It's been a
tough time for French fries lately. First there was the unpleasant
association with, well, all things French that led to one of
America's favourite foods
being renamed, at least within the Beltway, Freedom Fries. Then
there was the New York
City fat police "request" that restaurants
stop using trans fats for frying fries, a move that many restaurant
owners say will adversely affect the taste of fries. And finally
there was the new scientific study from the Harvard School of Public
Health that suggests that eating French fries as a child might lead
to breast cancer later in life.
The study, by a
team of researchers led by Karin Michels, looked at children's
preschool diets and the subsequent risk of breast cancer using the
long-running Nurses' Health Study which followed 115,195 nurses for
sixteen years. Michels' group studied 582 women with breast cancer
and 1569 breast cancer free women in 1993. Their interest was the
childhood diet of these women and its possible connection with their
risk for breast cancer.
In order to
find out about their diets, the subjects' mothers were asked to
complete a 30 item food frequency questionnaire about their
daughters' eating habits when they were aged 3-5. After correlating
the answers to the questionnaire with the women's health status they
discovered that women who had had as children one extra serving of
French fries per week had a 27% increased risk for breast cancer in
later life. Interestingly, foods like ice cream and hot dogs were
not associated with an increased risk for breast cancer. Commenting
on her study, Michels noted that "This study provides additional
evidence that breast cancer may originate during the early phases of
a woman's life and that eating habits during that phase may be
particularly important to reduce future risk of breast cancer."
Correlation and
Causation
But before we
have mass hysteria about the possible role of French fries in breast
cancer, there are several things about this study which suggest it
probably merits little serious attention. The first of these is the
confusion that appears to surround so much science (and even more
science reporting) between statistical associations and causal
connections. Doctors, for instance, have stethoscopes, but having a
stethoscope doesn't make one a doctor. In certain parts of the world
there is a strong statistical association between having malaria and
not wearing shoes, but failing to wear shoes is not a cause of
malaria.
One of the more
famous of these confusions between association and causation was a
study that appeared in the scientific journal Nature in May, 1999.
The research reported that babies, aged two and younger, who slept
in rooms with a nightlight, were more likely to be myopic as adults
than babies who did not. The scientists who reported the finding
were unable to provide any explanation for this strange connection
between nightlights and adult myopia. A second study, however, which
appeared the following year was able to explain the "connection"
between nightlights and myopia. According to the authors of this
report, it was due to the fact that nearsighted parents were more
likely to use nightlights in their babies' rooms than normal sighted
parents. In other words it was true that near-sighted adults had
slept as babies in rooms with nightlights, but their myopia was not
due to the nightlights but rather to the fact that their parents
were myopic. There was a causal connection at work, but it was not
the one assumed by the researchers. The first worry about the French
fries/breast cancer connection then is that it is an instance of the
nightlight-myopia confusion where it is true that the women with
breast cancer ate French fries but it is not true that eating French
fries was the cause of their breast cancer.
The
Mismeasurement Problem and Recall Bias
This worry is
strengthened by a second problem about this research, namely its
methodology, in this case using a questionnaire to find out what
someone ate decades before. One of epidemiology's fundamental
problems is what might be called its measurement problem. Part of
this revolves around the fact that its data is compiled from a
subject's or a surrogate's memories, often from decades before.
These memories, however, may be extremely unreliable. Many of us
have difficulty remembering what we had for lunch yesterday, let
alone remembering what we gave our children to eat years before. The
mothers in this study were asked to recall with considerable
precision not only what kinds but how much food their daughters ate
some 30-35 years previously.
But beyond
memory itself, there is also the problem of recall bias, where
memory is biased by something else. The classic instance is a study
of diesel exhaust and cancer in which the researchers asked elderly
subjects who were dying from cancer about their exposure to exhaust
and whether they thought it was connected with their cancer. Similar
recall problems have plagued studies on the alleged connection
between second-hand smoke and lung cancer, which are based on
non-smokers -- or in many instances on the relatives of deceased
non-smokers' -- memories of "how much" second-hand smoke they were
exposed to. In the fries/breast cancer study, the mothers were asked
to remember what their daughters ate AFTER they knew that their
daughters had breast cancer, something which could have easily
biased their recollections.
These sorts of
problems suggest how weak the foundations of much epidemiology are.
Science obtains its authority from the fact that its measurements
and its data are precise. Imprecise measurements are measurements
that cannot support scientific conclusions. The difficulty with
recall epidemiology is that one's statistics will be based not on
precise measurements but on years old memories. These memories may
be accurate or they may be inaccurate but there is no way to
know.
Low Statistical
Power
But there is
another problem with the fries/breast cancer study and that is its
terribly low statistical power. The OR (odds ratio) of increased
breast cancer risk from early childhood consumption of French fries
was only 1.27. In other words it was barely statistically
significant. The press, however, reported the increased risk as 27%,
which made the increased risk look enormous to readers who knew
little or nothing about statistics. But if the number -- 1.27 -- is
placed in perspective, its real lack of significance is apparent.
Risk ratios (RR's and OR's) are reported as numbers typically
ranging from 1.00 to 15 or 20. RRs under 2 are considered very weak;
the statistical association is likely due to chance rather than to
their being a real link between, for instance, fries and breast
cancer.
Still another
problem stems from the fact that the study's authors were unable to
provide any explanation as to how French fries would lead to breast
cancer some year years latter, just as the authors of that Nature
study were unable to explain how nightlights lead to adult myopia.
Indeed, lead author Michels, appearing on the Today Show, was unable
to provide any answer, when asked, as to how childhood consumption
of fries might lead to breast cancer. Epidemiologists refer to this
as the biological plausibility test. Epidemiological studies are,
after all, simply statistical reports. For them to have any
scientific authority they must be corroborated by a biological
explanation as to how something happens. The fact that this study
can provide no biological plausibility significantly weakens its
credibility.
Finally, the
fries/breast cancer connection is contradicted by other studies that
have examined the link between breast cancer and French fries. In a
study published in March of this year, researchers at the
Karonlinska Institute in Stockholm looked at the diets of
43,000 women in the Swedish Women's Lifestyle and Health Cohort. The
study found no statistically significant connection between the
intake of fries and increased risk of breast
cancer.
The world is
full of statistical associations like those found between
nightlights and myopia and malaria and being shoeless. No one,
however, really believes that not wearing shoes is the cause of
malaria. The numerous problems plaguing this study suggest that we
should be equally skeptical about French fries and breast
cancer.
John C. Luik is
writing a book about health care policy. He lives in
Canada.