A Two-Study Comparison of Clinical and MRI Markers of Transition from Mild Cognitive Impairment to Alzheimer’s Disease
Table 2
Predictive accuracy of specific combinations of predictor variables for classification of transition to Alzheimer’s disease (AD) by 3 years of followup in two independent samples (ADNI and QD) of older adults with Mild Cognitive Impairment, and comparisons of three predictor models.
ADNI
QD
Model
Predictor variables
AUC (SE)
Sensitivity at specificity = 80% (90%)
Correct classification %
AUC (SE)
Sensitivity at specificity = 80% (90%)
Correct classification %
Age
0.497
12.74 (5.73)
55.67
0.739
52.61 (29.85)
73.02
MMSE
0.655
37.88 (19.20)
65.54
0.778
41.41 (26.79)
76.00
Hippocampal vol.
0.725
48.05 (34.42)
64.60
0.753
62.07 (41.38)
80.51
Entorhinal volume
0.718
50.65 (35.71)
67.16
0.773
67.86 (50.00)
80.34
AVLT
0.756
49.47 (25.16)
44.33
0.849
71.63 (53.13)
80.00
FAQ
0.738
49.05 (35.90)
44.33
0.708
45.46 (32.83)
75.42
Age, MMSE
0.659
36.94 (18.79)
63.48
0.821
72.73 (39.39)
76.00
Hippocampal and entorhinal volumes
0.744
55.84 (35.71)
68.98
0.824
67.86 (67.86)
88.03
AVLT/SRT and FAQ
0.811
62.74 (42.68)
72.70
0.879
78.13 (59.38)
82.05
Model 1
Age, MMSE, AVLT/SRT and FAQ
0.828 (0.024)
73.25 (49.05)
73.40
0.921 (0.027)
90.63 (81.25)
87.07
Model 2
Age, MMSE, Hippocampal and entorhinal volumes
0.783 (0.028)
57.79 (40.26)
73.72
0.866 (0.046)
82.14 (71.43)
87.07
Model 3
Age, MMSE, AVLT/SRT, FAQ, Hippocampal and entorhinal volumes
0.865 (0.022)
75.33 (55.20)
77.01
0.940 (0.027)
92.59 (88.89)
89.72
Model comparisons
AUC difference
value
AUC difference
value
Model 1 versus Model 2
0.0396
0.2271
0.0428
0.3618
Model 1 versus Model 3
0.0428
0.0035**
0.0282
0.1979
Model 2 versus Model 3
0.0824
0.0001**
0.0710
0.0254*
A threshold of 0.5 was used on predicted risk derived from the logistic regression models. Area under the curve (AUC) was derived from receiver operating characteristic (ROC) analyses. (157 converters) in ADNI and (33 converters) in QD. The differences between models in AUCs are slightly different from the direct subtraction of AUCs between models because of missing data that ranged from 1% to 4% for the variables examined in ADNI and 1% to 5% for the variables examined in QD.
*
, **.