Evaluating the performance of propensity score matching based approaches in individual patient data meta-analysis

Table 7 Simulation results: Statistical Power and type I error of log(OR) by PSM-based approaches and data generation features

			Treatment Prevalence According to Study				50% Treatment Prevalence				30% Treatment Prevalence
OR	SD ^c	SD ^b	Across	Preferential	Random	Within	Across	Preferential	Random	Within	Across	Preferential	Random	Within
	(σ_1j)	(σ_0j)	Study	Study	Effect	Study	Study	Study	Effect	Study	Study	Study	Effect	Study
3	0.04	0.7	100	95	83	90	100	100	100	100	100	100	100	100
		1.4	100	89	79	84	100	99	100	99	100	99	99	99
		2.1	99	83	72	73	100	97	98	95	100	92	94	91
	0.08	0.7	100	95	83	90	100	100	100	100	100	100	100	100
		1.4	100	88	79	84	100	99	100	99	100	100	99	99
		2.1	98	83	72	72	100	96	98	94	100	91	94	90
1.5	0.04	0.7	64	50	22	47	92	80	69	80	90	77	72	76
		1.4	60	38	29	37	87	60	49	56	90	49	48	57
		2.1	53	27	22	25	88	32	29	24	91	30	28	25
	0.08	0.7	63	49	21	46	92	79	68	77	89	77	70	75
		1.4	59	36	27	36	86	58	47	53	89	60	46	55
		2.1	52	27	21	24	87	29	26	22	90	28	26	23

OR indicates pooled odds ratio used for simulation
^cSD is the standard deviation, σ_1j of study specific random slope, u_1j
^bSD is the standard deviation, σ_0j, of study specific random intercept, u_0j
Across-study indicates consideration of Single-level logit model with across-study matching
Prefer-study indicates consideration Single-level logit model with preferential-within study matching
Random-effect indicates consideration of Random-effect logit model with across-study matching
Within-study indicates consideration of Single-level logit model with within-study matching

ISSN: 1471-2288