For the 453 cases and 453 controls of the study sample, the median time between the enrolment in the E3N cohort and mammogram was 11.3 years (interquartile range (IQR): 9.0 to 13.2 years). Fifty percent of the women were born before 1939 (IQR: 1936 to 1943); median age at diagnosis of cases was 61 years (IQR: 59 to 66); the median age at mammogram was 61 years (IQR: 58 to 65 years); the median time between the age at mammograms and at diagnosis was 0.2 years (IQR: 0.1 to 1.6 years). Cases had higher BMI and DA than controls (median BMI: 23.1 versus 22.8 kg/m2, P =0.005; median DA: 35 versus 29 cm2, P<0.001); no statistically significant difference was observed for NDA (median NDA: 68 versus 72 cm2, P =0.36). When the contribution of cases and controls was weighted to properly account for their distribution in the general population, the mean BMI was 23.1 kg/m2 (SD, 3.1 kg/m2), the mean DA and NDA were 33 cm2 (SD: 21) and 79 cm2 (SD: 43) respectively, and the proportion of MHT ever users 0.76. The proportion of MHT ever users was higher among cases than controls (0.83 vs 0.76, P=0.009). Among MHT users, only 4.2% used exclusively estrogen compared to 68.3% who used exclusively estrogens plus progestogens (19.2% used either estrogens plus progesterone or dydrogesterone, 33.3% used estrogens plus progestogens other than progesterone or dydrogesterone, 15.8% used both). Other characteristics of the study sample are reported in Table 1.
Table 1 Characteristics of the women.
Characteristic
|
All (N=906)
|
Controls (N=453)
|
Cases (N=453)
|
Reference age (yrs)*
|
61 (59 to 66)
|
61 (59 to 66)
|
61 (59 to 66)
|
Age at mammogram (yrs)*
|
61 (58 to 65)
|
61 (58 to 65)
|
60 (58 to 65)
|
|
|
|
|
MHT at mammogram, N (%)
|
|
|
|
Never
|
189 (20.9)
|
110 (24.3)
|
79 (17.4)
|
Current
|
432 (47.7)
|
217 (47.9)
|
215 (47.5)
|
Past, 0-2 years since last use
|
154 (17.0)
|
50 (11.0)
|
104 (23.0)
|
Past, >2 years since last use
|
131 (14.5)
|
76 (16.8)
|
55 (12.1)
|
Type of MHT at mammogram, N(%)
|
|
|
|
None
|
189 (20.9)
|
110 (24.3)
|
79 (17.4)
|
Estrogen
|
30 (3.3)
|
15 (3.3)
|
15 (3.3)
|
Estrogens plus progestogens
|
490 (54.1)
|
233 (51.4)
|
257 (56.7)
|
Estrogen plus progesterone or dydrogesterone
|
138 (15.2)
|
79 (17.4)
|
59 (13.0)
|
Estrogen plus progestins
|
239 (26.4)
|
99 (21.9)
|
140 (30.9)
|
Both
|
113 (12.5)
|
55 (12.1)
|
58 (12.8)
|
Others
|
197 (21.7)
|
95 (21.0)
|
102 (22.5)
|
|
|
|
|
BMI at mammogram (kg/m2)*
|
22.9 (21.1 to 25.1)
|
22.8 (21.0 to 24.8)
|
23.1 (21.3 to 25.4)
|
|
|
|
|
PMD*
|
32 (20 to 45)
|
31 (17 to 43)
|
35 (23 to 49)
|
DA (cm2)*
|
33 (20 to 47)
|
29 (17 to 42)
|
35 (23 to 52)
|
NDA (cm2)*
|
70 (47 to 97)
|
72 (49 to 97)
|
68 (46 to 96)
|
|
|
|
|
ER and PR status, N (%)**
|
|
|
|
ER+ and PR+
|
|
|
224 (60.6)
|
ER+ and PR-
|
|
|
77 (20.8)
|
ER- and PR+
|
|
|
13 (3.5)
|
ER- and PR-
|
|
|
56 (15.1)
|
|
|
|
|
Family history of breast cancer in first degree relatives, N (%)
|
|
|
|
No
|
771 (85.1)
|
394 (87.0)
|
377 (83.2)
|
Yes
|
135 (14.9)
|
59 (13.0)
|
76 (16.8)
|
|
|
|
|
Age of menarche (yrs), N (%)
|
|
|
|
<12
|
162 (17.9)
|
69 (15.2)
|
93 (20.5)
|
12
|
244 (26.9)
|
125 (27.6)
|
119 (26.3)
|
>12
|
500 (55.2)
|
259 (57.2)
|
241 (53.2)
|
|
|
|
|
Past use of oral contraceptives, N (%)
|
|
|
|
No
|
452 (49.9)
|
232 (51.2)
|
220 (48.6)
|
Yes
|
454 (50.1)
|
221 (48.8)
|
233 (51.4)
|
|
|
|
|
Parity and lactation, N (%)**
|
|
|
|
Nulliparous
|
116 (13.6)
|
59 (14.0)
|
57 (13.3)
|
Parous without lactation
|
205 (24.1)
|
94 (22.3)
|
111 (25.8)
|
Parous with lactation for less than 4 months
|
246 (28.9)
|
123 (29.2)
|
123 (28.6)
|
Parous with lactation for 4 months or more
|
284 (33.4)
|
145 (34.4)
|
139 (32.3)
|
For each matched case-control pair, reference age is the age at diagnosis of the case.
MHT: menopausal hormone therapy; BMI: body mass index; PMD: percent mammographic density; DA: dense area; NDA, non-dense area; ER: estrogen receptor; PR: progesterone receptor.
*Median (interquartile range)
**Number of missing: ER/PR status, 83; Parity and lactation, 55
Compared to the whole E3N cohort, the control group of this study sample had higher proportions of women who at baseline never used MHT (87.4% versus 78.8%); never used oral contraceptives (51.2% versus 45.3%); had a BMI lower than 25 kg/m2 (88.1 versus 82.2); breastfed for a more than 4 months (40% versus 22%); had a family history of breast cancer in first-degree relatives (13.0% versus 8.1%) (Supplementary Table S1).
ASSOCIATION BETWEEN MENOPAUSAL HORMONE THERAPY AND MAMMOGRAPHIC DENSITY
MHT use status at mammogram was significantly associated with PMD, DA and NDA (P<0.001, <0.001 and 0.006, respectively): for PMD and DA, past users had significantly higher values than never users (P=0.003 and P=0.005 respectively) and significantly lower values than current users (P=0.004 and P=0.003 respectively); for NDA, both past and current users had lower levels than never users, but differences were statistically significant only for current versus never users (P=0.001). Table 2 shows the predicted mammographic density measures by MHT status for women aged 60 and 70 years. Distinguishing MHT ever users by type of MHT formulation (estrogen versus estrogen plus progesterone or dydrogesterone versus estrogen plus other progestogens) did not improve the fit for any of the mammographic density variables (P=0.92, 0.86 and 0.67, for PMD, DA and NDA respectively).
Table 2 Predicted mammographic measures in women 60 and 70 years old by menopausal hormone therapy use.
|
PMD (%)
|
DA (cm2)
|
NDA (cm2)
|
|
Predicted (95% CI)
|
Predicted (95% CI)
|
Predicted (95% CI)
|
For a woman aged 60 years
|
|
|
|
Never
|
24 (22 to 26)
|
24 (22 to 26)
|
78 (72 to 83)
|
Past use
|
29 (27 to 31)
|
29 (26 to 31)
|
71 (66 to 76)
|
Current use
|
33 (31 to 35)
|
33 (31 to 35)
|
67 (63 to 70)
|
For a woman aged 70 years
|
|
|
|
Never
|
20 (18 to 23)
|
24 (21 to 27)
|
94 (87 to101)
|
Past use
|
25 (22 to 27)
|
29 (26 to 32)
|
87 (80 to 93)
|
Current use
|
29 (26 to 32)
|
33 (30 to 37)
|
82 (75 to 89)
|
PMD: percent mammographic density; DA: dense area; NDA: non-dense area; MHT: menopausal hormone therapy.
To assess the effect of pattern of use of MHT on mammographic density, we fitted a model that included both duration and time since last use categorized according to the medians in all women (6 and 2 years, respectively). Among past users, time since last use was negatively associated to PMD (P=0.009) and DA (P<0.001), and positively but not significantly associated to NDA (P=0.24). There was no statistically significant association of duration with PMD, DA nor NDA (all P>0.05). For none of the mammographic density variables, adding the interaction between duration and time since last use significantly improved the model. Then duration was excluded from the model. Table 3 reports the corresponding predictions of PMD, DA and NDA for women aged 60 and 70 years. For all three mammographic density variables, the values for past users who stopped MHT less than 2 years earlier were not statistically significantly different from those of current users (P=0.65 for PMD, 0.88 for DA and 0.95 for NDA), whereas the values for past users who stopped more than 2 years earlier were not significantly different from those of never users (P=0.19, 0.39 and 0.44, respectively).
Table 3 Predicted mammographic measures in women 60 and 70 years old by pattern of menopausal hormone therapy use.
|
PMD (%)
|
DA (cm2)
|
NDA (cm2)
|
|
Predicted (95% CI)
|
Predicted (95% CI)
|
Predicted (95% CI)
|
For women aged 60 years
|
|
|
|
Never
|
24 (22 to 26)
|
24 (22 to 26)
|
78 (72 to 83)
|
Current
|
33 (31 to 35)
|
33 (31 to 35)
|
67 (63 to 70)
|
Past, 0-2 yrs since last use
|
32 (29 to 36)
|
33 (29 to 37)
|
67 (60 to 74)
|
Past, >2 yrs since last use
|
26 (24 to 29)
|
25 (23 to 28)
|
74 (68 to 81)
|
For women aged 70 years
|
|
|
|
Never
|
20 (18 to 23)
|
24 (21 to 27)
|
94 (86 to 101)
|
Current
|
29 (26 to 32)
|
33 (30 to 37)
|
82 (75 to 89)
|
Past, 0-2 yrs since last use
|
28 (24 to 32)
|
33 (29 to 38)
|
81 (73 to 91)
|
Past, >2 yrs since last use
|
23 (20 to 26)
|
26 (23 to 29)
|
90 (82 to 98)
|
PMD: percent mammographic density; DA: dense area; NDA, non-dense area.
For all three mammographic density variables, the best polynomial model included the first-grade polynomial for duration and the second-degree polynomial for time since last use (supplementary Table S2). The trends by age of PMD, DA and NDA predicted by such models are shown in Figure 1 for a woman who has never used MHT; a woman who has started MHT at the age of 55 and who has never stopped; and a woman who has started at the age of 55 and has stopped after 3, 6 or 8 years (second, third and fourth quartiles of the MHT duration). It takes less than one year of use of MHT for mammographic density to reach a plateau, as indicated by the discontinuity of the predicted curve between never and current users. According to the models, for a woman who has started MHT at 55 years and who has stopped after 3 years, the levels of PMD, DA and NDA returned to the levels of never users after approximately 8, 9 and 4 years respectively; after 8, 11 and 6 years if the same woman has stopped MHT after 6 years; after 8, 12 and 6 years if she has stopped after 8 years.
The adjustment for additional potential confounders did not materially change any of the above estimates.
MEDIATION ANALYSIS
Considering that the effect of MHT on mammographic density can be observed for up to 8 years after MHT discontinuation (as reported in the previous section), we conducted mediation analyses only on current and never MHT users. In the model adjusted for age and for the matching variables, the OR of BC associated with current versus never use of MHT was 1.67 (95% CI, 1.04 to 2.68); when PMD was added into the model, the OR became 1.40 (0.86 to 2.28) that corresponds to a 34% mediated effect on the log scale. When mediation analysis was conducted stratifying by hormone receptor status, it appeared that the association between MHT and BC risk was mainly due to hormone receptor positive breast cancers: the OR of hormone receptor positive BC associated to with current use of MHT was 1.81 (1.05 to 3.10); when PMD was added into the model the OR became 1.46 (0.84 to 2.57), that corresponds to a 36% mediated effect on the log scale. For hormone receptor negative BC, the association between MHT and BC was not significant, either without or with inclusion of PMD into the model (OR = 0.64, 0.15 to 2.81, and OR = 0.57, 0.12 to 2.70, respectively). Similar findings were obtained from the regression models adjusted for all potential confounders, where the proportion of the effect of MHT mediated by PMD was 40% for any BC and 41% for hormone receptor positive BC.
Table 4 reports the results from the mediation model that accounted for the joint mediation effect of PMD and BMI (Pearson’s correlation coefficient between square-rooted transformed PMD and BMI -0.39, P<0.001). In the model adjusted for age and matching variables, the OR associated with current use of MHT was 1.46 (0.42 to 2.41); the average direct effect was 1.37 (0.85 to 2.18); the average mediated effects were 1.20 (1.06 to 1.41) through PMD and 0.90 (0.78 to 0.98) through BMI. When mediation analysis was stratified by hormone receptor status, it appeared that the association between MHT and BC risk was entirely due to hormone receptor positive breast cancers: the total effect of MHT was 1.58 (0.93 to 2.64), resulting from a direct effect of 1.44 (0.85 to 2.34) and indirect effects of 1.22 (1.06 to 1.52) through PMD and 0.91 (0.78 to 0.99) through BMI. According to this model the proportion of the effect mediated by PMD on the log scale was 48% for any BC and 43% for hormone receptor positive BC.
No material changes were observed in the results of the mediation analysis when the models were adjusted for all potential confounders (results not shown). Qualitatively similar results were observed for MHT coded as ever versus never (supplementary Table S3).
Table 4 Mediation analysis of the effect of current versus never use (reference category) of menopausal hormone therapy on breast cancer risk, overall and by ER and PR status. The table reports the OR and 95% confidence intervals from the unconditional logistic models adjusted for age at mammogram and the matching variables (reference age, year of birth and menopausal status at baseline).
|
All
OR (95% CI)
|
ER+ or PR+
OR (95% CI)
|
ER- and PR-
OR (95% CI)
|
|
NCA=196/NCO=196
|
NCA=150/NCO=150
|
NCA=22/NCO=22
|
Total effect
|
1.46 (0.92 to 2.41)
|
1.58 (0.93 to 2.64)
|
0.63 (0.11 to 2.61)
|
Average direct effect
|
1.37 (0.85 to 2.18)
|
1.44 (0.85 to 2.34)
|
0.61 (0.10 to 2.48)
|
Average joint mediated effect
|
1.07 (0.92 to 1.25)
|
1.10 (0.93 to 1.37)
|
1.00 (0.48 to 2.52)
|
Average mediated effect by PMD
|
1.20 (1.06 to 1.41)
|
1.22 (1.06 to 1.52)
|
1.10 (0.61 to 2.63)
|
Average mediated effect by BMI
|
0.90 (0.78 to 0.98)
|
0.91 (0.78 to 0.99)
|
1.00 (0.49 to 1.40)
|
Mediation analysis evaluating the direct and indirect effect of current versus never use of MHT on breast cancer risk, overall and by ER and PR status in presence of PMD and BMI, correlated mediators.