Accuracy of random-forest-based imputation of missing data in the presence of non-normality, non-linearity, and interaction

Table 1 Parameter estimates of the regression from the case study

Method	Intercept		Regression coefficient for AFP		Regression coefficient for tumor diameter
Method	Estimate (95% CI)	p-value	Estimate (95% CI)	p-value	Estimate (95% CI)	p-value
Original data	0.54 (0.43, 0.64)	< 0.001	−0.20 (− 0.31, − 0.10)	< 0.001	− 0.65 (− 0.76,-0.54)	< 0.001
MAR data
PMM	0.55 (0.45, 0.65)	< 0.001	− 0.19 (− 0.29, − 0.09)	< 0.001	−0.66 (− 0.77, − 0.56)	< 0.001
missForest	0.54 (0.44, 0.64)	< 0.001	−0.17 (− 0.28, − 0.07)	0.002	−0.74 (− 0.85, − 0.62)	< 0.001
CALIBERrfimpute	0.54 (0.44, 0.64)	< 0.001	−0.19 (− 0.29, − 0.09)	< 0.001	−0.66 (− 0.77, − 0.56)	< 0.001
MCAR data
PMM	0.54 (0.44, 0.64)	< 0.001	−0.20 (− 0.31, − 0.10)	< 0.001	−0.66 (− 0.77, − 0.56)	< 0.001
missForest	0.54 (0.44, 0.64)	< 0.001	−0.21 (− 0.32, − 0.10)	< 0.001	−0.72 (− 0.84, − 0.61)	< 0.001
CALIBERrfimpute	0.55 (0.45, 0.65)	< 0.001	−0.22 (− 0.33, − 0.12)	< 0.001	−0.68 (− 0.79, − 0.58)	< 0.001

*Results were obtained from logistic regression
CI confidence interval, MCAR missing completely at random, MAR missing at random

ISSN: 1471-2288