Authors: Jerome Groopman
Mammography is routinely ordered by primary care physicians as a screening test to detect early cancer in women entering middle age. "Mammograms are the most monotonous type of work that we do," Orwig said. "And mammograms are the most anxiety provoking of all x-rays," he added. To miss a cancer is devastating, because the tumors that are found early are readily removed, and missing the cancer can result in metastases that are hard to control and rarely, if ever, cured. On the other hand, overreading a mammogram will subject a healthy woman to the emotional roller coaster of further imaging, a biopsy, and then the lingering doubt about whether there actually was a cancer that was missed despite the biopsy result.
Not surprisingly, mammography is a fertile field for medical-legal conflict, and radiologists are acutely aware that errors can result in a malpractice lawsuit. Even the best radiologists will inaccurately read a mammogram in 2 to 3 percent of the cases, while some series show that other doctors incorrectly read the images in 20 percent or more. The aim is to recommend a biopsy on the women who will prove to have a tumor, and not to recommend a biopsy for women with benign changes on the mammogram. The women who undergo a biopsy are said to be "called back." "In theory, it would be best to have a four or five percent callback rate," Orwig said. This is considered to be the optimal rate. "But the norm," Orwig said, "is about ten to eleven percent." This higher callback rate results in a larger number of women with benign changes who undergo further evaluation and biopsy.
There is a tradeoff here: causing emotional distress in women with benign changes versus the need to "capture" a number of breast cancers that otherwise would be missed. In Orwig's group of eleven radiologists, he falls in the callback rate of 10 to 11 percent, the norm, but one colleague has a 15 to 16 percent callback rate. Many of the women he calls back end up having benign biopsies. "He was sued," Orwig told me. "Years ago, he missed a breast cancer." This experience caused him to become more "aggressive," as Orwig put it, in assessing mammograms and calling more women back for further studies and biopsy. While his colleague's callback rate is still within "reasonable bounds," Orwig said, there is no doubt that the consequences of missing a malignant lesion and being sued caused him to think in a different way.
Dr. Potchen published a paper analyzing medical decisionmaking and concluded that what most influenced clinical choices was "the last bad experience." Potchen's conclusion mirrors the availability error, which Croskerry and Redelmeier highlighted earlier: what is most available in your mind strongly colors your thinking about a new case that has some similarities, but it can cause you to ignore important differences and come to an incorrect diagnosis.
Imaging the body using x-rays began at the end of the nineteenth century. Over the ensuing years, advances in technology such as the CT scan and the MRI scan have greatly enhanced a radiologist's ability to visualize our anatomy, but they pose new cognitive challenges. As we saw earlier, a chest x-ray shows a single, static view of the thorax along with the heart, lungs, bones, soft tissues, and mediastinum, the area of the chest that includes the aorta. The chest x-ray, again, is performed with a patient in two positions, so that two images are generated, one with a view of the front of the chest, another of the side. When CT scans were first developed, they contained dozens of images; the first MRI scans produced hundreds. When the images from these scans were presented on a film, they averaged twelve per film; this was termed a "tile display." Another way to view the images from these scans was one after another on a monitor, called a "cine" mode, referring to the experience of viewing a movie.
A landmark study compared tile with cine viewing of CT images of the chest, specifically the detection of lung nodules. These nodules are small, solid masses in the lungs that can be benign, such as after an infection, or malignant, indicating a lung cancer or a metastasis to the lungs. Radiologists were much more successful in detecting lung nodules using the cine mode than the tile display; they were also more accurate in identifying artifacts, such as blood vessels in the lungs that can sometimes resemble a nodule if viewed in a certain plane. The researchers noted that moving images created a novel cognitive cue, particularly with smaller nodules: they seemed to "pop out" on the cine display, and thus more frequently caught the eye of the radiologist. Throughout the 1990s, cine viewing became the preferred image display mode, allowing more efficient evaluation of scans that contained hundreds of images.
But as refinements were made in CT scanning, particularly rapid imaging over large parts of the body that allowed for the visualization of multiple organs and vessels simultaneously, radiologists were presented with a dilemma. There could be a thousand or more images with many tissues moving at once before the radiologist's gaze. As Dr. Herbert Kressel, the Stoneman Professor of Radiology at Harvard and a specialist in abdominal imaging, recently told me, "Over an average weekend in a busy emergency department, a radiologist may have to read a hundred fifty CT scans. It's impossible for him to look at a hundred fifty thousand or more images."
It is not only in the ER that scans have become routine. Between 1998 and 2002, the number of CT studies in the United States increased by 59 percent, MRI by 51 percent, and ultrasound by 30 percent, during traditional work hours, and each increased by 15 percent during on-call off-hours. Survey studies show that with this increased workload, radiologists reported more symptoms of blurred vision, eyestrain, difficulty focusing, and headache. Elizabeth Krupinski of the University of Arizona pointed out in a recent publication that the huge amount of imaging data generated has the high potential for fatigue, discontent, and possibly increased error rates.
Kressel extends this concern to the naïve reliance on high-tech imaging by busy clinicians. Patients are sometimes sent for studies without a detailed clinical history. Some doctors scribble "rule out pathology" on the referral slip. Others can be too directive, or show a poor understanding of how the newer CT and MRI scans work. For example, Kressel told me, recently a woman had been sent for an imaging study with the request "rule out pulmonary embolus." Pulmonary embolus is a life-threatening condition when a clot, often from the legs, breaks off and lodges in the pulmonary artery. "We performed the study and timed the images when the contrast material would be filling the pulmonary vessels," Kressel told me. The study showed nothing to suggest a pulmonary embolus, since there was no blunting or obstruction of the contrast material in the pulmonary arteries. A few days later, the reason for the woman's chest pain and difficulty breathing became apparent: she had a tear in her aorta. "We could have detected this tear by timing the images on the scan to follow the contrast material through the aorta," Kressel said. The shape of the aorta was not distorted, so the tear was not visualized in the absence of contrast material that would enter the torn part of the wall.
"The old idea based on a static study—that an image is an image is an image—is obsolete now with technology that is dynamic, that can show us active changes in blood flow and other aspects of physiology," Kressel said. "How you use the machine translates into what you get to see."
The problem with applying a methodical and rigorous approach to every image, Kressel said, is that with a CT scan or an MRI, "there is just so much data." There may be more than a thousand images per CT scan or per MRI, so that a single radiologist's entire day could well be occupied with looking at just one of these studies. The solution, in part, Kressel said, is to go organ by organ. In his field, abdominal radiology, he will rigorously view the liver, then the kidneys, then the spleen, and so on. Kressel's strategy, like other radiologists who specialize in MRI, is to analyze primarily the "data-rich sequences," those segments of the study that can provide the most information, then arrive at a tentative list of possible diagnoses and selectively look at other images to find data that support or contradict the possibilities.
Sometimes the technology itself sabotages this careful approach. The tracker ball, equivalent to a mouse on a computer, can accelerate rapidly the speed of the images flashing before the radiologist's eyes with small increases in pressure from his fingers. This means that the images may speed up without the radiologist's being aware of it. "You may not get each image coming at three per second," Kressel said. "You go to an emergency room, where a radiologist is working under an incredible load, and you watch him go through a CT scan case, and you will see he is flying through it." Kressel has seen how radiologists press on the tracker ball, consciously or subconsciously, in an effort to get through the thousand or so images. "With your hand on a tracker ball, you can literally skip three or four images and not even realize it." This is because while you are looking at images in two dimensions, "your mind is working to integrate the three-dimensional space that you are moving through." When Kressel supervises residents, "I drive them nuts. I force them to slow down so they have to see each image." There should be no shortcuts, either intended or subliminal. After decades of looking at images, Kressel's eye has become so acute that he can pick up "subtle differences in contour of a structure in the abdomen, even with a rapid look," he said, which a resident cannot do. For example, there may be a lymph node near the pancreas that is observable in a single image on a scan, "and the resident doesn't see it." Identifying this lymph node could be clinically meaningful in assessing whether a cancer of the pancreas had spread beyond the organ, a critical piece of information in its treatment and prognosis. "So I will go through the case and force him to go back and show him how he missed it."
"In some ways, we are victims of our own success," Dennis Orwig said. "We have so many excellent imaging techniques. Some doctors hardly examine patients or take histories anymore. They just order scans and say to the radiologist, 'Give me the diagnosis.'" In fact, the week we spoke, there was an article in the
New England Journal of Medicine
about whether the stethoscope had become a vestige of a bygone era, since cardiac imaging techniques were so advanced that the findings cardiologists traditionally make by listening had been rendered moot. "And when clinicians order sophisticated scans," Orwig continued, "they expect a definitive answer back. They don't want to hear a radiologist's description of a constellation of observations—they want one diagnosis. There is tremendous pressure on us to come to a conclusion," Orwig said, "and we have to resist that, because sometimes you can't make an exact diagnosis. The best you can do is to describe what you see."
A seasoned radiologist also learns not to give in when a clinician demands a discrete diagnosis. "Sometimes," Orwig allowed, "you can say, 'This is diverticulitis.' It's ninety-nine percent certain, and the clinician then can feel comfortable and go ahead and treat the patient with antibiotics. And that one percent may be a perforated colon cancer. But then there are patients who undergo CT scans and the best you can say is that there is 'a complex inflammatory process in the pelvis of this gentleman.' Many clinicians don't like to hear that. They think that the radiologist is waffling. But what the radiologist is doing is showing the doctor his thinking, sharing with him the most that can be said based on his expert observations."
Just as a clinician needs to choose his words carefully in communicating with patients, he must tailor the language of his request to a radiologist. "There is this notion that the clinician wants to keep the radiologist honest, so he doesn't tell him anything specific," Kressel noted. "In my field, you will get a referral that says 'patient with abdominal pain.' But without more specific history, you reduce the kind of clinical cues that are so important and actually make it much more difficult to assess the images.
"I always thought this was maximally stupid," Kressel said. "Why would you want to tie someone's arm behind his back?" Not only can this hamper perception and cognition, as Potchen's studies show, but, as Kressel elaborated, it can also affect technique and lead to errors. With the new multidetector CT scans, very large volumes of tissue are scanned in a short time, so for the best results the settings on the machine have to be adjusted to take into account the patient's clinical history. "If the clinician doesn't give us a full history, just the one question in his mind, then we will technically tailor the exam to that one question— like, Is there pulmonary embolus?—and risk missing something else that is important."
It is not only the clinician's language that can be misleading. Different radiologists use different terms to describe what they see. "People usually don't think about radiologists in terms of nuances of communication," Kressel said. "When you think of a doctor speaking, you imagine a clinician at the bedside explaining something to a patient. But radiologists become very impassioned about the words they use in their reports. And, of course, the language used reflects your type of thinking. Moreover, there is no agreement about different terms, no uniform structured approach to communicating findings." This is especially true with the more advanced CT and MRI scanners. "Even if different radiologists see the same thing on an image, just from the way they describe it, there are nuances and ambiguities communicated by the terms." Kressel referred again to the woman who was evaluated for a possible pulmonary embolus. "The radiologist reported that the aorta was 'not enlarged.' That term 'not enlarged' can be taken in many different ways. First, it doesn't mean that the radiologist saw the interior of the vessel. He is just making a descriptive statement. And 'not enlarged' is not the same as saying that the vessel is normal, although many clinicians would take that as the meaning of the radiologist's phrase. Because different terms mean different things to different doctors, a single term can guide thinking in different directions."