Skip to main content
. 2021 Jul 28;12:637939. doi: 10.3389/fneur.2021.637939

Table 4.

Variable importance (VIMP) and relative variable importance (RVIMP) values from conditional Random Forest algorithm (100,000 trees) of each candidate clinical, demographical, pathological, treatment, and coffee/tea consumption variables in explaining the variability of the ΔFS (log values).

Rank Variable VIMP RVIMP
1 Age at onset 0.2075 100.00%
2 Education 0.0298 14.34%
3 Site of onset 0.0118 5.67%
4 Country 0.0058 2.79%
5 Duration of coffee consumption 0.0049 2.36%
6 Current alchool drinker 0.0048 2.32%
7 Lifetime intensity of green tea consumption (cups/day) 0.0016 0.78%
8 Gender 0.0009 0.45%
9 BMI 0.0008 0.37%
10 Lifetime intensity of coffee consumption (cups/day) 0.0006 0.28%
11 Other types of tea load (cup-years) 0.0003 0.16%
12 Duration of other tea consumption 0.0001 0.06%
13 Lifetime intensity of other tea consumption (cups/day) 0.0000 0.00%
14 Tea consumption status 0.0000 0.00%
15 Green tea load (cup-years) 0.0000 0.00%
16 Tea intensity at interview 0.0000 0.00%
17 Duration of green tea consumption 0.0000 0.00%
18 Coffee load (cup-years) 0.0000 0.00%
19 Coffee consumption status 0.0000 0.00%
20 Riluzole 0.0000 0.00%
21 Current smokers 0.0000 0.00%
22 Coffee intensity at interview 0.0000 0.00%

Variables are ranked from the most to the less important (rank).

VIMP is the sum of the decrease in prediction (of log-ΔFS) error values when a tree split by that variable whereas RVIMP is the VIMP divided by the highest VIMP value so that values are bounded between 0 and 1 (or between 0 and 100%).