Research Program Lisa A. Weissfeld Professor and Associate Chair Dept. of Biostatistics
Methodology • Survival Analysis • Spline-based extensions of the Cox proportional hazards model: development of estimators for correlated outcome data, estimation of survival curve, goodness-of-fit, tests of proportionality • Correlated outcome data: copula approach • Missing data: copula approach, pattern mixture model, estimating equation approach
Collabarotive • Critical Care Medicine: sepsis research, health services research (2 GSRs funded) • Obesity and Nutrition Research Center: behavioral trials, metabolic studies, PET studies (2 GSRs funded) • Positron Emission Tomography: Pittsburgh compound B, late life depression (2 GSRs funded)
Other Research funding • Cancer training grant: funds two students and has funding for one postdoc
Critical Care Medicine • Spline based extensions of the Cox proportional hazards model based on Gray’s model. • Application: transplant data and ICU data. • Properties of model: does not require that proportionality assumption holds.
Critical Care Medicine • Missing and/or truncated data • Examples: inflammatory marker data has a lower limit of detection. Most “normal” samples are at this lower limit. • Development of statistical methodology: modeling techniques for accounting for the truncation of the outcome variable in the repeated measures setting, modeling techniques that account for truncation when the variable is a covariate, modeling techniques that allow for the inclusion of multiple correlated inflammatory markers.
Critical Care Medicine • Missing and/or truncated data (ctd.) Organ Failure assessment: how to handle large amounts of missing data. Examine the impact of “filling in” missing values on analyses. Informative censoring: how do you account for informative censoring in a repeated measures analysis.
Critical Care Medicine • Quality of Life Analyses Estimation of quality adjusted survival: methods in this area are different from those in cancer where there are discrete states. Missing data is also a problem with this type of data.
ONRC • Missing data • This is a big issue in behavioral intervention studies. • In the area of pediatric obesity, the problem is further confounded by the fact that the subjects are growing over the course of the study. • Received attention in the medical literature with an editorial in the New England Journal of Medicine
ONRC • Missing data (ctd.) • Also a problem in smoking cessation where individuals often miss visits. • Appears as a different problem in metabolic studies, where you may sample a small portion of a large cohort (outcome-dependent sampling).
ONRC • Definition of outcomes • Problem in pediatric obesity where many of the subjects recruited are > 95th percentile of body mass index. Need good definition of weight loss for individuals in this category.
PET • Development of methods for the analysis of a new ligand, Pittsburgh Compound B (PIB), which binds with amyloid • Development of discrimination rules for a diagnosis of Alzheimers disease and mild cognitive impairment from PIB results.
PET • Statistical Issues • Development of voxel-based methods for the analysis of PIB data, particularly across modalities. Currently, there are no methods that are computationally feasible. • Development of summary measures that can be readily used to discriminate diagnosis categories.
PET • Statistical Issues (ctd.) • Assessment of “best” parameter settings for voxel-based analyses. • Analysis of repeated PIB scans using a voxel-based approach