2008 Application Catalog
Preceptorships - Division of Cancer Prevention
Foundations of Prevention Research Groups
The Biometry Research Group engages in independent and cooperative research studies on cancer epidemiology, prevention, screening, and diagnosis using methods of mathematical and analytic statistics, and conducts independent and collaborative studies in biostatistical and epidemiologic methodology and in mathematical modeling of processes relevant to cancer prevention activities.
Chief: Philip C. Prorok, Ph.D.
Stuart G. Baker, Sc.D.
Causal inference, cancer screening, cancer biomarkers, and missing data.
- Evaluating the age to start periodic screening without data from randomized trial.
- Issues involved in using biomarkers as surrogate endpoints.
- Identifying biomarker combinations for further study as triggers of early intervention.
- Designing a study to compare digital versus analog mammography.
- Paired availability design for strengthening inference from historical controls.
- Adjusting for noncompliance in randomized trials.
- Adjusting for missing categorical data.
Vance Berger, Ph.D.
Design and analysis of medical studies, especially randomized clinical or prevention trials. This includes all types of biases that can interfere with valid comparisons of medical interventions.
- How to salvage valid between-group comparisons in the presence of selection bias arising from incomplete allocation concealment (predictable allocations).
- How to salvage valid between-group comparisons in the presence of selection bias arising from using run-in data as part of the entry criteria.
- The quantification of statistical evidence as it applies to all forms of medical decision making problems.
- Building robustness into statistical analyses so that they are free from unrealistic assumptions such as normality, proportional hazards, or common variances.
- Improved evaluation techniques for screening programs.
- Making efficient use of multiple endpoints.
Grant Izmirlian, Ph.D.
Applied statistician with interests in the areas of monitoring clinical trials, the analysis of microarray data and machine learning in the classification of proteomics data.
- Statistical issues in peak detection in proteomics studies. Random Forests is a classification algorithm that is suitable for classification problems in proteomics studies. My recent work has demonstrated that inference on the selection of important peaks is fraught with difficulties. I have several ideas about how these can be resolved. The interested fellow could join my efforts in this area. This project involves statistical theory as well as computer programming.
- Statistical issues in the analysis of microarray data. Another recent project has been the development of a new “per gene” test which borrows strength over the ensemble of genes in the estimate of the per gene covariance matrix. The test has advantages over standard anova tests in the form of greater efficiency, and this results in better estimates of the importance of truly biologically active genes that are potentially overshadowed in the conventional anova tests by genes of low signal that are tightly distributed about their means. The interested fellow can be involved with analyses that I am doing as well as the larger project of testing the methodology out via simulation.
- In the area of monitoring of clinical trials I have been highly active in the design of a monitoring plan in a recently initiated trial of lung cancer screening. While the standard methods of monitoring trials (the Lan-Demets method for boundary construction) have been around for quite some time, there really aren’t adequate methods for gauging the performance of candidate plans under crossing hazards when it is necessary to use a weighted statistic. Moreover, recent work of others suggests that the boundaries be constructed on a scientifically meaningful scale. Under proportional hazards this is just the relative risk estimate, which is obtained directly from the log-rank statistic. In the crossing hazards/weighted statistic arena there is much to be done to clarify this issue. Work for the interested fellow would be related to the methodology discussed here.
Victor Kipnis, Ph.D.
Biostatistical methodology, statistical modeling, statistical methods in nutritional epidemiology, and monitoring of clinical trials.
Statistical modeling of, and adjustment for, dietary measurement error in the nutritional epidemiologic studies.
Design and analysis of nutritional epidemiologic studies.
Design and analysis of studies to validate/calibrate dietary assessment instruments.
Energy adjustment models in nutritional epidemiology.
Aggregate level analysis in multi-centers epidemiological studies.
Pamela Marcus, Ph.D.
Cancer screening, including both methodology associated with and implementation of cancer screening trials.
- Randomized controlled trials of cancer screening modalities, including protocol development, study operations, and data analysis
- Methodology associated with assessment of cancer screening modalities, including case-control studies of screening and randomized controlled trials
Blossom Patterson, Ph.D.
Selenium kinetics, compartmental modeling, dietary assessment, and latent class analysis.
- Human metabolism of the trace element selenium (Se), a potential cancer preventive agent, has been investigated in a pharmacokinetic crossover study. Compartmental models for an inorganic (sodium selenite) and an organic (selenomethionine) form are being used to compare metabolism by gender and by fasting status (fasting and fed).
- An ongoing Se supplementation kinetics tracer study in humans is generating data that are being used to compare the kinetics of selenite and selenomethionine in subjects before and following long-term supplementation with selenomethionine. The models are being modified and will be used to test for changes in metabolism due to supplementation.
Philip Prorok, Ph.D.
Design and analysis of cancer screening programs, mathematical modeling of screening, analysis of cancer natural history data, analysis of censored duration data, and stochastic processes.
- Data from randomized trials of breast, colon, and lung cancer are analyzed to investigate properties of screening tests, age-specific effects of screening, and disease natural history. Models of the natural history of several major cancers are being developed with a screening component, including properties of the screening tests, screening interval, compliance, etc., to address screening policy questions and analysis of screening studies. Areas of methodologic development include estimation of detection properties of screening tests, validation of outcome measures for screening studies, case control studies of screening, study monitoring procedures, and estimation of lead time and length bias.
Jian-Lun Xu, Ph.D.
Stochastic modeling in cancer study, survival analysis, multivariate analysis, nonparametric inference, asymptotic theory, and analysis of biased sampling data.
- Modeling and estimating tumor size at metastasis.
- Modeling and estimating sensitivity and mean sojourn time in cancer screening.
- Estimating sensitivity and specificity for dependent repeated screening programs.
- Estimating the survival function based on cross-sectional sampling.
- Estimation of biased sampling model and adjustment of the bias.
- Modeling exchangeable binary data.