Is using multiple imputation better than complete case analysis for estimating a prevalence (risk) difference in randomized controlled trials when binary outcome observations are missing?

Table 1 MAR and efficacy rate 85 % versus 60 % (RD 0.250): estimated efficacy differences, coverage and bias for 5 %, 15 % and 30 % averages of 5000 simulated data sets, 50 imputations

Model	RD (RMSE)	Coverage	Bias
All outcomes recorded:	0.250 (0.061)	0.950	0.000
5 % of outcomes missing:
CC	0.250 (0.063)	0.946	0.000
MI: wt, hb, age, para	0.237 (0.063)	0.958	-0.013
MI: hb, age, para	0.238 (0.063)	0.961	-0.012
MI: hb, age, para, group	0.249 (0.062)	0.949	-0.001
MI: wt, hb, age, para, group	0.248 (0.062)	0.945	-0.002
15 % of outcomes missing
CC	0.250 (0.066)	0.945	0.000
MI: wt, hb, age, para	0.212 (0.066)	0.944	-0.038
MI: hb, age, para	0.214 (0.066)	0.947	-0.036
MI: hb, age, para, group	0.249 (0.066)	0.948	-0.001
MI: wt, hb, age, para, group	0.247 (0.066)	0.949	0.003
30 % of outcomes missing
CC	0.250 (0.073)	0.948	0.000
MI: wt, hb, age, para	0.175 (0.071)	0.887	-0.075
MI: hb, age, para	0.174 (0.071)	0.879	-0.076
MI: hb, age, para, group	0.248 (0.072)	0.946	-0.002
MI: wt, hb, age, para, group	0.249 (0.072)	0.946	-0.001

ISSN: 1745-6215