Statistical analysis
Statistical analyses were performed using SAS software package version 9.4. Standard descriptive statistics were used to describe the study population: continuous variables were described using mean and standard deviation (SD), whereas categorical variables were presented as percentages. Proportions of sociodemographic and reproductive health factors in Groups 1 and 2, as well as the variation between women who conceived and those who did not conceive within the last 12 months, were calculated using Pearson’s chi-square test and Fisher’s exact test, as appropriate.
In our study, TTP represented the elapsed time from attempting to conceive to confirmation of pregnancy among women who conceived within the last 12 months (retrospective design) or the time from attempting to conceive to the investigation time among women who had not still conceived (cross-sectional design). To estimate a TTP distribution, we used Kaplan–Meier curves for both populations, which could provide an estimation of TTP and allow censors for reason other than conception. To set up a better model, we censored observations longer than 120 months because of the small numbers as 95% extreme values. Cox regressions were subsequently applied to estimate the fecundability odds ratios (FORs) associated with sociodemographic and reproductive factors for TTP by both unadjusted and adjusted models. FORs estimate the odds of becoming pregnant in a woman’s current cycle, conditional on not being pregnant in the previous cycle.(8) An FOR <1 indicates reduced fecundity or, equivalently, a longer TTP, whereas an FOR >1 represents a shorter TTP. We used the expected low-risk group in a regression model as the reference group according to previous literatures or clinical consensus, and results were presented with odds ratios and 95% confidence intervals (CI) around the point estimate. Models were first run to estimate the crude FORs, and multivariable models were then fitted to adjust for potential risk factors for fecundability. Adjustments included age, ethnicity, educational level, employment status, history of toxic exposure, smoking and drinking status of couples, frequency of physical exercise in women, annual household income of the family, duration of cohabitation (years), menstrual cycle interval, history of abnormal pregnancy, and parity. Likelihood ratio tests were used to compare the fit of models containing different variables to test for departures from a linear trend in ordered categorical variables. Differences with a P value of <.05 were considered statistically significant.