On the use of the outcome variable “small for gestational age” when gestational age is a potential mediator: a maternal asthma perspective

Table 6 Confounding scenarios for indirect ICS effect on BW with heterogeneity of errors

Scenario (model, interpretation of effect)	mean	se	null ICS effect?
Confounding 1 (M₁, total)	−513.28	0.2388	no
Confounding 1 (M_1,V, total)	−366.76	0.2579	no
Confounding 1 (M₂, direct)	−0.2363	0.2328	yes
Confounding 1 (M_2,V, direct)	−0.2306	0.2353	yes
Confounding 1 (M₃, total)	1.5614	0.0015	no
Confounding 1 (M_3,V, total)	1.1601	0.0016	no
Confounding 1 (M₄, direct)	−0.0011	0.0015	yes
Confounding 1 (M_4,V, direct)	−0.0001	0.0016	yes
Confounding 2 (M₁, total)	−366.46	0.2600	no
Confounding 2 (M_1,V, total)	−366.58	0.2312	no
Confounding 2 (M₂, direct)	24.188	0.2168	no
Confounding 2 (M_2,V, direct)	−0.1823	0.2167	yes
Confounding 2 (M₃, total)	1.0759	0.0012	no
Confounding 2 (M_3,V, total)	1.1475	0.0012	no
Confounding 2 (M₄, direct)	0.0125	0.0013	no
Confounding 2 (M_4,V, direct)	0.0127	0.0013	no
Confounding 3 (M₁, total)	−406.71	0.2335	no
Confounding 3 (M_1,V, total)	−366.76	0.2593	no
Confounding 3 (M₂, direct)	−40.210	0.2226	no
Confounding 3 (M_2,V, direct)	−0.2394	0.2378	yes
Confounding 3 (M₃, total)	1.3095	0.0017	no
Confounding 3 (M_3,V, total)	1.1839	0.0019	no
Confounding 3 (M₄, direct)	0.1738	0.0014	no
Confounding 3 (M_4,V, direct)	0.0050	0.0016	no
Confounding 4 (M₁, total)	−553.28	0.2444	no
Confounding 4 (M_1,V, total)	−366.76	0.2579	no
Confounding 4 (M₂, direct)	−10.454	0.2361	no
Confounding 4 (M_2,V, direct)	−0.2306	0.2353	yes
Confounding 4 (M₃, total)	1.6301	0.0013	no
Confounding 4 (M_3,V, total)	1.1518	0.0014	no
Confounding 4 (M₄, direct)	0.1859	0.0014	no
Confounding 4 (M_4,V, direct)	0.0120	0.0015	no

Legend. Mean of the estimated exposure (ICS) effect (mean difference or log odds ratio) on the outcome (BW, LBW or SGA) based on 1000 samples of size 20,000, with Monte Carlo standard error (se)
Confounding Scenario 1: confounder between ICS and GA; Confounding Scenario 2: confounder between GA and BW; Confounding Scenario 3: confounder between ICS and BW; Confounding Scenario 4: confounder between ICS, GA and BW
M₁: BW ~ ICS; M₂: BW ~ ICS + GA; M₃: LBW ~ ICS; M₄: SGA ~ ICS;
null ICS effect = yes if 0 ∈ mean ± 1.96se
Models subscripted by V additionally include the V variable in the regression model
Abbreviations: BW birth weight, GA gestational age, ICS inhaled corticosteroids, LBW low birth weight, SGA small for gestational age

ISSN: 1471-2288