Tuesday, February 10, 2009

Names Can Kill

In my book, I point out that one problem with writing clear code is variable name rot (inspired by Ward Cunningham's discussion of variable names in his 2004 OOPSLA talk). You named a variable something like AccountBalance, but then later had to change the code so that contents of the variable are rounded to the nearest dollar. Someone else then, reading your code as they modify it, fails to realize that AccountBalance really should have been named RoundedAccountBalance, so they perform penny-wise calculations that they should not

When is "Carcinoma" Not Cancer?

As I point out in the book in a footnote, the problem of naming things correctly is not isolated to computer programming. In particular, medicine has naming mistakes that cause enormous problems.

For example, every year, women are diagnosed with Ductal Carcinoma In Situ, or DCIS. Well, we all know that "carcinoma" means "cancer", so it's no wonder that many of these women elect to have their breasts amputated even though the treatment of lumpectomy followed by radiation is just as effective in most cases. There's just one little problem with that: DCIS is not cancer.

People often imagine that cancer is some foreign invader that has horns and a tail when you look at it under a microscope. It's not. Cancer is something going wrong with your own cells, and that "wrong" happens in many stages, so deciding whether or not you have cancer is a judgment call made by a guy you'll never meet who spent a little time (not much -- he has lots of others to process) staring at some of your cells under a microscope.

What DCIS technically is, is a "pre-cancer". The guy with the microscope looked at your cells, said, "well, that's not cancer, but it's definitely funny looking and I think it's real likely to turn into cancer". You certainly should either get it treated, or monitor very frequently for the appearance of cancer.

But the medical community does a lousy job of grappling with this simple question: How many women would avoid breast amputation and elect for lumpectomy followed by radiation instead if DCIS was presented to them as "not cancerous yet" instead of being presented as "Stage Zero breast cancer"? Many doctors do at least take the time to feel a little bad about the astounding percentage of women who elect amputation even though they don't technically have cancer, but the only real result is periodic handwringing in a newspaper article somewhere. Don't get me wrong: electing mastectomy for DCIS can be a rational choice; it's just that the numbers strongly imply it often is an irrational choice, and calling a non-cancer "carcinoma" surely contributes to that irrationality.

The Misnamed "Tumor Cell"

DCIS is a more or less accidental bad naming choice as far as I can tell, but there are other bad naming choices that I suspect arise from a profit motive. One of those is in the news today, and it's called a circulating tumor cell.

Doctors have known for years that people with cancer are real likely to have cancerous cells floating by in the bloodstream. But only recently have people been creating technology that lets someone quantify how many cancer cells you have in your bloodstream in a repeatable way. Remember, deciding whether or not cells are cancerous is a judgment call. A little trick is required to achieve a highly repeatable count of circulating tumor cells.

Here's the trick: there's a particular kind of cell, the epithelial cell, that really shouldn't be floating around in the bloodstream. But if you have a tumor, there's a good chance it will shed these epithelial cells. Aha! Medical science has been getting really good at tagging particular types of cells, so in recent years we've seen the introduction of lab tests that semi-automatically locate and mark those epithelial cells that shouldn't be floating around if you don't have a tumor. Now, a guy still has to look through a microscope, but it's pretty much reduced to a job of counting how many green dots he sees, since the cells of interest have been chemically marked for him.

But wait, it gets even better. An American version of this technology, named CellSearch, managed to get FDA approval, because they showed that (in a very particular situation), if your circulating tumor cell (CTC) count didn't go down shortly after you started chemotherapy, you weren't likely to live as long as the folks whose CTC counts did go down after starting chemotherapy. Cool beans, because this implied a doctor might choose a chemo drug, see that the CTC was not going down, and get at least one more shot at quickly switching to another chemo drug in the hopes it might be more effective.

What's the Naming Problem?

Where does the naming problem come in? Well, as you can see, I've lapsed into saying the phrase "circulating tumor cell" (or "CTC"), which is exactly what the folks who make the CellSearch test want. But is that name really accurate? Remember, this tests works by assuming that an epithelial cell wouldn't be floating around out there unless it was from a tumor. So, the more accurate name (which the CellSearch folks fastidiously avoid) would be "circulating epithelial cell test", because no human actually goes in there and confirms that each of those epithelial cells really is cancerous.

But if that assumption (which is swept under the carpet by using the CTC name) were faulty, how did CellSearch get approved by the FDA? Well, they did a quite impressive study, that's how. Tested a bunch (about 400) of women who came in for a breast biopsy, followed them to see which actually turned out to have cancer, and found that almost nobody who didn't have cancer had a non-zero CTC. Cool beans, because a sample of 400 is about what you need to prove statistical significance... if your sample is representative of the general population. Is it possible that women who are getting a breast biopsy are just not quite representative of the general population of women? I claim it is.

See, we used the CellSearch test on my wife after her breast cancer treatment. Comfortingly, her CTC descended eventually to 0 or 1. Hooray! Until one year, the test came back 12 (very roughly speaking, a "bad" CTC count is 5 or bigger). We were using the test well outside its FDA-mandated application at that point (remember, it was only approved for use trying to guess if your first chemo choice was working), but there was no way around the fact that it looked like bad news. Hoping for the best, I wondered: was there any way those 12 circulating "tumor" cells were actually just circulating epithelial cells that weren't cancerous at all? In fact, even though the CellSearch folks had a great study that said the answer was "no", the answer was "yes".

As I re-educated myself on the whole CTC literature, I discovered that the Europeans have a competing technology, called MAINTRAC. But they had done a fascinating study that the CellSearch folks seem to have no interest in replicating. They wanted to look at the (scary, sorry) idea that cancer surgery itself can send tumor cells out into the blood, possibly raising the odds (but probably not hugely) of the patient getting a deadly metastasis. They tracked the CTC count in the hours after breast cancer surgery and, not to their surprise, found that they went up after surgery. But the part of the study that interested me was the control: they did the same CTC tracking on a benign case, a patient who got surgery but didn't have breast cancer. The CTC number peaked at over 50,000 in that patient. And now I'm back at the naming problem, because if your "circulating tumor cell" test reports a number of 50,000 in a patient that has no tumor, then maybe that's not quite the right name for it. Solution? If you read that paper, you'll see that they are careful to say "circulating epithelial cell" and not "circulating tumor cell". Now, even though they have a test just like CellSearch, the Europeans don't call it a circulating tumor cell test, because they know that name can be misleading. They know that what the test actually counts is epithelial cells, and know of at least one situation where those epithelial cells don't point to cancer at all.

But back to the cliffhanger! My wife gets the scary CTC of 12. Did she have a deep cut that could mimic the epithelial cell mobilization seen in surger? No, but it dawned on me. She exercises. Hard. In fact, she runs a lot of half-marathons, and those come with blisters and who-knows-what kinds of internal damage. So I proposed a little test. No exercise for 2 weeks straight, followed by another CellSearch "CTC" test. The result? Tada! A count of zero. Now what are the odds that you can pick 400 women who just got a breast biopsy, and none of them had run a half-marathon within the previous week? I'm guessing the odds are better than you think, because a) women getting a breast biopsy tend to be older and b) they had time to get real worried about having cancer and cancel any big events (like a half-marathon) between when the doctor found something suspicious and when the biopsy actually got scheduled and performed.

So my hypothesis is (backed up by one data point) that, in contrast to the underlying assertion of the name "circulating tumor cell", the CellSearch test is easy to generate false positives for.

Your Turn, Men

CellSearch is in the news today because it was used in a study of prostate cancer patients. I was dismayed to see that all the news reporting I could find faithfully repeated the name "circulating tumor cell", and absolutely, completely reinforced the idea that this test has no false positives -- even though the Europeans have proven that false. Why does it matter? If the test gets widespread use, some cancer patient is going to do something to get a false positive (exercise too hard? cut themselves shaving? who knows?). And their doctor is going to make a potentially devastating medical decision based on that false positive, a decision they might be much slower to make if the name of the test were not so inaccurate.

Names can kill.


rmosquin said...


Ron Burk said...

Very sorry to hear your wife has stage 4. But of course, if you have to choose between seeing improvement on scans or feeling better and seeing improvement on tumor markers, you definitely want the former!

Of all the areas of cancer I've studied, tumor markers are one of the toughest. Let's face it, they are messy, poorly understood, and highly imprecise -- but also can be the best indicator we've got if they give a really clear signal.

If the blood could give us a little number that tells us whether cancer is getting worse or better, how wonderful would that be? You could then just try one experiment after another and watch the little number to see if it's helping or hurting. We might have cancer licked in no time (or at least extend life way better than many billion-dollar drugs).

Unfortunately, the current set of tumor markers we have are nowhere near that precise. In fact, we often don't have a good answer to the question: "If treatment X makes tumor marker Y go down, does that guarantee it is reducing the cancer?" Clearly there are cases where the answer is "no".

As it happens, my wife recently had her highest post-treatment elevation of CA 27.29. Not high enough to represent, say, a greater than 80% chance of metastasis, but high enough the oncologist was willing to reach for a PET scan which, thankfully, showed no new disease (though of course, PET scans can miss centimeter sized tumors!).

In reviewing what was happening, I realized we had cut her Vitamin B dose in half a couple months earlier. She was on the Vitamin B because of the Indian research showing that you can reduce tumor markers in breast cancer survivors using a combination of Coenzyme Q10, riboflavin, and niacin (aka, the CORN protocol). The latter two molecules are part of the Vitamin B family.

Sure enough, we put the Vitamin B dose back up (to a much larger dose than the Indians use, FWIW), took another CA 27.29 measurement, and it seems to be going back down. Which sounds really nice until you realize we don't know the answer to this question: does using CORN to decrease breast cancer antigen numbers actually reduce the odds of metastasis? Obviously, we really, really hope so, but this is simply not part of the tiny island of cancer knowledge for which there is good, well-established proof.

More and more often, I think of cancer primarily as a challenge to our ability to think clearly.