It is impossible for one person to keep abreast of all new methods in biostatistical design and analysis. Therefore, biostatisticians in the Center for Biostatistics develop different areas of expertise by focusing their work in a few areas. We take advantage of our diversity of expertise by assigning a team of statisticians to each research project, depending on its needs.
Design and Analysis:
Center biostatisticians focus on study design and planning as their most critical contribution to collaboration. This requires frequent interactions that result in improvements in experimental design, convincing conclusions and revealing data analyses.
Our expertise in design and analysis includes:
- Multiple hypothesis testing strategies
- Laboratory experimental design
- Bioassay experiments
- Robust mixed modeling for experimental data with dependency structure
- Cancer control design and analysis
- Propensity score matching for observational study designs
- Complex modeling for longitudinal and other observational studies
- Re-estimation of sample size
- Evaluation of diagnostic techniques
Center biostatisticians work with OSUCCC principal investigators on clinical trials conducted in cancer prevention, detection, diagnosis and treatment.
Our expertise in clinical trials includes:
- Bayesian designs
- Interim analyses and monitoring
- Development of study design and decision rules
- Choice of outcome measures
- Sample size and power calculations
- Development of case report/data collection forms
- Analysis of correlative data including pathway analysis
- Chemoprevention trials
- Re-estimation of sample size
- Repeated measures designs including crossover and semi-crossover
High Dimensional Data Analysis Support:
There are seven biostatisticians in the Center who have expertise in the design and analysis of studies with high-dimensional data.
Our expertise in high-dimensional data analysis support includes:
- Variance and degrees of freedom smoothing for small sample size
- Differential expression/abundance
- Data visualization
- Prognostic/diagnostic multivariate modeling
- Sample size to control power distribution
- Methods of controlling false discoveries
- Feature selection/validation including pathway analysis
Data Type Experience:
- Methylation MassARRAY
- Netowrk and Pathway Analysis
- Biomarker Discovery and Validation
- Customized RT-PCR arrays
Adaptive Sample Size:
Developed a re-estimation method that retains blinding of group assignment, and allows for an increase in enrollment when the event rate (recurrence for one trial, death for the other) is lower than initially anticipated. A manuscript is in press (clinical trials) on a method of blinded re-estimation and bootstrapped sample size distributions, which provides confidence in settling on a sample size near the end of planned accrual.
Methods for Donor Cell Experiments:
Donor cell experiments involve repeated measures (mixed) modeling. The BSR has developed a method of robustly testing hypotheses in these experiments by using concepts from conditional mixed models to ensure the variance estimates (denominator of test statistics) are not biased [presented at Joint Statistical Meeting (JSM) August 2009]. This method avoids estimating a saturated co-variance structure (which is often computationally difficult because of the small sample sizes available for many of these experiments) by defining appropriate variance components for specific contrasts on fixed effects.
Mechanism Hypothesis Testing:
Many cell experiments involve up-and-down regulation to validate direct control of the target outcome. This requires significant differences for both up-and-down conditions vs. control. The BSR has identified a method of determining sample size that ensures, with chosen power, significance for both comparisons simultaneously. The same method is used when testing a combination of conditions against each condition alone. Similarly, the BSR is researching methods to draw broader conclusions from a series of experiments that add confidence to a hypothesis about pathway mechanisms compared to more limited conclusions for each experiment in a series.