Mining leachate contamination and subfecundity among women living near the United States-Mexico border

Design and sample
We conducted a retrospective cohort study on pregnancy cases of women living across the watershed area of the Sonora River, in those municipalities directly affected by the environmental disaster: Arizpe, Banámichi, Huépac, San Felipe de Jesús, Aconchi, Baviácora, Ures, and some villages belonging to the state capital Hermosillo (Molino de Camou, Topahue, San Francisco de Batuc, El Tronconal and San Pedro El Saucito) (Lammers., 2014). In addition, we included Magdalena de Kino as a non-exposed community, whose water supply source is different from the resting municipalities, but shares common climate and ecological characteristics (Figure 1).
Participants were first identified by the local health centers and then visited at home, where they were asked to participate before signing the informed consent. Inclusion criteria were women being older than 18 years, planned pregnancy, conception of pregnancy in municipalities considered as exposed and non-exposed in this study after the disaster or before up to 10 years prior, with the possibility of including more than one pregnancy for each participant. Among the criteria for exclusion were having a physical or mental disability to adequately respond the questionnaire or refused to sign the consent. The study sample size had enough statistical power according to the Freedman table (Freedman., 1982), taking into account that previous studies exploring fecundity and metal exposure worked with samples between 41 and 300 pregnancies (Snijders., 2012). The study was approved by either ethical review board of the University of Sonora, Mexico, and the Industrial University of Santander, Colombia.
Statistical methods
Univariate and bivariate descriptive analysis were made after completing the database, using chi-square, Student’s t, and Mann-Whitney U to test for differences between variables. Fecundability odds ratios (fOR) obtained from discrete time analogue of Cox’s proportional hazard models were used to estimate associations with logistic regression models using the macro dthaz (Dinno., 2011). For multiple analysis, variables included in the model had p<0.25 in the bivariate Wald-test or had effect changes > 20%. Models were adjusted for age at pregnancy, age at first pregnancy, pregnancy order, maternal occupation, father’s occupation, father’s age, and father history of hyperthyroidism.
Finally, for each approach to measure exposure (acute, acute plus non-affected zone, and pollution gradient), we developed three models: model 1 included all pregnancies with TTP up to 12 months; model 2 excluded pregnancies with TTP = 1; whereas model 3 only included first time pregnancies in order to control for obstetric history of participants (Idrovo et al., 2005; Weinberg et al., 1994 ). All assumptions and adjustments were evaluated, and Stata 14 was the statistical software used to carry out all the analysis (Stata Corporation, College Station, USA).