Mary E. Putt, Sc.D.

Mary Putt, Professor of Biostatistics, specializes in the application of statistical methods to data arising from biomedical research. She develops and modifies statistical methods, as needed, to meet the specific needs of a particular collaborative research projects.

Most recently I worked with Dr. Sung Han, a post-doctoral associate, developing new methods to estimate change-points in blood flow data where the baseline function is modelled using smoothing splines. These data were collected as part of a tumor biology study to better understand how a novel treatment for cancer, photodynamic therapy (PDT), can effectively kill cancer cells by shutting down blood flow to the tumor. The data for one of the papers resulting from this work is included on this site. Work is ongoing to develop better approaches for estimating confidence intervals, and for determining the number of change-points.

With a second post-doctoral associate, Dr. Yun Lu, I recently carried out a comparison of the relative strengths of two iterative logic regression approaches, logicFS and MC Logic, for identifying predictors of interest in high dimensional data that arose in a study of heart failure. We showed that the two approaches yield similar results based on frequency metrics, but that a variable importance measure (VIM) suggested by logicFS is fundamentally flawed. Through analysis and simulation we demonstrate the reasons for the misleading VIM scores that sometimes result from this procedure. A paper on this work is currently under revision.

Lastly, recently work graduate student, Laurel Bastone, yielded a permutation-based latent class method to understand heterogeneity of outcome in a family-based linkage study in genetics. I extended this work to identify classes of neurons that may fire synchronously as part of a collaborative project with the Coulter lab. This group is testing the hypothesis that seizure disorders may involve a few neurons acting out of synchrony with the majority of their network, which then triggers a pathological cascade of neuronal activity. Simulation studies to demonstrate the validity of the latent class method as described in Takano et al. (2012) appear below.

Selected Publications

  • Putt M and Chinchilli VM. A robust analysis of the two-treatment crossover trial using multi-sample generalized L-statistics. Journal of the American Statistical Association 95(452):1256-1262, 2000.
  • Putt ME. Power to detect clinically relevant carryover in a series of crossover studies. Statistics in Medicine 25(15):2567-2586, 2006.
  • Bastone L, Spielman RS, Wang X, TenHave T, Putt M. A latent class model for testing for linkage and classifying families when the sample may contain segregating and non-segregating families. Human Heredity 70:75–91, 2010.
  • Takano H, McCartney M, Ortinski PI, Yue C, Putt ME, Coulter DA: Deterministic and stochastic neuronal contributions to distinct synchronous CA3 network bursts. The Journal of Neuroscience 32(14): 4743-4754, 2012
  • Han, SW, Mesquita, R, Busch, T, Putt, M. A method for choosing the smoothing parameter in a semi-parametric model for detecting change-points in blood flow. Journal of Applied Statistics (in revision)
  • Lu, Y, Hannenhalli, S, Cappola, T and Putt, M. An evaluation of MC Logic and logicFS motivated by a study of the regulation of gene expression in heart failure. Journal of Applied Statistics (in revision)