Table 2.
Imputation errors of different methods in HT-data
| Skewness or proportion of minority class | MICE | missForest | GAIN | |
|---|---|---|---|---|
| Missingness rate = 20% | ||||
| Continuous variables | ||||
| Age, years | −0.018 | 0.063 ± 0.002 | 0.049 ± 0.001 a,b | 0.057 ± 0.004 a |
| SBP | 0.492 | 0.075 ± 0.001 | 0.058 ± 0.000 a | 0.048 ± 0.000 a,c |
| Charlson index | 0.146 | 0.154 ± 0.002 | 0.121 ± 0.001 a,b | 0.144 ± 0.003 a |
| TC/HDL-C ratio | 3.139 | 0.175 ± 0.003 | 0.137 ± 0.001 a | 0.115 ± 0.001 a,c |
| Hospital admission times | 7.037 | 2.379 ± 0.069 | 1.885 ± 0.042 a | 1.752 ± 0.141 a,c |
| Categorical variables | ||||
| Smoking | 7.45% | 0.133 ± 0.007 | 0.123 ± 0.003 a | 0.098 ± 0.010 a,c |
| Hypertensive drugs | 8.10% | 0.149 ± 0.006 | 0.126 ± 0.003 a | 0.098 ± 0.002 a,c |
| Lipid Lowering drugs | 9.99% | 0.173 ± 0.007 | 0.159 ± 0.003 a | 0.129 ± 0.006 a,c |
| Overweight | 37.89% | 0.433 ± 0.01 | 0.400 ± 0.005 a | 0.359 ± 0.003 a,c |
| Sex | 41.21% | 0.448 ± 0.019 | 0.412 ± 0.004 a | 0.405 ± 0.022 a |
| Missingness rate = 50% | ||||
| Continuous variables | ||||
| Age, years | −0.018 | 0.129 ± 0.002 | 0.102 ± 0.001 a | 0.094 ± 0.007 a,c |
| SBP | 0.492 | 0.115 ± 0.001 | 0.095 ± 0.001 a | 0.080 ± 0.002 a |
| Charlson index | 0.146 | 0.295 ± 0.001 | 0.239 ± 0.002 a | 0.241 ± 0.009 a |
| TC/HDL-C ratio | 3.139 | 0.279 ± 0.004 | 0.235 ± 0.003 a | 0.183 ± 0.002 a,c |
| Hospital admission times | 7.037 | 3.766 ± 0.12 | 3.199 ± 0.057 a | 3.004 ± 0.246 a,c |
| Categorical variables | ||||
| Smoking | 7.45% | 0.335 ± 0.006 | 0.277 ± 0.015 a | 0.267 ± 0.012 a,c |
| Hypertensive drugs | 8.10% | 0.368 ± 0.014 | 0.305 ± 0.004 a | 0.276 ± 0.005 a,c |
| Lipid Lowering drugs | 9.99% | 0.441 ± 0.015 | 0.319 ± 0.006 a | 0.304 ± 0.009 a,c |
| Overweight | 37.89% | 1.135 ± 0.018 | 1.029 ± 0.019 a | 0.850 ± 0.020 a,c |
| Sex | 41.21% | 1.149 ± 0.02 | 1.050 ± 0.013 a | 1.007 ± 0.055 a |
Notes
SBP Systolic Blood Pressure, TC Total Cholesterol, HDL-C High-Density Lipoprotein Cholesterol
Since NRMSE and PFC both followed normal distribution (Shapiro-Wilk normality test p value > 0.05), imputation errors of different methods were compared using one-way ANOVA. If p < 0.05, paired methods were compared using independent sample t-test;
aThe mean imputation error is significantly lower than that of MICE (p < 0.05)
bThe mean imputation error is significantly lower than that of GAIN (p < 0.05)
cThe mean imputation error is significantly lower than that of missForest (p < 0.05)