ave you ever noticed how research study conclusions can shift over time?
Take red wine. In the late 1980s, epidemiologists observed that French people had a low risk of dying from coronary heart disease despite their predilection for cheese and other fatty foods. They theorized that drinking red wine, which contains antioxidants from grape skins, was the reason. As a result, many white wine drinkers switched to red.
But then other evidence began to emerge. White wine also contains antioxidants, and some studies found it provides as many health benefits as red. Furthermore, antioxidant content depends on the type of grape, the soil where it’s grown, how it’s processed, and even how it’s stored. Some white wines contain more antioxidants than red.
Confusing results like these are common, and it’s not always because subsequent studies come to different conclusions. Sometimes data from the original study could have revealed more nuanced or even totally different conclusions, but we don’t learn about them because researchers don’t examine all the variables and how they relate to one another, says Jeffrey Wilson, a professor in the Department of Economics.
“In research papers in biostatistics and epidemiology, too often, people ignore correlations for the ease of calculations,” Wilson says. “When you ignore correlations, you are likely to see something as significant when it is not. You may make statements that aren’t true.”
ave you ever noticed how research study conclusions can shift over time?
Take red wine. In the late 1980s, epidemiologists observed that French people had a low risk of dying from coronary heart disease despite their predilection for cheese and other fatty foods. They theorized that drinking red wine, which contains antioxidants from grape skins, was the reason. As a result, many white wine drinkers switched to red.
But then other evidence began to emerge. White wine also contains antioxidants, and some studies found it provides as many health benefits as red. Furthermore, antioxidant content depends on the type of grape, the soil where it’s grown, how it’s processed, and even how it’s stored. Some white wines contain more antioxidants than red.
Confusing results like these are common, and it’s not always because subsequent studies come to different conclusions. Sometimes data from the original study could have revealed more nuanced or even totally different conclusions, but we don’t learn about them because researchers don’t examine all the variables and how they relate to one another, says Jeffrey Wilson, a professor in the Department of Economics.
“In research papers in biostatistics and epidemiology, too often, people ignore correlations for the ease of calculations,” Wilson says. “When you ignore correlations, you are likely to see something as significant when it is not. You may make statements that aren’t true.”
To help researchers avoid this pitfall, Wilson and a few PhD students began working on a new statistical model that analyzes multiple variables and observes how they affect one another as changes occur over time. They began their work in 2000, developing the model over several years. Today, Wilson’s Generalized Method of Moments marginal regression model has been accepted by leading academic journals. But because it’s abstract and complex, few people outside of academia understand how to use it.
An opportunity arises
Wilson found the chance to apply his model to real results when statistics graduate student Dan Xue asked for help interpreting a study Wilson had never heard of: the Chinese Longitudinal Healthy Longevity Study. The investigation began in 2005 and tracked the health of 2,000 people ages 64 and older in China for at least nine years. Wilson says the last follow-up survey was in 2018. The study is unusually rich in data, not only because it has lasted so long but also because it includes many variables that come from two separate methods of acquiring information.
In one method, researchers measured input variables: They asked participants health-related interview questions and reported their answers. Questions ranged from the number of vegetables the participants consumed to how much they exercised, and whether they could move about without help and make decisions.
In the other method, researchers measured response variables, observing participants as they performed physical tasks. Could they move about freely? Could they reach down and pick up a book from the floor?
The same people were studied in these same two ways, year in and year out (though the number of participants eventually dwindled to 1,100, likely because some passed away during the study). That created a host of possibilities for examining the results. Each input variable could be correlated with not only all the other input variables but also each response variable. Also, participants’ health habits and physical abilities changed over time, and each change produced new correlations to study.
Making sense of this much information would be a daunting task for most researchers. “But I have a model that can analyze it,” Wilson says.
Applying the model
Working with Xue, now a business analyst at Equality Health, and graduate student Elsa Vazquez-Arreola, now a statistician at the National Institute of Diabetes and Digestive and Kidney Diseases of the National Institutes of Health, Wilson applied his model to the Chinese study results from 2005 to 2014. The researchers recently published their results in a paper in BMC Medical Research Methodology.
They learned many things that could be useful to older adults and elder-care providers. For example, eating lots of vegetables positively affected health, but only for those who did so in the early stages of old age. Having the ability to make decisions also correlated with better health for younger seniors. Being able to move about without assistance was consistent with better health outcomes across all age groups over all time periods.
Some of these results suggest that feeling independent plays a larger role in overall health than previously supposed. “Making decisions is so important. It could lead to a different path in caring for older people and a better quality of life,” Wilson says.
The value of complexity
As older populations in industrialized nations increase, senior care is a matter of growing importance. The Chinese study offers more data for other researchers to mine, Wilson says, and the information they discover could be important in shaping the future of public health care.
But the value of Wilson’s model is not limited to health; researchers can apply it to any field. Modern computing makes complex models easier and faster to use, and researchers owe it to themselves to take advantage of them, Wilson says.
— Teresa Meek