My original research interest, dating back to undergraduate studies in language development, is analysis of longitudinal and other correlated data. Missing data prevent simple pre/post analysis of data collected longitudinally, and are a barrier to more detailed analysis as well. The missing data literature concentrates on implicit or explicit imputation of missing values, so that inference can be made as if the data were not missing. My research considers another possibility: when the “missing” data do not even exist, as for data missing due to death. When analyses accommodate “missing at random” longitudinal data, imputation beyond death can influence study conclusions. For example, since cognitive performance often declines as death approaches, implicit imputation of cognitive performance scores beyond death could lead to lowered estimates of average functioning for survivors. My doctoral dissertation formalized an additional approach: partly conditioning on survival status, using techniques from standard methods such as generalized linear models and weighted estimating equations. The target of inference for these partly conditional models may be referred to as “regression conditioning on being alive.”
More recently, as a cancer center biostatistician, I have worked with investigators to design early phase clinical trials that include placebo controls, pharmacodynamic endpoints, and/or correlative studies to assess measurement of tumor markers. My area of emphasis is clinical research involving functional imaging, specifically PET and breast MRI. Statistical research on incorporating biomarkers into clinical trial study design has emphasized tissue and blood biomarkers. Since functional imaging has different strengths (in vivo functional information and measurement of multiple tumors) and weaknesses (limited access to archived samples), refinements to address statistical treatment of quantitative functional imaging data are a valuable contribution both to the clinical trials literature and to functional imaging.
My independent research in quantitative imaging has encompassed both analysis of correlated data and study design. I have examined within-patient differences in estrogen receptor functioning in breast cancer patients with multiple lesions, to inform design of functional imaging predictive markers to guide use of endocrine therapy. Another current project uses data from PET test-retest studies, calibration studies, and clinical trials to provide parameters for simulation studies to inform the use of quantitative imaging in future clinical trials. For example, the benefits of using simpler imaging protocols at a larger number of sites can be weighed against the potential for greater precision and lower detection thresholds accomplished by sophisticated imaging protocols available at fewer sites.