As you’ve no doubt heard by now, there’s been a new recommendation issues
which proposes changing the breast-cancer screening protocol for women under
50, by eliminating mammograms for women who don’t have significant risk
factos. While Orac has done a terrific job of covering this here and
here, I wanted to throw
in a couple of notes and a personal perspective.
To begin with, there’s a bit of math which has been bandied about, and
I thought I’d just quickly walk through it.
When you look at things like screening programs, what you’re doing is
performing some kind of test on a very large population, in the hopes of
finding a comparatively small number of serious illnesses. Any process
like that is, necessarily a tradeoff.
I’m going to use totally fake numbers to explain this – so don’t
think that these numbers are real.
Suppose that you’ve got a non-contagious disease which will be caught by
one out of every 1,000 individuals. That’s a fairly rare disease. If
it were harmless, you’d completely ignore it – you wouldn’t even bother
spending money on researching a cure for it. It’s just not worth the
Now, suppose that the disease is universally fatal. Only one person out of
every 1,000 will catch it – but all of them will die. In this case,
there’s not much point in trying to figure out who has it – there’s nothing
you can do for them. But you would start spending money to figure out how
to cure it.
Now for the tricky case. Suppose that the disease isn’t universally
fatal. It’s almost always fatal if it’s had time to establish itself. But
if you catch it early enough, then you can with high probability, save the
person who has it. This is the case for cancers where we consider
In this last case, there’s a very complicated tradeoff. Should you
spend time and money trying to detect the disease? Or should you spend
the time and money trying to discover better treatments for the disease?
In fact, the tradeoff is ever worse that that, for two reasons. First, the
screening process has a huge false positive rate. The screening process comes
up positive in 5 out of every 1000 cases. So you wind up treating 4 people who
don’t have the disease. And the treatment isn’t always completely
benign. Most of the time, you do an additional test, and that’s it. But that
additional test has some amount of risk to it – some people will end up dying
as a result of the treatment for a disease that they didn’t have. And second,
the test itself isn’t risk free: it’s got a small but real chance of
causing exactly the disease that it’s being used to detect!
So how do you set a balance? You can screen people – and save some number
of lives by detecting disease that would have killed them had it gone
undetected. But by doing that, you’ll give the disease to some number of
people who wouldn’t have gotten it otherwise; and you’ll harm some number of
people who were caught in the screening process, but didn’t have the disease
It ends up coming down to a mathematical optimization process. You want to
maximize the survival rate of the population as a whole. You do that by
putting together a lot of factors: how many people will get the disease? How
many will miss detection if you don’t do the screening? How many will get the
disease as a result of the screening? How many will be harmed by procedures
done as a result of false positives? And how many could have been saved by
spending money on developing cures instead of screening?
The current situation appears to be that for women under
50 who don’t have other risk factors, it doesn’t make sense for them
to get the screening. The risks and costs of the screening outweigh any
benefits of it.
To shift to the personal side for a moment, I’d like to provide you
with a concrete example of how a false positive can do harm.
Close to 20 years ago now, my father had a muscular cancer in his leg. He
had surgery followed by radiation to have it removed. Everything went
beautifully. It turned out to be a highly aggressive cancer, but the surgery
appeared to have gotten it all. But because it was so aggressive, they wanted
to keep screening, looking for any trace of a recurrence. About two years
after the surgery, they saw something on an x-ray – it was either a
patch of scar tissue right by a vein, or it was the beginning of a new tumor.
They did numerous tests, but nothing was able to determine definitively what
the hell it was. A radiologist from Memorial Sloan-Kettering thought it was
just scar tissue, and recommended waiting. But the radiologist from the
hospital where the surgeon worked was equally sure that it was cancer. So they
decided to do surgery – it was an aggressive cancer with a very low survival
rate; why take a chance on letting it spread? So they went in and removed it.
The second surgery went very badly. It took over a year for the surgical wound
to heal, and there was enough circulation lost during the surgery to kill the
nerves in his leg. After that surgery, he could never feel or move that leg
beneath the knee. For the next 18 years, it caused constant trouble with poor
circulation. A blood clot in the vein affected by the surgery is what
eventually led to his death.
The point of that isn’t to tell you a sad story – but to illustrate the
very real risks of any intervention. The initial surgery – the one for the
confirmed cancer, was a complex procedure, due to the location and type of the
tumor. But the second one was quite routine – removing a one-centimeter,
well-isolated mass from the muscle below the knee. It wasn’t simple, but it
wasn’t by any means a particularly complicated surgery either: it was the sort
of procedure that the surgeon geon did, on average, twice a week! But even a
routine procedure has risks. Even the most routine procedure can, in rare
cases, wind up killing you.
That’s the point of the optimization problem that’s used to figure out
whether or not to screen for a potentially deadly disease: no intervention is
ever free – and I don’t mean that just in terms of money. Every intervention
comes with an associated risk. You have to find out where the balance point is
between the risks that you’re trying to protect people from, and the risks
that you’re going to inflict on people in the process of protecting them.
The best recent evidence, when put into that optimization problem, has
strongly suggested that the mammograms in women under 50 aren’t worth the
risk. The risk of the radiation of a yearly mammogram plus the risks of the
biopsies end up being worse, on average, than not doing the mammogram. The
balance point between risk and benefit doesn’t work out well until you
shift the pool of people being screened to be older – and thus at higher
The natural response to this is to say: but what about the women under 50
who have cancer that would have been detected early enough to be cured? Isn’t
refusing to give them mammograms harming them?
Yes, it is. But giving the mammograms will harm a different
group of women – and it appears that the group harmed by the earlier
mammograms is larger that the group helped by them.
To give a very different example of screening balance decision, one
which again has some personal resonance for me:
Most men will, at some point in their lives, suffer from gastric reflux.
One of the side-effects of severe reflux is esophageal cancer – which is
almost universally fatal.
Esophageal cancer is actually pretty easy to screen for. The
vast majority of cases start as pre-cancerous lesions in the esophagus
long before the cancer develops. Those lesions can easily
be detected by doing an upper endoscopy. Nearly every case of
reflux-driven esophageal cancer is preventable by
So – if we can save all of those people, why don’t we give every man over
40 a yearly endoscopy? Because the endoscopy has risks. They’re not
particularly common – but when they happen, they’re pretty severe (perforation
of the esophagus and/or stomach, infection). Even though those occur
very rarely, they do occur. And the risk of those injuries
caused by the procedure outweigh the benefits of the procedure for men with no
symptoms or risk factors for severe reflux-related disease. (The personal
connection here is that I’ve had 6 endoscopies, and surgery to try to
eliminate the reflux. I’ve also had several relatives die of esophageal cancer
caused by reflux.)
In contrast, we do routinely do colonoscopies to screen for colon cancer.
It’s another common, dangerous cancer. And the risks of doing a colonoscopy
are comparable (if not a bit greater!) to the risks of an endoscopy. But it’s
got no symptoms that we can use to identify people who are likely to develop
it. So we can’t do what we do with endoscopies – which is to select people at
risk based on symptoms. So we routinely screen people when they get to the
right age – because the number of lives saved is less than the number of lives
lost due to complications.
If we started doing colonoscopies at 30 instead of the current recommendation
of 50, we’d save some people from dying of colon cancer. But we’d also hurt
a whole lot of people without colon cancer. So we don’t do it.
It all comes back to the optimization problem: find the optimal point
where the benefits, costs, and risks balance out.