Decision tree analysis of genetic risk for clinically heterogeneous Alzheimer’s disease

BMC Neurology

Table 3 Score set evaluation statistics

Score	AUC ± SE	P-val vs. A	P-val vs. Prev.	ΔAUC
A	0.69 ± 0.04	N/A	N/A	N/A
B	0.75 ± 0.04	0.01	0.01	0.056
C	0.79 ± 0.03	0.0018	0.02	0.038
D	0.81 ± 0.03	0.00010	0.18	0.021
E	0.83 ± 0.03	<0.00001	0.13	0.023
F	0.84 ± 0.03	<0.00001	0.64	0.005
G	0.85 ± 0.03	<0.00001	0.36	0.011
H	0.85 ± 0.03	<0.00001	0.76	0.003
I	0.86 ± 0.03	<0.00001	0.35	0.008
J	0.86 ± 0.03	<0.00001	0.42	0.006
K	0.87 ± 0.03	<0.00001	0.27	0.010
L	0.86 ± 0.03	<0.00001	0.11	−0.011
M	0.87 ± 0.03	<0.00001	0.35	0.005
N	0.86 ± 0.03	<0.00001	0.55	−0.004
O	0.87 ± 0.03	<0.00001	0.57	0.004
P	0.87 ± 0.03	<0.00001	0.76	−0.001
Q	0.88 ± 0.03	<0.00001	0.02	0.007
R	0.87 ± 0.03	<0.00001	0.90	−0.001
S	0.87 ± 0.03	<0.00001	0.82	−0.001
T	0.88 ± 0.03	<0.00001	0.67	0.002
U	0.87 ± 0.03	<0.00001	0.88	−0.001
V	0.88 ± 0.03	<0.00001	0.30	0.005
W	0.88 ± 0.03	<0.00001	0.27	0.005
X	0.89 ± 0.03	<0.00001	0.89	0.001
Y	0.89 ± 0.03	<0.00001	0.90	0.001
Z	0.89 ± 0.02	<0.00001	0.25	0.007

Each score set was evaluated for AUC of the ROC curve to assess predictive ability. Score Q (in bold) was determined the best performing scoring set given the following two sets resulted in consecutively lower AUC values. This resulted in a final score set, Q that had a statistically significant better AUC than just APOE (set A) alone, P < 0.00001.

ISSN: 1471-2377