2.
The head of breast imaging at Memorial Sloan-Kettering Cancer Center, in New York City, is a physician named David Dershaw, a youthful man in his fifties, who bears a striking resemblance to the actor Kevin Spacey. One morning not long ago, he sat down in his office at the back of the Sloan-Kettering Building and tried to explain how to read a mammogram.
Dershaw began by putting an X-ray on a light box behind his desk. “Cancer shows up as one of two patterns,” he said. “You look for lumps and bumps, and you look for calcium. And, if you find it, you have to make a determination: is it acceptable, or is it a pattern that might be due to cancer?” He pointed at the X-ray. “This woman has cancer. She has these tiny little calcifications. Can you see them? Can you see how small they are?” He took out a magnifying glass and placed it over a series of white flecks; as a cancer grows, it produces calcium deposits. “That’s the stuff we are looking for,” he said.
Then Dershaw added a series of slides to the light box and began to explain all the varieties that those white flecks came in. Some calcium deposits are oval and lucent. “They’re called eggshell calcifications,” Dershaw said. “And they’re basically benign.” Another kind of calcium runs like a railway track on either side of the breast’s many blood vessels — that’s benign, too. “Then there’s calcium that’s thick and heavy and looks like popcorn,” Dershaw went on. “That’s just dead tissue. That’s benign. There’s another calcification that’s little sacs of calcium floating in liquid. It’s called ‘milk of calcium.’ That’s another kind of calcium that’s always benign.” He put a new set of slides against the light. “Then we have calcium that looks like this — irregular. All of these are of different density and different sizes and different configurations. Those are usually benign, but sometimes they are due to cancer. Remember you saw those railway tracks? This is calcium laid down inside a tube as well, but you can see that the outside of the tube is irregular. That’s cancer.” Dershaw’s explanations were beginning to be confusing. “There are certain calcifications in benign tissues that are always benign,” he said. “There are certain kinds that are always associated with cancer. But those are the ends of the spectrum, and the vast amount of calcium is somewhere in the middle. And making that differentiation, between whether the calcium is acceptable or not, is not clear-cut.”
The same is true of lumps. Some lumps are simply benign clumps of cells. You can tell they are benign because the walls of the mass look round and smooth; in a cancer, cells proliferate so wildly that the walls of the tumor tend to be ragged and to intrude into the surrounding tissue. But sometimes benign lumps resemble tumors, and sometimes tumors look a lot like benign lumps. And sometimes you have lots of masses that, taken individually, would be suspicious but are so pervasive that the reasonable conclusion is that this is just how the woman’s breast looks. “If you have a CAT scan of the chest, the heart always looks like the heart, the aorta always looks like the aorta,” Dershaw said. “So when there’s a lump in the middle of that, it’s clearly abnormal. Looking at a mammogram is conceptually different from looking at images elsewhere in the body. Everything else has anatomy — anatomy that essentially looks the same from one person to the next. But we don’t have that kind of standardized information on the breast. The most difficult decision I think anybody needs to make when we’re confronted with a patient is: Is this person normal? And we have to decide that without a pattern that is reasonably stable from individual to individual, and sometimes even without a pattern that is the same from the left side to the right.”
Dershaw was saying that mammography doesn’t fit our normal expectations of pictures. In the days before the invention of photography, for instance, a horse in motion was represented in drawings and paintings according to the convention of ventre à terre, or “belly to the ground.” Horses were drawn with their front legs extended beyond their heads, and their hind legs stretched straight back, because that was the way, in the blur of movement, a horse seemed to gallop. Then, in the 1870s, came Eadweard Muybridge, with his famous sequential photographs of a galloping horse, and that was the end of ventre à terre. Now we knew how a horse galloped. The photograph promised that we would now be able to capture reality itself.
The situation with mammography is different. The way in which we ordinarily speak about calcium and lumps is clear and unambiguous. But the picture demonstrates how blurry those seemingly distinct categories actually are. Joann Elmore, a physician and epidemiologist at the University of Washington Harborview Medical Center, once asked ten board-certified radiologists to look at 150 mammograms — of which 27 had come from women who developed breast cancer, and 123 from women who were known to be healthy. One radiologist caught 85 percent of the cancers the first time around. Another caught only 37 percent. One looked at the same X-rays and saw suspicious masses in 78 percent of the cases. Another doctor saw “focal asymmetric density” in half of the cancer cases; yet another saw no “focal asymmetric density” at all. There was one particularly perplexing mammogram that three radiologists thought was normal, two thought was abnormal but probably benign, four couldn’t make up their minds about, and one was convinced was cancer. (The patient was fine.) Some of these differences are a matter of skill, and there is good evidence that with more rigorous training and experience radiologists can become better at reading breast X-rays. But so much of what can be seen on an X-ray falls into a gray area that interpreting a mammogram is also, in part, a matter of temperament. Some radiologists see something ambiguous and are comfortable calling it normal. Others see something ambiguous and get suspicious.
Does that mean radiologists ought to be as suspicious as possible? You might think so, but caution simply creates another kind of problem. The radiologist in the Elmore study who caught the most cancers also recommended immediate workups — a biopsy, an ultrasound, or additional X-rays — on 64 percent of the women who didn’t have cancer. In the real world, a radiologist who needlessly subjected such an extraordinary percentage of healthy patients to the time, expense, anxiety, and discomfort of biopsies and further testing would find himself seriously out of step with his profession. Mammography is not a form of medical treatment, where doctors are justified in going to heroic lengths on behalf of their patients. Mammography is a form of medical screening: it is supposed to exclude the healthy, so that more time and attention can be given to the sick. If screening doesn’t screen, it ceases to be useful.
Gilbert Welch, a medical-outcomes expert at Dartmouth Medical School, has pointed out that, given current breast-cancer mortality rates, nine out of every thousand sixty-year-old women will die of breast cancer in the next ten years. If every one of those women had a mammogram every year, that number would fall to six. The radiologist seeing those thousand women, in other words, would read ten thousand X-rays over a decade in order to save three lives — and that’s using the most generous possible estimate of mammography’s effectiveness. The reason a radiologist is required to assume that the overwhelming number of ambiguous things are normal, in other words, is that the overwhelming number of ambiguous things really are normal. Radiologists are, in this sense, a lot like baggage screeners at airports. The chances are that the dark mass in the middle of the suitcase isn’t a bomb, because you’ve seen a thousand dark masses like it in suitcases before, and none of those were bombs — and if you flagged every suitcase with something ambiguous in it, no one would ever make his flight. But that doesn’t mean, of course, that it isn’t a bomb. All you have to go on is what it looks like on the X-ray screen — and the screen seldom gives you quite enough information.
3.
Dershaw picked up a new X-ray and put it on the light box. It belonged to a forty-eight-year-old woman. Mammograms indicate density in the breast: the denser the tissue is, the more the X-rays are absorbed, creating the variations in black and white that make up the picture. Fat hardly absorbs the beam at all, so it shows up as black. Breast tissue, particularly the thick breast tissu
e of younger women, shows up on an X-ray as shades of light gray or white. This woman’s breasts consisted of fat at the back of the breast and more dense, glandular tissue toward the front, so the X-ray was entirely black, with what looked like a large, white, dense cloud behind the nipple. Clearly visible, in the black, fatty portion of the left breast, was a white spot. “Now, that looks like a cancer, that little smudgy, irregular, infiltrative thing,” Dershaw said. “It’s about five millimeters across.” He looked at the X-ray for a moment. This was mammography at its best: a clear picture of a problem that needed to be fixed. Then he took a pen and pointed to the thick cloud just to the right of the tumor. The cloud and the tumor were exactly the same color. “That cancer only shows up because it’s in the fatty part of the breast,” he said. “If you take that cancer and put it in the dense part of the breast, you’d never see it, because the whiteness of the mass is the same as the whiteness of normal tissue. If the tumor was over there, it could be four times as big and we still wouldn’t see it.”
What’s more, mammography is especially likely to miss the tumors that do the most harm. A team led by the research pathologist Peggy Porter analyzed 429 breast cancers that had been diagnosed over five years at the Group Health Cooperative of Puget Sound. Of those, 279 were picked up by mammography, and the bulk of them were detected very early, at what is called Stage One. (Cancer is classified into four stages, according to how far the tumor has spread from its original position.) Most of the tumors were small, less than two centimeters. Pathologists grade a tumor’s aggression according to such measures as the “mitotic count” — the rate at which the cells are dividing — and the screen-detected tumors were graded “low” in almost 70 percent of the cases. These were the kinds of cancers that could probably be treated successfully. “Most tumors develop very, very slowly, and those tend to lay down calcium deposits — and what mammograms are doing is picking up those calcifications,” Leslie Laufman, a hematologist-oncologist in Ohio, who served on a recent National Institutes of Health breast-cancer advisory panel, said. “Almost by definition, mammograms are picking up slow-growing tumors.”
A hundred and fifty cancers in Porter’s study, however, were missed by mammography. Some of these were tumors the mammogram couldn’t see — that were, for instance, hiding in the dense part of the breast. The majority, though, simply didn’t exist at the time of the mammogram. These cancers were found in women who had had regular mammograms, and who were legitimately told that they showed no sign of cancer on their last visit. In the interval between X-rays, however, either they or their doctor had manually discovered a lump in their breast, and these “interval” cancers were twice as likely to be in Stage Three and three times as likely to have high mitotic counts; 28 percent had spread to the lymph nodes, as opposed to 18 percent of the screen-detected cancers. These tumors were so aggressive that they had gone from undetectable to detectable in the interval between two mammograms.
The problem of interval tumors explains why the overwhelming majority of breast-cancer experts insist that women in the critical fifty-to-sixty-nine age group get regular mammograms. In Porter’s study, the women were X-rayed at intervals as great as every three years, and that created a window large enough for interval cancers to emerge. Interval cancers also explain why many breast-cancer experts believe that mammograms must be supplemented by regular and thorough clinical breast exams. (Thorough is defined as palpation of the area from the collarbone to the bottom of the rib cage, one dime-size area at a time, at three levels of pressure — just below the skin, the midbreast, and up against the chest wall — by a specially trained practitioner for a period not less than five minutes per breast.) In a major study of mammography’s effectiveness — one of a pair of Canadian trials conducted in the 1980s — women who were given regular, thorough breast exams but no mammograms were compared with those who had thorough breast exams and regular mammograms, and no difference was found in the death rates from breast cancer between the two groups. The Canadian studies are controversial, and some breast-cancer experts are convinced that they may have understated the benefits of mammography. But there is no denying the basic lessons of the Canadian trials: that a skilled pair of fingertips can find out an extraordinary amount about the health of a breast, and that we should not automatically value what we see in a picture over what we learn from our other senses.
“The finger has hundreds of sensors per square centimeter,” says Mark Goldstein, a sensory psychophysicist who cofounded MammaCare, a company devoted to training nurses and physicians in the art of the clinical exam. “There is nothing in science or technology that has even come close to the sensitivity of the human finger with respect to the range of stimuli it can pick up. It’s a brilliant instrument. But we simply don’t trust our tactile sense as much as our visual sense.”
4.
On the night of August 17, 1943, two hundred B-17 bombers from the United States Eighth Air Force set out from England for the German city of Schweinfurt. Two months later, 228 B-17s set out to strike Schweinfurt a second time. The raids were two of the heaviest nights of bombing in the war, and the Allied experience at Schweinfurt is an example of a more subtle — but in some cases more serious — problem with the picture paradigm.
The Schweinfurt raids grew out of the United States military’s commitment to bombing accuracy. As Stephen Budiansky writes in his wonderful recent book Air Power, the chief lesson of aerial bombardment in the First World War was that hitting a target from eight or ten thousand feet was a prohibitively difficult task. In the thick of battle, the bombardier had to adjust for the speed of the plane, the speed and direction of the prevailing winds, and the pitching and rolling of the plane, all while keeping the bombsight level with the ground. It was an impossible task, requiring complex trigonometric calculations. For a variety of reasons, including the technical challenges, the British simply abandoned the quest for precision: in both the First World War and the Second, the British military pursued a strategy of morale or area bombing, in which bombs were simply dropped, indiscriminately, on urban areas, with the intention of killing, dispossessing, and dispiriting the German civilian population.
But the American military believed that the problem of bombing accuracy was solvable, and a big part of the solution was something called the Norden bombsight. This breakthrough was the work of a solitary, cantankerous genius named Carl Norden, who operated out of a factory in New York City. Norden built a fifty-pound mechanical computer called the Mark XV, which used gears and wheels and gyroscopes to calculate airspeed, altitude, and crosswinds in order to determine the correct bomb-release point. The Mark XV, Norden’s business partner boasted, could put a bomb in a pickle barrel from twenty thousand feet. The United States spent $1.5 billion developing it, which, as Budiansky points out, was more than half the amount that was spent building the atomic bomb. “At air bases, the Nordens were kept under lock and key in secure vaults, escorted to their planes by armed guards, and shrouded in a canvas cover until after takeoff,” Budiansky recounts. The American military, convinced that its bombers could now hit whatever they could see, developed a strategic approach to bombing, identifying, and selectively destroying targets that were critical to the Nazi war effort. In early 1943, General Henry (Hap) Arnold — the head of the Army Air Forces — assembled a group of prominent civilians to analyze the German economy and recommend critical targets. The Advisory Committee on Bombardment, as it was called, determined that the United States should target Germany’s ball-bearing factories, since ball bearings were critical to the manufacture of airplanes. And the center of the German ball-bearing industry was Schweinfurt. Allied losses from the two raids were staggering. Thirty-six B-17s were shot down in the August attack, 62 bombers were shot down in the October raid, and between the two operations, a further 138 planes were badly damaged. Yet, with the war in the balance, this was considered worth the price. When the damage reports came in, Arnold exulted, “Now we have got Schweinfurt!” He was
wrong.
The problem was not, as in the case of the Scud hunt, that the target could not be found, or that what was thought to be the target was actually something else. The B-17s, aided by their Norden Mark XVs, hit the ball-bearing factories hard. The problem was that the picture Air Force officers had of their target didn’t tell them what they really needed to know. The Germans, it emerged, had ample stockpiles of ball bearings. They also had no difficulty increasing their imports from Sweden and Switzerland, and, through a few simple design changes, they were able to greatly reduce their need for ball bearings in aircraft production. What’s more, although the factory buildings were badly damaged by the bombing, the machinery inside wasn’t. Ball-bearing equipment turned out to be surprisingly hardy. “As it was, not a tank, plane, or other piece of weaponry failed to be produced because of lack of ball bearings,” Albert Speer, the Nazi production chief, wrote after the war. Seeing a problem and understanding it, then, are two different things.
In recent years, with the rise of highly accurate long-distance weaponry, the Schweinfurt problem has become even more acute. If you can aim at and hit the kitchen at the back of a house, after all, you don’t have to bomb the whole building. So your bomb can be two hundred pounds rather than a thousand. That means, in turn, that you can fit five times as many bombs on a single plane and hit five times as many targets in a single sortie, which sounds good — except that now you need to get intelligence on five times as many targets. And that intelligence has to be five times more specific, because if the target is in the bedroom and not the kitchen, you’ve missed him.