Preterm birth is the leading cause of infant mortality, and bacterial vaginosis (BV), a common condition, has been consistently associated with a doubled risk of preterm birth. However randomized trials of treating BV to prevent preterm birth have produced mixed results. Systematic reviews using aggregate study data found no overall benefit of treatment, but raised possibilities of benefit to women at high preterm birth risk treatd at <20 weeks' gestation. To overcome the limitations of conventional meta- analysis, we propose pooling the raw data from the trials in an individual patient data meta-analysis. Using pooled raw data, we will have greater power to evaluate the benefit of treatment of BV in subgroups of individuals and measure the days of pregnancy prolongation in treated and untreated women. Use of aggregate data does not facilitate analyses based on individual patient characteristics, such as gestational age at randomization, as is possible with individual data. In particular, we will investigate the interaction between gestational age at treatment and a priori risk of preterm birth on prolongation of pregnancy. We have commitments from 10 study PIs, representing 72% of women randomized in all the trials, to provide data and help understand variables and analyses reported from their studies. Preparing the data for analysis and harmonizing variables across studies is a major effort. Our team has extensive experience and expertise in maternal fetal medicine in general and preterm birth and BV in particular as well as database management and statistics. Missing data methods, including multiple imputation, will be employed for some variables. Logistic regression models with random effects for specific studies and interactions between treatment (drug/placebo), gestational age at randomization, and a priori risk will be fit to identify subgroups of pregnant women in whom the risk of preterm birth is reduced. Time to event (i.e., birth) models with random effects and interactions will be used to estimate the number of days of pregnancy prolongation attributable to treatment of BV in pregnant women. A sensitivity analysis will be conducted to study the influence of single studies, inclusion of aggregate as well as individual data, and statistical models. Developments in this area have utilized Bayesian hierarchical models, which have the flexibility to incorporate data of different types. Computations will employ Monte Carlo Markov chain simulation in R and WinBugs. Statistical methods will be extended as appropriate. This project is novel in at least three ways: a unique collection of data sets and major effort to combine them, unique analyses of the impact of BV treatment on preterm birth, and unique sensitivity analyses and statistical methods. We expect to identify subgroups of women who benefit from treatment and the optimal window to initiate treatment and to have a significant positive impact on maternal and child health through screening and treatment recommendations.
Bacterial vaginosis (BV), a common condition, has been consistently associated with a doubled risk of preterm birth, a leading cause of infant mortality. Randomized trials and reviews using aggregate data have produced mixed results. A meta-analysis of individual patient data from multiple studies is a cost effective, necessary step in assessing the benefit of BV treatment and defining best clinical practice.
