Lifestyle and endometrial cancer risk: a cohort study from the Swedish twin registry
Case-control studies of lifestyle factors have been inconclusive in the study of endometrial cancer, and prospective data are scarce. Our aim was to examine the associations of physical activity, weight and weight change, fruit, vegetable, and alcohol consumption, socio-economic status, parity and presence of diabetes mellitus with the risk of endometrial cancer in a cohort study. In 1967, 11,659 women in the Swedish Twin Registry, born 1886–1925, answered a 107-item questionnaire, including questions about diet, physical activity and other lifestyle factors. Complete follow-up through 1992 was attained through record linkage to the Swedish Cancer and Death Registers. The relative risks for endometrial cancer were estimated in proportional hazards models that adjusted confidence limits for correlated outcomes. We observed 133 incident cases of endometrial cancer in the cohort. There was no clear pattern of risk over strata of alcohol or fruit and vegetable intake, although the data suggest an increased risk with very low fruit and vegetable intake. Increasing physical activity markedly decreased the risk of endometrial cancer (p for trend < 0.01), independently of weight and parity; the risk in the highest quartile, relative to the sedentary category, was 0.2 (95% CI 0.3–0.8). As expected, higher weight in middle age increased the risk (p for trend < 0.01), as did higher weight in early adulthood. Contrary to previous findings, weight gain did not have an effect independent of weight at enrollment. We did not find a genetic component to endometrial cancer. Our results confirm that environmental factors are the most important, especially physical activity, parity, and weight in young and middle age. Int. J. Cancer 82:38–42, 1999. © 1999 Wiley-Liss, Inc.
Endometrial cancer is the most common cancer of the female reproductive tract, with an expected 30,000 incident cases and 6,000 related deaths in the United States each year (Hill and Austin, 1996). Known risk factors for the disease include estrogen replacement therapy, obesity in middle age, diabetes mellitus and low parity (Grady and Ernster, 1996). Uncertainties remain about the effects of diet, exercise, weight in youth, change in weight, alcohol consumption and genetic heritability.
Body weight in different periods of life, and changes in weight between these periods, have emerged as potential risk factors for cancers of the breast (Magnusson et al., 1998) as well as the endometrium (Le Marchand and Wilkens, 1991; Swanson et al., 1993a; Olson et al., 1995). It is not clear whether the increased risks due to adult weight gain are independent of factors such as initial body weight, parity or physical activity, all of which have been associated with the risk of endometrial cancer (World Cancer Research Fund and American Institute for Cancer Research, 1997; Olson et al., 1997). Confounding by such factors may explain some of the observed variation in risk among these variables. The number of cohort studies of this malignancy is few. In this report, we examine the effects of lifestyle factors in a cohort study of Swedish twins.
MATERIAL AND METHODS
Our study was conducted within a cohort of twins followed by the Swedish Twin Registry, consisting of male and female same-sexed twin pairs, born 1886–1925 and both still living in Sweden in 1961, the year of compilation. In 1961, 12,186 women, approximately 85% of the 14,354 females in the Registry, responded to initial questionnaires. Zygosity was determined for over 95% through questions related to “sameness” of individuals within pairs. It was assumed that most of the missing individuals had either emigrated or were dead (Cederlof et al., 1977). In 1967, a 107-item questionnaire regarding lifestyle factors was mailed to registrees. Excluding those who died prior to assessment (506) and those with prevalent endometrial cancer at baseline (21), this analysis included the 11,659 women who responded to this questionnaire with at least minimal information on potential risk factors.
In this questionnaire, alcohol and smoking were assessed as number of drinks per week and number of cigarettes per day, respectively. We coded current cigarette smoking as “none,” “light” (1–10 cigarettes/day) and “regular” (11+/day). Fruit and vegetable consumption was assessed by a single item on a 4-point relative scale, with response alternatives “very little or no part,” “a small part,” “a moderate part” and “a large part” of the diet. Physical activity was assessed on a relative scale as “hardly any physical exercise,” “light exercise, e.g., regular walks, light gardening,” “regular exercise” and “hard physical training.” A childhood socio-economic variable was derived from subjects' fathers' occupations, e.g., unskilled or skilled labor, professional, etc., and was categorized as “low,” “middle” and “high.”
Subjects were followed from exposure assessment to 1 of 3 study endpoints: diagnosis of endometrial cancer, death, or the end of the study on December 31, 1992. Cancer incidence was ascertained by record linkage with the Swedish Cancer and Mortality Registries (documented to be 98% complete) (Mattsson and Wallgren, 1984). Similarly, death was ascertained by linkage to the Swedish Death Registry.
The genetic component (heritability of liability) in endometrial cancer was analyzed in 6,020 complete twin pairs who responded to the questionnaire in 1961. We estimated the relative importance of genetic and environmental effects using 2 × 2 contingency tables of disease status in twin pairs (twin A vs. twin B) stratified by zygosity status (Neale and Cardon, 1992). Relative risk estimates were obtained from proportional hazards analyses that adjusted confidence limits for multivariate outcomes (White, 1982). Relative risk estimates are presented age-adjusted, as well as with multivariate-adjustment for age, weight, physical activity and parity.
The average age at enrollment was 56.2 years. During a mean follow-up period of 20.4 years, there were 133 incident cases of endometrial cancer diagnosed at an average age of 66.2 years.
In an analysis using all 6,020 female twin pairs responding to the 1961 questionnaire, we observed disease concordance/discordance in 1:53 monozygotic and 3:107 dizygotic pairs. The estimated heritability for liability in endometrial cancer was 0 (95% confidence limits [CI] 0.00–0.50) (Table I). Shared environmental influences were estimated to account for 0.22 (95% CI 0.00–0.41) of the variation in liability, and unshared environmental factors for 0.78 (95% CI 0.50–0.99).
Table I. ESTIMATES OF GENETIC AND ENVIRONMENTAL EFFECTS FOR ENDOMETRIAL CANCER FROM STRUCTURAL EQUATION MODEL FITTING IN THE ENTIRE OLD COHORT OF THE SWEDISH TWIN REGISTRY
|MZ||1||53||0.17||0 (0.0–.50)||0.22 (0.0–0.41)||0.78 (0.50–0.99)||0.24||2||0.88|
High baseline weight was associated with an increased risk of endometrial cancer; the age-adjusted relative risk for the highest vs. the lowest quartile of weight was 2.4 (95% CI 1.4–3.8; p for trend < 0.01) (Table II). The associations were similar for weight at age 25 (relative risk 2.6; 95% CI 1.5–4.6) and weight at age 40 (relative risk 2.2; 95% CI 1.3–3.8). Relative risk estimates for these weight variables were not altered by control for height (data not shown) or changes in weight from age 25 to baseline (Table II).
Table II. ASSOCIATIONS OF ANTHROPOMORPHIC FACTORS AND ENDOMETRIAL CANCER
|Weight at enrollment (in kg)|
| Lowest quartile ≤57||2,888||23||59,847||1.0||referent|| || ||1.0||referent||1.0||referent|
| Second quartile 58–63||2,795||22||58,771||1.0||0.6–1.8|| || ||0.8||0.4–1.5||1.0||0.5–2.3|
| Third quartile 64–70||2,430||27||49,565||1.5||0.8–2.5|| || ||1.1||0.5–2.1||1.8||1.0–4.0|
| Highest quartile 71+||2,954||51||58,677||2.4||1.4–3.82,3|| || ||1.8||0.9–3.43||3.6||1.8–7.92,3|
|Weight at age 25 (in kg)|
| Lowest quartile ≤52||2,006||16||43,179||1.0||referent||1.0||referent|| || ||1.0||referent|
| Second quartile 53–56||2,143||16||46,309||0.9||0.5–1.8||0.9||0.4–1.8|| || ||1.0||0.5–2.0|
| Third quartile 57–62||2,001||20||43,329||1.2||0.6–2.4||1.1||0.5–2.2|| || ||1.3||0.7–2.7|
| Highest quartile 63+||2,113||43||44,185||2.6||1.5–4.62,3||1.9||1.0–3.83|| || ||2.7||1.5–5.32,3|
|Weight change (in kg, age 25 to enrollment)|
| Weight change (lowest quartile ≤1)||2,042||27||42,701||1.7||0.9–3.1||1.9||1.04–3.61||1.3||0.9–3.1|
| Weight change (second quartile 2–5)||1,985||17||44,121||1.0||referent||1.0||referent||1.0||referent|
| Weight change (third quartile 6–11)||2,074||18||45,197||1.1||0.6–2.1||0.8||0.4–1.6||1.1||0.6–2.5|
| Weight change (highest quartile ≥12)||2,151||33||44,809||2.1||1.2–3.81||1.0||0.5–2.0||2.1||1.3–4.71|
|Height (in cm)|
| Lowest quartile (≤157)||2,650||28||52,066||1.0||referent||1.0||referent||1.0||referent||1.0||referent|
| Second quartile (158–161)||2,650||28||52,066||1.2||0.7–1.9||0.9||0.6–1.6||0.7||0.4–1.4||0.9||0.5–1.6|
| Third quartile (162–164)||2,284||25||48,147||0.9||0.5–1.7||0.7||0.4–1.3||0.7||0.3–1.3||0.9||0.5–1.6|
| Highest quartile (165+)||3,409||37||71,157||1.1||0.6–1.7||0.6||0.4–1.1||0.7||0.4–1.3||1.1||0.6–1.9|
Weight gain from age 25 to baseline (an average duration of about 30 years) was also associated with risk (relative risk 2.1, 95% CI 1.2– 3.8), independently of weight before the weight change, physical activity and parity. Control for weight at enrollment, however, eliminated this effect entirely. An increased risk was observed for weight loss during middle age (weight-adjusted relative risk 1.9, 95% CI 1.04–3.6). The relative risk estimates for all weight variables were similar when comparing cancers occurring at or above the age of 60 with cancers occurring below the age of 60 (data not shown). Height was not related to endometrial cancer risk.
Increasing physical activity was associated with a markedly reduced risk of endometrial cancer (Table III). Among those who exercised heavily, the relative risk of endometrial cancer was 0.2 (95% CI 0.04–0.8) compared with those who reported very little exercise. The p-value for trend was highly significant (p < 0.01). Adjustment for other covariates, including body weight, did not alter these findings.
Table III. ASSOCIATIONS OF LIFESTYLE FACTORS AND ENDOMETRIAL CANCER1
|Fruit and vegetable consumption|
| Large part of the diet||2,191||32||46,327||1.0||referent||1.0||referent|
| Moderate part of the diet||4,961||57||104,705||0.8||0.5–1.2||0.7||0.5–1.1|
| Small part of the diet||1,673||11||32,886||0.5||0.2–0.971||0.4||0.2–0.92|
| Little or none||64||2||1,119||2.6||0.7–10.7||3.1||0.8–12.7|
| No children||2,553||31||49,344||1.0||referent||1.0||referent|
| 1–2 children||4,987||69||105,941||1.0||0.7–1.6||0.9||0.6–1.5|
| 3 or more children||2,997||19||62,458||0.5||0.3–0.83,4||0.4||0.2–0.83,4|
| 1–10 cigarettes/day||755||10||16,558||1.0||0.5–1.9||1.2||0.6–2.3|
| 11+ cigarettes/day||348||2||7,379||0.5||0.1–1.8||0.5||0.1–2.0|
| Some (up to 2 drinks/week)||1,211||22||26,367||1.6||1.01–2.62||1.7||1.03–2.82|
| 2 drinks, up to 4 drinks/week||830||10||18,041||1.1||0.6–2.1||1.2||0.6–2.4|
| 4 or more drinks/week||551||7||12,394||1.1||0.5–2.4||1.3||0.6–2.8|
There was no clear pattern of risk over strata of fruit and vegetable consumption, although the higher of the 2 middle intake categories showed an erratic decrease in risk relative to the highest intake (Table III). The indicator for very low fruit and vegetable consumption suggested an increased risk, but confidence intervals were wide.
Increasing parity decreased the risk (p for trend < 0.01), in both univariate and multivariate models (relative risk for 3 or more children vs. none was 0.4, 95% CI 0.2–0.8 (Table II). An increased risk was observed with the presence of diabetes, but confidence limits were wide (relative risk 2.4, 95% CI 0.6–9.7) (Table III). The highest socio-economic status, vs. the lowest, showed a statistically non-significant reduction of risk (relative risk 0.5, 95% CI 0.1–1.9), although no significant trend by socio-economic status was observed.
We found no clear association between smoking and risk of endometrial cancer (Table III). If anything, smoking 10 or more cigarettes per day tended to be protective of endometrial cancer, although confidence limits included unity. The smoking rates in this older cohort of women were relatively low: current smokers constituted only 13%, more than 75% of whom smoked less than 10 cigarettes per day. There was no clear association between alcohol consumption and endometrial cancer risk (Table III).
In this follow-up study, with exposure assessment long before most cases of endometrial cancer, we estimated the heritability of liability for endometrial cancer to be 0, but with a broad confidence interval (0.00–0.50). This analysis essentially compares monozygotic vs. dizygotic twin pairs with respect to disease frequency and assumes that greater disease concordance among monozygotic twins indicates a genetic component. Our findings indicate that genetic effects are, at the most, modest, and that environmental effects are of greater importance. Shared environmental influences were estimated to account for 0.22 (0.00–0.41) of the variation in liability and unshared environmental factors for 0.78 (0.50–0.99). Since the confidence limits for unshared environmental factors did not include unity, our data relect some degree of correlation in the risk of endometrial cancer between twins, due to genetics, shared environment or both.
Our data allowed us to examine some aspects of weight history that have previously received little attention. Our finding that higher weight at enrollment was associated with an increased risk for endometrial cancer is consistent with the results of many prior studies (Grady and Ernster, 1996; World Cancer Research Fund and American Institute for Cancer Research, 1997). Contrary to findings from two previous studies (Le Marchand and Wilkens, 1991; Swanson et al., 1993a), we found that higher weight in young adulthood also increased the risk. Also contrary to previous studies (Le Marchand and Wilkens, 1991; Tornberg and Carstensen, 1994), we found weight at enrollment and weight at age 25 to increase the risk for endometrial cancers occurring before the age of 60, as well as for those occurring at or after the age of 60. Contrary to the findings of Olson et al. (1995), we found that weight gain during middle age did not increased the risk for endometrial cancer. In our study, we were able to control for the potentially confounding effects of weight before the measured change, physical activity and parity, which may help explain some of the discrepancy between our findings and those of previous studies. Interestingly, women who had mostly lost weight during middle age (subjects in the lowest quartile of our weight change variable) appeared to be at increased risk for endometrial cancer in comparison with those in the second quartile, who gained very little weight. Restricting analyses to cancers occuring several years after exposure assessment did not alter these findings, suggesting that previously undetected cancers do not explain weight loss in these subjects.
We found that increasing physical activity lowered the risk of endometrial cancer, especially when subjects with little or no physical activity were compared with those who exercised heavily. Although there were only 2 cases of endometrial cancer in the highest category, physical activity had a significant inverse association with endometrial cancer risk. Relative risk estimates for the 2 middle exercise categories were very similar in magnitude, showing no trend by themselves. This is not surprising, since with a 4-point relative measurement scale there is likely to be considerable non-differential misclassification between the 2 middle categories. A prospective cohort study found increasing occupational physical activity to decrease the risk for endometrial cancer (Moradi et al., 1998), and physical activity was also associated with decreased risk for endometrial cancer in 2 case-control studies (Olson et al., 1997; Shu et al., 1993) although not clearly so in another (Levi et al., 1992). A proposed mechanism for the protective effect of exercise is lowered body mass (Shu et al., 1993), although our relative risk estimates for physical activity were not alterered by controlling for the effects of body mass. A case-control study (Shu et al., 1993) also found that the protective effects of physical activity are independent of body mass.
Alcohol drinking was not associated with an increased risk of endometrial cancer in our study. Approximately 72% of this population did not drink, however, and intake of more than 4 or 5 drinks per week was rare, leaving a limited range of exposure. A prior cohort study of endometrial cancer also found no effect of alcohol drinking (Gapstur et al., 1993). Prior case-control studies on alcohol have shown positive (Shu et al., 1993b; La Vecchia et al., 1986; Parazzini et al., 1995), inverse (Swanson et al., 1993b) and no associations (Austin et al., 1993; Newcomb et al., 1997), while one case-control study found a positive association that was confined to obese women (Webster and Weiss, 1989). We found no association when analyses were restricted to the heaviest women (data not shown), nor was such an association found in a prospective study (Gapstur et al., 1993). It has been suggested that alcohol consumption may increase circulating estrogen levels, thereby potentially increasing the risk for endometrial cancer (Hill and Austin, 1996).
No clear pattern for fruit and vegetable consumption was observed. Our findings were limited by a low number of cases in the lowest intake category. Moreover, we cannot ignore the possibility of confounding by an unknown correlate of this extreme type of diet. Our data suggest a potential for increased risk among those whose fruit and vegetable intake is very low.
Our dietary data did not allow adjustment for total energy intake. Our intake measure qualitatively addressed the relative contribution of fruit and vegetables to the diet as a whole, however, which to some extent may obviate the need for total energy adjustment. Moreover, in a case-control analysis of endometrial cancer, control for total energy intake did not diminish the protective effects of nutrients derived from fruit and vegetable consumption (Barbone et al., 1993). We were unable to control for the potentially confounding effects of a wide range of inter-correlated food groups, although the parameter estimates for fruit and vegetables were not altered by adding similar qualitative measures for meat, sausage, milk, eggs, cake or coffee, to our model. Another limitation of our study is the lack of repeated measures of exposure during the follow-up period.
Although we did not have information on use of oral contraceptives and hormone replacement therapy, confounding by these factors in our study is unlikely. Our cohort was born between 1886 and 1925, which would make the youngest woman in her forties when oral contraceptives began to be used commonly in Sweden. Similarly, the youngest woman would have been in her mid-fifties when hormone replacement therapy became common in Sweden, making widespread exposure in our cohort unlikely (Swedish Corporation of Pharmacies). The possibility remains, however, that some of the women in our cohort may have received exogenous hormones and that this exposure was correlated with some of our exposures of interest. A similar potential for bias might also be due to unmeasured hysterectomy status if this variable is related to any of our exposures of interest. As with hormone replacement therapy, we suspect a very low rate of hysterectomy in this cohort of Swedish women born around the turn of the century. The Swedish Twin Registry (1996; data not shown) found a history of surgery of the uterus in 50/868 women in a younger cohort born between 1926 and 1958 (about 5.8% of those sampled) and in 0/78 in our study cohort (twins born between 1886 and 1925).
The strengths of our study include the prospective assessment of exposure, which has been infrequent in prior studies of endometrial cancer. The potential for differential recall of exposure in prior studies is a concern, especially since associations between lifestyles and cancer have become well known among the public. Our data enabled us to examine the effects of weight at various ages and weight change, and also to control for these factors in our analysis of physical activity. Finally, the nearly complete end-point assertainment in our study reduces the potential for bias from differential follow-up.
In conclusion, our findings suggest that physical activity and parity lower the risk for endometrial cancer, while higher weight in early and middle adulthood increases the risk. No clear pattern of risk was observed for fruit and vegetable intake or with alcohol consumption.
We thank Dr. D.C. Thomas for valuable insights into the analysis of twin studies, Dr. P.D. Allison for providing a SAS Macro for performing hazard analyses with multivariate failure times, and Drs. A. Ahlbom and N. Pedersen and Mr. H. Malmström for assistance with the Swedish Twin Registry. The Swedish Twin Registry is supported by grants from the John D. and Catherine T. MacArthur Foundation and The Swedish Council for Planning and Coordination of Research (FRN).