Int. J. Cancer: 52, 750-758 (1992) 0 1992 Wiley-Liss, Inc.

Publication of the lnternailonal Unlon Agalnst Cancer Publication de I'Union lnternatlonaleContre le Cancer

RISK FACTORS FOR CERVICAL CANCER IN COLOMBIA AND SPAIN F.X. BOSCH ',I5, N. MUfiOZ', s. DE SANJOSE', 1. IZARZUGAZA', M. GILP,P. VILADIU',M.J. TORMO', P. MOREO~, N. ASCUNCE', N. ARISTIZABAL9, M. SANTAMARIA", P. ALONSO DE R U I Z ' ~ L.C. GONZALEZ~, L. TAFUR9, J.M. KALDOR'", E. GUERRERO", and K. SHAHl4 I Unit of Field and Intervention Stirdies, International Agency for Research on Cancer, 150, Coiirs Alber? Thomas, F-69372 Lyon Cedex 08, France; 2Registrode Cancer de Eiiskadi, Vitoria-Gasteiz; sUniversidad de Sevilla, Seville; 4Hospital Provincial Sta Caterina, Girona; 5Consejeriade Sanidad, Murcia; bRegistrode Cancer, Diputacion General de Aragon, Zaragoza; 'Departamento de Salud, Gobiemo de Navawa, Paiplona; 8Delegacibn Tem'torial de Bienestar Social, Salamanca, Spain; 91Jniversidaddel Valle, Cali, Colombia; '"National Centre in HIV Epidemiology and Clinical Research, Darlinghirrst,Australia; "Abbott Cientifica s.a., Madrid; {?HospitalProvincial de Navara, Pamplona, Spain; I3Hospital General de Mexico, UNAM, Mexico D.F., Mexico; and I4TheJohns Hopkins UniversityqSchool of Hygiene and Public Health, Baltimore, MD, USA. A population-based case-controlstudy of cervical cancer was conducted in Spain and Colombia to assess the relationship between cervical cancer and exposure to human papillomavirus (HPV), selected aspects of sexual and reproductive behaviour, use of oral contraceptives, screening practices and smoking. The study included 436 cases of histologically confirmed squamous-cell carcinoma and 387 age-stratified controls randomly selected from the general populationthat generated the cases. The presence of HPV DNA in cervical scrapes was assessed by PCR-based methods and was the strongest risk factor (OR = 23.8; 13.4-42.0). Risk estimates for any other factor were only slightly modified after adjusting for HPV status. Among women found positive for HPV DNA, only the use of oral contraceptives was a risk factor for cervical cancer (OR = 6.5; I .3-3 I .4 for ever vs. never use). Patients with cervical cancer who were HPV DNA-negative retained most of the established epidemiological features of this disease. This suggests that some instances of HPV infection went undetected or that other sexually transmitted factor(s) contribute to the causation of cervical cancer. Early age at first intercourse(OR = 4.3; 2.1-9.0 for age < 16 vf. 24+) and early age at first birth (OR = 5.0; I .&14.2 for age < 16 vs. 24+) were associated with increased risk of cervical cancer; these effects were independent of one another. Low educational level was a risk factor (OR = 2.5; I .6-3.9). Number of sexual partners was in our study a surrogate for HPV infection. Smoking and parity after age 24 were weakly and inconsistently associated with the risk of cervical cancer. Previous screening (OR = 0.7; 0.5- I .O) and ever having undergonea Caesarean section (OR = 0.4; 0.2-0.8) were protective factors. (01992 Wile).-Liss,Inc.

The most consistent risk factors for cervical cancer are a woman's lifetime number of sexual partners and her age at commencement of sexual activity. More recently, the relationship between cervical cancer and the number of sexual partners of the male partner has also been reported. Variables reflecting low socio-economic status, long-term use of oral contraceptives (OCs), smoking and high parity have been extensively studied but an association with cervical cancer has been less consistently found. Associations with some nutritional deficiencies, markers of hormonal balance and immunological status are less well documented (for reviews see Brinton, 1992; Armstronget al., 1992). In the late 1970s, the hypothesis that certain types of human papilloma-virus (HPV) were related to human genital cancers was proposed and throughout the 1980s gained rapid support. Based primarily on experimental and clinical evidence, certain types of HPV were recognized as causal factors of cervical cancer (zur Hausen, 1989). To date, 5 case-control studies on invasive cervical cancer (including our own) have been completed in different geographical areas and most of the early concerns about study design have been clarified (Mufiozet al., 1988). In spite of having used HPV DNA hybridization assays of differing validity, the results of the 5 studies are consistent and confirm strong associations

between invasive cervical cancer and HPV types 16 and 18, and occasionally with types 31, 33 and 35 (Reeves et a!., 1989; Schmauz et al., 1989; Donnan et al., 1989; Peng et al., 1991; Mufioz et al., 1992). Indirect evidence on the role of HPV has been provided by a study showing that some of the key risk factors for HPV infection and for cervical cancer are common (Ley et al., 1991). The identification of HPV as the sexually transmitted agent in cervical carcinogenesis has permitted, for the first time, adequate consideration of the role of other putative risk factors; 2 of the studies on invasive cervical cancer have attempted to do this. In one study in which HPV DNA was assessed by the filter in situ hybridization (FISH) method (later recognized as inaccurate) number of sexual partners, age at first sexual intercourse, high parity and absence of previous Pap smear were found to have independent effects (Brinton et al., 1989). In the other study which used polymerase chain reaction methods (PCR) for HPV diagnosis, none of the risk factors considered showed an independent effect after allowing for the strong HPV effect (Peng et al., 1991). However, in this latter study conducted in a rural population in China, the information collected on sexual behaviour was phrased as marital history and the prevalence of smoking and O C use was limited, making the results difficult to compare with those of other studies. In 1985, we initiated a population-based case-control study in Spain and Colombia to investigate the relationship of HPV infection to cervical cancer and to re-evaluate the role of the established risk factors. A particular objective was to assess if the association with sexual behaviour could be partly or entirely explained by HPV infection and if other environmental factors, including tobacco smoking and use of OCs, interacted with the virus in causing the disease. The main results concerning the association of cervical cancer with HPV are reported in a companion article (Mufioz et al., 1992); in this report we present an evaluation of the established risk factors for cervical cancer. METHODS

The design of the study has been described in detail (Mufioz et al., 1992). In brief, the field work was conducted in Spain from June 1985 to December 1987 and in Colombia from June 1985 to December 1988. Eligible cases were incident, histologically confirmed, invasive squamous-cell carcinomas of the cervix identified among residents (of at least 6 months' standing) in the predefined study areas. All CIN 111 cases diagnosed during the study period were also identified and the histological slides reviewed by a panel of pathologists to ensure I5Towhom correspondence and reprint requests should be sent.

Received: June 23,1992 and in revised form August 5 , 1992

CERVICAL CANCER RISK FACTORS

completeness of recruitment of the invasive cancer cases. Controls were an age-stratified random sample selected from the general population that generated the cases and the target number was one control per case. Individuals identified as controls who were ineligible or untraceable were replaced but those who refused were not. All women were asked to permit the taking of 2 cervical scrapes. Smears were prepared and stained by the Papanicolaou method and the remainingcells were eluted in phosphatebuffered saline (PBS), pelleted into 2 aliquots and stored at -20°C or -70°C. A fresh biopsy from the cancer was obtained and frozen without any processing. The histopathological sections from which the diagnosis of the cases had been made (invasive or CIN 111) as well as the cytological smears collected for the study (cases and controls) were submitted to a panel of 3 cytopathologists (P.A., M.S., N.Ar.), who reached a consensus diagnosis after several reviews. All protocols were cleared by the IARC and local ethical committees. Detection of HPV D N A by PCR HPV DNA sequences were sought for in cytological specimens obtained by cervical scraping. The laboratory methods are described in detail elsewhere (Guerrero et al., 1992). All specimens were first screened for HPV sequences using Virapap (TM) and Southern hybridization. PCR amplification was used in the remaining material using HPV LI consensus primers. The amplification products were hybridized sequentially with probes to HPV 6, 11, 16, 18, 31, 33 and 35 under high-stringency conditions. Subsequently, the filters were screened with a generic probe which contained a mixture of amplimers of HPV 16 and 18 (Bauer et al., 1991). The primers for preparation of amplimers were kindly supplied by Dr. M. Manos. Statistical analysis All established risk factors for cervical cancer, including HPV status, were evaluated using linear logistic regression models to estimate odds ratios (ORs) and 95% confidence intervals after controlling for the effects of potential confounders (Breslow and Day, 1980). Subjects without biological specimens were included in all analyses as a separate category (unknown HPV status). Because of their potential confounding effects and the stratified design, age and study area were included in all logistic models. To explore whether the effects of established risk factors were the same in women with and without HPV DNA, interaction terms between each variable and HPV status were evaluated while controlling for the effects of all other variables. This analysis was restricted to women for whom a PCR result was available. The significance of the interaction was tested by use of the likelihood ratio statistic. The relationship between age at first birth and the risk of cervical cancer was analysed in discrete categories and by use of different polynomial logistic regression models. The GLIM software package (Baker and Nelder, 1978) was used. RESULTS

Among the individuals identified as eligible, 80% of the cases and 74’3%of the controls participated. The number of subjects included in this analysis was 823; 250 cases and 238 controls in Spain and 186 cases and 149 controls in Colombia. Of these, 72% of the cases and 80% of the controls provided cytological specimens and in addition 93% of the cases provided histopathological sections for review. A consensus diagnosis was provided by the cytopathology panel for each subject, following which 66 cases originally diagnosed as CIN 111 were up-graded to invasive cancer. These cases were not different from the originally diagnosed invasive cases of the same age with respect to their exposure to the established risk

75 1

factors, but their inclusion resulted in slightly more cases than controls. Of the 823 participants, 457 (229 cases and 228 controls) provided sufficient samples for PCR testing. These were compared with the 206 cases and 151 controls who failed to provide a sample. No differences were observed in the prevalence of major risk factors except for OC use, which was more prevalent among cases who did not provide a sufficient sample (37%) than among cases who did (28%,p = 0.05). Table I shows the age distributions and prevalence of the main risk factors for cervical cancer by case-control status and country. Cases in Colombia were younger (mean ages 46.5 vs. 52.2; p > 0.05) and had higher exposures to all risk factors, except “low education” and “never had previous screening”, than cases in Spain. The HPV prevalence was almost identical among cases in the 2 countries (69.0% in Spain vs. 72.4% in Colombia) and was largely due to HPV type 16. Among controls, 91.6% of women in Spain reported only one lifetime sexual partner as compared to 58.4% in Cali where almost 40% declared 2 to 5 lifetime partners. Less than 3% of the control groups in both countries declared 6 or more partners. Of the control women in Spain, 22% had ever used OCs and 17% had smoked cigarettes; the corresponding values in control women in Cali were 36% and 40%. The prevalence of HPV DNA among controls as detected by PCR was 4.6% in Spain and 13.3% in Cali. Table I1 shows the OR’S and 95% confidence intervals for the established risk factors for cervical cancer. In addition to age and study area, the adjustment included HPV status, number of sexual partners, age at first birth, presence or absence of previous screening, and level of education. Number of partners, visits to prostitutes or smoking habits of the husbands modified in a limited way the risk estimates for the female characteristics and were not included in any of these analyses. The evaluation of the male role in the risk of cervical cancer will be reported elsewhere. In general terms, the risk factors identified in our study were consistent in both countries. Significant increases in risk were found for low education level, number of sexual partners, age at first sexual intercourse and age at first birth. Age at first birth was related to cervical cancer more consistently than age at first intercourse and was retained throughout the analysis as a regressor variable. When both variables were considered in the same model and categorized in the same age groups (i.e. ages = < 17, 18-20, 21+) the effect of age at first birth remained unchanged (ORc = 1.0, 2.5, 4.0; O R adjusted for age at first intercourse = 1.0,2.4,3.4) whereas the effect of age at first sexual intercourse became statistically non-significant (ORc = 1.0, 1.8, 2.5; O R adjusted for age at first birth = 1.0, 1.0,1.2). Smoking showed a weak and marginally significant association with cervical cancer. Ever having had a Caesarean section was a protective factor although only 16 cases and 33 controls reported at least one. The estimated effect of multiple Caesarean sections was also significant and a reduction of 40% (10-60%) of the risk for cervical cancer was estimated for each Caesarean section (data not shown). Previous abortions were reported by one-third of our population and stillbirths by less than 5%. None of these was associated with cervical cancer. Parity showed an increased risk for cervical cancer in Spain only and previous screening was protective in Cali. To further investigate the effect of an early first birth, an analysis was conducted restricted to women who had initiated sexual activity before the age of 22. Table I11 shows that women who before that age had had one pregnancy were at a 4-fold increased risk as compared to women who did not. For 2 or more pregnancies before 22 years of age, the risk estimates remained at the same 4-fold level. When the analysis was

752

BOSCH ETAL. TABLE I - DISTRIBUTION OF STUDY SUBJECTS BY ESTABLISHED RISK FACTORS FOR CERVICAL CANCER BY COUNTRY Spain

~

Number of subjects Age, mean (SD)

< 30 30-39 4034 45-54 2 55

Education (schooling) Ever Never Parity 0-2 3-5 6+

Colombia

Cases N (%)

Controls N (%)

Cases N (“.)

Controls N (%)

250 52.2 (12.7) 7 (2.8)

238 52.3 (12.5) 7 (2.9)

186 46.5 (12.7)

149 47.5 (12.4) 10 (6.7)

162 64.8 88 r35.21

179 75.2 59 t24.81

155 83.3 31 f16.71

140 94.0) 9 16.0)

102 (40.8 119 t47.61 29 11.6

136 (57.1 86 (36.11 16 (6.7)

41 (22.0 62 133.31 83 44.6

45 (30.2 53 t35.61 51 34.2

189 7 5 6 48 [19:2{ 13 (5.2)

218 91.6) 16 16.7 4 (1.71

33 (17.7

76 (30.4)

121 (50.8

111 (44.4

Number of sex partners 0- 1 2-5 6+

4 (2.7)

Age at first sexual intercourse 24 + 22-23 20-21 18-19 16-17 < 16

10 (5.3) 13 6.9) 22 111.8)

9 (3.6)

26 (17.5)

Age at first birth

+

24 22-23 20-21 18-19 16-17 < 16

nulliparous Oral contraceptives Never Ever Unknown Caesaren section Never Ever Smoking Never Ever Previous screening Never Ever HPV Negative Positive Unknown

145 60.9 36 115.11 22 9.2 6 12.51 1 0.4

22 11.8 19 110.2

46 3 0 9 18 [12:1{

28 (11.8)

3 (1.6)

19 12.8) 7 t4.7 3 (2.01

182 72.8) 64 f25.6) 4 (1.6)

175 735 53 [22:31 10 (4.2)

109 58.6 77 l41.41 0 -

1 (0.6)

241 96.4) 9 t3.6)

219 92.0) 19 18.0)

179 96.2) 7 j3.8)

135 90.6) 14 f9.4)

184 73.6) 66 126.4)

198 (83.2 40 (16.81

92 (49.5 94 (50.51

195 (78.0 55 (22.01

171 71.8 67 128.21

9 (3.6) - 25 (10.0)

- -

I

124 52.1) 6 2.5) 108 (45.4)

24 12.9 63 133.91 99 (53.2)

85 57.0) 13 $8.7, 51 (34.2)

‘“Ever screening” refers to those having had at least one Pap smear at least 12 months before selection for the study. restricted to women who had only had one lifetime pregnancy, a higher risk estimate was also found if this occurred before the age of 22; OR = 6.6 (0.6-70.5) (data not shown). Figure 1 shows the relationship between age at first birth and the risk of cervical cancer, taking as reference those women who had had their first child at age 22. The data are presented as discrete categories by age at first birth and after modelling the relationship, and show that the risk decreases steeply for each year of delay of the age at first birth until ages

26-28, after which it remains stable. The best description of

the data corresponded to a cubic model which significantly improved the fit of the linear and quadratic models. Table IV shows the distributions and risk estimates for selected factors according to HPV D N A status and includes only subjects for whom HPV status was determined. Risk estimates for number of sexual partners and OC use were significantly different between women who were positive and women who were negative for HPV (JI < 0.05). The effect of

753

CERVICAL CANCER RISK FACTORS

TABLE 11 - RISK FACTORS FOR INVASIVE CARCINOMA OF THE CERVIX BY COUNTRY

Risk factor

Education (schooling) Ever Never Ph Parity 0-2 3-5 6+ P Number of sexual partners 1 2-5 6+ P Age at first sexual intercourse4 24 f 22-23 20-21 18-19 16-17 < 16 P Age at first birth5 24 + 22-23 20-21 18-19 16-17

< 16 P

I

Oral contraceptives Never Ever P Caesarean section Never Ever P Smoking Never Ever P Previous screening Never Ever P HPV Negative Positive Unknown

Spain ORc'

ORa' (9%

Colombia CL)'

ORc

ORa (95% CL)

Spain and Colombia ORa (9% CL)

1.o 2.0

1 .o 2.4 (1.4 4.1) 0.003

1.o 3.3

1.0 3.3 (1.3 8.5) 0.003

1.0 2.5 (1.6 3.9) 0.000

1.0 1.9 2.6

1.0 2.5 (1.5 4.1) 1.9 (0.8 4.7) 0.002

1.0 1.3 2.1

1.o 1.3 0.6 2 6 1.2 (0.6 2141 0.7

1.0 1.9 1 3 2 7 1.6 (0:Y 2:6] 0.05

1 .o 3.6 4.4

1.0 3.4 1.7 6.9) 4.8 11.3 18.5) 0.000

1.0 1.5 9.4

1.0 1.2 0.7 2.0) 6.9 t2.0 23.8) 0.003

1.o 1.7 1.1 2.5) 5.6 12.4 13.4) 0.000

1.o 1.5 2.9 3.3 10.0 19.2

1 .o 1.5 (0.8 2.8 2.8 (1.4 5.51 2.7 1.45.4 7.5 2.4 23.7) 13.9 11.5 131.9) 0.000

1.o 1.3 3.0 2.4 2.6 4.6

1.o

1.o 1.2 (0.7 2.1)

2.5 0 9 7.1 2.9 {1:0 8.61 0.012

3.7 1.9 7.0 4.3 j2.1 9.0) 0.000

1.0

1.0

1.0 1.8 2.6 9.2 13.8

1.0

1.0 2.1 2.2 3.3 4.4

4.3 (1.8 10.3)

14.7 (1.5 145.1) 6.1

6.8 (2.0 23.4) 0.000

5.0 (1.8 14.2) 0.000

0.000 1.0 1.2

1.o 1.1 (0.6 2.0) 1.0

1.0 1.3

1.0 1.5 (0.8 2.9) 0.3

1.0 1.3 (0.9 2.0) 0.3

1.o 0.4

1.o 0.5 (0.2 1.3) 0.15

1.o 0.4

1.0 0.3 (0.1 1.1) 0.06

1.o 0.4 (0.2 0.8) 0.01

1.o 2.1

1.o 1.6 (0.8 2.9) 0.14

1.o 1.7

1.0 1.4 (0.8 2.4) 0.22

1.o 1.5 (1.0 2.2) 0.05

1.0 0.7

1.o 0.9 (0.5 1.5) 0.5

1.o 0.5

1.0

1.o 0.7 (0.5 1.0) 0.06

0.5 (0.3 0.9) 0.02

1.n 1.o 1.o 1.0 17.1 46.2 (18.5 115.1) 15.6 6.9 34.7 23.8 13.4 42.0) 6.8 2.4 (1.5 4.0) ' 8.3 l4.4 15.61 4.0 f2.7 5.6) n n nnn 0.000 r, 0.000 'ORc, odds ratio adjusted for age and centre.-?ORa, odds ratio adjusted for age, centre, HPV status assessed by PCR, number of sexual partners, education, age at first birth and ever previous screening as a~propriate.-~(95%CL), confidence limit~.-~Notadjusted for age at first birth.5Excluding nulliparous.-6p, Likelihood ratio test for the difference between 2 categories or test for trend across more than 2 categories.

1.0

49.5 2.8

754

BOSCH E T A L . TABLE 111 - EFFECT O F CHILDBIRTH BEFORE 22 YEARS O F AGE

Number of childbirths before 22

Cases/ controls

ORc’

519

1.o 3.5 3.7

ORa2 (950% CII

bearc of ape

0 1

105154 225 / 110

1 or more

1.o 4.1 (1.1 17.8) 4.0 (1.0 16.1)

‘ORc, adjusted for age and centre.-?ORa, adjusted for age, centre, education, number of sexual partners, age at first sexual intercourse (continuous), ever screening, HPV status and additional number of children born after 22 years of age. 10

515

16-

18-

-

20-

22-

24-

26-

28-

30-

232

Age at first birth Categoricalmodel

-

Cubic modal

FIGURE1 -Relationship between age at first birth and risk of cervical cancer. Categorical model: Risk is expressed for discrete age at first birth groups. Age 22 is taken as the reference group. Cubic model: Risk is expressed after modelling the relationship taking age at first birth as a continuum.

increased number of partners remained among the HPVnegative only (test for trendp = 0.000) and use of OCs showed a strong effect among the HPV-positive. An analysis by age at first exposure indicated that ever-use of OCs before the age of 20 conveyed an additional risk over that of women who initiated OC use after that age. OR = 2.9 (1.4-6.2) after adjustment for number of years of O C use (data not shown). Among HPV-positive women, no other risk factor remained significantly associated with cervical cancer. Among HPVnegative women, early age at first birth, having never attended school and history of previous screening were associated with cervical cancer. Early age at first intercourse and high parity showed some degree of association. Cigarette smoking was not related to cervical cancer in either of the 2 groups. Table V presents the analysis of the joint effects of O C use and HPV status. Relative to HPV-negative women who never used OC, the OR for HPV-positive and ever-users of O C was 113.4 (44.4-289.0). The interaction was statistically significant (p = 0.007). DISCUSSION

Our study has provided the opportunity to evaluate risk factors for cervical cancer as assessed by comprehensive methodology while allowing for the very strong effect of PCR-detected HPV DNA. It has been concurrently conducted

in 2 populations with important differences in the incidence of cervical cancer and yet with some common cultural features. Our most striking finding is a strong effect of the use of OCs in increasing risk of cervical cancer in women who were HPVpositive. We also observed a large increase in the risk with early age at first sexual intercourse and with early first birth. HPV-negative cases showed all the established risk factors for cervical cancer, suggesting that in these cases HPV infection may have remained undetected. Alternatively, other sexually transmitted agents may have contributed to these cases. Risk factors were in general consistent across the 2 populations.

HPV detectiori In our study the link between HPV and cervical cancer was confirmed as the major risk factor, and at least 70% of cases are associated with the viral infection. The very high OR’S linked to HPV DNA dominate the interpretation of risk estimates for any other factor and it is important to discuss in the first place the results of the hybridization assay. The prevalence of HPV DNA among cases was lower than in several published and unpublished studies, some of which have reported HPV prevalences (in tissue specimens as opposed to the cervical scrapes used in our study) of more than 70% and sometimes more than 90% (Van den Brule et al., 1991; Lorincz et al., 1992). In contrast, the results of the laboratory procedures followed to prevent sample contamination and the results obtained for the internal controls established suggest that the fraction of HPV DNA false positives is rather low. If 15-20% of our cases (and a few additional controls) are misclassified as HPV-negatives, the adjustment presented in Table I1 is insufficient and justifies the interpretation of the results based on comparison of the analyses presented in Tables I1 and IV. However, our analysis by HPV status, presented in Table IV, is limited because only 19 controls were identified as HPV-positive. making the results highly vulnerable to any additional misclassification on the variables of interest. This is clearly shown by the wide confidence intervals around some of the risk estimates. In a study in progress, biopsies from the cervical cancer cases that were HPVnegative in cervical scrapes have been re-examined and at least 50% have been found HPV-positive when identical amplification and detection systems have been used. Age at first sexual intercourse, age at first birth and OC use The results presented in Table I1 and Figure 1 provide additional support for the hypothesis that early sexual intercourse and giving birth at early ages are both related to the risk of cervical cancer and suggest in addition that relationship is not linear after a certain age (24+ in our study). The majority of our subjects had their first child within the year following initiation of sexual relationships and women who had multiple pregnancies also tended to have their first pregnancy earlier. Because of the close correlation between first intercourse and first birth (r = 0.85 after exclusion of the 55 nulliparous women), it was sometimes difficult to fully separate their individual effects. However, the biological implications of a risk for cervical cancer associated with early pregnancy were of considerable interest and this was further studied in the group of women who had early sexual intercourse with or without concurrent early pregnancy and birth. The results presented in Table I11 as well as other analyses performed using other age limits (data not shown) strongly suggest that early pregnancy makes an additional contribution to the risk of cervical cancer, over and above the effects of early intercourse. O u r results also indicate that further pregnancies later in life may add little to the risk. In our study, the risk associated with ever-use of OCs was significantly elevated among women who were HPV-positive. In the absence of HPV DNA, there was no measurable effect (Tables 111 and IV). These results suggest a synergistic effect of

755

CERVICAL CANCER RISK FACTORS TABLE IV - DISTRIBUTION AND RISK ESTIMATES FOR INVASIVE CARCINOMA OF THE CERVIX BY HPV-DNA STATUS. SPAIN AND COLOMBIA

HPV-DNA-negaiwe

Education (Schooling) Ever Never

Cases/

ORa

controls

(95% CL)'

471178 21131

1.o 2.2 (1.1 4.7)

HPV-DNA-porilive Cases/ ORa (95% CL)

controls

121117 4012

Tesl for interaction />-value*'

1.0 2.6 (0.6 11.6) 1.0

P3 Parity 0-2 3-5 26

0.003 18175 32179 18/55

1.o 1.8 (0.9-3.7 1.6 (0.7-3.71

0.16 4314 5714 61111

1.o 1.4 (0.3 6.2) 0.6 (0.2 2.0) 0.3

0.039

P Number of sexual partners 1

2-5 26

491165 12143 711

1.o 1.1 (0.5 2.6) 35.6 (3.9 329.2)

0.65

99/10 4417 1812

1.0 0.7 0.2 2.1 1.2 10.2 6.01 0.043

P Age at first sexual intercourse 21 + 18-20 < 17

0.000

381127 12140 18/42

1.o 1.2 (0.5 2.7) 3.0 (1.2 7.3)

0.34

6819 4713 4617

1.0 2.9 0.7 12.1) 1.8 10.5 6.0) 0.09

0.10

P Age at first births 21 + 19-20 < 18

451157 13135 10117

.o

1

2.8 1 1 7 3 3.4 [1:2 9:71

0.23 85110 4816 2813

1.o 1.5 (0.4 5.3) 2.4 (0.6 10.2) 0.7

P PAP screening Never Ever

0.00 471 104 211105

1.o 0.5 (0.2 0.9)

0.27 104112 5717

1.o 0.9 (0.3 2.6) 0.27

P Oral contraceptivesh Never Ever

0.90

0.005 521148 14158

1.o 0.8 (0.4 1.9)

110117 5012

1.o 6.5 (1.3 31.4) 0.007

0.26

P Years of O C use6 Never 1-9 2 10

521148 5123 9/35

1.o 0.8 (0.2 2.6 0.9 (0.3 2.11

0.05 110117 1211 3511

I .o 0.032

0.460

P Caesarean' Never Ever

591184 2/18

P Smoking Never Ever

.o

1

0.5 (0.1 2.3)

0.027 145116 611

0.78

0.17 461 153 21156

1.o 1.5 (0.7-3.0)

1.o 0.7 (0.1 7.3) 0.7

108111 5318

1.0 0.6 (0.2 1.8)

0.18 -0.42 'ORa, Odds ratio adjusted for age, centre, education, number of sexual partners, age at first birth and ever screening PAP as appropriate.-'p*, Likelihood ratio test of an interaction term between HPV and the characteristic under study. It evaluates the difference in the risk estimates for the relevant factor between HPV-negative and HPV-positive women.-?p, Likelihood ratio test for the difference between 2 categories or test for trend across categories.-JAdjustment does not include age at first birth.-5Nulliparous included in group 21 +.-hExcluding those with unknown information.'Excluding nulliparous; analysis adjusted also for total number of children. n

0.47

756

BOSCH ET AL. TABLE V DISTRIBUTION AND RISK ESTIMATES FOR INVASIVE CARCINOMA OF THE CERVIX FOR THE ASSOCIATION OF ORAL CONTRACEPTIVE USE AND HPV STATUS, SPAIN AND COLOMBIA ~

HPV-DNAneeative

Oral contraceptives Never Ever use 0~b4

HPV-DNAoositive

.w

1 [52/14813 1.4 (0.8 2.3) [14/58] 1 .o

ORa'

1.o 1.4 (0.7 2.7) 29.3 ii6.i 53.4)

'ORa, Odds ratio for oral contraceptive use adjusted in addition for HPV status.-20dds ratio adjusted for age, centre, number of sexual partners, education, age at first birth and ever screening. ( ) = 95% confidence 1, number of casesinumber of c~ntrols.-~ORb,Odds ratio of HPV status adjusted in addition for OC use. OCs and HPV in the origin of cervical cancer. However, a word of caution should be added because only 2 controls were HPV-positive and ever-users of OC. Risk estimates based on such small numbers are unstable and even a moderate misclassification of the variables used in the analysis has a great impact on the risk estimates and potentially on the interpretation of the analysis. Our finding should clearly be replicated by other studies also based on high-quality HPV detection methods and, if confirmed, might have some clinical and publichealth implications. Several publications support some of these conclusions. Virtually all studies on cervical cancer have reported that risk estimates associated with age at first intercourse approach a plateau after 20-23 years. This is consistent in studies that did or did not adjust for HPV status (reviewed by Brinton, 1992). The focus in most of these studies was on age at first sexual intercourse rather that age at first birth, and this may reflcct the differences across populations in sexual and reproductive behaviour, but the results consistently suggest that the period of high risk has an age limit perhaps linked to aspects of maturation of the transitional zone of the cervix. The independence of the effects of early pregnancy from the effects of early intercourse has been occasionally described (Cuzick et a/., 1990), but not always confirmed (Brinton et al., 1989). Studies in populations with longer intervals between first sexual intercourse and first pregnancy might be useful to confirm our results and to further assess the specific role of early pregnancies in the risk of cervical cancer. OC use has been inconclusively evaluated as a risk factor for squamous-cell carcinomas in studies which did or did not consider HPV status (reviewed by Brinton, 1992). The differences in the results presented in Tables I1 and IV in relation to OC use indicate that lack of proper HPV status asscssment is a major difficulty in interpreting most of the previous studies on O C use and cervical cancer. A few studies have specifically addressed the relationship between HPV and steroids in patients with cervical cancer. One Southern hybridization (SH)-based survey reported a higher prevalence of HPV D N A in thc canccr tissuc of OC users than in non-OC users (Hildesheim et al., 1990). Analysis of sex steroid hormone receptors in HPV-induced cervical lesions suggested that oestrogen and, particularly, progesterone receptors may act as co-factors in promoting HPV-related cervical neoplasia (Monsonego et al., 1991). Finally, experimental work is producing additional evidence of possible interactions between steroids and HPV. Oestrogen increases the transcription of the HPV-16-transforming proteins in the SiHa cell line (Mitrani-Rosenbaum et al., 1989) and progesterone has been shown in a rat model to induce oncogenic transformation in the presence of HPV 16 DNA and the ras oncogene (Pater et al., 1990). HPV-16-immortalized cclls from the transitional zone of the cervix show enhanced C16 alpha-hydroxylation of oestradiol, a metabolic pathway that has been proposed as part of a model for cervical

carcinogenesis linked to the combined effects of HPV and oestrogens (Auborn et al., 1991). Other investigations are currently producing evidence on progesterone- and dexamethasone-dependent elements in the non-coding region of genital HPVs (reviewed by zur Hausen, 1991). These recent epidemiological and experimental results generate interesting hypotheses for testing in the future.

Panty and Caesarean section In our data, the effect of parity (full-term pregnancies) was moderate and inconsistent (Table 11). In Spain, the increase in risk observed was largely explained by the effect of early age at first birth. In Cali, where 15% of our population reported 10 or more pregnancies, there was no detectable effect in any of the analyses performed. Table IV shows that the risk estimates were elevated (although not significantly) only among the HPV-negative subjects. The results obtained in Colombia contrast with the findings of a recent study in 4 Latin American populations that reported a strong effect of parity (Brinton et al., 1989). Several clinical studies have indicated that HPV detection is higher among pregnant than non-pregnant women and that the HPV prevalence increases as the pregnancy progresses (for a review see Schneider and Koutsky, 1992). If the effects of HPV in relation to cervical carcinogenesis were enhanced during pregnancy, a dose-response effect with parity would be expected, contrary to our findings in Colombia. This apparent contradiction could be explained by the powerful effect of age at first birth, and is consistent with the interpretation that subsequent pregnancies after 24 years have little additional effect. The protective effect of the Caesarean section appears to support the vicw of a dclcterious effect of vaginal delivery through trauma. It also provides indirect evidence for the postulated independence of the risk linked to early first birth. However, assessment of other traumatic events during birth, such as the use of forceps or the presence of clinical complications during and after delivery, has been inconclusive (Brinton et al., 1989). Moreover, it is uncertain by what biological mechanism traumatic events in the cervix could promote neoplastic growth, unless we postulate an increased risk of HPV infection or some non-specific effect of increased cell proliferation. Number of sexual partners The number of sexual partners had no relationship to cervical cancer among women who were HPV-positive, and retained a strong effect among HPV-negatives. The results among HPV-positive subjects suggest that, once the HPV infection relevant to the carcinogenic process is contracted, further increasing the number of partners does not significantly influence the risk of progression to cancer. The results among HPV-negative cases and controls are more difficult to interpret due to the recognized underdetection of HPV status, at least among the cases. Among the controls the prevalence of

757

CERVICAL CANCER RISK FACTORS

HPV DNA increased significantly with the number of partners (Table IV). In summary, our results d o not support the independence of number of partners from HPV as a risk factor for cervical cancer and clearly show that this variable is to a large extent a surrogate measure of HPV infection.

Smoking, screening and education level In our data, ever-smoking was marginally associated with cervical cancer after adjustment for HPV status (Table 11). HPV-negative women showed a non-statistically-significant increase in the risk estimates and n o effect was observed among HPV-positives (Table IV), providing limited support for an independent role of cigarette smoking. Two reports have produced relevant information on smoking and HPV. One of them used S H methods in a series of biopsies from cervical cancer cases and showed no association between smoking and presence of HPV D N A (Hildesheim et al., 1990). In contrast, a FISH-based case-control study reported an association of smoking with cervical cancer among HPV-positive women but no association among HPV-negatives (Herrero et al., 1989). These inconsistencies between studies underline the importance of using accurate measurement of HPV status to evaluate risk factors such as smoking for which the risk estimates are only incompletely adjusted by number of sexual partners or other aspects of sexual behaviour. Previous screening is protective in the population of Cali where, in our study, 70% of the control women reported at least one screening test before the 12-month period preceding the interview. In Spain only 20% had been screened and the protective effect was not statistically significant. Table IV shows that the reduction in risk is mostly observed among the HPV-negatives. Low educational level is a general indicator of social status or social behaviour, which have commonly been associated with cervical cancer. The effect of never having attended school, found in our study, confirms that some important, yet imprecise exposure is linked to educational status and that this is perhaps independent of HPV. The lower education level and screening history reported by the population controls in Spain could be partly due to the inclusion of less educated rural populations, whereas in Cali only urban residents were included. The corresponding differences observed also among cases could indicate that, in Spain, cervical cancer occurs largely among women of lower socio-economic status, whereas in the high-risk area in Cali, cervical cancer also occurs among the more educated social classes. An additional and more general difficulty in the interpretation of questionnaire-based information on screening history lies in the almost unavoidable confusion in the participants’ recall between cytology taking and any other gynaecological examination. Finally, studies of the risk factors for HPV infection are being useful in confirming the existence of risk factors in common between HPV infection and cervical cancer. One PCR-based case-control study among women under 50 years of age with normal cytology reported on 4 risk factors for HPV DNA detection; high number of sexual partners, use of OCs, young age and black ethnicity. Smoking was related to HPV detection in a crude analysis but was entirely explained by the

underlying association of smoking with number of sexual partners (Ley et al., 1991). These results are strikingly consistent with our findings on cervical cancer. In conclusion, the results of our study are consistent with the interpretation that early HPV infection acquired through sexual intercourse is the key risk factor for cervical cancer and that concurrent early pregnancy and exposure to OCs are important co-factors. Since most cervical cancers occur only years after HPV exposure and in a fraction of the HPV ever-exposed women, it is conceivable that early pregnancy and O C use are relevant to the establishment of chronic type-specific HPV infection following exposure. The HPV chronic carrier state will in turn place women at high risk of developing cervical cancer later in life. Our results are also consistent with a promoting role of O C use as a risk factor for progression from chronic HPV infections to cervical cancer. Our results allow the speculation that the tendency towards initiating sexual intercourse at younger ages and delaying first birth may help to explain the secular trend of overall decreasing incidence and mortality from cervical cancer observed in most developed countries. This impressive finding has been partially attributed to screening practices. However, it remains largely unexplained for the time periods before the introduction of screening and for countries in which screening programmes have a limited coverage within the population at risk. Furthermore, if the interaction observed between HPV and OC use is confirmed, it could contribute to explain the upward trend in incidence of and mortality from cervical cancer among women under 40 in populations which had concurrent rises in the incidence of sexually transmitted diseases and massive access to OC during the last 30 years. Topics of particular interest for future studies are the epidemiology of HPV infections and its interaction with hormonal factors in the period from sexual initiation throughout young adulthood and during pregnancy. Studies on the dynamics of type-specific HPV infections and its markers should better characterize the chronic carrier status and identify its specific risk factors, notably the suspected fundamental effect of age at infection.

ACKNOWLEDGEMENTS

The authors are indebted to all the study participants, to the gynaecologists, pathologists and oncologists who facilitated the identification and contribution of the participants and to the supervisors of the field work. Thanks are also due to Ms. S. Teuchmann and Ms. N. Charnay for data management, to Ms. D. Magnin for handling of the specimens, to Drs. N.E. Day, D. English and B.K. Armstrong for comments and advice and to Dr. X. CastellsaguC for contributions to data analysis. Financial support was received from the IARC, the European Community [CI 1-0371-F (CD)] and the Fondo d e Investigaciones Sanitarias (FIS) of the Spanish Government (86/753, 871 1513, 88/2049, 90/0901). The authors acknowledge the support received in Spain from the Health Departments of the autonomous communities and from the local administrations in the areas that participated in the study.

REFERENCES

ARMSTRONG, B.K., MuNoz, N. and BOSCH,F.X.,Epidemiology of cancer of the ccwir. In: M. Coppleson, J.M. Monaghan, C.P. Morrow, pp. 11-29, Churchill and M.H.N. Tattersall (eds.), Gy~zecologi~uticu/ogy, Livingstone, Edinburgh (1992).

K.J., WOODWORTH, C., DIPAOLO, J.A. and BRADLOW, H.L., AUBORN, The interaction between HPV infection and estrogen metabolism in cervical carcinogenesis. hi. J. Cancer, 49,867-869 (1991). BAKER,R.J. and NELDER,J.A., The GLIM system, Release 3. General-

ized linear interactive modelling. Numerical Algorithms Group, Oxford (1978). C.E., CHAMBERS, J.C., TASHIRO, C.J., BAUER,H.M., Yt TING,GREER. A. and MANOS,M., Genital human papillomaCHIMERA, J., REINGOLD, virus infection in female university students as determined by a PCR-based method. J. Amer. med. Ass., 265,472477 (1991). BRESLOW, N.E. and DAY,N.E., Statistical methods in cancer research. Vol. I: The analysis of case-control studies, U R C Scientific Publications, 32, International Agency for Research on Cancer, Lyon (1980).

758

BOSCH ETAL.

BRINTON, L.A., Epidemiology of cervical cancer-an overview. In: N. human papillomavirus related lesions. Int. J. Cancer, 48, 533-539 MuAoz. F.X. Bosch, K. Shah, and A. Meheus (eds.), The epidemiolop (1991). of cervical cancer and human papillo~navinrs,IARC Scientific Publications, 119, pp. 3-22. International Agency for Research on Cancer, MUNOZ,N., BOSCH,F.X. and KALDOR,J.M., Does human papillomavirus cause cervical cancer'? The state of the epidemiological evidence. Lyon (1992). Brit. J. Cancer, 57,l-5 (1988). BRINTON,L.A.. REEVES,W.C., BRENES,M.M., HERRERO,R., DE I., BRITTON,R.C.. GAITAN,E., TENORIO, F.. GARCIA,M. and RAWLS, MUGOZ, N.. BOSCH, F.X., SANJOSE,S., TAFUR,L., IZARZUGAZA, GILI, M., VILADIU,P., NAVARRO.C., MARTOS,C., ASCUNCE,N., W.E., Parity as a risk factor for cervical cancer. Amer. J. Epidemiol., GONZALEZ. L.C., KALDOR, J.M., GUERRERO, E.. LORINCZ, A., SANTAM130,486496 (1989). ARIA. M., ALONSODE Ruiz, P., ARISTIZABAL, N. and SHAH,K., The CUZICK, J., SINGER,A,, DE STAVOLA, B.L. and CHOMET,J., Case- causal link between human papillomavirus and invasive cervical control study of risk factors for cervical intraepithelial neoplasia in cancer: a population-based case-control study in Colombia and Spain. young women. Europ. J. Cancer. 26,684-690 (1990). Int. J. Cancer, 52, This issue (1992). DONNAN, S.P.B., WONG,F.W.S., Ho, S.C.. LAU,E.M.C.. TAKASHI,K. PATER,A., BAYATPOUR, M. and PATER,M.M., Oncogenic transformaand E S T ~ V EJ.,, Reproductive and sexual risk factors and human tion by human papillomavirus type 16 deoxyribonucleic acid in the papilloma virus infection in cervical cancer among Hong Kong presence of progesterone or progestins from oral contraceptives. Chinese. Int. J. Epiderniol., 18,32-36 (1989). Amer. J. Obster. Gynecol., 162, 1099-1 103 (1990). GUERRERO, E., DANIEL,R.W., BOSCH,F.X., CASTELLSAGUE, X.. MuNoz. PENG, H., L I U ,S., MANN,V., ROHAN,T. and RAWLS,W., Human N.. G w , M., VILADIU, P., NAVARRO, C., MARTOS,C., ASCUNCE,N., papillomavirus types 16 and 33, herpes simplex virus type 2 and other GONZALEZ,L.C., TAFUR,L., IZARZUGAZA,I. and SHAH, K., A risk factors for cervical cancer in Sichuan Province, China. Int. J. comparison of Virapap, Southern hybridization and polymerase chain Cancer, 47,711-716 (1991). reaction methods for human papillomavirus (HPV) identification in an L A . , GARCIA,M., BRtNES, M.M.. HERRERO, epidemiological investigation of cervical cancer. J. din. Micrabiol. REEVES,W.C., BRINTON, R., GAITAN,E., TENORIO, P., DE BRITTON,R.C. and RAWLS,W.E., (1992) (In press). Human papillomavirus infection and cervical cancer in Latin America. HERRERO,R., BRINTON, L.A., REEVES,W.C., BRENES,M.M., TENO- New Engl. J. Med., 320,1437-1441 (1989). RIO, F.. DE BRITTON,R.C., GAITAN. E.. GARCIA, M. and RAWLS,W.E., Invasive cervical cancer and smoking in Latin America. J. nut. Cancer SCHMAUZ, R., DE VILLIERS,E.M., DENNIN,R., LWANGA,S.K. and OWOR,R., Multiple infections in cases of cervical cancer from a Inst., 81, 205-211 (1989). HILDESHEIM, A,, REEVES,W.C., BRINTON, L A . , LAVERY. C., BnENES, high-incidence area in tropical Africa. Int. J. Cancer, 43, 805-809 (1989). M., DE LA GUARDIA. M.E.. GODOY,J. and RAWLS.W.E., Association of oral contraceptive use and human pa illomaviruses in invasive SCHNEIDER, A. and KOUTSKY, L., Natural history and epidemiological cervical cancers. Inr. J. Cancer, 45,860-864 6990). features of genital HPV infection. In: N. Mu~ioz,F.X. Bosch, K. Shah LEY.C.. BAUER,H.M., REINCOLD, A,, SCHIFFMAN, M.H., CHAMBERS, and A. Meheus (eds.), The epidemiology of cervical cancer and human J.C.. TASHIRO, C.J. and MANOS,M.M., Determinants of genital human papillomavirus, IARC Scientific Publications, 119, pp. 23-52, Internapapillomavirus infection in young women. J. nut. Cancer Inst., 83, tional Agency for Research on Cancer, Lyon (1992). 997-1003 (1991). VANDEN BRLLE,A.J., WALBOOMERS, J.M.M., DU MAINE,M., KENELORINCZ. A,, REID. R., BENNETJENSON,A,. GREHNBERG, M.D.. MANS, P. and MEIJER.C.J.L.M., Difference in prevalence of human LANCASTER, W. and KURMAN, R.J., Human papillomavirus infection of papillomavirus genotypes in cytomorphologically normal cervical smears the cervix: relative risk associations of 15 common anogenital types. is associated with a history of cervical intraepithelial neoplasia. Int. J. Cancer, 48,404408 (1991). Obstet. Gynecol., 79, 1-10 (1992). MITRANI-ROSENBAUM, S., TSVIELLI, R. and TUR-KASPA, R., Oestrogen z u n HAUSEN,H., Papillomaviruses in anogenital cancer as a model to stimulates differential transcription of human papillomavirus type 16 understand the role of viruses in human cancers. Cancer Res., 49, 46774681 (1989). in SiHa cervical carcinoma cells. J. gen. Virol., 70,2227-2232 (1989). MONSONEGO, J., MAGDAL~NAT, H., CATALAN, F., COSCAS, Y., ZtRAr, ZUR HAUSEN,H., Human papillomaviruses in the pathogenesis of anogenital cancer. Virology, 184,9-13 (1991). L. and SASTRE,X., Estrogen and progesterone receptors in cervical