Improving Causal Inference Tools for Addiction Researchers

点击翻译按钮获取中文摘要

Project Summary Finding truly effective treatment for addiction and substance abuse requires not only knowing that a treatment works on average for a population, but also knowing the mechanisms or pathways through which a treatment works and for whom treatment is most effective. Methodologists have been developing new statistical techniques that rely on fewer assumptions and, as such, are more robust at providing accurate answers than ever before. However, it can be difficult for many applied researchers to understand how to implement these new methodologies and few such methods have been implemented in easy-to-use software. This application requests support for a three-year R01 study to improve the use of robust causal inference methods for understanding mechanisms and moderators of treatment effectiveness in the field of addiction. We propose to do so by enhancing existing methods, creating software to implement those methods, and embarking upon a strategic outreach campaign to facilitate researchers’ use of the new methods and software. Over the past decade, our team has successfully worked to encourage broader use of propensity score methods for causal research in the field of addiction by (1) creating and enhancing a statistical software package called TWANG or Toolkit for Weighting and Analysis of Nonequivalent Groups, which is available in R, SAS, and Stata, (2) contributing more than 30 methodological articles devoted to implementing propensity score methods, and (3) conducting over 20 short courses and workshops. Analytic tools such as TWANG which allow researchers to assess the comparability of groups receiving different treatments and estimate the causal impact of substance abuse treatment programs using observational data are increasingly valuable as funding for large scale randomized studies is dwindling. Our team plans to engage in a series of methodological and software developments that will help improve the next generation of addiction health services research. We aim to create 20 additional software commands and tutorials that will provide researchers with tools and training for studying how and for whom treatments work, studying the impact of treatment dose, and assessing the sensitivity of their estimates to uncontrolled factors. To enhance the utility of our tools, we will develop them in the R and Stata and in a new computing environment which analyst access through a web browser and is menu driven. In addition to methodological innovations, the proposed project will include several outreach efforts, such as short courses, webinars, and peer-review manuscripts to promote best practices among researchers applying the newly developed tools. Through these efforts, this project aims to not only develop new methods, tools and software, but also to improve the statistical practices of addiction researchers, greatly strengthening the scientific information upon which decisions are made to improve care in our country and directly meeting NIDA’s call for “methods that will support a new generation of health services research”.