Epidemiologic COVID-19 Response Corps
The Causal Lab is also involved in the Epidemiologic COVID-19 Response Corps. This COVID-Corps is a collaboration between BU faculty, students, alumni, and colleagues to address important epidemiologic and public health issues related to the COVID-19 pandemic, with a focus on impact, equity, and justice.
Improving epidemiologic research through the use of causal diagrams
Causal diagrams, such as directed acyclic graphs (DAGs), are an excellent tool for summarizing and assessing assumptions and knowledge about relationships between variables in order to make causal inference. However, there are few existing resources to guide the use of DAGs by applied researchers. To address this issue, we have formed a DAG Working Group, and begun to conduct research into the current state of applied DAG research (see: pre-print 1, pre-print 2).
Prevention of cervical cancer in young women with perinatal HIV infection through HPV vaccination
This project, funded by the NICHD, aims to understand preliminary evidence of a reduction in effectiveness of the human papillomavirus (HPV) vaccine in girls and young women with perinatal HIV and develop recommendations for improved HPV vaccine dosing schedules to reduce future risk of cervical cancer in these young women as they age. Much of this work also intersects with projects on improving the use of applied causal graphs, improving observational causal inference, and improving the use of individual-level simulation models for causal inference. For more details see: R21: HPV VACCINATION EFFICACY FOR CERVICAL CANCER PREVENTION IN YOUNG WOMEN WITH PERINATAL HIV INFECTION.
Improved causal inference for observational data
Obervational data presents a number of unique challenges for causal inference. Although a detailed theoretical framework exists for understanding how to address these challenges and estimate causal effects from observational data, application of these methods to real data and research questions is not always straight-forward. The goal of this project is to improve the usefulness and uptake of causal inference methods for observational data analysis. The first paper of this series focuses on understanding the challenges of using distance to care as an exposure or instrument in studies of health outcomes (AJE 2019). The second paper describes a method for determining the optimal treatment strategy when resource or other constraints prevent implementation of the true optimal strategy (preprint).
Causal inference using individual-level simulation models
The aim of this project is to develop methods for building and assessing individual-level simulation models to allow estimation of causal effects and improved decision-making. Individual-level simulation models require specifying hypotheses about the structure of complex systems and, when correct, can be used for inference about the effects of interventions on health behaviors at a population level. However, identifying whether a model has been correctly specified presents a number of challenges. This project will develop tools for describing model assumptions, identifying potential violations of these assumptions, and assessing the impact of these violations on model-driven decision-making. The three main publications from this project are available in the Publications list (AJE 2017, MDM 2018, MDM 2020).
Design and analysis of pragmatic randomized trials
Pragmatic randomized trials are randomized trials designed for the purpose of identifying population-level impacts of medical interventions. As such, they are typically of longer duration, have looser inclusion/ exclusion criteria, and are less tightly controlled than traditional randomized clinical trials. These features allow for potentially more generalizable trial results, but come with a cost. Pragmatic trials are subject to potential biases in the design or analysis phase, and no clear guidance exists on how best to design these trials when causal effects other than the effect of randomization are desired. This project aims to provide this guidance by developing and demonstrating appropriate methods for pragmatic trial analysis through case studies, and by identifying patient-centered causal effects — causal effect information that is useful to patients for participating in shared medical decision-making. See Publications for the papers from this project: (1) describing patient and investigator preferences for causal effects (Journal of Clinical Epidemiology, 2018); (2) impact of time between study visits on feasibility of controlling for post-randomization confounding (Trials, 2019); and (3) draft guidelines for the design and analysis of pragmatic trials (preprint).
Adjustment for (null) effects of adherence to placebo in randomized trials
As part of the above project on design and analysis of pragmatic randomized trials, this project aims to demonstrate the possibility of adjusting for predictors of adherence without introducing bias in to the analysis of randomized trials. Since at least the 1980s, adherence-adjustment in randomized trials has been viewed with skepticism. Much of this skepticism is due to a compelling paper from the Coronary Drug Project trial, comparing adherers to non-adherers in the placebo arm of the trial and finding a large increased risk of death over 5 years associated with non-adherence to placebo. This was seen as proof of intractible confounding due to healthy behaviors and other unknown drivers of adherence. However, when we applied modern statistical methods to the Coronary Drug Project data we were able completely remove this effect. We continue this project by applying these methods to several other randomized trials for which placebo-arm adherence analyses have previously been reported. See Publications for these papers, including re-analyzing the Coronary Drug Project (Clinical Trials 2016, Trials 2018), and re-analyzing the Candesartan in Heart Failure Morbidity and Mortality trail (Contemporary Clinical Trials, 2020).
History of epidemiology
Dr Murray is also interested in the history of epidemiology. This began with an investigation into the sketch of ‘Captain John Graunt’ which is likely not a true represenation of John Graunt (AJE 2019). A second history of epidemiology project that Dr Murray is working on is as Associate Editor for Social Media at the American Journal of Epidemiology. In this role, she is the editorial lead on a series of Profiles in Epidemiology, highlighting epidemiologists who have made unique contributions to the development of the field, or followed non-standard paths in their epidemiologic career.
Work absence and disability among medical professionals
This set of projects assessed risk factors of, and interventions to reduce, workplace absence and work disability among nurses and other medical professionals. Medical professionals suffer high rates of burnout and mental health-related work disabiltiy, and can be required to perform intense physical labor, especially when lifting and transfering patients. In addition, medical professionals in rural areas may be subject to high levels of violence, as well as challenges due to professional isolation. In collaboration with the Occupational Health and Safety Agency for Healthcare (OHSAH), the Canadian Institute for the Relief of Pain and Disability (CIRPD), and the Institute for Work & Health, Dr Murray conducted three systematic reviews of the literature on occupational health and safety issues affecting medical professionals. See Publications for the resulting 7 peer-reviewed publications.
Time-series methods for assessing seasonality of novel infections
Analysis of H5N1 influenza (bird flu) in humans in Egypt and Indonesia using Fourier analysis of time series to determine whether H5N1 cases occurred with a predictable seasonality, and to identify climate variables for which the seasonality was associated with H5N1 seasonality and develop hypotheses about drivers of influenza seasonality. Results of this project, which concluded that cases likely occur seasonally and may be related to absolute humidity levels, were published in PLoS One (2011).