An expert panel of 10 international cancer researchers and practicing oncologists met in Boston to discuss the past, present, and future uses of antiestrogens in the treatment of breast cancer.
BOSTON--When epidemiologic tools are used to dissect clinical trial results, potential biases abound, said panel member Robert W. Morgan, MD, an epidemiologist and president of Environmental Health Strategies, Menlo Park, Calif.
"Epidemiology is a blunt instrument and delivers rough justice," he said, so that small increases in risk found by epidemiologic methods may not be meaningful, as appears to be the case in the analyses of endometrial cancer risk in patients enrolled in trials of tamoxifen(Drug information on tamoxifen) (Nolvadex).
Dr. Morgan described six potential areas of bias in cohort studies that, he said, severely limit our ability to gather conclusions from these studies about the risk of second cancers in women taking tamoxifen therapy. The same problems can apply to case-control studies, he added. The possible biases are:
- Post hoc design. The original tamoxifen studies were designed to look at efficacy, not at endometrial cancer risk, so that all the necessary data to analyze a possible relationship were not collected. The studies looking at a possible relationship of tamoxifen and endometrial cancer were designed after the fact.
- Surveillance bias. "Any time you have a group of people under study, they will report an increased rate of almost any disease imaginable," Dr. Morgan said.
- Ascertainment bias. This is the likelihood of diagnosing a preexisting disease through increased medical surveillance. At least two of the cohort studies have shown that women in the tamoxifen arm have triple the rate of gynecological exams as the control group. "If you're tripling the rate of examination, then you are more likely to find the underlying tumors if they are there," he said.
- Patient compliance problems. In many of these studies, it was found that not all patients in the study group actually received tamoxifen and that some control patients did manage to get the drug.
- Latency issues. Endometrial cancer can exist in a latent, subclinical phase for many years, Dr. Morgan said. When endometrial cancers are found just a few months after initiation of tamoxifen treatment, "it seems against common sense to attribute these to tamoxifen," he said.
- Lack of data concerning estrogen replacement therapy (ERT). Although ERT is not generally recommended in breast cancer patients, Dr. Morgan said that it is being prescribed on a fairly regular basis for women with breast cancer without regard as to whether they are enrolled in a trial, and "virtually none of these studies have good data on ERT, which, I believe, is a major confounder of the findings."
The effect of each of these biases, he said, is to increase the apparent relative odds and relative risk of endometrial cancer attributed to tamoxifen.
In his review of 18 studies of endometrial cancer risk in tamoxifen trials, Dr. Morgan found that only two showed any significant increased risk for tamoxifen users. "And in both these studies, the magnitude of the risk depends on how you look at it," he said. For example, in NSABP B-14, the relative risk ranged from as low as 2.1 to as high as infinity, depending on which set of controls were used as a comparison group.
The NSABP researchers used three comparison groups: the nontamoxifen control group within the study (which had no endometrial cancers, prompting the investigators to seek other comparison groups); a control group from another study; and a control group based on SEER data on cancer incidence.
The last group gave the lowest relative risk of 2.1. "And in epidemiology, relative risks of 2 are not very high," Dr. Morgan said.
He also pointed out that since breast cancer and endometrial cancer share a common epidemiology, women who have had breast cancer are already at increased risk of endometrial cancer whether or not they take tamoxifen.
Thus, much of the increased risk seen when the incidence of endometrial cancer in tamoxifen users was compared with that of the general population from SEER data could be explained by the common epidemiology.
The other study to show a significant increase in risk (Stockholm, 1993, reported by Fornander) gave a range of relative risk of 3 to 6, depending on which assumptions the investigators made, "whether you throw out the people who did not have true cancer, whether you include people who never received the drug and cross over the controls who did take it," he said.
Furthermore, in both of these studies, many of the endometrial cancer cases were reported within the first 2 years after tamoxifen therapy began. "This short latency period makes it implausible that tamoxifen is the primary cause, although the data are compatible with tamoxifen either promoting an existing tumor or producing changes such as bleeding that lead to diagnosis," he said.
Dr. Morgan ended with the plea that investigators plan clinical studies with epidemiologic concerns in mind from the beginning. This would include close monitoring of patient compliance, to make sure patients are not crossing over to the other side of the trial, and data on confounders such as use of ERT.
