© International Epidemiological Association 2000
Printed in Great Britain
International Journal of Epidemiology 2000;29:214–223
Variation in the interaction between familial and reproductive factors on the risk of breast cancer according to age, menopausal status, and degree of familiality N Andrieu,a T Prevost,b TE Rohan,c E Luporsi,d MG Lê,a M Gerber,e DG Zaridze,f Y Lifanova,b R Renaud,g HP Leeh and SW Duffyb
Background Studies have found that reproductive factors might have a variable effect on the occurrence of breast cancer (BC) according to the existence or not of a family history of BC. The effect of a family history of BC on the risk of BC may also vary according to the age at diagnosis and the degree of kinship. This may confound the relation between familial risk and reproductive factors. A combined analysis was performed to study the interaction between familial risk and reproductive factors according to degree of familiality, age at interview and menopausal status. Methods
The present analysis included 2948 cases and 4170 controls in seven case-control studies from four countries. The combined relative risks were estimated using a Bayesian random-effects logistic regression model.
Results
The main effects of reproductive life factors on the risk of BC are in agreement with previous studies. Two-way interactions between subject’s age or menopausal status and a family history of BC were not significant. Although the threeway interaction between age, familial risk and parity was not significant, familial risk seemed to be increased slightly for women with high parity compared with women with low parity in the older age group, and seemed to be slightly decreased for women with high parity compared with women with low parity in younger women. The subject’s age also appeared to have an effect on the interaction between familial risk and the age at first childbirth (P = 0.1).
Conclusions A possible influence of reproductive and menstrual factors on familial risk of BC has been suggested previously and was also evident in the present study. Threeway interactions between age, family history and parity or age at first childbirth might exist and they merit further investigation. Keywords
Breast cancer, familial risk, reproductive factors, interactions
Accepted
15 October 1999
a Unité INSERM 521, Institut Gustave Roussy, 39 rue Camille Desmoulins,
94805 Villejuif Cedex, France. b MRC Biostatistics Unit, Institute of Public Health, University Forvie Site,
Robinson Way, Cambridge CB2 2SR, UK. c Cancer Epidemiology Unit, University of Toronto, 12 Queen’s Park Crescent
W, 3rd Floor, McMurrich Bldg, Toronto, Ontario, M5S 1A8, Canada. d Centre Alexis Vautrin, 54511 Vandoeuvre les Nancy Cedex, France. e Groupe d’Epidémiologie Métabolique, INSERM-CRLC, Centre de Recherche
en Cancérologie, Rue des Apothicaires, Parc Euromédecine, 34094 Montpellier Cedex 5, France. f Department of Epidemiology, Cancer Research Center, Russian Academy of
The association of a family history of breast cancer (BC) with an increased risk of BC has been well documented. However, genetic factors do not explain all of the variation in BC rates and several reproductive factors are now well-established risk factors. These include, for example, an early age at menarche, a late age at menopause, a late age at first childbirth, and nulliparity.1 There is also considerable interest in gene-environment interactions. Studies have found that some reproductive factors might have a variable effect on the occurrence of BC according to the existence or not of a family history of BC.2–7 Also, there
Sciences, 24 Kashirskoje Shosse, 115478 Moscow, Russia. g Département de Gynécologie Obstétricale, Hospices civils, 67000 Strasbourg,
France. h Department of Community, Occupational and Family Medicine, National
University of Singapore, Singapore.
Reprint requests to: Dr Nadine Andrieu, Unité INSERM 521, Institut Gustave Roussy, 39 rue Camille Desmoulins, 94805 Villejuif Cedex, France. E-mail:
[email protected]
214
BREAST CANCER: REPRODUCTIVE AND FAMILIAL FACTORS, AN INTERACTION STUDY
may be a variable effect according to age or menopausal status.8 Moreover, the effect of a family history of BC on the risk of BC may vary according to the age at diagnosis and the number of affected relatives and degree of kinship.1 In previous work, we investigated two-way interactions between familial and reproductive factors by combining seven case-control studies from various countries. Our findings suggested synergism between familial factors and history of abortions,9 and a slightly increasing familial risk with increasing number of children, age at first childbirth and duration of breast cell mitotic activity (BCMA).10 However, a possible difference in the familial effect on BC risk according to age, menopause, and degree of familiality might have confounded the estimates and led to inaccurate inferences. The effect of those potential confounders was not presented in our previous paper10 due to space limitations. To date, interactions between the different possible causes of heterogeneity in BC risks have not been studied. Therefore, we performed a combined analysis to study the interaction between familial risk and reproductive factors according to type of relationship with affected relatives, age at interview and menopausal status. Despite the possible lack of power for such an analysis, the analysis had the potential to generate hypotheses on the interrelation between familial risk, reproductive factors, and potential confounders.
Materials and Methods The analysis included case-control studies from four countries: France, Australia, Russia and Singapore. The data sets were chosen because they had information on family history of BC as well as reproductive factors of interest (such as age at menarche; number of children, number and type of abortions, age at first childbirth, menopausal status, and age at menopause). For all studies, data were collected by an in-person interview and family history of BC was recalled by the subjects and was not verified from medical records. The present analysis included 2948 cases and 4170 controls. In this analysis, no family history includes unknown family history. Most studies in the combined analysis have been published and the main design features have also been described elsewhere.10 Nevertheless, the studies are briefly described in Table 1, and the designs are outlined below. In a case-control study from South Australia,11 the cases were obtained from the population-based South Australian Central Cancer Registry between 1982 and 1984. Cases were 20–74
215
years, with a histologically verified first diagnosis of BC. For each case, one control was selected at random from the electoral roll from among women of approximately the same age as that of the case at diagnosis. Study subjects were interviewed in their homes by trained interviewers. Information on family history of cancer in sisters, mother and grandmothers was recorded. For the present study, information about first-degree relatives only was provided. In a case-control study from Singapore,12 BC patients were consecutive admissions to Singapore General Hospital and the National University Hospital between 1986 and 1988. Approximately twice as many hospital controls as cases were selected within 5-year age groups. Subjects were 24–88 years. They were interviewed in hospital by experienced investigators. History of BC in the subject’s mother, sisters, or maternal aunts was recorded. Data were obtained from a case-control study performed in Moscow (Russia) (Lifanova et al., unpublished) where subjects were interviewed from 1992 to 1994. Cases were aged 23–82 years with histologically confirmed primary carcinoma of the breast and were recruited from four Moscow hospitals. Controls were women with minor non-chronic complaints registered in primary care polyclinics in Moscow. Information was recorded on the occurrence of BC in the family (sisters, mother, aunts and grandmothers). In a case-control study carried out in Montpellier13 (France), subjects were interviewed between 1983 and 1987. Cases were women aged 26–66 years with histologically confirmed primary carcinoma of the breast who were hospitalized in the Montpellier Cancer Institute and had not previously undergone any therapy. Controls were women of the same age range admitted for the first time into three different wards: neurology, neurosurgery and general surgery. Information was recorded on the occurrence of BC in the family (sisters, mother and aunts) and the number of sisters and aunts. A case-control study was carried out in the Nancy Cancer Institute14 (France) between 1985 and 1987 in which cases were aged 24–83 years, with a histologically-confirmed infiltrating breast carcinoma. Controls were women admitted into general surgery or general medicine wards, and they were examined to eliminate a diagnosis of cancer. Controls were matched to cases by age at interview (±3 years), residential area and occupational status. Each case was matched to two controls. Information was recorded on the occurrence of BC in the family (sisters, mother, aunts and grandmothers) and the number of sisters and aunts.
Table 1 Studies included in the combined analysis Study
Country
No. of cases
Rohan et al., 1988
Australia
451
Lee et al., 1991
No. and typea of controls 451
P
Age at interview 20–74
Year of interview and type of interview 1982–1984
in-personb
Singapore
200
420
H
24–88
1986–1988
in-person
Lifanova et al., unpublished
Russia
681
834
H
21–87
1992–1994
in-person
Richardson et al.,1991
France
450
603
H
21–66
1983–1987
in-person
Luporsi,1988
France
406
812
H
24–83
1985–1987
in-person
Le et al., 1984
France
265
265
H
22–46
1982–1984
in-person
Clavel et al., 1991
France
495
785
H
20–56
1983–1987
in-person
Total
2948
4170
a H: hospital-controls; P: population-based controls. b Type of interview: in-person = in-person interview.
216
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
In a multicentre case-control study performed in France15 between 1981 and 1984, cases were ø46 years, with a histologically verified breast carcinoma diagnosed less than a year before the interview. Each case was matched with one control with respect to hospital, date of interview and age. The controls were chosen from patients with non-malignant diseases, excluding benign breast disease and severe or moderate cervical dysplasia. Information was recorded on the occurrence of BC in the family (sisters, mother, aunts and grandmothers) and the number of sisters and aunts. Data were obtained from a case-control study in five French hospitals16 between 1983 and 1987 where cases were 20–56 years and had a histologically confirmed infiltrating or in situ breast carcinoma. Three types of controls were eligible for each case: friends, colleagues or patients hospitalized for a nonmalignant disease. Controls were matched to cases on centre, age at interview (±5 years) and year of interview (±14 months). Each case and her matched controls was interviewed by the same interviewer. Information was recorded on the occurrence of BC in the family (sisters, mother, aunts and grandmothers) and the number of sisters and aunts.
Statistical methods The combined relative risks were estimated using a Bayesian random-effects logistic regression model. A random effects analysis assumes that the true effects in each study are not necessarily equal (as is assumed in a fixed effect logistic regression analysis), but are random perturbations about some common mean effect. Inferences based on this model were obtained by the simulation technique known as Gibbs Sampling17 implemented by the BUGS software.18 This model allows for between study heterogeneity in effects and hence gives wider 95% CI on combined effects when heterogeneity is present. The risk of BC associated with a family history of BC was estimated according to the type of relationship between the subject and her affected relative (first-degree or second-degree relative), the age at interview, and menopausal status. The risks associated with reproductive factors were also calculated for each subgroup with or without a family history of BC stratified according to age at interview or menopausal status. The presence of heterogeneity across studies and the significance of interaction effects were assessed using Wald test statistics. The reproductive variables studied were age at menarche, age at first childbirth, number of children, number and type of abortions (spontaneous, induced or both), age at menopause, and estimated duration of ovarian activity as a measure of BCMA. Among nulliparous women, duration of BCMA until first childbirth (pre-first-childbirth BCMA) was calculated by totalling the years between menarche and interview for premenopausal women or menopause for post-menopausal women. Among parous women, pre-first-childbirth BCMA was calculated as the years between menarche and first childbirth. The pre-first-childbirth BCMA was categorized into four classes (Τable 2). Lifetime BCMA was calculated by totalling the years of reproductive life between menarche and interview for premenopausal women, or between menarche and menopause for post-menopausal women, and then subtracting the estimated total years of full term pregnancy which is the number of children multiplied by 0.75 (0.75 years = average length of
Table 2 Combined odds ratios (OR) of breast cancer associated with reproductive factors Reproductive factors
OR
95% CI
Pa
Age at menarche (years)b 15
0.73
(0.61–0.87)
,0.001
Age at first child (years)c No childbirth
1
30
1.10
(0.89–1.35)
0.002
No. of childrend No child
1
1–2
0.89
(0.76–1.05)
>3
0.71
(0.60–0.86)
,0.001
No. of abortionse 0
1
1
1.05
(0.91–1.22)
>2
1.02
(0.86–1.21)
ns
Type of abortionsf Any abortion
1
Spontaneous only
0.99
(0.79–1.22)
Induced only
1.28
(1.02–1.66)
Both
1.23
(0.87–1.68)
–
Age at menopauseg Premenopausal
1
Menopausal ,50 years
0.61
(0.48–0.78)
Menopausal >50 years
0.92
(0.70–1.21)
–
Pre-first-childbirth-BCMA (years)h 21
1.45
(1.17–1.82)
,0.001
Lifetime BCMA (years)i 38
2.16
(1.77–2.65)
,0.001
a P value for trend test. b Adjusted for age at interview, age at first child, number of abortions,
number of children, menopausal status and family history of BC. c Adjusted for age at interview, number of abortions, age at menarche,
number of children, menopausal status and family history of BC. d Adjusted for age at interview, age at first child, number of abortions, age at
menarche, menopausal status and family history of BC. e Adjusted for age at interview, age at menarche, age at first child, number of
children, menopausal status and family history of BC. f Adjusted for age at interview, age at first child, number of abortions,
number of children, age at menarche and family history of BC. g Adjusted for age at interview, number of abortions, number of children,
menopausal status and family history of BC. h Adjusted for age at interview, age at first child, number of abortions,
menopausal status and family history of BC. i Adjusted for age at interview, age at first child, number of abortions, and
family history of BC.
BREAST CANCER: REPRODUCTIVE AND FAMILIAL FACTORS, AN INTERACTION STUDY
217
pregnancy) (lifetime BCMA is equal to pre-first-childbirth BCMA in nulliparous women). Then lifetime BCMA was categorized in four classes (Table 2). Lifetime BCMA was not investigated stratified by age at interview because of its strong correlation with age at interview. Because detailed information on incomplete pregnancies was not available for most studies, a precise assessment of BCMA was not possible. However, analyses were performed adjusting for total number of abortions (both spontaneous and induced). Lee et al.’s data12 were excluded from analyses on the effect of abortions, and Rohan et al.’s and Lee et al.’s data11,12 were excluded from analyses on the familial effect of an affected second-degree relatives due to lack of such information in these studies.
with >3 children (P trend , 0.001). A decreased risk of BC with an age at first childbirth of ø24 years (OR = 0.84; 95% CI : 0.72–1.00) and an increasing risk of BC with increasing age at first childbirth among parous women (P trend = 0.002) were found. Although no increased risk was found with the number of abortions, a significant increased risk was associated with induced abortions (OR = 1.28; 95% CI : 1.02–1.66). A significant decrease in risk was found associated with menopause, especially when menopause occurred before 50 years (OR = 0.60; 95% CI : 0.48–0.78). A significant increased risk of BC associated both with an increased duration of pre-first-childbirth BCMA and with an increased duration of lifetime BCMA (P trend , 0.001) was found.
Results
Variation in the familial risk of BC according to degree of familiality, age and menopausal status
Main effects of reproductive factors The main effects of most of the reproductive factors were investigated previously.10 The odds ratios (OR) are summarized in Table 2. A significant decrease in risk of BC was found for women with an age at menarche of ù15 years compared with those with an age at menarche ,13 years and for women
The main effect of family history of BC according to the degree of relationship, age at interview and menopausal status is shown in Table 3. The OR estimated from the combined analysis associated with a history of BC among first-degree relatives was 2.34 (95% CI : 1.79–3.13), among second-degree relatives, 2.07 (95% CI : 1.22–3.60) and among either first- or second-degree
Table 3 Combined odds-ratios (OR) of breast cancer (BC) associated with family history of BC according to type of relationship and stratified by age at interview and menopausal status Without a family history
With a family history
Type of relationship
Cases
Controls
Cases
Controls
1st degree relativesb
OR
95% CI
Pa 0.001
2702
4008
246
162
2.34
(1.79–3.13)
Age 3 children than for those having one or two in the subgroup of younger women (1–2 children, ORFR = 2.78: 95% CI : 1.49–6.00; >3 children, ORFR = 1.44; 95% CI : 0.50–3.94). When the effect of a family history of BC was estimated across reproductive life factors stratified by menopausal status (data not shown), the results were almost similar to those when stratified by age at interview. However, one difference was that the point estimates of OR FR for induced abortion were similar in the two subgroups pre-menopausal and postmenopausal women. Also the inverse pattern in the point estimates of ORFR for number of children observed when stratified by age at interview was less clear when stratified by menopausal status. An increasing familial risk as lifetime BCMA increased was observed, except in the last category in which a slight decrease was seen in both pre- and post-menopausal women subgroups. For duration of BCMA ,28 years, 28–32 years, 33–37 years and .37 years, ORFR equals 1.60, 1.97, 2.56 and 2.51 in the postmenopausal women subgroup and 2.14, 2.31, 3.03 and 2.22 in the pre-menopausal women subgroup.
BREAST CANCER: REPRODUCTIVE AND FAMILIAL FACTORS, AN INTERACTION STUDY
219
Table 4 Variation of breast cancer (BC) risk associated with reproductive factor according to the presence or not of a family history of BC in first degree relatives and variation of familial risk according to reproductive factors (familial risk associated with 1st degree or either 1st or 2nd degree relative)
Reproductive factors
Risk associated with reproductive factor in those:
Familial risk associated with
without a 1st degree relative with BC
1st degree relative
cas/cont.
OR
with a 1st degree relative with BC
95% CI
cas/cont.
OR
95% CI
1st or 2nd degree relative
OR
95% CI
Pa
OR*
95% CI
Age at menarcheb 15 yrs
487/ 902
0.73 (0.59–0.87)
53/37
0.88 (0.49–1.55)
2.86
(1.71–4.79)
1
86/50
1
2.29 (1.61–3.31) 0.58
2.18 (1.60–3.01) 2.25 (1.47–3.41)
Age at first child birthc no child birth
416/534
1
34/13
1
3.34
(1.72–6.83)
2.07
(1.42–2.91)
2.74 (1.64–4.75)
30 yrs
327/375
1.10 (0.88–1.35)
37/17
0.92 (0.40–2.20)
2.68
(1.44–5.26)
2.80 (1.61–5.08)
0.83 (0.70–0.98)
102/84
0.55 (0.28–1.10)
0.63
1.95 (1.41–2.72)
Number of childrend 0 (1–2) >3
3.29
(1.73–6.81)
1532/2097
416/534
1.12 (0.89–1.37)
1
134/88
34/13
0.76 (0.33–1.50)
1
2.18
(1.60–3.00)
754/1372
0.94 (0.72–1.19)
78/61
0.67 (0.29–1.40)
2.30
(1.56–3.38)
2.72 (1.67–4.67) 0.55
2.01 (1.50–2.67) 2.38 (1.66–3.34)
Number of abortionse 0
1401/2043
2.11
(1.51–2.99)
1
501/719
1.04 (0.89–1.21)
1
129/96 49/28
1.26 (0.73–2.19)
1
2.54
(1.49–4.39)
>2
599/824
1.02 (0.86–1.20)
54/28
1.31 (0.75–2.25)
2.66
(1.54–4.57)
1.93 (1.50–2.54) 0.73
2.07 (1.43–3.11) 3.17 (2.13–4.80)
Type of abortionse any abortion
2.01
(1.37–2.92)
spontaneous only
1117/1674 313/515
0.97 (0.78–1.20)
1
112/88 41/27
1.20 (0.68–2.18)
1
2.49
(1.40–4.50)
1.87 (1.37–2.69)
induced only
545/676
1.27 (1.02–1.69)
40/20
1.43 (0.74–2.72)
2.45
(1.33–4.96)
2.96 (1.80–5.24)
both
119/144
1.19 (0.85–1.63)
11/2
4.82 (1.12–34.6)
7.82
(1.86–58.4)
4.44 (1.66–13.3)
0.43
2.09 (1.32–3.34)
Age at menopausef pre-menopausal
2.39
(1.68–3.48)
,50 yrs
1323/1900 606/1174
0.62 (0.50–0.77)
1
120/76 58/46
0.65 (0.39–1.11)
1
2.46
(1.56–3.90)
>50 yrs
762/919
0.95 (0.75–1.19)
68/40
0.87 (0.51–1.46)
2.10
(1.32–3.33)
2.41 (1.72–3.64) 0.87
2.27 (1.45–3.58) 2.07 (1.30–3.32)
Pre-first-childbirth-BCMAg 21 yrs
457/520
1.43 (1.19–1.73)
39/17
1.66 (0.86–3.39)
2.70
(1.49–5.13)
2.27 (1.44–3.71)
1
90/72
1
2.09 (1.55–2.81) 0.74
2.05 (1.51–2.83)
Lifetime BCMAh 38 yrs
419/452
2.06 (1.68–2.53)
44/19
2.57 (1.34–5.03)
2.58
(1.38–5.04)
2.38 (1.44–4.02)
1
66/61
1
2.03 (1.50–2.88) 0.95
2.12 (1.51–3.04)
* Numbers of study subjects are given in table 5a (age 3
81/140
1
284/507
1.13
(0.77–1.67)
22/11
68/49 0.86 (0.37–2.05) 1147/1504
1
309/375
1.12 (0.88–1.45) 167/125 0.84 (0.45–1.49) 0.30
1
38/21
1
95/139
1.44
(0.90–2.37)
14/13 0.65 (0.22–1.91) 612/1191
0.86 (0.67–1.15)
111/90 0.89 (0.47–1.01)
Number of abortions e 0
260/415
1
1
93/151
0.90
(0.54–1.43)
61/50
19/13 1.25 (0.49–3.28)
1
1050/1537 383/534
1.06 (0.89–1.26)
1
159/137
64/49 1.09 (0.71–1.73) 0.48
1
>2
65/138
0.76
(0.47–1.32)
21/7 2.45 (0.87–7.44)
486/665
1.00 (0.83–1.21)
81/42 1.47 (0.94–2.32)
Type of abortionse any abortion
225/340
1
spontaneous only 53/84
0.99
(0.60–1.65)
13/11 1.09 (0.39–3.20)
236/403
0.95 (0.75–1.19)
52/44 1.11 (0.68–1.83) 0.67
74/138
0.85
(0.52–1.47)
20/7 2.43 (0.88–7.65)
432/522
1.31 (1.03–1.82)
59/29 1.66 (0.97–2.89)
22/35
1.00
(0.51–1.90)
4/1 4.96 (0.53–180.0)
89/105
1.21 (0.78–1.75)
15/5 2.73 (0.96–9.15)
induced only both
55/47
1
811/1249
1
138/126
1
Pre-first-childbirth-BCMAf 21 yrs
53/59
1.80
(1.13–2.83)
15/10 1.17 (0.44–3.21)
378/442
1.36 (1.09–1.71)
50/26 1.63 (0.91–2.94)
31/26
1
834/1467
1
121/110
1
101/84 1.06 (0.69–1.64)
a P value of test for three-way interaction. b adjusted for age at first child, number of abortions, number of children. c adjusted for age at menarche, number of abortions, number of children. d adjusted for age at menarche, age at first child, number of abortions. e adjusted for age at menarche, age at first child, number of children. f adjusted for number of abortions, number of children.
The findings of this study with regard to the main effects of reproductive life factors on the risk of BC are in agreement with those of previous studies.1 An increased risk of BC was found in association with an early age at menarche, a late age at first childbirth, nulliparity, premenopausal status and increasing durations of BCMA until the first childbirth and through life. Exposure to menstrual activity has been used as a surrogate for assessing BCMA, which is mainly controlled by oestrogens. In assessing BCMA before first childbirth and over the entire reproductive life, we did not account for periods of oral contraceptive use. However, this may not lead to inaccurate measurement because an increase in BCMA has been found in the later weeks of the oral contraceptive cycle. Indeed, total BCMA may be very similar over an oral contraceptive cycle and a normal cycle.19 There was no association between the risk of BC and the total number of abortions, spontaneous abortions only and both
spontaneous and induced abortions. However, a significant increase in risk of BC was found in association with induced abortions. Published results on the risk of BC associated with abortions are disparate and controversial.1,20 The bias caused by cases being more aware of, or more candid about abortions than controls might explain the increased risk of BC found for induced abortions. Non-homogeneity in the measurement of family history of BC across studies (four studies recorded information on firstand second-degree relatives14–16 [Lifanova, unpublished], two in first- and second-degree relatives but not in grandmothers12,13 and one in first-degree relatives11) might have induced interference in the estimation of an interaction between familial risk and reproductive factors. Our results showed no significant difference in familial risk estimates according to the degree of relationship, with and without stratification on reproductive factors. This homogeneity in results may be partly due to the
BREAST CANCER: REPRODUCTIVE AND FAMILIAL FACTORS, AN INTERACTION STUDY
Table 5b Variation of familial risk according to reproductive factors stratified by age at interview (familial risk associated with either 1st or 2nd degree relative) Familial risk of BC Reproductive factors
age < 40 yrs
age . 40 yrs
OR*
OR*
95% CI
Pa
95% CI
Age at menarcheb 15 yrs
1.75 (0.49–5.81) 2.29 (1.50–3.49)
0.29
Age at first child birthc no child birth
3.48 (1.59–8.67) 2.38 (1.34–4.45)
30 yrs
0.97 (0.28–3.09) 3.92 (2.09–7.71)
0.10
Number of childrend 0
3.90 (1.52–13.3) 2.36 (1.32–4.45)
(1–2)
2.78 (1.49–6.00) 1.83 (1.32–2.47)
>3
1.44 (0.50–3.94) 2.49 (1.73–3.57)
0.30
Number of abortionse 0
2.06 (1.26–3.66) 1.87 (1.38–2.56)
1
2.53 (1.12–4.82) 1.97 (1.26–3.06)
>2
6.62 (2.66–18.9) 2.79 (1.80–4.36)
0.48
Type of abortionse any abortion
1.86 (1.09–3.49) 1.84 (1.32–2.57)
spontaneous only
1.93 (0.73–5.17) 2.12 (1.29–3.48)
induced only
5.57 (2.17–17.6) 2.54 (1.53–4.46)
both
8.72 (0.92–2.63) 3.94 (1.36–13.8)
0.67
Pre-first-childbirth-BCMAf 21 yrs
1.67 (0.65–5.05) 2.46 (1.47–4.19)
0.12
* Numbers of study subjects are given in table 5a. a P value of test for three-way interaction. b adjusted for age at first child, number of abortions, number of children. c adjusted for age at menarche, number of abortions, number of children. d adjusted for age at menarche, age at first child, number of abortions. e adjusted for age at menarche, age at first child, number of children. f adjusted for number of abortions, number of children.
measure of a family history of BC remaining a measure of mixed genetic susceptibilities (and within-family correlated environmental factors) even after having attempted homogenization of the analysis groups by specifying the degree of relationship. Byrne et al.21 found that different factors could modify in different directions the effects of an affected mother and the effects of an affected sister. However, aside from the fact that dissociate mother and sisters will probably not define homogeneous groups of genetic susceptibility, the number of cases in our study was not large enough to subdivide subjects with a first-degree family history of BC according to affected mother or sister. The category ‘no family history’ in this study included those with an unknown family history. This could have biased the results if cases were more aware of such a history than controls. However, cases and controls had a similar proportion of relatives
221
with an unknown cancer status and this proportion was small (about 2%). Also, although this bias might have affected the estimation of the relative risk for the main effect of family history, there is no reason to assume that it would have varied according to the reproductive factors. The number of female relatives was available only in French data sets. In these data sets, no difference in family sizes between cases and controls was observed. Because this information was missing for three out of seven data sets, the study of potential bias in combined familial risk estimates was made impossible. The sensitivity and specificity of family history of BC as surrogate for genetic factors may be questioned because it may reflect either genetic factors or shared environmental factors or both. However, family history is easily measured and not expensive to obtain and is actually a ‘genetic’ factors surrogate available in most epidemiological studies on BC. Thus, without a priori candidate genes, family history measurement as a surrogate for genetic factor may be useful to generate new hypotheses. Some studies have investigated variations in familial risk, whereas other studies have investigated variations in reproductive risk factors. The tests performed to detect variations in risks were similar whichever of these approaches was used. We chose to present and comment on the results for family history, as we were interested in the modifications of the familial risk due to reproductive factors and age at interview or menopausal status. None of the interactions between familial factor, reproductive factors and age at interview or menopausal status was statistically significant. One possible explanation for this is a lack of power to detect interactions, even with large sample sizes. Two-way interactions between age or menopausal status and a family history of BC were not significant in our study. Our results agree with those of a recent meta-analysis that showed a BC risk associated with a family history of BC in first-degree relatives of 1.8 in women ù50 years and of 2.3 in women ,50 years.22 The difference in familial risks did not appear large. In previous studies, familial effects were observed to increase in younger or pre-menopausal women. Because a difference in familial effects on BC risk according to the age or menopausal status of subjects might introduce confounding, in the studied interactions (even with adjustment for age or menopause), three-way interactions (family history—reproductive factor— age or menopausal status of subjects) were performed. Age and menopausal status had no effect on the interaction between familial risk and age at menarche in our study. The protective effect of late menarche has been shown only for premenopausal women in some studies4,23 and for women at all ages in others.24,25 Among eight studies which have investigated the variation in BC risk associated with age at menarche by family history, three found such a variation.3–7,26–28 The variation always reflected the same trend, namely an increased risk associated with a late age at menarche for women with a family history of BC and a decreased risk associated with a late age at menarche for women without family history.5,26,27 All others studies found no increased risk associated with a late age at menarche among women with a family history and they failed to observe a decreasing risk among women without a family history.3,4,6,7,28 A similar increasing familial risk of BC associated with an increasing number of abortions was shown whatever the subgroups, with a slightly higher familial risk associated with ù2
222
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
abortions and with induced abortions in the subgroup of younger women. A possible explanation for these increased familial risks is that recall and/or reporting of abortions is more difficult for older women because of the elapse of time and the past illegality of induced abortion, thereby smoothing the tendency of increasing familial risk with the number of abortions among older women. Previous findings have suggested a synergism between familial factors and abortions and a plausible biological mechanism has been proposed.9 Across pre-first-childbirth BCMA levels, the patterns of the familial risk are similarly unclear for all subgroups. Familial risk variation across lifetime BCMA levels was investigated according to menopausal status and similar patterns were found. That is, an increasing familial risk associated with an increasing lifetime BCMA. To our knowledge, only two studies have examined the interaction between familial risk and BCMA.26,29 These two analyses were performed on the Nurses’ Health Study and used the interval between menarche and menopause, or age at interview if women were premenopausal, as BCMA. In agreement with the trend of our results, they showed an increased familial risk among women who menstruated for ù35 years in the first report based on prevalent BC cases26 but not in the prospective data.29 The potential importance of mitotic activity on genetically susceptible cells has been discussed elsewhere10 and further studies including a very large number of cases may verify the existence of a plausible effect. On the other hand, age might have an effect on the interaction between familial risk and the number of children. Indeed familial risk seemed to increase slightly for women with a high parity (>3 children) compared with women with a low parity (1–2 children) in the older women group and seems to decrease slightly for women with a high parity compared with women with a low parity in younger women group. This could explain the discrepancy between epidemiological studies that have researched an interaction between parity and family history of BC without stratifying on age at interview. Four out of seven studies did not observe a variation in familial risk according to number of children.6,26,29,30 In the three others, no protection from multiple births was observed when compared with nulliparity among women with a family history of BC.4,5,7 This may also explain the discrepancy between family studies. Indeed, a study on linked-to-BRCA1 families found an increased risk for BC associated with low parity (,3) among susceptible related women from highly selected pedigrees where the mean age of BC diagnosis was about 41 years.31 Another family study on non-selected pedigrees where the mean age of BC diagnosis was about 51 years did not find a decreased risk for BC associated with high parity (ù3) among susceptible women.32 Moreover, epidemiological studies that considered the two-way interaction between age at diagnosis and parity have reported a crossover effect of parity on BC risk; i.e. increasing parity was associated with an increased risk of BC in women ,40 years, whereas in older women increasing parity appeared to be protective.8 Further studies are needed to clarify the interrelationship between age, parity and family history of BC. Our results showed that age might modify the interaction between familial risk and the age at first childbirth. Indeed familial risk seems to increase slightly as age at first childbirth increases only in the older group (P = 0.1). The two-way interaction between age at first full-term pregnancy and family
history has been studied in eight published studies. No evident increase in risks according to age at first childbirth was observed among women with a family history of BC in three studies.4,21,29 In two others, a similar increase in risk with age at first child was observed.3,5 In the three remaining studies, in agreement with our finding, particularly in the subgroup of older women, an increase in risk was described which was stronger among women with a family history than among women without a family history.7,30,33 Again, among epidemiological studies considering the two-way interaction between age at diagnosis and age at first childbirth, some have reported that a positive association of age at first birth and BC risk was restricted to younger women,34,35 and others found that it was restricted to older women.36 An interrelationship between age, age at first childbirth, and family history might also exist. A possible influence of reproductive and menstrual factors on familial risk has been suggested previously and emerges from the present study. Three-way interactions between age, family history and parity or age at first childbirth might exist and merit further investigation. However, interaction studies, especially those involving multi-way interactions, are rapidly limited by power. The necessary number of subjects to detect a moderate interaction effect may be prohibitive.37–39 Indeed, Smith and Day37 showed that detecting interactions of the same magnitude as postulated main effects always required increases in the study sizes by at least a factor of four. Methods to detect interaction have been reviewed40–42 and most methods may not be appropriate for the study of interaction involving uncommon exposures or moderate interaction effects. This approach is also limited by multiple testing which weakens the potential discovery of interactions and requires confirmation by other investigations (as generally recommended for any new observation). For now, such studies may be the only way to detect interactions and generate new hypotheses concerning the aetiology of complex diseases such as BC. Thus, further studies with increased power achieved either by using larger pooled data sets, and/or by using more homogeneous methods for the assessment of genetic factors than a family history of BC and/or by selecting a sub-population according to one of the factors involved in the multi-way interaction, are needed to confirm our findings.
References 1 Kelsey JL, Horn-Ross PL. Breast cancer: magnitude of the problem
and descriptive epidemiology. Epidemiol Rev 1993;15:7–16. 2 Adami HO, Hansen J, Jung B, Rimsten A. Familiality in breast cancer:
a case-control study in a Swedish population. Br J Cancer 1980; 42:71–77. 3 Brinton L, Hoover R, Fraumeni J. Interaction of familial and hormonal
risk factors for breast cancer. J Natl Cancer Inst 1982;69:817–22. 4 Negri E, La Vecchia C, Bruzzi P et al. Risk factors for breast cancer:
pooled results from three Italian case-control studies. Am J Epidemiol 1988;128:1207–15. 5 Parazzini F, La Vecchia C, Negri E, Franceschi S, Bocciolone L.
Menstrual and reproductive factors and breast cancer in women with family history of the disease. Int J Cancer 1992;51:677–81. 6 Sellers TA, Potter JD, Severson RK et al. Difficulty becoming pregnant
and family history as interactive risk factors for postmenopausal breast cancer: the Iowa Women’s Health Study. Cancer Cause Control 1993;4:21–28.
BREAST CANCER: REPRODUCTIVE AND FAMILIAL FACTORS, AN INTERACTION STUDY
223
7 Colditz GA, Rosner BA, Speizer FE. Risk factors for breast cancer
25 Hsieh C-C, Trichopoulos D, Katsouyanni K et al. Age at menarche, age
according to family history of breast cancer. J Natl Cancer Inst 1996;88: 365–71.
at menopause, height and obesity as risk factors for breast cancer: associations and interactions in an international case-control study. Int J Cancer 1990;46:796–800.
8 Lipworth L. Epidemiology of breast cancer. Eur J Cancer Prev 1995;4:
7–30. 9 Andrieu N, Duffy S, Rohan T et al. Familial risk, abortion and
interactive effect on the risk of breast cancer—a combined analysis of six case-control studies. Br J Cancer 1995;72:744–51. 10 Andrieu N, Smith T, Duffy S et al. The effects of interaction between
familial and reproductive factors on breast cancer risk—a combined analysis of seven case-control studies. Br J Cancer 1998;77:1525–36. 11 Rohan T, McMichael AJ, Baghurst PA. A population-based case-
26 Bain C, Speizer FE, Rosner B, Belanger C, Hennekens CH. Family
history of breast cancer as a risk indicator for the disease. Am J Epidemiol 1980;111:301–08. 27 Malone KE, Daling JR. Family history as a modifier of breast cancer
risk factors. Am J Epidemiol 1992;136:964. 28 Andrieu N, Clavel F, Auquier A et al. Variations in the risk of breast can-
cer associated with a family history of breast cancer according to age at onset and reproductive factors. J Clin Epidemiol 1993;46:973–80.
control study of diet and breast cancer in Australia. Am J Epidemiol 1988;128:478–89.
29 Colditz GA, Willett WC, Hunter DJ et al. Family history, age, and risk
12 Lee HP, Gourley L, Duffy SW, Estève J, Lee J, Day NE. Risk factors for
30 Sellers TA, Kushi LH, Potter JD et al. Effect of family history, body-fat
breast cancer by age and menopausal status: a case-control study in Singapore. Cancer Cause Control 1992;3:313–22.
distribution, and reproductive factors on the risk of postmenopausal breast cancer. N Engl J Med 1992;326:1323–29.
13 Richardson S, Gerber M, Cenee S. The role of fat, animal protein and
31 Narod SA, Goldgar D, Cannon-Albright L et al. Risk modifiers in
vitamin consumption in breast cancer. A case-control study in Southern France. Int J Cancer 1991;48:1–9.
32 Andrieu N, Demenais F. Interactions between genetic and repro-
14 Luporsi E. Breast Cancer and Alcohol. PhD thesis, University of Paris-
Sud, 1988.
of breast cancer. JAMA 1993;270:338–43.
carriers of BRCA1 mutations. Int J Cancer 1995;64:394–98. ductive factors in breast cancer risk in a French family sample. Am J Hum Genet 1997;61:678–90.
15 Lê MG, Bachelot A, Doyon F, Kramar A, Hill C. Oral contraceptive use
33 Dupont WD, Page DL. Breast cancer risk associated with proliferative
and breast cancer or cervical cancer: preliminary results of a French case-control study. In: Wolff J-P, Scott JD (eds). Hormones and Sexual Factors in Human Cancer Aetiology. Amsterdam: Elsevier Science, pp.139–47.
disease, age at first birth, and a family history of breast cancer. Am J Epidemiol 1987;125:769–79.
16 Clavel F, Andrieu N, Gairard B et al. Oral contraception and breast
cancer: a French case-control study. Int J Epidemiol 1991;20:32–38. 17 Gelfand AE, Smith AFM. Sampling-based approaches to calculating
marginal densities. J Am Statist Ass 1990;85:398–409. 18 Thomas A, Spiegelhalter DJ, Gilks WR. BUGS: a program to perform
34 Ewertz M, Duffy SW. Risk of breast cancer in relation to reproductive
factors in Denmark. Br J Cancer 1988;58:99–104. 35 Leon DA. A prospective study of the independent effects of parity
and age at first birth on breast cancer incidence in England and Wales. Int J Cancer 1989;43:986–91. 36 Lubin JH, Burns PE, Blot WJ et al. Risk factors for breast cancer in
women in northern Alberta, Canada, as related to age at diagnosis. J Natl Cancer Inst 1982;68:211–17.
Bayesian inference using Gibbs sampling. In: Bernardo JM, Berger JO, Davis AP et al. (eds). Bayesian Statistics. Oxford: Clarendon Press, 1992, pp.837–42.
37 Smith PG, Day NE. The design of case-control studies: the influence of
19 Pike MC, Spicer DV, Dahmoush L, Press MF. Cell proliferation and
38 Hwang SJ, Beaty TH, Liang KY, Coresh J, Khoury MJ. Minimum
Breast Cancer. Epidemiol Rev 1993;15:17–35. 20 Rosenberg L. Induced abortion and breast cancer: more scientific data
are needed. J Natl Cancer Inst 1994;86:1569–70. 21 Byrne C, Brinton LA, Haile RW, Schairer C. Heterogeneity of the
effect of family history on breast cancer risk. Epidemiology 1991;2: 276–84. 22 Pharoah P, Day N, Easton D, Duffy S, Ponder B. Family history and
the risk of breast cancer: a systematic review and meta-analysis. Int J Cancer 1997;71:800–09. 23 Kampert JB, Whittmore AS, Paffenbarger RS. Combined effect of
childbearing, menstrual events, and body size on age-specific breast cancer risk. Am J Epidemiol 1988;128:962–79. 24 Brinton LA, Schairer C, Hoover RN, Fraumeni JF. Menstrual factors
and risk of breast cancer. Cancer Invest 1988;6:245–54.
confounding and interaction effects. Int J Epidemiol 1984;13:356–65. sample size estimation to detect gene-environment interaction in case-control designs. Am J Epidemiol 1994;140:1029–37. 39 Goldstein AM, Falk RT, Korczak JF, Lubin JH. Detecting gene-
environment interactions using a case-control design. Genet Epidemiol 1997;14:1085–89. 40 Khoury MJ, Flanders WD. Nontraditional epidemiologic approaches
in the analysis of gene-environment interaction: case-control studies with no controls. Am J Epidemiol 1996;144:207–13. 41 Yang Q, Khoury MJ. Evolving methods in genetic epidemiology.III.
Gene-environment interaction in epidemiologic research. Epidemiol Rev 1997;19:33–43. 42 Andrieu N, Goldstein A. Epidemiological and genetic approaches in
the study of gene-environment interaction: An overview of available methods. Epidemiol Rev 1998;20:137–47.
