A comparison of methods to estimate the survivor average causal effect in the presence of missing data: a simulation study

Table 1 Log odds ratio estimates from simulation study

OR_UY	OR_UZ	Estimation method	Monotonicity
			Valid					Violated
			Estimate^a	SE	SB	MSE	Coverage^b	Estimate^a	SE	SB	MSE	Coverage^b
0.5	0.5	Average τ	0.92	0.09				0.92	0.09
		Marginal structural model	−0.54	0.12	−29	0.01	91.6	− 0.53	0.11	−14	0.01	94.6
		Sensitivity analysis
		SP = 0.5	−0.34	0.11	164	0.04		−0.42	0.10	92	0.02
		SP = 1	−0.54	0.11	−26	0.01		−0.52	0.10	−4	0.01
		SP = 2	−0.84	0.11	− 305	0.12		−0.67	0.10	− 159	0.04
0.5	2.0	Average τ	1.10	0.11				1.10	0.11
		Marginal structural model	−0.48	0.11	27	0.01	93.6	−0.50	0.10	11	0.01	92.4
		Sensitivity analysis
		SP = 0.5	−0.33	0.10	176	0.04		−0.42	0.10	95	0.02
		SP = 1	−0.48	0.10	30	0.01		−0.49	0.10	20	0.01
		SP = 2	−0.71	0.10	− 197	0.05		−0.61	0.10	− 106	0.02
2.0	0.5	Average τ	1.12	0.09				1.12	0.09
		Marginal structural model	−0.46	0.10	54	0.01	90.4	−0.49	0.09	23	0.01	92.6
		Sensitivity analysis
		SP = 0.5	−0.26	0.09	266	0.07		−0.38	0.09	143	0.02
		SP = 1	−0.47	0.09	46	0.01		−0.48	0.09	33	0.01
		SP = 2	−0.76	0.09	− 271	0.07		−0.64	0.09	− 140	0.02
2.0	2.0	Average τ	0.90	0.08				0.89	0.07
		Marginal structural model	−0.53	0.09	− 25	0.01	92.9	−0.52	0.08	−9	0.01	94.3
		Sensitivity analysis
		SP = 0.5	−0.36	0.08	183	0.03		−0.43	0.08	104	0.01
		SP = 1	−0.52	0.08	−7	0.01		−0.50	0.08	9	0.01
		SP = 2	−0.75	0.08	− 290	0.06		−0.62	0.08	− 144	0.02

^aEstimates of the log odds ratio have been averaged over 1200 simulated datasets from each scenario
^bCoverage indicates the percentage of datasets in each scenario where the true value of the SACE was within the bias-corrected bootstrap confidence interval of the marginal structural model
MSE Mean square error, SACE Survivor average causal effect, SB Standardized bias as a percentage, SE Empirical standard error, SP Sensitivity parameter.
OR_UY is the odds ratio effect of U on the outcome. OR_UZ is the odds ratio effect of U on survival. τ is the ratio of the odds of the outcome following high iron intake between compliant-survivors and always-survivors. True SACE log odds ratio = ln(0.6) = − 0.511.

ISSN: 1471-2288