, which corresponds to ≤ −1 z using age-corrected Cogstate norms and norms for change respectively. Annualized change percentiles used in ∆OBJ-LR and ∆OBJ-LME correspond to our cognitively unimpaired reference group (n=732 aged 50–65 at baseline); the corresponding cut-off for annualized change for OCL accuracy raw score (arcsine transformed) is provided in parentheses. Optimal cutoffs were derived using the Youden index [37], which maximizes the difference between the true positive and false positive rate across all possible cut-point values. A = amyloid; T = tau; OCL= One Card Learning; sOBJ = subtle objective cognitive impairment at baseline; ∆OBJ = objective longitudinal cognitive decline; WSD = within subjects standard deviation; LR = linear regression model; LME = linear mixed effects model; A−T− group N = 211; A+T+ group N = 32; A+T− group N = 68.