|Home | About | Journals | Submit | Contact Us | Français|
In 2005, mammography rates in the U.S. dropped nationally for the first time among age-eligible women. Increased risk of breast cancer related to hormone therapy (HT) use reported in 2002 led to a dramatic drop in its use by 2005. Because current users of HT also tend to have higher mammography rates, we examined whether concurrent drops in HT and mammography use were associated.
Using multivariate logistic regression we tested for an interaction between HT use and survey year, controlling for a range of measurable factors in data from the 2000 and 2005 National Health Interview Surveys (NHIS).
Women ages 50–64 were more likely to report a recent mammogram if they also reported more education, a usual source of care, private health insurance, any race except non-Hispanic Asian, talking with an obstetrician/gynecologist or other physician in the past 12 months, or were currently taking HT. Women 65 and older were more likely to report a recent mammogram if they also reported younger age (65–74 years), more education, a usual source of care, having Medicare Part B or other supplemental Medicare insurance, excellent health, any race except non-Hispanic Asian, talking with an obstetrician/gynecologist or other physician in the past 12 months, or were currently taking HT.
We found the change in HT use was associated with the drop in mammography use for women age 50–64, but not for women age 65 and older. NHIS data explained 70–80% of the change in mammography use.
In 2005, U.S. data showed the first-ever drop in mammography rates.1,2 This was surprising because mammography rates had consistently risen since first monitored in 1987. In 2002, JAMA published a report from the Women's Health Initiative that HT use was associated with increased risk of breast cancer.3 This widely-publicized finding alerted physicians and women to potential problems with HT, and led to a dramatic decline in the use of HT between 2000 and 2005.4,5 Because current users of HT also tend to have higher mammography rates,6 we speculated that women who stopped taking HT also might have stopped screening with mammography. Our reasoning was that if women typically need to see a doctor to renew their HT prescription, and physicians typically take that opportunity to discuss mammography and order a mammogram, then stopping the HT prescription visits would result in a lost opportunity for doctors to remind women about mammograms.
In this article we test whether HT is associated with mammography use among the population of US women 50 and older. Because the average age at menopause is slightly older than 50, and HT use is most common among post-menopausal women, we examined women 50 and older. We used a blended model (Figure 1). The model presents a patient-centered perspective and locates mammography appraisal within a multi-level population-based public health approach.
The literature on mammography use suggests that the most important predictor of mammography use is the medical encounter.7,8 Primary care constitutes the entry point to health care markets in the US,9 and provides the gateway to mammography. Although most Americans report having a usual source of health care, they need a specific reason to access the system. For example, if women want to obtain a mammogram or renew their HT prescription, they need to contact a physician. In 2000 and 2003, approximately 70% of women age 40 and older reported a recent mammogram; 30% did not. Nearly 50% of women 40–64 and 75% of women 65 and older who did not report a recent mammogram stated that they did not recall receiving a physician recommendation for one in the previous year.10 Different rates of cancer screening utilization are associated with different practice settings,11 and greater HMO market share is associated with greater use of mammography.12,13 Since use of HT is directly correlated with use of mammography,6,14 we examine whether women who used HT would continue to screen at the same higher levels when they stop taking HT.
We tested whether the drop in HT was directly correlated with the drop in mammography using data from 2000 and 2005 National Health Interview Surveys (NHIS). Questions on HT and mammography asked in 2000 and 2005 allowed us to test whether the drop in HT use was associated with the drop in mammography use. Our study examined women 50 and older interviewed in 2000 (N=7,125) and 2005 (N=7,387).
The NHIS is the principal source of health information on the civilian, non-institutionalized, household population of the United States and the largest population-based national sample on mammography use. The NHIS provides self- reported information. Questions on mammography use have been fielded periodically since 1987. Questions about mammography and HT were administered in 2000 and 2005. Periodic cancer control supplements are administered to a randomly selected sample adult in each household surveyed using computer-assisted personal interviewing (CAPI). African-American and Hispanic populations were oversampled in 2000 and 2005 to allow for more precise estimation of these groups. Data are collected using a complex sample design involving stratification, clustering, and multistage sampling. NHIS public use microdata files are released on an annual basis and survey descriptions, including response rates are available for each release at the NHIS website. Final response rates for the adult sample were 72% in 2000 and 69% in 2005.15,16
Our dependent variable is receipt of recent mammography defined as a mammogram reported within 2 years of the interview.
Our covariates are described below. Categories are mutually exclusive.
Educational attainment was categorized into less than high school vs. high school graduate only vs. some college or AA degree vs. college graduate (BA/BS) or higher.
Race-ethnicity was categorized into five groups, using federal government definitions for Hispanic, Non-Hispanic white, Non-Hispanic black, Non-Hispanic American Indian/Alaska Native, Non-Hispanic Asian.17
Immigrant status categorized variables on place of birth and time in the US according to: in the US less than 10 years vs. in the US 10 years or more vs. born in the US.
Income was not included in the model because most mammograms are paid for by insurance.18
Medical or health insurance was categorized slightly differently for women 50–64 and for women 65 and older, due to near-universal coverage by Medicare for women 65 and older. For women 50–64, insurance was grouped according to: private HMO vs. private non-HMO vs. public only vs. uninsured. For women 65 and older, insurance was grouped according to: Medicare HMO vs. Fee for Service (Private, Medicaid/Military/other government) vs. Medicare Fee for Service vs. Uninsured or Medicare Part A (hospital coverage) only.
Usual source of medical care was coded as yes vs. no or hospital emergency room (ER).
Saw or talked to a general doctor in the past 12 months was coded as yes vs. no.
Saw or talked to an ob-gyn in the past 12 months was coded as yes vs. no. Whether a woman was currently taking hormone therapy (HT) is the key covariate of interest and response items were coded yes vs. no. HT requires a prescription which requires contact with a physician.
Self-reported health status was grouped as: excellent/very good vs. good/fair/poor.
We applied the same computational methods used in 2000 to both years of data.19 We used statistical techniques to evaluate change over time in the age-eligible population.
First, to examine the relationship between population characteristics measured in the NHIS and the probability of receiving a recent mammogram, we fit a multivariate logistic regression model to pooled 2000 and 2005 data. The logistic model was stratified into two age groups because lack of insurance and unmet health service needs are more likely for women 50–6420 than for women 65 and older. Survey year was included as a covariate in the logistic model to represent change in mammography not captured in the variables directly measured in the NHIS surveys. We then calculated two types of change in the population using the logistic regression model: i) change in mammography rates associated with changes in population characteristics measured in the NHIS that have been shown in other studies to be associated with mammography, including HT use;21 and ii) change in screening rates attributable to factors not captured in the items we selected from the NHIS survey. We start with the observed difference in the rate of recent mammography between 2000 and 2005 for each age group. Next we calculate predicted marginals.22 The difference between predicted marginals in 2000 and in 2005 represents change in recent mammography rates attributable to factors not captured in the items we selected from the NHIS survey. The remainder of the observed difference is associated with the variables included in the logistic regression. Partitioning the overall change in screening into these two separate categories provides insight into the portion of the change in mammography use associated with factors measured by the NHIS and the portion related to factors that are not directly measured by the NHIS.
Second, we tested for an association between changes in HT use and changes in mammography use in the population of women 50 and older. Our hypothesis has two explicit scenarios illustrated in Figure 2. Because current users of HT also tend to have higher rates of current screening than do women who do not use HT,6,14 a question that follows is whether women who stop taking HT would continue to screen at higher levels. Our objective was to test whether women who stopped using HT between 2000 and 2005 also reduced their mammography use. Because NHIS is a cross-sectional survey, it does not capture longitudinal data on HT use. Therefore we tested for an interaction between HT and survey year (2000 and 2005) as a predictor of mammography use in the population of age-eligible women. With this interaction test, we can evaluate whether the influence of HT on mammography is different in 2005 than in 2000. It is unclear whether HT is an independent predictor of mammography screening, or if mammography use and HT are jointly correlated with other factors such as education, insurance, income, use of preventive services, and physicians visits. We therefore considered two extreme possibilities to illustrate how changes in current HT use would affect mammography rates. The numbers presented in Figure 2 were chosen for ease of computation.
If HT and mammography use are related only through factors other than HT, women who stop taking HT would maintain the same higher screening level as when they were current users of HT and mammography use would remain the same for the overall population and for the women who are current users of HT. However, for the group of women who are not current users of HT, screening rates would increase because this group would now include the women who stopped HT with their higher level of mammography use. This scenario would be represented in the logistic model by a significant interaction term between HT use and survey year because it would change the relationship between HT and mammography between 2000 and 2005.
If HT is an independent predictor of mammography use, women who stop taking HT would change their mammography use to resemble levels seen in non-users of HT. Thus, when a portion of the population stops using HT, the rates of mammography use for the overall population would decline because more of the population is now in the HT non-user group. This scenario would be represented in the logistic model by no significant interaction between HT use and survey year because mammography rates in the current HT user groups remain the same and mammography rates among non users of HT remain the same in 2000 and in 2005.
To see whether women who stopped HT use changed their mammography use, we tested an interaction term for current HT use and survey year in pooled logistic regression for each of the two age groups. The logistic regression included all covariates described in the methods section. All analyses were weighted by the NHIS sample weights to account for the unequal probabilities of selection and the complex survey design. SUDAAN release 9 was used to calculate frequencies and to compute logistic regression results. The sample weights for analyses using pooled data were modified by adding the NHIS sample weights for each survey and dividing them by two.
Table 1 shows the characteristics of the population in 2000 and 2005 for women age 50–64 years and age 65 and older. Population distributions for each group changed significantly between 2000 and 2005. It is difficult to predict the net effect on screening rates because some characteristics are positively associated with screening (e.g., education, recent doctor visit) while others are negatively associated (e.g., drop in HMO coverage and current HT use).
To evaluate the net effect, we examined the data using multivariate logistic regression. Table 2 shows adjusted results for each age group. Women 50–64 were more likely to report a recent mammogram if they also reported more education, a usual source of care, private health insurance, any race except non-Hispanic Asian, talking with an obstetrician/gynecologist or other physician in the past 12 months, or were currently taking HT. Women 65 and older were more likely to report a recent mammogram if they also reported younger age (65–74 years), more education, a usual source of care, having Medicare Part B or other supplemental Medicare insurance, excellent health, any race except non-Hispanic Asian, talking with an obstetrician/gynecologist or other physician in the past 12 months, or were currently taking HT. Findings for interactions for each age group will be discussed in the next section.
Figure 3 compares the change in recent mammography between 2000 and 2005 for the two age groups associated with changes in the population characteristics modeled in Table 2. These associations predict most of the observed change in mammography usage between 2000 and 2005 in both age groups. Recent screening declined 4.7 percentage points for women age 50–64 and 2.5 percentage points for women age 65 and older. Using the predicted marginal values for survey year, we found that 0.8 percentage points (standard error of 1.1 percentage points) of the estimated decline for women age 50–64 and 0.8 percentage points (standard error of 1.3 percentage points) of the estimated decline for women age 65 and over was not associated with the variables included in the regression analysis. These estimates suggest that 3.9 of the observed 4.7 percentage point decline in recent screening (80% of the total decline) for ages 50–64 and 1.7 of the observed 2.5 percentage point decline (70% of the total decline) for age 65 and older are associated with factors included in the modeling. Although we are able to estimate the portion of the decline associated with factors in the model, the large standard errors demonstrate the difficulty in obtaining precise estimates for these small effects. Nevertheless, we think they are worth reporting because this is our best estimate.
We also tested for an association between changes in HT use and changes in mammography use in the population of women 50 and older. Table 2 shows the results of the multivariate logistic regression for women 50–64 and 65 and older separately. The interaction between current HT use and survey year was significant only for women 65 and older, so results for women ages 50–64 are shown without this interaction.
For women age 50–64 we found no significant interaction between current use of HT and survey year. This scenario, shown in Figure 2, Hypothetical 2, suggests that HT and mammography use are related. Women who stopped HT had lower screening rates after they stopped than when they were taking HT, thus reducing their mammography rates to the level previously seen in non-users of HT.
For women 65 and older, a significant interaction was found. This scenario, shown in Figure 2, Hypothetical 1, suggests that HT and mammography use are not related. Women who reported being current users of HT had lower rates of mammography in 2005 than in 2000, while women who reported non-use of HT had similar rates of mammography use in 2005 and 2000. Our model suggests that the change in HT use had less effect on mammography use in older women compared to younger women.
Our study is the first to show that a drop in HT use was associated with a drop in mammography use in the general population. Variables from the NHIS were associated with 70% to 80% of the change in mammography for the women we studied. In addition to HT, age was strongly associated with mammography use. Our study is the first to show a difference by age group: change in HT use was associated with the drop in mammography use for women age 50–64, but not for women 65 and older.
Population changes in breast cancer incidence related to HT use may reflect a combination of a reduction in breast cancer risk and slightly lower screening rates23. We found an association between HT and mammography in a representative sample for women 50 and older. In an earlier article,1 we indicated that the extent to which the decline in incidence rates for breast cancer is associated with the drop in mammography needs to be investigated; however, because there is no measure of breast cancer incidence in NHIS, our study couldn't examine associations between mammography, HT, and incidence. Caan et al used HMO data to examine whether women who were regular users of HT had different patterns of mammography compared with women who continued using or never used hormones and found that, among women who stopped using HT after publication of the WHI study, both mammography use and breast cancer incidence dropped.24 Robbins et al compared secular trends in HT use and breast cancer incidence in California and found that each 1% decrease in the prevalence of HT use was associated with a decrease of 3.1 cases per 100,000 women in breast cancer incidence.25 More recently, a study of trends in invasive breast cancer incidence in women undergoing regular screening mammography showed a significant decrease in invasive breast cancer incidence between 2002 to 2006 among women age 50–69 and 70–79 years.26
To understand the potential impact of the drop in mammography use compared to the drop in HT use on breast cancer incidence, it is important to note that HT use dropped from 41% to 16% among women 50–64 and from 19% to 10% among women 65 and older; and biennial mammography use dropped from 78% to 73% among women ages 50–64 and from 67% to 65% among women 65 and older. The drop in HT use represented about 6.4 million women ages 50–64 and about 2 million aged 65 and older while mammography represented far less: about 1.2 million women ages 50–64 and about .5 million women aged 65 and older. While we conclude that the reduction of HT use is associated with reduced mammography, our analysis cannot assess the separate contributions of each to the reduction in incidence. Our results are consistent with the preponderance of evidence from other studies reporting the falling incidence is largely accounted for by a reduction in risk due to HT cessation and continues to raise the possibility that reduced mammography utilization may also have played a role in the relationship between HT and incidence in the population.
Our findings are significant and important in understanding the relationship between HT and mammography use, and mammography's impact on mortality. Observed disparities in cancer mortality, survival, and incidence have motivated the study of societal-level influences on the etiology of cancer.27 The literature shows that the practice setting is key to whether a woman gets a mammogram, and the NHIS captures this aspect of screening for individual women with a national sample. A limitation is that self-report may overestimate use.28–30 While ideally this analysis would be done with longitudinal data in order to directly examine the screening rates in women who stopped using HT,24 our data would not allow that. Other available sources of data were also problematic: HMO data would not be representative of the general population and Medicare data is limited to women 65 and older when we expect the biggest impact to be in women ages 50–70.
Finally, it is worth noting the considerable controversy over details of the mortality benefit of mammography in the scientific literature during the period that mammography dropped.31–35 An opportunistic system of screening leaves it to individual women to assess information in making a decision about the importance of having a mammogram relative to her resources. Therefore, different communication channels, because they reach women with different educational levels, may lead to different behaviors. In an earlier analysis we found that women with higher education and more resources were significantly less likely to screen in 2005 than in 2000.1 An important exception was women enrolled in organized systems. Organized systems develop and deploy protocols based on scientific evidence, and women enrolled in organized systems tended to have high rates of mammography use regardless of personal characteristics.36 Organized systems often add to their advantage by using electronic health records to maintain patient records, remind patients and doctors when tests are due and provide feedback to clinicians to help them improve their practice patterns.37,38
We wish to acknowledge Penny Randall-Levy for expert help in reference management and manuscript formatting.
The findings and conclusions in this report are those of the author(s) and do not necessarily represent the official position of the Centers for Disease Control and Prevention.
None of the authors have financial disclosures or conflicts of interest.