Int. J . Cancer: 48, 34-38 (1991) 0 1991 Wiley-Liss, Inc.
Publication of the International Union Against Cancer Publication de I'Union lnternationale Contre 1e Cancer
DIETARY AND SERUM CAROTENOIDS AND CERVICAL INTRAEPITHELIAL NEOPLASIA Juliet VANEENWYK'.3*4,Faith G . DAVIS'and Phyllis E. BOW EN^ 'Department of Epidemiology and Biostatistics, School of Public Health, University of Illinois at Chicago; and 2Department of Nutrition and Medical Dietetics, College of Associated Health Professions, University of Illinois at Chicago, Chicago, IL, USA. A case-control study examined the association between cervical intra-epithelial neoplasia (CIN) and serum and dietary alpha-carotene, beta-carotene, cryptoxanthin, lutein, and lycopene. Cases (n = 102) had biopsy confirmed C I N I, II or 111. Controls matched for age. ethnic origin and clinic (n = 102) had normal Pap smears. Participants completed health history and food frequency questionnaires. Fasting venous blood samples were assayed for serum carotenoids. Multivariable conditional logistic regression analyses yielded odds ratios and 95% confidence intervals (CIS) for those in quartiles 3.2, and I (lowest compared to quartile 4 (highest) of serum lycopene of 3.5 ( .1-11.5), 4.7 (1.2-17.7) and 3.8 (l.l-l2.4), respectively. Similar analyses yielded adjusted odds ratios ( 0 4 s ) and 95% CIS Of 4.6 (l.l-l9.7), 5.8 (1.621.3) and 5.4 (I .3-23.3) for dietary intake of lycopene. The findings for lycopene-rich foods (tomatoes) were consistent with this result. C I N was not associated with the lutein. Findings for alphacarotene, beta-carotene and cryptoxanthin were ambiguous. Quartile of vitamin C intake was also inversely associated with CIN with 0 4 s and 95% CIS of 3.7 (0.9-14.6), 4.1 (I.& 17.2). and 6.4 (I .4-30.0) for those in quartiles 3.2. and I compared to quartile 4.
I
Beta-carotene represents 1 of approximately 600 known carotenoids. Its activity with respect to cancer prevention has received attention, because it is one of the major provitamin A carotenoids (Olson, 1988), it functions as an antioxidant, and it enhances the immune response (Bendich, 1989). Epidemiologic evidence has shown increased consumption and/or serum levels of beta-carotene to be protective against many types of cancer (Wald, 1987; Zeigler, 1989). Until recently, the ability to assess intake and serum levels of other carotenoids has been limited. However, due to recent advances in high-pressure liquid chromatography procedures (Stacewicz-Sapuntzakis et al., 1987), it is now possible to measure the relationship between cancer and additional serum carotenoids, including alphacarotene, cryptoxanthin, lutein and lycopene. Concurrently, nutrient data-bases have been expanded to include these carotenoids. While it is expected that these data-bases may need to be refined as field validation studies become available, the data-base developed by Block et al. (1987-1989), offers a useful tool for estimating dietary intake of individual carotenoids. A case-control study of cervical intra-epithelial neoplasia (CIN) was designed to investigate the potential protective effect of several nutrients, including carotenoids. This premalignant condition was chosen to mitigate physiologic changes which may ensue from, rather than foreshadow, neoplasia. Previous studies of CIN, including those of carcinoma in situ, have shown no association between dietary intake of beta-carotene and disease (Wylie-Rosett et al., 1984; Brock et al., 1988; La Vecchia et al., 1988; Slattery et al., 1990) after controlling for other risk factors. In contrast, inverse associations have been noted for serum beta-carotene and disease (Hanis et al., 1986; Brock et al., 1988). MATERIAL AND METHODS
Selection of cases and controls
Cases were identified from non-pregnant, non-post-partum women with biopsy-confirmed diagnoses of CIN I, I1 or 111. Non-pregnant, non-post-partum women who attended the same clinics as the cases and who had a Pap smear as part of their
visit were identified as potential controls. These women became eligible to participate if their Pap smear showed no abnormalities of a severity greater than or equal to those of benign atypia. The controls were matched to the cases by clinic, age (18-24, 25-31, 32-38, 3 9 4 9 ) and ethnicity (black, nonblack). Recruiting took place at the University of Illinois Hospital Clinics and the Cook County Hospital Clinics in Chicago, IL between April 1987 and June 1989. Of 166 potential cases, 102 women participated in the project, yielding a volunteer rate of approximately 61%. To enroll an equal number of controls, study staff approached 195 women, obtaining a volunteer rate of approximately 52%. Informed consent was obtained from all participants, following the requirements of the participating institutions. Measurement of exposure
Participants returned to the clinic for venipuncture following a 10-hr fast. The blood samples were processed and frozen following the specifications of the laboratory (PB) contracted to assay the serum carotenoids. The assays were conducted using high-pressure liquid chromatography techniques (Stacewicz-Sapuntzakis et al., 1987). All laboratory personnel were blind to the case or control status of the blood samples. A failure to withdraw blood from four women resulted in laboratory data from 98 matched pairs. Participants completed the food frequency section (FFQ) of the Health Habits and History Questionnaire provided by the National Cancer Institute, Division of Cancer Prevention and Control (NCI-DCPC), version 2.1 (Block et al., 1987-1989). This questionnaire asks about the frequency of consumption of 99 food and beverage items and provides an open-ended question to record additional foods consumed at least once a week. Forms were self-administered and available in English and Spanish. A trained interviewer, blind to the study hypotheses and case or control status, questioned the participants about responses which were incomplete, internally inconsistent or seemed to contain recording errors. If necessary, responses were changed subsequent to this query. One woman failed to complete the FFQ and one woman was incapable of answering the questions even with help from the interviewer. These FFQs were eliminated from the analysis, resulting in dietary data for 100 matched pairs. The conversion of foods to nutrients was accomplished using NCI-DCPC software, version 2.2, August 1989 (Block et al., 1987-1989). To control for over- and under-reporting of actual intake, adjustment for calories was made using the regression procedure of Willett and Stampfer (1986). Because there is a lack of consensus concerning the use of Willett and Stampfer's regression procedure, the relationship of dietary carotenoids 3Currently at lllinois Department of Public Health, Office of Health Policy and Planning.
4T0whom correspondence and reprint requests should be sent, at the Illinois Department of Public Health, Suite 6-600, 100 W. Randolph, Chicago, 1L 60601, USA. Received; October 4, 1990 and in revised form December 19, 1990.
35
CAROTENOIDS AND CERVICAL DYSPLASIA
and CIN was further assessed without adjustment for total energy intake. The results from this analysis did not differ substantively from those reported.
TABLE I - PERCENTAGE DISTRIBUTIONS' OF CASES AND CONTROLS (UNMATCHED) ON DEMOGRAPHIC, MEDICAL. SMOKING, SEXUAL AND REPRODUCTIVE VARIABLES Cases
Variable
Measurement of potential confounders A self-administered questionnaire was used to obtain information about the participants' background, health and pregnancy history, and smoking and sexual habits. These data were used to assess independent contributors to risk in this sample and to control for potential confounders of the diseaseexposure association. In addition to using the FFQ to measure dietary intake of carotenoids, it was used to assess dietary intake of nutrients which were potential independent contributors to risk andlor confounders of the carotenoid-CIN association. The relationships of dietary vitamins A, C and E to CIN were assessed for this purpose. Statistical analysis Crude (OR,) and adjusted (OR,) odds ratios and 95% confidence intervals (CIS) were calculated using the SUGI MCSTRAT procedure (SAS, 1986) which performs an iterative conditional maximum-likelihood fit of a logistic regression model for case-control studies with matched pairs. For the nutrient analyses, quartiles of serum carotenoids and calorieadjusted nutrient intake were defined from the distribution of the controls. Those in the highest quartile, i.e., quartile 4, served as the comparison group. The minimum number of cases per quartile ranged from 10 to 17. In addition to the exposure variables, adjusted models included the independent contributors to risk of CIN in this sample and known confounders of the disease-exposure relationship. Natural log transformations of non-nutrient continuous variables were used to improve linearity. Tests of trend were conducted to assess doseresponse gradients, i . e . , whether odds ratios increased with decreasing quartile. The p-values for the response gradient were estimated by treating the quartiles of a nutrient as a single ordinal variable in the logistic model.
Ethnic origin
Monthly personal income Ever smoked Current smokers Years of education Age at menarche
18-24 25-3 1 >3 1 Black Hispanic White Other $800
I2 13
1 Years between >I Pap smears Self-reported history of genital herpes genital warts Quetelet index >I623 >2>26 >2630 >30-52 Age at first intercourse < 16 years 1617 years > 17 or never 5 RESULTS
Assessment of independent contributors to risk Table I summarizes the demographic characteristics of cases and controls, as well as the distribution of cases and controls on non-dietary factors associated with CIN. The matching effort achieved concordance between cases and controls for age and ethnic origin. Cases had lower incomes and were more likely to be smokers than controls. A larger number of cases than controls had onset of menarche before age 13, had more than one year between Pap smears, reported a history of genital warts and were in the lowest quartile of Quetelet index. More cases than controls had 5 or more sexual partners, while more controls than cases reported fewer than 3 partners. A greater percentage of cases than of controls had their first full-term pregnancy and delivery before age 19 and reported having used spermicidal foams or gels. Current smoking status, lower monthly personal income, more than 1 year between Pap smears, use of spermicidal foams or gels, self-reported history of genital warts, and lower Quetelet index were independent contributors to risk of CIN when entered simultaneously into a conditional logistic model. All of these variables, as well as the number of sexual partners, were retained in the adjusted models. Because carotenoids and vitamin C occur in many of the same foods, they have the potential to serve as surrogate measures for each other. For this reason and because of previous associations between vitamin C and CIN (Wassertheil-Smoller et al., 1981), intake of vitamin C was evaluated. Controlling for the non-dietary risk factors, OR,s (95% CIS) of 3.7 (0.914.6), 4.1 (1 .o-17.2) and 6.4 (1 A 2 9 . 0 ) were apparent for those in quartiles 3, 2 and 1 of vitamin C consumption, re-
Age at first delivery
19
=
never Number of pregnancies 0 I L
Ever users of
>2 oral contraceptives spermicidal foams
~---
o r PCIP -_
'Due
to
(n = 102)
Conmls (n = 102)
45.1 34.3 20.6 70.6 12.7 15.7 01.0 66.7 23.5 08.8 58.8 52.0 28.4 23.5 48.1 21.6 32.4 22.5 23.5 70.6 29.4 00.0 14.7 38.2 26.5 15.7 19.6 37.3 38.3 24.4 20.6 33.3 46.1 45.5 23.8 30.7 20.6 15.7 26.5 37.3 85.3 32.3
45.1 35.3 19.6 70.6 14.7 10.8 3.9 55.9 19.6 22.6 45.1 33.3 26.5 30.4 43.1 14.7 20.6 40.2 24.5 83.3 15.7 02.9 05.9 23.5 26.5 24.5 25.5 35.3 41.2 23.5 33.3 33.4 33.3 28.4 35.3 36.3 23.5 21.6 30.4 24.5 85.3 19.6
missing data, summed percentages do not always equal 100%.
spectively. The test of trend for increased risk with decreased quartile was statistically significant at p = 0.03. Assessment of the effect of carotenoids Table U presents the quartiles of serum carotenoids for the controls. Except for lycopene, overall serum values in this sample are low compared to those from other studies (Stacewicz-Sapuntzakis et al., 1987; Wald, 1987). Table 111 presents the ORs and 95% CIS by quartile of serum carotenoid. Elevated OR,s are apparent for those in the lowest 2 quartiles of serum cryptoxanthin and lowest quartile of serum alphacarotene. These associations are attenuated in the adjusted model where the approximately 2-fold elevations in the OR,s are not statistically significant and there is little evidence of a trend toward increasing odds with decreasing quartile. Similarly, the more than 2-fold increases in ORs for those in the lowest 3 quartiles of beta-carotene are not statistically significant and a dose gradient is not apparent. Although there appears to be no relationship between serum lycopene and CIN in the crude analysis, the adjusted model reveals elevations in the
36
VANEENWYK ET AL.
TABLE 11 - SERUM LEVELS FOR CAROTENOIDS (@g/dl)IN CONTROLS (N
= 98)
POPIJI.ATION NORMS
CONTRASTED TO
Quartile Carotenoid
Alpha-carotene Beta-carotene Cryptoxanthin Lutein Lycopene
1
2
3
4
0.0- 1 . 1
1.2- 1.7 6.0- 9.6 6.0- 8.6 10.9-13.0 21.4-29.4
1.8- 3.0 9.7-17.3 8.7-14.6 13.1-16.1 29.541.2
3.1- 20.0 17.4-116.8 14.7- 62.8 16.2- 42.6 41.3- 91.2
1 . 0 - 5.9 1 . 0 - 5.9 2.7-10.8 3.4-21.3
Norm'
Percent below norm
3.8 23.7 10.0 19.2 18.8
80.6 87.7 62.2 85.7 18.4
IThese norms are based on the mean levels of carotenoids in a healthy population of 55 middle-aged women (StacewiczSapuntzakis et al.. 1987). The norm for beta-carotene is similar to values found in 4 cohort studies of beta-caroteneand cancer (Wald, 1987).
TABLE 111 - CRUDE (OR,) AND ADJUSTED (OR,)' ODDS RATIOS, 95% CIS AND TESTS OF TREND FOR SERUM CAROTENOIDS
Quanile Carotenoid
4
3
2
1.7 0.7- 4.4 1.5 0.4- 5.3 2.5 1.G 6.2 2.5 0.8- 8.1 2.7 1 . 0 - 7.6 2.6 0.7- 9.4 1.2 0.6- 2.5 1.6 0.6- 4.2 1.9 0.8- 4.4 3.5 1.1-11.5
2.8 1 . 0 - 7.8 1.5 0.4- 5.7 2.2 0.9- 5.6 2.3 0.6- 8.4 3.4 1.2-9.3 2.7 0.7- 9.9 0.4 0.1- 1.0 0.6 0.2- 2.0 1.6 0.7- 3.8 4.7 1.2-17.7
(high)
Alpha-carotene
Beta-carotene
Cryptoxanthin
Lutein
Lycopene
OR, 95% CI ORa 95% CI OR, 95% CI ORa 95% CI OR, 95% CI ORa 95% CI OR, 95% CI ORa 95% CI OR, 95% CI ORa 95% CI
1 .O
1 .o 1 .o 1 .o
1 .o 1 .o 1 .o
1 .o 1 .o
1 .o
Test of vend p-value
1
(low)
-
3.2 1.2- 8.1 2.6 0.7- 9.7 2.4 0.9- 6.7 2.8 0.7-1 1.6 4.5 1.5-13.2 1.9 0.5- 7.9 1.6 0.8- 3.4 1 .o 0.4- 2.6 1.9 0.8- 4.5 3.8 1.1-12.4
0.02 0.15 0.22 0.32
0.01 0.49 0.52 0.54 0.19 0.04
'OR adjusted for current smoking status, monthly personal income bracket. number of sexual partners (lo&-transformed), quartile of dietary vitamin C, frequency of Pap smear (annual vs. less often), ever use of spermicidal contraceptive agents, self-reported history of genital warts, and Quetelet index (lo&-transformed).
ORas which are more than 3-fold for those in the lowest 3 quartiles of serum lycopene. The p-value for the test of trend is statistically significant, even though the observed OR, is lower in the 4th than in the 3rd quartile. Most of the change between the crude and adjusted models for lycopene results from controlling for current smoking status and monthly personal income. ORs (95% CIS) for quartiles 3, 2 and 1 of serum lycopene adjusted for smoking and income are 2.7 (1.1-6.8), 2.9 (1.1-7.8) and 2.6 (1.0-6.6), respectively. After adjustment for smoking and income, control for ever use of spermicidal contraceptive agents and a self-reported history of genital warts account for most of the remaining change. Thus, when these variables are added to a model with quartile of serum lycopene, smoking and income, the ORs (95% CIS)for quartiles 3 , 2 and 1 of serum lycopene are 3.1 (1.1-8.6), 4.6 (14-15.0) and 3.7 (1.3-10.9), respectively. The ORs and 95% CIS for quartiles of carotenoid intake are presented in Table IV. For dietary lycopene, the OR,s are strengthened in the adjusted models, yielding elevations in the ORas which are more than 4-fold for those in the lowest 3 quartiles. Most of the change from the crude to the adjusted ORs is the result of controlling for monthly personal income, ever use of contraceptive spermicidal agents and frequency of Pap smears. With adjustment for these 3 variables, the ORs (95% CIS) for quartiles 3, 2 and 1 of dietary lycopene are 4.1 (1.2-13.9), 6.0 (1.9-18.3) and 5.2 (1.5-17.8), respectively.
As with the serum lycopene measure, the test of trend is statistically significant in spite of an apparent decrease in the OR for those in the 4th quartile. For dietary cryptoxanthin, the elevation in the OR, for those in quartile 2 and the trend for increased risk with decreased quartile are attenuated in the adjusted model. Neither the crude nor the adjusted models reveal associations between CIN and alpha-carotene, beta-carotene, or lutein. Lycopene is responsible for the red color of foods and its presence has been documented in tomatoes, red peppers and ruby-red grapefruit. It is believed that tomatoes and tomato products represent the largest sources of lycopene in the American diet (Bauerfeind, 1981). In this study, ORas for quartiles 3,2, and 1 of intake of fresh tomatoes and tomato juice are 1.3 (0.4-4.2), 2.0 (0.6-7.2) and 2.8 (0.9-8.8), demonstrating a dose-response gradient (p = 0.05). DISCUSSION
Issues of bias While it is unknown how the relatively low response rate may have affected the results, the possibility of bias due to differential response rates of cases and controls was partially addressed by collecting information on age, ethnic origin, zip code and type of payment for medical services for all women
37
CAROTENOIDS A N D CERVICAL DYSPLASIA TABLE IV - CRUDE (OR,)’ AND ADJUSTED (OR,? ODDS RATIOS, 95% CIS AND TESTS OF TREND FOR DIETARY CAROTENOIDS
Quartile ~
Carotenoid
4
2
llow)
2.0 0.8- 5.0 2.0 0 . 6 6.8 I .6 0.7- 3.7 1.7 0 . 6 5.1 1.4 0.5- 3.5 1.3 0.4- 4 . 3 0.8 0.3- 1.9 0.6 0.2- 1.9 2.7 0.9- 8.1 4.6 I . 1-19.7
2.4 I.& 5.5 1.3 0.4- 4.1 1.6 0.7- 3.9 1.2 0.3- 3.9 2.1 1 . 1 - 6.8 1.5 0.5- 4.7 1.6 0.6- 3.8 1.1 0.3- 3.6 3.1 1.4-10.1 5.8 1.621.3
1.8 0.7- 4.6 1.5 0.4- 4 . 9 1.8 0.7- 4 . 6 2.0 0.6- 6 . 5 2.4 I.& 6.0 1.6 0.5- 5 . 4 1.7 0.7- 4 . 2 1.1 0.3- 4 . 0 2.6 0.9- 7 . 6 5.4 1.3-23.3
lhwhb
Alpha-carotene
Beta-carotene
OR, 95% CI OR, 95% CI OR,. 95% ‘cl
OR,
1 .o
.o .o .o
95% CI Cryptoxanthin
OR, 95% CI
OR,
Lutein
Lycopene
95% CI OR, 95% CI OR, 95% CI OR, 95% CI OR, 95% CI
.o .o 1 .o 1 .o 1 .o
1 .o
I
3
Test of trend p-value
0 . I5
0.63 0.24 0.34 0.03 0.48 0.14 0.75
0.05 0.02
‘Adjustment for total caloric intake using linear regression methodology (Willett and Stampfer, 1986) prior to conditional logistic regression.-*Adjusted for total caloric intake as in I; adjustment in the logistic model for current smoking status, monthly personal income bracket. number o f sexual partners (1og;transformed). quartile of dietary vitamin C, frequency o f Pap smear (annual YS. less often), ever use of spermicidal contraceptive agents, self-reponed history of genital warts, and Quetelet index (lo&-rransformed)
who were approached for participation. Response rates were lower L(I < 0.05) among controls than among cases for women aged 18 to 24 years. However, because cases and controls were matched for age, the effect of this discordance on the ORs cannot be estimated. The response rate for women who paid for their medical expenses was lower for controls than for cases (p < 0.05). It is not known in what direction this may have biased the ORs, since controls in both higher (privately insured) and lower (public aid recipients) socio-economic groups may have been over-represented. Adjustment for the nonnutrient factors in the logistic model may have helped to mitigate the effect of bias resulting from the non-random distribution of cases and controls concerning socio-economic factors due to differential response rates. An inherent limitation of case-control studies is the determination of exposure after the onset of disease. This is particularly troublesome in studies of hematologic nutritional status, since nutritional status often changes subsequent to the onset of illness. Since CIN may represent an early stage of the neoplastic process, the potential effect of the disease process on exposure may be attenuated. Nonetheless, the possibility that lower blood levels of nutrients result from CIN cannot be ruled out. While diet may also change after the onset of disease, in this study, women were asked to record their dietary habits for the previous 5 years, making it less likely that the associations between diet and CIN reflect changes subsequent to the onset of illness. Number of sexual partners is considered one of the strongest risk factors for CIN (Brinton er al., 1987; Kelsey and Hildreth, 1983). That neither this variable nor other variables related to sexual factors remained as independent contributors to risk after controlling for smoking and income may indicate that all participants were selected from a high-risk population. If this is the case, caution must be exercised in generalizing the findings of this study to populations at low risk of CIN. Although unlikely, it is possible that information about sexual habits was not accurately reported due to the sensitive nature of these questions. Independent validation of the participants’ responses was not undertaken.
The Pearson product-moment correlation coefficient for quartile of dietary vitamin C and dietary cryptoxanthin is 0.54. The correlation coefficients for quartile of dietary vitamin C and the other carotenoid measures range from 0.007 for serum lycopene to 0.24 for both dietary alpha-carotene and serum cryptoxanthin. Given these relatively low correlations, the inclusion of dietary vitamin C in the fully adjusted model is unlikely to be problematic with respect to collinearity. Analyses performed without the addition of dietary vitamin C do not yield substantively different results. With a sample size of approximately 100 pairs, the power to detect ORs of less than 2.5 is limited. Therefore, the presence of non-statistically significant 2-fold elevations in the OR for some variables may reflect this limitation. Carotenoids In this study, both dietary and serum lycopene manifest a strong inverse association with CIN after adjustment for independent contributors to risk and possible confounders. The concordance of the findings for serum and dietary lycopene and tomatoes, the magnitude of the observed associations, and the observation of a dose-response gradient support the conclusion that higher levels of lycopene or substances contained in lycopene-rich foods protect against CIN. Given the limited number of food sources for this carotenoid, it is unlikely that lycopene represents a surrogate measure for another known nutrient, but it may be a surrogate for unknown protective phytochemicals in tomatoes or other solanaceae. The masking of the association between lycopene and CIN by several confounding variables in the crude analysis is unusual and the reasons are not clear. One might hypothesize that people with lower incomes tend to eat more lycopene-rich foods, such as spaghetti and chili. However, this hypothesis cannot be verified, since no research on the relationship of lycopene status to socio-economic status has been reported. While the relatively high levels of serum lycopene in this sample might substantiate this hypothesis, the relatively younger mean age of the women in this sample compared to the nor-
38
VANEENWYK ET A L
mative sample is likely to account for some of the difference, because serum lycopene levels decline with age (PB). That lycopene or phytochemicals associated with lycopene may be anti-tumorigenic is suggested in 2 nested case-control studies in which lower levels of serum lycopene were observed in those who subsequently developed bladder (Helzlsouer et al., 1989) and pancreatic (Burney et al., 1989) cancer. If lycopene has anti-tumorigenic capacities, this may be due to its function as a quencher of singlet oxygen. Recent research indicates that lycopene’s capacity to quench singlet oxygen is comparable to that of beta-carotene and alpha-tocopherol (DiMascio et al., 1989). That no association was observed for dietary intake of betacarotene is in accord with other well-controlled investigations of CIN (Brock et al., 1988;La Vecchia et al., 1988;Slattery er al., 1990). In contrast, the absence of an association between serum beta-carotene and CIN reported here is not consistent with 2 studies which noted inverse associations between CIN and serum (Harris et al., 1986)and plasma (Brock et al., 1988)beta-carotene. Two explanations for this discrepancy are apparent. Firstly, if there is a threshold effect for the protective action of particular nutrients in anti-tumorigenesis, the low levels of serum carotenoids other than lycopene in this study sample may have impeded the ability to demonstrate this effect. In both of the other studies, the levels of beta-carotene for the controls were substantially higher than in this sample. Secondly, the inclusion in the adjusted model of all the independent contributors to risk in this sample may have yielded conservative ORs, because some of these factors may be steps in a causal chain or may be associated with disease through a correlation with exposure. For example, if poor nutrition leads to impaired immune status and increased susceptibility to viral infection, a history of genital warts may be a step in a causal chain. Obtaining an annual Pap smear may be associated with good nutritional habits, because both behaviors involve being aware of and following medical advice. To allow for comparison with the other 2 investigations, models were constructed to be as similar as possible to those
reported. When this is done, comparable findings emerge. Specifically, with adjustment for current smoking status, income, number of sexual partners, and ever use of oral contraceptives, the ORs (95% CIS) for quartiles 3,2,and I of beta-carotene are 3.3 (1.2-8.7),3.1 (1.1-8.8)and 2.7(0.9-8.1).These ORs are similar to those of Hams et al. (1986)whose estimates range from 1.7 to 3.7 for the lowest 4 quintiles of serum betacarotene with no dose-response relationship. Brock et al. (1988)noted an OR of 5.0 (95% CI, 1.0-10.0) for those in the lowest quartile of plasma beta-carotene compared to the highest, as well as a strong dose-response pattern @ = 0.02). In this partially adjusted model, inverse associations between CIN and both alpha-carotene and cryptoxanthin are suggested with ORs of 1.8 (0.7-5.1), 3.1 (1.0-9.9)and 2.9 (1.17.8)for quartiles 3,2,and 1 of serum alpha-carotene; and ORs of 3 .O ( 1 .o-9.0),3.2( 1.1-9.3)and 3.8 (1.1-12.4)for quartiles 3, 2, and 1 of serum cryptoxanthin. These findings suggest that lycopene and lycopenecontaining foods deserve further scrutiny as agents for the prevention of CIN. The combined evidence from this and other studies suggests a 2-to 3-fold effect for beta-carotene and CIN. While this investigation does not indicate an association between lutein and CIN, the findings concerning alpha-carotene and cryptoxanthin remain inconclusive. Since CIN can progress to invasive cervical cancer, these findings may have implications for the prevention of cervical neoplasia, as well as other epithelial cell cancers. ACKNOWLEDGEMENTS
The authors acknowledge the support of Dr. S. Gall, University of Illinois Hospital, Dr. M. Makii, Cook County Hospital, and Dr. R . Murphy, Illinois Department of Public Health. They appreciated the advice and comments of Drs. J. Kelsey, Columbia University, and W. Haenszel, J. Goldberg, V. Persky and M. Stacewicz-Sapuntzakis, University of Illinois, Chicago. This work was funded in part by a grant from the American Cancer Society, Illinois Division, and by the State Cancer Plan of the Illinois Cancer Council.
REFERENCES
BAUERNFEIND, J.C., Carotenoids as colorants and vitamin A precursors: technological and nutritional applications. Academic Press, New York ( 1981). BENDICH, A., Carotenoids and the immune response. J. Nutr., 119, 112115 (1989).
LA VECCHIA, C . , DECARLI. A , , FASOLI,M., PARAZZINI, F., FRANCESCHI, S., GENTILE,A. and NEGRI,E., Dietary vitamir. A and the risk of intraepithelial and invasive cervical neoplasia. Gynec. Oncol.. 30, 187-195 ( 1988). OLSON,J.A., Provitamin A function of carotenoids: the conversion of BLOCK,G . , COYLE,L., SMUCKER, R., HARLAN, L., HARTMAN, A. and beta-carotene into vitamin A. J. Nutr., 119, 105-108 (1988). SAS INSTITUTE, SUGI Supplemental Library User’s Guide, 5th ed., pp. KESSLER,L.. Health habits and history questionnaire: diet history and 307-328, SAS Institute, C a y (1986). other risk factors. Personal computer system packet. National Cancer Institute, Division of Cancer Prevention and Control, Bethesda (1987-1989). SLATTERY, M.L., ABBOTT,T.M., OVERALL,J.C., ROBISON,L.M., J.W. and WEST, D.W., Dietary BRINTON, L.A., TASHIMA, K.T., LEHMAN, H.F., LEVINE,R.S., MALLIN, FRENCH.T.K., JOLLES,C., GARDNER, K. and SAVITZ,D.A., Epidemiology of cervical cancer by cell type. vitamins A, C, and E and selenium as risk factors for cervical cancer. Epidemiology, 1, 8- 15 ( 1990). Cancer Res.. 47, 17061711 (1987). M.,BOWEN,P.E., KIKENDALL, J.W. and BROCK,K., BERRY,G., MOCK, P.A., TRUSWELL, L.A. and BRINTON, STACEWICZ-SAPUNTZAKIS, L.A., Nutrients in the diet and plasma and risk of in situ cervical cancer. BURGESS,M., Simultaneous determination of serum retinol and various carotenoids: their distribution in middle-aged men and women. J . MicroJ . nat. Cancer fnst., 80, 58Cb-585 (1988). nutrient Anal., 3, 2745 (1987). BURNEY, P.G.C., COMSTOCK, G.W. and MORRIS,J.S., Serum micronutrients and subsequent risk of pancreatic cancer. Amer. J. d i n . Nutr., 49, WALD,N., Retinol, beta-carotene and cancer. Cancer Surv., 6 , 63-51 (1987). 895-900 (1989). WASSERTHEIL-SMOLLER, ROMNEY, S.L., WYLIE-ROSETT, J., SLACLE, DIMASCIO.P.. KAISER.S. and SIES. H.. Lvcooene as the most efficient S., MILLER,G., LUCIDO,S., D., DUTTAGUPTA, C . and PALAN,P.R., Dietary biological carotenoid singlet oxygen quench&. hrch. Biochem. Biophys., vitamin C and uterine cervical dysplasia. Amer. J. Epidemiol.. 114, 7 1 4 274, 532-538 (1989). 724 (1981). HARRIS,R.W.C., FORMAN,D., DOLL, R., VESSEY,M.P. and WALD, WILLETT,W.C. and STAMPFER, M.J., Total energy intake: implications N.J., Cancer of the cervix uteri and vitamin A. Brit. 1. Cancer, 53, for epidemiologic analyses. Amer. J. Epidemiol.. 124, 17-27 (1986). 653-659 (1986). WYLIE-ROSETT, J.A., ROMNEY,S.L., SLAGLE,N.S., WASSERTHEILHELZLSOUER, K.J., COMSTOCK,G. and MORRIS,J.S., Selenium, lyco- SMOLLER,S., MILLER,G.L., PALAN,P.R., LUCIDO,D.J. and DUTpene, alpha-tocopherol, beta-carotene, retinol, and subsequent bladder TAGUFTA, c . , Influence of vitamin A on cervical dysplasia and carcinoma cancer. Cancer Res., 49, 6144-6148 (1989). in situ. Nutr. Cancer, 6 , 49-57 (1984). KELSEY,J.L. and HILDRETH,N.G., Breast and gynecologic cancer and ZEIGLER,R.,A review of epidemiologic evidence that carotenoids reduce epidemio/ogy. CRC Press, Boca Raton (1983). the risk of cancer. J . Nutr.. 119, I16122 (1989).
Publication of the International Union Against Cancer Publication de I'Union lnternationale Contre 1e Cancer
DIETARY AND SERUM CAROTENOIDS AND CERVICAL INTRAEPITHELIAL NEOPLASIA Juliet VANEENWYK'.3*4,Faith G . DAVIS'and Phyllis E. BOW EN^ 'Department of Epidemiology and Biostatistics, School of Public Health, University of Illinois at Chicago; and 2Department of Nutrition and Medical Dietetics, College of Associated Health Professions, University of Illinois at Chicago, Chicago, IL, USA. A case-control study examined the association between cervical intra-epithelial neoplasia (CIN) and serum and dietary alpha-carotene, beta-carotene, cryptoxanthin, lutein, and lycopene. Cases (n = 102) had biopsy confirmed C I N I, II or 111. Controls matched for age. ethnic origin and clinic (n = 102) had normal Pap smears. Participants completed health history and food frequency questionnaires. Fasting venous blood samples were assayed for serum carotenoids. Multivariable conditional logistic regression analyses yielded odds ratios and 95% confidence intervals (CIS) for those in quartiles 3.2, and I (lowest compared to quartile 4 (highest) of serum lycopene of 3.5 ( .1-11.5), 4.7 (1.2-17.7) and 3.8 (l.l-l2.4), respectively. Similar analyses yielded adjusted odds ratios ( 0 4 s ) and 95% CIS Of 4.6 (l.l-l9.7), 5.8 (1.621.3) and 5.4 (I .3-23.3) for dietary intake of lycopene. The findings for lycopene-rich foods (tomatoes) were consistent with this result. C I N was not associated with the lutein. Findings for alphacarotene, beta-carotene and cryptoxanthin were ambiguous. Quartile of vitamin C intake was also inversely associated with CIN with 0 4 s and 95% CIS of 3.7 (0.9-14.6), 4.1 (I.& 17.2). and 6.4 (I .4-30.0) for those in quartiles 3.2. and I compared to quartile 4.
I
Beta-carotene represents 1 of approximately 600 known carotenoids. Its activity with respect to cancer prevention has received attention, because it is one of the major provitamin A carotenoids (Olson, 1988), it functions as an antioxidant, and it enhances the immune response (Bendich, 1989). Epidemiologic evidence has shown increased consumption and/or serum levels of beta-carotene to be protective against many types of cancer (Wald, 1987; Zeigler, 1989). Until recently, the ability to assess intake and serum levels of other carotenoids has been limited. However, due to recent advances in high-pressure liquid chromatography procedures (Stacewicz-Sapuntzakis et al., 1987), it is now possible to measure the relationship between cancer and additional serum carotenoids, including alphacarotene, cryptoxanthin, lutein and lycopene. Concurrently, nutrient data-bases have been expanded to include these carotenoids. While it is expected that these data-bases may need to be refined as field validation studies become available, the data-base developed by Block et al. (1987-1989), offers a useful tool for estimating dietary intake of individual carotenoids. A case-control study of cervical intra-epithelial neoplasia (CIN) was designed to investigate the potential protective effect of several nutrients, including carotenoids. This premalignant condition was chosen to mitigate physiologic changes which may ensue from, rather than foreshadow, neoplasia. Previous studies of CIN, including those of carcinoma in situ, have shown no association between dietary intake of beta-carotene and disease (Wylie-Rosett et al., 1984; Brock et al., 1988; La Vecchia et al., 1988; Slattery et al., 1990) after controlling for other risk factors. In contrast, inverse associations have been noted for serum beta-carotene and disease (Hanis et al., 1986; Brock et al., 1988). MATERIAL AND METHODS
Selection of cases and controls
Cases were identified from non-pregnant, non-post-partum women with biopsy-confirmed diagnoses of CIN I, I1 or 111. Non-pregnant, non-post-partum women who attended the same clinics as the cases and who had a Pap smear as part of their
visit were identified as potential controls. These women became eligible to participate if their Pap smear showed no abnormalities of a severity greater than or equal to those of benign atypia. The controls were matched to the cases by clinic, age (18-24, 25-31, 32-38, 3 9 4 9 ) and ethnicity (black, nonblack). Recruiting took place at the University of Illinois Hospital Clinics and the Cook County Hospital Clinics in Chicago, IL between April 1987 and June 1989. Of 166 potential cases, 102 women participated in the project, yielding a volunteer rate of approximately 61%. To enroll an equal number of controls, study staff approached 195 women, obtaining a volunteer rate of approximately 52%. Informed consent was obtained from all participants, following the requirements of the participating institutions. Measurement of exposure
Participants returned to the clinic for venipuncture following a 10-hr fast. The blood samples were processed and frozen following the specifications of the laboratory (PB) contracted to assay the serum carotenoids. The assays were conducted using high-pressure liquid chromatography techniques (Stacewicz-Sapuntzakis et al., 1987). All laboratory personnel were blind to the case or control status of the blood samples. A failure to withdraw blood from four women resulted in laboratory data from 98 matched pairs. Participants completed the food frequency section (FFQ) of the Health Habits and History Questionnaire provided by the National Cancer Institute, Division of Cancer Prevention and Control (NCI-DCPC), version 2.1 (Block et al., 1987-1989). This questionnaire asks about the frequency of consumption of 99 food and beverage items and provides an open-ended question to record additional foods consumed at least once a week. Forms were self-administered and available in English and Spanish. A trained interviewer, blind to the study hypotheses and case or control status, questioned the participants about responses which were incomplete, internally inconsistent or seemed to contain recording errors. If necessary, responses were changed subsequent to this query. One woman failed to complete the FFQ and one woman was incapable of answering the questions even with help from the interviewer. These FFQs were eliminated from the analysis, resulting in dietary data for 100 matched pairs. The conversion of foods to nutrients was accomplished using NCI-DCPC software, version 2.2, August 1989 (Block et al., 1987-1989). To control for over- and under-reporting of actual intake, adjustment for calories was made using the regression procedure of Willett and Stampfer (1986). Because there is a lack of consensus concerning the use of Willett and Stampfer's regression procedure, the relationship of dietary carotenoids 3Currently at lllinois Department of Public Health, Office of Health Policy and Planning.
4T0whom correspondence and reprint requests should be sent, at the Illinois Department of Public Health, Suite 6-600, 100 W. Randolph, Chicago, 1L 60601, USA. Received; October 4, 1990 and in revised form December 19, 1990.
35
CAROTENOIDS AND CERVICAL DYSPLASIA
and CIN was further assessed without adjustment for total energy intake. The results from this analysis did not differ substantively from those reported.
TABLE I - PERCENTAGE DISTRIBUTIONS' OF CASES AND CONTROLS (UNMATCHED) ON DEMOGRAPHIC, MEDICAL. SMOKING, SEXUAL AND REPRODUCTIVE VARIABLES Cases
Variable
Measurement of potential confounders A self-administered questionnaire was used to obtain information about the participants' background, health and pregnancy history, and smoking and sexual habits. These data were used to assess independent contributors to risk in this sample and to control for potential confounders of the diseaseexposure association. In addition to using the FFQ to measure dietary intake of carotenoids, it was used to assess dietary intake of nutrients which were potential independent contributors to risk andlor confounders of the carotenoid-CIN association. The relationships of dietary vitamins A, C and E to CIN were assessed for this purpose. Statistical analysis Crude (OR,) and adjusted (OR,) odds ratios and 95% confidence intervals (CIS) were calculated using the SUGI MCSTRAT procedure (SAS, 1986) which performs an iterative conditional maximum-likelihood fit of a logistic regression model for case-control studies with matched pairs. For the nutrient analyses, quartiles of serum carotenoids and calorieadjusted nutrient intake were defined from the distribution of the controls. Those in the highest quartile, i.e., quartile 4, served as the comparison group. The minimum number of cases per quartile ranged from 10 to 17. In addition to the exposure variables, adjusted models included the independent contributors to risk of CIN in this sample and known confounders of the disease-exposure relationship. Natural log transformations of non-nutrient continuous variables were used to improve linearity. Tests of trend were conducted to assess doseresponse gradients, i . e . , whether odds ratios increased with decreasing quartile. The p-values for the response gradient were estimated by treating the quartiles of a nutrient as a single ordinal variable in the logistic model.
Ethnic origin
Monthly personal income Ever smoked Current smokers Years of education Age at menarche
18-24 25-3 1 >3 1 Black Hispanic White Other $800
I2 13
1 Years between >I Pap smears Self-reported history of genital herpes genital warts Quetelet index >I623 >2>26 >2630 >30-52 Age at first intercourse < 16 years 1617 years > 17 or never 5 RESULTS
Assessment of independent contributors to risk Table I summarizes the demographic characteristics of cases and controls, as well as the distribution of cases and controls on non-dietary factors associated with CIN. The matching effort achieved concordance between cases and controls for age and ethnic origin. Cases had lower incomes and were more likely to be smokers than controls. A larger number of cases than controls had onset of menarche before age 13, had more than one year between Pap smears, reported a history of genital warts and were in the lowest quartile of Quetelet index. More cases than controls had 5 or more sexual partners, while more controls than cases reported fewer than 3 partners. A greater percentage of cases than of controls had their first full-term pregnancy and delivery before age 19 and reported having used spermicidal foams or gels. Current smoking status, lower monthly personal income, more than 1 year between Pap smears, use of spermicidal foams or gels, self-reported history of genital warts, and lower Quetelet index were independent contributors to risk of CIN when entered simultaneously into a conditional logistic model. All of these variables, as well as the number of sexual partners, were retained in the adjusted models. Because carotenoids and vitamin C occur in many of the same foods, they have the potential to serve as surrogate measures for each other. For this reason and because of previous associations between vitamin C and CIN (Wassertheil-Smoller et al., 1981), intake of vitamin C was evaluated. Controlling for the non-dietary risk factors, OR,s (95% CIS) of 3.7 (0.914.6), 4.1 (1 .o-17.2) and 6.4 (1 A 2 9 . 0 ) were apparent for those in quartiles 3, 2 and 1 of vitamin C consumption, re-
Age at first delivery
19
=
never Number of pregnancies 0 I L
Ever users of
>2 oral contraceptives spermicidal foams
~---
o r PCIP -_
'Due
to
(n = 102)
Conmls (n = 102)
45.1 34.3 20.6 70.6 12.7 15.7 01.0 66.7 23.5 08.8 58.8 52.0 28.4 23.5 48.1 21.6 32.4 22.5 23.5 70.6 29.4 00.0 14.7 38.2 26.5 15.7 19.6 37.3 38.3 24.4 20.6 33.3 46.1 45.5 23.8 30.7 20.6 15.7 26.5 37.3 85.3 32.3
45.1 35.3 19.6 70.6 14.7 10.8 3.9 55.9 19.6 22.6 45.1 33.3 26.5 30.4 43.1 14.7 20.6 40.2 24.5 83.3 15.7 02.9 05.9 23.5 26.5 24.5 25.5 35.3 41.2 23.5 33.3 33.4 33.3 28.4 35.3 36.3 23.5 21.6 30.4 24.5 85.3 19.6
missing data, summed percentages do not always equal 100%.
spectively. The test of trend for increased risk with decreased quartile was statistically significant at p = 0.03. Assessment of the effect of carotenoids Table U presents the quartiles of serum carotenoids for the controls. Except for lycopene, overall serum values in this sample are low compared to those from other studies (Stacewicz-Sapuntzakis et al., 1987; Wald, 1987). Table 111 presents the ORs and 95% CIS by quartile of serum carotenoid. Elevated OR,s are apparent for those in the lowest 2 quartiles of serum cryptoxanthin and lowest quartile of serum alphacarotene. These associations are attenuated in the adjusted model where the approximately 2-fold elevations in the OR,s are not statistically significant and there is little evidence of a trend toward increasing odds with decreasing quartile. Similarly, the more than 2-fold increases in ORs for those in the lowest 3 quartiles of beta-carotene are not statistically significant and a dose gradient is not apparent. Although there appears to be no relationship between serum lycopene and CIN in the crude analysis, the adjusted model reveals elevations in the
36
VANEENWYK ET AL.
TABLE 11 - SERUM LEVELS FOR CAROTENOIDS (@g/dl)IN CONTROLS (N
= 98)
POPIJI.ATION NORMS
CONTRASTED TO
Quartile Carotenoid
Alpha-carotene Beta-carotene Cryptoxanthin Lutein Lycopene
1
2
3
4
0.0- 1 . 1
1.2- 1.7 6.0- 9.6 6.0- 8.6 10.9-13.0 21.4-29.4
1.8- 3.0 9.7-17.3 8.7-14.6 13.1-16.1 29.541.2
3.1- 20.0 17.4-116.8 14.7- 62.8 16.2- 42.6 41.3- 91.2
1 . 0 - 5.9 1 . 0 - 5.9 2.7-10.8 3.4-21.3
Norm'
Percent below norm
3.8 23.7 10.0 19.2 18.8
80.6 87.7 62.2 85.7 18.4
IThese norms are based on the mean levels of carotenoids in a healthy population of 55 middle-aged women (StacewiczSapuntzakis et al.. 1987). The norm for beta-carotene is similar to values found in 4 cohort studies of beta-caroteneand cancer (Wald, 1987).
TABLE 111 - CRUDE (OR,) AND ADJUSTED (OR,)' ODDS RATIOS, 95% CIS AND TESTS OF TREND FOR SERUM CAROTENOIDS
Quanile Carotenoid
4
3
2
1.7 0.7- 4.4 1.5 0.4- 5.3 2.5 1.G 6.2 2.5 0.8- 8.1 2.7 1 . 0 - 7.6 2.6 0.7- 9.4 1.2 0.6- 2.5 1.6 0.6- 4.2 1.9 0.8- 4.4 3.5 1.1-11.5
2.8 1 . 0 - 7.8 1.5 0.4- 5.7 2.2 0.9- 5.6 2.3 0.6- 8.4 3.4 1.2-9.3 2.7 0.7- 9.9 0.4 0.1- 1.0 0.6 0.2- 2.0 1.6 0.7- 3.8 4.7 1.2-17.7
(high)
Alpha-carotene
Beta-carotene
Cryptoxanthin
Lutein
Lycopene
OR, 95% CI ORa 95% CI OR, 95% CI ORa 95% CI OR, 95% CI ORa 95% CI OR, 95% CI ORa 95% CI OR, 95% CI ORa 95% CI
1 .O
1 .o 1 .o 1 .o
1 .o 1 .o 1 .o
1 .o 1 .o
1 .o
Test of vend p-value
1
(low)
-
3.2 1.2- 8.1 2.6 0.7- 9.7 2.4 0.9- 6.7 2.8 0.7-1 1.6 4.5 1.5-13.2 1.9 0.5- 7.9 1.6 0.8- 3.4 1 .o 0.4- 2.6 1.9 0.8- 4.5 3.8 1.1-12.4
0.02 0.15 0.22 0.32
0.01 0.49 0.52 0.54 0.19 0.04
'OR adjusted for current smoking status, monthly personal income bracket. number of sexual partners (lo&-transformed), quartile of dietary vitamin C, frequency of Pap smear (annual vs. less often), ever use of spermicidal contraceptive agents, self-reported history of genital warts, and Quetelet index (lo&-transformed).
ORas which are more than 3-fold for those in the lowest 3 quartiles of serum lycopene. The p-value for the test of trend is statistically significant, even though the observed OR, is lower in the 4th than in the 3rd quartile. Most of the change between the crude and adjusted models for lycopene results from controlling for current smoking status and monthly personal income. ORs (95% CIS) for quartiles 3, 2 and 1 of serum lycopene adjusted for smoking and income are 2.7 (1.1-6.8), 2.9 (1.1-7.8) and 2.6 (1.0-6.6), respectively. After adjustment for smoking and income, control for ever use of spermicidal contraceptive agents and a self-reported history of genital warts account for most of the remaining change. Thus, when these variables are added to a model with quartile of serum lycopene, smoking and income, the ORs (95% CIS)for quartiles 3 , 2 and 1 of serum lycopene are 3.1 (1.1-8.6), 4.6 (14-15.0) and 3.7 (1.3-10.9), respectively. The ORs and 95% CIS for quartiles of carotenoid intake are presented in Table IV. For dietary lycopene, the OR,s are strengthened in the adjusted models, yielding elevations in the ORas which are more than 4-fold for those in the lowest 3 quartiles. Most of the change from the crude to the adjusted ORs is the result of controlling for monthly personal income, ever use of contraceptive spermicidal agents and frequency of Pap smears. With adjustment for these 3 variables, the ORs (95% CIS) for quartiles 3, 2 and 1 of dietary lycopene are 4.1 (1.2-13.9), 6.0 (1.9-18.3) and 5.2 (1.5-17.8), respectively.
As with the serum lycopene measure, the test of trend is statistically significant in spite of an apparent decrease in the OR for those in the 4th quartile. For dietary cryptoxanthin, the elevation in the OR, for those in quartile 2 and the trend for increased risk with decreased quartile are attenuated in the adjusted model. Neither the crude nor the adjusted models reveal associations between CIN and alpha-carotene, beta-carotene, or lutein. Lycopene is responsible for the red color of foods and its presence has been documented in tomatoes, red peppers and ruby-red grapefruit. It is believed that tomatoes and tomato products represent the largest sources of lycopene in the American diet (Bauerfeind, 1981). In this study, ORas for quartiles 3,2, and 1 of intake of fresh tomatoes and tomato juice are 1.3 (0.4-4.2), 2.0 (0.6-7.2) and 2.8 (0.9-8.8), demonstrating a dose-response gradient (p = 0.05). DISCUSSION
Issues of bias While it is unknown how the relatively low response rate may have affected the results, the possibility of bias due to differential response rates of cases and controls was partially addressed by collecting information on age, ethnic origin, zip code and type of payment for medical services for all women
37
CAROTENOIDS A N D CERVICAL DYSPLASIA TABLE IV - CRUDE (OR,)’ AND ADJUSTED (OR,? ODDS RATIOS, 95% CIS AND TESTS OF TREND FOR DIETARY CAROTENOIDS
Quartile ~
Carotenoid
4
2
llow)
2.0 0.8- 5.0 2.0 0 . 6 6.8 I .6 0.7- 3.7 1.7 0 . 6 5.1 1.4 0.5- 3.5 1.3 0.4- 4 . 3 0.8 0.3- 1.9 0.6 0.2- 1.9 2.7 0.9- 8.1 4.6 I . 1-19.7
2.4 I.& 5.5 1.3 0.4- 4.1 1.6 0.7- 3.9 1.2 0.3- 3.9 2.1 1 . 1 - 6.8 1.5 0.5- 4.7 1.6 0.6- 3.8 1.1 0.3- 3.6 3.1 1.4-10.1 5.8 1.621.3
1.8 0.7- 4.6 1.5 0.4- 4 . 9 1.8 0.7- 4 . 6 2.0 0.6- 6 . 5 2.4 I.& 6.0 1.6 0.5- 5 . 4 1.7 0.7- 4 . 2 1.1 0.3- 4 . 0 2.6 0.9- 7 . 6 5.4 1.3-23.3
lhwhb
Alpha-carotene
Beta-carotene
OR, 95% CI OR, 95% CI OR,. 95% ‘cl
OR,
1 .o
.o .o .o
95% CI Cryptoxanthin
OR, 95% CI
OR,
Lutein
Lycopene
95% CI OR, 95% CI OR, 95% CI OR, 95% CI OR, 95% CI
.o .o 1 .o 1 .o 1 .o
1 .o
I
3
Test of trend p-value
0 . I5
0.63 0.24 0.34 0.03 0.48 0.14 0.75
0.05 0.02
‘Adjustment for total caloric intake using linear regression methodology (Willett and Stampfer, 1986) prior to conditional logistic regression.-*Adjusted for total caloric intake as in I; adjustment in the logistic model for current smoking status, monthly personal income bracket. number o f sexual partners (1og;transformed). quartile of dietary vitamin C, frequency o f Pap smear (annual YS. less often), ever use of spermicidal contraceptive agents, self-reponed history of genital warts, and Quetelet index (lo&-rransformed)
who were approached for participation. Response rates were lower L(I < 0.05) among controls than among cases for women aged 18 to 24 years. However, because cases and controls were matched for age, the effect of this discordance on the ORs cannot be estimated. The response rate for women who paid for their medical expenses was lower for controls than for cases (p < 0.05). It is not known in what direction this may have biased the ORs, since controls in both higher (privately insured) and lower (public aid recipients) socio-economic groups may have been over-represented. Adjustment for the nonnutrient factors in the logistic model may have helped to mitigate the effect of bias resulting from the non-random distribution of cases and controls concerning socio-economic factors due to differential response rates. An inherent limitation of case-control studies is the determination of exposure after the onset of disease. This is particularly troublesome in studies of hematologic nutritional status, since nutritional status often changes subsequent to the onset of illness. Since CIN may represent an early stage of the neoplastic process, the potential effect of the disease process on exposure may be attenuated. Nonetheless, the possibility that lower blood levels of nutrients result from CIN cannot be ruled out. While diet may also change after the onset of disease, in this study, women were asked to record their dietary habits for the previous 5 years, making it less likely that the associations between diet and CIN reflect changes subsequent to the onset of illness. Number of sexual partners is considered one of the strongest risk factors for CIN (Brinton er al., 1987; Kelsey and Hildreth, 1983). That neither this variable nor other variables related to sexual factors remained as independent contributors to risk after controlling for smoking and income may indicate that all participants were selected from a high-risk population. If this is the case, caution must be exercised in generalizing the findings of this study to populations at low risk of CIN. Although unlikely, it is possible that information about sexual habits was not accurately reported due to the sensitive nature of these questions. Independent validation of the participants’ responses was not undertaken.
The Pearson product-moment correlation coefficient for quartile of dietary vitamin C and dietary cryptoxanthin is 0.54. The correlation coefficients for quartile of dietary vitamin C and the other carotenoid measures range from 0.007 for serum lycopene to 0.24 for both dietary alpha-carotene and serum cryptoxanthin. Given these relatively low correlations, the inclusion of dietary vitamin C in the fully adjusted model is unlikely to be problematic with respect to collinearity. Analyses performed without the addition of dietary vitamin C do not yield substantively different results. With a sample size of approximately 100 pairs, the power to detect ORs of less than 2.5 is limited. Therefore, the presence of non-statistically significant 2-fold elevations in the OR for some variables may reflect this limitation. Carotenoids In this study, both dietary and serum lycopene manifest a strong inverse association with CIN after adjustment for independent contributors to risk and possible confounders. The concordance of the findings for serum and dietary lycopene and tomatoes, the magnitude of the observed associations, and the observation of a dose-response gradient support the conclusion that higher levels of lycopene or substances contained in lycopene-rich foods protect against CIN. Given the limited number of food sources for this carotenoid, it is unlikely that lycopene represents a surrogate measure for another known nutrient, but it may be a surrogate for unknown protective phytochemicals in tomatoes or other solanaceae. The masking of the association between lycopene and CIN by several confounding variables in the crude analysis is unusual and the reasons are not clear. One might hypothesize that people with lower incomes tend to eat more lycopene-rich foods, such as spaghetti and chili. However, this hypothesis cannot be verified, since no research on the relationship of lycopene status to socio-economic status has been reported. While the relatively high levels of serum lycopene in this sample might substantiate this hypothesis, the relatively younger mean age of the women in this sample compared to the nor-
38
VANEENWYK ET A L
mative sample is likely to account for some of the difference, because serum lycopene levels decline with age (PB). That lycopene or phytochemicals associated with lycopene may be anti-tumorigenic is suggested in 2 nested case-control studies in which lower levels of serum lycopene were observed in those who subsequently developed bladder (Helzlsouer et al., 1989) and pancreatic (Burney et al., 1989) cancer. If lycopene has anti-tumorigenic capacities, this may be due to its function as a quencher of singlet oxygen. Recent research indicates that lycopene’s capacity to quench singlet oxygen is comparable to that of beta-carotene and alpha-tocopherol (DiMascio et al., 1989). That no association was observed for dietary intake of betacarotene is in accord with other well-controlled investigations of CIN (Brock et al., 1988;La Vecchia et al., 1988;Slattery er al., 1990). In contrast, the absence of an association between serum beta-carotene and CIN reported here is not consistent with 2 studies which noted inverse associations between CIN and serum (Harris et al., 1986)and plasma (Brock et al., 1988)beta-carotene. Two explanations for this discrepancy are apparent. Firstly, if there is a threshold effect for the protective action of particular nutrients in anti-tumorigenesis, the low levels of serum carotenoids other than lycopene in this study sample may have impeded the ability to demonstrate this effect. In both of the other studies, the levels of beta-carotene for the controls were substantially higher than in this sample. Secondly, the inclusion in the adjusted model of all the independent contributors to risk in this sample may have yielded conservative ORs, because some of these factors may be steps in a causal chain or may be associated with disease through a correlation with exposure. For example, if poor nutrition leads to impaired immune status and increased susceptibility to viral infection, a history of genital warts may be a step in a causal chain. Obtaining an annual Pap smear may be associated with good nutritional habits, because both behaviors involve being aware of and following medical advice. To allow for comparison with the other 2 investigations, models were constructed to be as similar as possible to those
reported. When this is done, comparable findings emerge. Specifically, with adjustment for current smoking status, income, number of sexual partners, and ever use of oral contraceptives, the ORs (95% CIS) for quartiles 3,2,and I of beta-carotene are 3.3 (1.2-8.7),3.1 (1.1-8.8)and 2.7(0.9-8.1).These ORs are similar to those of Hams et al. (1986)whose estimates range from 1.7 to 3.7 for the lowest 4 quintiles of serum betacarotene with no dose-response relationship. Brock et al. (1988)noted an OR of 5.0 (95% CI, 1.0-10.0) for those in the lowest quartile of plasma beta-carotene compared to the highest, as well as a strong dose-response pattern @ = 0.02). In this partially adjusted model, inverse associations between CIN and both alpha-carotene and cryptoxanthin are suggested with ORs of 1.8 (0.7-5.1), 3.1 (1.0-9.9)and 2.9 (1.17.8)for quartiles 3,2,and 1 of serum alpha-carotene; and ORs of 3 .O ( 1 .o-9.0),3.2( 1.1-9.3)and 3.8 (1.1-12.4)for quartiles 3, 2, and 1 of serum cryptoxanthin. These findings suggest that lycopene and lycopenecontaining foods deserve further scrutiny as agents for the prevention of CIN. The combined evidence from this and other studies suggests a 2-to 3-fold effect for beta-carotene and CIN. While this investigation does not indicate an association between lutein and CIN, the findings concerning alpha-carotene and cryptoxanthin remain inconclusive. Since CIN can progress to invasive cervical cancer, these findings may have implications for the prevention of cervical neoplasia, as well as other epithelial cell cancers. ACKNOWLEDGEMENTS
The authors acknowledge the support of Dr. S. Gall, University of Illinois Hospital, Dr. M. Makii, Cook County Hospital, and Dr. R . Murphy, Illinois Department of Public Health. They appreciated the advice and comments of Drs. J. Kelsey, Columbia University, and W. Haenszel, J. Goldberg, V. Persky and M. Stacewicz-Sapuntzakis, University of Illinois, Chicago. This work was funded in part by a grant from the American Cancer Society, Illinois Division, and by the State Cancer Plan of the Illinois Cancer Council.
REFERENCES
BAUERNFEIND, J.C., Carotenoids as colorants and vitamin A precursors: technological and nutritional applications. Academic Press, New York ( 1981). BENDICH, A., Carotenoids and the immune response. J. Nutr., 119, 112115 (1989).
LA VECCHIA, C . , DECARLI. A , , FASOLI,M., PARAZZINI, F., FRANCESCHI, S., GENTILE,A. and NEGRI,E., Dietary vitamir. A and the risk of intraepithelial and invasive cervical neoplasia. Gynec. Oncol.. 30, 187-195 ( 1988). OLSON,J.A., Provitamin A function of carotenoids: the conversion of BLOCK,G . , COYLE,L., SMUCKER, R., HARLAN, L., HARTMAN, A. and beta-carotene into vitamin A. J. Nutr., 119, 105-108 (1988). SAS INSTITUTE, SUGI Supplemental Library User’s Guide, 5th ed., pp. KESSLER,L.. Health habits and history questionnaire: diet history and 307-328, SAS Institute, C a y (1986). other risk factors. Personal computer system packet. National Cancer Institute, Division of Cancer Prevention and Control, Bethesda (1987-1989). SLATTERY, M.L., ABBOTT,T.M., OVERALL,J.C., ROBISON,L.M., J.W. and WEST, D.W., Dietary BRINTON, L.A., TASHIMA, K.T., LEHMAN, H.F., LEVINE,R.S., MALLIN, FRENCH.T.K., JOLLES,C., GARDNER, K. and SAVITZ,D.A., Epidemiology of cervical cancer by cell type. vitamins A, C, and E and selenium as risk factors for cervical cancer. Epidemiology, 1, 8- 15 ( 1990). Cancer Res.. 47, 17061711 (1987). M.,BOWEN,P.E., KIKENDALL, J.W. and BROCK,K., BERRY,G., MOCK, P.A., TRUSWELL, L.A. and BRINTON, STACEWICZ-SAPUNTZAKIS, L.A., Nutrients in the diet and plasma and risk of in situ cervical cancer. BURGESS,M., Simultaneous determination of serum retinol and various carotenoids: their distribution in middle-aged men and women. J . MicroJ . nat. Cancer fnst., 80, 58Cb-585 (1988). nutrient Anal., 3, 2745 (1987). BURNEY, P.G.C., COMSTOCK, G.W. and MORRIS,J.S., Serum micronutrients and subsequent risk of pancreatic cancer. Amer. J. d i n . Nutr., 49, WALD,N., Retinol, beta-carotene and cancer. Cancer Surv., 6 , 63-51 (1987). 895-900 (1989). WASSERTHEIL-SMOLLER, ROMNEY, S.L., WYLIE-ROSETT, J., SLACLE, DIMASCIO.P.. KAISER.S. and SIES. H.. Lvcooene as the most efficient S., MILLER,G., LUCIDO,S., D., DUTTAGUPTA, C . and PALAN,P.R., Dietary biological carotenoid singlet oxygen quench&. hrch. Biochem. Biophys., vitamin C and uterine cervical dysplasia. Amer. J. Epidemiol.. 114, 7 1 4 274, 532-538 (1989). 724 (1981). HARRIS,R.W.C., FORMAN,D., DOLL, R., VESSEY,M.P. and WALD, WILLETT,W.C. and STAMPFER, M.J., Total energy intake: implications N.J., Cancer of the cervix uteri and vitamin A. Brit. 1. Cancer, 53, for epidemiologic analyses. Amer. J. Epidemiol.. 124, 17-27 (1986). 653-659 (1986). WYLIE-ROSETT, J.A., ROMNEY,S.L., SLAGLE,N.S., WASSERTHEILHELZLSOUER, K.J., COMSTOCK,G. and MORRIS,J.S., Selenium, lyco- SMOLLER,S., MILLER,G.L., PALAN,P.R., LUCIDO,D.J. and DUTpene, alpha-tocopherol, beta-carotene, retinol, and subsequent bladder TAGUFTA, c . , Influence of vitamin A on cervical dysplasia and carcinoma cancer. Cancer Res., 49, 6144-6148 (1989). in situ. Nutr. Cancer, 6 , 49-57 (1984). KELSEY,J.L. and HILDRETH,N.G., Breast and gynecologic cancer and ZEIGLER,R.,A review of epidemiologic evidence that carotenoids reduce epidemio/ogy. CRC Press, Boca Raton (1983). the risk of cancer. J . Nutr.. 119, I16122 (1989).
