Citation: For foundational, creative, and original contributions to mathematical statistics; for methodological developments in empirical processes and machine learning; for advancement of precision health; and for mentoring of students, postdocs, and junior faculty.

Sebastien Haneuse and Eric Laber write:

Michael Kosorok is the consummate biostatistics methodologist. His work is creative, rigorous, and motivated by urgent and impactful problems in biomedicine. He has authored more than 200 peer-reviewed manuscripts appearing in top-tier journals and conference proceedings. In 2008 he published his monograph Introduction to Empirical Processes and Semiparametric Inference, which quickly became the canonical introduction to the area. Shortly thereafter, Kosorok focused his research efforts on artificial intelligence and precision medicine. He was among the first to provide rigorous theoretical results for machine learning methods for estimation of optimal treatment regimes including some of the earliest applications of Q-learning and direct search estimation. His paper on outcome weighted learning (which has been cited nearly 750 times) began long and fruitful lines of research on direct-search estimation and revealed important connections between optimal treatment regimes and classification.

In addition to his prolific publication record, Kosorok has shaped the field through his service and mentoring. He served as the head of the Biostatistics Department at UNC from 2006–20, chair of the COPSS Presidents’ Awards committee, and is currently President-elect of the IMS. He has mentored more than 50 PhD students, many of whom now hold prominent positions at academic institutions or in industry.

In the publication cited in the award (Nguyen et al, 2020), Kosorok along with his co-authors, developed a novel direct-search estimator of the optimal regime under a 2×2 crossover design. Such crossover designs are common in pilot testing, rare diseases, and other settings in which recruiting a large pool of participants is difficult. Nevertheless, prior to this publication, there were no direct search estimators for estimating an optimal treatment regime under such a design. The proposed method accounts for carryover effects, uses a convex relaxation for computational efficiency, and is Fisher consistent.

This publication is an illustration of Kosorok’s research modus operandi. He identifies an important practical problem, develops a novel methodological approach, and then provides a rigorous and complete description of the method’s operating characteristics.