Human papillomavirus and risk factors for cervical cancer in Chennai, India: A case-control study
To evaluate the role of human papillomavirus (HPV) and other risk factors in the aetiology of invasive cervical carcinoma (ICC), we conducted a hospital-based case-control study in Chennai, Southern India. A total of 205 ICC cases (including 12 adenocarcinomas) and 213 frequency age-matched control women were included. HPV DNA in cervical cells was evaluated by means of a polymerase chain-reaction assay. Odds ratios (ORs) and corresponding 95% confidence intervals (CIs) were computed by means of unconditional multiple logistic regression models. HPV infection was detected in all but one ICC cases and in 27.7% of control women (OR = 498). Twenty-three different HPV types were found. HPV 16 was the most common type in either cases or controls, followed by HPV 18 and 33. The association of ICC with HPV 18 and HPV 16-associated types was somewhat stronger than the one with HPV 16. Multiple HPV infections did not show a higher OR for ICC than single infections. Other than HPV infection, high parity (OR for >4 vs. ≤2 births = 7.3), a woman's report of her husband's extramarital sexual relationships (OR = 10.0) and early menopause (OR for <45 vs. ≥45 years = 4.2) were significantly associated with ICC, also after restricting the analysis to HPV-positive cases and controls. Poor hygienic conditions were associated with an increased risk of HPV infection among control women but not with ICC risk among HPV-positive women. A vaccine against HPV 16 and 18 may be effective in more than three-quarters of ICC in the study area. © 2003 Wiley-Liss, Inc.
Approximately 125,000 new cervical cancer cases have been estimated to have occurred in India in 2000 (i.e., nearly one-third of all new cervical cancer cases worldwide).1 Incidence rates in the country, however, varied between 11 per 100,000 in Trivandrum and 30 per 100,000 in Chennai (former Madras), all in Southern India.2 In Chennai, a woman's estimated probability of developing cervical cancer up to age 74 was more than 3%. As in other developing countries, squamous cell carcinoma predominates, whereas adenocarcinoma of the cervix accounts for a small percent of all cervical cancer, even in the urban areas (e.g., Mumbai, 7%).3 A high prevalence of human papillomavirus (HPV) 16 and 18 in cervical cancer specimens has been shown in India2 as elsewhere,4 but limited information is available on the distribution of HPV types in the country.
Case-control studies on invasive cervical carcinoma (ICC) in India have been relatively few.6, 7, 8, 9, 10, 11 Low educational and socioeconomic level, young age at marriage or first sexual intercourse, high parity and male partner's sexual behaviour were the most consistently reported risk factors. No case-control study from India has provided thus far an estimate of the association between cervical cancer and infection with HPV, the necessary cause of the disease worldwide.12
To further elucidate the role of different HPV types and its co-factors in ICC aetiology, we carried out a case-control study in Chennai, Southern India.
MATERIAL AND METHODS
Our study is part of an international multicenter case-control study of ICC and HPV coordinated by the International Agency for Research on Cancer (IARC).13 From June 1998 to May 1999, 222 women with a diagnosis of ICC were recruited into the Cancer Institute in Chennai, Southern India.14 On account of the high burden of ICC cases at the Cancer Institute, the first woman to be newly diagnosed with ICC each working day was asked to participate in our study and was administered an informed consent. Inclusion criteria for case subjects were: (i) histologic confirmation of ICC diagnosis; (ii) no previous cancer treatment; and (iii) lack of physical or mental impairments that would have made the interview impossible. Seventeen women were excluded based on the revision of histologic reports: 2 had no neoplastic lesions, 14 women had cervical intra-epithelial neoplasia (CIN) I or II and 1 had carcinoma in situ. Among 205 eligible ICC cases, the distribution by FIGO stage was the following: stage 1, 10.2%; stage 2, 44.9%; stage 3, 38.5%; stage 4, 6.3%. A squamous cell carcinoma was diagnosed in 193 cases and an adeno/adenosquamous carcinoma in 12. Twelve cancer cases refused to provide cervical samples for our study, although they were willing to be interviewed, leaving a total of 193 cervical cancer patients with cervical exfoliated cells available for HPV testing.
Control women were identified at the Cancer Institute among in-patients and visitors of patients other than women with cervical cancer, and were frequency-matched on age to ICC cases within 5-year age groups. Exclusion criteria for control women included: (i) a diagnosis of ano-genital tract cancers (i.e., cervix, vagina, vulva, anal canal), cancer of the breast, endometrium, ovary or colon, benign genital tumours and tobacco-related diseases (e.g., coronary heart disease, chronic bronchitis and cancer of the head and neck, lung and bladder); (ii) a history of hysterectomy or cervical conization; and (iii) physical or mental problems. A total of 179 visitors and 37 in-patients were thus contacted. Of the eligible 216 controls, 3 declined to participate in our study and 3 refused to have a gynaecologic examination. Thus, a sample of exfoliated cells from the cervix was available for 210 control women.
Data and specimen collection
Each participant was administered a standardized questionnaire on socioeconomic status, eating habits, paan chewing, sexual behavior, reproductive history, contraceptive practices, hygienic conditions and history of sexually transmitted infections and cervical cytologic screening. Two trained female interviewers administered the questionnaire to all case and control women in the hospital.
A pelvic examination was performed on ICC cases and control women to obtain exfoliated cells for the preparation of a Pap smear and determination of the presence of HPV. The material was collected by sampling the ectocervix with 2 wooden spatulas and the endocervix with a cytobrush. The collection instruments were rinsed in phosphate-buffered saline (PBS), and the cell suspension was centrifuged for 10 min at 3,000g at room temperature. The cell pellets were stored at −70°C in the field until they were shipped to Lyon, France, or to Amsterdam, the Netherlands, for testing. A 10 ml sample of peripheral blood and, from case subjects, tumour biopsy specimens were also collected and frozen immediately at −70°C.
Our study was approved by the ethical review committees of the Cancer Institute in Chennai and the IARC in Lyon.
Detection of HPV DNA
To analyze the quality of target DNA in exfoliated cells from cases and controls, beta (β)-globin gene-specific primers were used. Cervical biopsies were used in 30 ICC cases for whom exfoliated cells were unavailable or β-globin-negative. The remaining β-globin-negative and HPV-negative samples (i.e., 2 ICC cases and 26 control women) were considered inadequate and excluded from HPV-related analyses, thus leaving 191 invasive cancer cases (179 squamous cell carcinoma cases and 12 adeno/adenosquamous carcinoma cases) and 184 control women for HPV analysis.
The presence of HPV DNA in cervical cells was assessed using general primer-mediated GP5+/6+-PCR. Polymerase chain-reaction (PCR) positivity was assessed by means of hybridization of PCR products in an enzyme immunoassay (EIA) using 2 HPV oligoprobe cocktails that together detect the following 36 HPV types: HPV 16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 58, 59, 66, 68, 6, 11, 26, 34, 40, 42, 43, 44, 53, 54, 55, 57, 61, 70, 71 (equivalent to CP8061), 72, 73, 81 (equivalent to CP8304), 82 (IS39 and MM4 subtypes), 83 (equivalent to MM7), 84 (equivalent to MM8) and CP 6108. Probes and procedures used for EIA detection are described elsewhere.15 In addition, HPV positivity was assessed by low-stringency Southern blot analysis of PCR products with a cocktail probe of HPV-specific DNA fragments. Subsequently, GP5+/6+ PCR was repeated on positive samples in triplicate to generate sufficient products for further typing. After pooling these PCR products, typing was performed using EIA and HPV type-specific oligoprobes for the HPV types described above.16
Finally, E7 open-reading frame type-specific PCR assays for 14 high-risk HPV types (16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 58, 59, 66, 68)15 were applied to 5 ICC case specimens that were positive for β-globin but negative for HPV-DNA by GP5+/GP6+ PCR. Reamplification of specimens with E7 was not done among control women, since it is unlikely that HPV L1 open-reading frame had been disrupted by the integration of HPV DNA in the cellular genome in cancer-free women.17
For analysis purposes, HPV types were grouped into high-risk types and low-risk types.18 The HPV 16-associated types included those types phylogenetically associated with HPV 16 (i.e., HPV 31, 33, 35, 52 and 58) and HPV 18-associated types included those associated with HPV 18 (i.e., HPV 39, 45, 59 and 68).10
To estimate the risk of cervical cancer associated with various HPV types and the other risk factors, we calculated odds ratios (ORs) and 95% confidence intervals (CIs) by means of unconditional logistic regression. The ORs were adjusted for age and area of residence (Chennai and out of Chennai) and for other risk correlates, as reported. Tests for linear trend were performed, assigning an increasing score for each level of categorical variables. Scores were then fitted into the model as a continuous variable. To elucidate the influence of HPV infection on our findings, major risk correlates were finally reevaluated in a fully adjusted model: (i) among all cases and controls; (ii) among controls only, by presence of HPV infection; and (iii) among HPV-positive women only.
Table I shows that HPV infection was detected in all but one ICC cases (99.4% of squamous cell carcinomas and 100% of adenocarcinomas) and in 27.7% of control women. A total of 224 HPV infections were found among 191 ICC cases and 60 among 184 control women. Twenty-three different HPV types were identified in either single or multiple infections. HPV 16 was by far the most common type among cases as well as control women, followed by HPV 18 and 33. Only HPV 16 and 18 infections were found among the 12 adenocarcinoma cases. No ICC cases and 11 control women were infected with low-risk HPV types only.
Table I. Distribution of HPV Infections by HPV Type and Multiplicity of Infection Among 191 Cases of Cervical Carcinoma and 184 Controls,1 Chennai, India, 1998–99
| High-risk types2||178||99.4||12||100.0||40||21.7|
| Low-risk types||-||-||-||-||11||6.0|
|HPV types|| || || || || || |
| Single infections||148||82.7||12||100.0||44||23.9|
| Multiple infections||30||16.8||-||-||7||3.8|
ORs for cervical carcinoma according to the presence of different HPV types or groups of types are shown in Table II. HPV infection of any type was associated with 498-fold increased ICC risk. The presence of multiple HPV infections did not confer an increased risk compared to a single HPV infection (OR = 1.2). To compare the strength of the association of ICC with different HPV types (alone or in multiple infections), types were ranked as in Table II. Compared to women who were infected by HPV 16, those infected with HPV 18 showed an increased OR of 3.9, of borderline statistical significance. Also the types associated with HPV 16 (OR = 7.1; 95% CI 0.9–54.4) but not those associated with HPV 18 (OR = 0.6) showed a more elevated ICC risk than HPV 16. The OR for all other types was 0.3 (95% CI 0.1–0.7).
Table II. OR and Corresponding 95% CI by Type(s) of HPV Infection Among 191 Cases of Cervical Carcinoma and 184 Controls,1 Chennai, india, 1998–99
|HPV|| || || || || |
| Positive (any type)||190||(99.5)||51||(27.7)||497.9 (67.7–999)|
|Multiple infection|| || || || || |
| Yes||30||(15.7)||7||(3.8)||1.2 (0.5–3.0)|
|HPV type(s)|| || || || || |
| HPV 16||115||(60.2)||32||(17.4)||13|
| HPV 18||28||(14.7)||2||(1.1)||3.9 (0.9–17.4)|
| HPV 16 and HPV18||5||(2.6)||2||(1.1)||0.7 (0.1–3.8)|
| HPV 16-associated types4||25||(13.1)||1||(0.5)||7.1 (0.9–54.4)|
| HPV 18-associated types5||8||(4.2)||4||(2.2)||0.6 (0.2–2.0)|
| Other types||9||(4.7)||10||(5.4)||0.3 (0.1–0.7)|
Table III shows ORs and corresponding 95% CIs according to age, demographic characteristics and selected indicators of poor hygienic conditions. Illiteracy (OR vs. high education = 4.8), lack of a toilet (OR = 4.8) or running water (OR = 2.0) inside the house and not washing genitals after sexual intercourse (OR = 4.5) were significantly associated with ICC risk. An increased ICC risk was also found among widows (OR vs. currently married women = 8.3) and separated or divorced women (OR = ∞) (Table III).
Table III. OR and corresponding 95% CI1 of Cervical Carcinoma Among 205 Cases and 213 Controls According to Age, Sociodemographic Factors and Hygienic Conditions, Chennai, India, 1998–99
|Age (years)|| || || |
| <41||59 (28.8)||53 (24.9)|| |
| 41–47||47 (22.9)||54 (25.4)|| |
| 48–55||55 (26.8)||55 (25.8)|| |
| ≥56||44 (21.5)||51 (23.9)|| |
|Education|| || || |
| High||22 (10.7)||62 (29.1)||13|
| Primary||22 (10.7)||48 (22.5)||1.4 (0.7–2.8)|
| Illiterate||161 (78.5)||103 (48.4)||4.8 (2.7–8.4)|
|χ2 or trend1|| || ||35.28; p = 0.00|
|Inside toilet|| || || |
| Yes||30 (14.6)||96 (45.1)||13|
| No||175 (85.4)||117 (54.9)||4.8 (3.0–7.8)|
|Running water|| || || |
| Yes||60 (29.3)||97 (45.5)||13|
| No||145 (70.7)||116 (54.5)||2.0 (1.3–3.0)|
|Wash genitals after intercourse|| || || |
| Often||185 (90.2)||208 (97.7)||13|
| Sometimes or never||20 (9.8)||5 (2.4)||4.5 (1.7–12.4)|
|Marital status|| || || |
| Married||143 (69.8)||201 (94.4)||13|
| Widowed||55 (26.8)||12 (5.6)||8.3 (4.2–17.6)|
| Separated-divorced||7 (3.4)||0 (0.0)||∞ (4.2–∞)|
Age at first intercourse <15 years (OR vs. ≥21 years = 2.2) and having had ≥2 lifetime sexual partners (OR vs. 1 = 4.0) were significant risk factors for ICC (Table IV). The use of oral contraceptives (0 cases and 11 controls) and intra-uterine device (3 cases and 11 controls) was rare and unrelated to ICC risk (data not shown). Tubal ligation was the most commonly reported contraceptive method (47 cases and 67 controls) and was associated with an OR of 0.6 (95% CI 0.4–1.0) (data not shown). With respect to a woman's report of her husband's extramarital sexual relationships and sexual intercourse with prostitutes, ORs were 4.3 and 5.8 for women who were uncertain and 8.7 and 10.5, respectively, for those who answered affirmatively (Table IV). Significant associations with ICC risk also emerged for a husband's history of syphilis (23 cases and 6 controls, OR = 4.7); gonorrhoea (22 cases and 2 controls, OR = 13.9); genital warts (16 cases and 2 controls, OR = 10.1); and herpes genitalis (9 cases and 0 controls, OR = ∞) (data not shown). No cases or control women reported a personal history of sexually transmitted diseases.
Table IV. OR and corresponding 95% CI1 of Cervical Carcinoma Among 205 Cases and 213 Controls According to Sexual Habits and Husband's Extramarital Sexual Relationships, Chennai, India, 1998–99
|Age at first intercourse (years)|| || || |
| ≥21||20 (9.8)||33 (15.5)||13|
| 19–20||36 (17.7)||41 (19.3)||1.4 (0.7–2.9)|
| 17–18||55 (27.0)||68 (31.9)||1.3 (0.7–2.5)|
| 15–16||47 (23.0)||39 (18.3)||1.9 (0.9–3.8)|
| <15||46 (22.6)||32 (15.0)||2.2 (1.1–4.6)|
|χ2 for trend|| || ||5.39; p = 0.02|
|No. of sexual partners|| || || |
| 1||195 (95.1)||210 (98.6)||13|
| ≥2||10 (4.9)||3 (1.4)||4.0 (1.1–14.8)|
|Husband's sexual habits|| || || |
|Extramarital sex|| || || |
| No||61 (30.0)||156 (73.6)||13|
| Uncertain||43 (21.2)||25 (11.8)||4.3 (2.4–7.8)|
| Yes||99 (48.8)||31 (14.6)||8.7 (5.2–14.6)|
|Intercourse with prostitutes|| || || |
| No||63 (31.0)||166 (78.3)||13|
| Uncertain||48 (23.7)||21 (9.9)||5.8 (3.2–10.6)|
| Yes||92 (45.3)||25 (11.8)||10.5 (6.1–18.2)|
The influence of menstrual and reproductive factors is shown in Table V. Age at menarche was not associated with ICC risk, whereas women who reported 7 or more births had a higher risk than those with 1 or 2 births (OR = 5.7). The OR for age at first pregnancy <17 years (vs. ≥19 = 2.2) was lowered to 1.7 (nonsignificant) after adjustment for number of pregnancies (data not shown). Spontaneous abortions showed a risk pattern similar to births. An inverse trend in risk emerged among the few women who reported induced abortions. Menopause <45 years, compared to ≥50 years, was associated with an OR of 8.8 (Table V). After further adjustment for body mass index, which was associated directly with age at menopause but inversely with ICC risk, the OR for age at menopause <45 years became 6.3 (95% CI 2.3–17.0, data not shown).
Table V. OR and Corresponding 95% CI1 of Cervical Carcinoma Among 205 Cases and 213 Controls According to Menstrual and Reproductive Factors, Chennai, India, 1998–99
|Age at the menarche (years)|| || || |
| ≥15||76 (37.1)||82 (39.0)||13|
| 13–14||84 (41.0)||90 (42.3)||0.9 (0.6–1.5)|
| <13||45 (22.0)||41 (19.3)||1.1 (0.7–1.9)|
|χ2 for trend|| || ||0.15; p = 0.70|
|Age at first pregnancy (years)|| || || |
| ≥19||93 (45.6)||115 (56.9)||13|
| 17–18||52 (25.5)||55 (27.2)||1.1 (0.7–1.8)|
| <17||59 (28.9)||32 (15.8)||2.2 (1.3–3.7)|
|χ2 for trend1|| || ||7.83; p = 0.01|
|No. of births|| || || |
| None||3 (1.5)||12 (5.6)||0.5 (0.1–2.1)|
| 1–2||37 (18.1)||78 (36.6)||13|
| 3–4||91 (44.4)||88 (41.3)||2.6 (1.6–4.3)|
| 5–6||50 (24.4)||23 (10.8)||5.7 (3.0–11.1)|
| ≥7||24 (11.7)||12 (5.6)||5.7 (2.4–13.3)|
|χ2 for trend1|| || ||29.11; p < 0.0001|
|Spontaneous abortion|| || || |
| None||172 (84.3)||178 (88.1)||13|
| 1||22 (10.8)||19 (9.4)||1.2 (0.6–2.4)|
| ≥2||10 (4.9)||5 (2.5)||2.1 (0.7–6.5)|
|χ2 for trend1|| || ||1.98; p = 0.16|
|Induced abortion|| || || |
| None||190 (92.7)||167 (78.4)||13|
| 1||11 (5.4)||30 (14.1)||0.3 (0.2–0.7)|
| ≥2||4 (2.0)||16 (7.5)||0.2 (0.1–0.7)|
|χ2 for trend1|| || ||13.2; p = 0.0003|
|Age at menopause (years)|| || || |
| ≥50||28 (23.7)||36 (46.2)||13|
| 45–49||31 (26.3)||33 (42.3)||1.2 (0.6–2.4)|
| <45||59 (50.0)||9 (11.5)||8.8 (3.7–20.8)|
|χ2 for trend1|| || ||23.9; p = 0.00|
Table VI shows the ORs from 3 different multiple logistic regression models where all the strongest risk factors in the age- and area of residence-adjusted models were included simultaneously. Among all study women, all major risk factors for ICC retained a strong effect, except for age at first sexual intercourse. Lack of inside toilet and husband's extramarital sexual relationships were significantly associated with HPV infection among control women. In the comparison restricted to HPV-positive cases and HPV-positive controls, a risk pattern similar to the one among all study women emerged, except for inside toilet that was no longer associated with ICC. ORs for education, being a widow, separated or divorced, having had ≥5 births, menopause <45 years and a woman's report of her husband's extramarital sexual relationships tended to be, if anything, higher among HPV-positive women than in the overall study population.
Table VI. OR and Corresponding 95% CI1 According to Selected Characteristics Among all Women, Controls Only, by HPV Infection and HPV-Positive Women Only, Chennai, 1998–99
|Education|| || || || || || |
| Illiterate||159/102||2.1 (1.1–3.9)||23/63||0.5 (0.2–1.1)||148/23||6.8 (2.5–18.8)|
|Marital status|| || || || || || |
| Widow, separated, divorced||62/12||5.3 (2.3–12.1)||2/8||0.5 (0.1–3.5)||56/2||10.9 (1.8–65.5)|
|Age at first intercourse (years)|| || || || || || |
| 17–20||90/109||0.9 (0.4–2.2)||30/69||0.9 (0.3–2.6)||86/30||1.1 (0.3–4.4)|
| <17||92/70||0.6 (0.2–1.5)||14/41||0.6 (0.2–2.3)||83/14||0.7 (0.2–3.1)|
|χ2 for trend1||2.10; p = 0.15||0.59; p = 0.44||0.44; p = 0.51|
|No. of births|| || || || || || |
| 3–4||90/88||3.8 (2.0–7.4)||25/52||1.2 (0.6–2.6)||85/25||4.0 (1.5–10.5)|
| >4||73/35||7.3 (3.3–16.0)||2/28||0.2 (0.0–0.8)||69/2||94.4 (12.3–725.3)|
|χ2 for trend1||25.20; p < 0.001||2.72; p = 0.10||22.46; p < 0.001|
|Age at the menopause|| || || || || |
| <45||57/9||4.2 (1.5–11.9)||1/7||0.2 (0.0–2.7)||48/1||7.4 (0.7–77.7)|
| Premenopause||86/135||0.5 (0.2–1.2)||30/84||0.3 (0.1–1.2)||83/30||1.0 (0.2–5.4)|
|Inside toilet|| || || || || || |
| No||173/116||3.8 (2.0–7.4)||37/66||4.0 (1.7–9.4)||159/37||0.9 (0.3–2.8)|
|Husband's extramarital sex4|| || || || || || |
| Uncertain||45/25||4.9 (2.3–10.3)||10/15||3.8 (1.2–11.7)||40/10||1.5 (0.5–4.8)|
| Yes||99/31||10.0 (5.1–19.5)||9/17||2.4 (0.8–7.2)||94/9||12.4 (3.7–41.3)|
HPV infection is confirmed in our study to be a necessary event for the development of cervical carcinoma. All but one cervical carcinoma cases were positive for HPV DNA. The types most often found among cases (i.e., in the order, HPV 16, 18, 33, 35 and 45) are consistent with the type distribution shown in cervical carcinoma worldwide.5, 18 HPV 58 also seems relatively frequent, as found particularly in studies from Asia.5 The prevalence of HPV infection among control women (27.7%) and, notably, the frequency of HPV 16 and other high-risk types were among the highest reported so far, at least in a middle-aged population.19, 20, 21 Multiple HPV infections were found among approximately one-fifth of HPV-positive cases and control women and were not associated with more elevated OR than single infections. In respect to HPV type, the association of cervical carcinoma with HPV 18 seemed to be somewhat stronger than the one with HPV 16. Compared to other high-risk types, HPV 18 has been associated with increased oncogenic potential in cell culture,22 faster transition to malignancy23 and poorer ICC prognosis.24 In addition, HPV 18 has been shown to be 1.7-fold more frequent in ICC than in high-grade squamous intra-epithelial lesions25 and in adenocarcinoma than in squamous cell carcinoma of the cervix.5 Our study was too small to provide an accurate estimate of the OR for rare HPV types, but those phylogenetically associated with HPV 16 also tended to show higher ORs than HPV 16.
Other than HPV, indicators of poverty, high parity and the sexual behaviour of a woman's husband seemed to be the most important risk factors for cervical carcinoma in our study. An excess of cervical cancer among low socioeconomic class women is one of the earliest and most consistent findings of epidemiologic studies of this tumour.26 The difference between social classes in Chennai is certainly not attributable to different access to cervical cancer screening since only one woman (a case) reported to have ever had a Pap smear. Among the many correlates of low socioeconomic status we found associated with cervical cancer risk in Chennai (e.g., illiteracy, poor nutrition, underweight, paan chewing),14 indicators of poor hygienic conditions showed an especially strong effect. A direct role of poor hygienic conditions on risk of HPV infection and cervical cancer has been reported.27 In our study, however, poor hygienic conditions were associated with a higher prevalence of HPV infection among control women but not with an increased ICC risk among HPV-positive women. Smoking was reported very rarely (2 ICC cases and 0 control women), whereas paan chewing was reported by 18.5% of cases and 7.0% of control women.14 Chewing was more common among less-educated women but a 2-fold increased ICC risk was also found after adjustment for education, occupation, and other major risk factors.
Regarding the influence of parity in a pooled analysis of 10 case-control studies coordinated by IARC,13 women who had had 5–6 or 7 or more full-term pregnancies showed, compared to nulliparous women, ORs of 5.0 and 8.3, respectively. A similar, although apparently weaker, association with parity was reported in developed countries,28, 29 where a smaller parity range can be studied. Little progress has been made to explain the adverse effect of full-term pregnancies, but hormonal influences on biology of HPV and labour-related trauma have been suggested as possible mechanisms.13 Also in agreement with previous findings,13 the apparent influence of early age at first pregnancy was diminished by adjustment for number of pregnancies. Very few women in our study reported use of contraceptive methods other than tubal ligation, and no relationship with ICC risk emerged. The strong inverse association we found with age at menopause is a new finding.13 The adverse effect of early menopause was not explained by low body mass index, another risk factor for ICC in our study.14
A large number of studies have shown a relationship between sexual activity indicators and ICC risk,8, 26, 30 as expected since the necessary cause of the disease, HPV, is sexually transmitted. In our study, however, a very small proportion of women reported 2 or more lifetime sexual partners and age at first sexual intercourse or first marriage were not significantly associated with cancer risk after allowance for sociodemographic and reproductive factors. Being a widow or separated or divorced, characteristics associated with a marked excess risk for cervical carcinoma in our study, was not associated with HPV infection among control women. In agreement with previous work,10, 31 a husband's sexual behaviour retained a great influence on ICC risk even in the fully adjusted models and in the analysis restricted to HPV-positive women. It was also associated with a higher prevalence of HPV infection among control women. The reliability of a wife's reporting of her husband's extramarital sexual relationships and STIs and the scope for recall bias are unknown.
As in all hospital-based case-control studies, we cannot rule out selection bias. As we had observed in a case-control study on oral cancer in Southern India,32 it is conceivable that the poorest, illiterate women go to the hospital for advanced cancers, but they seldom attend for less severe conditions or to visit relatives and friends. This may partly explain the strength of some of the associations that we found, notably with some socioeconomic indicators, but do not eclipse our major findings related to HPV.
In conclusion, although large studies that are able to adjust or stratify for HPV status are becoming available,13, 33 the best way to account for the effect of HPV in the evaluation of other risk factors remains an open issue. One way of potentially eliminating confounding by HPV infection is to restrict analyses to HPV-positive women. When we did so, the role of education, marital status, high parity, early menopause and a woman's report of her husband's extramarital sexual relationships were similar to those obtained without such restriction. Conversely, the apparent effect of poor hygienic conditions disappeared. HPV 16 and 18 accounted for three-quarters of cervical cancer cases. Vaccines against HPV 16 and 18, currently under development,34 therefore offer hope for the control of the majority of the disease in this area, if they can be produced and delivered at an affordable cost.
The authors thank Ms. A. Arslan and Dr. J. Smith for technical assistance and comments.