Table 4.
Variable importance (VIMP) and relative variable importance (RVIMP) values from conditional random forest algorithm (100,000 trees) of each candidate’s clinical, demographical, pathological, treatment, and smoking/alcohol consumption variables for explaining the variability of the log(ΔFS) values. Variables are ranked from the most to the least important (rank).
| Variable | Conditional VIMP |
Conditional RVIMP |
|---|---|---|
| Diagnostic delay | 0.6302 | 100.0% |
| Age at onset | 0.1680 | 26.7% |
| El Escorial classification | 0.0413 | 6.6% |
| Education | 0.0278 | 4.4% |
| Site of onset | 0.0072 | 1.1% |
| Alcohol load (drink-years) | 0.0043 | 0.7% |
| Alcohol intensity (drinks/day) | 0.0043 | 0.7% |
| Smoking intensity (cigarettes/day) | 0.0016 | 0.3% |
| Country | 0.0014 | 0.2% |
| Riluzole | 0.0007 | 0.1% |
| Alcohol duration | 0.0005 | 0.1% |
| Smoking load (pack-years) | 0.0002 | 0.0% |
| BMI | 0.0000 | 0.0% |
| Smoking duration | 0.0000 | 0.0% |
| Alcohol drinking status | 0.0000 | 0.0% |
| Smoking status | 0.0000 | 0.0% |
| Gender | 0.0000 | 0.0% |
The VIMP of a specific variable is the sum of the decrease in prediction error values (of log(ΔFS)) when a tree of the forest splits due to that variable, whereas RVIMP is the VIMP divided by the highest VIMP value such that values are bounded between 0 and 1 (or between 0 and 100%).