An important analytical issue for many of the Division's research initiatives is the characterization of time to an event. Many studies include repeated events (i.e., time to pregnancy, habitual pregnancy loss), and it is important to examine how various factors such as environmental factors affect these event times. Standard statistical methods in survival analysis, when the event time is death or the occurrence of disease, are well developed. However, time-to-pregnancy and other outcomes related to maternal and child health pose new analytic challenges. For example, unlike traditional survival analysis, time-to-pregnancy analyses must account for the fact that there is no risk for pregnancy without intercourse during the fertile window. Further, approaches for modeling time-to-pregnancy must account for the fact that a percentage of women will not become pregnant regardless of number of attempts or duration of trying. The BBB is uniquely poised to address these issues. Examples of specific research in this area are included below.
Reproductive epidemiologists are increasingly interested in ascertaining the influence of environmental exposures on reproductive health of humans; however, research is not readily amenable to experimental designs. One quantity of considerable interest -human fecundability - has necessitated development of better statistical methods and models for its study. Human fecundability is measured through the probability of conception in a menstrual cycle for a couple who is having regular intercourse and is not using contraception. Because of the inherent hierarchical data structure and measurement error involved in various quantities, such as day of ovulation, identification of "fertile window" length, and other biological quantities (some of them unmeasured) with considerable impact on conception and on probability and the time to conception, the statistical methods required are very specific to this area of research. BBB investigators have developed methods that address joint modeling of longitudinal intercourse data and time-to-pregnancy data. The Branch has also studied issues related to incorporating chemical mixtures, a process which results in highly correlated covariates, into survival models to assess the effects of environmental exposures on time to pregnancy. These methods have applications for various prospective pregnancy cohort studies, such as the LIFE Study, that involve pre-pregnancy recruitment of women or couples.
Developing new methods for survival analysis is an important and dynamic biostatistics research area. Although survival is not usually an endpoint in Division studies, many survival analyses techniques can be adapted to analyzing time to a particular event, such as time to pregnancy, time to an important milestone in fetal or child growth, gestational age, time to ovulation, or time to a crash in the teen driving studies. Adapting and developing new statistical methods for survival analysis that are applicable to the type of event times related to child and maternal health is complex and complicated. BBB investigators are working to create new methods of analyzing recurrent events (such as the timing of multiple pregnancies), correlated-survival data (such as time to ovulation when multiple cycles of the same women are observed), multistate events (such as time to different milestones in child development), and event-times with competing risks (such as time-to-poor delivery outcome).
Gestational age can also be viewed as a type of survival data (i.e., time from conception to birth). BBB investigators have been developing innovative new statistical methods for estimating the incidence curve for gestational age that can be used to predict whether women with prior history of complications in pregnancy are more likely to have pre-term birth. This methodology allows investigators to estimate these incidence curves when there is a competing risk. For example, this method is used for estimating the incidence of pre-term birth for a spontaneous birth by treating elective deliveries as a competing risk.