lnt. J . Cancer: 46, 597-603 (1990) 0 1990 Wiley-Liss, Inc.
Publication of the International Union Against Cancer Publication de I’Union Internationale Contre le Cancer
AGE AT FIRST BIRTH, PARITY AND RISK OF BREAST CANCER: A META-ANALYSIS OF 8 STUDIES FROM THE NORDIC COUNTRIES Marianne EwERTZ’.~,Stephen W. D u r n 2 , Hans-Olov ADAI~II~, Gunnar KV/(LE4,Eiliv LUND~, Olav MEIRIK~, Anders MELLEMGAARD’, Irma SOINI?and Hrafn TULINIUS~ ‘Danish Cancer Registry, Institute of Cancer Epidemiology, Danish Cancer Society, Rosenvcengets Hovedvej 35, Box 839, DK-2100 Copenhagen, Denmark; 2MRC Biostatistics Unit, Fairview Lodge, 5 Shafiesbury Road, Cambridge CB2 2BW, United Kingdom; 3University Hospital, Dept. of Surgery, S-751 85 Uppsala, Sweden; 4University of Bergen, Dept. of Hygiene and Social Medicine, Haukeland Sykehus, N-5021 Bergen, Norway; sUniversity of Trams@, Institute of Community Medicine, Box 41 7, N-9001 Tromso, Norway; 6World Health Organization, 1211 Geneva 27, Switzerland; ?Dept. of Public Health, University of Tampere, SF-30101 Tampere, Finland; ‘Icelandic Cancer Society, PO Box 5420, IS-125 Reykjavik, Iceland. Several large epidemiological studies in the Nordic countries have failed to confirm an association between age at first birth and breast cancer independent of parity. To assess whether lack of power or heterogeneity between the countries could explain this, a meta-analysis was performed of 8 population-based studies (3 cohort and 5 case-control) of breast cancer and reproductive variables in the Nordic countries, including a total of 5,568 cases. It confirmed that low parity and late age at first birth are significant and independent determinants of breast-cancer risk. Nulliparity was associated with a 30% increase in risk compared with parous women, and for every 2 births, the risk was reduced by about 16%. There was a significant trend of increasing risk with increasing age at first birth, women giving first birth after the age of 35 years having a 40% increased risk compared to those with a first birth before the age of 20 years. Tests for heterogeneity between studies were not significant for any of the examined variables. In the absence of bias, this suggests that several individual Nordic studies may have had too little power to detect the weak effect of age at first birth observed in the meta-analysis.
Before 1970, low parity was considered an established risk factor for breast cancer. MacMahon et al. (1970) demonstrated in an international multicentre case-control study a protective effect of low age at first birth, indicating that this accounted in large measure for the protective effect of high parity. Since then, numerous studies have confirmed the association between breast cancer and age at first birth, employing various methods, such as hospital-based case-control studies (Paffenbarger et al., 1980; Helmrich et al., 1983), population-based case-control studies (Hunt et al., 1980; Paul et al., 1986), cohort studies (Trapido, 1983; Pathak et a l . , 1986; Leon, 1989), and case-control studies nested in cohorts (Bain et al., 1981; Brinton et al., 1983). Some of these have found an additional protective effect of parity (Brinton et al., 1983; Helmrich et al., 1983; Pathak et al., 1986; Leon, 1989). Practically all breast-cancer studies conducted in the Nordic countries support a protective effect of high parity. However, the effect of age at first birth does not appear consistent. Some studies (Soini, 1977; Tulinius et a l . , 1978; Lund, 1989; Mellemgaard et al., 1990) have shown a clear protective effect of an early first birth, one study (Kvile et al., 19876) a weak effect, while others (Adami et al., 1980; Meirik et al., 1986; Ewertz and Duffy, 1988) have failed to detect a significant association between age at first birth and breast-cancer risk. The latter studies were reasonably large, population-based and without obvious sources of bias. Since a traditional review of studies from the Nordic countries did not explain the varying effect of age at first birth on breast cancer risk, we decided to approach the problem quantitatively, in this case by conducting a meta-analysis. This was designed to address specifically the following 4 questions: 1. Are the observed differences in the effects of parity and age at first birth on breast-cancer risk in the Nordic countries
sufficiently systematic to be statistically significant or are they compatible with chance variation? 2. If the differences are significant, are they best described by time, place or methodology? 3. If the differences are not significant, what are the overall estimates of risk associated with parity and age at first birth? 4. Are the effects consistent across age groups? MATERIAL AND METHODS
The analysis included only population-based studies of the cohort and case-control type from the 5 Nordic countries (Denmark, Finland, Iceland, Norway and Sweden). Authors of recently published studies from these countries on breast-cancer epidemiology were approached and contacts with the cancer registries and other research institutes ensured that no major unpublished material was omitted. Some small studies (with less than 100 cases) were excluded from the analysis, also studies where the subjects were selected on the basis of specific exposures ( e . g . , Storm et a l . , 1986; Bergkvist et a l . , 1989). Since all studies in the meta-analysis have been published, only a brief description is given of each (appearing alphabetically by country) and pertinent details are summarized in Table 1. Denmark A case-control study based on the entire Danish population (Ewertz and Duffy, 1988) was conducted, including 1,487 breast-cancer cases diagnosed over a 1-year period (19831984). They were identified from the files of the nationwide trial of the Danish Breast-Cancer Co-operative Group and the Danish Cancer Registry. The control group was an agestratified random sample of 1,338 women without breast cancer, from the general population identified from the Central Population Register. Data on risk factors were collected by self-administered (mailed) questionnaires, completed by close to 90% of cases and 80% of controls. In 1967, a cervical-cancer screening programme started in Maribo County of all resident women, born 1918-1939, including a PAP-smear and an interview on reproductive history (Lynge and Poll, 1986; Mellemgaard et a l . , 1990). Information was obtained on number of pregnancies and age at first pregnancy, but not whether these pregnancies continued to term. Breast-cancer cases were identified by record linkage to the Danish Cancer Registry. Data were available for analysis of 270 breast cancer cases in a total cohort of 14,421 women.
T o whom reprint requests should be addressed. Received April 4, 1990 and in revised form June 23, 1990.
598
EWERTZ ET AL.
TABLE I - STUDIES INCLUDED IN THE META-ANALYSIS OF THE EFFECT OF REPRODUCTIVE FACTORS ON BREAST CP.NCER RISK IN SCANDINAVIA Study
County
Type of
Number of
Number of
studv
cases
controlslpopulation
Ewertz and Duffy, (1988) Mellemgaard ef a/. (1990) Soini (1977) Tulinius et al. (1978) (updated)
Denmark
Case-control Cohort Case-control
1487 270 122 769 1466 1 04 317 1034
1338
Kvile et al. (1987a,b) Meirik et al. (1986)
Adami et al. 11980)
Denmark
Finland Iceland Norway Norway
Sweden
Sweden
Cohort
Cohort Case-control Case-control Case-control
Finland This case-control study (Soini, 1977) included 122 breastcancer cases, of which 95 were treated at Tampere Central Hospital between 1970 and 1975, and 27 were diagnosed in a breast-cancer screening programme conducted in 1974-1 975 in Tampere. A total of 534 controls were selected in the screening, by taking the first subject with a normal finding following a case with an abnormal finding. Data on risk factors were obtained by interviews. Iceland Since 1964, a national cervical-cancer screening programme has been in operation in Iceland. At each visit to the clinic, the women completed a questionnaire on reproductive history. By the end of 1975, 216 breast-cancer cases were identified by the Icelandic Cancer Registry in a cohort of 3 1,453 women from which results were reported by Tulinius et al. (1978). For the present analysis, the dataset was updated with a further 10 years of follow-up. By the end of 1985, 769 women, out of 58,706 women attending the programme, had developed breast cancer. Norway In 19561959, a screening programme for early diagnosis of breast cancer was carried out in 4 counties in Norway, including a standardized interview and a clinical breast examination (Kvile et al., 1987a,b). The analysis was restricted to 3 counties (Nord-Tr~ndelag,Aust-Agder, Vestfold) and to women who had complete data on reproductive history. By record linkage to the Norwegian Cancer Registry, 1466 breast cancer cases were identified between 1961 and 1980 in a cohort of 59,405 women. As a part of a joint international case-control study of oral contraceptives (Meirik et al., 1986), 105 breast-cancer cases aged less than 40 years were traced from all (71) surgical departments of Norwegian hospitals in 1984-1985 and verified against the Norwegian Cancer Registry. Two controls per case, born on the same day and year as the case, were chosen from the general population. Data on reproductive history were obtained by personal interviews. Sweden The other part of the oral-contraceptives study (Meirik et al., 1986) included personal interviews with 3 17 breast-cancer cases, aged less than 45 years, identified in 1984-1985 from the 6 regional cancer registries in Sweden, and with one control per case selected from the general population. The controls were born in the same year and month ( +/ - 1) as the cases and were resident in the same county. An earlier case-control study was carried out in 1978 covering about a third of the Swedish population (Adami et al., 1980). Cases were reported to the investigators by,.all pathology departments in the 3 regions of Uppsala, Orebro and Umei. Controls were chosen from the female population in these regions and matched for age with the cases. Data on parity and age at first birth were collected by mailed question-
14421 534 58706 59405 208 317 1086
Age at diagnosis
Time of diagnosis
25-70 25-69 41-60 25-79 32-94 25-39 25-44 26-93
1983-84 1967-84 197C-75 1964-85 1961-80 1984-85 1984-85 1978
naires completed by 94% of cases and 92% of controls. The present analysis includes 1,034 cases and 1,086 controls, the data from the study as published, plus a further 28 cases and 80 controls. Analysis Data from each study were submitted for analysis in a standardized format specifying categories of age and time of diagnosis, levels of parity and age at first birth for cases and controls or person-years at risk. The age limits for inclusion were 25-94 years. In the first stage of the analysis, each study was analyzed separately, the case-control studies by logistic regression and the cohort studies by log-linear (Poisson) regression (Berry, 1983). Both methods produced estimates of the relative risk. These were combined by taking a multivariate weighted average of the logarithms of the relative risk estimates, weighting by the inverse covariance matrices of the logarithm of the estimates. The univariate analogue of this is already commonly used where dichotomous risk factors are to be combined (Prentice and Thomas, 1987). The method provides a framework for point and interval estimation of relative risks and testing of effects on risk and heterogeneity from study to study. It is intuitively appropriate as it gives greater weight to larger studies and lesser weight to smaller ones. Mathematical details are given in the Appendix. RESULTS
Table I1 shows the effect of age at first birth on breast-cancer risk in each study separately and in ;a11 studies combined, adjusted for age and parity. In 3 studies (Soini, 1977; Meirik et al., 1986; Ewertz and Duffy, 1988), there was no effect of age at first birth, while a significant trend appeared of increasing risk with increasing age at first birth in the 5 other studies, the effect being most pronounced in Ilceland (Tulinius et al., 1978). The combined analysis confirmed this trend, with a smooth increase in risk to a RR = 1.40 (95% CI 1.15-1.70) for women with a first birth after the age of 35, as compared with women whose first birth occurred before the age of 20 years. There was no significant heteaogeneity among studies. The results of the analysis of parity ,arepresented in the same way in Table 111. Except for one study (Meirik et al., 1986), all studies showed a decreasing risk with increasing parity. In the combined analysis, the trend was highly significant with a RR = 0.69 (95% CI 0.61-0.79) for women with 5 or more births compared with those who had 1-2 births. No significant heterogeneity among studies or age groups was observed. There was no significant interaction (xi = 10.99, p > 0.2) between parity and age at first birth (‘Table IV). The effect of nulliparity (Table V) was not as consistent as for increasing parity among parous women, the test for heterogeneity approaching significance, p == 0.09. In the combined analysis, nulliparity was associated with a RR = 1.30 (95% CI 1.20-1.41).
599
PARITY A N D BREAST CANCER RISK I N NORDIC COUNTRIES TABLE I1 - EFFECT OF AGE AT FIRST BIRTH IN EACH STUDY SEPARATELY FOR PAROUS WOMEN, AND LN ALL STUDIES COMBINED (IN EACH ONE, ADJUSTED FOR AGE AND PARITY)
Study
Age at first birth'
Ewertz and Duffy (1988)
Publication of the International Union Against Cancer Publication de I’Union Internationale Contre le Cancer
AGE AT FIRST BIRTH, PARITY AND RISK OF BREAST CANCER: A META-ANALYSIS OF 8 STUDIES FROM THE NORDIC COUNTRIES Marianne EwERTZ’.~,Stephen W. D u r n 2 , Hans-Olov ADAI~II~, Gunnar KV/(LE4,Eiliv LUND~, Olav MEIRIK~, Anders MELLEMGAARD’, Irma SOINI?and Hrafn TULINIUS~ ‘Danish Cancer Registry, Institute of Cancer Epidemiology, Danish Cancer Society, Rosenvcengets Hovedvej 35, Box 839, DK-2100 Copenhagen, Denmark; 2MRC Biostatistics Unit, Fairview Lodge, 5 Shafiesbury Road, Cambridge CB2 2BW, United Kingdom; 3University Hospital, Dept. of Surgery, S-751 85 Uppsala, Sweden; 4University of Bergen, Dept. of Hygiene and Social Medicine, Haukeland Sykehus, N-5021 Bergen, Norway; sUniversity of Trams@, Institute of Community Medicine, Box 41 7, N-9001 Tromso, Norway; 6World Health Organization, 1211 Geneva 27, Switzerland; ?Dept. of Public Health, University of Tampere, SF-30101 Tampere, Finland; ‘Icelandic Cancer Society, PO Box 5420, IS-125 Reykjavik, Iceland. Several large epidemiological studies in the Nordic countries have failed to confirm an association between age at first birth and breast cancer independent of parity. To assess whether lack of power or heterogeneity between the countries could explain this, a meta-analysis was performed of 8 population-based studies (3 cohort and 5 case-control) of breast cancer and reproductive variables in the Nordic countries, including a total of 5,568 cases. It confirmed that low parity and late age at first birth are significant and independent determinants of breast-cancer risk. Nulliparity was associated with a 30% increase in risk compared with parous women, and for every 2 births, the risk was reduced by about 16%. There was a significant trend of increasing risk with increasing age at first birth, women giving first birth after the age of 35 years having a 40% increased risk compared to those with a first birth before the age of 20 years. Tests for heterogeneity between studies were not significant for any of the examined variables. In the absence of bias, this suggests that several individual Nordic studies may have had too little power to detect the weak effect of age at first birth observed in the meta-analysis.
Before 1970, low parity was considered an established risk factor for breast cancer. MacMahon et al. (1970) demonstrated in an international multicentre case-control study a protective effect of low age at first birth, indicating that this accounted in large measure for the protective effect of high parity. Since then, numerous studies have confirmed the association between breast cancer and age at first birth, employing various methods, such as hospital-based case-control studies (Paffenbarger et al., 1980; Helmrich et al., 1983), population-based case-control studies (Hunt et al., 1980; Paul et al., 1986), cohort studies (Trapido, 1983; Pathak et a l . , 1986; Leon, 1989), and case-control studies nested in cohorts (Bain et al., 1981; Brinton et al., 1983). Some of these have found an additional protective effect of parity (Brinton et al., 1983; Helmrich et al., 1983; Pathak et al., 1986; Leon, 1989). Practically all breast-cancer studies conducted in the Nordic countries support a protective effect of high parity. However, the effect of age at first birth does not appear consistent. Some studies (Soini, 1977; Tulinius et a l . , 1978; Lund, 1989; Mellemgaard et al., 1990) have shown a clear protective effect of an early first birth, one study (Kvile et al., 19876) a weak effect, while others (Adami et al., 1980; Meirik et al., 1986; Ewertz and Duffy, 1988) have failed to detect a significant association between age at first birth and breast-cancer risk. The latter studies were reasonably large, population-based and without obvious sources of bias. Since a traditional review of studies from the Nordic countries did not explain the varying effect of age at first birth on breast cancer risk, we decided to approach the problem quantitatively, in this case by conducting a meta-analysis. This was designed to address specifically the following 4 questions: 1. Are the observed differences in the effects of parity and age at first birth on breast-cancer risk in the Nordic countries
sufficiently systematic to be statistically significant or are they compatible with chance variation? 2. If the differences are significant, are they best described by time, place or methodology? 3. If the differences are not significant, what are the overall estimates of risk associated with parity and age at first birth? 4. Are the effects consistent across age groups? MATERIAL AND METHODS
The analysis included only population-based studies of the cohort and case-control type from the 5 Nordic countries (Denmark, Finland, Iceland, Norway and Sweden). Authors of recently published studies from these countries on breast-cancer epidemiology were approached and contacts with the cancer registries and other research institutes ensured that no major unpublished material was omitted. Some small studies (with less than 100 cases) were excluded from the analysis, also studies where the subjects were selected on the basis of specific exposures ( e . g . , Storm et a l . , 1986; Bergkvist et a l . , 1989). Since all studies in the meta-analysis have been published, only a brief description is given of each (appearing alphabetically by country) and pertinent details are summarized in Table 1. Denmark A case-control study based on the entire Danish population (Ewertz and Duffy, 1988) was conducted, including 1,487 breast-cancer cases diagnosed over a 1-year period (19831984). They were identified from the files of the nationwide trial of the Danish Breast-Cancer Co-operative Group and the Danish Cancer Registry. The control group was an agestratified random sample of 1,338 women without breast cancer, from the general population identified from the Central Population Register. Data on risk factors were collected by self-administered (mailed) questionnaires, completed by close to 90% of cases and 80% of controls. In 1967, a cervical-cancer screening programme started in Maribo County of all resident women, born 1918-1939, including a PAP-smear and an interview on reproductive history (Lynge and Poll, 1986; Mellemgaard et a l . , 1990). Information was obtained on number of pregnancies and age at first pregnancy, but not whether these pregnancies continued to term. Breast-cancer cases were identified by record linkage to the Danish Cancer Registry. Data were available for analysis of 270 breast cancer cases in a total cohort of 14,421 women.
T o whom reprint requests should be addressed. Received April 4, 1990 and in revised form June 23, 1990.
598
EWERTZ ET AL.
TABLE I - STUDIES INCLUDED IN THE META-ANALYSIS OF THE EFFECT OF REPRODUCTIVE FACTORS ON BREAST CP.NCER RISK IN SCANDINAVIA Study
County
Type of
Number of
Number of
studv
cases
controlslpopulation
Ewertz and Duffy, (1988) Mellemgaard ef a/. (1990) Soini (1977) Tulinius et al. (1978) (updated)
Denmark
Case-control Cohort Case-control
1487 270 122 769 1466 1 04 317 1034
1338
Kvile et al. (1987a,b) Meirik et al. (1986)
Adami et al. 11980)
Denmark
Finland Iceland Norway Norway
Sweden
Sweden
Cohort
Cohort Case-control Case-control Case-control
Finland This case-control study (Soini, 1977) included 122 breastcancer cases, of which 95 were treated at Tampere Central Hospital between 1970 and 1975, and 27 were diagnosed in a breast-cancer screening programme conducted in 1974-1 975 in Tampere. A total of 534 controls were selected in the screening, by taking the first subject with a normal finding following a case with an abnormal finding. Data on risk factors were obtained by interviews. Iceland Since 1964, a national cervical-cancer screening programme has been in operation in Iceland. At each visit to the clinic, the women completed a questionnaire on reproductive history. By the end of 1975, 216 breast-cancer cases were identified by the Icelandic Cancer Registry in a cohort of 3 1,453 women from which results were reported by Tulinius et al. (1978). For the present analysis, the dataset was updated with a further 10 years of follow-up. By the end of 1985, 769 women, out of 58,706 women attending the programme, had developed breast cancer. Norway In 19561959, a screening programme for early diagnosis of breast cancer was carried out in 4 counties in Norway, including a standardized interview and a clinical breast examination (Kvile et al., 1987a,b). The analysis was restricted to 3 counties (Nord-Tr~ndelag,Aust-Agder, Vestfold) and to women who had complete data on reproductive history. By record linkage to the Norwegian Cancer Registry, 1466 breast cancer cases were identified between 1961 and 1980 in a cohort of 59,405 women. As a part of a joint international case-control study of oral contraceptives (Meirik et al., 1986), 105 breast-cancer cases aged less than 40 years were traced from all (71) surgical departments of Norwegian hospitals in 1984-1985 and verified against the Norwegian Cancer Registry. Two controls per case, born on the same day and year as the case, were chosen from the general population. Data on reproductive history were obtained by personal interviews. Sweden The other part of the oral-contraceptives study (Meirik et al., 1986) included personal interviews with 3 17 breast-cancer cases, aged less than 45 years, identified in 1984-1985 from the 6 regional cancer registries in Sweden, and with one control per case selected from the general population. The controls were born in the same year and month ( +/ - 1) as the cases and were resident in the same county. An earlier case-control study was carried out in 1978 covering about a third of the Swedish population (Adami et al., 1980). Cases were reported to the investigators by,.all pathology departments in the 3 regions of Uppsala, Orebro and Umei. Controls were chosen from the female population in these regions and matched for age with the cases. Data on parity and age at first birth were collected by mailed question-
14421 534 58706 59405 208 317 1086
Age at diagnosis
Time of diagnosis
25-70 25-69 41-60 25-79 32-94 25-39 25-44 26-93
1983-84 1967-84 197C-75 1964-85 1961-80 1984-85 1984-85 1978
naires completed by 94% of cases and 92% of controls. The present analysis includes 1,034 cases and 1,086 controls, the data from the study as published, plus a further 28 cases and 80 controls. Analysis Data from each study were submitted for analysis in a standardized format specifying categories of age and time of diagnosis, levels of parity and age at first birth for cases and controls or person-years at risk. The age limits for inclusion were 25-94 years. In the first stage of the analysis, each study was analyzed separately, the case-control studies by logistic regression and the cohort studies by log-linear (Poisson) regression (Berry, 1983). Both methods produced estimates of the relative risk. These were combined by taking a multivariate weighted average of the logarithms of the relative risk estimates, weighting by the inverse covariance matrices of the logarithm of the estimates. The univariate analogue of this is already commonly used where dichotomous risk factors are to be combined (Prentice and Thomas, 1987). The method provides a framework for point and interval estimation of relative risks and testing of effects on risk and heterogeneity from study to study. It is intuitively appropriate as it gives greater weight to larger studies and lesser weight to smaller ones. Mathematical details are given in the Appendix. RESULTS
Table I1 shows the effect of age at first birth on breast-cancer risk in each study separately and in ;a11 studies combined, adjusted for age and parity. In 3 studies (Soini, 1977; Meirik et al., 1986; Ewertz and Duffy, 1988), there was no effect of age at first birth, while a significant trend appeared of increasing risk with increasing age at first birth in the 5 other studies, the effect being most pronounced in Ilceland (Tulinius et al., 1978). The combined analysis confirmed this trend, with a smooth increase in risk to a RR = 1.40 (95% CI 1.15-1.70) for women with a first birth after the age of 35, as compared with women whose first birth occurred before the age of 20 years. There was no significant heteaogeneity among studies. The results of the analysis of parity ,arepresented in the same way in Table 111. Except for one study (Meirik et al., 1986), all studies showed a decreasing risk with increasing parity. In the combined analysis, the trend was highly significant with a RR = 0.69 (95% CI 0.61-0.79) for women with 5 or more births compared with those who had 1-2 births. No significant heterogeneity among studies or age groups was observed. There was no significant interaction (xi = 10.99, p > 0.2) between parity and age at first birth (‘Table IV). The effect of nulliparity (Table V) was not as consistent as for increasing parity among parous women, the test for heterogeneity approaching significance, p == 0.09. In the combined analysis, nulliparity was associated with a RR = 1.30 (95% CI 1.20-1.41).
599
PARITY A N D BREAST CANCER RISK I N NORDIC COUNTRIES TABLE I1 - EFFECT OF AGE AT FIRST BIRTH IN EACH STUDY SEPARATELY FOR PAROUS WOMEN, AND LN ALL STUDIES COMBINED (IN EACH ONE, ADJUSTED FOR AGE AND PARITY)
Study
Age at first birth'
Ewertz and Duffy (1988)
