The Use of Mixed Models for the Analysis of Mediated Data with Time-Dependent Predictors

Table 5

The total effect of heavy alcohol consumption on CD4 cell count from a prospective cohort study of HIV-infected subjects on antiretroviral therapy (𝑛=319) [29]. Longitudinal regression analyses were performed using linear mixed models and structural equation models, and adjusted mean differences (SE) are reported.

SEM				LMM
Time point	Constant effect	Delayed effect	Unrestricted	Constant effect	Delayed effect	Time interaction	Unrestricted
𝑡1	−3.7 (9.6)	0.41 (11.6)	−4.8 (19.6)	−3.0 (11.3)	−2.3 (13.8)	0.44 (22.0)	−7.7 (23.9)
𝑡2	−3.7 (9.6)	0.41 (11.6)	0.13 (20.1)	−3.0 (11.3)	−2.3 (13.8)	−0.41 (18.1)	6.2 (24.1)
𝑡3	−3.7 (9.6)	0.41 (11.6)	−1.4 (20.4)	−3.0 (11.3)	−2.3 (13.8)	−1.3 (14.7)	6.9 (23.7)
𝑡4	−3.7 (9.6)	0.41 (11.6)	5.7 (18.9)	−3.0 (11.3)	−2.3 (13.8)	−2.1 (12.3)	0.68 (22.3)
𝑡5	−3.7 (9.6)	−10.3 (14.0)	−3.4 (20.2)	−3.0 (11.3)	−4.1 (16.5)	−3.0 (11.3)	−4.6 (24.1)
𝑡6	−3.7 (9.6)	−10.3 (14.0)	3.8 (22.0)	−3.0 (11.3)	−4.1 (16.5)	−3.8 (12.3)	9.5 (25.1)
𝑡7	−3.7 (9.6)	−10.3 (14.0)	−13.1 (25.9)	−3.0 (11.3)	−4.1 (16.5)	−4.7 (14.8)	−13.6 (30.3)
𝑡8	−3.7 (9.6)	−10.3 (14.0)	−41.8 (31.6)	−3.0 (11.3)	−4.1 (16.5)	−5.5 (18.2)	−15.8 (35.1)