Epidemiology and Cancer Prevention
Carotenoids, antioxidants and ovarian cancer risk in pre- and postmenopausal women
Article first published online: 24 JUL 2001
Copyright © 2001 Wiley-Liss, Inc.
International Journal of Cancer
Volume 94, Issue 1, pages 128–134, 1 October 2001
How to Cite
Cramer, D. W., Kuper, H., Harlow, B. L. and Titus-Ernstoff, L. (2001), Carotenoids, antioxidants and ovarian cancer risk in pre- and postmenopausal women. Int. J. Cancer, 94: 128–134. doi: 10.1002/ijc.1435
- Issue published online: 23 AUG 2001
- Article first published online: 24 JUL 2001
- Manuscript Accepted: 8 MAY 2001
- Manuscript Revised: 2 MAY 2001
- Manuscript Received: 9 FEB 2001
- National Cancer Institute. Grant Number: CA54419
- ovarian neoplasms;
- vitamin A
An inverse association between ovarian cancer risk, carotenoids and antioxidant vitamins has been suggested by several epidemiologic studies and 1 experimental trial of a vitamin A analogue. From a population-based study of 549 cases of ovarian cancer and 516 controls, we estimated the consumption of the antioxidant vitamins A, C, D and E and various carotenoids, including alpha- and beta-carotene and lycopene, using a validated dietary questionnaire. Multivariate logistic regression was used to calculate the exposure odds ratios adjusted for established ovarian cancer risk factors. Intakes of carotene, especially alpha-carotene, from food and supplements were significantly and inversely associated with risk for ovarian cancer, predominantly in postmenopausal women. Intake of lycopene was significantly and inversely associated with risk for ovarian cancer, predominantly in premenopausal women. Food items most strongly related to decreased risk for ovarian cancer were raw carrots and tomato sauce. Consumption of fruits, vegetables and food items high in carotene and lycopene may reduce the risk of ovarian cancer. © 2001 Wiley-Liss, Inc.
Carotenoids include substances able to be converted into vitamin A, such as alpha- and beta-carotene, as well as other compounds, which cannot be converted into vitamin A but have more potent antioxidant properties, such as lycopene from tomatoes. Together with vitamins A, C and E, the carotenoids are believed to be important in cancer prevention because of their properties as antioxidants or their ability to affect cell differentiation or proliferation. Whether these substances affect risk for ovarian cancer is of interest given its high case-fatality ratio. Three case-control studies have reported that either beta-carotene or vitamin A might be inversely related to ovarian cancer risk.1–3 In addition, an experimental trial to assess whether the vitamin A analogue, fenretinide, might prevent a second breast cancer yielded the surprising finding of a decreased risk for ovarian cancer.4 In our report, we assessed the association between ovarian cancer and vitamins A, C, E and the carotenoids using data from a population-based case-control study. Taking the lead from an investigation of breast cancer suggesting that menopausal status might modify the association between carotenoids and breast cancer risk,5 we examined whether menopausal status also influenced associations with ovarian cancer.
MATERIAL AND METHODS
The parent study has been described previously.6 Briefly, between May 1992 and March 1997, 1,080 incident cases of ovarian cancer among women residing in eastern Massachusetts or New Hampshire were identified through hospital tumor boards and statewide cancer registries. We excluded 203 of these cases because they had died (n = 91), moved or had no phone (n = 51), did not speak English (n = 14) or had a nonovarian primary on review (n = 47); 877 eligible women remained. Physicians denied access to 126 (14%) of these women, 136 (16%) declined to participate and 52 (6%) cases with nonepithelial tumors were excluded, leaving 563 cases with epithelial ovarian cancer, including tumors of borderline malignancy, who were studied.
We identified controls using a combination of random digit dialing (RDD) and selection from voter lists (townbooks). In both groups women who had had bilateral oophorectomy were ineligible. The RDD controls were selected to match the ages and telephone prefixes of cases. In approximately 10% of RDD calls to households, the person answering the phone declined to provide an age census within the household; in about 80% of calls there was no eligible match. Of the remaining households with a potentially eligible control, 72% agreed to participate. Because RDD proved inefficient for identifying controls over age 60 in Massachusetts, we selected older controls from voter lists of residents available in Massachusetts to match cases by community of residence and age within 4 years. Of 328 sampled controls, 21% could not be reached, 18% were ineligible and 30% declined to participate. A total of 523 (421 RDD and 102 townbook) controls were studied.
After obtaining written informed consent, we conducted an in-person interview to assess menstrual, reproductive, medical and family histories, as well as demographic information. Subjects were unaware of specific hypotheses; and exposures were assessed prior to a reference date, defined as 1 year before the date of diagnosis for cases and date of interview for controls. A previously validated self-administered food-frequency questionnaire7 assessed the frequency of consumption of stated portions of specified foods ranging from “never or less than once per month” to “6+ per day.” Subjects were excluded if they failed to complete a dietary questionnaire or their energy intakes were implausibly low or high (<600 or >5,000 kcal per day), leaving 549 cases and 516 controls in the final analyses. Average daily intakes of micronutrients were calculated based on algorithms that assigned a nutrient score to each individual food item using information from United States Department of Agriculture (USDA) sources8 and further adjusted by use of supplements. Food composition data for specific types of carotenoids were based on the USDA-NCI carotenoid database developed by Chug-Ahuja et al.9 and Mangels et al.10 The carotenoid content of tomato-based food products was updated with values from the USDA.11 The questionnaire and algorithms used in this study produced scores that correlated with plasma carotenoid levels in men and women.12
We used unconditional multivariate logistic regression to analyze the effect of carotenoid or vitamin intake on ovarian cancer risk.13 All analyses were performed using the SAS system (SAS Institute, Cary, NC). Based on the distributions for vitamin or carotenoid antioxidant intake in the entire study population,14 we categorized subjects as to their quintile of intake. Multivariate logistic regression models were run, first including the quintiles as categorical variables in comparison to the lowest quintile of consumption, and then testing for linear trends by modeling the quintiles as an ordinal variable, so that odds ratios (ORs) were calculated per quintile increase in consumption of the carotenoid or vitamin in question. Analyses were also performed on the intake of selected food items, similarly modeling frequency of food item intake first as categorical variables in comparison to the baseline of least frequent consumption, and then testing for a linear trend.
In all our models, we adjusted for the matching variables, age (categorical, in decades) and study site (Massachusetts, New Hampshire). Based on previous analyses from this study,6 which suggested that controls were more likely than cases to have gone beyond high school, married, had children and used oral contraceptives, we adjusted for education (high school or less, more than high school), marital history (ever married, never married), parity (nulliparous, 1 live birth, 2 live births, 3 or more live births) and oral contraceptive use (never or less than 3 months, 3 months or more). Other risk factors for ovarian cancer included as potential confounders were family history of breast, ovarian or prostate cancer in a first-degree relative (no, yes) and tubal ligation (no, yes). Finally, we adjusted for body mass index (continuous) and total caloric intake as a suggested method15 to account for the potential under (or over-) reporting of food intake by subjects.
We conducted subgroup analyses according to menopausal status (premenopausal or postmenopausal), in which menopausal case subgroups were compared to like-controls, and then according to histologic subtype of ovarian cancer. Women were defined as postmenopausal if they indicated they had ceased having periods for at least 1 year and (for cases) had stopped having periods before the cancer surgery. Of 40 cases and 40 controls who had had a hysterectomy before their periods had ceased, 8 cases and 11 controls were assigned to the postmenopausal group based on a reference age of older than 50 years. Regarding histologic subtypes of epithelial ovarian cancer, a distinction between serous and mucinous subtypes has been emphasized.16 We have further suggested that a distinction between serous borderline and serous invasive tumors is necessary based on the fact that these tumors have different molecular genetic “fingerprints.”17 This distinction is unnecessary for the mucinous borderline and mucinous invasive tumors that frequently coexist in the same specimen.18 Thus the categories chosen were serous invasive, serous borderline, mucinous (both borderline and invasive), and other.
Table I indicates that, for all subjects, micronutrients associated with decreased ovarian cancer risk based on a significant trend (final column) included total carotene (p = 0.03) and its separate alpha and beta components (p = 0.02 and p = 0.03, respectively), and lycopene (p = 0.003). These associations for all subjects reflected stronger associations among either premenopausal or postmenopausal women. The reduced risk associated with carotene, especially alpha-carotene, was most apparent among postmenopausal women, whereas the lycopene association was most apparent among premenopausal women. The adjusted OR for ovarian cancer among postmenopausal women associated with consumption of alpha-carotene in the highest quintile compared to the lowest was 0.42 [95% confidence interval (CI) 0.22–0.77]. Similarly, the adjusted OR for ovarian cancer among premenopausal women associated with consumption of lycopene in the highest quintile compared to the lowest was 0.37 (95% CI 0.20–0.66). Levels of consumption in Table I are based on all sources including supplements. For the most part, the associations shown in Table I were similar when intake was based only on food sources (data not shown). However, decreasing risk for ovarian cancer among all (p = 0.05) women was associated with vitamin A obtained from food sources only, but not with vitamin A from all sources including supplements (Table I). Other micronutrients studied and not found to relate to risk for ovarian cancer included vitamins C, D and E.
|Nutrient||Quintile2||95% CI for quintile 5||p for Trend|
|No. of cases/controls||130/83||104/109||114/99||106/107||95/118|
|No. of cases/controls||74/48||52/63||62/60||47/55||48/62|
|No. of cases/controls||56/35||52/46||52/39||59/52||47/56|
|No. of cases/controls||127/86||108/105||115/97||101/112||98/116|
|No. of cases/controls||64/52||50/54||68/59||51/67||50/56|
|No. of cases/controls||63/34||58/51||47/38||50/45||48/60|
|No. of cases/controls||128/84||109/104||108/105||106/107||98/116|
|No. of cases/controls||72/47||59/60||55/62||49/60||48/59|
|No. of cases/controls||56/37||50/44||53/43||57/47||50/57|
|No. of cases/controls||116/96||121/92||98/115||115/98||99/115|
|No. of cases/controls||68/52||57/51||45/66||64/54||49/65|
|No. of cases/controls||48/44||64/41||53/49||51/44||50/50|
|No. of cases/controls||127/79||113/106||115/98||99/114||95/119|
|No. of cases/controls||69/35||59/64||54/49||49/67||52/73|
|No. of cases/controls||58/44||54/42||61/49||50/47||43/46|
|No. of cases/controls||97/88||129/111||106/103||112/103||105/111|
|No. of cases/controls||57/47||81/69||51/55||51/54||43/63|
|No. of cases/controls||40/41||48/42||55/48||61/49||62/48|
|Vitamin A (IU/day)||≤6,478||--9,183||--13,370||--18,721||>18,721|
|No. of cases/controls||119/93||116/97||99/114||111/102||104/110|
|No. of cases/controls||68/58||63/55||50/68||51/48||51/59|
|No. of cases/controls||51/35||53/42||49/46||60/54||53/51|
|Vitamin A from food only||≤5,678||--7,875||--10,649||--16,128||>16,128|
|No. of cases/controls||124/88||108/105||109/104||110/103||98/116|
|No. of cases/controls||71/47||54/65||61/57||49/61||48/58|
|No. of cases/controls||53/41||54/40||48/47||61/42||50/58|
|Vitamin C (mg/day)||≤97||--146||--204||--337||>337|
|No. of cases/controls||108/104||116/97||117/96||100/113||108/106|
|No. of cases/controls||68/57||62/51||52/52||45/67||56/61|
|No. of cases/controls||40/47||54/46||65/44||55/46||52/45|
|Vitamin D (IU/day)||≤162||--257||--386||--584||>584|
|No. of cases/controls||120/92||97/116||102/111||110/103||120/94|
|No. of cases/controls||67/55||53/64||54/57||49/64||60/48|
|No. of cases/controls||53/37||44/52||48/54||61/39||60/46|
|Vitamin E (IU/day)||≤4||--6||--14||--37||>37|
|No. of cases/controls||110/101||100/113||108/106||121/92||110/104|
|No. of cases/controls||61/56||47/63||61/67||66/54||48/48|
|No. of cases/controls||49/45||53/50||47/39||55/38||62/56|
Table II shows the ORs for specific histologic types of ovarian cancer associated with carotene and lycopene consumption. The protective effect of alpha-carotene consumption was most apparent for invasive serous cancers and for mucinous tumors (borderline and invasive combined), whereas the protective effect of lycopene consumption was most apparent for serous tumors of borderline malignancy, which tend to occur more frequently in premenopausal women.
|Nutrient||Quintile2||95% CI for quintile 5||p for trend|
|No. of cases/controls||50/83||45/109||39/99||46/107||41/118|
|No. of cases/controls||23/83||14/109||18/99||15/107||16/118|
|No. of cases/controls||19/83||16/109||16/99||16/107||10/118|
|No. of cases/controls||38/83||29/109||41/99||29/107||28/118|
|No. of cases/controls||51/86||50/105||44/97||38/112||38/116|
|No. of cases/controls||19/86||13/105||20/97||18/112||16/116|
|No. of cases/controls||21/86||15/105||17/97||11/112||13/116|
|No. of cases/controls||36/86||30/105||34/97||34/112||31/116|
|No. of cases/controls||47/84||46/104||42/105||41/107||45/116|
|No. of cases/controls||22/84||17/104||17/105||13/107||17/116|
|No. of cases/controls||22/84||15/104||12/105||17/107||11/116|
|No. of cases/controls||37/84||31/104||37/105||35/107||25/116|
|No. of cases/controls||46/79||45/106||51/98||46/114||33/119|
|No. of cases/controls||17/79||25/106||17/98||14/114||13/119|
|No. of cases/controls||22/79||10/106||13/98||12/114||20/119|
|No. of cases/controls||42/79||33/106||34/98||27/114||29/119|
Table III shows the distribution of all cases and controls by frequency of consumption of foods that contribute to carotenoid intake, together with associated ORs. Consistent with the strong effects of lycopene and carotene, consumption of tomato sauce and raw carrots were the food items most clearly associated with decreasing risk for ovarian cancer. Relative to those who reported eating tomato sauce less than monthly, women consuming 2 or more one-half-cup servings of tomato sauce per week had a 40% decrease in risk for ovarian cancer (OR = 0.60; 95% CI, 0.37–0.99). Relative to those who reported eating raw carrots less than monthly, women consuming 5 or more servings (4 cut sticks) of raw carrots per week had a 54% decrease in their risk for ovarian cancer (OR = 0.46; 95% CI, 0.28–0.78).
|Frequency of consumption||No. of cases1||No. of controls1||Adjusted2 odds ratio (95% CI)|
|Less than monthly||60||39||1.00|
|1–3 times/month||156||108||0.92 (0.56–1.51)|
|1 time/week||204||214||0.68 (0.43–1.09)|
|2+ times/week||121||144||0.60 (0.37–0.99)|
|p for trend = 0.009|
|Less than monthly||50||41||1.00|
|1–3 times/month||108||87||0.97 (0.57–1.64)|
|1 time/week||155||119||1.09 (0.66–1.80)|
|2–4 times/week||168||201||0.69 (0.43–1.13)|
|5+ times/week||67||66||0.88 (0.50–1.54)|
|p for trend = 0.10|
|Less than monthly||395||370||1.00|
|1–3 times/month||92||87||1.00 (0.71–1.41)|
|1 time/week||34||29||1.05 (0.61–1.79)|
|2+ times/week||20||25||0.65 (0.34–1.22)|
|p for trend = 0.40|
|Less than monthly||111||100||1.00|
|1–3 times/month||245||218||1.05 (0.74–1.49)|
|1 time/week||157||162||1.04 (0.70–1.54)|
|2+ times/week||31||34||0.84 (0.45–1.57)|
|p for trend = 0.79|
|Less than monthly||229||212||1.00|
|1–3 times/month||201||180||1.02 (0.76–1.35)|
|1 time/week||89||91||0.91 (0.63–1.31)|
|2+ times/week||28||30||0.86 (0.49–1.53)|
|p for trend = 0.55|
|Less than monthly||369||317||1.00|
|1–3 times/month||129||138||0.88 (0.65–1.18)|
|1 time/week||31||42||0.77 (0.46–1.27)|
|2+ times/week||13||12||1.07 (0.47–2.46)|
|p for trend = 0.39|
|Less than monthly||124||109||1.00|
|1–3 times/month||168||154||0.99 (0.70–1.42)|
|1 time/week||158||149||1.01 (0.70–1.45)|
|2+ times/week||94||100||0.83 (0.55–1.26)|
|p for trend = 0.47|
|Less than monthly||187||123||1.00|
|1–3 times/month||125||123||0.78 (0.54–1.11)|
|1 time/week||103||116||0.69 (0.47–1.00)|
|2–4 times/week||96||95||0.79 (0.53–1.17)|
|5+ times/week||34||56||0.46 (0.28–0.78)|
|p for trend = 0.01|
|Less than monthly||364||326||1.00|
|1–3 times/month||132||133||0.84 (0.63–1.14)|
|1 time/week||36||40||0.83 (0.51–1.36)|
|2+ times/week||14||14||0.84 (0.38–1.86)|
|p for trend = 0.26|
|Less than monthly||186||162||1.00|
|1–3 times/month||213||203||1.01 (0.75–1.36)|
|1 time/week||106||111||0.91 (0.64–1.31)|
|2+ times/week||37||34||1.21 (0.71–2.07)|
|p for trend = 0.91|
|Less than monthly||102||79||1.00|
|1–3 times/month||178||182||0.80 (0.55–1.16)|
|1 time/week||179||164||0.96 (0.65–1.40)|
|2+ times/week||83||83||0.84 (0.53–1.32)|
|p for trend = 0.82|
|Less than monthly||66||58||1.00|
|1–3 times/month||153||127||1.22 (0.78–1.90)|
|1 time/week||178||175||1.12 (0.72–1.73)|
|2–4 times/week||130||130||1.09 (0.69–1.71)|
|5+ times/week||17||22||0.90 (0.42–1.92)|
|p for trend = 0.76|
|Less than monthly||155||139||1.00|
|1–3 times/month||157||136||1.19 (0.85–1.68)|
|1 time/week||89||103||0.91 (0.62–1.33)|
|2–4 times/week||93||82||1.14 (0.77–1.69)|
|5+ times/week||44||47||0.87 (0.53–1.43)|
|p for trend = 0.71|
Our study suggests that risk for ovarian cancer may be reduced by the consumption of certain types of carotenoids, adding to existing evidence that these substances may decrease risk for a variety of cancers including cervix,19 prostate20 and breast.21 The studies of prostate cancer and breast cancer may be especially relevant to our findings on ovarian cancer because rates for the 3 diseases correlate internationally.22 In the largest study addressing prostate cancer,20 both total dietary lycopene and tomato sauce consumption were associated with decreased risk, similar to the findings of our study. The potential importance of processed tomato products over raw tomatoes has been attributed to greater bioavailability of lycopene from tomatoes cooked in an oily medium.23 For breast cancer, a meta-analysis of studies conducted between 1982 and 1997 found an inverse association with beta-carotene.21 Of 6 major studies conducted since then,5, 24–28 an inverse relationship with beta-carotene or carrot consumption was found in all but 1.28 In a recent study of breast cancer and the carotenoids, the protective effect was most apparent for beta-carotene and was confined to premenopausal women.5
Regarding ovarian cancer, 2 studies found an inverse (protective) association with carotene.2, 3 Another study found an inverse association with vitamin A, especially from fruits and vegetables, in women under age 50,1 whereas another study found an inverse association between ovarian cancer and consumption of green vegetables and carrots.29 In contrast, a small case-control study of ovarian cancer reported no association with estimated vitamin A consumption,30 whereas a prospective study found there might be a slight increase in risk associated with dietary vitamin A and no association with beta-carotene.31 In another study, Helzlsouer et al.32 did not find lower measured levels of vitamin A, carotene or lycopene in sera stored from women who subsequently developed ovarian cancer, although statistical power was limited in this study involving only 35 cases and 67 controls. Inconsistencies may also relate to differences in the dietary instruments, types of algorithms used to calculate micronutrient consumption, whether supplements are included and differences in the numbers of pre- and postmenopausal women studied. The latter variable may be important because our study suggests that the effect of carotenoids on ovarian cancer risk may vary by menopausal status.
In our study, first considering all cases and controls, we found inverse associations for ovarian cancer risk with vitamin A from food sources only, total carotene and its alpha and beta components, and lycopene. We then examined how associations varied by menopausal status and by histologic type of ovarian cancer. We chose to examine variation by menopausal status because of data suggesting that breast cancer risk associated with carotenoid consumption5 and ovarian cancer risk associated with caffeine consumption may be modified by menopausal status.33 There appeared to be some variation by menopausal status with the inverse association with alpha-carotene stronger for postmenopausal women and the inverse association with lycopene stronger for premenopausal women. However, a formal test for interaction was close to significance (p = 0.08) only for lycopene (data not shown). We also chose to examine associations by histologic subtype of ovarian cancer based on the observation that reproductive and other risk factors may vary especially between the serous and mucinous subtypes.16 We found the association with alpha-carotene was more apparent in women with invasive serous and mucinous tumors, which occur more predominantly in postmenopausal women and the association with lycopene stronger for borderline serous tumors, which occur more often in premenopausal women. Issues of study power prevent a firm statement about whether menopausal status or histologic type is a more important modifier of the risk for ovarian cancer associated with carotenoid consumption.
Major features distinguishing individual carotenoids are their ability to be converted into vitamin A and their relative antioxidant potential. As noted earlier, alpha- and beta-carotene are provitamins that can be converted into vitamin A, whereas lycopene is 1 of the most efficient singlet oxygen quenchers.34 If this distinction relates to the basis for the possible modifying effect of menopause, then it would suggest that provitamin A effects are more important postmenopausally and antioxidant effects more important premenopausally.
In the discussion of the antineoplastic effects of vitamin A or provitamin A, apoptosis is a key mechanism that has been considered. Vitamin A, carotene or synthetic retinoids are capable of producing apoptosis in rat colonic tumor cells,35 human prostate36 and cervix,37 and bovine mammary cells.38 At least part of this effect may be mediated through the ability of the retinoids to block the effect of growth factors such as insulin-like growth factor-1 (IGF-1).38 Although IGF-1 is largely produced by the liver, there is evidence that IGF-1 production may also occur within the ovary and be stimulated by gonadotropins.39 Circulating IGF-1 levels generally decline after menopause, but levels within the ovary could be high in response to the dramatically elevated levels of gonadotropins postmenopausally. Thus, vitamin A or carotene blockage of local effects of gonadotropin-induced growth factors could offer a mechanism for vitamin A or carotene protection in postmenopausal women. If it is reasonable to invoke the higher levels of gonadotropins to explain the effects of carotene on postmenopausal disease, then estrogen could be a factor mediating the effects of lycopene on premenopausal disease. Here, we should consider the possibility that the pro-oxidant and potentially DNA-damaging activity of catechol estrogens,40 which would be higher in premenopausal women, could be countered by the strong free-radical scavenging activity of lycopene.
Our exercise in constructing biologic plausibility is incomplete without a discussion of the factors that could affect the validity of our findings, including the inability to assess long-term diet and possibilities for recall bias, selection bias or confounding. We asked our cases to recall average dietary patterns at least 1 year before diagnosis, which may only partially reflect long-term dietary patterns that are likely more relevant to cancer risk. It is also possible that recent dietary patterns in response to the cancer diagnosis among cases may have influenced recall of diet prior to the disease. A selection bias could have been introduced if any of the dietary factors we studied influenced survival, because 91 potential cases could not be interviewed due to death shortly after diagnosis. Finally, we should consider the potential role of confounding factors in the associations we have described. We have adjusted for key factors that relate to ovarian cancer risk such as age, parity, oral contraceptive use and family history, as well as factors such as body mass index and total calories consumed which might be associated with diet. We also performed additional analyses (not shown) including the addition of caffeine and total, saturated or animal fat consumption in the models and found they made no difference to our estimates.
Given the inconsistency among studies of diet and ovarian cancer and the inability to infer causality from associations found in retrospective epidemiologic studies, it may be premature to suggest that women modify their diets based on the evidence from our study. Still, any lifestyle change must be considered that may assist in the primary prevention of a disease that is generally diagnosed at late stages with poor survival. However, the results from 2 clinical trials of beta-carotene supplements and lung cancer,41, 42 which yielded the surprising finding of an increased risk from the intervention, suggest that any modification should involve foods rather than dietary supplements. Our study suggests that a diet high in fruits and vegetables containing carotenoids, including raw carrots and tomato products, may be important in the prevention of a very lethal form of cancer in women.
- 7The use of a self-administered questionnaire to assess diet four years in the past. Am J Epidemiol 1989;18: 858–67., , , et al.
- 8U.S. Department of Agriculture. Composition of foods—raw, processed, and prepared, 1963–1991. Agriculture handbook no. 8 series. Washington, D.C.: Department of Agriculture, Government Printing Office, 1992.
- 13BreslowNE, DayNE, eds. IARC scientific publication no. 32. Statistical methods in cancer research. The analysis of case-control studies. vol. 1. Lyon: IARC, 1980.
- 15Nutritional epidemiology, 2nd ed. New York: Oxford University Press, 1998..
- 18Tumors of the ovary and maldeveloped gonads. PP83-84. Washington DC: Armed Forces Institute of Pathology, 1979..
- 24Intake of carrots, spinach, and supplements containing vitamin A in relation to risk of breast cancer. Cancer Epidemiol Biomarkers Prev 1997;11: 887–92., , , et al.
- 37Growth retardation in human cervical dysplasia-derived cells by beta carotene through down regulation of epidermal growth factor receptor. Am J Clin Nutr 1995;62: 15355S–405S., , , et al.