miR 92/19 cluster in the ERK context

点击翻译按钮获取中文摘要

The Tsepamo birth outcomes surveillance study has accrued many thousands of women living with HIV from Botswana. The Tsepamo Plus study (Project 1) will continue to collect ARV exposures in pregnancy which are rarely studied with high precision. With such large samples, Tsepamo can detect differences in rare pregnancy outcomes such as neural tube defects, and can provide precise estimates of effects for common outcomes such as prematurity. Because Tsepamo can provide high precision when there is much uncertainty about safety in pregnancy and Botswana typically rolls out new ARVs before other African countries, Tsepamo may be the first to report on novel safety signals. This in turn makes Tsepamo highly influential in the understanding of adverse ARV effects in pregnancy. A natural question is if first and subsequent analyses should apply group sequential statistical methodology to plan for “when to look” and to enhance the understanding of uncertainty. Further, guidance is sorely needed for understanding uncertainty and interpreting safety signals for unplanned analyses. Moreover, surveillance data alone describes circumstances as they are and are not directly set up to inform personal, clinical, or societal decision-making. For example, while a surveillance database may help detect an increase in neural tube defects, the data alone do not inform the optimal decision about what’s to be done to reduce neural tube defects in future pregnancies. Ideally, such decisions would be informed by randomized trials, but trials are often too costly, unethical, or not timely enough. Instead, we can only empirically inform decisions by emulating target trials using observational data. The principle behind target trial emulation is simple: lacking a randomized trial for a given research question, we describe in detail the protocol of the randomized trial we would like to conduct, and then combine subject matter expertise and appropriate statistical analyses to emulate that hypothetical trial using observational data. By embracing this framework, many of the common pitfalls of current observational research practices can be avoided; however, target trial emulation requires multidisciplinary collaboration and often novel analytic approaches. The objectives of this proposal are to evaluate methodologies and provide guidance on methods for making best use of Tsepamo data and similar studies. Specifically, to evaluate the use of group sequential methodology for surveillance systems in studies of pregnancy and to create guidance on statistical adjustments for unplanned analyses we have created the following aims: Aim 1: To create guidance on when to publicly report on safety signals based on unplanned analyses. We will compare the statistical properties of group sequential methods from Aim 1 to fixed sample methods which were used in the NTD example, and provide recommendations on the timing of public release of unplanned analyses. Considerations will include the probability of decision reversal, the probability of missing a true signal, and a comprehensive scientific understanding of the range of true effects consistent with the analysis. Aim 2: To evaluate the statistical properties (e.g. Power, Type one error) of various group sequential methods (e.g. Lan-DeMets error spending approach) when applied in the setting of large sample sizes with rare and common pregnancy outcomes. We will compare numerous approaches to account for tests of accumulating surveillance data in the Tsepamo study, and develop methods to account for uncertainty in the maximal number of women taking specific ART regimens. To further develop methods for meaningful and valid causal inferences in Tsepamo and related databases, and develop a sustainable program for target trial emulation in Botswana, we also aim: Aim 3: To develop and implement methods to benchmark an emulated target trial using Tsepamo data with results from a randomized trial. We will benchmark Tsepamo analyses against the VESTED trial to validate and calibrate our emulation approach, and then expand upon the VESTED trial findings to estimate effects on rare (e.g. stillbirth) and new (e.g. weight gain) outcomes and within subgroups (e.g. by maternal nutrition status) that the original trial was not able to assess. Aim 4: To develop and improve triangulation strategies for emulating target trials within pregnancyrelated surveillance data. We will develop guidelines, procedures, and software to support triangulating evidence, including integrating methods for transporting results from a randomized trial to Tsepamo’s underlying study population, as well as non-conventional study designs based on multiple pregnancies, proposed instrumental variables, and negative controls.