Return to Biostatistics
Ritesh Ramchandani, Jaclyn Scholl, Adam Sullivan, and Florence Yong are second year students in the Biostatistics doctoral program. They are currently focused on coursework, and are currently taking core biostatistics and epidemiology courses. Last summer Ritesh worked on chronic myeloid leukemia (CML) data, estimating rates of resistance to the drug lmatinib, and how those rates are affected by time from diagnosis to treatment. In addition they modeled CML progression and death using multi-state models to estimate probabilities of remission, progression, and relapse for patients. Ritesh is also interested, and is beginning work on methods for interval-censored data for progression and death from cancer and neurological diseases. Jackie has been working on Bayesian adaptively randomized clinical trial designs for glioblastoma research. More specifically, she has been investigating whether or not adaptively randomized designs are worthwhile from the drug company's point of view. Currently, common practice is to employ fixed randomization schemes and research is now taking the patient's vantage point to see if adaptive designs would allow for more efficient and effective treatments. Because drug companies are also important, she have been trying to determine whether these designs are convenient for them too. Costs and benefits are considered as monetary losses and gains. Since glioblastomas are not common and their natural progression is quite rapid, it was an appropriate and realistic setting to apply these clinical trial designs. Adam's research this past summer was to help analyze data for 4 different papers dealing with the aspects of Religious and Spiritual Care given to end of life cancer patients. These papers looked at patient's, nurse's and physician's perceptions of religious and spiritual care. Also the analysis of what predictors are there of nurses and physicians actually giving religious and spiritual care. The National Consensus Project for Quality Palliative Care defined religious and spiritual care to be an important part of palliative care. From this research the next step will be looking at structural equation models for the doctors and nurses in order to better understand what leads them to give spiritual care, along with this a mediation analysis. This will lead to methods work in terms of a sensitivity analysis for linear structural equation models which Adam will work on this summer. In addition Adam will consult on a paper looking at what kind of inhibitions that physicians and nurses have with giving religious and spiritual care to end of life cancer patients. Florence is very interested in Personalized Medicine in cancer research. Florence's summer project involved identifying subpopulations that would benefit most from the treatment for future patients using statistical methods. They hope to develop a parametric candidate scoring system using different covariates to estimate subject-specific treatment differences in clinical trials, with a goal to develop novel statistical methods to analyze cancer outcomes in observation studies. The potential contribution of such techniques could go beyond cancer research to many other disease areas. In addition she is involved in a book chapter project focusing on the analysis of survival outcomes.
Huang is a fourth year student (dissertation advisor: Cheng Li) in the Biostatistics doctoral program. He has completed all his courses and is working on his thesis. For the first project he reviewed various statistical methodologies for integrating copy number (CN) and gene expression (GE) data. This paper was published last year.
For the second project, he is exploring the response prediction problem through Support Vector Machines (SVM). The classical setup of SVM in response prediction (using GE) is to first filter the genes, then scale and select a kernel, and finally train the SVM algorithm through a validation dataset. In his setup however, he wants to incorporate network information into the setup and approach the problem through the use of *graphs. This requires working in the space of graphs and also a kernel specifically designed for this space. He is currently working on this paper and anticipates publishing it at the end of the summer this year.
For the third project, he wants to extend the methodology introduced in the second project to incorporate CN data. He hopes that the added dimensionality of CN data can increase the overall response accuracy. This final project is still in the planning stages.
Danielle Braun is a fourth year student (dissertation advisor: Giovanni Parmigiani) in the Biostatistics doctoral program. She has completed all her coursework and is working on her thesis. Currently she is working on a project to develop methods for handling misreported family history in Mendelian risk prediction models for cancer. In this project they introduce an approach that allows adjustment for misreporting of family history by modeling the measurement error process in this survival context, and using this to weigh the risk prediction estimates. They propose different models for the measurement error process using data from UCI. The goal is to extend BRCAPRO, a Mendelian risk prediction model for breast and ovarian cancer, to handle misreported family history. They will extend BRCAPRO using the methods they develop. They also will illustrate the results using specific cases, as well as validate using CFR data.
Cudhea is currently a fifth year doctoral student (dissertation advisor: Paul Catalano) in Biostatistics. He has completed all his courses and is working on his thesis. Fred's thesis focuses on developing a new way to fit dose-response models for developmental toxicity data using the Plackett-Dale distribution. Specifically, the goal is to model three outcomes: death, malformation, and fetal weight. The ultimate goal of fitting these dose-response models is to use them to establish a 'safe' dose for the toxin of interest.
The major challenge in analyzing this data is to account for the relevant litter-level and inter-outcome correlations. The hierarchical nature of the outcomes (malformation and fetal weight cannot be observed unless the fetus does not die) adds an additional layer of complexity and the primary goal of their method is to address this relationship between death and malformation/weight. They need to do this for proper inference on the model parameters, as well as estimation of the inter-outcome correlations. This also important as it can help them calculate joint risk of the multiple outcomes, which we can use to estimate an overall benchmark dose (or a 'safe" dose), as opposed to a benchmark dose for each outcome. Previous methods have used on the latent normal distribution to as a framework to analyze this data but Fred believes using the Plackett-Dale distribution offers some advantages in terms of interpretability of results, flexibility in modeling, and estimation of relevant associations within the data. The concept of dose-response models was first used in cancer studies but can be used in several applications.
Li is a fifth year Biostatistics doctoral student (dissertation advisor: Robert Gray). She has completed her coursework and is working on her thesis. She is currently working on three projects that are related to adjusting for treatment noncompliance in randomized cancer clinical trials. In the first project, they developed a weighted method to estimate the treatment effect, and possibly the interaction effects between treatment and covariates, in randomized clinical trials under treatment noncompliance, and compared the method with existing methods. Using simulations, the newly developed method is shown to be more robust to certain model misspecifications.
In the second project, they are interested in identifying biomarkers that can predict patient survival, and identifying patient subgroups (defined by the biomarker information) that will most likely benefit from specific treatment decisions. And the particular setting that they are mostly interested is that when it is not feasible to collect the biomarker information from all subjects, due to cost issue or others such as patient not consenting, etc, and further when treatment noncompliance is present. They extended the methods described above to fit in this more complex setting.
In the third project, they are interested in that when treatment noncompliance is present, how the various methods are affected in terms of the type 1 and type 2 errors, powers, etc, and simulations are run to explore what is the best strategy to apply different methods under different scenarios and conditions.
Anna Snavely is a fifth year doctoral student in Biostatistics (dissertation advisors: Yi Li and David Harrington) who recently defended her thesis, "Multivariate Data Analysis with Applications to Cancer". In her thesis Anna developed semiparametric and parametric latent variable transformation models that are useful in the setting where multiple outcomes are collected when the outcome of interest cannot be directly measured. The models allow for multiple outcomes of mixed types, including censored outcomes. This work was motivated by a head and neck cancer study at DFCI where multiple outcomes were collected to measure dysphagia in patients after treatment. Following graduation Anna will be a Clinical Assistant Professor at the University of North Carolina at Chapel Hill in the Departments of Medicine and Biostatistics.
Anna has also been working with the Head and Neck group at DFCI on clinical papers relating to dysphagia. In the clinical papers they used a dysphagia score that they developed (1 published, 1 in preparation).
In addition she has been working with the Survivorship group at DFCI. Anna has been primarily involved in a project where both patients and providers were surveyed to ask about survivorship care (what is currently happening and what they think should be happening).
- "Semiparametric Latent Variable Transformation Models for Multiple Outcomes of Mixed Types," University of Pennsylvania, Department of Biostatistics & Epidemiology. February 2012.
- "Semiparametric Latent Variable Transformation Models for Multiple Outcomes of Mixed Types," Vanderbilt University, Department of Biostatistics. March 2012.
- "A Semiparametric Latent Variable Transformation Approach for Modeling Multiple Outcomes of Mixed Types," ENAR, Washington, DC. April 2012.
- "Semiparametric Latent Variable Transformation Models for Multiple Outcomes of Mixed Types," University of North Carolina, Lineberger Comprehensive Cancer Center. April 2012
Dave Sihai Zhao is a fifth year doctoral student in Biostatistics (dissertation advisor: Yi Li) who recently defended his thesis, "Survival Analysis with High-Dimensional Covariates with Applications to Cancer Genomics". Following graduation Dave will be a Research Fellow at the University of Pennsylvania in the Department of Statistics.
Dave is interested in methods for analyzing genomic data with survival outcomes, especially for the purpose of generating prediction rules. One project he worked on in the past year was a method for quickly screening high-dimensional covariates to remove the ones most likely unassociated with the observed outcome. He studied the theoretical property of this procedure and showed that it can improve the predictive ability of reguarlized regression procedures.
In addition, Dave is currently working on a model for predicting high-risk multiple myeloma patient. He has applied his screening method as well as another method for regularized estimation using estimating equations to develop a model that can achieve 80% AUC in a validation dataset.
- "Regularized estimation for estimating equations", Statistics Department and Department of Biostatistics and Epidemiology, University of Pennsylvania, Philadelphia, PA, November 2011 (invited)
- "Sure screening for estimating equations in ultra-high dimensions", Department of Operations Research and Financial Engineering, Princeton University, Princeton, NJ, November 2011 (invited)
- "Sure screening for estimating equations in ultra-high dimensions", Department of Biostatistics, Harvard School of Public Health, Boston, MA, October 2011 (invited)
- "Grouped variable selection via hierarchical linear models", JSM, Miami Beach, FL, August 2011
Cristian Tomasetti is a postdoctoral fellow in Biostatistics (mentor Professor Giovanni Parmigiani, HSPH/DFCI). Biostatistics at the Harvard School of Public Health and the Dana-Farber Cancer Institute. Dr. Tomasetti received his PhD in Applied Mathematics and a MA in Mathematics, from the University of Maryland, College Park.
Dr. Tomasetti has been able to fully engage in developing novel mathematical methods and models with applications to the study of cancer. The main focus of his research has been the formulation and analysis of mathematical models, most often of mixed-type combining stochastic processes with differential equations, for modeling the development of drug resistance in cancer, the mode of division of cancer stem cells, and the accumulation of somatic point mutations in cancer. For example, he has been able to consider probabilistic cancer models where the typical exponential growth curve has been replaced by more realistic tumor growth curves.
Under the direction of his post-doctoral mentor, Professor Giovanni Parmigiani, Dr. Tomasetti has also been exposed to the statistical analysis of various cancer datasets (e.g. the TCGA sequencing data), thus enormously expanding his research potential by allowing the integration of these data analysis skills with his mathematical modeling approach. This interplay between biostatistics and biomathematics has resulted, for example, in a new mathematical model able to describe the accumulation of passenger and driver mutations in a way that fits the sequencing data.
Dr. Tomasetti has published various papers both on cancer mathematical modeling as well as on clinically relevant issues, like the effect of imatinib on leukemic stem cells. In addition to these papers, he and Dr. Parmigiani are close to submit some work with Professor Bert Vogelstein that will contribute to distinguish between drivers and passengers mutations found in tumor tissue. Finally, Dr. Tomasetti has also been invited to contribute a chapter on the phenomenon of drug resistance in leukemia. (Tomasetti C. Drug Resistance in Leukemia. In S. Corey & M. Kimmel & (Eds.), A Systems Approach to Blood. New York: Springer, submitted.)
Dr. Tomasetti has presented his scholarly work at international conferences and meetings, such as the 8th European Conference on Mathematical and Theoretical Biology (ECMTB) and Annual Meeting for the American Society of Mathematical Biology (SMB) in Krakow, Poland in July 2011, the 35th Conference on Stochastic Processes and their Applications, Oaxaca, Mexico in June 2011, as well as the Quantitative Issues in Cancer Research Working Seminar Group, Harvard School of Public Health, in November 2011. The opportunity to present his work at conferences has allowed Dr. Tomasetti to greatly expand his network of colleagues in cancer research and biomathematics.
- Mathematical modeling of random genetic mutations for a general class of tumor growth curves with applications (invited), Quantitative Issues in Cancer Research Working Seminar Group, School of Public Health -- November 2011, Harvard School of Public Health, Boston, MA, USA
- Modeling drug resistance dynamics in cancer stem cells for various types of tumor growth (invited), Institute of Applied Mathematics, HGS Mathematical and Computational Methods for the Sciences -- July 2011, University of Heidelberg, Heidelberg, Germany
- The role of symmetric and asymmetric division of cancer stem cells in developing drug resistance for various types of tumor growth, 8th European Conference on Mathematical and Theoretical Biology (ECMTB) and Annual Meeting for the Society of Mathematical Biology (SMB) -- June 2011, Krakow, Poland
- On the probability of random genetic mutations for various types of tumor growth, 35th Conference on Stochastic Processes and their Applications (SPA) -- June 2011, Oaxaca, Mexico
- The role of symmetric and asymmetric division of cancer stem cells in developing drug resistance, Dana-Farber BCB Retreat -- May 2011, Dana-Farber Cancer Institute, Boston, MA, USA
- On the probability of random genetic mutations for various types of tumor growth (invited), Branching Processes and Derived Processes: Transient and Asymptotic Behaviors -- April 2011, Centre International de Rencontres Mathematiques (CIRM), Luminy, France