Christopher H. Schmid, Professor and Chair of Biostatistics, Brown University
Constantine was born in Velanidia Kozanis—a small village in Macedonia, Greece—and came to the US to study first at Union College and then Princeton, where he focused on mathematics and graduated in 1976. He then completed a PhD in mathematical statistics at Cornell University in 1981, writing his dissertation on decision theory. While at Cornell, Constantine became involved in statistical applications through consultations with researchers from the agricultural school and participated in the Seminar of Bayesian Inference, organized and run by Arnold Zellner through the National Bureau of Econometric Research. Connections made through this seminar with Rob Kass, Jim Hodges, and Nozer Singpurwalla and through a two-year visiting professorship at Carnegie Mellon led to the popular Case Studies in Bayesian Statistics meetings that took place at Carnegie Mellon in the 1990s and produced a series of published proceedings. These demonstrated the power of solving complex problems with Bayesian methods that flowed from fundamental advances in computation arising from new techniques such as posterior approximation and Markov Chain Monte Carlo.
Following academic appointments at Rutgers and the University of Massachusetts and a growing interest in applied statistics stimulated through work on the analysis of data from longitudinal studies of childhood depression with Marika Kovaks in Pittsburgh, Constantine moved to Harvard to help start the department of health care policy in 1988. He wrote several influential papers while there, applying Bayesian hierarchical models in health services research studying variations in utilization, outcomes, and quality of health care. A 1997 paper in the Journal of the American Statistical Association with Sharon-Lise Normand and Mark Glickman established a new approach to medical provider profiling that augmented risk adjustment with methods for identifying aberrant providers through posterior predictions.
At this time, Constantine also worked with Mike Daniels to develop ordinal hierarchical models for analyzing studies of diagnostic accuracy. This enabled simultaneous modeling of location and scale parameters, accounted for clustering among readers and institutions, and incorporated external information about parameters such as disease prevalence through prior distributions. Constantine also started collaborating with Harvard faculty such as Fred Mosteller, Art Dempster, and Tom Chalmers on analyses of Medicare claims data and meta-analysis using the hierarchical modeling techniques he was pioneering. This led to methods for constructing hierarchical summary ROC curves for diagnostic tests through a series of papers with his Harvard post-doc, Carolyn Rutter, between 1995 and 2001 and to a seminal 1994 methods paper for meta-analysis of diagnostic tests led by Les Irwig, who was visiting Harvard. Through Irwig, Constantine became involved with the Cochrane Collaboration and convened a working group that developed a manual and template for conducting meta-analysis of diagnostic test studies housed in the well-known Cochrane Library. While at Harvard, Constantine also helped start the Section on Health Policy Statistics for the ASA, serving as its first chair and co-chairing the first ICHPS, held in Boston in 1995. This led to founding the journal Health Services and Outcomes Research Methods, which Constantine edited for many years.
In 1995, he left Harvard to found the Center for Statistical Sciences (CSS) at Brown and start a track that later became a section in biostatistics within the department of community health in the medical school. Students soon began arriving, and the first PhD was awarded in 2005. In 2011, biostatistics became a department and, in 2013, became one of the four founding departments of the school of public health under Constantine’s leadership. It serves about 60 graduate students today and is the home for a growing undergraduate major in statistics.
Almost since its beginning, the major research program at CSS has been the data coordinating center for the American College of Radiology Imaging Network (ACRIN), which eventually merged with the Eastern Cooperative Oncology Group to form ECOG-ACRIN. Constantine serves as the group statistician and leads the analysis of its NCI-funded multi-center clinical studies. These studies involve large-scale clinical trials to evaluate the efficacy of diagnostic screening tests, as well as health services investigations of their use and cost-effectiveness. Influential studies include DMIST, which established the superiority of digital mammography as compared to plain film mammography; NLST, which established that screening for early detection of lung cancer reduced overall mortality and was cost-effective; and TMIST, which is assessing the impact of tomosynthesis—three-dimensional mammography—in 165,000 women.
Recently, the center has branched out to diagnostic brain imaging in the IDEAS study, which is looking at the impact on clinical care of testing for amyloid plaques in patients with mild cognitive impairment using a registry linked to Medicare claims data. IDEAS is also comparing health care use outcomes between those who had the test and a control group of similar patients without the test constructed from Medicare claims data. The control group is derived using sophisticated matching algorithms developed by Brown Biostatistics Professor Roee Gutman and department students.
In addition to his work at Brown, Constantine has chaired the Committee on Applied and Theoretical Statistics of the National Academy of Sciences and served on the Committee on National Statistics. He co-chaired a National Research Council committee that produced an influential 2009 report on the major scientific and organizational problems facing forensic science. It found a lack of conclusive evidence about the accuracy of many common criminal investigative tests such as the analysis of fingerprints, tire marks, ballistics, and bites. It also documented excessive variation in the practice of forensics across the US due to different rules, requirements, and qualifications. The report led to increased research funding for forensic science and spurred interest in the problems among statisticians. Constantine also served on the committee that defined comparative effectiveness research and recommended the formation of the Patient-Centered Outcomes Research Institute (PCORI). He later chaired the group that developed PCORI’s guidelines for evaluating diagnostic tests.
Nowadays, Constantine continues his work in diagnostic screening and its connections to the analysis of biomarkers and medical images but focuses on the application of modern methods for big data and electronic health records in the context of causal inference. Though no longer chair of the biostatistics department at Brown, he is still focused on its role as a driving force in health data science and its close connections to health policy.
Key Publications of Constantine Gatsonis
C.A. Gatsonis. 1995. Random effects models for diagnostic test accuracy. Academic Radiology 2 S14–S21.C.A Gatsonis; Normand, S.L.; Liu, C.; and Morris, C. 1993. Geographic variation of procedure utilization: A hierarchical approach. Medical Care YS54–YS59.
Institute of Medicine. 2009. Initial national priorities for comparative effectiveness research. National Academies Press, Washington, DC.
L. Irwig; Tosteson, A.; Gatsonis, C.A.; Lau, J.; Colditz, G.; Chalmers, T.C.; and Mosteller, F. 1994. Guidelines for meta-analyses evaluating diagnostic tests. Annals of Internal Medicine 120, 667–676.
National Lung Screening Trial Research Team. 2011. Reduced lung cancer mortality with low-dose computed tomographic screening. New England Journal of Medicine 365(5), 395–409.
National Research Council. 2009. Strengthening forensic science in the United States: A path forward.
S.L. Normand; Glickman, M.; and Gatsonis, C.A. 1997. Statistical methods for profiling providers of medical care. Issues and applications. Journal of the American Statistical Association 92, 803–814.
E.D. Pisano; Gatsonis, C.A.; Hendrick, E., et al. 2005. Diagnostic performance of digital versus film mammography for breast-cancer screening. New England Journal of Medicine 353, 1773–1783.
C.M. Rutter and Gatsonis, C.A. 1995. Regression methods for meta-analysis of diagnostic tests. Academic Radiology 2, S48–S56.
C.M. Rutter and Gatsonis, C.A. 2001. A hierarchical regression approach to meta-analysis of diagnostic test accuracy evaluations. Statistics in Medicine 20, 2865–2884.
Leave a Reply