With increasing availability of massive quantities of health care data and advances in technology for data analyses, novel investigations of health delivery and outcomes are proliferating. In this issue of JAMA Internal Medicine, Harding et al1 investigate the health effects of screening mammography through the merging of 2 large data sources using an ecological study design. The study correlates US county-level estimates of mammography, based on national surveys of women who recount prior mammography examinations, with breast cancer incidence and mortality rates from the Surveillance, Epidemiology, and End Results registry. Harding et al report increased rates of breast cancer diagnoses in areas where more women have undergone screening mammography examinations, but no apparent correlation between increased mammography and subsequent breast cancer mortality.
Ecological studies use an intuitively appealing research design that relates the frequency with which some exposure or intervention (eg, cancer screening) and some outcome of interest (eg, cancer diagnosis or mortality) occur in the same geographic area or care setting. If the intensity of an intervention varies across the areas and is effective, then one should observe a gradient in the outcome from areas with lower frequencies of the intervention to areas with higher frequencies. Through the years, ecological studies have been helpful in identifying associations between smoking rates and subsequent lung cancer and between screening with Papanicolaou tests and cervical cancers. Ecological studies have even been used previously to study mammography in relation to breast cancer mortality.2,3
However, much has also been written about the caution needed when interpreting ecological analyses. It is well known, for example, that ecological studies provide no information as to whether the people who were actually exposed to the intervention were the same people who developed the disease, whether the exposure or the onset of disease came first, or whether there are other explanations for the observed association. Ecological analyses also may not properly reflect group-level associations because of area-level variations in confounding factors or other practices affecting the outcome.
The analysis by Harding et al1 did not detect a significant gradient in breast cancer mortality across areas with higher vs lower mammography use. In their analysis, the mammography frequencies in the counties ranged from 39% to 78%; however, the range in the largest counties, which essentially determined the reported weighted correlations varied between about 60% and 75%. This narrow interval might not have been enough to induce a gradient in disease mortality given the many other factors, such as population behaviors and treatment patterns, that typically vary across areas and affect breast cancer mortality. Prior ecological studies of mammography conducted at the larger state level with a wider range of mammography frequencies showed a decline in breast cancer mortality associated with more screening.2,3
Regarding incidence, there is no question that we have seen a dramatic increase in the incidence of ductal carcinoma in situ (DCIS) since the introduction of screening mammography. In the study by Harding et al,1a higher incidence of DCIS and invasive breast cancer was observed in areas in which mammography was more common. Furthermore, the investigators found that increased incidence of smaller tumors (≤2 cm) in areas with more screening was not counterbalanced by a decline in the incidence of larger tumors. These findings, together with their nonsignificant mortality result, led the investigators to conclude that mammography screening is leading to overdiagnosis.
Overdiagnosis is the diagnosis of a tumor that would not have become clinically apparent in the absence of screening. Treatment of an overdiagnosed tumor cannot provide benefit, but it can lead to harm. Overdiagnosis and overtreatment are now widely acknowledged to be an important harm of medical practice, including cancer screening.
Whereas most previous studies have attempted to estimate breast cancer overdiagnosis from trends in disease incidence at a larger population level, this study uses smaller study units and examines patterns at the county level. As with the ecological analysis of mortality data, some caution is required when interpreting the county-level patterns of incidence. Increased screening generally leads to increased incidence, but determining how patterns of smaller and larger tumors might be expected to change under increased screening is quite complex. While observations by Harding et al could have several causes and cannot be unequivocally attributed to overdiagnosis, they add to a growing body of literature on the topic.
Most scientists now acknowledge that there is some level of overdiagnosis in breast cancer screening, but the frequency of overdiagnosis has not been conclusively established. Estimates in the literature cover a frustratingly broad range, from less than 10% to 50% or more of breast cancer diagnoses. Published studies of overdiagnosis vary widely not only in their results but also in their populations, methods, and measures.4In practice, the frequency of overdiagnosis is likely quite different for DCIS and invasive tumors.
Sadly, we are left in a conundrum. Women will increasingly approach their physicians with questions and concerns about overdiagnosis, and we have no clear answers to provide. We do not know the actual percentage of overdiagnosed cases among women screened, and we are not able to identify which women with newly diagnosed DCIS or invasive cancer are overdiagnosed. Many screening guidelines now mandate shared and informed decision making in the patient-physician relationship, but this is not an easy task.
As health care professionals, we want to do better at educating our patients about all of their health care options, including possible consequences and harms as well as benefits. While almost all women (96.3%) report that their physicians discussed the pros of screening mammography with them, women also report that only 1 of 5 physicians (19.5%) mentioned the potential harms, such as overdiagnosis.5 Qualitative studies have shown that women considering initiating screening mammography were unaware of overdiagnosis but could understand the idea and placed value on this information.6 A recent randomized clinical trial that evaluated the effects of a decision aid providing information about the concept and frequency of overdiagnosis reported that women support the inclusion of such information.7
We need clear communication and better tools to help women make informed decisions regarding breast cancer screening mammography. When talking with women about premalignant conditions (eg, DCIS), we encourage using care with word choices and restricting use of the word cancer to invasive cancers.8Perhaps most important, we need to learn how to communicate with our patients about uncertainty and the limits of our scientific knowledge. In the end, we all need to become comfortable with informing women that we do not know the actual magnitude of overdiagnosis with precision. Part of informed decision making is providing all the information, even our uncertainty.