Introduction

Tissue density, a reflection of the physical and biochemical composition of the tissue, can be estimated from the Hounsfield Unit of computed tomography images. Tissue density is particularly useful for capturing a variety of obesity-related health risks as the Hounsfield Unit provides an indication of adipose infiltration into tissue. For example, low liver density reflects fat accumulation, a risk factor for insulin resistance and type 2 diabetes [1,2]. Low skeletal muscle density is also associated with insulin resistance and type 2 diabetes [3,4] as well as increased mortality risk [5], fracture [6] and mobility limitation in old age [7]. Within adipose tissue, denser tissue (smaller adipocytes versus larger lipid filled adipocytes) is associated with increased mortality risk in older adults [8]. Given the general consistency of tissue characteristics in the body (ie. fatty liver is positively correlated with visceral adipose and skeletal muscle adipose [2]) it is likely that the characteristics of other tissues may also be associated with health risks.

Mammography for breast cancer screening is one of the most widely used forms of imaging but the prognostic value of mammographic images outside of breast cancer risk prediction is largely unknown. In mammography, three metrics which reflect the tissue composition of the breast have been commonly characterized in epidemiologic studies: 1) nondense area which represents breast fat tissue, 2) dense area which refers to fibroglandular tissue and stromal tissue, and 3) percent density, the amount of dense tissue relative to total breast area. Density measures can be estimated quantitatively including planimetry [9], computerized thresholding techniques [10] or qualitatively via visual assessment, for example the Breast Imaging and Data System (BI-RADS) breast composition categories which range from almost entirely fatty to extremely dense [11]. Within the breast cancer literature, numerous studies have demonstrated that greater dense area and percent density are strong risk factors for breast cancer [12-14]. Nondense area may also confer risks for breast cancer independent of breast density although the direction of the association is unclear [15,16]. Although the reasons underlying these associations are incompletely understood, mammographic density has emerged as an important factor for the prediction of breast cancer risk [17].

It is less clear if mammographic features relate to survival among women diagnosed with breast cancer. Several studies have reported no association between breast density and breast cancer mortality [18-21] or increased risk of breast cancer mortality [20] among breast cancer cases. A study of breast tissue density among breast cancer cases relative to cases and non-breast cancer controls combined, reported that high density breast tissue was associated with decreased mortality risk [22]. Even less is known regarding potential relationships of breast density with all-cause mortality in women without breast cancer. Improved understanding of potential relationships is important for helping to identify mechanisms which may be targeted for interventions. Hypothesized mechanisms linking breast density and breast cancer mortality such as paracrine growth factors and genetic damage [23] may be pertinent to the development of disease beyond breast cancer such as cardiovascular disease [24] and arthritis [25]. Therefore, the aim of this study is to prospectively investigate relationships between nondense and dense mammographic area, breast area and percent density with risk of all-cause mortality in women without breast cancer.

Methods

Results

Table 1 presents transformed mammographic measures and additional characteristics of the analytical sample. The median (IQR) age of the analytic sample was 50 years (44-57 years). More than 94% of the analytic sample is white, and 45% have postsecondary education. Generally, participants with higher nondense area are older, less educated, have heavier BMI, and are more likely to have diabetes and heart disease (P<0.05 for all).

Table 2 depicts correlations between age, mammographic features and BMI. Age was positively correlated with nondense area (r=0.29, P<0.001), breast area (0.19, P<0.001) and BMI (r=0.12, P<0.001), and inversely correlated with dense area (r=-0.18, P<0.001) and percent density (r=-0.28, P<0.001). Nondense area was positively correlated with BMI (r=0.68, P<0.001), and breast area (r=0.83, P<0.001) and inversely correlated with dense area and percent density (r=-0.42, P<0.001 and -0.80, P<0.001, respectively). Dense area was weakly but significantly correlated with breast area (r=0.07, P<0.001) and inversely correlated with BMI (r=-0.17, P<0.001). Breast area was negatively correlated with percent density (r=-0.37, P<0.001) and positively correlated with BMI (r=0.64, P<0.001). As expected due to its derivation, percent density was positively correlated with dense area (r=0.85, P<0.001). In summary, older women had higher BMI and larger breasts characterized by greater nondense area and lower dense area.

The mean (SD) follow-up period was 28.2 (4.75) years during which time 1361 women died (mortality rate 139/1000 person years). 45 women who were not known to have breast cancer during the BCDDP follow-up study died due to breast cancer. Associations between continuous mammographic features and risk of mortality per SD difference are shown in Table 3. In unadjusted models (Model 1), greater nondense area and breast area were associated with increased risk of mortality. Conversely, greater dense area and percent density were associated with lower risk of mortality. All associations were reduced but remained significant after adjustment for age (Model 2) and did not change appreciably after adjustment for additional covariates (Model 3). With additional adjustment for BMI (Model 4), associations were diminished for all features but remained significant for dense area (HR 0.94, 95% CI 0.89-0.99, P=0.02) and percent density (HR 0.93, 95% CI 0.87-0.98, P=0.01). Within models older age and greater BMI were consistently associated with increased mortality risk (P<0.001).

Associations between percentiles of mammographic features and risk of death yielded similar results (Table 4). Compared with women in the 15^th percentile, the upper three groups of nondense area and breast area had increased risk of mortality (Model 1). Associations were reduced with adjustment for age (Model 2) and but remained significant for the upper 2 groups of nondense area and the upper 3 groups of breast area (Model 3). Inverse relationships were observed for dense area and percent density; compared with the 15th percentile, women in the upper three groups had a lower risk of mortality (Model 1) that persisted after adjustment for age and additional covariates (Models 2 and 3). Adjustment for BMI (Model 4) attenuated associations except for marginal associations with the upper group of dense area (HR=0.83, 95% CI 0.68-1.00), and the upper groups of percent density: 40^th percentile (HR=0.86, 95% CI 0.74-0.99), 65^th percentile (HR=0.84, 95% CI 0.71-0.996) and 85^th percentile (HR=0.80, 95% CI 0.64-1.00). Results were similar when models of percentiles and continuous mammographic features were adjusted for menopause status and age at first birth.

Discussion

This study contributes to our understanding of the prognostic value of mammographic features by examining associations with mortality risk in healthy women, a seldom studied area. Greater nondense area and total breast area were associated with increased mortality risk but associations were not independent of BMI. In contrast, greater dense area and percent density were both associated with lower mortality risk independent of risk factors and BMI. For every SD increase of dense area, there was a 6% reduction in mortality risk and for every SD increase of percent density, there was a 7% reduction in mortality risk. These results suggest that mammographic images may capture features that are independently related to survival. Thus, when evaluating mammograms in relation to breast cancer risk, the characteristics of breast tissue and specifically, dense tissue and percent density may provide novel prognostic information for healthy women.

In this population the association between mammographic features and risk of death appears to largely reflect risk carried by overweight or obesity. All mammographic features were correlated with BMI (nondense area and breast area positively, dense area and percent density inversely). Our risk models also seem to reflect the risk for mortality attributable to excess body weight [29-31]. The upper percentile groups of nondense area and total breast area that carried the greatest mortality risk in unadjusted analyses predominately consisted of women who were overweight or obese. Moreover, adjustment for BMI reduced associations and significantly attenuated associations with nondense area and breast area. However, inverse associations remained statistically significant for continuous measures of dense area and percent density after adjustment for BMI. This suggests that associations between mammographic features and mortality risk may extend beyond risk attributable to BMI alone.

The basis for the observed associations between mammographic density characteristics and mortality risk are unclear and likely complex. Relationships between adipose tissue, disease and disease-related mortality are possibly mediated through carbohydrate metabolizing characteristics of adipose tissue, secretion of cytokines [32] and/or adipokines [33]. Thus the metabolic characteristics of adipose tissue may contribute to our finding of greater mortality risk with greater nondense fatty tissue and help explain why associations were attenuated with adjustment for overall adiposity (BMI). Proposed mechanisms linking mammographic density and breast cancer risk have included steroid sex hormones and growth factors since they are generally positively associated with breast density and are involved in the development of breast cancer [23]. These stimuli also contribute to the pathogenesis of multiple diseases that impact overall life expectancy [24,25,34]. However, our data raises questions regarding this hypothesis as we observed an inverse association between dense area, mammographic density and risk of death. Alternatively, genetic factors may play a role in risk relationships. The variation in breast density attributed to lifestyle factors including hormone levels, age, age at parity, and BMI has been estimated to be only 20 to 30% [35] with the remainder possibly attributable to genetic factors [36]. As genes involved in breast density are identified, our understanding of associations between breast density and mortality risk may be expanded.

Direct comparison of our results is limited by the lack of studies on mammographic features and health outcomes in healthy women but parallels can be drawn to studies of breast cancer mortality. The association between greater dense area and lower mortality we observed is consistent with a study of breast cancer cases from a Danish screening study that reported lower risk of all-cause mortality in women with “mixed/dense” breasts [22]. The positive association we found between nondense fat tissue and mortality risk (although not independent of BMI) is consistent with a study of breast cancer cases in which Gierach et al. [18] reported increased risk of breast cancer mortality in obese women with BI-RADS 1 density (almost entirely fat). In contrast, Chiu et al. [37] reported significant increased mortality from breast cancer among women with greater dense tissue, however, breast tissue was broadly categorized as “dense” or “nondense”. Our results vary from studies of breast cancer cases that found null associations between breast density and mortality, however these studies incompletely adjusted for confounders of prognosis such as comorbidities, treatment, tumor size or stage of cancer [19,20,38].

It is likely that the associations we observed with mammographic features and mortality risk are not specific to a body compartment or tissue, but rather represent a systemic effect. For example, obesity manifests as fatty infiltration of the liver [2] and skeletal muscle [39], larger adipocytes in adipose tissue [8] and as in this study, greater breast fat tissue. To that end, our results are consistent with studies of computed tomography density that show non-breast tissue density provides important prognostic information related to health outcomes [2,6,40]. Specifically, lower density skeletal muscle based on the Hounsfield Unit, has been associated with cancer-related mortality [41] and more dense adipose tissue is a risk factor for all-cause mortality in healthy older adults [8]. Although the measurement of tissue “density” differs across studies, together these findings suggest consistent associations between radiographic characteristics of tissue and mortality. Studies with whole body imaging would improve our understanding of the underlying biological characteristics of tissue density throughout the body.

Strengths of this study include a long follow-up period of more than 30 years and the quantitative continuous measures of breast nondense and dense areas. Continuous measures overcome the limitations of BI-RADS classification which can have substantial interobserver variability [42], potentially resulting in misclassification. A limitation of this analysis concerns the design of the BCDDP cohort which was originally conceived as a general population study to demonstrate the feasibility of large-scale breast cancer screening. As a result, data on smoking, heart disease and diabetes were not collected at the time of the screening mammogram. Status was based upon recalled history in follow-up questionnaires possibly leading to misreporting. There has also been a shift towards heavier body weights since this study was initiated [43] which may limit the generalization of our results to contemporary populations. Studies in breast cancer suggest that associations between mammographic measures and mortality risk are greatest among women with BMI >30kg/m² [44] and it is possible that we did not fully capture the risk relationships. The distribution of body weight in the population also limited our ability to stratify risk analyses by BMI. Given the strong attenuating effect of BMI on associations between mammographic features and mortality, there is the potential for residual confounding.

It is also important to acknowledge potential bias within our participant population when considering the generalizability of our results. Our sample of healthy women included women with benign breast disease and women recommended for biopsy at baseline screening, who may have had a higher risk of breast cancer than the general population. However, we excluded women with incident breast cancer during follow-up screening and only a small number of women in our population (N=45) died from breast cancer. It is likely that our analytical sample is healthier than the general population. The prevalence of smoking is low, and the prevalence of diabetes is lower than women in the Second National Health and Nutrition Examination Survey which was conducted during a similar time period as the BCDDP [45]. Additional analyses in large populations to see whether our results replicate would be valuable.

Conclusions

Mammographic images are often not exploited beyond assessment of breast abnormalities and breast cancer risk but as we demonstrate here tissue characteristics captured with mammography may carry broader health risks. There are potential public health implications from our results given the availability of mammographic images from widespread annual mammographic screening. Further studies are needed to replicate these results and explore potential mechanisms. Although these questions require clarification, the results of this study suggest a secondary role for analyses of mammographic images that may impact estimation of mortality risk in healthy women.