Research Interests

• Development of methodology for causal inference, accounting for missing data and comorbidity summary measures
• Application of state-of-the-art approaches, including hierarchical Bayesian, propensity score, competing risk, cost effectiveness and latent variable methods
• Examining the statistical properties of natural language processing algorithms.

Lab Overview

I am a Research Professor and Biostatistician at the Fox Chase Cancer Center. I have had papers published in both the methodological literature (e.g. Journal of the American Statistical Association, Biometrics) and the substantive literature (e.g. New England Journal of Medicine). I am actively involved in a number of behavioral clinical trials, and have experience accounting for missing data in such a context. I am specifically trained in the statistical field of causal inference, and many of my publications are related to methods for the analysis of observational data when confounding by comorbidities or other variables is of concern. I am also involved in pre-clinical biomarker validation and therapeutic research. More recently, I have been investigating the statistical properties of natural language processing algorithms. The mix of research projects allows me to engage in a broad spectrum of cancer and health services research.

Lab Description

1. I am the statistician on behavioral studies and clinical trials examining the utility of remote counseling for genetics. This is an increasingly important issue as genetic testing becomes more widespread and diffuse. Many physicians may desire that their patients have genetic counseling so that patients can better understand the implications of their test results, both for the patients themselves, and their families. However, many practices cannot afford to hire their own genetic counselors. Remote counseling, either through videoconferencing or telephone, may provide a low cost solution to the genetic counseling needs of both physicians and their patients.
2. I have increasingly become involved in natural language processing projects. In particular, I am interested in the statistical properties of these algorithms. This is work I have been conducting jointly with Dr. Slobodan Vucetic, a professor at Temple University. I find the work exciting, as the inferential properties (e.g. standard errors) of many techniques have not been well characterized. There is hence ample opportunities for novel and impactful statistical research in this field.
3. I am involved in biomarker and pre-clinical therapeutic research in a wide range of disease sites. I often use Gamma growth curve models with Generalized Estimating Equations (GEE) to examine dose response effects of drugs in animal models. I am excited to participate in projects that have the potential of resulting in new clinical treatments and etiologic findings.
4. Comorbidity adjustment is an important goal of health services research and clinical prognosis. When adjusting for comorbidities in statistical models, researchers can include comorbidities individually or through the use of summary measures such as the Charlson Comorbidity Index or Elixhauser score. While many health services researchers have compared the utility of comorbidity scores using data examples, there has been a lack of mathematical rigor in most of the evaluations. We have developed a research program examining the conditions under which individual versus summary measures are most appropriate. We have provided theoretical justification showing that comorbidity scores created analogously to the Charlson Comorbidity Index may be appropriate balancing scores for prognostic modeling and comorbidity adjustment. This work encompasses our general interest in methods for measuring and accounting for comorbidities.
5. I have collaborated extensively on projects using National Cancer Database (NCDB) and linked SEER-Medicare data. The Surveillance Epidemiology and End Results (SEER) database is maintained by the National Cancer Institute and collects data on tumor characteristics and demographics information about incident cancers. Studies using observational data, such as Medicare claims data, can provide useful information about treatment effectiveness. Of course, well run randomized clinical trials provide the best level of evidence of treatment effects. However, in many fields of medicine, randomized trials have not been feasible in the United States. We have used the SEER-Medicare claims and NCDB data for a number of comparative effectiveness research projects.