Laura Stevens, a doctoral candidate, and David Kao, MD, were invited to present their findings
at the annual scientific meetings of the American Heart Association.
Photo by Mark Couch
By Mark Couch
(May 2018) A cup of coffee a day might not keep the doctor away, but an extra cup or two could keep the cardiologist at bay.
from the University of Colorado Anschutz Medical Campus recently
presented research that shows drinking coffee is linked with reduced
risks for heart failure, stroke and coronary disease.
Just don’t ask if coffee is a magic elixir to will let you live longer.
was the main question we received: ‘Should I drink more coffee to live
longer?’” said David Kao, MD, assistant professor of medicine in the
Division of Cardiology. “I don’t think you can take it quite that far
because we don’t know whether it’s the coffee itself or something else.”
Kao and Laura Stevens, a doctoral candidate in the
University’s computational biosciences program, were invited to present
their findings at the annual scientific meetings of the American Heart
Association (AHA) last December.
While they can’t say
conclusively that drinking coffee extends your life, they were able to
show that drinking coffee is associated with a lower risk of some
serious life-limiting conditions.
According to their
review of data collected in the Framingham Heart Study, which has
tracked the eating patterns and cardiovascular health of more than
15,000 people since the 1940s, they found that every extra cup of coffee
consumed per day reduced heart failure by 5 percent, and stroke by 6
The news perked interest around the globe, with
reports in the International Business Times in London, Time magazine,
and in the online pages of the “the voice of Clarksville, Tennessee.”
drinks coffee,” Kao said. “It’s a high impact idea that a lot of people
can relate to. Plus, it validates what a lot of people had hoped would
be true – that more coffee is better. It justifies their behavior. I
figured that that would resonate.”
Stevens and Kao
explained that grinding through the data is a massive project that
required “machine learning” to identify associations that would be
buried under the multiple possible connections between numerous
Machine learning is a way of getting computers
to recognize patterns and make predictions, rather than simply
executing pre-programmed tasks. It’s teaching the computers to discover
or ‘learn’ new patterns within large amounts of data, and it works the
same way that online shopping sites aim to predict a shopper’s preferences based on previous purchases or an email provider tries to separate spam from other messages.
machine learning allows us to do is to identify factors that may be
important when we don’t know what we’re looking for in large pools of
information,” Kao said.
“In this case we were interested
in stroke, heart failure, and cardiovascular disease and then we used
machine-learning to determine what lifestyle, dietary factors, and
medical conditions are most important for predicting each disease.,”
Stevens said. “Basically, what you do with machine learning is you put
in the hundreds of factors into a model, and results will tell you which
factors are the most important for predicting a given outcome, such as
heart failure or stroke.”
Their study of the Framingham
data confirmed that certain well-known factors – such as smoking and
high cholesterol – had strong association with stroke, heart failure,
and cardiovascular disease. Sifting through the results they found the
tantalizing potential benefit of coffee drinking.
the machine learning process found that coffee may be important for
predicting heart disease, it didn’t establish a connection to increased
or decreased risk of getting the disease. It simply showed that coffee
was in the top 15 percent of important factors among the hundreds of
factors poured into the mix.
“Once you have an idea that
coffee is important at predicting risk, what you don’t know from machine
learning is whether it is associated with increased risk or decreased
risk, so from there we used more traditional methods to evaluate if
coffee drinking was harmful or protective,” Stevens said.
this case we used survival analyses to evaluate whether people who
drink coffee survived longer, and determined increased coffee
consumption was associated with decreased risk of heart failure and
While coffee consumption was associated with the decreased risk, the study doesn’t show a cause and effect relationship.
specific data analysis doesn’t give you that,” Stevens said. “And
that’s what we’re looking into now. We are curious if it is some other
habit that people who drink coffee have that is related, or if it is the
caffeine or the antioxidants in the coffee itself?”
check their results, Stevens, Kao and their colleagues used traditional
analysis in two other studies with similar sets of data – the
Cardiovascular Heart Study and the Atherosclerosis Risk In Communities
Study. The association between drinking coffee and a decreased risk of
heart failure and stroke was found in all three studies.
Now that the AHA presentation has been completed, Stevens and Kao are working on drafts of papers on the work.
think the tricky part is where to stop with this first paper because
people want to go all the way down the rabbit hole and you just can’t
take on everything at once,” Kao said. “So that’s where we are right now
is trying to decide where to stop with this presentation with an eye
toward what happens next.”
The importance of the study
wasn’t just the finding about coffee, but also a recognition that the
process of discovery could be useful in the design of future studies.
findings suggest that machine learning could help us identify
additional factors to improve existing risk assessment models, said
Stevens, who is a data scientist for the Precision Medicine Institute at
the AHA. “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.”
Stevens conducts research seeking to connect genetic information to lifestyle, environmental, and clinical factors.
think the kind of work that Laura did is the next wave in personalized
medicine,” said Kao, who is a member of the Colorado Center for
Personalized Medicine. “There are so many things that could be
important, that you could modify, that you can’t study in the way we’ve
studied medicine in the past.
“This was an interesting
result in and of itself, but as just interesting is that the method can
work and find novel information in other datasets that are important to
people, and that may represent a potential intervention, positive or
negative. I think without that personalized medicine is going to be very
hard to pull off.”