Cervical Carcinoma and Sexual Behavior: Collaborative Reanalysis of Individual Data on 15,461 Women with Cervical Carcinoma and 29,164 Women without Cervical Carcinoma from 21 Epidemiological Studies

The association of cervical carcinoma with sexually transmitted infection has long been suggested by demographic and epidemiologic data (1, 2). A woman's lifetime number of sexual partners, becoming sexually active at an early age (3), and multiple sexual partners of a woman's husband have all been found to be consistent risk factors for cervical carcinoma (4). High-risk types of human papillomavirus (HPV), which are sexually transmitted, were first associated with cervical carcinoma worldwide in the 1990s (5, 6) and are now considered a necessary cause of cervical cancer.

The International Collaboration of Epidemiological Studies of Cervical Cancer was set up in 2003 to bring together, reanalyze, and publish the worldwide data on hormonal contraceptive use and cervical cancer risk (7). The collaboration has also published reports on the role of smoking and reproductive factors (8-10). The present report concerns the role played by a woman's lifetime number of sexual partners and age at first sexual intercourse. Unlike the risk factors previously studied by the collaboration, sexual behavior is intimately connected to the acquisition of HPV, which causes cervical cancer. Our goal in collating the available data is to analyze sexual behavior in more detail than previously possible.

The methods of study identification and data verification, collection, and correction have been described elsewhere (8, 9). Briefly, cohort (prospective) studies were eligible if they included at least 30 cases of invasive cervical carcinoma (ICC) or cervical intraepithelial neoplasia grade 3 (CIN3)/carcinoma in situ; case-control studies were eligible if they had at least 100 ICC cases or 200 CIN3/carcinoma in situ cases. When invasive and in situ carcinoma were both present, the carcinoma was classed as invasive. Cohort studies were analyzed as nested case-control studies with up to four randomly selected controls per case matched by age at diagnosis and date of entry into the study.

Analyses were restricted to women between 16 and 89 years of age. Controls who had undergone a hysterectomy were excluded, as were women who reported no previous sexual partners (27 cases and 2,324 controls).

We refer to individual studies included in the analysis by the names found in the first column of Table 1. Some studies collected information on lifetime number of sexual partners as aggregated categories. All but one study (Male Factor) had information that allowed categorization as 1, 2-5, and ≥6 sexual partners and hence we used these groups for most analyses. A more detailed classification (1, 2, 3, 4, 5-10, and ≥11) was also done; however, this entailed the exclusion of 604 ICC cases, 477 CIN3/carcinoma in situ cases, and 916 and 951 controls, respectively, from seven studies (Brinton US, Male Factor, Daling Seattle, WHO, IARC Colombia, IARC Paraguay, IARC Spain) that did not allow for unambiguous classification into one of these groups. One study (Johannesburg) had information on lifetime number of sexual partners, but not age at first intercourse, and therefore contributed only to analyses that did not require the latter variable. Age at first intercourse was recorded in single years in all other studies save one (Male Factor).

Conditional logistic regression was used to calculate odds ratio (OR) estimates, which approximate to relative risk (RR) estimates, and 95% confidence intervals (95% CI). When more than two groups were compared, the method of floating absolute risk (11, 12) was used to calculate floating SEs for the log-relative risk. For the figures, floating confidence intervals were calculated from the floating SEs to represent the dose-response relationship in a way that is independent of the choice of reference category.

Heterogeneity tests were carried out by calculating the likelihood ratio between two models: one in which the effect of the risk factor of interest was allowed to vary between strata and another where it was constrained to be the same across strata. Tests for trend were carried out using lifetime number of sexual partners and age at first intercourse as continuous variables.

All the analyses were conditioned on single year of age and on study, or study center in the case of multicentric studies (WHO, Brinton Latin America, IARC). Conditioning on other variables was also done as reported in Results. When conditioning on sexual factors (i.e., lifetime number of sexual partners, age at first intercourse), the broad classification of 1, 2-5, and ≥6 was used for lifetime number of sexual partners, and single years were used for age at first intercourse in the range 15 to 24 years, with additional categories of ≤14 years and ≥25 years for women outside of this range. When conditioning on reproductive factors [i.e., number of full-term pregnancies (FTP) and age at first FTP], the number of FTPs was categorized as 0, 1-2, 3-4, and ≥5, and age at first FTP was categorized as <17, 17-19, 20-24, ≥25 years, and nulliparous. FTP was defined as a pregnancy lasting ≥26 weeks. Subjects with missing data in any conditioning variable were dropped from the corresponding analysis.

Results in the text are presented as RRs and their appropriate CIs. Where results are presented in the form of plots, RRs are represented by squares and their corresponding CIs/floating CIs by horizontal lines. The position of the square indicates the point estimate of the RR, and the area of the square is inversely proportional to the variance of the logarithm of the RR estimate, thus providing an indication of the amount of statistical information available for that particular estimate. Where summary RRs have been calculated, they are shown as open diamonds.

Thirty-six eligible studies were identified. Data could not be retrieved for 10 (13-23) and one research group declined to take part in the collaboration (24). Four of the 25 studies that joined the collaboration did not have information on sexual behavior (25-28), leaving 21 studies available for the present reanalysis (29-63).

Table 1 shows the studies included in the analysis, ordered by study design and, within studies of the same design, by median year of diagnosis of the cases. Three multicentric studies (WHO, Brinton Latin America, IARC) were counted as single studies, even when results from individual centers had been published separately. Likewise, two studies on invasive and in situ carcinoma in Milan were considered as two arms of a single study. Table 1 thus shows data on 21 studies in total, of which 16 separate studies or groups of studies included data on ICC and 15 studies included data on CIN3/carcinoma in situ. The combined study population included 10,773 ICC cases, 4,688 cases of CIN3/carcinoma in situ, and 29,164 women without carcinoma of the cervix. When a study included both invasive and in situ cases, the same controls were used (n = 19,121) for both outcomes. The median year at diagnosis by study ranged between 1980 and 1999, and the median age at diagnosis ranged between 26 and 55 years. Among control women, the proportion reporting ≥2 sexual partners ranged between 1% and 94% and the median age at first intercourse varied between 15 and 23 years. Table 1 also shows the studies that included assessment of HPV infection in both cases and controls. Five studies of ICC included HPV testing by PCR.

Table 2 shows the associations between sexual factors and potential confounding variables among controls. The associations are quantified by the ORs for having ≥2 versus 1 sexual partner and for having first intercourse at age <19 years versus ≥19 years. There was a strong association between the two sexual behavior variables: women with first intercourse at age <17 years were more likely to have more than one sexual partner than women with first intercourse at age ≥25 years (OR, 8.75; 95% CI, 7.63-10.03). Women with ≥3 FTPs were less likely to have more than one sexual partner than nulliparous women (OR, 0.24; 95% CI, 0.20-0.28) and more likely to have first intercourse before age 19 years (OR, 7.65; 95% CI, 6.74-8.68). Women with earlier first FTP were less likely to have more than one sexual partner. Among parous women, there was a strong correlation between age at first intercourse and age at first FTP, because, in many of the study populations, first FTP occurred shortly after first intercourse.

The effect on the RR estimates of conditioning on different confounding variables was examined in a subset of 5,383 cases and 9,500 control women for whom data on all potential confounders were available. Of the many potential confounding factors considered, age at first intercourse reduced the effect of lifetime number of sexual partners (Table 3). The converse was also true. Conditioning on reproductive factors did not strongly affect the RR estimate for lifetime number of sexual partners, but the RR estimate for age at first intercourse was strongly attenuated and the width of the CI greatly increased so that the effect of age at first intercourse was no longer significant in this subset. Conditioning on other factors such as education, tobacco smoking, hormonal contraceptive use, and history of Pap smear did not materially change the RRs or the CIs. Due to the strong effect of number of FTPs and age at first FTP on the RR estimates, it was decided to control for these reproductive factors in all further analyses. However, because the strong association between age at first intercourse and age at first FTP may result in overadjustment, results are also presented without controlling for reproductive factors.

Figure 1 shows the RR of ICC by lifetime number of sexual partners. The RR for ≥6 partners versus 1 was 2.27 (95% CI, 1.98-2.61) conditioned on age at first intercourse and increased to 2.78 (95% CI, 2.22-3.47) when additionally conditioned on reproductive factors. The fully adjusted RR for ≥11 sexual partners versus 1 in the studies that allowed finer classification was 3.15 (95% CI, 2.19-4.52).

Figure 2 shows the RR of ICC by age at first intercourse. The risk increased steadily with decreasing age at first intercourse, but the RRs conditioned on lifetime number of sexual partners were substantially reduced after additional conditioning on reproductive factors. The RR for age at first intercourse ≤14 versus ≥25 years was 3.52 (95% CI, 3.04-4.08) conditioning on lifetime number of sexual partners and 2.05 (95% CI, 1.54-2.73) after additional conditioning on reproductive factors.

Lifetime number of sexual partners and age at first intercourse were examined in combination (Fig. 3). The RR for women who reported both ≥2 sexual partners and age at first intercourse <17 years compared with those who reported both 1 partner and an age at first intercourse of ≥25 years was 5.94 (95% CI, 5.24-6.74). This RR decreased to 3.60 (95% CI, 2.95-4.38) after additional conditioning on reproductive factors. A relation between age at first intercourse and ICC risk was also seen among nulliparous women (RR for age at first intercourse <17 versus ≥25, 2.28; 95% CI, 1.62-3.19) and in different strata of age at first FTP (Fig. 4).

The consistency of the associations between ICC and lifetime number of sexual partners or age at first intercourse across individual studies and by study design is shown in Supplementary Fig. S1 and S2. The RR for ≥2 sexual partners versus 1, conditioned on age at first intercourse and reproductive factors, showed significant heterogeneity between individual studies and was lower in case-control studies with hospital controls (1.47; 95% CI, 1.29-1.67) than studies with population controls (RR, 2.89; 95% CI, 2.25-3.71; Supplementary Fig. S1). However, women with ≥2 sexual partners reported a much lower median lifetime number of sexual partners (2 for both cases and controls) in hospital-based case-control studies than the corresponding women in population-based case-control studies (four among cases and five among controls; Supplementary Fig. S1). Only one cohort study (Guanacaste) could be included and had too few cases to estimate the RR controlling for all the variables in the model.

The overall RR of ICC per 5-year decrease in age at first intercourse was 1.31 (95% CI, 1.21-1.41), conditioned on lifetime number of sexual partners and reproductive factors (Supplementary Fig. S2). RRs by age at first intercourse did not show significant heterogeneity by study or by study design (Supplementary Fig. S2).

Data on stage were available on 5,395 ICC cases from eight studies. A subgroup analysis of these subjects found no evidence of heterogeneity by stage for either age at first intercourse or lifetime number of sexual partners (data not shown).

Figures 5 and 6 show results analogous to Figs. 1 and 2 for the 15 studies that included cases of CIN3/carcinoma in situ. The pattern of risk by lifetime number of sexual partners was very similar to that for ICC in both shape and magnitude (Fig. 5). The RR for ≥6 versus 1 sexual partner was 2.51 (95% CI, 2.13-2.97) conditioning on age at first intercourse and 2.31 (95% CI, 1.83-2.92) after additionally conditioning on reproductive factors. The risk of CIN3/carcinoma in situ increased with decreasing age at first intercourse, as observed for ICC. The RR of CIN3/carcinoma in situ for age at first intercourse ≤14 versus ≥25 years was 2.33 (95% CI, 1.88-2.91) conditioning on lifetime number of sexual partners and decreased to 2.03 (95% CI, 1.41-2.91) after additional conditioning on reproductive factors.

Supplementary Figs. S3 and S4 show the consistency of the relationship between lifetime number of sexual partners and age at first intercourse and risk of CIN3/carcinoma in situ across different studies and study designs. As for ICC (Supplementary Fig. S1), the fully adjusted RR for ≥2 versus 1 sexual partner was higher in population-based case-control studies (RR, 3.23; 95% CI, 2.10-4.96) than in hospital-based case-control studies (RR, 1.51; 95% CI, 1.24-1.85). In cohort studies, it was 2.10 (95% CI, 1.12-3.93). Heterogeneity was significant across individual studies and study designs. This may be attributable to the higher lifetime number of sexual partners reported by women in cohort and population-based case-control studies than in hospital-based case-control studies (Supplementary Fig. S3).

The inverse association between CIN3/carcinoma in situ and age at first intercourse, conditioned on lifetime number of sexual partners and reproductive factors (RR per 5-year decrease in age at first intercourse, 1.32; 95% CI, 1.16-1.50) showed no significant heterogeneity between individual studies nor by study design (Supplementary Fig. S4).

This collaborative reanalysis of individual data from more than 15,000 women with ICC or CIN3/carcinoma in situ confirms the relationship between major indicators of sexual behavior and the risk of cervical carcinoma. On account of the large number of women involved, this reanalysis allowed the examination of the joint effect of two closely correlated sexual variables: lifetime number of sexual partners and age at first intercourse.

Various aspects of sexual behavior are related to the acquisition of HPV infection, the primary cause of cervical cancer. The probability of HPV transmission per sexual act with an infected partner is unknown, but available evidence suggests that it is higher than other viral sexually transmitted diseases (64). Furthermore, the prevalence of HPV infection increases rapidly among young women after they become sexually active (65). Hence, first exposure to HPV probably occurs soon after first intercourse in many women (66).

The majority of HPV infections become undetectable within 1 to 2 years of exposure, although it is unclear whether they resolve completely or remain in a latent state in the basal cell epithelium of the cervix (66). The onset of microscopically detectable precancer (e.g., CIN3) may occur rapidly after infection, possibly within 5 years (66). The rate at which CIN3 progresses to invasive cancer is not known precisely. The unique study in which CIN3 was left untreated by design showed a cumulative incidence of invasive cancer of ∼30% over 30 years (67). This is consistent with an earlier estimate, based on modeling of cancer mortality data in the United Kingdom, that 40% of untreated CIN3 will eventually progress to invasive cancer (68). This natural history of HPV infection can be interrupted if CIN3 is detected by screening (68) and is treated (67), but this is unlikely to have consistently occurred in most of the study populations included in this reanalysis, where screening was suboptimal or nonexistent during the study period.

With this understanding of the natural history of HPV infection, we interpret the increasing risk of both ICC and CIN3/carcinoma in situ with lifetime number of sexual partners as an effect of increased exposure to HPV infection. As might be expected, lifetime number of sexual partners has been found to be associated with HPV prevalence among women without cervical carcinoma (69).

The results on age at first intercourse are more difficult to interpret. One interpretation is in terms of confounding by reproductive factors. Age at first FTP and number of FTPs have previously been found to be strongly associated with ICC risk in our collaborative reanalysis (9). The strength of the association between age at first intercourse and ICC risk was substantially attenuated after controlling for age at first FTP, and a further influence of residual confounding cannot be ruled out. Nevertheless, an association between cervical carcinoma and age at first intercourse was found even among nulliparous women. Moreover, because first intercourse was shortly followed by first FTP in many of the study populations, controlling for reproductive factors may have resulted in overadjustment when examining the effect of age at first intercourse.

It is also conceivable that age at first intercourse is related to ICC risk through HPV acquisition. One possibility is that cervical cancer risk may increase with duration of HPV infection (47). As noted above, it is likely that women who have earlier first sexual intercourse are also exposed to HPV earlier. Therefore, among women of the same age, as our cases and controls were by design, those who had earlier first intercourse might have a longer duration of infection. A second possibility is that early first intercourse is a marker of high-risk behavior for HPV exposure. This would be consistent with the high incidence of HPV infection in young women shortly after the first sexual intercourse (65). A third possibility is that younger women are more susceptible to HPV infection (1, 2). Although this possibility cannot be ruled out, it is unlikely to involve a specific vulnerability of the cervix to HPV infections in the interval shortly after menarche because we observed a steady increase in ICC risk with decreasing age at first intercourse across the age range considered and not exclusively in the ages surrounding menarche. This conclusion is also supported by a prospective study of HPV infection risk in the first years after menarche (70).

Further interpretation of the effect of lifetime number of sexual partners and age at first intercourse on cervical cancer risk is made difficult by the fact that these variables do not fully describe a woman's risk profile for HPV infection. In many of the study populations, most women reported only one sexual partner. For these women, the risk of exposure to HPV, and consequently of developing cervical cancer, chiefly depends on the lifetime number of sexual partners of their husband (69). With few exceptions, we did not have data on husbands' sexual behavior or other data relevant to HPV transmission, such as condom use. Differences in these unmeasured variables may contribute to the heterogeneity between studies.

In addition to these problems of interpretation, there are the usual epidemiologic concerns over misclassification, bias, and confounding. The acceptability and reliability of questions on sexual behavior have always been of great concern (3) and data on the accuracy of self-reports of sexual behavior are scant (71). The data on sexual behavior were also incomplete. Among studies that contributed to the present collaborative reanalysis, four cohort studies did not include information on sexual behavior (25-28). Findings on sexual behavior have been published from eight studies that were not included and were consistent with our present results (14, 16-23).

The fact that the evidence on sexual behavior in the present study derives, in large part, from case-control studies raises the possibility of recall bias. The RR for each additional sexual partner was lower in hospital-based than in population-based case-control studies and in studies carried out in developing countries. This suggests a possible variation in the quality of the information available.

With respect to potential confounding factors other than reproductive factors, the strength of the associations we found between risk of cervical carcinoma and sexual factors was not materially affected by tobacco smoking, use of hormonal contraceptives and condoms, or history of Pap smear.

The consistency of the association between ICC and sexual behavior by histologic type has previously been investigated as part of a review of all risk factors considered by the collaboration (10). No evidence was found for a difference in between squamous cell carcinoma and adenocarcinoma for either age at first intercourse or lifetime number of sexual partners.

Previous publications by this collaboration on tobacco, reproductive factors, and hormonal contraceptives have included analyses restricted to HPV-positive women in the studies that used PCR-based detection of HPV (8-10). Supplementary Fig. S5 shows this subgroup analysis for lifetime number of sexual partners and age at first intercourse. Both associations were attenuated by restriction to HPV-positive women. Further control for reproductive factors rendered the relative risk estimates unstable and neither association was statistically significant. There are good reasons, however, for skepticism over the utility of this analysis, primarily due to the asymmetrical interpretation of HPV-positive findings in cases and controls. A current HPV infection in a case is almost certainly a long-term infection, possibly acquired at a much younger age, whereas a current infection in a control may be a recently acquired transient infection. Restriction to HPV-positive women does not, therefore, yield a subset of cases and controls with similar age at infection. The existence of a current HPV infection in a middle-aged woman gives little information about lifetime infection history, which is the information required to fully interpret the findings on sexual behavior.

Future prospective studies, including current trials of prophylactic vaccines against HPV16 and HPV18 (72), should shed further light on the natural history of HPV infection; however, they will seldom include lesions more severe than CIN3. Therefore, our present collaborative reanalysis represents an overview of the majority of information that has been, and probably ever will be, collected on sexual behavior and ICC.

No potential conflicts of interest were disclosed.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

T. Perdrix-Thoma provided skilful editorial assistance. We thank the women with and without cervical carcinoma that took part in this research.