Faculty Spotlight: Min Qian

Min Qian, PhD, Associate Professor of Biostatistics

My long-term goal is to develop rigorous and generalizable statistical tools to make better use of big data to provide optimal treatment decisions. Toward this goal, I have pioneered machine learning and statistical inference methods for data-driven decision-making in precision medicine. Precision medicine aims to develop evidence-based approaches to personalizing treatment based on individual characteristics to optimize clinical outcomes. Influenced by recent advances in biomedical technologies, which have created and continue to create rich personal data (such as genomics, electronic health records, and digital health data), precision medicine has become a national priority since the announcement of the Precision Medicine Initiative in 2015. It holds promise for improving many aspects of healthcare, such as advancing the diagnosis and treatment of complex diseases, alleviating unnecessary side effects that result from conventional one-size-fits-all approaches, and ultimately reducing healthcare costs without compromising quality or outcomes.

My primary research interest is in the development of automatic decision-support systems for mobile health applications. This research is motivated by the IntelliCare suite of apps for anxiety or depression. The suite contains 12 smartphone apps that provide users with different psychological treatment strategies. A Hub app was developed to help users navigate the 12 apps in the IntelliCare ecosystem and coordinate their experience. This experience includes managing messages and notifications from the other clinical apps within the IntelliCare suite and making weekly recommendations for new apps to encourage app usage and exploration. I have been working, in collaboration with students and colleagues, on the development of reinforcement learning methods to support making mobile intervention decisions to optimize both proximal (e.g., app usage) and distal (e.g., clinical) outcomes in a highly personalized manner. Examples include estimating personalized decision algorithms to optimize response rates within the generalized linear mixed model framework and developing decision algorithms to optimize app usage in the presence of high-dimensional decision options.

Another of my research interests is in high-dimensional inference. A key challenge in precision medicine is to identify variables among a large set of candidate covariates that are predictive of the outcome of interest. Most of the literature in this area focuses on variable selection methods, which provide only point estimates of the effects of the selected covariates. To measure the strength of the association between the outcome and the selected covariates, researchers will undoubtedly seek some sort of inferential guarantees (such as p values or confidence intervals) for the selected covariates. This is a challenging task, since we must account for the effects of the selection; otherwise, it can greatly exaggerate the strengths of relationships. I have been working on various topics in this setting. Methods developed here can be used to screen prognostic or predictive risk factors, select treatment-beneficial subgroups, identify potential mediators, and so forth.