John S. Preisser, PhD

John Preisser, PhD, is a research professor in the Department of Biostatistics at the Gillings School of Global Public Health at the University of North Carolina at Chapel Hill (UNC) and deputy director of the biostatistics service with the North Carolina Translational & Clinical Sciences Institute (NC TraCS) where he directs biostatistics collaboration. He is a research fellow in the program on Aging, Disability, and Long-Term Care in the UNC Cecil G. Sheps Center for Health Services Research where he collaborates with experts in the field to conduct research that improves the care of the elderly in nursing homes and residential care facilities. Dr. Preisser is also deeply engaged in oral health research in the UNC School of Dentistry. These vibrant, multi-disciplinary, collaborative research environments motivate him to develop new statistical models and methods, and to promote state-of-art statistical analysis techniques that improve human health and welfare.

Honors and Awards

Elected Fellow

2010, American Statistical Association

Elected Member

2006, International Statistical Institute

Research Fellow of the Cecil G. Sheps Center for Health Services Research

2006-Present, University of North Carolina at Chapel Hill

James E. Grizzle Distinguished Alumnus Award

2001, University of North Carolina at Chapel Hill

Representative Courses

Models and Methodology in Categorical Data (BIOS 765), 2006-2018 (even years)

Research Activities

Dr. Preisser works on interdisciplinary research teams with clinical trialists and implementation scientists to study the impact of interventions on behavioral and health outcomes in real world settings. He currently serves as the biostatistician on a cluster-randomized trial (CRT) that uses a stepped wedge design to evaluate an intervention to improve the care of patients during transitions from skilled nursing facilities to home. He has helped devise stepped wedge and cluster-crossover designs in communities, emergency departments, hospitals and prisons for interventions targeting the opioid epidemic. He has also served as biostatistician on many intervention studies using more traditional CRT designs including a match-pairs CRT in nursing homes that improves oral health to reduce pneumonia rates and a non-randomized repeated cross-sectional CRT to combat underage drinking in communities.

Dr. Preisser’s methodological interests include modeling clustered binary data in a number of settings. First, the use of estimating equations with finite-sample corrections for population-averaged models has application for the analysis of CRTs, which typically have a small number of clusters (e.g., 8-30). In such settings, finite-sample bias corrections are needed to reduce the bias in the estimation of intervention effects. The development of methods for sample size estimation and statistical analysis of stepped wedge and other CRT designs provides an integrated approach to their design and analysis.

Second, Dr. Preisser is interested in health sciences applications involving partial-cluster sampling of correlated binary outcomes from anatomical regions where multi-stage sampling is employed for disease surveillance in populations. For example, large epidemiological investigations such as NHANES may employ partial-mouth oral exams where a subset of tooth surfaces are sampled (in the final stage following sampling of individuals) to estimate person-level disease prevalence for conditions such as chronic periodontitis that have full-mouth definitions. While the logistical and cost-saving motivations for such sampling strategies are strong, a correlated binary data model for the pattern of disease within the mouth is recommended to eliminate or mitigate the bias of prevalence estimates.

Finally, Dr. Preisser is interested in marginalized two-part models for semi-continuous and count outcomes with excess zeros. The general approach is to model the marginal mean of the outcome (i.e., of all the data) instead of the mean of an unobservable subset (latent class) of individuals. Applications include the estimation of overall effects in marginalized models for dental caries counts and medical costs where the two parts correspond to models for the marginal mean and the excess zeros. While interpretations are similar to those of traditional one-part generalized linear models, simulations show that “two parts are better than one” when zeros comprise greater than twenty percent of the data.

Service Activities

2018-present Team Science Promotions and Tenure Committee, North Carolina Translational and Clinical Sciences (TraCS) Institute
2018-present Serve on Data Safety and Monitoring Board for an NIDCR-funded study
2014-2017 Ad-hoc reviewer for the National Institute of Dental and Craniofacial Research
2011-present Chair, Pranab K. Sen Visiting Professorship Committee, Department of Biostatistics, UNC-Chapel Hill

Key Publications

Optimal allocation of clusters in cohort stepped wedge designs. Fan Li, Elizabeth Turner, John Preisser (2018). Statistics & Probability Letters, 137(2018), 257-263.

Two parts are better than one: modeling marginal means of semi-continuous data. Valerie Smith, Brian Neelon, Matthew Maciejewski, John Preisser (2017). Health Services Outcomes and Research Methodology, 17(3-4), 198-218.

Matching the statistical model to the research question for dental caries indices with many zero counts. John Preisser, D. Leann Long, John Stamm (2017). Caries Research, 51(3), 198-208.

A new way to estimate disease prevalence from random partial-mouth samples. John Preisser, Sarah Marks, Anne Sanders, Aderonke Akinkugbe, James Beck (2017). Journal of Clinical Periodontology, 44(3), 283-289.