Our cohort of 801 women had a mean age of 47 years, a mean BMI of 26 kg/m2, and almost half were non-Hispanic white (). Percent mammographic density was nearly normally distributed (skewness = −0.1; kurtosis = 0.9), with a mean of 44.5 (standard deviation = 20.5) and a median of 45.9 (interquartile range = 29.2). BMI had a strong inverse association with percent mammographic density: density increased by 2% for each one kg/m2 decrease in BMI. Mean percent mammographic density differed by race, was lower among older women, current smokers, and among women who previously used oral contraceptives (). Percent density was similar for women with and without a family history of breast cancer.
Study population characteristics by mean percent mammographic density (N = 801)
Results of the unadjusted linear regression models showed that age at menarche was positively associated with mammographic density; those with age at menarche greater than 13 years had 11% higher percent density compared to women with age at menarche less than 12 years (). Three premenstrual symptom groupings were associated with lower percent density: anxiety/mood changes, cramps/back pain, and cravings/bloating. Nulliparous women had a greater mean percent density (mean = 46.6, SD = 23.0), than parous women (mean = 44.0, SD = 19.8, ANOVA P = 0.15). Both younger age at FFTB and greater number of births were inversely associated with percent density in unadjusted models. Premenopausal women had higher density, compared to women in early perimenopause. After adjustment for potential confounders, nearly all menstrual/reproductive factor associations were attenuated and lost statistical significance. Associations with shorter menstrual cycle length and premenstrual groupings for breast pain and headaches were somewhat stronger in adjusted models.
Results from individual linear regression models for each menstrual and reproductive factor in relation to percent mammographic density
When examined together in a single multivariable model, the following menstrual and reproductive variables remained as important predictors of mammographic density: age at menarche, premenstrual cravings and bloating, number of births, and menopausal status (). This model explained 7.0% of the variation in percent mammographic density. Secondary analyses showed that age at FFTB (β = −8.12, P < 0.01 for ≤23 years versus no births), and the combined variable for age at FFTB and number of births (β = −9.19, P < 0.01 for ≤23 years and ≥3 births versus no births) were also important predictors, in separate models with age at menarche, premenstrual cravings and bloating, and menopausal status. These models explained 8.0% and 8.2% of the variation in percent mammographic density, respectively. Menstrual and reproductive factors were not strongly correlated with each other; the strongest correlation was between premenstrual cravings and bloating and age at menarche (r = −0.13). In addition, no evidence for collinearity was found in these models; variance inflation factors were less than 1.7.
Stepwise regression results for percent mammographic density in relation to menstrual and reproductive characteristics
With adjustment for covariates, associations with percent mammographic density remained for age at menarche (positive), number of births (negative) and premenopausal status (positive), however, only number of births remained statistically significant (). In secondary analyses, we observed that age at FFTB (β = −4.59, P = 0.12 for ≤23 years versus no births) and the combined variable for age at FFTB and number of births (β = −4.09, P = 0.13 for ≤23 years and ≥3 births versus no births) were both inversely associated with density, in separate models with age at menarche, premenstrual cravings and bloating, and menopausal status.
The adjusted menstrual/reproductive factor model presented in explained 39.4% of the variance in percent mammographic density. Without BMI, the factor most strongly associated with percent mammographic density in these data, the variance explained by the adjusted base model was 11.8%. To be certain that BMI was not masking associations with additional menstrual or reproductive variables, we used stepwise regression after forcing BMI in the model, but did not identify additional factors associated with percent mammographic density.
Analyses stratified by BMI tertiles revealed modification of menstrual/reproductive factor associations for density (), although no interaction terms with BMI were statistically significant. For example, the inverse association with greater number of births was confined to women within the lowest BMI tertile, or with a BMI less than 21.3 kg/m2 (P for interaction = 0.11) The same trend, although a somewhat weaker association was seen with younger age at FFTB (≤23 years versus no births) for density, among the lowest tertile (β = −8.85, P = 0.02), mid-tertile (β = −1.05, P = 0.96), and highest tertile (β = −2.19, P = 0.50) (P for interaction = 0.12). In contrast, the positive association with premenopausal status was strongest among the heaviest women (P for interaction = 0.23) (). Somewhat stronger associations were observed with age at menarche and number of births among early perimenopausal women, and among ever smokers, compared to premenopausal and never smokers (data not shown). Age (<, ≥ median) or race/ethnicity-study site did not appear to modify the associations between mammographic density and individual reproductive variables (age at menarche, premenstrual cravings and bloating, age at FFTB, number of births, or menopausal status).
Results from a single linear regression model for percent mammographic density by BMI