Skip to main content

Table 3 Variable selection results by each of 3 methods, with the best imputation method and threshold value of π for BI-BART and BI-XGB suggested in simulations. BART was used with missForest and π=0.1, XGB with missForest and π=0.3. Definitions of the variable names appear in Web Table 1. RR-BART and BI-BART both selected 17 variables, and BI-XGB selected 16 variables

From: A flexible approach for variable selection in large-scale healthcare database studies with missing covariate and outcome data

Variables RR-BART BI-BART BI-XGB
TRIGRES Yes Yes Yes
SYSBP Yes Yes Yes
LPA Yes Yes Yes
DIABP Yes Yes Yes
WAIST Yes Yes Yes
INSULIN Yes No Yes
LUCRES Yes No No
APOARES Yes Yes Yes
EDUCATION Yes No Yes
BP Yes Yes No
WHRATIO Yes Yes Yes
BMI Yes Yes Yes
RACE Yes Yes No
TPA Yes Yes Yes
DTTLIN Yes Yes No
SHBG Yes Yes Yes
RESTLES Yes Yes No
GLUCOSE No Yes Yes
T No Yes No
E2AVE No No Yes
PAI1 No No Yes