Burning wood in the kitchen increases the risk of cervical neoplasia in HPV-infected women in Honduras

Authors


Abstract

There is suggestive evidence that the use of wood for cooking increases the risk of invasive cervical cancer. We investigated this association in women with cervical neoplasia in Honduras. Women aged 20–64 years with cervical intraepithelial neoplasia (CIN) grade I (n = 44), CIN II (n = 36) or CIN III (n = 45) were recruited from screening programs in Tegucigalpa City and each was matched by age and clinic to 2 controls (241 total) without cervical abnormalities. The clinics selected women of low socioeconomic status. Cervical scrapes were tested for the presence of human papillomavirus (HPV) DNA using a general primer set directed against the L1 open reading frame, and HPV genotyping was performed. Odds ratios (ORs) were computed through conditional logistic regression; p-values were from tests for linear trend of risk with increasing exposure. HPV DNA was detected in 48% of women with CIN I, 67% with CIN II and 89% with CIN III. The ORs were 1.5, 2.5 and 38.3 respectively. At univariate analysis, age at first intercourse was consistently lower among cases than controls. Risk was reduced by 50% or more in all 3 CIN classes when initiation of sexual activity at age 20 years or older was compared with initiation before age 16 years (p = 0.013 for CIN I). No effect was observed for smoking, oral contraceptives or previous cytologic screening. Effects for number of sexual partners, parity, age at first pregnancy and education were in the expected directions but never persisted after adjustment for HPV. Chronic exposure to wood smoke significantly increased the risk of CIN III (p = 0.022). However, women who said “No” when asked if they ever used wood in the kitchen had a higher risk than those with low or intermediate exposure. This was taken as evidence that the initial screening question had either been misunderstood or that answers were biased. Restricting the analysis to women who reported exposure yielded positive associations in all CIN classes with for CIN III ORs of 2.3 for 25–34 and 9.5 for 35+ years compared with women who had 1–14 years of exposure (p = 0.017). A multivariate analysis of the complete dataset (n = 366) allowed for separate ORs for HPV in each CIN class. Inclusion of age at first intercourse significantly improved this model (p = 0.021). Adding exposure to wood smoke further improved the model only if an interaction between woodsmoke and HPV was allowed for. If, as the data suggest, it was assumed that wood smoke had its effect among HPV-positives only, there was a significant linear dose-response relationship between exposure to woodsmoke and risk of CIN (p = 0.026). This association was independent of other risk factors including education, parity and number of sexual partners. ORs in the final model were 0.37 for age at first intercourse 20 years or higher and 5.69 for more than 35 years of exposure to wood burning in the kitchen. The present study suggests that the use of wood for cooking is a risk factor for cervical neoplasia that deserves further study, given its high prevalence in developing countries. © 2001 Wiley-Liss, Inc.

Inhalation of air contaminated by chemical substances as a cause of cancer is a well-known concept. Inhalation of air contaminated through the combustion of tobacco (i.e., cigarette smoking) has been amply documented to cause cancer and also cervical neoplasia and invasive cancer.1–5 Inhalation of air contaminated through the combustion of coal, wood and straw has been related to lung cancer6, 7 and cancers of the head and neck.8, 9 In a case-control study of invasive cervical cancer in Honduras,10 we presented evidence suggestive of an increased risk of invasive cervical cancer among women who had used wood as fuel in the kitchen over many years. Human papillomavirus (HPV)-positive women who had been exposed to woodsmoke in the kitchen for 35 years or more had a 4 times higher risk than women who had been exposed for less than 15 years.

In the same research project, we also collected data from women with cervical intraepithelial neoplasia (CIN) and matching controls. The relationship between type-specific HPV status and risk has already been documented.11 Here we report our findings concerning other risk factors, particularly the burning of wood in the kitchen, which remains a highly prevalent practice among women of low socioeconomic status in Honduras.

MATERIAL AND METHODS

Subjects

The present study is part of a wider HPV and cervical neoplasia project in Honduras10–13 in which women with cervical dysplasia or neoplasia are compared with women without cervical abnormalities as determined by cytologic screening. Women with cervical dysplasia or cancer in situ who had not been treated previously for the disease were recruited from 14 public hospitals, family planning clinics and other health centers where cytologic screening was offered. These centers were all located in Tegucigalpa, the capital of Honduras and were serving populations of low socioeconomic status. Overall, 58% of the cases were detected at the San Felipe cancer center and another 17% at a major family planning clinic. Cases from other clinics (25%) were often detected in the process of selecting control women for the present study or for another study. They were then evaluated at the San Felipe Hospital through a Pap smear, colposcopic inspection of the cervix and a colposcopically directed cervical biopsy. The histologic diagnosis based on the biopsy decided the classification of each case in the study. Cases accrued over the years 1993–1995, but 65% of cases of CIN I and CIN II were recruited in 1995.

Controls were individually matched to cases by age. Each control was recruited within 1 month of detection of the case from the same clinic where the case was first seen and had the same age (±5 years). When an interviewer visited a clinic to select controls, all eligible women were selected from the women scheduled for cytologic screening that day and 2 for each case were randomly selected for inclusion in the study. Controls had a normal cervix and a Pap smear that exhibited no abnormalities or only inflammatory changes and had never been treated for cervical abnormalities.

All women gave informed consent and none refused participation.

Data collection

Interviews took place in the clinics and were conducted by a trained nurse or social worker responsible for all data collection. The interviews lasted approximately 20 minutes.

Each woman provided 2 cytologic scrapes taken from the transformation zone with an Ayre spatula; 1 smear was prepared for staining by the Papanicolau method and the remaining material was processed for virologic studies. Smears were read by 1 cytotechnician and all positive slides plus a sample of negative slides were reviewed by a single pathologist.

Methods for detection of HPV have been described in detail previously.11 After DNA extraction in Honduras, PCR was performed in The Netherlands using a set of general primers directed against the L1 open reading frame,14 followed by PCR with type-specific primers and sequencing if necessary. To prevent contamination, strict spatial partitioning of the different technical steps of the PCR was practised and the recommendations of Kwok and Higuchi15 were followed.

Statistical analysis

Odds ratios (ORs) quantifying the association of potential risk factors with disease status were estimated separately for each CIN class by means of conditional logistic regression,16 taking into account the individual matching of cases and controls by age, clinic and calendar time. Dummy variables were used to estimate the ORs for each exposure category separately.

Most variables were categorized and categories were assigned numbers 1, 2, 3, etc. This linear variable was used to assess the statistical significance of the linear trend of risk with successive exposure categories by means of the likelihood ratio test, which in this situation has a chi-square distribution with 1 degree of freedom. If comparison of the goodness-of-fit of models with a linear variable and with dummy variables suggested departure from linear trend, significance was assessed by adding the full set of dummy variables to the model. In this situation the likelihood ratio test has a chi-square distribution with degrees of freedom equal to the number of categories minus 1. Unless otherwise indicated in the tables, chi-square statistics were of the former type, with 1 degree of freedom.

Categorization of exposure variables

Parity was computed as the sum of completed pregnancies and abortions, whether spontaneous or induced. Only 2 women in our study had never been pregnant and these were excluded from the analyses involving parity.

Women answered 3 questions regarding the use of wood in the kitchen, as follows: “Have you used wood for cooking?” and if yes, “for how many years have you used it?” and “for how many years have you no longer cooked with wood?”. A last question inquired into the use of wood in the kitchen during childhood. From these variables, the exposure to woodsmoke after childhood was conservatively estimated by subtracting the years that wood was no longer used from the number of life years after age 14 and reducing this number further if the reported number of years of use was smaller. The minimum exposure time was set at 1 year. Zero years of exposure was assumed if the initial question was answered with “No.”

Women were asked how many times they had undergone cytologic screening previously and how long ago the most recent screen had taken place. There was a suspicion that women included 1 or both of the screening events related to their enrollment into the study when reporting the number of previous cytologic screens. The number of previous screens was therefore reduced by 1 in the analysis if the most recent event was reported to have taken place within 12 months prior to the interview.

RESULTS

A total of 366 women were enrolled in the study, 44 with CIN I, 36 with CIN II and 45 with CIN III; each woman was matched with 2 controls (241 total) by age and clinic. Their ages varied from 17 to 64 years with a mean (± SE) of 35.1 (± 1.32) years for CIN I, 35.6 (± 1.59) years for CIN II and 40.3 (± 1.54) years for CIN III. Four women were younger than 20 years, of whom 1 case had CIN II, and 4 women were older than 60 years, including 2 cases (both CIN III). The clinics selected women of low socioeconomic status. Thirteen percent of women had not completed primary school, and 36% had continued education beyond primary school. Women were of mixed race (mestizo), and 91% reported family incomes below 125 US dollars per month. Only 12% of women had ever smoked and 8% were current smokers; no significant associations of smoking with outcome were observed.

Univariate analysis

The distribution of cases and controls by CIN class over various potential risk factors is shown in Table I. The ratio of controls to cases was 1.93. Matched ORs and 95% confidence intervals (CIs) are presented in Table II. HPV was detected in 48% of CIN I cases, in 67% of CIN II cases and in 89% of patients with carcinoma in situ. Matched ORs increased from 1.48 for CIN I to 2.49 for CIN II and 38.28 in CIN III. The association was statistically significant for CIN II and CIN III.

Table I. Distribution of Cases and Controls Over Risk Factors for Cervical Neoplasia, Honduras, 1993–19951
 CIN ICIN IICIN III
CasesControlsCasesControlsCasesControls
  • 1

    Data are numbers, with percents in parentheses. CIN, cervical intraepithelial neoplasia; HPV, human papillomavirus.

Education
 None6 (13.6)8 (9.3)6 (16.7)4 (5.9)8 (17.8)16 (18.4)
 Primary26 (59.1)51 (59.3)23 (63.9)39 (57.4)29 (64.4)55 (63.2)
 Secondary+12 (27.3)27 (31.4)7 (19.4)25 (36.8)8 (17.8)16 (18.4)
Parity
 1–424 (54.5)51 (59.3)23 (65.7)43 (64.2)13 (28.9)39 (44.8)
 5, 615 (34.1)24 (27.9)7 (20.0)15 (22.4)11 (24.4)23 (26.4)
 7–105 (11.4)10 (11.6)3 (8.6)8 (11.9)14 (31.1)19 (21.8)
 11+0 (−)0 (−)2 (5.7)1 (1.5)7 (15.6)6 (6.9)
No. of sex partners
 120 (45.5)43 (50.0)11 (30.6)33 (48.5)12 (26.7)31 (35.6)
 2–322 (50.0)38 (44.2)20 (55.6)31 (45.6)25 (55.6)45 (51.7)
 4+2 (4.5)5 (5.8)5 (13.9)4 (5.9)8 (17.8)11 (12.6)
Age at 1st intercourse (yr)
 12–1512 (27.3)18 (20.9)10 (27.8)14 (20.6)15 (33.3)20 (23.0)
 16–1715 (34.1)19 (22.1)10 (27.8)15 (22.1)15 (33.3)27 (31.0)
 18–1911 (25.0)21 (24.4)8 (22.2)18 (26.5)10 (22.2)25 (28.7)
 20+6 (13.6)28 (32.6)8 (22.2)21 (30.9)5 (11.1)15 (17.2)
Age at 1st pregnancy (yr)
 12–1717 (38.6)28 (33.3)19 (52.8)24 (36.4)20 (44.4)33 (37.9)
 18–1915 (34.1)19 (22.6)6 (16.7)14 (21.2)14 (31.1)24 (27.6)
 20+12 (27.3)37 (44.0)11 (30.6)28 (42.4)11 (24.4)30 (34.5)
Previous screening
 None4 (9.1)6 (7.0)2 (5.6)4 (6.0)7 (15.6)13 (15.5)
 1–414 (31.8)27 (31.4)12 (33.3)23 (34.3)16 (35.6)17 (20.2)
 5–916 (36.4)39 (45.3)13 (36.1)29 (43.3)11 (24.4)32 (38.1)
 10+10 (22.7)14 (16.3)9 (25.0)11 (16.4)11 (24.4)22 (26.2)
HPV
 Negative23 (52.3)54 (62.8)12 (33.3)37 (54.4)5 (11.1)52 (59.8)
 Positive21 (47.7)32 (37.2)24 (66.7)31 (45.6)40 (88.9)35 (40.2)
Oral contraceptives
 Never25 (56.8)34 (39.5)13 (36.1)31 (45.6)25 (55.6)46 (52.9)
 Ever19 (43.2)52 (60.5)23 (63.9)37 (54.4)20 (44.4)41 (47.1)
Table II. Matched Odds Ratios and Confidence Intervals Quantifying Risk of Cervical Neoplasia for Different Risk Factors, Honduras, 1993–19951
 CIN ICIN IICIN III
OR95% CIOR95% CIOR95% CI
  • 1

    χ2 and p-values are of tests for linear trend unless degrees of freedom have been given in parentheses. CIN, cervical intraepithelial neoplasia; HPV, human papillomavirus.

Education
 None1.001.001.00
 Primary0.670.21–2.070.430.12–1.581.100.38–3.16
 Secondary+0.600.17–2.080.200.05–0.901.090.28–4.33
χ2 = 0.51/p = 0.477χ2 = 5.03/p = 0.025χ2 = 0.01/p = 0.91
Parity
 1–41.001.001.00
 5, 61.470.59–3.650.900.32–2.511.930.66–5.68
 7–10(7+) 1.100.25–4.850.890.24–3.364.301.17–15.7
 11+6.361.36–29.8
χ2 = 0.20/p = 0.652χ2 = 0.05/p = 0.832χ2 = 7.38/p = 0.007
No. of sex partners
 11.001.001.00
 2–3(2+) 1.250.59–2.651.980.79–4.981.440.62–3.33
 4+3.810.80–18.11.880.60–5.89
χ2 = 0.33/p = 0.567χ2 = 3.98/p = 0.046χ2 = 1.34/p = 0.25
Age at 1st intercourse (yr)
 12–151.001.001.00
 16–171.100.40–3.040.840.28–2.570.750.29–1.95
 18–190.630.22–1.820.600.17–2.060.540.19–1.54
 20+0.190.05–0.810.500.16–1.600.480.15–1.53
χ2 = 6.12/p = 0.013χ2 = 1.76/p = 0.185χ2 = 2.14/p = 0.144
Age at 1st pregnancy (yr)
 12–171.001.001.00
 18–191.240.50–3.100.560.18–1.781.030.44–2.40
 20+0.370.13–1.070.510.21–1.220.630.27–1.51
χ2 = 5.63(2)/p = 0.060χ2 = 2.50/p = 0.114χ2 = 0.93/p = 0.335
Previous screening
 None1.00(0–1) 1.001.00
 10.810.14–4.772.080.51–8.47
 20.630.11–3.680.900.34–2.410.770.22–2.62
 3+1.150.18–7.481.560.50–4.871.210.30–4.95
χ2 = 1.35(3)/p = 0.717χ2 = 0.41/p = 0.523χ2 = 3.64(3)/p = 0.303
HPV
 Negative1.001.001.00
 Positive1.480.73–2.982.491.04–5.9638.285.16–283.8
χ2 = 1.17/p = 0.279χ2 = 4.47/p = 0.035χ2 = 38.33/p < 0.0001
Oral contraceptives
 Never1.001.001.00
 Ever0.500.24–1.061.550.64–3.780.900.44–1.85
χ2 = 3.41/p = 0.065χ2 = 0.94/p = 0.330χ2 = 0.09/p = 0.769

The OR for lifetime number of sexual partners was greater than unity in all CIN classes, but the association was significant only for CIN II. Women who initiated sexual activity after age 19 had a risk of cervical dysplasia that was reduced to 50% (CIN II and III) or even 20% (CIN I) compared with women who had become sexually active before age 16 years. In this univariate analysis, the association was statistically significant only in the CIN I group. Similar associations were observed for age at first pregnancy, but none were statistically significant.

An increasing risk of cancer in situ with parity was observed (p < 0.007); women with more than 10 pregnancies had a 6-fold increased risk compared with women with less than 5 pregnancies.

History of cytologic screening was not associated with risk in any of the CIN classes, nor was ever use of oral contraceptives.

A trend of decreasing risk with increasing education was observed for CIN I and CIN II, which was significant for CIN II (p < 0.025).

At first inspection, associations of long-term utilization of wood as fuel in the kitchen with cervical neoplasia appeared to be negative and nonsignificant (Table III), except for an association with CIN III, which showed a significant departure from linear trend. Fifty-six women denied any utilization of wood in the kitchen, of whom 23 (6 cases and 17 controls) had also lived in households during childhood where wood was not used in the kitchen. By contrast, 21 cases and 12 controls had been exposed during childhood but reported zero exposure as adults. Of the 5 women who only used wood after childhood, 4 were in the lowest exposure category.

Table III. Matched Odds Ratios and Confidence Intervals Indicating Risk of Cervical Neoplasia for Different Levels of Exposure to Wood Burning in Kitchen1
Exposure to wood smoke (yr)CIN ICIN IICIN III
Cases (no.)Controls (no.)OR95% CICases (no.)Controls (no.)OR95% CICases (no.)Controls (no.)OR95% CI
  • 1

    χ2 and p-values are of tests for linear trend unless degrees of freedom have been given in parentheses. CIN, cervical intraepithelial neoplasia.

09101.009111.00981.00
1–1420400.610.20–1.9011340.280.07–1.1511340.360.11–1.18
15–2410230.450.13–1.5510140.900.19–4.258230.350.09–1.39
25–3451225+:0.320.06–1.584625+:0.660.08–5.638161.340.20–9.18
35+0123964.890.51–47.1
χ2 = 2.63/p = 0.105χ2 = 5.28(3)/p = 0.152χ2 (3) = 11.41/p = 0.022
1–1420401.0011341.0011341.00
15–2410232.110.37–12.110142.060.49–8.618230.970.20–4.67
25–3451225+:1.600.21–12.14625+:1.980.35–11.18162.260.34–14.8
35+0123969.471.16–77.4
χ2 = 0.05/p = 0.82χ2 = 0.78/p = 0.376χ2 = 5.73/p = 0.017

When the analysis was restricted to women who reported exposure to woodsmoke after childhood, the association was positive for all CIN classes and was statistically significant for CIN III, for which the data consistently showed a linear trend of increasing risk with increasing years of exposure (Table III).

Multivariate analysis

The above associations were assessed in the full dataset comprising cases of the 3 CIN classes and all controls by means of conditional logistic regression so as to make full use of all the available information and taking the matching into account. The basic model only included presence/absence of HPV and 2 Interaction variables to allow for a separate estimate of the effect of HPV for the different CIN classes. This interaction was statistically significant (χ2 = 17.9, 2 df, p = 0.0001). As shown in Table IV, the analysis showed a statistically significant linear trend of declining risk with increasing age at first intercourse (p = 0.0021). Tests for interaction between age at first intercourse and CIN class or HPV status were not significant. Inclusion of education, parity and number of sex partners in this model yielded similar results. Effect estimates, presented in Table V, showed that the risk of developing any stage of cervical dysplasia was reduced to approximately one-third for women who first had sex after age 19 compared with women who started sexual activity before age 16.

Table IV. Conditional Logistic Regression of Risk of CIN Among 366 Cases and Controls in Tegucigalpa, Honduras, 1993–19951
Variables in modelAdjusted for HPV2Adjusted for HPV and other variables3
−2 × MLDFX2(df)/p-value−2 × MLDFX2(df)/p-value
  • A1S, age at first intercourse; wood, exposure to woodsmoke (Wood1 indicates women reporting zero exposure to woodsmoke; Class, CIN class I, II or III; DF, degrees of freedom; ML, maximum (log) likelihood; CIN, cervical intraepithelial neoplasia; HPV, human papillomavirus.

  • 1

    Only p-values below 0.1 are shown. Each X2 test compares models with and without the last variable listed.

  • 2

    Basic model is HPV + HPV × Class.

  • 3

    Basic model includes HPV, HPV × Class, education, parity, number of sex partners, age at first intercourse.

  • 4

    This model is identical to the basic model.

Basic model223.393210.187
+A1S213.9349.46 (1)/p = 0.0021210.18477.11  (1)/p = 0.008
+A1S + A1S × HPV213.5050.44 (1)209.8280.36  (1)
+A1S + A1S × Class213.2460.70 (2)209.6290.56  (2)
+Wood223.0940.30 (1)209.3980.79  (1)
+Wood + Wood × Class219.9463.15 (2)206.29103.10  (2)
+Wood + Wood × HPV218.5654.53 (1)/p = 0.033204.7494.65  (1)/p = 0.031
+Wood1217.4145.98 (1)/p = 0.014203.8686.32  (1)/p = 0.012
+Wood1 + Wood1 × Class217.3160.95 (2)203.70100.16  (2)
+Wood1 + Wood1 × HPV217.3550.05 (1)203.8590.005 (1)
+Wood1 + Wood215.6151.80 (1)203.0490.82  (1)
+Wood1 + Wood + Wood × Class214.1971.42 (2)201.51111.53  (2)
+Wood1 + Wood + Wood × HPV212.4763.14 (1)/p = 0.076199.71103.14  (1)/p = 0.068
+Wood1 + Wood × HPV212.4854.93 (1)/p = 0.026199.8494.02  (1)/p = 0.045
Table V. Conditional Logistic Regression Models Fitted to the Data of All 366 Women in the Study, Tegucigalpa, 1993–19951
 HPV + HPV × CIN+Age 1st sex+Age 1st sex + SES+Wood1 + Wood + Wood × HPV+Wood1 + Wood × HPV+Age 1st sex +Wood1 + Wood + Wood × HPV + SES+Age 1st sex + Wood1 + Wood × HPV + SES
OR95% CIOR95% CIOR95% CIOR95% CIOR95% CIOR95% CIOR95% CI
  • 1

    SES, means education, parity, number of lifetime sex partners; HPV, human papillomavirus; CIN, cervical intraepithelial neoplasia.

  • 2

    Because of sparse data, categories 25–34 and 35+ years have been merged together in this model.

HPV × CIN I1.480.73–2.981.420.67–2.981.430.67–3.031.280.57–2.961.250.35–5.211.280.54–3.081.280.54–3.02
HPV × CIN II2.491.04–5.962.541.03–6.292.511.01–6.232.320.86–6.262.130.83–5.442.220.79–6.242.140.81–5.63
HPV × CIN III38.285.16–28342.615.70–31841.963.92–22125.693.20–20623.172.98–18025.072.99–21022.412.79–179
Age at first intercourse (yr)
 12–15reference catg.reference catg.reference catg.reference catg.
 16–171.060.54–2.071.100.56–2.151.130.56–2.281.080.53–2.18
 18–190.640.32–1.280.690.34–1.420.780.37–1.660.750.35–1.59
 20+0.340.16–0.750.380.17–0.850.390.17–0.910.370.16–0.87
Exposure to woodsmoke (yr)
 HPV negatives
  02.711.25–5.912.761.22–6.23
  1–14reference catg.reference catg.
  15–241.510.51–4.531.280.42–3.93
  25–340.440.08–2.4620.350.06–2.002
  35+
 HPV positives
  02.711.25–5.912.661.29–5.492.761.22–6.232.781.30–5.93
  1–14reference catg.reference catg.reference catg.reference catg.
  15–241.720.62–4.731.760.67–4.631.430.35–5.741.590.56–4.52
  25–343.721.54–9.0023.470.85–14.13.321.23–8.2023.230.74–14.0
  35+5.311.00–28.25.691.00–32.7

Adding utilization of wood to the basic model as a linear trend did not significantly improve the fit of the model (Table IV). However, the interaction of wood utilization and HPV was statistically significant, suggesting that the effect of woodsmoke is different for those infected with HPV compared with uninfected women.

Because of the problems with risk estimation identified in Table III, we then singled out women who reported zero exposure to woodsmoke by including an indicator variable (wood1) taking the value 1 for women reporting zero exposure and 0 otherwise. The analysis showed that women reporting zero exposure had a significantly higher risk of CIN than women reporting exposure. This effect was not different for women with or without HPV infection or in different CIN classes. However, again adding exposure to woodsmoke and the interaction of exposure with HPV to this model showed that the interaction of exposure to woodsmoke and HPV was borderline significant, again pointing to a different effect of woodsmoke in women with or without HPV infection. This is also suggested by the risk estimates in Table V, which show no association of exposure and risk in HPV-negative women but a trend of increasing risk with increasing exposure in HPV-positives. If we assume, as the results suggest, that the effect of woodsmoke is only present in HPV-infected women and if we, therefore, only include the interaction variable for HPV and woodsmoke, leaving out the main effect, the effect of woodsmoke is again statistically significant (χ2 = 4.93, 1df, p = 0.026; Table IV). Women who utilized wood in the kitchen for 35 years or more had a more than 5-fold increased risk of CIN compared with women who had less than 14 years of exposure (Table V).

The above reasoning is not changed when other variables indicative of socioeconomic status such as education, parity and number of sex partners are included, even though the actual values are slightly different (Table V). None of these 3 variables contributed significantly to the fit of the model after age at first intercourse had been included. There was no significant interaction of education, parity and number of sex partners with either HPV or CIN class.

DISCUSSION

Besides HPV, the 2 variables in the present study significantly associated with cervical dysplasia of any stage were age at first sexual intercourse and the use of wood as fuel in the kitchen. Ours was only a small study, with fewer than 50 cases in each CIN class. Although the number of controls was higher (n = 241), OR estimates lacked stability and had wide CIs. By testing in the full dataset the effects of variables for which some association was observed in the individual CIN classes, statistical power was increased.

Age at first sexual intercourse has been known as a risk factor for cervical neoplasia for many years,3, 17, 18 although its effect could not be demonstrated in all studies, e.g., because of confounding with age at first birth5 or number of sexual partners.2, 19, 20 We note that in the present study only 8% of control women had more than 3 lifetime sexual partners (Table I), which is low compared with other published studies. In addition, 27% of control women in the present study started sexual activity after age 19, giving a reasonably wide range of variation over this important risk factor, which could be demonstrated among both HPV-positive and HPV-negative women. Consistent with this finding, age at first intercourse was a risk factor for CIN III in the well-known Columbia-Spain study,21 both in the overall study group and among HPV-positive women.

The present results showed that women who had not been exposed to woodsmoke after childhood had a higher risk than those who were exposed in 8 of the 10 non-zero exposure categories (Table III). By contrast, data within those 8 exposure categories showed consistently elevated ORs and even a statistically significant linear trend in the CIN III group when high exposure was compared with low rather than zero exposure. These observations seemed highly inconsistent and we therefore hypothesized that answers to the initial screening question (“Have you used wood for cooking?”) were somehow biased. Since no further questions were asked if the answer was “No,” it was impossible to verify whether misclassification had occurred. Women who said they did not use wood both during childhood and adulthood were more often controls than cases, but their number was too small to act as a reference category. We note for the record that this problem did not occur when we studied this factor among cases with invasive cervical cancer,10 since only 3 cases and 10 controls in that study reported zero exposure to woodsmoke.

In that earlier case-control study of women with invasive cervical cancer,10 the association of exposure to woodsmoke with risk was much stronger among women with HPV infection, suggesting that woodsmoke acted as a promotor in those who had been infected with HPV. In the present study, not the utilization of wood as such, but the interaction of wood utilization and HPV was significantly associated with risk. It was therefore reasonable to assume an effect of woodsmoke only in HPV-positive women. The fact that the models with and without the main effect for wood had a comparable goodness of fit (212.47 vs. 212.48) in the present study (Table IV) served to confirm that assumption. In addition, the ORs for utilization of wood in HPV-negative women showed no association with risk, whereas the corresponding ORs in HPV-positive women showed a dose-response relationship.

Inhalation of cigarette smoke has been linked to the secretion of tobacco-specific substances in the cervical mucus,22–24 such as cotinine and nicotine. The presence of the tobacco-specific nitrosamine NNK, a known carcinogenic substance,25 has been demonstrated in the cervical mucus of smokers.26 Thus by analogy, exposure to woodsmoke could result in the secretion of inhaled substances or their metabolites in the cervical mucus where they could initiate or (more likely) promote a carcinogenic process. During the combustion of wood, carbon monoxide, particulates, sulfur and nitrogen oxides and mixtures of poly-cyclic aromatic hydrocarbons (PAHs) are released. Most PAHs have been classified by the World Health Organization as possibly carcinogenic to humans; 3 have been classified as probably carcinogenic to humans.27

Even though the association reported here is plausible, we cannot exclude the possibility that other exposures, closely related to the use of wood in the kitchen, are the real culprits. Candidates might be ashes, charred wood, the soot covering the walls of many kitchens in which wood is used and dietary factors. However, the association of exposure to woodsmoke and the incidence of cervical neoplasia cannot be explained by an underlying association with socioeconomic status. All women in our study were of low socioeconomic status, and the association did not change when we adjusted for education or age at first sexual intercourse, both of which are related to socioeconomic status. Similarly, differences in screening behavior cannot explain the observed association since women in our study were free of symptoms and the frequency of previous screening tests was similar among cases and controls.

The present study suggests that the use of wood for cooking is a risk factor for cervical neoplasia in HPV-infected women; given its high prevalence in developing countries, this risk factor deserves further study.

Acknowledgements

This study was carried out under grant CI1*CT92-0003 from the commission of the European community. J.P.V. received partial support from the Vereniging Trust Fonds Erasmus Universiteit Rotterdam. The authors thank Mrs. L. Diaz and Mrs. D. Maradiaga for collecting the data and Drs. Moreno and Castellsague for helpful comments during various stages of the work.

Ancillary