Cancer risks in twins: Results from the Swedish family-cancer database
Twin studies on cancer have addressed two general questions, one about the possible carcinogenic effects of twinning and the second about heritable effects of cancer. The first question is answered by comparing the occurrence of cancer in twins to that in singletons; the second is answered in probandwise analysis of monozygotic twins compared to dizygotic twins or siblings. We used the nationwide Swedish Family-Cancer Database on 10.2 million individuals and 62,574 0–66-year-old twins to calculate standardized incidence ratios (SIRs) and 95% confidence intervals (CIs) for all main cancer compared to cancer in singletons. In probandwise analysis, the SIR was calculated for the co-twin of an affected twin. The overall risk of cancer in same or opposite sex twins was at the level of the risk for singletons. Testicular cancer was increased among same sex twins and all twins to an SIR of 1.43. Melanoma was decreased in these groups of twins to an SIR of 0.84. Some other cancer sites were increased or decreased in some groups of twins, but none in all twins. The SIR of breast cancer was 1.01 and 1.04 in same and opposite sex twins, respectively. Probandwise analysis showed increased risks for Hodgkin's disease in males and breast cancer and childhood acute lymphoid leukemia among females. The data on this unselected population of twins suggest that twinning per se is not a risk factor of cancer. However, because twins are smaller than singletons at birth, some possible effects could be masked by such differences. In utero hormonal exposures may be related to the risk of testicular cancer. The protective effects in melanoma may be due to socioeconomic factors. © 2002 Wiley-Liss, Inc.
Twin studies have been a favored approach in the estimation of heritability of a disease.1 However, because of the rareness of twinning and the rather recent establishment of twin registers, this approach has been limited in the cancer field, and modeling studies on cancer causation have reached statistical power only for the main sites of cancer.2, 3 In genetic studies, risks are calculated for twins of affected probands, i.e., probandwise, whereas in comparisons of twins to singletons, risks are calculated for all twins irrespective of the proband status. Twins differ from singletons in birth weight and in intrauterine environment, which have been the subject of a number of studies, particularly in testing the maternal estrogen hypothesis for breast cancer.4, 5 Maternal serum estrogen levels are higher in twin pregnancies, particularly in dizygotic twin pregnancies, compared to singletons, and this has motivated a number of breast cancer studies.6–13 Some studies have found an increased risk of breast cancer among dizygotic or monozygotic twins compared to singletons,5, 9 whereas others have not observed a difference.8, 14, 15 It has also been realized that a number of physiologic parameters are changed in twin pregnancies, in addition to the levels of estrogens. In contrast to breast cancer, a limited number of other cancer sites have been studied among twins.14–17
Data on twins are now available in many databases and registers, which have been collected in a number of ways. However, in almost all twin databases used in cancer studies, the initial step has been to identify living twins or living twin pairs who have responded to a call or questionnaire.3, 9, 15 The condition for survival would imply initially a selection toward a cancer-free population (“healthy twin effect”), which would be expressed as a reduced cancer incidence compared to the whole population. On the other hand, responding to a questionnaire may bias toward health-consciousness. We use here another approach to study cancer in twins, the nationwide Swedish Family-Cancer Database, where the birthdates of the whole population are recorded.18 The Database has been updated in 2001 to include more than 10 million individuals and more than 1 million registered tumors. It offers unique possibilities for reliable estimation of cancer risks in twins compared to singletons, because the data on family relationships and cancers were obtained from registered sources of practically complete coverage. We analyzed separately cancer risks in twins from families of multiple twin pregnancies, and we tested the hypothesis that the co-twin of an affected twin tends to maintain a constant rather than an increasing incidence of breast cancer.19 However, data on twins are only available on a 0–66-year-old population. The Database has been used in some 100 cancer studies so far.
MATERIAL AND METHODS
The Swedish Family-Cancer Database was initially created in the middle of the 1990s by linking an administrative family register on all Swedish families to the Swedish Cancer Registry.18, 20 For each child there are data on both parents at the time of birth. Each person has been assigned a unique technical identification number (which is different from the national identification number, “personal number”), allowing construction of families, for example, through the mother. The Database includes all persons born in Sweden after 1931 with their biological parents, totaling more than 10.2 million individuals. It was updated in the beginning of 2001 to include cancers from the nationwide Swedish Cancer Registry from years 1958–1998. The Database is organized into 3.2 million families, with parents and offspring. Twins and siblings can be identified through their birthdate and parents; thus they could only be identified in the 0–66-year-old offspring population and no data were available on their zygosity.
The completeness of cancer registration in the 1970s has been estimated to be higher than 95% and is now considered to be close to 100%. The percentage of cytologically or histologically verified cancers has been close to 100%.21 The Family-Cancer Database has an incomplete linkage from deceased offspring to parents, particularly among those offspring born from 1932–1940.18 Of a total of 7.0 million offspring, 216,000 have died by the end of follow-up, December 31, 1998. Parental information was missing from 15,000 offspring who had a diagnosis of cancer (9.9% of all offspring cancers). This deficit will prevent the identification of twinships and will cause a small reduction in the number of twin pairs.
The Swedish Cancer Registry is based on compulsory notification of cases.21 A 4-digit diagnostic code according to the 7th revision of the International Classification of Diseases (ICD-7) was used. The following ICD-7 codes were pooled: “upper aerodigestive tract” cancer, codes 140–148 (lip, mouth, pharynx), except for code 142 (salivary glands), and “leukemia,” codes 204–207 (leukemias), 208 (polycythemia vera) and 209 (myelofibrosis). Skin cancer includes only squamous cell carcinomas.
All tumor incidence rates were based on the data in the Family-Cancer Database. Follow-up was started at birth or January 1, 1961, whichever came latest. Follow-up was terminated on diagnosis of cancer, death, emigration, or the closing date of the study, December 31, 1998. Standardized incidence ratios (SIRs) among twins were calculated as the ratio of observed (O) to expected (E) number of cases. The expected numbers were calculated from 5-year-age-, sex-, tumor type-, period- (10-year bands), socioeconomic status- (4 groups: manual worker, blue collar worker, professional, other) and residential area- (2 groups: 3 large cities, rest of the country) specific standard incidence rates among singletons, aged 0–66 years. Confidence intervals (95% CI) were calculated assuming a Poisson distribution. Probandwise risks were calculated by the cohort method described earlier and given in an abbreviated form below, specifically for twins; for singletons the calculation is extended to all siblings.22 Confidence intervals were adjusted for dependence between twin or sibling pairs.
For calculation of the probandwise risk among twins, the following the definitions were used: n1 = 1 affected twin, n2 = 2 affected twins in a family; p0 = total person-years contributed by the twins in families of 0 affected twins; p1 = total person-years contributed by the unaffected twin in families of 1 affected twin; y1 = total person-years contributed by the affected twin in families of 1 affected twin; y2 = total person-years contributed by the affected twins in families of 2 affected twins.
According to the cohort method: A = n2/(p1 + y2); reference rate: B = (n1 + n2)/(p0 + p1 + y1 + y2); SIR = A/B.
The Family-Cancer Database covered years 1961–1998 from the Swedish Cancer Registry and included 1997 cancers among twins aged 0–66 years among 62,574 identified twins. Of these, 1,233 were the same sex as the co-twin, showing an SIR for all male and female cancer of 1.00 and 0.95, respectively (Table I). The data in this and subsequent tables were adjusted for sex, age, period, residential area and socioeconomic status. Among male twins, testicular cancer was increased to an SIR of 1.51 (95% CI 1.15–1.95), and melanoma was decreased to an SIR of 0.61 (0.41–0.89). Among females no SIR was increased between twins and singletons but colon cancer was decreased to an SIR of 0.56 (0.33–0.90). Among all same sex twins, testicular (SIR 1.51) and endocrine (SIR 1.34) tumors were in excess and rectal cancer (0.61) and melanoma was decreased (SIR 0.79). Because multiple comparisons are of concern in these analyses, we show 99% CIs for the significant SIRs in the footnote. Only the SIRs for testicular cancer and male melanoma remained significant at the 1% level.
Table I. Risk for Cancer in Same Sex Twins1
|Upper aerodigestive tract||19||1.11||0.67||1.74||10||1.29||0.61||2.38||29||1.17||0.78||1.68|
|Other female genitals||0||5||1.07||0.34||2.53||5||1.07||0.34||2.53|
|Other male genitals||2||0.75||0.07||2.78||0||2||0.75||0.07||2.78|
Among the population of opposite sex twins, cancer cases numbered 764, and the total male and female SIRs were close to unity, 1.03 and 1.06, respectively (Table II). Bone cancer was increased among men to an SIR of 2.44 (1.11–4.65) but 99% CIs included 1.00 (0.79–5.38). However, among all different sex twins no SIR differed from unity.
Table II. Risk for Cancer in Different Sex Twins1
|Upper aerodigestive tract||15||1.46||0.81||2.41||5||1.14||0.36||2.67||20||1.36||0.83||2.11|
|Other female genitals||0||2||0.75||0.07||2.75||2||0.75||0.07||2.75|
|Other male genitals||3||1.92||0.36||5.68||0||3||1.92||0.36||5.68|
We also calculated risks for all twins (data not shown). The increases in testicular (n = 88, SIR 1.43, 1.14–1.76) and bone cancer (19, 1.77, 1.06–2.76) remained even among all male twins, but for bone cancer the increase failed to be significant for all twins (23, 1.25, 0.79–1.88). The decrease for melanoma in men (52, 0.72, 0.54–0.94) and all twins (145, 0.84, 0.71–0.99) remained. Both colon and rectal cancers were fewer than expected but the SIRs were not significant (colon: 73, 0.85, 0.66–1.06; rectum: 36, 0.77, 0.54–1.06).
Probandwise risks for concordant cancer among male siblings from non-twin (singleton) families and twins are shown in Table III. For singletons, data are shown only for the sites, for which affected twin pairs were identified, and only for singleton families. Data are additionally shown for siblings who were born more or less than 5 years apart from each other. For prostate cancer the birth interval made no difference, but for testicular cancer, melanoma and Hodgkin's disease, those born less than 5 years apart had higher risks, however with overlapping 95% CIs. Siblings born less than 5 years apart showed SIR for prostate (4.14) and testicular (11.07) cancer, melanoma (3.97) and Hodgkin's disease (9.98). We carried out a separate analysis on siblings who had the same father; SIRs increased marginally (data not shown). Among 9 affected twins, all had a same sex co-twin, except for a male twin with melanoma, who had a twin sister with melanoma. Increases were only found for non-Hodgkin's lymphoma (11.86) and Hodgkin's disease (15.84), but only one concordant twin pair was identified for these malignancies and only the SIR for Hodgkin's disease was significant.
Table III. Probandwise Risk for Concordant Cancer in Male Siblings1
Probandwise analysis of female singletons and twins is presented in Table IV. Because of breast cancer, the number of concordant singletons was higher than that in Table III. The SIR for breast cancer was 1.68 and 1.66 when sisters were born less or more than 5 years apart, respectively. SIR was also increased for melanoma (2.75 when sisters were born less than 5 years apart), nervous system cancer (2.60, when sisters were born 5 years or more apart) and leukemia (4.73 when sisters were born less than 5 years apart). Only the SIRs for breast cancer (4.52) and leukemia (30.53) were increased for twins. The leukemia patients were diagnosed with acute lymphoid leukemia before age 20 years.
Table IV. Probandwise Risk for Concordant Cancer in Female Siblings1
We identified a number of families with multiple twins and some families with higher orders of multiple births. However, no cancers were identified in triplets or in higher-order multiple births. In families with triplets, twins were also born but their SIR for cancer did not deviate from the expected rates (Table V). Similarly, in families of two pairs of twins, the SIRs were close to unity. The only exception was families of 3 twin pairs. A significant increase in breast cancer was noted but all cases came from a single family.
Table V. SIR for Cancer in Twins and Siblings in Families With Multiple Twins and Triplets1
| Nervous system||1||0.58||0.00||3.31||3||3.20||0.60||9.46||6||1.24||0.45||2.71||4||1.66||0.43||4.30|
| <15 years||1||0.78||0.00||4.48||2||3.47||0.33||12.74||3||1.22||0.23||3.62||1||0.96||0.00||5.49|
We wanted to test the hypothesis that the co-twin of an affected twin tends to maintain a constant incidence of breast cancer (Table VI). The sister who was first diagnosed with breast cancer was the proband and incidence of breast cancer was calculated for the second sister at two different time intervals. The incidence was 0.8% in the first interval and 0.1% in the second interval, but no more than 17 twin pairs were available for the analysis.
Table VI. Subsequent Annual Incidence (%) of Concordant Breast Cancer in Twins
Multiple comparisons are a problem in this kind of study, and undoubtedly some of the observed changes were fortuitous. The way to judge the plausibility of a result is to evaluate it in terms of biologic plausibility, previous findings and consistency between same and different sex twins and between male and female twins. Although the response may differ between genders and monozygotic and dizygotic twins, the first assumption is that an effect should be reproduced in several twin groups, as discussed below. Higher significance levels may also be required in exploratory type of studies, and we also showed 99% CIs for the findings that were significant with 95% CIs.
Among studies comparing cancer risks on many sites between twins and the general population, our study on 1997 cancers was smaller than the Norwegian study on more than 3,200 cases but larger than the U.K. and U.S. studies on 1,063 and 1,918 cases, respectively.14, 15, 23 Our study did not include an initial population selection but was confined to 0–66-year-old individuals. In the Norwegian study, the overall cancer incidence was decreased, as it was for female cancers in a study based on the Finnish Twin Cohort, which was established in 1975 on live pairs of twins.8 By contrast, in our study on an unselected population of twins, the SIRs of all cancer in twins were exactly at the level of the singleton population, suggesting that twinning per se is no risk factor for cancer.
At specific cancer sites, the comparison of cancer risks between twins and singletons showed increases for testicular cancer among same sex twins and a modest increase in different sex twins, resulting in a significant overall excess of 1.43. The Norwegian study reported a nonsignificant increase in testicular cancer in all twins (SIR 1.20, 95% CI 0.79–1.74), and a UK study reported an increase in dizygotic twins compared to monozygotic twins.7, 15 In our study, the risks were higher for same sex twins, which may experience lower in utero estrogen exposures than different sex twins.24 Thus, the data lent limited support to the assumed testicular cancer risk posed by in utero exposure to estrogens or other pregnancy hormones.25–27 Among the other sites where increases were observed in one group of twins, including tumors in the endocrine glands and bone, no support was found in other groups of twins, and the results may be fortuitous. These results were not significant when 99% CIs were applied.
Apparent protection was found for colon and rectal cancer and melanoma. However, only the decrease in melanoma was reproduced among all twins; colon and rectal cancer were decreased among all twins but the results were not significant. Melanoma and male colon cancer were also decreased in the Norwegian study.15 Although we cannot exclude some biologic mechanism for the protection against melanoma, we suggest that it depends on socioeconomic factors. Twin birth limits the family's possibilities for sun holidays in southern countries and thus would reduce spells for short-term excessive sun exposure for twins.28 Our data were adjusted for the socioeconomic factors but could not consider constraints imposed by twin pregnancies. We have shown elsewhere that children born to large families are protected against melanoma, probably also for socioeconomic reasons.27 Breast cancer showed an SIR of 1.01 in the same sex and 1.04 in different sex twins. This is in agreement with the above Norwegian study and the joint estimate from the 5 studies discussed in the above Finnish study.8, 15 The data are in disagreement with a U.S. cohort study that found an increased risk of breast cancer among dizygotic twins.9 However, recent data suggest that high birth weight and long gestational age among twins are risk factors for breast cancer but whether these would modify the comparisons to the singleton population remains to be established.11–13 The studies in families with multiple twins showed no excess risks of cancer among twins. In one family with 3 twins and an excess of breast cancer, the risk is unlikely to be related to twinning.
The second part of our study covered probandwise analysis of twin cancers but this was limited because of paucity of data. Some of the present twin data would be part of the published Nordic analysis of adult twins.3 Significant increase in cancer risk was found at 3 sites: Hodgkin's disease was increased among male twins (SIR 15.84, but included only 1 pair), and breast cancer (4.52) and leukemia (30.53) were increased among female twins. These risks were much higher than those observed among siblings from singleton pregnancies in the same population, 9.98, 1.68 and 4.73, respectively. On the other hand, sibling but not twin risks were increased for prostate and testicular cancer and melanoma in men and nervous system cancer and melanoma in women. The reason for the differences is probably in the paucity of the twin data. When siblings were analyzed in 2 groups, depending on their age difference, no systematic difference was noted for solid tumors. However, siblings born less than 5 years from each other had a higher risk for Hodgkin's disease and leukemia than siblings with a large age difference. These findings may have etiologic implications, as discussed below.
For leukemia, all concordant cases were acute lymphoid leukemias diagnosed before age 20 years. We have observed these cases previously and consider them as evidence for the hypothesis that the initial transformation event took place during the fetal period in one twin and the co-twin received transformed cells through a shared placenta, which is common among monozygotic twins.29, 30 For Hodgkin's disease, a previous study reported a 99-fold risk in monozygotic twins.31 One may speculate whether the mechanism could be similar to that of leukemia, i.e., spreading of transformed cells between twins who share the placenta. Two aspects deserve comment on breast cancer. The probandwise SIRs for siblings and twins (1.68 and 4.52) were lower than the odds ratios reported for twins in the Nordic study,3 but the present sibling risks were equal to numerous studies.32–34 One reason for the difference is that our twin population contains young monozygotic and dizygotic twins.35, 36 Peto and Mack have suggested that the co-twin of an affected twin maintains a constant rather than an increasing incidence of breast cancer.19 Our limited data showed that the incidence of the co-twin was higher in the first 5 years of follow-up than later. Our dataset was limited and the follow-up time was truncated to 66 years, but the findings were not inconsistent with the hypothesis that has no obvious biologic explanation. The age-incidence relationships in contralateral breast cancer are also apparently paradoxical.37
In summary, our data on medically verified diagnosis and an unselected twin population showed that the risk of cancer in same or different sex twins was at the level of the risk for singletons, with a few possible exceptions. Testicular cancer was increased among same sex twins and all twins, and melanoma was decreased for these twins. Probandwise analysis showed increased risks for Hodgkin's disease in men and breast cancer and childhood acute lymphoid leukemia among women.
The Family-Cancer Database was created by linking registers maintained at Statistics Sweden and the Swedish Cancer Registry.