Model Analysis

A&M Statistician Probes Missing Link Between Diet and Cancer

By Shana Hutchins
 

Cancer. It’s a leading cause of death in the United States and a leading inspiration behind the current research agenda of Texas A&M University Distinguished Professor of Statistics Raymond J. Carroll.

According to the American Cancer Society, about one-third of the 564,830 cancer deaths expected to occur in 2006 will be related to nutrition, physical inactivity or obesity and, thus, could be prevented.

Carroll has dedicated many of his research efforts to doing just that, helping Americans beat their odds of contracting the nation’s second deadliest disease by working to ensure the validity of the studies that help predict those odds.

Since 1991, Carroll has been analyzing the missing link between diet and cancer — or, more specifically, why so many studies find no relationship between the two due to errors in reporting and analysis.

“Typical nutritional surveys feature questions like, ‘How many calories a day do you consume?’ and ‘What percentage of your diet has fat in it?’” Carroll explains. “These are questions nobody can really answer.”

In the quest to offset ambiguity, Carroll and his legion of protégés have made many fundamental contributions to the area of statistical expertise known as measurement error modeling — providing reliable analyses in situations where variables and exposures are measured with error.

In Carroll’s current case, the situation is breast cancer and the possible nutritional causes behind it, particularly fat intake. His statistical weapon of choice involves nutritional epidemiology, a high-stakes field in which the name of the game is to develop questionnaires that elicit more accurate answers and, therefore, more reliable data.

Years of using statistics to quantify survey uncertainty have led Carroll and his research group to two main conclusions: Size matters. So does method.

“One of the impacts of my work has been to show that there is a true need for large samples, because it’s incredibly hard to measure and quantify survey results,” he says. “To be reliable, sample sizes need to be in the hundreds of thousands.”

Carroll notes that survey designers typically rely on three measurement instruments to help gauge American dietary habits: food frequency questionnaires, 24-hour recalls and food diaries.

To put the inexactness of this science into proper perspective, he cites a classic food frequency questionnaire staple — the “pizza question,” in which respondents are asked how often they eat pizza, how many pieces they eat and whether or not they include meat.

“In just three questions, they try to capture your lifetime fat intake from pizza,” he explains. “Then they have to convert your answers into nutrient values. There are uncertainties in every step of the process, from how people answer, to the conversions. The reason for requiring large samples is that the instruments aren’t very good.”

In addition to introducing error and uncertainty, food frequency questionnaires can’t measure caloric intake, which Carroll says has a huge impact on obesity studies.

He considers the 24-hour recall method equally problematic, because it looks at a very small snapshot of dietary history when the long-term is more important. Moreover, it’s expensive and, therefore, highly cost-prohibitive.

A better alternative to both methods is food diaries, in which respondents report actual eating habits for a week at a time on three or four different occasions throughout the year.

“Food diaries are really good instruments because most Americans would be appalled at what they report,” Carroll adds. “They are a much better way of measuring, not to mention modifying.”

In a study he recently submitted to the Women’s Health Initiative using food diaries and a sample of 30,000 women, Carroll and his colleagues found a statistically significant relationship between fat intake and breast cancer.

“I’m not a nutritional epidemiologist; I’m a hard-core statistician,” he notes. “But what I’ve done is design a statistical method that goes into the analysis of these big data sets. Our group has come up with statistical methods and analyses unique to the problem of measuring diet.”

In 2001 Carroll came up with a novel way to train the next generation of statistical scientists, helping to shape a new genre of interdisciplinary research known as bioinformatics in the process. Backed by a $1.6 million National Cancer Institute (NCI) grant recently renewed through 2011, he established the Bioinformatics, Biostatistics and Nutrition Training Program at Texas A&M. The unique post-doctoral program — the nation’s only one in biostatistics — seeks to build bridges between life sciences (biology and genetics) and computational sciences (statistics) to better train future statisticians to function as independent researchers as they continue to explore links between nutrition and cancer.

“I was developing statistical methods for more interdisciplinary projects by myself by reading books on biology,” Carroll recalls. “I had a wonderful student at the time named Jeff Morris who helped a lot, but essentially, I proved that a statistician could learn enough biology to be useful to a biologist.”

Last year Carroll was rewarded for his pioneering efforts in nutritional epidemiology and biology with an NCI Method to Extend Research in Time (MERIT) Award. Less than five percent of all National Institutes of Health-funded investigators merit selection for the highly selective award, which includes up to 10 years of grant support. Carroll, who was one of eight nationwide to earn the distinction in 2005, is the first statistician to be chosen since the inception of the program in 1987.

Never one to rest on his laurels, Carroll is already honing in on his next challenge — analysis of radiation exposure and cancer risk as part of a Nevada bomb test site study.

“All of us living in the U.S. in the 1950s were exposed to radiation when the government tested bombs at sites in Nevada,” he explains. “Regardless of what the weather conditions were like back then, there were a lot of deposits of radiation in the ground. Cows ate the grass; people drank the cows’ milk; people developed thyroid disease.

“Statisticians always talk in terms of probability. We never make categorical statements. It’s extremely difficult to say with absolute certainty that this caused you to have this problem or condition. It’s a population science.”

One that never gets boring for Carroll, who notes with a slight degree of amusement that he’s changed research areas four times since entering a profession he loves back in the 1970s.

“Statistics is just fun and cool,” he adds. “You don’t actually know what the answers are going to be when you start. Plus, the skills are transferable to interesting areas.”