Biostatistics, by its very nature, is a collaborative profession, says biostatistician Babette Brumback, Ph.D., who serves on research teams across the university on topics as varied as child obesity, lung cancer and the BP oil spill.
In fact, it is those collaborations, in which Brumback works with investigators to design studies and analyze data, that lead to new ideas for her own research.
“You can have these new interests come up when you find out that there’s a gap in methodology that needs to be filled with a new method, but you won’t learn that necessarily until you collaborate with the other investigator,” said Brumback, an associate professor in the department of biostatistics in UF’s colleges of Public Health and Health Professions and Medicine.
Two recent collaborations have led to the award of a National Science Foundation grant to support Brumback’s development of methods for analyzing complex survey data and adjusting for confounding variables.
In the first collaboration, Brumback worked with Amy Dailey, Ph.D., a former UF assistant professor of epidemiology now at Gettysburg College in Pennsylvania, on Dailey’s study of what factors may predict whether or not a woman will get a mammogram. Using data from the CDC’s National Health Interview Survey, they sought to separate individual level predictors, like race, income and education, from unmeasured predictors at the neighborhood level, like community resources.
“I became interested in the problem of assessing the effects of individual level predictors, adjusting for the fact that it may not be the individual level predictors that are inducing the mammography screening. Instead, the individual effect could be confounded by these unmeasured neighborhood effects. For instance, maybe it’s not the education of the individual, but rather the health literacy of her community, that is sending her to mammography screening,” Brumback said.
The statistical methods Brumback is creating with Dailey could potentially be used by researchers analyzing any of the major health surveys using complex sampling design that have clusters, like the neighborhoods evaluated in the mammogram study. The new methods may also have widespread use because they are easier to program and can be used with standard software, Brumback said.
Another recent study also involved complex sampling design with unmeasured confounding variables. Brumback collaborated with Richard Rheingans, Ph.D., an associate professor in the department of environmental and global health, on his study of water quality and sanitation in rural Kenyan schools and whether interventions like providing soap and building latrines would decrease students’ absenteeism. Schools were randomized to receive one of three interventions, but adherence to those interventions varied among the schools, making for some tricky data analysis and leading to the development of new statistical methods.
“The methods we developed for this study can be applied any time there is a desire to estimate the effect of the implementation of the intervention rather than the assignment of the intervention,” Brumback said.
Brumback came to the field of biostatistics almost by accident. After completing a bachelor’s degree in electrical engineering, she took a job babysitting the children of David Donoho, Ph.D., and Miriam Gasko Donoho, Ph.D., two well-known statisticians then on the faculty at the University of California, Berkeley. Brumback had been all set to begin Berkeley’s graduate program in mathematics, but the Donohos convinced her that statistics was a better match for her interests and helped her switch to the statistics program. Brumback went on to complete a postdoctoral fellowship in biostatistics at Harvard University.
“It was a pretty random path which led to my current profession,” she said. “But I was very fortunate to switch to statistics, and then biostatistics. Subsequent jobs have been very interesting indeed, and the field has expanded so much due to the incredible advances in computing that we have witnessed since the late 1980s.”