Choosing marginal or random-effects models for longitudinal binary responses: application to self-reported disability among older persons

Table 6 Marginal model and random-effects models analysing time-fixed and time varying covariates

	Marginal model* GEE			Random model* likelihood integration
	Estimate	SE	p value	Estimate	SE	p value
Baseline
Living alone	-0.31	0.09	0.0004	-0.66	0.18	0.0002
BMI (kg/m²)
< 25 (ref)	0			0
25 – 29	0.30	0.10	0.002	0.69	0.20	0.0005
≥ 29	0.75	0.12	< 0.0001	1.46	0.23	< 0.0001
Visual acuity
≤ 2/10	0.64	0.20	0.001	1.21	0.35	0.0006
> 2/10 (ref)	0			0
Perceived health
Bad or very bad	0.76	0.15	< 0.0001	1.43	0.28	< 0.0001
Follow-up
Hospitalised (past year)	0.26	0.05	< 0.0001	0.49	0.11	< 0.0001
Temporarily bed confined (past year)	0.39	0.07	< 0.0001	0.73	0.15	< 0.0001
Number of falls ≥ 2 (past year)	0.23	0.08	0.004	0.49	0.15	0.0013

*adjusted on time since inclusion, age at inclusion, age × time, baseline disability and death

ISSN: 1471-2288