By CARL BIALIK,
Critics of new recommendations to scale back breast-cancer screening complain that cold numbers trumped medical judgment.
But despite decades of extensive mammography study, those numbers are inexact and don’t point to a clear course of action.
The U.S. Preventive Services Task Force, a 16-doctor, federally funded panel, recommended against routine mammograms for women in their 40s, and to reduce the frequency from annual to biennial starting at age 50. That sparked a backlash among many groups, including the American Cancer Society, which continues to recommend annual screenings starting at age 40.
One set of numbers was seized upon by critics of the report: that 1,900 women in their 40s would need to be screened annually to save one life, compared with 1,300 women in their 50s. The task force recommended a more relaxed standard of individual discretion on screening for 40-somethings. That suggested to some breast-cancer survivors that the lives of individual women were being undervalued.
The 1-out-of-1,900 figure was derived from comparing breast-cancer death rates in women who were being screened with those who weren’t. For every 1,900 women in the screened group, one fewer died of breast cancer than did among 1,900 women who weren’t screened.
The task force amassed results from various trials, using a statistical technique for combining research studies called meta-analysis, to arrive at the 1-in-1,900 figure. But that figure comes with massive uncertainty. Because the women in these studies represented a sample of all women, the results have a margin of error, as with political polls.
Yet the uncertainty here wouldn’t pass muster at Gallup. Researchers said there was a 95% chance that to avert one cancer death, somewhere between 900 and 6,000 women in their 40s would have to be screened, though somewhere around 1,900 was their best estimate. That range is so broad that it is possible far more lives are saved with annual screening among 40-year-olds than the studies indicated.
“We don’t have a lot of precision here,” says Heidi Nelson, a research professor of medical informatics and clinical epidemiology and medicine at the Oregon Health & Science University who helped review the literature for the task force.
Even to get to that low level of precision required making some concessions about the quality and age of studies used. Of seven included studies, most involved screening in the 1970s or 1980s.
Only one of the studies was based in the U.S., and it began in 1963. Dr. Nelson says the task force could find no flawless study, rating none of the trials used as better than fair, but she notes that they all had fairly consistent findings on survival gains from mammograms.
Another potential problem with the new mammogram recommendations is that the task force excluded studies that might have cast screening of younger women in a more positive light. The panel didn’t factor in any so-called observational studies, where instead of randomly assigning women to receive or not receive screening, women were tracked after making their own decisions about mammograms. Such studies are considered inferior because, without randomization, other factors could affect improved survival for women who are screened, such as better overall health and health care.
Robert A. Smith, director of cancer screening for the American Cancer Society, says failing to include results from observational studies was a missed opportunity. He noted a 2003 study he co-wrote in the Lancet medical journal with a longer follow-up period that found that only 726 women in their 40s must be screened to save one life.
The 1,900 figure also has been portrayed misleadingly, in ways that can either understate or overstate mammography’s benefits. For one thing, one life was saved per 1,900 women who were invited to be screened; many didn’t comply, meaning that mammograms could save one life for fewer than 1,900 women. Dr. Nelson says about 70% to 80% of women complied in the studies, but everyone invited was included because of potential differences between those who did and didn’t choose to be screened.
Also, the 1,900 figure is based on a relatively brief follow-up period. Include more follow-up, and more lives could have been saved, because more women at the end of a follow-up period who hadn’t been screened may be dying of breast cancer, as Dr. Smith of the American Cancer Society notes. And saving women in their 40s saves more life-years than saving someone older.
On the other hand, since the conclusion is based on annual tests over the entire decade, the 1,900 figure represents one life saved for as many as 19,000 mammograms. And the benefits of the mammograms are being compared with an unrealistic opposite scenario, in which no women in their 40s are screened.
Diana Petitti, a professor in biomedical informatics at Arizona State University and vice chairwoman of the panel, said the task force looked at a range of evidence in making its recommendation. “This is purposely a qualitative assessment and not an assessment based on some magic number,” she said in an email.
There isn’t enough research to say exactly what effect screening once every two years would have, so the task force drew upon computer simulations conducted by researchers. They factored in treatment, the range of cancer life cycles, and data on breast-cancer screening accuracy and survival to estimate how many life-years are saved for each 1,000 women who are screened — and, on the downside, how many false positives and unnecessary biopsies accumulate.
The results from these models received far less attention than the 1,900 figure, but provide a key to understanding the task force’s thinking. Take, for instance, numbers produced by one model from a team at Stanford University that the task force highlighted. Screen women annually from ages 50 to 69, and 132 life-years will be saved per 1,000 women. A biennial screening regimen resulted in 33 more life-years lost per 1,000 women than annual screening, but it succeeded in avoiding 8,815 mammograms, 430 false-positive results and 30 unnecessary biopsies.
“Researchers aren’t going to calculate [their] way out of this problem,” says H. Gilbert Welch, professor of medicine at the Dartmouth Institute for Health Policy and Clinical Practice. “This is still a close call.”