Finding breast cancer as early as possible seems to be a great idea, like trying to diagnose high blood pressure before it damages the heart or the kidneys. And mammograms can occasionally detect tumors as small as an eighth of an inch across, whereas the lower limit for tumors diagnosable by palpation (examining the breast manually) is about half an inch across. Yet one would like to be sure that this difference actually translates into a higher likelihood that treatment will be successful. An official nationwide mammography program would be a huge commitment: 51.5 million American women are aged forty or above. And one must bear in mind the cost of needless medical procedures generated by the huge number of false-positive mammograms—two to four false positives for every true positive, according to some measures. (A false positive shows a mass or lump that proves after further testing not to be cancerous.)We continue to consider creating a national screening program, but I believe it has never been proved that such a program would, on balance, be beneficial—even if it served the secondary purpose of bringing into the health-care system women who otherwise could not afford it or would not see a doctor frequently.
To prove the value of mammography scientifically is more difficult than it might seem. In some studies investigators ask women to volunteer for screening, and then report the number of breast-cancer cases and the percentage of women who survive five years after diagnosis. This figure is compared with the percentage in the population at large. In these studies researchers often go to considerable trouble to eliminate potential sources of confusion. For example, they may try to match by age the women undergoing regular mammography with other women. Or they may match by race or socioeconomic class. No matter how hard researchers try, though, such studies remain susceptible to three of the most common sources of bias in medical research.
Mammography may find a tumor as early as two years before it could have been detected by palpation. Let us, however, consider a hypothetical case in which the cancer has already spread to other parts of the body by the time it is discovered, and the woman goes to her grave on exactly the same day she would have if the tumor had been discovered later. In that case the sole effect of early detection has been to stretch out the time in which the woman bears the knowledge of her condition. But that is not how the woman would appear statistically if she had happened to become part of a research study. Pushing back the date of first diagnosis would increase the interval between diagnosis and death, apparently lengthening her survival. Statisticians call this effect "lead-time bias." Although nothing has actually changed, a woman who would have died, say, three years after treatment now dies five to six years after treatment—manufacturing an apparent victory for medicine.
A second problem with measuring the benefits of mammography is known as "length bias." Women typically have mammograms every year or every other year. Any cancers that are found between mammograms will be detected by palpation—very possibly by the woman herself. Such tumors are likely to be fast-growing; indeed, their rapid onset often explains why they were not picked up by the previous mammogram. The more aggressive tumors tend not to be diagnosed mammographically, and thus the tumors that are discovered by mammograms are often less dangerous. Mammography will be made to look good in a study comparing the survival rate of women whose tumors were diagnosed by mammography with that of women whose tumors were diagnosed by palpation, because the tumors discovered by mammography tend to be those that grow relatively slowly and thus take longer to kill patients.
The third and most important problem is "selection bias." This occurs when researchers measure the effect of a treatment on a group of people without realizing that those people are different from the general population. The risks that selection bias will occur are high, because women who participate in medical experiments are often not like the general population. Researchers typically work in teaching hospitals and thus draw their subjects from the patients who frequent them. These people may be more affluent than most Americans, and thus more prone to diseases of affluence. Or they may be more worried about their health, and thus more likely to seek expert medical care. In either case, the results of a test on such a select group can be misleading.
Many researchers agree that lead-time, length, and selection biases may flaw the optimistic accounts of the efficacy of mammography that have appeared in the scientific press and the popular media. Nonetheless, they support the idea of routinely screening women. The principal reason is that benefits from mammography have been observed in a special kind of study known as a prospective randomized clinical trial. In such a trial researchers randomly divide large numbers of volunteers into two groups at the outset (prospectively): a control group, which receives ordinary medical care; and a test group, which receives the medication or procedure under scrutiny. After a given period of time the two groups are compared. Properly conducted, such trials avoid all three kinds of bias. Even prospective randomized clinical trials have their pitfalls, though, because doctors can't control the actions of their patients. Members of the test group may fail to take their medicine or to show up for their medical procedures, and members of the control group may seek out a drug or procedure they are not supposed to have. As hard as researchers may try to ascertain levels of compliance, misclassification of a certain number of participants is inevitable. In addition, the medical care provided at research centers, which often conduct clinical trials, may not be representative of the care received by most people.
Nonetheless, by ensuring from the beginning that the test group and the control group are statistically similar and by tracking everyone in both groups, these trials can produce data that are as solid as medical research gets. And several big prospective randomized clinical trials have reported that women who regularly undergo mammography have, roughly speaking, 25 to 30 percent less chance of dying from breast cancer in the decade after initial screening than women who are not screened. Most breast-cancer specialists thus endorse mammography.
In fact the evidence from these trials is weaker than it sounds. In April of last year an article in Cancer summarized all eight of the major mammography trials that have been conducted to date. Six of the trials saw no significant decreases in breast-cancer mortality as a result of mammography. "Significant," an important term, means that statistical tests indicate that the effect is probably not due to chance. Also, the two significant clinical trials were the first ones completed. "Should the early trials be accepted as the gold standard and the later ones dismissed as somehow incompetent?" Charles J. Wright, of the University of British Columbia, and C. Barber Mueller, of McMaster University, asked in The Lancet last July. "Surely not, in view of the increasing rigour of trial design over the past 30 years and the vast improvement in quality of mammography." Indeed, the two earliest trials have serious potential shortcomings. Explaining these shortcomings involves delving still further into technical details; some readers may wish to skip the next section of this article entirely. The gist of my argument is that the benefits of frequent mammography as opposed to palpation performed during regular checkups and also by a woman herself are not well established; if they do exist, they are not as great as many women hope.