Causal inference in the effect of a dynamic treatment process with applications to reproductive health


 Dr. Bo Lu, Department of Public Health, Biostatistics
Rank at time of award: Associate Professor



A major goal in many research fields is to identify and estimate the causal effect of a certain intervention. It becomes very challenging in population research, where most of the studies are observational. In those studies, the treatment is not randomly assigned to participants. Therefore, the differences in outcomes by treatment groups could be due to the differences in covariates prior to the treatment. Many statistical methods have been developed to deal with such selection bias, including propensity score matching, propensity score stratification, inverse probability of treatment weighting (Rosenbaum and Rubin, 1983; Robins et al., 2000). When applying those methods, researchers usually consider the treatment as static with dichotomous classification, such as treated and not-treated. In demography and population health research, many important characteristics may change over time, such as drug taking behavior, marital/cohabitation status, risky health habits (smoking, drinking, etc.). To fully understand their impact on future health outcomes, simply ignoring the dynamic nature of those processes might not be appropriate. In this proposal, we will consider the treatment assignment over time as dynamic and develop innovative statistical methodology to account for the selection effect due to various factors, which can be time-varying.

Our research is motivated by an observational dataset of patients seen at the Prematurity Prevention Clinic at the Ohio State University. The U.S. preterm birth rate rose by more than 20 percent between 1990 and 2006, accounting for nearly one in eight births in 2006. More than a third of infant deaths are related to preterm births. Infants born preterm also have higher rates of health complications and lifelong disabilities (Maxtin et al., 2006; Williamson et al., 2008). The underlying causes of preterm birth are poorly understood, although genetic, social, and environmental factors all likely play a role. Weekly injections of 17 alpha-hydroxyprogestcrone caproate (17P) have been found to reduce the risk of preterm birth and birth complications in randomized trials (Meis et al., 2003; Bergholla et al., 2010). Although positive results for clinical use of 17P have been shown, the mechanisms by which it works have not been fully determined and it remains unclear whether weekly 17P injections prevent the cervix from shortening. Treatment with 17P has been available at this Prematurity Prevention Clinic since 1998 and by 2005, 17P began to be offered to nearly all women. With this clinical database, we propose several innovative statistical methods to estimate the causal effect of 17P use on cervix change and time to delivery. We will examine the first reported pregnancy for each women with a singleton gestation resulting in births from 2005 and later (n = 315).