Optimizing Neuropsychological Assessments for Cognitive, Behavioral, and Functional Impairment Classification: A Machine Learning Study
Table 10
Top 10 features most frequently found as best predictors across all 10 rounds and all 100 iterations using the features chosen by the neuropsychologists.
Level of impairment
Features
CDR = 1 versus CDR = 0
(1) LDELTOTAL (Logical Memory Test)
80%
(2) TOTALMOD (ADAS)
50%
(3) FAQ total (FAQ)
29%
(4) TOTAL11 (ADAS)
18%
(5) CATVEGESC (Semantic Fluency Test)
13%
(6) Q4 (ADAS)
13%
(7) LIMMTOTAL (Logical Memory)
9%
(8) Q8 (ADAS)
5%
(9) MMSCORE (MMSE)
5%
(10) Q1 (ADAS)
3%
CDR = 0.5 versus CDR = 0
(1) FAQ total (FAQ)
81%
(2) LDELTOTAL (Logical Memory Test)
77%
(3) Q4 (ADAS)
44%
(4) TOTALMOD (ADAS)
39%
(5) CATVEGESC (Semantic Fluency Test)
36%
(6) LIMMTOTAL (Logical Memory)
30%
(7) MMSCORE (MMSE)
30%
(8) Q8 (ADAS)
23%
(9) TOTAL11 (ADAS)
19%
(10) Q1 (ADAS)
19%
CDR = 1 versus CDR = 0.5
(1) FAQ total (FAQ)
82%
(2) CLOCKSCOR (CLOCK Test)
36%
(3) Q8 (ADAS)
35%
(4) LDELCUE (Logical Memory Test)
33%
(5) TOTAL11 (ADAS)
30%
(6) Q4 (ADAS)
29%
(7) TOTALMOD (ADAS)
22%
(8) LDELTOTAL (Logical Memory Test)
20%
(9) CATVEGESC (Semantic Fluency Test)
19%
(10) Q1 (ADAS)
18%
10 rounds of the nested CV and across 100 iterations.