BETHESDA, MdMost oncologists probably never thought
they would need to worry about overdiagnosing cancer. Yet overdiagnosisthe
detection of an actual cancer that will never harm the patientis emerging as
a clinical concern. In this interview, medical oncologist Barnett
"Barry" Kramer, MD, discusses the issue of overdiagnosis with Patrick
Young, ONI’s Washington Bureau Chief.
Dr. Kramer is now director of the Office of Medical
Applications and Research at the National Institutes of Health and
editor-in-chief of the Journal of the National Cancer Institute. He joined NCI,
where he retains an appointment as a medical scientist, in 1986 after 8 years
on the faculty of the University of Florida, and moved to his present NIH
position in 1999.
ONI: What are the major concerns about overdiagnosis?
Dr. Kramer: Cancer treatments have toxicity, which can be very
substantial, and clearly, you don’t want to apply such treatments to people
who don’t need them in the first place. So the main problem of overdiagnosis
is subjecting people to the morbidities and risks of treatment without any real
ONI: Is this common now?
Dr. Kramer: It’s very difficult, outside of randomized
trials, to estimate exactly how much overdiagnosis there is. The more we learn
about the diagnosis of cancer, the more we’re learning that it is going to be
the rule rather than the exception.
As new screening tests come on board, we will be diagnosing a
new spectrum of disease, and some of it will not be as life threatening as the
traditional cancer diagnoses.
In the case of neuroblastoma, the estimates are that most of
the screen-detected cases in infants actually represent overdiagnosis. In the
case of lung cancerwhich is a bit more of a surprise because it is such a
fearsome diseaseit looks like a substantial minority of cases in randomized
trials that were picked up through chest x-ray screenings could have
Prostate cancer has a real potential for overdiagnosis because
there are so many competing causes of death during the age range at which men
are at highest risk of the disease. In prostate cancer, estimates of
overdiagnosis range from about 10% or 15% of screen-detected cases to more than
ONI: Why is the question of over-diagnosis emerging at this
Dr. Kramer: Although it has been proposed for more than 2
decades as a theoretical possibility, it has only been in the last 10 years or
so that we have had evidence to suggest that it is more than simple theory.
The evidence that has emerged from screening programs for
neuroblastoma, both in Canada and Japan, has been compelling. Likewise, it was
only the recent analysis of the long-term follow-up of the Mayo lung project
that raised the serious specter of over-diagnosis in lung cancer. And it has
been raised as a serious issue in prostate cancer since PSA [prostate-specific
antigen] screening became popular in the United States.
ONI: Is there a sense we are experiencing more overdiagnosis
Dr. Kramer: The risk of overdiagnosis increases as the
screening tests become more sensitive. And so, as we have more sensitive
screening tests and they are applied more widely, we will learn more and more
about overdiagnosis of cancer.
ONI: Which is more troublesome for a cancer patient, a false
positive or an overdiagnosis?
Dr. Kramer: Well, a false-positive finding is more easily
detected, and so reaches the level of consciousness more quickly. A patient who
has an abnormal screening test goes on to a workup and possibly biopsy, and if
the biopy is negative, then, relatively quickly, the patient finds that it was
likely to have been a false positive.
A false diagnosis produces immediate concerns and worries to
the patientthe diagnostic workup, the fear that there is cancer, and then
ultimately the possible morbidity of an invasive biopsy. Over-diagnosis, on the
other hand, is almost impossible for the individual person to detect.
A person is diagnosed with cancer and undergoes therapy, and
sometimes it can be quite morbid and even potentially lethal in the case of
some cancer treatments. But there would have been no way for the patient to
know that he or she represented a case of overdiagnosis. It is only in the
context of large, randomized clinical trials that it becomes clear that a
population of patients was over-diagnosed and treated without benefit.
ONI: You mentioned the Mayo lung screening study and the fact
that we are getting more sensitive tests. The argument has been made with
spiral CT scanning that lung cancer is so virulent that overdiagnosis is not
going to be a big problem.
Dr. Kramer: I would say we already have strong and mounting
evidence of overdiagnosis for lung cancer from chest x-rays, and a recent paper
suggests that it could be even more so for spiral CT.
Almost everything we know about cancer is derived from the
natural history and the cases that became symptomatic, which are often rapidly
lethal. So our concept of the lethality of lung cancer comes from centuries of
observation without any screening tests involved.
Suddenly, when a screening test comes along, we see a different
spectrum of disease whose natural history is less well defined, and we are
trying to extrapolate as best we can from the natural history of a different
spectrum of disease. But a new diagnostic test can perturb the natural history
in unpredictable ways.
The Japanese have had a mass screening program with spiral CT
for several years. A recent paper looked at the tumor doubling times of people
who were screened in Japan by spiral CT (Hasegawa: British Journal of
Radiology, December 2000).
Among 18,000 spiral CT scans, there were people who were
diagnosed with tumors that were visible on both chest
x-ray and spiral CT, as well as those diagnosed with tumors visible only on
spiral CT. The mean tumor doubling time was 536 days in the tumors that could
be seen only on spiral CT, compared with 250 days in those that could be seen
on both spiral CT and chest x-ray.
You can see that the mean doubling time of these tumors was
extremely long if they were picked up by spiral CT scans and not seen on chest
x-ray. And so right away that suggests that there is length bias sampling; that
is, the spiral CT scan has a propensity to pick up slower growing cancers than
does chest x-ray.
And with doubling times this long, obviously people can die of
intervening causes and would never have been diagnosed with lung cancer if they
had not been screened.
ONI: We’re moving into an era of molecular screening. What
will happen in terms of overdiagnosis when you can detect cancer before it can
Dr. Kramer: The more sensitive our tests and the more we get
down to the molecular level, where there are accumulating genetic and
epigenetic abnormalities, the more chance there is for over-diagnosis. The
further you move from the end stage of the disease, the harder it is to predict
what will happen.
As we move further and further back into time, we can pick up
these molecular abnormalities that predispose to death from a particular
disease, but our certainty of prediction starts to deteriorate.
At its extreme, we could screen everyone in utero for all known
disease-predisposing genes, and we could theoretically overdose everybody who
has a "pre-existing condition."
For example, if you pick up a BRCA1 mutation in utero, you
still don’t know for a fact that the person will die of breast cancer. That
same person may harbor scores of other disease-predisposing genes, and only one
can cause death. So it is easy to see how you could get to the situation of
ONI: What can and is being done to try to blunt overdiagnosis?
Dr. Kramer: Any new medical intervention should be carefully
studied for both its benefits and its potential harms. It is particularly
important to do so for interventions that will affect healthy peoplenot just
hundreds of thousands, but many millions.
The trick is to make our predictions as specific as possible.
As we increase our sensitivity to pick up problems, we need to do research on
how to improve the specificity, that is, how to correctly identify people who
are not going to have a problem with the disease.
That is very tricky, but there are several possible approaches.
One is to look simultaneously at all the predictors and try to fit them into
statistical models that will help us gauge the effect of multiple gene
It is unlikely that most diseases result from a single gene
alteration. There will be all sorts of modulations by other gene mutations and
by environmental exposures, diet, and lifestyle. The trick is trying to put all
that into one predictive model.
Right now, we’re just at the formative stages of developing
those models. One research program that we hope to use to approach the problem
is the NCI-sponsored Early Detection Research Network, which is developing
multiple markers for early detection.
We also have theoretical statisticians working with the
laboratory scientists to try to figure out predictive models that won’t get
us into the problem of over-diagnosis.