Background:

Identifying risk factors for aggressive forms of breast cancer is important. Tumor factors (e.g., stage) are important predictors of prognosis, but may be intermediates between prediagnosis risk factors and mortality. Typically, separate models are fit for incidence and mortality postdiagnosis. These models have not been previously integrated to identify risk factors for lethal breast cancer in cancer-free women.

Methods:

We combined models for breast cancer incidence and breast cancer–specific mortality among cases into a multi-state survival model for lethal breast cancer. We derived the model from cancer-free postmenopausal Nurses’ Health Study women in 1990 using baseline risk factors. A total of 4,391 invasive breast cancer cases were diagnosed from 1990 to 2014 of which 549 died because of breast cancer over the same period.

Results:

Some established risk factors (e.g., family history, estrogen plus progestin therapy) were not associated with lethal breast cancer. Controlling for age, the strongest risk factors for lethal breast cancer were weight gain since age 18: > 30 kg versus ± 5 kg, RR = 1.94 [95% confidence interval (CI) = 1.38–2.74], nulliparity versus age at first birth (AAFB) < 25, RR = 1.60 (95% CI = 1.16–2.22), and current smoking ≥ 15 cigarettes/day versus never, RR = 1.42 (95% CI = 1.07–1.89).

Conclusions:

Some breast cancer incidence risk factors are not associated with lethal breast cancer; other risk factors for lethal breast cancer are not associated with disease incidence.

Impact:

This multi-state survival model may be useful for identifying prediagnosis factors that lead to more aggressive and ultimately lethal breast cancer.

This article is featured in Highlights of This Issue, p. 1515

For cancer, tumor factors (e.g., stage, grade) and treatment are key predictors of survival. However, growing evidence suggests that prediagnosis factors may determine the developmental pathways of carcinogenesis that can impact tumor aggressiveness and may increase lethality. For example, prediagnosis factors have been associated with increased risk of lethal prostate cancer (1–3) and rapidly fatal ovarian cancer (4, 5).

We have previously developed a risk prediction model for postmenopausal breast cancer incidence based on lifestyle factors that represents a simplified version of the validated Rosner-Colditz model (6, 7). However, not all breast cancer risk factors are associated with survival after diagnosis. For example, current use of postmenopausal hormone therapy (HT) is a strong positive risk factor for incident breast cancer, but has been inversely associated with breast cancer mortality (8, 9). In addition, a previous analysis looked at associations between updated smoking status and mortality and reported a HR = 1.3 for current versus never smoking and breast cancer mortality (10). However, both women with prevalent and incident breast cancer were included in this analysis. Furthermore, all patients with breast cancer are encouraged to avoid smoking, or engage in smoking cessation if currently smoking (11).

Little is known about risk factors for lethal breast cancer among disease-free women. Furthermore, to our knowledge, analytic methods for determining associations of risk factors with lethal cancer among disease-free subjects are not well developed. In this article, we propose a novel multi-state model evaluating the role of prediagnosis factors in lethal breast cancer risk using data from disease-free postmenopausal women in the Nurses’ Health Study (NHS).

The NHS was established in 1976 among 121,700 U.S. female registered nurses, ages 30 to 55 years. Participants completed an initial questionnaire about their lifestyle factors, health behaviors, and medical history, and, since baseline, have been followed biennially by questionnaires to update exposure status and disease diagnoses (12). Analyses began in 1990, as there were an appreciable number of postmenopausal women age ≥ 45 with no prior cancer (except for nonmelanoma skin cancer; N = 56,346). The goal was to identify risk factors assessed in 1990 that were associated with developing incident breast cancer and dying with breast cancer as the cause of death through December, 2014.

Postmenopausal breast cancer and death ascertainment

Breast cancer cases were identified through self-report on biennial questionnaires or the National Death Index. We requested permission to access medical records on all cases; investigators blinded to exposure status reviewed these records to confirm the diagnosis. This analysis included all confirmed invasive postmenopausal breast cancer cases (n = 4,391) diagnosed between July 1, 1990 and December 31, 2014.

Deaths were identified via family members, the U.S. Postal Service, or the National Death Index (13, 14). Cause of death was based on review of medical records and death certificates by study physicians blinded to exposure status. Breast cancers were considered lethal if the death certificate listed breast cancer as the primary cause of death or was determined to be the cause of death by medical record review. We observed 1,568 deaths after incident breast cancer through December 31, 2014 of which 549 were due to breast cancer.

Exposure variables

All exposures investigated in this study were assessed as of 1990 and have been studied for their association with incidence of total invasive breast cancer (Table 1; refs. 15–19). In addition, we also considered smoking because it has been previously associated with breast cancer prognosis (20–22). All analyses were conducted in SAS v9.4 (23), and all P values are two sided.

Table 1.

Univariate comparisons in 1990 of postmenopausal women who developed breast cancer between July 1, 1990 and December 31, 2014 versus those who did not develop breast cancer over the same time period, NHS.

Breast cancer
Yes (N = 4,391)No (N = 51,955)
VariableCategoryN (%)N (%)χ2aP
Age (yrs) ≥45, <55 1,019 (23) 13,171 (25) (ref)  
 ≥55, <60 1,250 (28) 14,246 (27) 8.10 0.004 
 ≥60, <65 1,210 (28) 13,593 (26) 9.92 0.002 
 ≥65 912 (21) 10,945 (21) 2.38 0.12 
Duration of menopause (yrs) ≥0, <5 1,062 (24) 12,071 (23) (ref)  
 ≥5, <10 1,232 (28) 13,287 (26) 1.39 0.24 
 ≥10 2,097 (48) 26,597 (51) 7.71 0.006 
 Mean (SD) 9.5 (6.1) 10.1 (6.5)  <0.001b 
Pregnancy history Nulliparous 307 (7) 3,103 (6) 13.17 <0.001 
 AAFB <25 1,951 (44) 24,945 (48) (ref)  
 AAFB ≥25, <30 1,650 (38) 18,760 (36) 11.26 <0.001 
 AAFB ≥30 483 (11) 5,147 (10) 11.62 <0.001 
Benign breast disease (BBD)c No 3,294 (75) 41,721 (80) (ref)  
 Yes 1,097 (25) 10,234 (20) 70.46 <0.001 
First degree family history of breast cancer No 3,735 (85) 46,454 (89) (ref)  
 Yes 656 (15) 5,501 (11) 80.26 <0.001 
Current HT use No 2,457 (56) 32,221 (62) (ref)  
 Estrogen only 872 (20) 9777 (19) 14.41 <0.001 
 Estrogen & progestin 718 (16) 6093 (12) 95.76 <0.001 
 Other 344 (8) 3,864 (7) 6.50 0.011 
Height (cm.) Mean (SD) 64.6 (2.3) 64.4 (2.4)  0.001b 
BMI at age 18 (kg/m2≥15, <18.5 612 (14) 6,489 (12) 2.39 0.12 
 ≥18.5 <23 2,973 (68) 33,905 (65) (ref)  
 ≥23, <25 484 (11) 6,374 (12) 7.87 0.005 
 ≥25 322 (7) 5,187 (10) 32.54 <0.001 
Weight change since age 18 (kg) <−5 110 (3) 2,508 (5) 20.64 <0.001 
 ≥−5, ≤5 908 (21) 12,942 (25) (ref)  
 >5, ≤10 885 (20) 10,258 (20) 17.61 <0.001 
 >10, ≤20 1,395 (32) 15,431 (30) 32.68 <0.001 
 >20, ≤30 748 (17) 7,165 (14) 59.70 <0.001 
 >30 345 (8) 3,651 (7) 20.19 <0.001 
Alcohol intake (g/day) 1,641 (37) 20,885 (40) (ref)  
 >0, <11 1,959 (45) 22,307 (43) 10.11 0.001 
 ≥11, <22 480 (11) 5,523 (11) 3.38 0.066 
 ≥22 311 (7) 3,240 (6) 9.40 0.002 
Cigarette smoking Never 1,910 (44) 22,288 (43) (ref)  
 Past 1,802 (41) 20,692 (40) 0.21 0.65 
 Current, 1–14 cigarettes/day 239 (5) 3,231 (6) 4.14 0.042 
 Current, ≥15 cigarettes/day 440 (10) 5,744 (11) 4.07 0.044 
Breast cancer
Yes (N = 4,391)No (N = 51,955)
VariableCategoryN (%)N (%)χ2aP
Age (yrs) ≥45, <55 1,019 (23) 13,171 (25) (ref)  
 ≥55, <60 1,250 (28) 14,246 (27) 8.10 0.004 
 ≥60, <65 1,210 (28) 13,593 (26) 9.92 0.002 
 ≥65 912 (21) 10,945 (21) 2.38 0.12 
Duration of menopause (yrs) ≥0, <5 1,062 (24) 12,071 (23) (ref)  
 ≥5, <10 1,232 (28) 13,287 (26) 1.39 0.24 
 ≥10 2,097 (48) 26,597 (51) 7.71 0.006 
 Mean (SD) 9.5 (6.1) 10.1 (6.5)  <0.001b 
Pregnancy history Nulliparous 307 (7) 3,103 (6) 13.17 <0.001 
 AAFB <25 1,951 (44) 24,945 (48) (ref)  
 AAFB ≥25, <30 1,650 (38) 18,760 (36) 11.26 <0.001 
 AAFB ≥30 483 (11) 5,147 (10) 11.62 <0.001 
Benign breast disease (BBD)c No 3,294 (75) 41,721 (80) (ref)  
 Yes 1,097 (25) 10,234 (20) 70.46 <0.001 
First degree family history of breast cancer No 3,735 (85) 46,454 (89) (ref)  
 Yes 656 (15) 5,501 (11) 80.26 <0.001 
Current HT use No 2,457 (56) 32,221 (62) (ref)  
 Estrogen only 872 (20) 9777 (19) 14.41 <0.001 
 Estrogen & progestin 718 (16) 6093 (12) 95.76 <0.001 
 Other 344 (8) 3,864 (7) 6.50 0.011 
Height (cm.) Mean (SD) 64.6 (2.3) 64.4 (2.4)  0.001b 
BMI at age 18 (kg/m2≥15, <18.5 612 (14) 6,489 (12) 2.39 0.12 
 ≥18.5 <23 2,973 (68) 33,905 (65) (ref)  
 ≥23, <25 484 (11) 6,374 (12) 7.87 0.005 
 ≥25 322 (7) 5,187 (10) 32.54 <0.001 
Weight change since age 18 (kg) <−5 110 (3) 2,508 (5) 20.64 <0.001 
 ≥−5, ≤5 908 (21) 12,942 (25) (ref)  
 >5, ≤10 885 (20) 10,258 (20) 17.61 <0.001 
 >10, ≤20 1,395 (32) 15,431 (30) 32.68 <0.001 
 >20, ≤30 748 (17) 7,165 (14) 59.70 <0.001 
 >30 345 (8) 3,651 (7) 20.19 <0.001 
Alcohol intake (g/day) 1,641 (37) 20,885 (40) (ref)  
 >0, <11 1,959 (45) 22,307 (43) 10.11 0.001 
 ≥11, <22 480 (11) 5,523 (11) 3.38 0.066 
 ≥22 311 (7) 3,240 (6) 9.40 0.002 
Cigarette smoking Never 1,910 (44) 22,288 (43) (ref)  
 Past 1,802 (41) 20,692 (40) 0.21 0.65 
 Current, 1–14 cigarettes/day 239 (5) 3,231 (6) 4.14 0.042 
 Current, ≥15 cigarettes/day 440 (10) 5,744 (11) 4.07 0.044 

Abbreviations: AAFB: age at first birth; BBD: biopsy-confirmed benign breast disease; BMI: body mass index; HT: hormone therapy; yrs: years.

aPearson χ2 (1 df) versus reference group.

bTwo-sample t test.

cBiopsy-confirmed BBD.

Multi-state models

We sought to identify risk factors for lethal breast cancer among disease-free women. This is a special case of a type of multi-state model referred to as an illness-death model (24–25). For example, if state 1 = healthy, state 2 = incident disease (D), and state 3 = death due to D, then we wish to determine the probability of transitioning from state 1 to state 3 over some time period, t. In some illness-death models, it is possible to transition from state 1 to state 3 without going through state 2. However, in our application, state 1 = no previous cancer other than nonmelanoma skin cancer, state 2 = incident breast cancer, and state 3 = death due to breast cancer, and one must go through state 2 to transition from state 1 to state 3.

Competing risks

There are competing risks both for transitions from state 1 to 2 and from state 2 to 3. There are two competing events for the transition from state 1 to 2: (i) women may die of another cause, or (ii) women may develop another type of cancer (other than nonmelanoma skin cancer) before they develop breast cancer, but if they do and subsequently die, it may be difficult to identify which was the cause of death. Also, treatment for the other cancer may modify the effects of the baseline risk factors on breast cancer. Similarly, in the transition from state 2 to 3, a competing event is death from another cause after they develop breast cancer.

Multi-state model for lethal breast cancer

We consider a group of postmenopausal women age ≥ 45 who are free of cancer (except for nonmelanoma skin cancer) as of 1990. We consider the cumulative incidence of developing breast cancer and then dying of breast cancer through December 31, 2014. We use time since July 1, 1990 as the time scale for incidence and time since diagnosis as the time scale for mortality due to breast cancer. Suppose also that the main exposure is x and other risk factors are denoted by | $\underline {z} $ |⁠. If | $CIF( {t,x,\underline{z} } )$ | = cumulative incidence of fatal breast cancer over t years among disease-free women at time 0 (i.e., 1990), then to die of breast cancer at some time over t years, one must first develop breast cancer at time t1 years (0 < t1 < t), and then die of breast cancer at some time over the next tt1 years, or

formula

where

| ${I}_1( {{t}_1,x,\underline{z} } )$ || ${\rm{\ }}$ |= incidence of breast cancer at time t1 years after t0 (1990), adjusted for competing risks,

| $CI{F}_1( {{t}_1,x,\underline{z} } )$ | = cumulative incidence of breast cancer over t1 years, adjusted for competing risks,

| $CI{F}_2( {t - {t}_1,x,\underline{z} } )$ | = cumulative incidence of breast cancer mortality among patients with breast cancer during tt1 years after diagnosis, adjusted for competing risks.

We have:

formula

where

formula

= 1 − cumulative incidence for breast cancer adjusted for competing risks (25).

Also,

formula

= subdistribution hazard function for breast cancer incidence, and

formula

Similarly, for breast cancer mortality, we have:

formula

where

formula

Thus,

formula

We now compare the cumulative incidence of lethal breast cancer between two women who differ by 1 unit on x and have the same | $\underline{z} $ |⁠, and quantify it by the relative risk | $\ \equiv RR( t ) = CIF( {t,x + 1,\underline{z} } )/$ || $CIF( {t,x,\underline{z} } ).$ | On the basis of Eq. (B), RR(t) is a nonlinear function of | $\underline{\beta } $ | and | $\underline{\gamma } $ |⁠. Also, since risk factors are only available in discrete time if we approximate | ${\rm{ln}}[ {RR( t )} ]$ | by a first-order Taylor series about | $\underline{\beta } $ | = 0 and | $\underline{\gamma } $ | = 0 and replace the integrals by sums (Supplementary Materials and Methods) we obtain:

formula

We note that | ${w}_1\ {\rm{and}}\ {w}_2$ | depend on the baseline hazard and survival function for incidence and mortality postdiagnosis over the entire follow-up period. Thus, in our application, | ${\rm{ln}}[ {RR( t )} ]$ | is the log relative risk for fatal breast cancer over t years among cancer-free women (other than nonmelanoma skin cancer) at baseline, where

formula

| $CI{F}_{01}( {{t}_1} )\ {\rm{and}}\ CI{F}_{02}( {t - {t}_1} )$ | are the baseline cumulative incidence functions (CIF) for incidence and mortality for a subject with average covariate levels using the Fine–Gray approach (26) and | ${\beta }_1,\, {\beta }_2$ | are the regression coefficients for x in the subdistribution hazard models for incidence and mortality, respectively.

We estimate ln[RR(t)] by ln[ | $\widehat {RR}( t )] = {w}_1{\hat{\beta }}_1 + {w}_2{\hat{\beta }}_2$ | and note that:

formula

Thus, a large sample 100% × (1 − α) CI for RR(t) = | $[\exp ( {{c}_1} ),\exp ( {{c}_2} )],$ | where:

formula

and zp = pth percentile of a N(0,1), distribution.

In Eq. (D), we have assumed that | ${w}_1$ | and | ${w}_2$ | are constants, or have low variability compared with | ${\beta }_1$ | and | ${\beta }_2.$ | To check the validity of this assumption, we have performed a bootstrap analysis creating 100 bootstrap samples with replacement each of size 56,346 from the original sample of 56,346 women. We then computed | ${R}_k = {\rm{ln}}({A}_k/{B}_k),\ $ | where | ${A}_k\ $ |= sample variance from the 100 bootstrap estimates of ln[RR(t)] and | ${B}_k = $ | large sample variance of Var{ln[ | $\widehat {RR}$ |(t)]} for the kth regression coefficient from Eq. (D), k = 1,…, K, and

formula

is an overall index of comparability of the bootstrap variance to the large sample variance. A 95% confidence interval (CI) associated with | $\bar{R}$ | is given by | $\bar{R}$ || $\pm \ 1.96\ \mathop \sum \nolimits_{k = 1}^K \{ {[({R}_k - {\overline {R)} }^2] / [K( {K - 1} )]\} }^{1/2}$ |⁠, which will include 0 if the large sample variance is adequate.

Data availability

The data used in this analysis are available from the authors upon reasonable request.

We compare the risk factor profile in 1990 of women who developed breast cancer between 1990 and 2014 (case group, n = 4,391) versus women who did not develop breast cancer (control group, n = 51,955) based on univariate comparisons (Table 1).

We observed that women in the case group were slightly older, and had a slightly shorter time since menopause versus control women. In addition, case women were slightly more likely to either have a late AAFB or be nulliparous than control women, and were more likely than control women to have reported previous biopsy-confirmed benign breast disease (BBD) or have a first-degree family history of breast cancer. Case women were also more likely than control women to currently use Estrogen & Progesterone (E&P) HT, and less likely to be overweight at age 18. Furthermore, case women had substantially greater weight gain since age 18 and reported slightly greater alcohol intake than control women. Also, cases and controls had a similar cigarette smoking profile.

Overall, among the 4,391 breast cancer cases, 1,568 women died by the end of follow-up (549 deaths due to breast cancer; 1,019 deaths due to other causes) and 2,823 women were still alive. In Table 2, among breast cancer cases, we compared the risk factor profile of women who died of breast cancer (breast cancer deaths) with those who did not die of breast cancer as of the end of follow-up (breast cancer controls) based on univariate comparisons. We see that there was a strong positive association with age; 26% of breast cancer deaths versus 20% of breast cancer controls were age ≥65 in 1990 (age ≥ 65 vs. 45–54, P < 0.001). Women who died of breast cancer also had a longer duration of menopause (≥10 years vs. < 5 years, P = 0.005). No associations were apparent for either pregnancy history or presence of biopsy-confirmed BBD. However, there was a nonsignificant inverse trend with family history of breast cancer; 12% of breast cancer deaths versus 15% of breast cancer controls had a first-degree family history of breast cancer (P = 0.084). Similarly, there was an inverse association between E&P HT use and breast cancer mortality; 11% of breast cancer deaths versus 17% of breast cancer controls were current users of E&P HT in 1990 (P < 0.001). There was no association with breast cancer mortality for either height, body mass index (BMI) at age 18, or alcohol intake. However, there was a positive association with weight change since age 18; 30% of breast cancer deaths versus 25% of breast cancer controls had a weight change of >20 kg since age 18 (> 20, ≤30 kg vs. no change, P = 0.029; >30 kg. vs. no change, P = 0.059). Finally, there was a positive association of heavy current smoking with breast cancer mortality (13% of breast cancer deaths vs. 10% of breast cancer controls smoked ≥ 15 cigarettes/day in 1990, P = 0.041).

Table 2.

Univariate comparisons among cases who died due to breast cancer (breast cancer deaths) between July 1, 1990 and December 31, 2014 versus those who did not die due to breast cancer over the same time interval, NHS.

Breast cancer death
Yes (N = 549)No (N = 3,842)
VariableCategoryN (%)N (%)χ2aP
Age (yrs) ≥45, <55 94 (17) 925 (24) (ref)  
 ≥55, <60 133 (24) 1,117 (29) 1.10 0.30 
 ≥60, <65 179 (33) 1,031 (27) 15.45 <0.001 
 ≥65 143 (26) 769 (20) 18.03 <0.001 
Duration of menopause (yrs) ≥0, <5 114 (21) 948 (25) (ref)  
 ≥5, <10 134 (24) 1,098 (29) 0.00 0.97 
 ≥10 301 (55) 1,796 (47) 7.78 0.005 
 Mean (SD) 10.4 (6.3) 9.4 (6.1)  <0.001b 
Pregnancy history Nulliparous 44 (8) 263 (7) 1.19 0.27 
 AAFB <25 233 (42) 1,718 (45) (ref)  
 AAFB ≥25, <30 207 (38) 1,443 (38) 0.25 0.62 
 AAFB ≥30 65 (12) 418 (11) 0.69 0.41 
Benign breast disease (BBD)c No 415 (76) 2,879 (75) (ref)  
 Yes 134 (24) 963 (25) 0.08 0.78 
First degree family history of breast cancer No 481 (88) 3,254 (85) (ref)  
 Yes 68 (12) 588 (15) 2.99 0.084 
Current HT use No 339 (62) 2,118 (55) (ref)  
 Estrogen only 113 (21) 759 (20) 0.32 0.57 
 Estrogen & progestin 59 (11) 659 (17) 15.27 <0.001 
 Other 38 (7) 306 (8) 1.73 0.19 
Height (cm) Mean (SD) 64.5 (2.3) 64.6 (2.3)  0.37b 
BMI at age 18 (kg/m2≥15, <18.5 77 (14) 535 (14) 0.01 0.91 
 ≥18.5 <23 382 (70) 2,591 (67) (ref)  
 ≥23, <25 51 (9) 433 (11) 1.82 0.18 
 ≥25 39 (7) 283 (7) 0.08 0.77 
Weight change since age 18 (kg) <−5 18 (3) 92 (2) 2.01 0.16 
 ≥−5, ≤5 102 (19) 806 (21) (ref)  
 >5, ≤10 103 (19) 782 (20) 0.04 0.85 
 >10, ≤20 161 (29) 1,234 (32) 0.03 0.87 
 >20, ≤30 112 (20) 636 (17) 4.77 0.029 
 >30 53 (10) 292 (8) 3.56 0.059 
Alcohol intake (g/day) 219 (40) 1,422 (37) (ref)  
 >0, <11 236 (43) 1,723 (45) 1.25 0.26 
 ≥11, <22 55 (10) 425 (11) 1.01 0.31 
 ≥22 39 (7) 272 (7) 0.09 0.77 
Cigarette smoking Never 236 (43) 1,674 (44) (ref)  
 Past 214 (39) 1,588 (41) 0.69 0.16 
 Current, 1–14 cigarettes/day 28 (5) 211 (5) 0.03 0.86 
 Current, ≥15 cigarettes/day 71 (13) 369 (10) 4.17 0.041 
Breast cancer death
Yes (N = 549)No (N = 3,842)
VariableCategoryN (%)N (%)χ2aP
Age (yrs) ≥45, <55 94 (17) 925 (24) (ref)  
 ≥55, <60 133 (24) 1,117 (29) 1.10 0.30 
 ≥60, <65 179 (33) 1,031 (27) 15.45 <0.001 
 ≥65 143 (26) 769 (20) 18.03 <0.001 
Duration of menopause (yrs) ≥0, <5 114 (21) 948 (25) (ref)  
 ≥5, <10 134 (24) 1,098 (29) 0.00 0.97 
 ≥10 301 (55) 1,796 (47) 7.78 0.005 
 Mean (SD) 10.4 (6.3) 9.4 (6.1)  <0.001b 
Pregnancy history Nulliparous 44 (8) 263 (7) 1.19 0.27 
 AAFB <25 233 (42) 1,718 (45) (ref)  
 AAFB ≥25, <30 207 (38) 1,443 (38) 0.25 0.62 
 AAFB ≥30 65 (12) 418 (11) 0.69 0.41 
Benign breast disease (BBD)c No 415 (76) 2,879 (75) (ref)  
 Yes 134 (24) 963 (25) 0.08 0.78 
First degree family history of breast cancer No 481 (88) 3,254 (85) (ref)  
 Yes 68 (12) 588 (15) 2.99 0.084 
Current HT use No 339 (62) 2,118 (55) (ref)  
 Estrogen only 113 (21) 759 (20) 0.32 0.57 
 Estrogen & progestin 59 (11) 659 (17) 15.27 <0.001 
 Other 38 (7) 306 (8) 1.73 0.19 
Height (cm) Mean (SD) 64.5 (2.3) 64.6 (2.3)  0.37b 
BMI at age 18 (kg/m2≥15, <18.5 77 (14) 535 (14) 0.01 0.91 
 ≥18.5 <23 382 (70) 2,591 (67) (ref)  
 ≥23, <25 51 (9) 433 (11) 1.82 0.18 
 ≥25 39 (7) 283 (7) 0.08 0.77 
Weight change since age 18 (kg) <−5 18 (3) 92 (2) 2.01 0.16 
 ≥−5, ≤5 102 (19) 806 (21) (ref)  
 >5, ≤10 103 (19) 782 (20) 0.04 0.85 
 >10, ≤20 161 (29) 1,234 (32) 0.03 0.87 
 >20, ≤30 112 (20) 636 (17) 4.77 0.029 
 >30 53 (10) 292 (8) 3.56 0.059 
Alcohol intake (g/day) 219 (40) 1,422 (37) (ref)  
 >0, <11 236 (43) 1,723 (45) 1.25 0.26 
 ≥11, <22 55 (10) 425 (11) 1.01 0.31 
 ≥22 39 (7) 272 (7) 0.09 0.77 
Cigarette smoking Never 236 (43) 1,674 (44) (ref)  
 Past 214 (39) 1,588 (41) 0.69 0.16 
 Current, 1–14 cigarettes/day 28 (5) 211 (5) 0.03 0.86 
 Current, ≥15 cigarettes/day 71 (13) 369 (10) 4.17 0.041 

Abbreviations: AAFB: age at first birth; BBD: biopsy-confirmed benign breast disease; BMI: body mass index; HT: hormone therapy; yrs: years.

aPearson χ2 (1 df) versus reference group.

bTwo-sample t test.

cBiopsy-confirmed BBD.

In Table 3, we present multivariate Cox proportional hazards models for breast cancer incidence, survival among breast cancer cases, each based on a Fine–Gray analysis, and cumulative incidence of lethal breast cancer. We see that there was an increased risk with age for both incidence and mortality resulting in an overall increased risk for lethal breast cancer [age ≥ 65 vs. 45–54: RR = 2.78, 95% CI = (1.92–4.04)]. Duration of menopause was inversely associated with lethal breast cancer [RR per 5 years = 0.80, 95% CI = (0.72–0.88)], due to a significant inverse association with incidence and a nonsignificant inverse trend with mortality. Regarding pregnancy history, only nulliparity was positively associated with lethal breast cancer [nulliparous vs. AAFB < 25: RR = 1.60, 95% CI = (1.16–2.22)], due to significant positive associations with incidence and nonsignificant positive trends with mortality. Age at first birth ≥25 was weakly positively, but significantly associated with incidence, and was null for mortality, resulting in a nonsignificant association with lethal breast cancer. BBD was positively associated with lethal breast cancer [RR = 1.27, 95% CI = (1.04–1.55)] due to a strong positive association with incidence. Interestingly, family history was significantly positively associated with incidence, but significantly inversely associated with mortality resulting in a null association with lethal breast cancer [RR = 1.09, 95% CI = (0.84–1.40)]. Current use of all types of HT was positively associated with incidence, especially for E&P. However, E&P use was strongly inversely associated with mortality [current E&P use vs. no current use: HR = 0.63, 95% CI = (0.47–0.83)], while current use of other types of HT were not associated with mortality. Overall, only use of E-alone was positively associated with lethal breast cancer [current use of E-alone vs. no current use: RR = 1.24, 95% CI = (1.00–1.55)].

Table 3.

Multivariate models of breast cancer incidencea, breast cancer mortalityb, and lethal breast cancer among disease-free womenc, NHS, July 1, 1990 to December 31, 2014.

VariableCategoryHR_incd (95% CI)HR_mortd (95% CI)P_heteRR_lethalc,f (95% CI)P
Age (yrs) ≥45, <55 1.0 (ref) 
 ≥55, <60 1.30 (1.19–1.42) 1.11 (0.85–1.47) 0.30 1.42 (1.09–1.87) 0.011 
 ≥60, <65 1.60 (1.45–1.78) 1.56 (1.16–2.11) 0.87 2.40 (1.78–3.24) <0.001 
 ≥65 1.78 (1.56–2.03) 1.65 (1.13–2.40) 0.71 2.78 (1.92–4.04) <0.001 
Duration of menopause Per 5 years 0.84 (0.81–0.87) 0.94 (0.85–1.05) 0.042 0.80 (0.72–0.88) <0.001 
Pregnancy history Nulliparous 1.35 (1.20–1.53) 1.21 (0.88–1.67) 0.53 1.60 (1.16–2.22) 0.004 
 AAFB < 25 1.0 (ref) 
 AAFB ≥ 25, <30 1.13 (1.06–1.21) 0.98 (0.81–1.20) 0.17 1.11 (0.92–1.35) 0.28 
 AAFB >= 30 1.21 (1.10–1.34) 1.00 (0.76–1.33) 0.22 1.21 (0.91–1.61) 0.19 
Benign breast disease (BBD) No 1.0 (ref) 
 Yes 1.28 (1.20–1.37) 1.00 (0.82–1.22) 0.020 1.27 (1.04–1.55) 0.017 
First degree family history of breast cancer No 1.0 (ref) 
 Yes 1.43 (1.31–1.55) 0.76 (0.59–0.97) <0.001 1.09 (0.84–1.40) 0.52 
Current HT use No 1.0 (ref) 
 Estrogen only 1.25 (1.15–1.35) 1.00 (0.81–1.25) 0.067 1.24 (1.00–1.55) 0.049 
 Estrogen & progestin 1.54 (1.41–1.68) 0.63 (0.47–0.83) <0.001 0.98 (0.74–1.30) 0.90 
 Other 1.23 (1.09–1.37) 0.86 (0.61–1.20) 0.049 1.05 (0.75–1.47) 0.77 
Height (cm) per 15 cm. 1.07 (1.00–1.15) 0.88 (0.71–1.09) 0.078 0.95 (0.76–1.17) 0.62 
BMI at age 18 (kg/m2≥15, <18.5 0.99 (0.90–1.08) 0.95 (0.74–1.22) 0.77 0.94 (0.73–1.20) 0.61 
 ≥18.5 <23 1.0 (ref) 
 ≥23, <25 0.94 (0.85–1.03) 0.75 (0.56–1.00) 0.15 0.72 (0.53–0.96) 0.025 
 ≥25 0.83 (0.74–0.93) 0.84 (0.60–1.19) 0.94 0.71 (0.50–1.00) 0.047 
Weight change since age 18 (kg) <−5 0.70 (0.57–0.86) 1.55 (0.91–2.62) 0.006 1.06 (0.63–1.81) 0.82 
 ≥−5, ≤5 1.0 (ref) 
 >5, ≤10 1.21 (1.10–1.33) 1.06 (0.81–1.40) 0.37 1.27 (0.97–1.67) 0.083 
 >10, ≤20 1.29 (1.18–1.40) 1.06 (0.82–1.36) 0.15 1.35 (1.05–1.73) 0.018 
 >20, ≤30 1.55 (1.40–1.71) 1.40 (1.06–1.84) 0.49 2.09 (1.58–2.75) <0.001 
 >30 1.47 (1.30–1.67) 1.36 (0.97–1.92) 0.68 1.94 (1.38–2.74) <0.001 
Alcohol intake (g/day) 1.0 (ref) 
 >0, <11 1.09 (1.02–1.16) 0.98 (0.81–1.18) 0.28 1.06 (0.88–1.28) 0.53 
 ≥11, <22 1.10 (0.99–1.22) 0.88 (0.65–1.20) 0.18 0.98 (0.72–1.32) 0.87 
 ≥22 1.22 (1.08–1.39) 0.97 (0.68–1.38) 0.21 1.18 (0.83–1.68) 0.37 
Cigarette smoking Never 1.0 (ref) 
 Past 1.01 (0.94–1.07) 0.97 (0.80–1.17) 0.71 0.97 (0.81–1.18) 0.79 
 Current, 1–14 cigarettes/day 0.96 (0.84–1.10) 1.01 (0.68–1.52) 0.80 0.97 (0.65–1.45) 0.88 
 Current, ≥15 cigarettes/day 1.02 (0.92–1.14) 1.43 (1.08–1.89) 0.029 1.42 (1.07–1.89) 0.015 
VariableCategoryHR_incd (95% CI)HR_mortd (95% CI)P_heteRR_lethalc,f (95% CI)P
Age (yrs) ≥45, <55 1.0 (ref) 
 ≥55, <60 1.30 (1.19–1.42) 1.11 (0.85–1.47) 0.30 1.42 (1.09–1.87) 0.011 
 ≥60, <65 1.60 (1.45–1.78) 1.56 (1.16–2.11) 0.87 2.40 (1.78–3.24) <0.001 
 ≥65 1.78 (1.56–2.03) 1.65 (1.13–2.40) 0.71 2.78 (1.92–4.04) <0.001 
Duration of menopause Per 5 years 0.84 (0.81–0.87) 0.94 (0.85–1.05) 0.042 0.80 (0.72–0.88) <0.001 
Pregnancy history Nulliparous 1.35 (1.20–1.53) 1.21 (0.88–1.67) 0.53 1.60 (1.16–2.22) 0.004 
 AAFB < 25 1.0 (ref) 
 AAFB ≥ 25, <30 1.13 (1.06–1.21) 0.98 (0.81–1.20) 0.17 1.11 (0.92–1.35) 0.28 
 AAFB >= 30 1.21 (1.10–1.34) 1.00 (0.76–1.33) 0.22 1.21 (0.91–1.61) 0.19 
Benign breast disease (BBD) No 1.0 (ref) 
 Yes 1.28 (1.20–1.37) 1.00 (0.82–1.22) 0.020 1.27 (1.04–1.55) 0.017 
First degree family history of breast cancer No 1.0 (ref) 
 Yes 1.43 (1.31–1.55) 0.76 (0.59–0.97) <0.001 1.09 (0.84–1.40) 0.52 
Current HT use No 1.0 (ref) 
 Estrogen only 1.25 (1.15–1.35) 1.00 (0.81–1.25) 0.067 1.24 (1.00–1.55) 0.049 
 Estrogen & progestin 1.54 (1.41–1.68) 0.63 (0.47–0.83) <0.001 0.98 (0.74–1.30) 0.90 
 Other 1.23 (1.09–1.37) 0.86 (0.61–1.20) 0.049 1.05 (0.75–1.47) 0.77 
Height (cm) per 15 cm. 1.07 (1.00–1.15) 0.88 (0.71–1.09) 0.078 0.95 (0.76–1.17) 0.62 
BMI at age 18 (kg/m2≥15, <18.5 0.99 (0.90–1.08) 0.95 (0.74–1.22) 0.77 0.94 (0.73–1.20) 0.61 
 ≥18.5 <23 1.0 (ref) 
 ≥23, <25 0.94 (0.85–1.03) 0.75 (0.56–1.00) 0.15 0.72 (0.53–0.96) 0.025 
 ≥25 0.83 (0.74–0.93) 0.84 (0.60–1.19) 0.94 0.71 (0.50–1.00) 0.047 
Weight change since age 18 (kg) <−5 0.70 (0.57–0.86) 1.55 (0.91–2.62) 0.006 1.06 (0.63–1.81) 0.82 
 ≥−5, ≤5 1.0 (ref) 
 >5, ≤10 1.21 (1.10–1.33) 1.06 (0.81–1.40) 0.37 1.27 (0.97–1.67) 0.083 
 >10, ≤20 1.29 (1.18–1.40) 1.06 (0.82–1.36) 0.15 1.35 (1.05–1.73) 0.018 
 >20, ≤30 1.55 (1.40–1.71) 1.40 (1.06–1.84) 0.49 2.09 (1.58–2.75) <0.001 
 >30 1.47 (1.30–1.67) 1.36 (0.97–1.92) 0.68 1.94 (1.38–2.74) <0.001 
Alcohol intake (g/day) 1.0 (ref) 
 >0, <11 1.09 (1.02–1.16) 0.98 (0.81–1.18) 0.28 1.06 (0.88–1.28) 0.53 
 ≥11, <22 1.10 (0.99–1.22) 0.88 (0.65–1.20) 0.18 0.98 (0.72–1.32) 0.87 
 ≥22 1.22 (1.08–1.39) 0.97 (0.68–1.38) 0.21 1.18 (0.83–1.68) 0.37 
Cigarette smoking Never 1.0 (ref) 
 Past 1.01 (0.94–1.07) 0.97 (0.80–1.17) 0.71 0.97 (0.81–1.18) 0.79 
 Current, 1–14 cigarettes/day 0.96 (0.84–1.10) 1.01 (0.68–1.52) 0.80 0.97 (0.65–1.45) 0.88 
 Current, ≥15 cigarettes/day 1.02 (0.92–1.14) 1.43 (1.08–1.89) 0.029 1.42 (1.07–1.89) 0.015 

a4,391 cases among 56,346 women; Competing risks: non-breast cancer death, 18,540; other cancers (other than non-melanoma skin cancer), 2,959.

b549 breast cancer deaths among 4,391 breast cancer cases; Competing risks: non-breast cancer death, 1,019.

c549 breast cancer deaths among 56,346 women.

dSubdistribution HR adjusted for competing risks.

eP for heterogeneity between associations of specific risk factors for breast cancer incidence versus breast cancer–specific mortality among cases.

fw1 = 0.972, w2 = 0.934.

Regarding anthropometric variables, height was not associated with lethal breast cancer. However, early life BMI (at age 18) was inversely associated with lethal breast cancer [BMI at age 18 ≥ 25 vs. 18.5–<23: RR = 0.71, 95% CI = (0.50–1.00)] due to inverse associations with both incidence and mortality. In contrast, excessive weight gain since age 18 was strongly and monotonically positively associated with lethal breast cancer (>30 kg vs. ± 5 kg: RR = 1.94 (95% CI = 1.38–2.74)] due to positive associations with both incidence and mortality. However, weight loss of >5 kg was not associated with lethal breast cancer because the inverse association with incidence was countered by the positive association with mortality.

Alcohol intake showed weak positive associations with incidence, null associations with mortality and was not associated with lethal breast cancer, although the CI was wide. However, current smoking ≥ 15 cigarettes/day, while not a risk factor for incidence, was strongly associated with mortality, resulting in a positive association with lethal breast cancer [≥15 cigarettes/day vs. never smoking: RR = 1.42 (95% CI = 1.07–1.89)]. A summary of the incidence, mortality and lethal cancer analyses is provided in Fig. 1.

Figure 1.

For categorical variables with more than two categories. Plot of HR (95% CI) for breast cancer incidence (black), HR (95% CI) for mortality due to breast cancer among cases (blue), and RR (95% CI) for two-stage lethal breast cancer (red) versus reference category for age, AAFB, current HT use, BMI at age 18, weight change since age 18, alcohol intake, and smoking. For categorical variables with two categories or continuous exposures. Plot of HR (95% CI) for breast cancer incidence (black), HR (95% CI) for mortality due to breast cancer among cases (blue), and RR(95% CI) for two-stage lethal breast cancer (red) for duration of menopause (per 5 years), BBD (yes vs. no), first-degree family history of breast cancer (yes vs. no), and height (per 15 cm).

Figure 1.

For categorical variables with more than two categories. Plot of HR (95% CI) for breast cancer incidence (black), HR (95% CI) for mortality due to breast cancer among cases (blue), and RR (95% CI) for two-stage lethal breast cancer (red) versus reference category for age, AAFB, current HT use, BMI at age 18, weight change since age 18, alcohol intake, and smoking. For categorical variables with two categories or continuous exposures. Plot of HR (95% CI) for breast cancer incidence (black), HR (95% CI) for mortality due to breast cancer among cases (blue), and RR(95% CI) for two-stage lethal breast cancer (red) for duration of menopause (per 5 years), BBD (yes vs. no), first-degree family history of breast cancer (yes vs. no), and height (per 15 cm).

Close modal

We also examined heterogeneity of associations of specific risk factors based on Wald tests comparing beta coefficients between incidence and mortality (Table 3, P_het). There were significant differences for duration of menopause, BBD, family history, current HT use, weight loss, and heavy current smoking, justifying the use of a multi-state model to analyze the data.

Finally, because the impact of risk factors with the magnitude of risk for lethal breast cancer is not always clear from RR estimates, in Table 4, we also provide estimates of cumulative incidence over 25 years for each of lethal breast cancer, nonlethal breast cancer and total breast cancer, for selected categories of long-term weight change, cigarette smoking, and pregnancy history, assuming mean levels for the other risk factors in Table 3. The overall 25-year cumulative incidence of lethal, nonlethal, and total breast cancer was 0.0081, 0.0661, and 0.0742, respectively. Both weight change since age 18 and pregnancy history were associated with all three endpoints, but the magnitude of association with lethal breast cancer was higher. However, for cigarette smoking, there was a strong association with lethal breast cancer, but no association with either nonlethal or total breast cancer. Also, the relative risk estimates in Table 3 show excellent agreement with the relative risk estimates for cumulative incidence in Table 4, indicating that the Taylor series approximation in Eq. (C) works very well.

Table 4.

Cumulative incidence of lethal, non-lethal and total breast cancer over 25 years based on selected risk factors.a

Risk factorCategoryCIF lethalbRRCIF nonlethaldRRCIF totalcRR
Overall  0.0081 — 0.0661 — 0.0742 — 
Weight change since age 18 ±5 kg 0.0060 1.0 (ref) 0.0560 1.0 (ref) 0.0620 1.0 (ref) 
 >30 kg 0.0117 1.95 0.0780 1.39 0.0897 1.45 
Pregnancy history AAFB < 25 0.0074 1.0 (ref) 0.0611 1.0 (ref) 0.0685 1.0 (ref) 
 Nulliparous 0.0119 1.61 0.0796 1.30 0.0915 1.33 
Cigarette smoking Never smoker 0.0079 1.0 (ref) 0.0661 1.0 (ref) 0.0740 1.0 (ref) 
 Current smoker ≥15 cigarettes/day 0.0112 1.42 0.0643 0.97 0.0755 1.02 
Risk factorCategoryCIF lethalbRRCIF nonlethaldRRCIF totalcRR
Overall  0.0081 — 0.0661 — 0.0742 — 
Weight change since age 18 ±5 kg 0.0060 1.0 (ref) 0.0560 1.0 (ref) 0.0620 1.0 (ref) 
 >30 kg 0.0117 1.95 0.0780 1.39 0.0897 1.45 
Pregnancy history AAFB < 25 0.0074 1.0 (ref) 0.0611 1.0 (ref) 0.0685 1.0 (ref) 
 Nulliparous 0.0119 1.61 0.0796 1.30 0.0915 1.33 
Cigarette smoking Never smoker 0.0079 1.0 (ref) 0.0661 1.0 (ref) 0.0740 1.0 (ref) 
 Current smoker ≥15 cigarettes/day 0.0112 1.42 0.0643 0.97 0.0755 1.02 

aAssuming mean levels of other risk factors.

bCumulative incidence of lethal breast cancer (Eq. (B)) = CIFlethal.

cCumulative incidence of total breast cancer using the Fine–Gray approach = CIFtotal.

dCumulative incidence of non-lethal breast cancer = CIFtotal minus CIFlethal.

Sensitivity analyses

On the basis of Eq. (B), it is clear that the CIF is a function of time. Thus, the relative risk estimates of specific risk factors may depend on time as well. Therefore, to assess the sensitivity of relative risk estimates with follow-up time (t), we refit the model in Eq. (C) using a 15-year instead of a 25-year follow-up time. The results are shown in Table 5. Overall, there were 3,024 breast cancer cases from July 1, 1990 to December 31, 2004 and 262 breast cancer deaths over the same period. In general, associations were similar for 15- and 25-year cumulative incidence. Some RR estimates for lethal breast cancer were higher for the 15-year period than the 25-year period (e.g., nulliparity vs. AAFB ≤ 25, 15-year period: RR = 2.11, 95% CI = 1.33–3.34; 25-year period: RR = 1.60, 95% CI = 1.16–2.22), while others were lower (e.g., weight change since age 18 > 30 kg vs. no change, 15-year period: RR = 1.78, 95% CI = 1.08–2.93; 25-year period: RR = 1.94, 95% CI = 1.38–2.74).

Table 5.

Multivariate models of breast cancer incidencea, breast cancer mortalityb, and lethal breast cancer among disease-free womenc, NHS, July 1, 1990 to December 31, 2004.

VariableCategoryHR_incd (95% CI)HR_mortd (95% CI)P_heteRR_lethalc,f (95% CI)P
Age (yrs) ≥45, <55 1.0 (ref) 
 ≥55, <60 1.46 (1.31–1.63) 0.72 (0.48–1.08) <0.001 1.06 (0.71–1.59) 0.78 
 ≥60, <65 2.02 (1.78–2.29) 1.16 (0.75–1.89) 0.017 2.30 (1.49–3.55) <0.001 
 ≥65 2.34 (2.00–2.73) 0.93 (0.54–1.62) 0.002 2.16 (1.24–3.74) 0.006 
Duration of menopause Per 5 years 0.85 (0.81–0.89) 0.96 (0.83–1.12) 0.12 0.82 (0.70–0.95) 0.009 
Pregnancy history Nulliparous 1.41 (1.22–1.63) 1.54 (0.97–2.43) 0.72 2.11 (1.33–3.34) 0.002 
 AAFB <25 1.0 (ref) 
 AAFB ≥25, <30 1.17 (1.08–1.27) 1.31 (0.97–1.74) 0.45 1.51 (1.14–2.00) 0.004 
 AAFB >=30 1.30 (1.16–1.47) 0.84 (0.53–1.35) 0.079 1.10 (0.69–1.76) 0.67 
Benign breast disease (BBD) No 1.0 (ref) 
 Yes 1.31 (1.21–1.42) 1.01 (0.76–1.34) 0.086 1.32 (0.99–1.75) 0.059 
First degree family history of breast cancer No 1.0 (ref) 
 Yes 1.45 (1.31–1.60) 0.56 (0.37–0.85) <0.001 0.83 (0.55–1.24) 0.37 
Current HT use No 1.0 (ref) 
 Estrogen only 1.30 (1.18–1.43) 0.89 (0.64–1.23) 0.027 1.15 (0.83–1.59) 0.39 
 Estrogen & progestin 1.72 (1.55–1.90) 0.63 (0.42–0.94) <0.001 1.09 (0.73–1.62) 0.68 
 Other 1.21 (1.05–1.39) 0.91 (0.56–1.48) 0.28 1.10 (0.68–1.78) 0.70 
Height (cm) per 15 cm. 1.12 (1.02–1.22) 0.91 (0.67–1.23) 0.19 1.01 (0.75–1.37) 0.93 
BMI at age 18 (kg/m2≥15, <18.5 0.98 (0.88–1.09) 0.83 (0.57–1.22) 0.43 0.82 (0.56–1.20) 0.30 
 ≥18.5 <23 1.0 (ref) 
 ≥23, <25 0.94 (0.84–1.06) 0.75 (0.48–1.16) 0.33 0.71 (0.46–1.10) 0.13 
 ≥25 0.86 (0.74–0.98) 1.00 (0.63–1.58) 0.53 0.85 (0.54–1.36) 0.51 
Weight change since age 18 (kg) <−5 0.72 (0.57–0.90) 1.93 (1.00–3.71) 0.005 1.34 (0.69–2.61) 0.38 
 ≥−5, ≤5 1.0 (ref) 
 > 5, ≤10 1.18 (1.06–1.32) 1.13 (0.76–1.67) 0.81 1.32 (0.89–1.96) 0.17 
 >10, ≤20 1.17 (1.06–1.30) 1.14 (0.80–1.64) 0.89 1.33 (0.93–1.91) 0.12 
 >20, ≤30 1.45 (1.29–1.63) 1.54 (1.04–2.29) 0.78 2.18 (1.47–3.23) <0.001 
 >30 1.42 (1.22–1.65) 1.28 (0.78–2.11) 0.70 1.78 (1.08–2.93) 0.023 
Alcohol intake (g/day) 1.0 (ref) 
 >0, <11 1.05 (0.97–1.13) 0.85 (0.64–1.13) 0.17 0.90 (0.68–1.18) 0.44 
 ≥11, <22 1.08 (0.95–1.22) 0.88 (0.57–1.37) 0.38 0.95 (0.62–1.48) 0.83 
 ≥22 1.18 (1.02–1.37) 0.93 (0.57–1.53) 0.38 1.10 (0.67–1.81) 0.72 
Cigarette smoking Never 1.0 (ref) 
 Past 1.08 (1.00–1.17) 0.88 (0.66–1.17) 0.17 0.95 (0.71–1.26) 0.72 
 Current, 1–14 cigarettes/day 1.00 (0.84–1.17) 0.89 (0.48–1.64) 0.73 0.89 (0.49–1.63) 0.71 
 Current, ≥15 cigarettes/day 1.16 (1.02–1.31) 1.73 (1.19–2.52) 0.045 1.95 (1.34–2.85) <0.001 
VariableCategoryHR_incd (95% CI)HR_mortd (95% CI)P_heteRR_lethalc,f (95% CI)P
Age (yrs) ≥45, <55 1.0 (ref) 
 ≥55, <60 1.46 (1.31–1.63) 0.72 (0.48–1.08) <0.001 1.06 (0.71–1.59) 0.78 
 ≥60, <65 2.02 (1.78–2.29) 1.16 (0.75–1.89) 0.017 2.30 (1.49–3.55) <0.001 
 ≥65 2.34 (2.00–2.73) 0.93 (0.54–1.62) 0.002 2.16 (1.24–3.74) 0.006 
Duration of menopause Per 5 years 0.85 (0.81–0.89) 0.96 (0.83–1.12) 0.12 0.82 (0.70–0.95) 0.009 
Pregnancy history Nulliparous 1.41 (1.22–1.63) 1.54 (0.97–2.43) 0.72 2.11 (1.33–3.34) 0.002 
 AAFB <25 1.0 (ref) 
 AAFB ≥25, <30 1.17 (1.08–1.27) 1.31 (0.97–1.74) 0.45 1.51 (1.14–2.00) 0.004 
 AAFB >=30 1.30 (1.16–1.47) 0.84 (0.53–1.35) 0.079 1.10 (0.69–1.76) 0.67 
Benign breast disease (BBD) No 1.0 (ref) 
 Yes 1.31 (1.21–1.42) 1.01 (0.76–1.34) 0.086 1.32 (0.99–1.75) 0.059 
First degree family history of breast cancer No 1.0 (ref) 
 Yes 1.45 (1.31–1.60) 0.56 (0.37–0.85) <0.001 0.83 (0.55–1.24) 0.37 
Current HT use No 1.0 (ref) 
 Estrogen only 1.30 (1.18–1.43) 0.89 (0.64–1.23) 0.027 1.15 (0.83–1.59) 0.39 
 Estrogen & progestin 1.72 (1.55–1.90) 0.63 (0.42–0.94) <0.001 1.09 (0.73–1.62) 0.68 
 Other 1.21 (1.05–1.39) 0.91 (0.56–1.48) 0.28 1.10 (0.68–1.78) 0.70 
Height (cm) per 15 cm. 1.12 (1.02–1.22) 0.91 (0.67–1.23) 0.19 1.01 (0.75–1.37) 0.93 
BMI at age 18 (kg/m2≥15, <18.5 0.98 (0.88–1.09) 0.83 (0.57–1.22) 0.43 0.82 (0.56–1.20) 0.30 
 ≥18.5 <23 1.0 (ref) 
 ≥23, <25 0.94 (0.84–1.06) 0.75 (0.48–1.16) 0.33 0.71 (0.46–1.10) 0.13 
 ≥25 0.86 (0.74–0.98) 1.00 (0.63–1.58) 0.53 0.85 (0.54–1.36) 0.51 
Weight change since age 18 (kg) <−5 0.72 (0.57–0.90) 1.93 (1.00–3.71) 0.005 1.34 (0.69–2.61) 0.38 
 ≥−5, ≤5 1.0 (ref) 
 > 5, ≤10 1.18 (1.06–1.32) 1.13 (0.76–1.67) 0.81 1.32 (0.89–1.96) 0.17 
 >10, ≤20 1.17 (1.06–1.30) 1.14 (0.80–1.64) 0.89 1.33 (0.93–1.91) 0.12 
 >20, ≤30 1.45 (1.29–1.63) 1.54 (1.04–2.29) 0.78 2.18 (1.47–3.23) <0.001 
 >30 1.42 (1.22–1.65) 1.28 (0.78–2.11) 0.70 1.78 (1.08–2.93) 0.023 
Alcohol intake (g/day) 1.0 (ref) 
 >0, <11 1.05 (0.97–1.13) 0.85 (0.64–1.13) 0.17 0.90 (0.68–1.18) 0.44 
 ≥11, <22 1.08 (0.95–1.22) 0.88 (0.57–1.37) 0.38 0.95 (0.62–1.48) 0.83 
 ≥22 1.18 (1.02–1.37) 0.93 (0.57–1.53) 0.38 1.10 (0.67–1.81) 0.72 
Cigarette smoking Never 1.0 (ref) 
 Past 1.08 (1.00–1.17) 0.88 (0.66–1.17) 0.17 0.95 (0.71–1.26) 0.72 
 Current, 1–14 cigarettes/day 1.00 (0.84–1.17) 0.89 (0.48–1.64) 0.73 0.89 (0.49–1.63) 0.71 
 Current, ≥15 cigarettes/day 1.16 (1.02–1.31) 1.73 (1.19–2.52) 0.045 1.95 (1.34–2.85) <0.001 

a3,024 cases among 56,346 women; Competing risks: non-breast cancer death, 6,436; other cancers (other than nonmelanoma skin cancer), 2,694.

b262 breast cancer deaths among 3,024 breast cancer cases; Competing risks: nonbreast cancer death, 183.

c262 breast cancer deaths among 56,346 women.

dSubdistribution HR adjusted for competing risks.

eP for heterogeneity between associations of specific risk factors for breast cancer incidence versus breast cancer–specific mortality among cases.

fw1 = 0.985, w2 = 0.954.

One assumption of the variance estimates of ln[ | $\widehat {RR}( t )]\ $ |for individual risk factors in Eq. (D), is that the weighting factors | ${w}_1$ | and | ${w}_2$ | are approximate constants with relatively little variability compared with | $Var( {{{\hat{\beta }}}_1} ){\rm{\ or}}\ Var( {{{\hat{\beta }}}_2} ).\ $ |To assess the validity of this assumption, we performed a bootstrap analysis. We used PROC SURVEYSELECT of SAS 9.4 to select 100 bootstrap samples with replacement, each of size 56,346, and repeated the 25-year analyses in Table 3 for each of the samples. We then compared (i) the mean estimates of regression coefficients from the bootstrap samples versus the original estimates in Table 3 and (ii) the empirical variance from the bootstrap samples with the large sample variance estimates from the original data in Table 3. The results are given in Supplementary Table S1. We see that overall there is little bias in estimation of ln[RR(t)] for individual regression coefficients [mean = −0.005, 95% CI = (−0.012 to 0.002)]. In addition, there is little bias in variance estimates of ln[RR(t)] parameters (mean | ${\rm{ln}}\{ Var[{\rm{ln}}({\widehat {RR( t )]}}_{bootstrap}\} $ |⁠, minus | ${\rm{ln}}{\{ {Var\{ {\ln [ {\widehat {RR}( t )} ]} \}} \}}_{large\ sample\ }$ | = 0.012 (95% CI = −0.055 to 0.079). Thus, the large sample variance estimates are valid at least for the sample sizes used in this study. Finally, the estimates of | ${w}_1$ | ( | $mean \pm sd = 0.968 \pm 0.001$ |⁠), and | ${w}_2$ | ( | $mean \pm sd = 0.935 \pm 0.003$ |⁠) had little variability over the bootstrap samples, were similar to the estimates in the original sample ( | ${w}_1 = 0.972,\ {w}_2 = 0.934)\ $ |and had low variability compared with | $Var( {{{\hat{\beta }}}_1} ){\rm{\ and}}\ Var( {{{\hat{\beta }}}_2} ).$ | Thus, the assumption of | ${w}_1$ |and | ${w}_2$ | as approximate constants in estimation of | $Var[{\rm{ln}}( {\widehat {RR}( t )} ]$ | is reasonable.

In this article, we presented a novel multi-state model for estimating the risk of developing cancer that will ultimately be lethal among a population of initially healthy individuals, and applied this method to breast cancer in the NHS. This is a special type of multi-state model where it is impossible to go from a disease-free state (state 1) to death due to a disease (state 3) without developing incident disease (state 2). We used time since 1990 as the time scale in the proportional hazards model for incidence, and time since diagnosis in the proportional hazards model for mortality, that is, the clock-reset approach (24), which is appropriate when mortality risk depends on time since diagnosis.

In our analyses, we did not include women who had a cancer other than non-melanoma skin cancer (referred to as other cancer, below) before 1990. In addition, we censored women if they developed an incident other cancer after 1990. We determined that there were 75 women who developed an incident other cancer after 1990, but before the diagnosis of breast cancer (38 colorectal, 17 melanoma, 9 lung, and 11 other types). This represents approximately 1.7% additional cases beyond the 4,391 incident breast cancers identified in Table 1. However, we considered the other cancer to be a competing risk because (i) the treatment for the other cancer might modify the association of 1990 risk factors with breast cancer and (ii) if a death occurs, it might be difficult to attribute the cause to a specific cancer, if multiple primaries are present.

There are some limitations of our model. First, the model is expressed in terms of risk factors that are defined at a fixed point in time. An extension of this model would allow for updating exposure both before and after diagnosis, particularly for a disease such as breast cancer where women can survive many decades. In addition, some postdiagnosis risk factors may have different associations shortly after diagnosis (e.g., within the first 5 years postdiagnosis), than long after diagnosis (e.g., after 5 years postdiagnosis). We deliberately did not include tumor characteristics (e.g., stage, grade) in our breast cancer mortality model because they are unknown prediagnosis, and may be intermediates between risk factor levels in 1990 and survival postdiagnosis. Second, although this model could be used to understand the risks of other potentially lethal diseases, it does not incorporate the situation where patients may die immediately after diagnosis (e.g., sudden cardiac death). Third, power is more limited for assessing risk factors for breast cancer mortality than for breast cancer incidence, thus reducing power for assessing heterogeneity of associations between incidence and mortality. Four, although the long follow-up of the NHS cohort is a strength, advances in treatment over 25 years have substantially impacted survival and may influence risk factor associations with lethal cancer.

Our multi-state model aligns well with the carcinogenic continuum in modern cancer biology. Many cancers arise from abnormally growing cells in the epithelial compartments and there are multiple steps going from (1) normal epithelial tissue (state 1) to dysplasia/invasive breast cancer (state 2) to metastatic lethal disease (state 3). People usually will not die of the original tumor, but rather due to metastasis (lethal disease). A mitogenic signal may be associated with both breast cancer onset and progression. For example, white adipose tissue promotes transformation of benign epithelial cells, leading to estrogen production in postmenopausal women (27), and inducing epithelial–mesenchymal transition for cancer metastasis (28).

Weight gain in adulthood is an established risk factor for breast cancer (29–32), and an emerging prognostic factor for survival (33). Our multi-state model captures the aggregated effects of adulthood weight gain on breast cancer risk and prognosis. Other risk factors may only be associated with either disease onset or prognosis. For example, family history is a strong risk factor for incident breast cancer (34–36), but its prognostic value remains mixed. In some studies, family history is associated with prolonged survival (37–39), possibly due to more frequent mammographic screening and early detection; while in others inverse (40) or no associations (41–43) were observed. In our model, family history was positively associated with incidence, but was also associated with prolonged survival, resulting in a null association with lethal breast cancer.

In general, despite being a well-known carcinogen, smoking is not a consistent risk factor for breast cancer incidence. However, previous studies suggest that smoking is associated with worse breast cancer survival, possible reflecting a more systemic effect of smoking and oxidative stress (20–22). In our multi-state model, heavy current cigarette smoking was not associated with incidence, but was associated with lethal cancer, suggesting the importance of smoking cessation. Because some risk factor associations may be mediated by surveillance, we fitted an additional survival model for breast cancer cases controlling for stage (Supplementary Table S2) and found some evidence of mediation for family history and weight change, but not for smoking or E&P use.

Most of the variables in our lethal cancer model are established risk factors for breast cancer incidence. Some of the risk factors are also useful predictors of lethal breast cancer (e.g., weight gain since age 18). However, there may be other risk factors that are predictors of lethal breast cancer (e.g., cigarette smoking). This is an interesting area for future research.

In summary, the multi-state model presented here is a novel approach to understand whether certain prediagnostic risk factors may predispose individuals toward the most aggressive forms of chronic disease. This model may be a useful tool in understanding the biology and epidemiology of chronic disease.

B. Rosner reports grants from NIH during the conduct of the study. R.J. Glynn reports grants from AstraZeneca, Kowa, Novartis, and Pfizer outside the submitted work. A. Heather Eliassen reports grants from NIH during the conduct of the study. R.M. Tamimi reports grants from NIH/NCI during the conduct of the study; personal fees from Sterigenics outside the submitted work. M.D. Holmes reports non-financial support from Bayer AG during the conduct of the study. W.C. Willett reports grants from NIH and Breast Cancer Research Foundation during the conduct of the study. S.S. Tworoger reports grants from NIH/NCI during the conduct of the study; grants from DOD, Florida Department of Health, NIH, and BMS outside the submitted work; and receives honoraria from AACR for teaching and senior editor of CEBP (paid to S.S. Tworoger), Ponce Health Sciences University for membership on external advisory committee (EAC) paid to S.S. Tworoger, Ovarian Cancer Research Alliance for membership on scientific advisory board paid to S.S. Tworoger, German Cancer Research Center for grand rounds speaking paid to S.S. Tworoger; teaching at the Harvard T.H. Chan School of Public Health paid to S.S. Tworoger; NIH Study Section renumeration paid to S.S. Tworoger and non-paid relationships: Member of external advisory committee of California Teachers Study (City of Hope) and The Tomorrow Project (Alberta Cancer Center). No disclosures were reported by the other authors.

B. Rosner: Conceptualization, data curation, formal analysis, funding acquisition, validation, investigation, methodology, writing–original draft, writing–review and editing. R.J. Glynn: Data curation, validation, investigation, methodology, writing–review and editing. A.H. Eliassen: Methodology, writing–review and editing. S.E. Hankinson: Writing–review and editing. R.M. Tamimi: Writing–review and editing. W.Y. Chen: Writing–review and editing. M.D. Holmes: Formal analysis, validation, investigation, methodology. Y. Mu: Software, formal analysis. C. Peng: Writing–review and editing. G.A. Colditz: Writing–review and editing. W.C. Willett: Conceptualization, project administration, writing–review and editing. S.S. Tworoger: Data curation, validation, investigation, methodology, writing–review and editing.

All authors received support from NCI grants UMI CA 186107 and P01 CA87969.

We would like to thank the following state cancer registries for their help: AL, AZ, AR, CA, CO, CT, DE, FL, GA, ID, IL, IN, IA, KY, LA, ME, MD, MA, MI, NE, NH, NJ, NY, NC, ND, OH, OK, OR, PA, RI, SC, TN, TX, VA, WA, and WY.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Richman
EL
,
Kenfield
SA
,
Stampfer
MJ
,
Giovannucci
EL
,
Chan
JM
.
Egg, red meat, and poultry intake and risk of lethal prostate cancer in the prostate-specific antigen-era: incidence and survival
.
Cancer Prev Res
2011
;
4
:
2110
21
.
2.
Wilson
KM
,
Kasperzyk
JL
,
Rider
JR
,
Kenfield
S
,
van Dam
RM
,
Stampfer
MJ
, et al
.
Coffee consumption and prostate cancer risk and progression in the health professionals follow-up study
.
J Natl Cancer Inst
2011
;
103
:
876
84
.
3.
Mucci
LA
,
Stampfer
MJ
.
Mounting evidence for prediagnostic use of statins in reducing risk of lethal prostate cancer
.
J Clin Oncol
2014
;
32
:
1
2
.
4.
Poole
EM
,
Merritt
MA
,
Jordan
SJ
,
Yang
HP
,
Hankinson
SE
,
Park
Y
, et al
.
Hormonal and reproductive risk factors for epithelial ovarian cancer by tumor aggressiveness
.
Cancer Epidemiol Biomarkers Prev
2013
;
22
:
429
37
.
5.
Fortner
RT
,
Poole
EM
,
Wentzensen
NA
,
Trabert
B
,
White
E
,
Arslan
AA
, et al
.
Ovarian cancer risk factors by tumor aggressiveness: an analysis from the ovarian cancer cohort consortium
.
Int J Cancer
2019
;
145
:
58
69
.
6.
Rosner
B
,
Tamimi
RM
,
Kraft
P
,
Gao
C
,
Mu
Y
,
Scott
C
, et al
.
Simplified breast risk tool integrating questionnaire risk factors, mammographic density, and polygenic risk score: development and validation
.
Cancer Epidemiol Biomarkers Prev
2021
;
30
:
600
07
.
7.
Rosner
BA
,
Colditz
GA
,
Hankinson
SE
,
Sullivan-Halley
J
,
Lacey
JV
Jr
,
Bernstein
L
.
Validation of rosner-colditz breast cancer incidence model using an independent data set, the california teachers study
.
Breast Cancer Res Treat
2013
;
142
:
187
202
.
8.
Rauh
C
,
Schuetz
F
,
Rack
B
,
Stickeler
E
,
Klar
M
,
Orlowska-Volk
M
, et al
.
Hormone therapy and its effect on the prognosis in breast cancer patients
.
Geburtshilfe Frauenheilkd
2015
;
75
:
588
96
.
9.
Cheek
J
,
Lacy
J
,
Toth-Fejel
S
,
Morris
K
,
Calhoun
K
,
Pommier
RF
.
The impact of hormone replacement therapy on the detection and stage of breast cancer
.
Arch Surg
2002
;
137
:
1015
9
.
10.
Carter
BD
,
Abnet
CC
,
Feskanich
D
,
Freedman
ND
,
Hartge
P
,
Lewis
CE
, et al
.
Smoking and mortality–beyond established causes
.
N Engl J Med
2015
;
372
:
631
40
.
11.
Runowicz
CD
,
Leach
CR
,
Henry
NL
,
Henry
KS
,
Mackey
HT
,
Cowens-Alvarado
RL
, et al
.
American Cancer Society/American Society of Clinical Oncology Breast Cancer survivorship care guideline
.
CA Cancer J Clin
2016
;
66
:
43
73
.
12.
Colditz
GA
,
Hankinson
SE
.
The Nurses' Health Study: lifestyle and health among women
.
Nat Rev Cancer
2005
;
5
:
388
96
.
13.
Rich-Edwards
JW
,
Corsano
KA
,
Stampfer
MJ
.
Test of the national death index and equifax nationwide death search
.
Am J Epidemiol
1994
;
140
:
1016
19
.
14.
Stampfer
MJ
,
Willett
WC
,
Speizer
FE
,
Dysert
DC
,
Lipnick
R
,
Rosner
B
, et al
.
Test of the National Death Index
.
Am J Epidemiol
1984
;
119
:
837
9
.
15.
Colditz
GA
.
Epidemiology of breast cancer. Findings from the Nurses' Health Study
.
Cancer
1993
;
71
:
1480
9
.
16.
Chen
WY
,
Manson
JE
,
Hankinson
SE
,
Rosner
B
,
Holmes
MD
,
Willett
WC
.,et al.
Unopposed estrogen therapy and the risk of invasive breast cancer
.
Arch Intern Med
2006
;
166
:
1027
32
.
17.
Colditz
GA
,
Kaphingst
KA
,
Hankinson
SE
,
Rosner
B
.
Family history and risk of breast cancer: Nurses' Health Study
.
Breast Cancer Res Treat
2012
;
133
:
1097
104
.
18.
Colditz
GA
,
Rosner
B
.
Cumulative risk of breast cancer to age 70 years according to risk factor status: data from the Nurses' Health Study
.
Am J Epidemiol
2000
;
152
:
950
64
.
19.
Rosner
B
,
Eliassen
AH
,
Toriola
AT
,
Chen
WY
,
Hankinson
SE
,
Willett
WC
, et al
.
Weight and weight changes in early adulthood and later breast cancer risk
.
Int J Cancer
2017
;
140
:
2003
14
.
20.
Warren
GW
,
Kasza
KA
,
Reid
ME
,
Cummings
KM
,
Marshall
JR
.
Smoking at diagnosis and survival in cancer patients
.
Int J Cancer
2013
;
132
:
401
10
.
21.
Braithwaite
D
,
Izano
M
,
Moore
DH
,
Kwan
ML
,
Tammemagi
MC
,
Hiatt
RA
, et al
.
Smoking and survival after breast cancer diagnosis: a prospective observational study and systematic review
.
Breast Cancer Res Treat
2012
;
136
:
521
33
.
22.
Holmes
MD
,
Murin
S
,
Chen
WY
,
Kroenke
CH
,
Spiegelman
D
,
Colditz
GA
.
Smoking and survival after breast cancer diagnosis
.
Int J Cancer
2007
;
120
:
2672
7
.
23.
SAS Institute, Inc
.
SAS/STAT 13.2 Users Guide
;
2014
.
Available from
: https://www.sas.com.
24.
Putter
H
,
Fiocco
M
,
Geskus
RB
.
Tutorial in biostatistics: competing risks and multi-state models
.
Stat Med
2007
;
26
:
2389
430
.
25.
Putter
H
,
van der Hage
J
,
de Bock
GH
,
Elgalta
R
,
van de Velde
CJ
.
Estimation and prediction in a multi-state model for breast cancer
.
Biom J
2006
;
48
:
366
80
.
26.
Fine
JP
,
Gray
RJ
.
A proportional hazards model for the subdistribution of a competing risk
.
J Am Stat Assoc
1999
;
94
:
496
509
.
27.
Nelson
LR
,
Bulun
SE
.
Estrogen production and action
.
J Am Acad Dermatol
2001
;
45
:
S116
24
.
28.
Lengyel
E
,
Makowski
L
,
DiGiovanni
J
,
Kolonin
MG
.
Cancer as a matter of fat: the crosstalk between adipose tissue and tumors
.
Trends Cancer
2018
;
4
:
374
84
.
29.
Eliassen
AH
,
Colditz
GA
,
Rosner
B
,
Willett
WC
,
Hankinson
SE
.
Adult weight change and risk of postmenopausal breast cancer
.
JAMA
2006
;
296
:
193
201
.
30.
Vrieling
A
,
Buck
K
,
Kaaks
R
,
Chang-Claude
J
.
Adult weight gain in relation to breast cancer risk by estrogen and progesterone receptor status: a meta-analysis
.
Breast Cancer Res Treat
2010
;
123
:
641
9
.
31.
Emaus
MJ
,
van Gils
CH
,
Bakker
MF
,
Bisschop
CN
,
Monninkhof
EM
,
Bueno-de-Mesquita
HB
, et al
.
Weight change in middle adulthood and breast cancer risk in the EPIC-PANACEA study
.
Int J Cancer
2014
;
135
:
2887
99
.
32.
Rosner
B
,
Eliassen
AH
,
Toriola
AT
,
Hankinson
SE
,
Willett
WC
,
Natarajan
L
, et al
.
Short-term weight gain and breast cancer risk by hormone receptor classification among pre- and postmenopausal women
.
Breast Cancer Res Treat
2015
;
150
:
643
53
.
33.
Playdon
MC
,
Bracken
MB
,
Sanft
TB
,
Ligibel
JA
,
Harrigan
M
,
Irwin
ML
.
Weight gain after breast cancer diagnosis and all-cause mortality: systematic review and meta-analysis
.
J Natl Cancer Inst
2015
;
107
:
djv275
.
34.
Pharoah
PD
,
Day
NE
,
Duffy
S
,
Easton
DF
,
Ponder
BA
.
Family history and the risk of breast cancer: a systematic review and meta-analysis
.
Int J Cancer
1997
;
71
:
800
9
.
35.
Brewer
HR
,
Jones
ME
,
Schoemaker
MJ
,
Ashworth
A
,
Swerdlow
AJ
.
Family history and risk of breast cancer: an analysis accounting for family structure
.
Breast Cancer Res Treat
2017
;
165
:
193
200
.
36.
Collaborative Group on Hormonal Factors in Breast Cancer
.
Familial breast cancer: collaborative reanalysis of individual data from 52 epidemiological studies including 58,209 women with breast cancer and 101,986 women without the disease
.
Lancet
2001
;
358
:
1389
99
.
37.
Malone
KE
,
Daling
JR
,
Weiss
NS
,
McKnight
B
,
White
E
,
Voigt
LF
.
Family history and survival of young women with invasive breast carcinoma
.
Cancer
1996
;
78
:
1417
25
.
38.
Mohammed
SN
,
Smith
P
,
Hodgson
SV
,
Fentiman
IS
,
Miles
DW
,
Barnes
DM
, et al
.
Family history and survival in premenopausal breast cancer
.
Br J Cancer
1998
;
77
:
2252
6
.
39.
Gajalakshmi
CK
,
Shanta
V
,
Hakama
M
.
Survival from contralateral breast cancer
.
Breast Cancer Res Treat
1999
;
58
:
115
22
.
40.
Slattery
ML
,
Berry
TD
,
Kerber
RA
.
Is survival among women diagnosed with breast cancer influenced by family history of breast cancer?
Epidemiology
1993
;
4
:
543
8
.
41.
Anderson
DE
,
Badzioch
MD
.
Survival in familial breast cancer patients
.
Cancer
1986
;
58
:
360
5
.
42.
Israeli
D
,
Tartter
PI
,
Brower
ST
,
Mizrachy
B
,
Bratton
J
.
The significance of family history for patients with carcinoma of the breast
.
J Am Coll Surg
1994
;
179
:
29
32
.
43.
Schouten
LJ
,
Hupperets
PS
,
Jager
JJ
,
Volovics
L
,
Wils
JA
,
Verbeek
AL
, et al
.
Prognostic significance of etiological risk factors in early breast cancer
.
Breast Cancer Res Treat
1997
;
43
:
217
23
.

Supplementary data