Because of the large proportion of missing data for some early-life exposures, we repeated our main analyses and considered whether there was additional confounding among early-life exposures after multiple imputation of missing data. Results from these secondary analyses were largely similar to the main analyses and showed little confounding among the early-life exposures. Because maternal vital status at baseline was one of the strongest and most consistent factors that influenced whether data were missing for early-life exposures, we also repeated our main analyses after restricting to the approximately 60% of participants whose mothers were alive at baseline. Results from the restricted subset were less precise, but differences were generally minor except that pre-pregnancy diabetes was no longer associated with very early menarche (rRR = 1.06 [95% CI = 0.32-3.56]). However, because having diabetes influences lifespan, our restricted subset had fewer mothers with pre-pregnancy diabetes who were alive at baseline.
