Table 3.

Causal effects^a of BMI on FEV1/FVC and FEF2575 in predictive and in long-term cross-sectional models.

	N	βc1,βc2	SE	p-value
FEV1/FVC
Predictive model
BMI main effect log(BMIs1s2) → FEV1/FVCs2,s3	2,853	−0.468	0.455	0.300
BMI*Age interaction effect log(BMIs1s2):Ages1s2 → FEV1/FVCs2,s3		0.011	0.015	0.490
Long-term cross-sectional model
BMI main effect log(BMIs1s2s3) → FEV1/FVCs1,s2,s3	2,731	−0.479	0.565	0.400
BMI*Age interaction effect log(BMIs1s2s3):Ages1,s2,s3 → FEV1/FVCs1,s2,s3		0.006	0.016	0.730
FEF2575
Predictive model
BMI main effect log(BMIs1s2) → FEF2575s2,s3	2,850	−9.932	6.622	0.134
BMI*Age interaction effect log(BMIs1s2):Ages1s2 → FEF2575s2,s3		0.269	0.225	0.231
Long-term cross-sectional model
BMI main effect log(BMIs,1s2,s3) → FEF2575s1,s2,s3	2,728	−9.111	8.117	0.262
BMI*Age interaction effect log(BMIs1,s2,s3):Ages1,s2,s3 → FEF2575s1,s2,s3		0.163	0.234	0.486

Childhood BMI genetic score: (Felix; 12 SNPs).

^a${\beta }_{c1}$, causal BMI main effect per one BMI-increasing allele; β_c2, causal BMI^*Age interaction effect per one BMI-increasing allele.

The negative sign of the causal main effect means that, keeping all other predictors fixed, at age 18 (which has been chosen as the origin in our analysis) BMI has a causal negative effect on LF. The positive sign of the Age × BMI causal interactive effect implies that, as age increases, the detrimental effect of BMI on LF decreases. As a consequence, the total effect of BMI becomes null at middle ages and protective at older ages.