Table 2.
Multiple-exposure elastic net penalized regression modelsa (βEN) for term birth weight.
| Potential predictor (increment) | Adjusted | Plus gestational age | Further adjusted |
|---|---|---|---|
| ln-MEHHP (1.70 ng/mL) | –64.67b | –59.43b | –48.61 |
| ln-MEOHP (1.29 ng/mL) | –0.15 | 0 | 0 |
| ln-MECPP (1.42 ng/mL) | 0 | 0 | 0 |
| ln-MHiNP (2.74 ng/mL) | 0 | 0 | 0 |
| ln-MOiNP (2.22 ng/mL) | 23.81 | 22.26 | 16.31 |
| ln-MCiOP (2.32 ng/mL) | 0 | 0 | 0 |
| ln-PFHxS (1.24 ng/mL) | –3.49 | 0 | 0 |
| ln-PFHpA (1.84 ng/mL) | 0 | 0 | 0 |
| ln-PFOS (1.60 ng/mL) | 0 | 0 | 0 |
| ln-PFOA (1.18 ng/mL) | –11.51 | –10.11 | –38.82 |
| ln-PFNA (1.03 ng/mL) | –7.05 | –7.69 | 0 |
| ln-PFDA (1.40 ng/mL) | 0 | 0 | 0 |
| ln-PFUnDA (2.10 ng/mL) | 0 | 0 | 0 |
| ln-PFDoDA (1.67 ng/mL) | –22.56 | 0 | 0 |
| ln-PCB-153 (2.43 ng/g) | 0 | 0 | 0 |
| ln-p,p´-DDE (1.82 ng/g) | –106.39b | –76.63b | –47.02b |
| Regression coefficients (βEN) represent the change in mean birth weight (g) for term infants per increment: a 2-SD increase in ln-transformed exposure biomarker levels. βEN for the modeled, unpenalized covariates are not shown. aThe cross-validated optimum penalization was α = 1.00, λ = 3.32 (MSE = 205,061) for the adjusted model (minimal sufficient set: study population, maternal age, prepregnancy BMI, and parity); α = 1.00, λ = 3.32 (MSE = 177,179) for the model additionally adjusted for gestational age; and α = 0.98, λ = 2.46 (MSE = 166,329) for the further adjusted model (plus infant sex, maternal height, alcohol consumption, serum cotinine, and vitamin D). All models, n = 1,250. bCovariance test (Lockhart et al. 2014) p < 0.05: p-values were 0.002, < 0.001, and 0.345 for MEHHP and < 0.001, 0.017, 0.046 for p,p´-DDE in the minimally adjusted, plus gestational age, and further adjusted models, respectively. p = 0.071 for PFOA in the further adjusted model; all other p-values were > 0.10. | |||