Skip to main content

Table 4 Comparison of Monte Carlo standard error, mean squared error, model and empirical standard error using non-response imputation, impute-before-dichotomizing and dichotomize-then-impute multiple imputation1

From: Imputation strategies when a continuous outcome is to be dichotomized for responder analysis: a simulation study

Dropout model

Imputation method

MCSE

MSE

SEmod

SEemp

1: Lack of efficacy

NRI

0.12

40.84

4.46

4.65

DTI MI

0.13

28.94

5.99

5.31

DTI MI with CV

0.13

27.31

5.75

5.23

IBD MI

0.14

29.26

6.14

5.41

IBD MI with CV

0.13

28.27

5.84

5.25

2: Differing mechanism

NRI

0.12

50.99

4.66

4.72

DTI MI

0.14

31.11

6.10

5.49

DTI MI with CV

0.13

29.10

5.84

5.39

IBD MI

0.14

30.03

6.14

5.48

IBD MI with CV

0.13

29.70

5.93

5.38

3: Differing mechanism, reversed

NRI

0.12

35.34

4.60

4.72

DTI MI

0.14

31.63

6.03

5.60

DTI MI with CV

0.14

30.36

5.72

5.49

IBD MI

0.14

31.39

6.17

5.60

IBD MI with CV

0.14

30.30

5.95

5.45

4: Differential dropout rates

NRI

0.12

24.27

4.67

4.76

DTI MI

0.14

31.26

6.00

5.59

DTI MI with CV

0.13

29.13

5.69

5.36

IBD MI

0.13

28.54

6.10

5.34

IBD MI with CV

0.13

27.13

5.79

5.20

5: Lack of efficacy, sensitivity of mechanism

NRI

0.12

48.88

4.41

4.63

DTI MI

0.14

31.97

6.13

5.56

DTI MI with CV

0.14

29.59

5.89

5.44

IBD MI

0.14

31.04

6.23

5.57

IBD MI with CV

0.14

30.00

6.01

5.42

6: Lack of efficacy and tolerability

NRI

0.12

39.14

4.58

4.75

DTI MI

0.14

30.73

6.00

5.50

DTI MI with CV

0.13

28.74

5.73

5.36

DTI MI with AE

0.14

32.73

5.84

5.70

IBD MI

0.14

31.44

5.94

5.61

IBD MI with CV

0.14

30.13

5.73

5.43

IBD MI with AE

0.14

31.43

5.91

5.60

  1. MCSE: Monte Carlo standard error; MSE: Mean squared error; SEmod: Average standard error of the risk difference; SEemp: Empirical standard error of the risk difference; NRI: Non-response imputation; DTI MI: Dichotomize then impute multiple imputation; IBD MI: Impute before dichotomizing multiple imputation; CV: Correlated variable; AE: Adverse event status
  2. 1 Results are from a linear response profile with 30% data missing at random, N = 200. In fully observed data, % responders in Treatment A and B was 25.6 and 10.6, respectively for a difference of 15.0 and power = 0.80