Faculty Profile

Thomas  Braun, PhD

Thomas Braun, PhD

  • Professor Of Biostatistics
  • M4063 SPH II
  • 1415 Washington Heights
  • Ann Arbor, Michigan 48109-2029

Thomas Braun is a Professor in the Department of Biostatistics. He received his PhD in Biostatistics from the University of Washington in 1999 and joined the University of Michigan faculty that same year. Prior to his graduate study in Seattle, Tom was an actuary with Metropolitan Life Insurance Company. Dr. Braun's current research interests include: (1) adaptive Phase I trial design methodology, including optimizing schedule of adminstration, accomodating patient heterogeneity, and incorporating multiple toxicity grades; (2) methodology for modeling tooth-level and site-level correlation of teeth in longitudinal dental and periodontal studies, including multivariate longitudinal modeling of clinical attachment level (CAL), pocket depth (PD), and bleeding on probing (BOP), as well as multivariate longitudinal clustering of those outcomes as predictors of site-specific and subject-average progression of oral disease. 

  • PhD, Biostatistics, University of Washington, 1999
  • M.S., Biostatistics, University of Washington, 1996
  • B.B.A, Actuarial Science, University of Wisconsin-Madison, 1990

  • Adaptive Phase I Trial Designs in Oncology
    Sponsor: University of Michigan Comprehensive Cancer Center (UMCCC)

    Many of the motivating examples for my research develop from my collaborations with the UMCCC Bone Marrow Transplantation and their studies of acute-graft-versus-host disease (aGVHD) and relapse in allogeneic bone marrow transplant recipients.

    Although sophisticated designs for dose-finding in Phase I trials have grown in number over the past two decades, little attention has been given to determining how many administrations of the dose can be given, as well as when each dose should be given. I have developed statistical models that allow for assessing the toxicity and efficacy of multiple administration schedules of an agent as well as determining which of the schedules leads to optimal rates of toxicity and/or efficacy. The schedule finding algorithm relies upon Bayesian methods that require the user to develop a prior distribution that is sufficiently informative for the beginning of the trial, but sufficiently non-informative for the end of the trial. Therefore, my research also focuses upon how to derive usable prior distributions from basic information supplied by investigators, allowing statisticians and clinicians to "communicate" more effectively.

    I am also researching: (1) models for studies of two or more agents, one of which is often a biologic target used to weaken the cancer and "boost" the myeloablative ability of standard chemotherapy and radiation; (2) the effect of ignoring and the benefit of incorporating patient heterogeneity on the performance of adaptive designs; (3) models for inclusion of mild, non-dose-limiting toxicities into determining the overall probability of dose-limiting toxicities (DLTs); (4) the impact of premature death, a competing risk for DLT, on the performance of adaptive designs.

    This research is partially funded through my R01 grant from the National Institutes of Health entitled "Statistical Methods and Issues for Implementing Adaptive Phase I Trials"

  • Multivariate Longitudinal Modeling of Periodontal Outcomes
    Sponsor: Michigan Center for Oral Health Research (MCOHR)

    I am the primary statistician for clinical studies at the University of Michigan School of Dentistry.  Much of my collaborative effort involves prediction, diagnosis, and treatment of periodontal disease.  Most periodontal studies collect data with multiple levels of correlation including (a) the same outcome, i.e. clinical attachment level (CAL), pocket depth (PD), and bleeding on probing (BOP), measured longitudinally on the same site, (b) different outcomes measured longitudinally on the same site, (c) multiple sites measured on the same tooth, and (d) multiple teeth measured on the same subject.  However, hierarchical data analysis methods like random effects models and GEE have yet to be the "gold-standard" in this setting.  Much of my research is focused upon increasing the use of these data analysis techniques in periodontal studies, as well as developing new approaches for correlated data useful for this setting, such as Bayesian spatial modeling and permutation tests.

    A secondary area of interest is methodology is developing predictive models for the progression of periodontal disease and gingivitis using longitudinally measured site- and tooth-specific salivary biomarkers and plaque baterial counts, either through regression modeling strategies or multivariate clustering algorithms. 

  • Biometrics Society
  • American Statistical Association