Genetic and environmental determinants of risk for cholangiocarcinoma via Opisthorchis viverrini in a densely infested area in Nakhon Phanom, northeast Thailand
Article first published online: 14 JUN 2005
Copyright © 2005 Wiley-Liss, Inc.
International Journal of Cancer
Volume 117, Issue 5, pages 854–860, 10 December 2005
How to Cite
Honjo, S., Srivatanakul, P., Sriplung, H., Kikukawa, H., Hanai, S., Uchida, K., Todoroki, T., Jedpiyawongse, A., Kittiwatanachot, P., Sripa, B., Deerasamee, S. and Miwa, M. (2005), Genetic and environmental determinants of risk for cholangiocarcinoma via Opisthorchis viverrini in a densely infested area in Nakhon Phanom, northeast Thailand. Int. J. Cancer, 117: 854–860. doi: 10.1002/ijc.21146
- Issue published online: 10 OCT 2005
- Article first published online: 14 JUN 2005
- Manuscript Accepted: 23 FEB 2005
- Manuscript Received: 7 SEP 2004
- Cancer Research Foundation of the National Cancer Institute
- Ministry of Public Health of Thailand
- Ministry of Education, Culture, Sports, Science and Technology of Japan
- population-based case-control study;
- alcohol drinking;
Infection with Opisthorchis viverrini (OV) is associated with cholangiocarcinoma. OV is common in northeast Thailand, but less than 10% of the inhabitants develop cholangiocarcinoma. Animal experiments suggest that OV infection alone does not cause cholangiocarcinoma, and thus other environmental and genetic factors may play a role in causation. We conducted a population-based case-control study in which sex, age and place of residence were matched individually. Polymorphisms of GSTM1 and GSTT1 alone were not associated with risk for cholangiocarcinoma, while an elevated level of antibodies against OV (ELISA) ≥0.200 was the strongest risk indicator (odds ratio as compared to that <0.200 = 27.09 [95% confidence interval (CI): 6.30–116.57]. Compared to subjects who had a normal antibody range and the wild-type GSTM1 gene, those who had elevated antibodies had higher odds ratios of 12.32 (95% CI: 1.60–94.85) for wild-type GSTM1 and 23.53 (95% CI: 4.25–130.31) for the null variant thereof, respectively. Past and current regular drinkers of alcohol had higher risk [odds ratio = 5.39 (95% CI: 1.11–26.06) and 4.82 (95% CI: 1.29–18.06), respectively]. Eating fermented products was an independent risk factor. Smokers or consumers of fermented fish had substantially increased risk if they were past or current drinkers. Infection with OV correlates strongly with cholangiocarcinoma, susceptibility to which may be possibly associated with GSTM1 polymorphism. Alcohol may affect metabolic pathways of endogenous and exogenous nitrosamines. © 2005 Wiley-Liss, Inc.
Cholangiocarcinoma is uncommon, and the aetiology has been elucidated only for a limited portion of cases.1 Northeast Thailand has an exceptionally high incidence of liver cancer of up to 87.5 (male) and 37.2 (female) per 100,000 (world population-adjusted incidence rate). Most of these cases have been attributed to cholangiocarcinoma.2 Infestation of the liver fluke, Opisthorchis viverrini (OV), has been causally related to cholangiocarcinoma in this region.3 The prevalence of OV infection in Nakhon Phanom province was between ∼30 and ∼70% with substantial variation between villages;4 the technique for estimation of prevalence was not explicitly mentioned but supposed OV egg counting in the faeces. However, it is unknown why only a limited proportion of the inhabitants (<10%) develop cholangiocarcinoma.
In our study, we hypothesised that genetic background may be a contributing factor in the occurrence of the cholangiocarcinoma. Hence, we examined the association between polymorphisms of genes for glutathione-S-transferase enzymes, GSTM1 and GSTT1, and the risk for cholangiocarcinoma. Because both enzymes function in carcinogen detoxification systems,5 it seems reasonable to speculate that null variants of these genes may be associated with increased risk of cancer.6
The association between OV infection and cancer risk has been a common assumption, but adequate analytic epidemiological studies are sparse. Parkin and colleagues7 reported in their case-control study that past or present infection with OV, as indicated by serum antibodies against OV, is associated with a 5-fold increase in cancer risk. Although the anti-OV antigen titre was elevated and reached a plateau at an approximately 2 months after OV inoculation in hamsters,8 the transition of the antibody level over the lifespan remains uncertain in both humans and animals. We thus examined the risk of foods possibly containing OV metacercariae as well as that of OV infection as indicated by serum antibody response.
In animal experiments, OV infection causes cholangiocarcinoma with coadministration of nitrosamine9 in a dose-dependent manner.10 Diet and cigarette smoking are major sources of nitrosamine exposure,11 and several foods and foodstuffs consumed in the study area reportedly contain nitrosamines.12 Ethanol may inhibit first-pass clearance in the liver and allow carcinogens to pass to the peripheral blood.13 Ethanol may also induce enzymes responsible for carcinogen activation.14 We therefore examined the possible risk of smoking, alcohol consumption and diet for cholangiocarcinoma. Subsequently we investigated possible interactions between genetic polymorphisms of the genes encoding GSTM1 and GSTT1 and smoking as well as other lifestyle-related factors on the risk for cholangiocarcinoma. We examined each of the lifestyle-related factors individually for possible interaction with the genetic polymorphisms.
Material and methods
Our study was performed after receiving approval from the ethics committees of the University of Tsukuba, Ibaraki, Japan and the National Cancer Institute, Bangkok, Thailand.
All cases with cholangiocarcinoma were identified between September 1999 and November 2001 upon visitation to the Nakhon Phanom provincial hospital or one of the community hospitals in the province. All cases from the community hospitals had been referred to the Nakhon Phanom Provincial hospital for confirmation of the diagnosis. Diagnosis was based on abdominal ultrasonography by a single radiologist (P.K.) at the provincial hospital with serological supportive evidence including a raised CA 19-9 (≥40 μ/ml) and a normal level of alpha fetoprotein (AFP; <20 ng/ml), although the latter was not obligatory. Histological diagnosis was made for only 9 cases as the diagnosis without liver biopsy was usual in our study area. Because all clinically diagnosed cases both with and without ultrasound confirmation and serum tumour marker measurement were to be reported to the provincial cancer registry, the number of cases in our study should be smaller than that from the cancer registry for the corresponding time period. Local health personnel selected 1 candidate as a control, being matched by sex, age (within 5 years) and place of residence (amphur, district). Because the health personnel were familiar with the inhabitants in the local communities prior to our study, they easily found candidates for the matched controls. Although the health personnel did not intentionally choose such candidates among inhabitants who apparently were cooperative with the study, systematic random sampling was not employed and apparently ill persons were not contacted. Each control candidate was invited to the provincial hospital and underwent an abdominal ultrasonographic examination by the same radiologist used for the cases. Control candidates that had no ultrasonographic findings suggestive of cholangiocarcinoma or hepatocellular carcinoma were verified as controls. Only in 1 case did a candidate control have ultrasonographic findings suggestive for cholangiocarcinoma with subsequent serological findings of raised CA19-9 level and normal AFP. This candidate was diagnosed with cholangiocarcinoma and subsequently became a case. Another control was identified for this case, whereas a second control candidate was identified for the first proband case.
We found no material changes in the results if we excluded 3 cholangiocarcinoma cases with both abnormally increased serum AFP and CA19-9 levels via statistical analyses. Therefore, we treated these 3 subjects as cholangiocarcinoma in the subsequent analyses. All the cases and controls were inhabitants of the Nakhon Phanom province with a population of 340,539 males and 343,905 females as of July 1, 2000. None of the subjects we contacted declined to participate in our study, and a total of 85 male and 44 female case-control pairs were obtained. Although the study area is an economically deprived rural region, inhabitants were generally educated and literate (as shown in Table I).
|Variable||Category||Cases1||Controls1||Adjusted2 OR||95% CI3||p value|
|Anti-OV Ab (ELISA)||<0.200||61||119||1.00||Reference|
|Toilet||Inside the house||59||55||1.00||Reference|
|Outside or none||69||73||0.70||0.31||1.59||0.40|
|Source of water||Pipe water||66||79||1.00||Reference|
|Treatment with praziquantel||No||69||98||1.00||Reference|
|Betel nut chewing||Never||94||96||1.00||Reference|
|Eating frequency in thepast 10 years|
|Salty fish or meat||0||48||45||1.00||Reference|
|Fermented fish or pork||0||28||41||1.00||Reference|
|Chinese or north-east sausage||0||48||58||1.00||Reference|
|Fermented vegetables or fruits||<4/week||59||50||1.00||Reference|
|Cooked fresh vegetables||<3/week||39||51||1.00||Reference|
Determination of serum markers for liver fluke and hepatitis virus infection
OV-specific IgG was determined by indirect ELISA developed by Sripa and Kaeskes,8 and we used a cut-off value at 0.200 optical density for OV-positive subjects. Hepatitis B antigen was also measured for 69% of the cases and 61% of the controls; and 5.6% and 1.3% of those examined gave positive results, respectively.
Genetic polymorphism analysis of GSTM1 and GSTT1
The polymorphism of the GSTM1 and GSTT1 genes was determined according to the procedure of Abdel-Rahman and colleagues using multiplex PCR15 with slight modification. Briefly, the reaction mixture of 50 μl contained ∼150 ng of DNA, 30 pmol each of the mixture of the GSTM1 primers 5′-GAA CTC CCT GAA AAG CTA AAG C-3′ and 5′-GTT GGG CTC AAA TAT ACG GTG G-3′ or the GSTT1 primers 5′-TTC CTT ACT GGT CCT CAC ATC TC-3′ and 5′-TCA CCG GAT CAT GGC CAG CA-3′. A fragment of the CYP1A1 gene was amplified as an internal control in the same reaction tube using the primers 5′-GAA CTG CCA CTT CAG CTG TCT-3′ and 5′-CAG CTG CAT TTG GAA GTG CTC-3′. The reaction was conducted in the presence of 200 μM dNTPs, 50 mM KCl, 10 mM Tris-HCl, pH 8.5, 1.5 mM MgCl2 and 2 U of Amplitaq Gold polymerase (Cetus PerkinElmer, Norwalk, CT). The conditions of PCR were 95°C for 10 min, and 35–45 cycles of 94° for 1 min, 59°C for 1 min and 72°C for 2 min, followed by 72°C for 7 min. The products were separated electrophoretically in 2% agarose gel as described.15 Specimens were not available for 2 cases and 6 controls.
Data collection on lifestyle-related factors and preparation for statistical analysis
Personal interviews were conducted for both cases and controls at the hospital by local health personnel using a structured questionnaire. Inquiry was made regarding educational attainment, type of toilet used, source of water and personal history of treatment with the anthelmintic, praziquantel, smoking, alcohol consumption and dietary habits. Smoking habits were categorised into 4 categories including never, occasional, ex-regular and regular smoker. The regular smoker was defined as smoking at least 1 cigarette per day, whereas the ex-regular smoker was defined as having stopped smoking regularly at least 1 year prior to our study. For ex-regular and regular smokers, the type and number of cigarettes smoked per day and age at beginning and/or cessation of smoking were noted. Cigarette-year was calculated for ex-regular and currently regular smokers as the number of cigarettes smoked daily multiplied by the number of years of smoking, regardless of cigarette type. Fifteen cases and 5 controls did not give information on smoking status and/or amount and duration of smoking. Alcohol consumption status was divided into 4 categories including never, occasional, ex-regular and regular drinker. The regular drinker was defined as drinking more than once a week, while the ex-regular drinker was defined as having stopped regular drinking at least 1 year prior to the study. For ex-regular and currently regular drinkers, the number of glasses consumed weekly for each of beer, whiskey and local whiskey were also noted. Lifetime ethanol intake was calculated for ex-regular and currently regular drinkers using the number of the glasses, ethanol concentration and duration for each beverage type. Thirteen cases and 5 controls did not give information on alcohol consumption status and/or the amount and duration of alcohol consumption.
Dietary habits at the time of the study and during the previous 10 years were obtained for 12 selected food items with respect to eating frequency. Eating frequency for each of the selected food items was categorised into 3 classes, if possible, with the intention of keeping nearly equal numbers across the categories among the control group. We ignored the present diet in analysis of risk for cholangiocarcinoma, considering the past diet to be more aetiologically related to cancer occurrence.
Risk for cholangiocarcinoma was evaluated as an odds ratio due to the genetic polymorphism of GSTM1 and GSTT1, employing subjects with wild-type enzymes as the reference group and maintaining matched pairs. Risk of the demographic and other lifestyle-related factors was estimated similarly. The odds ratio was calculated using a conditional logistic regression model, keeping matched case-control pairs.16 An adjusted odds ratio was calculated with an allowance for anti-OV antibody (≥0.200 vs. <0.200), cigarette smoking and alcohol consumption. However, the elevated serum antibody was not adjusted for in the analysis of the risk due to eating raw fish and sticky rice because the former may contain metacercariae of OV and the latter is eaten oftentimes with undercooked fish, koi-pla (prepared and eaten the same day3). Actual current eating habits of these 2 items correlated with an increased risk for elevated antibody among cases (data not shown).
Possible interactions between each of the 2 genetic polymorphisms of GSTM1 and GSTT1 and each lifestyle-related factor was first examined in a case-control design using cases only, treating cases with the null variant as a case and ones with the wild type as control,17 with adjustment for sex and age (<54, 54–66, and >66 years). When the odds ratio varied from unity (p < 0.10) in the case-only analysis, then the odds ratio due to combined categories of the genetic polymorphism and the lifestyle factor was calculated with a conditional logistic regression model using cases and controls.16
Because we found an increased risk for cholangiocarcinoma due to alcohol drinking as shown below and allowed for the possibility that ethanol may affect carcinogen metabolism,13, 14 we performed likelihood ratio tests16 to examine the effect that alcohol consumption might have on the association of anti-OV antibody, smoking and dietary habit with risk for cholangiocarcinoma. In those tests, alcohol consumption was categorised into 2 classes, nondrinkers and everdrinkers. The latter comprised occasional, ex-drinkers and current regular drinkers. The tests compared a main effect, no interaction terms, with a full model containing all possible interaction terms for the variables concerned. When the p value for the log-likelihood ratio statistic was <0.1, we calculated the odds ratio due to the combination of alcohol consumption and each of variables concerned.
All the statistical analyses were performed with a statistical package, STATA,18 and statistical significance was defined as p < 0.05 unless indicated otherwise.
Genetic polymorphism of GSTM1 or GSTT1 alone did not correlate with risk for cholangiocarcinoma. Odds ratios (ORs) for null variants of GSTM1 and GSTT1 genes were 0.94 [95% confidence interval (CI) = 0.57–1.54] and 0.72 (0.44–1.20), respectively. Using subjects with wild-type alleles for both reference groups, the OR of having null variants for both enzymes was not significantly different from unity (OR = 0.68, 95% CI = 0.34–1.38).
The relationship of serum anti-OV antibody, smoking, alcohol use, diet or other environmental factors to risk for cholangiocarcinoma are shown in Table I. Elevated serum anti-OV antibody correlated with a nearly 30-fold increase in risk. Adjustment for education or literacy did not materially change the risk (data not shown). Ex-drinkers and currently regular drinking were also at increased risk. Using nondrinkers as a reference group, the odds ratio of the ever drinkers was significantly increased (OR = 3.16, 95% CI = 1.02–9.83). The relationship between regular smoking and the risk for cholangiocarcinoma was not significant (Table I). When we examined the risk from alcohol consumption and smoking in terms of cumulative exposure (ethanol intake and cigarette-years), a dose-response relationship was observed both for drinking and for smoking (data not shown). A source of water other than pipe water correlated with increased risk (p = 0.07). Reported treatment by the anthelmintic, praziquantel, correlated positively with risk in a univariate analysis (data not shown) but not when adjusted for anti-OV antibody, smoking and alcohol consumption. An adjustment for the reported use of praziquantel did not materially change the increased OR due to OV infection (data not shown). Among dietary habits during the past 10 years, raw fish intake and fermented fish or pork intake were significantly associated with an increased risk (p for trend <0.05).
The possible effect modification due to a genetic polymorphism of GSTM1 and GSTT1 was examined among cases only using a case-control design. Cases with elevated anti-OV antibody were twice as likely to have the null variant of GSTM1, although this difference was not statistically significant (OR = 2.10, p = 0.08). Cases reporting a history of treatment with praziquantel were more likely to have a null variant of GSTT1 (OR = 2.68, p = 0.04). Although not significant, current regular smokers were less likely to have the null variant of GSTM1 (OR = 0.40, p = 0.18) or GSTT1 (OR = 0.32, p = 0.08), respectively, as compared to cases who never smoked. Ex-regular drinkers were 7-fold more likely to have the null variant of GSTM1 (OR = 7.44, p < 0.05). Alcohol drinkers, especially current regular ones, were less likely to have the null variant of GSTT1 (OR = 0.22, p = 0.04). Table II presents ORs due to the combination of GSTM1 polymorphism and each of the selected variables showing statistical significance or approaching significance for interaction (p < 0.10). Increased risk for cholangiocarcinoma due to elevated antibody tended to be more clearly observed when GSTM1 was null. On the other hand, the null genotype tended to be protective in non-OV infected subjects. Ex-regular drinking tended to be associated with an increased risk if GSTM1 was null. Similarly, ORs due to a combination of GSTT1 polymorphism and each of selected factors with a p < 0.10 for interaction are shown in Table III. Increased risk of a history of treatment with an anthelmintic was only observed if GSTT1 was null, although this was not significant (p = 0.31). There was an insignificant increased risk due to smoking, predominantly if the GSTT1 polymorphism was wild type. There was an elevated risk for current regular drinkers compared to nondrinkers only when subjects had wild-type GSTT1, whereas ex-regular drinking correlated with increased risk only among subjects carrying the GSTT1 null variant.
|Adjusted1 OR||95% CI||p value||Adjusted1 OR||95% CI2||p value|
|Toilet||Inside the house||1.00||Reference||0.22||0.06||0.88||0.03|
|Outside or none||0.20||0.04||1.02||0.05||0.25||0.07||0.91||0.04|
|Cooked fresh vegetables||<3/week||1.00||Reference||0.39||0.08||2.02||0.26|
|Adjusted1 OR||95% CI2||p value||Adjusted1 OR||95% CI2||p value|
|Treatment with praziquantel||No||1.00||Reference||0.52||0.17||1.57||0.24|
Finally, we examined the possibility that alcohol consumption affects risk. We considered anti-OV antibody, smoking and dietary habit during the past 10 years, and only the risks due to smoking and eating of fermented fish (pla-ra and/or pla-chao) were found to be altered with alcohol consumption (p < 0.01 and 0.07, respectively). Table IV presents the ORs due to the combined categories of alcohol consumption and those 2 variables.
|Never drinkers||Ever2 drinkers|
|Adjusted3 OR||95% CI||p value||Adjusted3 OR||95% CI||p value|
In our study, elevated anti-OV antibody as well as the consumption of raw fish correlated with increased risk for cholangiocarcinoma. A few points should be considered. First, all fishes do not have OV metacercariae in their flesh. Second, individual cases may possibly differ with respect to susceptibility to the infection. Third, it is unclear as to how precisely the duration and intensity of the OV infection was estimated in cases with elevated anti-OV antibody. Finally, the reported history of treatment with praziquantel should be considered because this drug is effective after only a single administration;19 however, statistical adjustment for use of this drug did not affect the increased risk due to elevated anti-OV antibody. It may be difficult to measure exactly the magnitude of the effect of OV infection and eating raw fish. However, if the prevalence of infection, as indicated by an elevated antibody level, is assumed to be 6% (8/127) among the study population, then ∼60% of cholangiocarcinomas may be preventable by eliminating the infection. Thus, avoiding the consumption of raw and undercooked fish must be a fundamental approach for preventing cholangiocarcinoma in the study area.
The magnitude of the increased risk due to the elevated antibody was more than 5-fold compared to that in a previous study.7 The previous study was conducted based on the 3 hospitals including 2 in northeast Thailand and the National Cancer Institute in Bangkok, and the controls were also recruited from patients at the hospitals. On the other hand, all the counterparts in our study were apparently healthy persons from local communities and further confirmed not to have abnormalities on ultrasonography. Thus, our estimated risks associated with OV infection and other risk factors may be possibly overestimated. Discordant findings regarding the risk related to sticky rice between the previous7 and the present studies may also be due to different backgrounds of the study populations. Because sticky rice is a staple food in the present study area, it might be difficult to differentiate the risk due to this food in the present study. Betel nut chewing was associated with the increased risk in the previous study,7 but it did not correlate with the risk in our study. We have no clear explanation for this discrepant finding.
Although infectious and inflammatory conditions have been linked to carcinogenesis,20, 21, 22, 23 OV infection alone generally did not cause cholangiocarcinoma in animal models.9, 10 Therefore, a few behavioural modifications other than the cessation of eating raw fish must be considered for the prevention of cholangiocarcinoma in the present study area. First, the inhabitants should refrain from alcohol consumption and smoking. Alcohol drinking elevated the risk of cancer and positively modified the risks due to smoking and due to the fermented fish consumption. It may be possible that ethanol augments the effect of cigarette carcinogens and fermented products, and mechanisms for such alcohol-related cancer risk have been proposed.13, 14 As for smoking, cigarettes per se contain carcinogenic compounds, and cigarette smoking has been linked to immunosuppression.24, 25 Second, the inhabitants should limit their intake of foods containing nitrosamines. Most fermented pork or fish dishes, including pla-ra and/or pla-chao, contain N-nitrosodimethylamine, N-nitrosopiperidine and N-nitrosopyrrolidine.12 Finally, although unexpected, the relationship between nonpiped water use (>92% of which was well water) and increased risk approached significance (p = 0.07). Nitrates and other chemicals in drinking water may impact the risk of cancer,26 and high concentrations of nitrogen compounds were detected in well water in an area of China having a high incidence of oesophageal cancer.27 Therefore, it may be worthwhile to measure the levels of nitrates and related chemicals in well water in the present study area.
Ascorbic acid contained in various kinds of vegetables and fruits was supposed to inhibit endogenous nitrosamine production among infected subjects with OV.28 In our study, however, the consumption of fresh vegetables and fruits did not correlate with decreased cancer risk. The study area is economically deprived, and the diet as a whole is substandard. Hence, it may be difficult to find possible effect from consumption of foods with protective anti-oxidant activity in the present epidemiological study area. It is also possible that vegetables may be a source of nitrate and nitrite for nitrosamine production in infected subjects with OV, in whom the potential for nitrosation may be increased.22
Genetic polymorphism of GSTM1 or GSTT1 alone did not correlate with the risk of cholangiocarcinoma. The increased risk from OV infection, as indicated by elevated serum antibodies against OV, was more pronounced among subjects with the null variant of GSTM1. It may be argued that GSTM1, a detoxifying enzyme, is related to cancer risk because endogenous nitrosamine production may be increased in OV-infected subjects, as mentioned above.22
Our study has several limitations. First, although the sample size was large compared to the previous study on cholangiocarcinoma in Thailand,7 a much larger number of subjects is required to detect even a 2-fold increase in risk. Ninety discordant pairs are required to detect such mildly increased risk with α of 0.05 and power of 90%, providing that the proportion of null variants among controls is 0.4 or 0.7.29 Second, controls were not chosen on the basis of random sampling, although we did not intentionally select controls from apparently cooperative persons only. If the controls in our study were more health conscious than the rest of inhabitants, then the risks from smoking, drinking and elevated anti-OV antibody level may have been attributed, at least in part, to difference in relation to general health status between the cases and controls. On the other hand, if cases were overmatched with regard to the district, then the cancer risk due to OV infection and some of the foods might have been underestimated. Third, because more than 10 cases did not provide information required to calculate the cumulative exposure to smoking and drinking, a precise estimation of risk due to these exposures was not possible. Such low response rates to those questions might have led to underestimated risk.
Cholangiocarcinoma is relatively unusual worldwide; however, it has a rather high estimated incidence of 15% among liver cancer.1 Although aetiological agents1 and molecular changes30 in cholangiocarcinoma tissues may differ at least in part between the northeast of Thailand and other areas of the world, the present findings further our understanding of the aetiology of this cancer and the relationship between inflammation and carcinogenesis.
We thank the cases and controls for their participation in our study. We thank the staffs of Nakhon Phanom Provincial Office, Nakhon Phanom Hospital and all of the community hospitals for their generous assistance in interviewing and collecting the specimens from all the subjects. We are grateful to Dr. D.M. Parkin, International Agency for Research on Cancer, Lyon, for helpful comments on the manuscript.
- 2Cancer in Thailand Vol. II, 1992–1994 IARC Technical report, vol. 34 Lyon: International Agency for Research on Cancer, 1999. 135., , , , , , , , .
- 3International agency for research on cancer. Infection with liver flukes (Opisthorchis viverrini, Opisthorchis felineus and Clonorchis sinensis). IARC Monogr Eval Carcinog Risks Hum 1994; 61: 121–75.
- 16Statistical methods in cancer research. Vol. I. The analysis of case-control studies, vol. 32. Lyon: International Agency for Research on Cancer, 1980., .
- 18Statacorp. Stata Statistical Software, ed. Release 7.0 College Station, TX: Stata Corporation, 2001.
- 29Case-control studies: design, conduct, analysis. Oxford: Oxford University Press, 1982..