BusinessDay 17 Dec 2017
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Sunday <strong>17</strong> <strong>Dec</strong>ember 20<strong>17</strong><br />
C002D5556<br />
SUNDAY<br />
BD<br />
45<br />
Health&Science<br />
Even brain images can be biased<br />
...Study samples that are too rich and too well-educated may give a biased picture of brain development<br />
BETHANY BROOKSHIRE<br />
An astonishing number<br />
of things that scientists<br />
know about<br />
brains and behavior<br />
are based on small<br />
groups of highly educated, mostly<br />
white people between the ages<br />
of 18 and 21. In other words, those<br />
conclusions are based on college<br />
students.<br />
College students make a convenient<br />
study population when<br />
you’re a researcher at a university.<br />
It makes for a biased sample, but<br />
one that’s still useful for some<br />
types of studies. It would be easy to<br />
think that for studies of, say, how<br />
the typical brain develops, a brain<br />
is just a brain, no matter who’s<br />
skull its resting in. A biased sample<br />
shouldn’t really matter, right?<br />
Wrong. Studies heavy in rich,<br />
well-educated brains may provide<br />
a picture of brain development<br />
that’s inaccurate for the<br />
American population at large, a<br />
recent study found. The results<br />
provide a strong argument for<br />
scientists to pay more attention<br />
to who, exactly, they’re studying<br />
in their brain imaging experiments.<br />
It’s “a solid piece of evidence<br />
showing that those of us in neuroimaging<br />
need to do a better<br />
job thinking about our sample,<br />
where it’s coming from and who<br />
we can generalize our findings<br />
to,” says Christopher Monk, who<br />
studies psychology and neuroscience<br />
at the University of Michigan<br />
in Ann Arbor.<br />
The new study is an example<br />
of what happens when epidemiology<br />
experiments — studies of<br />
patterns in health and disease<br />
— crash into studies of brain<br />
imaging. “In epidemiology we<br />
think about sample composition<br />
a lot,” notes Kaja LeWinn, an<br />
epidemiologist at the University<br />
of California in San Francisco.<br />
Who is in the study, where they<br />
live and what they do is crucial to<br />
finding out how disease patterns<br />
spread and what contributes to<br />
good health. But in conversations<br />
with her colleagues in psychiatry<br />
about brain imaging, LeWinn<br />
realized they weren’t thinking<br />
very much about whose brains<br />
they were looking at. Particularly<br />
when studying healthy populations,<br />
she says, there was an idea<br />
that “a brain is a brain is a brain.”<br />
But that’s a dangerous assumption.<br />
“The brain does not exist in a<br />
vacuum, destined to follow some<br />
predetermined developmental<br />
pathway without any deviation,”<br />
LeWinn says. “Quite the opposite,<br />
our brains, especially in early<br />
life, are exquisitely sensitive to<br />
environmental cues, and these<br />
cues shape how we develop.” She<br />
wondered whether the sampling<br />
used in brain imaging studies<br />
might affect the results scientists<br />
were seeing.<br />
To find out, LeWinn and her<br />
colleagues turned to the Pediatric<br />
Imaging, Neurocognition<br />
and Genetics — or PING — study.<br />
“It’s probably the best study we<br />
have of pediatric brain imaging,”<br />
she says.<br />
Conducted across eight cities<br />
(including San Diego, New York<br />
and Honolulu), the study included<br />
more than 1,000 children<br />
from ages of 3 to 20. It recorded<br />
information about the children’s<br />
genetics, mental development<br />
and emotional function. And of<br />
course, it contains lots of images<br />
of their brains. The goal was to<br />
gain a comprehensive set of data<br />
on how children’s brain develop<br />
over time.<br />
The PING database is large,<br />
well-organized and free for any<br />
scientists to look at. LeWinn and<br />
her colleagues examined the<br />
dataset for the race, sex, parental<br />
education and household income<br />
of its participants.<br />
The end sample of 1,162 brains<br />
was a bit more diverse than the<br />
U.S. population. According to the<br />
2010 census, the U.S. population<br />
is about 70 percent white, 14<br />
percent black and 7.5 percent<br />
Hispanic. By contrast, the racial<br />
breakdown of the PING study<br />
was 42 percent white, 10 percent<br />
black and 24 percent Hispanic,<br />
with a larger percentage of “other”<br />
or mixed-race participants.<br />
“It was more diverse. That’s<br />
not common,” LeWinn says. This<br />
could be because the study sites<br />
were in large cities with diverse<br />
populations, she notes.<br />
The PING study participants<br />
weren’t like the average American<br />
in other ways as well. The<br />
children were from richer households<br />
than Americans in general,<br />
and their parents were more<br />
highly educated. While only 11<br />
percent of Americans have a<br />
post-college education, 35 percent<br />
of the PING study’s children had<br />
parents who had attended graduate<br />
school.<br />
So LeWinn and her colleagues<br />
set out to make the data in the<br />
PING study look more like the<br />
data from the U.S. population as<br />
a whole. They applied sample<br />
weights to the brain imaging data,<br />
giving more weight to the brains<br />
of kids with poorer, less educated<br />
families, and adding additional<br />
weights to match the racial demographics<br />
of the United States.<br />
In the newly weighted data,<br />
LeWinn and her group noticed<br />
that children’s brains matured<br />
more quickly. The cortex of the<br />
brain reached a peak surface area<br />
2.4 years earlier than the original<br />
data would have suggested. Some<br />
brains areas — such as the amygdala,<br />
an area associated with<br />
emotional processing — appeared<br />
to reach maturity a full four<br />
years faster. “Low socioeconomic<br />
status is associated with faster<br />
brain development, so that’s one<br />
potential explanation,” LeWinn<br />
notes. The group reported their<br />
findings October 12 in Nature<br />
Communications.<br />
Unfortunately, this study can’t<br />
tell scientists if children’s brains<br />
actually are maturing faster than<br />
we think they are. The weighted<br />
sample isn’t a representation of<br />
what average brain development<br />
looks like in the United States.<br />
Instead, it’s just closer to what it<br />
might look like. “I would like to<br />
see this replicated in an actual<br />
sample of people who do represent<br />
the population,” says Kate<br />
Mills, a cognitive neuroscientist<br />
at the University of Oregon in<br />
Eugene.<br />
But brain development wasn’t<br />
the point. Instead, the point is to<br />
show that when there’s a bias in<br />
the sample of participants in a<br />
brain imaging study, the data are<br />
biased, too. Even a large sample<br />
may not provide an accurate<br />
picture of brain development — if<br />
that sample has biases of its own.<br />
It’s a strong argument for an<br />
unbiased sample, no matter the<br />
type of study. “It’s illustrating the<br />
impact of sample composition on<br />
these measures,” Mills says. “It’s<br />
not something we can disregard<br />
anymore.” She’s optimistic that<br />
change is nigh. “The datasets being<br />
collected now [in brain imaging<br />
studies] are already taking this<br />
more seriously.”<br />
But it can be difficult to get<br />
study volunteers who represent<br />
a particular population. “A representative<br />
sample is expensive and<br />
challenging,” Monk notes. For his<br />
own recent brain imaging work,<br />
Monk has teamed up with a large<br />
existing project to get a larger<br />
sample, but even then, he says,<br />
“it’s still questionable whether<br />
or not the sample can be made<br />
representative.” People may not<br />
respond to the call. Volunteers<br />
may not show up. But unless scientists<br />
put in the extra legwork<br />
to make sure those people are<br />
accounted for, our picture of how<br />
human brains work won’t apply<br />
to everyone.