Biostatistics Research

Medical and public health research are concerned with understanding the causes of diseases and developing methods for their prevention, treatment and palliation. The role of the biostatistician in this enterprise is to formulate the study design-the plan for conducting the study-and to interpret its results by means of a formal statistical analysis. As medical science progresses, so too do the means for conducting medical studies. Consequently, there is a continuing need for the development of new methods for the design and analysis of medical research studies. The research of biostatisticians is concerned with the formulation of new methods for design and analysis, the elucidation of their mathematical properties, and their correct and efficient implementation in computer software.

Faculty of the Department of Biostatistics are active in a number of areas of methodologic research. The list below gives a few major areas of interest.

Categorical data:
Many variables naturally are measured on categorical scales-i.e., as classifications into discrete categories rather than on a continuous scale. The modeling of such data poses an array of statistical challenges. Recent areas of interest have been the modeling of missing categorical data and clustered categorical data, as arise for example in periodontics.

Begg, Levin, Parides, Petkova

Clinical trials:
It is widely acknowledged that the randomized trial, in which patients are assigned randomly between study arms, constitutes the best possible method for comparing two treatments. However, in specific trials many questions may arise, such as how best to effect the randomization, how many subjects to enroll, how often and how long to observe them during the course of the study, how to analyze data prior to the termination of the study, what model to use for data analysis, and how to handle data from subjects who contribute only partial observations or who do not fully comply with their assigned treatment.

Bagiella, Levin, Meier, Parides, Petkova, Thompson, Troxel, Desai, Cheung, Friedewald

Clustered and longitudinal data:
Many studies in the health sciences involve periodic observation of outcome variables on a sample of subjects. Analyses of such data must posit reasonable statistical models for the evolution of outcomes over time, and must properly account for the correlation within individuals over time. Recent methodologic research in this areas has been concerned with formulating appropriate models for categorical observations, for clustered observations at each time (such as visual acuity in right and left eyes), and for series that are incomplete due to patient dropout.

Begg, Liu, Paik, Petkova, Troxel, Waternaux

Epidemiologic methods:
Studies of the causes of disease are typically observational, in that the assignment of treatments to subjects is not under the control of the investigator. Analyses of data that are collected in this way are much more heavily dependent on underlying statistical models than are the analyses of data from randomized trials. Recent statistical research in this area has been concerned with models for causal inference, and with methods for exploiting new epidemiologic study designs and data structures.

Bagiella, Hodge, Levin, McKeague Wickramaratne

Incomplete data:
Most medical research studies are affected to some degree by missing data; for example, subjects may miss visits or refuse to answer some questions. It is critical to understand how missing data arise, how they can affect design and analysis, and what can be done to extract correct information from studies that are subject to missing observations. Recent research has focused on methods for imputing missing observations, and for coping with the biases that can arise when the incomplete cases are not a random sub-sample from the study population.

Liu, Paik, Petkova, Troxel, Tsai, Waternaux

Statistical genetics (Genetics of Complex Disease Training Program):
A central goal of contemporary health research is to understand the molecular and genetic basis of disease. Statistical modeling has always been an essential part of this enterprise, and statisticians with expertise in genetics have been much in demand in recent years.

Hodge, Lo, Desai, Friedewald, Gorroochurn

Survival data:
In many medical studies, the most important outcome variable is time to an event, such as time to death, heart attack, development of AIDS symptoms, or recurrence of cancer. One feature of such data is that typically some observations are censored, in that they have not reached the endpoint (death or disease) by the time the study ends. Research is ongoing on how best to handle such data, with minimal assumptions, in a wide variety of applications.

Bagiella, Liu, Lo, McKeague Meier, Paik, Troxel, Tsai, Weinberg, Wickramaratne, Jin

Department of Biostatistics . 722 West 168th Street . 6th Floor . New York . NY 10032
Phone:(212) 305-9398 . Fax: (212) 305-9408