Type of Document Dissertation Author Mukherjee, Sudeshna URN etd-10082014-205320 Title Timing of gestational arrest prior to miscarriage Degree PhD Department Epidemiology Advisory Committee
Advisor Name Title Digna R. Velez Edwards Committee Chair Aaron M. Kipp Committee Member Katherine E. Hartmann Committee Member Pingsheng Wu Committee Member Keywords
- reproductive epidemiology
- misclassification and bias
- gestational arrest
Date of Defense 2014-07-18 Availability unrestricted AbstractRisk of miscarriage (i.e. a pregnancy loss before 20 completed weeks of gestation) is known to differ by race but timing of loss is not well established in the literature. The gap between biological pregnancy loss identified by ultrasound and clinical manifestation of that loss may bias effect estimates for early-pregnancy exposures associated with miscarriage. Right from the Start (RFTS) is a unique and diverse prospective pregnancy cohort that captures uniform early first-trimester ultrasound information and pregnancy-related behaviors from first-trimester interviews in order to study the distribution of this gap.
Nearly 13% of women in this cohort experienced a pregnancy loss (n=697), the majority of whom have ultrasound data available (73%, n=509). Ultrasounds were conducted between 40 and 95 days gestation from last menstrual period (LMP) for this cohort. Gestational arrest prior to miscarriage was observed in 38.7% of losses (n=197). The mean gap between LMP and estimated gestational age at arrested development (GAAD) was 19.3 ± 15.0 days (median GAAD gap was 19 days). The GAAD gap did not differ by race or pregnancy intention.
In order to determine if failing to account for this gap influences effect estimates we assessed exposures commonly associated with pregnancy loss. We compared models that estimated gestational age based on self-reported LMP and models that incorporated gestational age at time of arrested development (GAAD). We used bootstrap methods to determine the magnitude of bias for both models. Smoking during pregnancy was not modified by race and was not associated with miscarriage risk within this cohort for either current or former smokers compared to never smokers in either model. Stratified by race and adjusted for confounding, the protective effect of vitamin use on miscarriage risk was stronger among White women than Black women when using the LMP models (Whites aHR=0.34, 95% CI [0.21, 0.54]; Blacks aHR=0.53, 95% CI [0.33, 0.84]), while no substantial difference by race was observed with the GAAD models (Whites aHR=0.43, 95% CI [0.24, 0.76]; Blacks aHR=0.44, 95% CI [0.26, 0.74]).
Models that use self-reported LMP to estimate gestational age underestimate the true value of first-trimester smoking exposure on miscarriage risk by as much as 15% for current smokers and 5% of former smokers when compared to models that use GAAD (the bootstrap bias ratio between models for current smokers ratio=0.85, 95% CI [0.75, 0.94]; for former smokers ratio=0.95, 95% CI [0.92, 0.97]). When stratified by race, the bias was nearly 20% for both Whites and Blacks for miscarriage risk associated with early pregnancy vitamin exposure (Whites bias ratio= 0.79, 95% CI [0.62, 0.87]; Blacks bias ratio=1.19, 95% CI [1.13, 1.45]). These results suggest that early-pregnancy exposures associated with miscarriage risk are influenced by proper classification of gestational arrest prior to loss, and that the magnitude and direction of bias differs by race. By more accurately identifying which insults have occurred prior to pregnancy arrest and differentiating them from exposures that occur after developmental arrest but before the onset of bleeding, we have a more optimal method to assess miscarriage risk by not mis-assigning exposure time.
Filename Size Approximate Download Time (Hours:Minutes:Seconds)
28.8 Modem 56K Modem ISDN (64 Kb) ISDN (128 Kb) Higher-speed Access Mukherjee_Dissertation_Dec.2014.pdf 3.81 Mb 00:17:36 00:09:03 00:07:55 00:03:57 00:00:20