Skip to main content
. 2023 Feb 3;13:1981. doi: 10.1038/s41598-023-27568-6

Table 5.

Variable importance for the predictive models (frequent users 3).

Variable LR GBM NN RF1 RF2
Age
18–34 Ref Ref Ref 562.0 271.1
35–54 7.0 5.1E−4 3.5
55–64 9.0 1.1E−3 2.7
65–74 7.5 4.4E−4 3.6
75–84 2.5 4.0E−3 2.8
 ≥ 85 3.2 8.5E−3 3.7
PPDIP
Regular Ref Ref Ref 465.6 226.3
GIS 13.2 7.5E−3 4.2
Not admissible 8.5 1.1E−2 3.9
LRFA 17.0 1.4E−2 4.7
CCI
0 Ref Ref Ref 882.1 439.6
1–2 19.8 1.7E−2 4.1
3–4 21.3 1.3E−2 5.0
 ≥ 5 24.7 1.6E−2 4.3
PV
 ≤ 1 Ref Ref Ref 5255.2 2621.2
2–3 74.3 0.2 3.7
4 66.2 0.1 5.3
5 61.7 0.1 6.0
 ≥ 6 96.4 0.3 5.3
COPD 24.4 2.6E−2 5.1 475.9 234.7
Injury 5.1 2.2E−3 3.7 205.3 98.3
SMD 5.4 2.2E−3 4.8 182.0 88.5
CMD 12.0 1.8E−2 2.9 266.9 130.8
CNCP 11.6 5.7E−3 3.3 180.1 86.7
Alcohol 4.3 2.0E−3 5.3 164.6 79.7
Drugs 8.1 2.9E−3 4.8 228.4 111.2

The higher the value, the higher the importance (relative to a model).

LR logistic regression, GBM gradient boosting machine, NN neural network, RF random forests (1: binary outcome, 2: continuous outcome).