I, for one, can't function without a large cup of coffee in the morning. And at least another cup in the afternoon, to get me out of that post-lunch slump that makes me want to put my head on my desk and take a nap.
Fortunately, it seems that my coffee-drinking habits may mean that I'm less exposed to certain health risks, according to new research by the University of Colorado School of Medicine in Aurora.
Laura Stevens, a Ph.D. student at the University of Colorado, and Drs. Carsten Görg and David Kao, who both conducted this study, used machine learning alongside traditional data analysis techniques to uncover an inverse relationship between how much coffee we drink per week and how exposed we are to heart failure and stroke.
Their results were recently presented at the American Heart Association's (AHA) Scientific Sessions 2017, held in Anaheim, CA.
Could more coffee make a difference?
In the first instance, the researchers employed the random forests algorithm in machine learning to examine data sourced from the Framingham Heart Study, which has been running since 1948, providing crucial information about cardiovascular health.
Machine learning can make predictions based on data associations, and it has been increasingly used in healthcare and health-related research in the past few years. This is, in part, because it allows researchers to perform data mining — the process of identifying patterns based on very large amounts of data — more efficiently.
Following the machine learning data analysis, Stevens and colleagues found that an extra cup of coffee every week is associated with a 7 percent lower risk of heart failure and an 8 percent lower risk of stroke.
Additionally, the researchers performed traditional data analysis — Cox proportional hazards — on the information sourced from two other large population studies: the Cardiovascular Health Study and the Atherosclerosis Risk In Communities Study.
The same association between drinking coffee and a lowered risk of heart failure and stroke was found after analyzing the additional data, confirming the results indicated by machine learning.
Although the findings were consistent, the researchers emphasize that the association is not necessarily causal, so we shouldn't jump to any conclusions just yet.
Machine learning for health research
Stevens and team also found a correlation between how much red meat we include in our diets and the risk of stroke and heart failure. In this case, red meat consumption was identified as a potential risk factor, although here, the correlation was less striking.
Data analysis of the Framingham Heart Study suggested that red meat eaters, much like coffee drinkers, are at a lower risk of experiencing either stroke or heart failure. That being said, it is currently difficult to verify these results because the definition of what constitutes "red meat" differs between studies.
Currently, the AHA recommend consuming less red meat — including beef, pork, and lamb — due to its higher cholesterol and saturated fat content. Instead, they suggest replacing it with chicken, fish, and beans.
Stevens and team also designed a predictive risk model targeting congestive heart failure and stroke. To do this, they relied on known risk factors obtained using the Framingham Risk Score, to which they also added the newly revealed correlation between coffee consumption and cardiovascular health.
"By including coffee in the model, the prediction accuracy increased by 4 percent," says Dr. Kao. "Machine learning may a useful addition to the way we look at data and help us find new ways to lower the risk of heart failure and strokes."
The scientists will go on to explore risk factors for those two events and aim to better understand how our diets influence our cardiovascular health. They hope that machine learning methods will be effective in uncovering hitherto unknown culprits.
"Our findings suggest that machine learning could help us identify additional factors to improve existing risk assessment models. The risk assessment tools we currently use for predicting whether someone might develop heart disease, particularly heart failure or stroke, are very good but they are not 100 percent accurate."