Single time point comparisons in longitudinal randomized controlled trials: power and bias in the presence of missing data

Table 1 Comparison of t-test, mixed model for repeated measures with compound symmetric variance-covariance, and mixed model for repeated measures with unstructured variance-covariance, with respect to bias percent and power; simulation results for linear trajectory

	ρ = 0.7		ρ = 0.5		ρ = 0.3
	(N = 10,000)		(N = 10,000)		(N = 10,000)
	Bias %	Power	Bias %	Power	Bias %	Power
Complete
t-test^a	0	80	0	80	0	80
MMRM-CS^b	0	80	0	81	0	80
MMRM-UN^c	0	80	0	80	0	80
MCAR with equal dropout of 40 % in each group
t-test	0	58	0	58	0	58
MMRM-CS	0	70	0	65	0	61
MMRM-UN	0	70	0	65	0	60
MAR with unequal dropout of 30 % and 50 % in each group, one reason
t-test	−15	44	−11	48	−6	51
MMRM-CS	0	69	0	64	0	60
MMRM-UN	0	69	0	64	0	59
MAR with unequal dropout of 30 % and 50 % in each group, two reasons
t-test	−5	53	−3	53	−2	54
MMRM-CS	0	68	0	63	0	59
MMRM-UN	0	68	0	62	0	59
MAR with equal dropout of 40 % in each group
t-test	−1	57	0	57	−1	56
MMRM-CS	0	69	0	64	0	59
MMRM-UN	0	69	0	64	0	59
MNAR with unequal dropout of 30 % and 50 % in each group, one reason
t-test	−18	42	−16	43	−15	44
MMRM-CS	−6	64	−9	56	−13	49
MMRM-UN	−6	64	−9	56	−13	49
MNAR with unequal dropout of 30 % and 50 % in each group, two reasons
t-test	−7	52	−6	51	−6	51
MMRM-CS	−2	67	−4	60	−5	54
MMRM-UN	−2	67	−4	60	−5	55
MNAR with equal dropout of 40 % in each group
t-test	−3	55	−3	55	−3	55
MMRM-CS	−1	69	−2	63	−2	58
MMRM-UN	−1	69	−2	63	−2	58

^aIndependent two-sample t-test for the difference between group means at the final time point
^bMixed model for repeated measures, compound symmetric variance-covariance matrix, contrast between group means at the final time point
^cMixed model for repeated measures, unstructured variance-covariance matrix, contrast between group means at the final time point

ISSN: 1471-2288