*2.4. Statistical Analysis*

We used a linear regression to assess the association between elevated cadmium exposure during pregnancy and birth weight, and a logistic regression to assess the association between elevated cadmium exposure and preterm birth. In our study, cadmium was severely right-skewed. Despite log transformation, the skewness did not improve (results using the continuous log Cd variables are available in Supplemental Tables S1 and S2). As cadmium exposure is ubiquitous, we assessed cadmium in quartiles and defined high cadmium exposure as having a cadmium blood level in the highest quartile (mean Cd (ng/g) per quartile: 25th percentile: 0.12, 50th percentile: 0.24, 75th percentile: 0.46). We identified potential confounders a priori based on the literature and substantive knowledge and selected a final set of confounders using Bayesian Information Criteria (BIC). We included prepregnancy BMI, smoking during pregnancy, and sex of the infant as confounders in our final models. In our analysis of birth weight, we also included gestational age as a confounder and in our analysis of preterm birth we included birth weight as a confounder in the model. We assessed e ffect measure modification by Mediterranean diet adherence by including an interaction term in our models and by stratification of high and low maternal Mediterranean diet adherence. E ffect measure modification analyses are limited to the 185 women who completed first trimester FFQs and had cadmium measures available. We conducted supplemental analysis to explore possible changes in the association between prenatal cadmium exposure and birth outcomes using di fferent cut points for high and low Mediterranean diet adherence. Additionally, we explored the association between prenatal cadmium exposure and (1) infant Apgar score, (2) infant ponderal index at birth.
