- Email: [email protected]
Oral contraceptives and liver cancer
ELSEVIER
Oral Contraceptives
Results of the Multicentre study (MILTS) The Collaborative
and Liver Cancer
International
MILTS Project Teams
Many, but not all, previous epidemiological studies indicated a greater risk of hepatocellular cancer (HCC) in women who have used combined oral contraceptives for a long period of time, but no one has analyzed this risk based upon use of different formulations. It was decided to analyze specifically the risk of OC containing cyproterone acetate (CPA) after toxicological experiments in animals found hints for a potential genotoxicity. This report describes the risk associated with ever having used combined oral contraceptives (OC) among 317 cases of primary hepatocellular cancer (HCC) in women under age 65, compared with 1060 age-matched hospital and 719population controls in a case-control study, which was conducted in six European countries. The adjusted odds ratio (unconditional logistic regression) for ever having used any OC was found to be 0.75 (0.54 to 1.03) when all cases were compared w.th all controls, and compared to hospital and population controls separately: 1.13 (0.86 to 1.48) and 0.78 (0.59 to 1.03), respectively. The adjusted odds ratios for OC containing all progestins of the CPA group were 0.89 (0.49 to 1.61); and 0.89 (0.37 to 2.18) for OC containing only CPA. There was no increase in risk for HCC with increasing duration of OC use among the different groups of OCs in the total group of cases with pooled controls. The risk estimates were not related to time since first or last use of any of the types of OCs considered. The most important risk factors for HCC were confirmed as a prior history of hepatitis B and C (adjusted odds ratio
6 This article was oreuared bv L.A.J. Heinemann.?# T. DoM1nh.t I. Guaaenboos-Holzmann (ie&ased),i C. Thiel,t and E. &&be.*t# Scleniific Ad%ory Committee, A.R. Felnstein (chair), D. Thomas, C. Brechot, W.0. Spltzer, S. Watanabe, V Beral, 0. Meink (see the Appendix); Collaborating centers and key personnel, other advisory committees, and reference expert groups were published elsewhere and are listed in the Appendix *Potsdam Institute of Pharmacoepldemiology and Technology Assessment (PIPTA), Potsdam, Germany, #Department of Epidemiology and Biostatistics, McGill University, Montrbal, Quhbec, Canada, tCentre for Epidemiology and Health Research (ZEG), Zepernick, Germany; *institute of Medical Statistics and lnformatics, Free University Berlin, Berlin, Germany. Name and address for correspondence, Prof. Dr. Lothar A.J. Helnemann, Centre for Epidemiology and Health Research Berlin, Schoenerlinder Strasse I I, 16341 Zepernick, Germany, Tel. 4930 9451 ,OiZO; Fax, 4930 9451 0123; e-mail zegr&nx.de Submitted for publication July 18, 1997 Revised August 22, 1997 Accepted for publication August 25, 1997
@ 1997 Elsevier Science lnc 655 Avenue of the Americas,
Liver Tumor
All rights reserved. New York, NY 10010
3.1 (2.2; 4.3) and 37.9 (20.2: 70.9) for HBV and HCV, respectively). In the small subgroup of HCC cases without liver cirrhosis and with negative serology for HBV and HCV, there was evidence of an association with duration of OC use. No such trend was observed for the CPA group of OCs. Altogether, there is no evidence for an increased risk of HCC associated with CPA or CPA-like OCs. Oral contraceptives in the aggregate may enhance the risk of liver carcinomas not associated with HBV or HCV infection, but if so, this is an extremely rare adverse effect of their use, CONTRACEPTION ~997i56~275-284 0 1997 Elsevier Science Inc. All rights reserved. WORDS: hepatocellular cancer, oral contraceptive, cyproterone acetate, case-control study, MILTS Study
KEY
Introduction he Multicentre International Liver Tumour Study (MILTS)’ was initiated after experimental reports suggested that cyproterone acetate (CPA] is genotoxic in rat hepatocytesz,3 and carcinogenic,4 but there were many other reasons for a new study. Although nine case-control studie+l4 and some reviews15J6 had been published on steroid contraceptives and liver cancer, no pharmacoepidemiological studies had been carried out to distinguish between the various formulations [estrogens, progestins] of oral contraceptives (OC). Also, studies in hepatitis B-endemic area&l3 generally did not show a relationship between OC and risk of hepatocellular cancer (HCC), whereas studies in nonendemic areaP2 tended to show an increase in risk in long term users, suggesting the need for a new study. In addition to hepatitis B virus (HBV), other risk factors for HCC have been identified. These include: hepatitis C, liver cirrhosis, possibly diabetes mellitus, alcohol abuse, possibly smoking, and nutritional factors (for example, aflatoxin), employment in the rubber or polyvinyl chloride (PVC) industry, occupational contact with oil products, hydrocarbons, dyes, and other chemicals, and use of potentially hepatotoxic drugs (for example, stanozolol, clofibrate, or
T
ISSN 0010.7624/97/$17.00 Pll SOOlO-7824(97)00158-3
276
MILTS Project
Team
azathioprine). These have not consistently been considered as possible confounders in prior studies of liver cancer and OC. The current study was aimed mainly at elucidating possible associations between risk of HCC and use of various types of OC with special emphasis on formulations containing CPA or progestins of the same family. Secondary aims were to investigate possible associations with OC use in subgroups of women with and without other risk factors for HCC, including HBV and hepatitis C virus (HCV).
Methods The study methods have been published in detail e1sewhere.l In brief, this hospital based case-control study (with hospital and community controls) was undertaken in six European countries between July 1994 and June 1996, designed to get a mix of population with lower and higher prevalence of OC use, and variation in the incidence of HCC as well hepatitis B and C. Women under the age of 6.5 with and without HCC were recruited. The project involved 64 clinics in 31 mainly university related settings, and combined for analyses in 12 regions (seven in Germany and one each in the United Kingdom, France, Italy, Greece, and Spain). The study covered two time periods: “historical” accrual for the 4 years before implementation of the study (July 1990 to June 1994) and “concurrent” accrual from July 1994 to June 1996. A monitoring system was established in each collaborating hospital to identify all female cases under the age of 65. Cases were classified as having a probable or definite diagnosis of HCC. A probuble diagnosis of HCC was based on a valid imaging result (sonography, computerized tomography, magnetic resonance tomography), and a clearly elevated blood a-fetoprotein level [500+ units). A definite diugnosis required histological confirmation. An expert group [including a reference pathologist) evaluated and classified all candidate cases and decided on inclusion or exclusion of equivocal cases by consensus. On average, four controls were originally to be obtained for each HCC case: two general hospital controls without malignancy; one hospital control with an eligible tumor diagnosis (as noted in the previously published list’); and one population control. Frequency matching was done in the same 5 year age groups as the cases; living controls were obtained for cases in which patients were deceased. Cases were identified in specified liver centers in the participating countries by monitoring all admission records and regular visits to relevant hospital wards. Similarly, the hospital controls were selected
Contraception 1997;56:275-284
from among new admissions with prespecified diagnoses by regular visits to specific wards on a rotating basis. Population controls were randomly selected from a complete population register in the respective and visited in their regions, invited to participate homes for a face-to-face interview (details have been published elsewhere’). All cases and controls were interviewed in a standardized manner. The questionnaire was designed to obtain information on demographic variables, personal disease history, family history of cancer, reproductive history, including lifetime history of OC use, lifetime use of particular drugs, smoking and alcohol use, and occupational history and related exposure indicators. Three other forms were used to abstract relevant information from the patients’ medical records (various diagnostic information, data on potential exposures or confounders). An extensive system of quality control measures was applied (details have been published elsewherel). For deceased cases or those who could not be interviewed for other reasons, a next-of-kin was interviewed. Proxies were individuals who were sufficiently well acquainted with the case to be able to respond to the interview, and included the person’s mother, sister, husband, or friend. When a proxy of the case was interviewed, a concurrent control was interviewed; that is, neither historical controls nor proxies of controls were used in this study. This method, which avoids high costs, is justified by the high congruence reported in the literature between results of interviews of proxies and interviews with controls.“~‘* Differences in duration of observation and time of possible exposure were controlled for in the analyses. Oral contraceptive use was the primary exposure variable of interest. We also distinguished in the analyses exposure to the subgroup of OC containing CPA-like progestins, that is, the chemically similar substances: cyproterone acetate (CPA), medroxyprogesterone acetate (MPA), and chlormadinone acetate (CMA), as well as CPA alone. The history of a hepatitis B (HBV) or C (HCV) virus infection was defined as negative if all serological tests that were made (HBsAG, AntiHBs, AntiHB=, AntiHC) were negative, and as positive if any of the tests were positive. Persons with missing data were excluded from this analysis. The lab tests were performed with commercially available test kits in the individual collaborating liver centers. Details will be published elsewhere. Unconditional logistic regression models were used to estimate crude and adjusted odds ratios using STATA@ software.19 Conditional logistic regression was performed as a second choice to make cases and
Contraception 1997:56:275-284
controls as similar as possible regarding the frequency matching variables (age, center, and catchment area). Variables were included in regression models if their inclusion altered the odds ratio by 10% or more, or if previous findings in the literature suggested that they were likely to be confounders.
Results Response Rates Of 368 eligible cases of liver cancer that were identified, 317 (86.1?’ 0 ) were included in the study: 179, 40, 29, 15, 14, and 40 came from Germany, UK, France, Italy, Greece and Spain, respectively. A total of 1060 hospital controls and 719 population controls were included in the study. The response rates for hospitalized patients (cases and hospital controls) ranged between 68% and 100% among the participating centers; the lowest rates were for cases that either were not identified while hospitalized or who died with no next-of-kin willing to participate. The response rates for population controls were lower than for hospitalized women and ranged from about 80% (UK, Italy, Greece) to between 60% and 70% (Germany, France, Spain).
Comparability of Case Subgroups Some of the HCC cases (who were either severely ill or deceased) were not able to give a personal interview; therefore, the information was obtained from a relative, friend, or other source (136 out of 317 cases). Cases with proxy information had a slightly higher prevalence of OCs at any time compared with those with personal interview (55.9% vs 48.1%, 46.3% vs 39.2%, and 7.4% vs 8.8% for all OC types, non-CPA groups, and CPA groups, respectively). The type of OC was unknown in three cases without a personal interview (2.2%), but for none with a personal interview. Both subgroups were not significantly different regarding exposure and the types of OC analyzed. As prespecified in the protocol, we pooled the subgroups with and without personal interview in the main analysis and do not present them separately here.
Comparability of Control Groups Table 1 shows some differences between hospital and population controls in such features as history of gallbladder disease, history of chronic disease, previous major surgery, and year of birth that are related to OC use. Blood samples were only taken from cases and hospital controls.
Liver Cancer
and OC Use
277
Prevalence of OC Use The percentage of control women who ever used any OC, those of the CPA group, and those that contain CPA alone differed among the participating countries. These percentages for all controls, respectively, were as follows: Germany (69%, 15%, and 5%), UK (71%, 2%, and 2%), France (50%, 3%, and 2%), Italy (49%, 4%, and 4%), Greece (21%, 4%, and 4%), and Spain (23%, 0%, and 0%).
Validation of OC Use Differential recall of the OC type was considered as a potential problem in German centers because of a mass media campaign about an increased liver cancer risk for users of CPA containing OC. There was much less attention in other European countries regarding the risks of CPA. Therefore, a validation of the OC brands ever used according to the recall of the respondents was performed against the lifelong prescription data in a sample of cases and controls in Germany (125 of 285 invited persons who had ever used OC and who agreed to return to their prescribing doctors). Problems of the study included: no recall of doctor’s name/address; up to four doctors per respondent needed to be found and contacted; doctor moved away/died; offices closed; old files vanished or destroyed; and others. From the information obtained, we can summarize the following points. The congruence of information from doctor’s files and interview is high (8 1% and 9 1% for ever use in a certain period, for cases and controls, respectively). There was neither a significant difference between information obtained from personal interview or proxy information nor among OC types, particularly regarding CPA. The degree of congruence also did not significantly change with the time since last use, although the amount of information available from patient’s files declined the more we went back in time. A fairly high correlation coefficient was found between the recalled duration of use and the information available from physician’s files of prescriptions. Details will be published elsewhere (doctoral thesis of S. Hollmann in preparation; personal communication).
Risk Factors for HCC and Potential Confounders We combined all hospital and community controls for the main analysis. The odds ratios were significantly elevated in persons testing positive for hepatitis B (HBsAG, AntiHBs, AntiHBJ, and especially hepatitis C (AntiHC): 3.05 (2.18 to 4.25) and 37.88 (20.24 to 70.89), respectively. In data not shown here, these odds ratios did not differ appreciably when calculated separately using the two different hospital control
278
MILTS Project
Table 1. Frequency rows due to missing
Contraception 1997;56:275-284
Team
distribution information
and means (SD) for cases and two control
groups;
the total numbers
(N = 317)
~25 25-34 3544 45-54 55-64 Mean age (SD) t Mean year of birth (SD) t * Mean year of diagnosis/interview (SD)t+ Region: North (GER, FRA, UK)t Hepatitis serology available: yes-t Abdominal imaging tests: yes5 Occupational exposure: yest# Alcohol [regular use): yest Smoker (ever): yest Family history any cancer: yes Diabetes: yest l l Varicose veins: yest l * Gallbladder disease: yest l * Jaundice: yes* * Liver cirrhosis: yest l l Other liver disease: yes* l Surgery: yest * l Chronic diseases historyt’j Educational level: basict Town size: smallt
in some
Controls
Cases
Age:
differ slightly
Population
(N = 719)
Hospital
(N = 1060)
N or Mean
% or (SD)
N or Mean
% or (SD)
N or Mean
% or (SD)
20 21 47 74 155 50.4 42.2
6.3 6.6 14.8 23.3 49.9
2.5 8.5 9.3 26.3 53.4
;;;:;j
18 61 67 189 384 52.1 42.4
67 126 179 285 403 47.7 46.9
6.3 11.9 16.9 26.9 38.0* (12.6)* (12.6)*
93.1 248 260 317 125 253 114 139 26 76 62 109 116 16 205 237 179 52
(1.71 78.2 82.0 100.0 42.2 80.0 37.2 48.4 8.2 25.1 20.1 35.2 37.7 5.3 64.9 76.0 63.9 18.5
95.1 638 0 322 285 605 299 344 38 223 109 105 1 18 448 472 395 82
95.1 901 789 803 386 780 464 512 79 257 247 127 6 22 817 838 604 139
85.0* 74.4* 75.8* 36.4 73.6* 43.8 49.7 7.5 24.3* 23.4* 12.0 0.6 2.1 77.1* 79.1* 59.5 13.1
0 44.8 39.6 84.1 41.6 49.3 5.3 31.0 15.2 14.6 0.1 2.5 62.3 65.7 56.5 11.4
il.11
*Significant difference between control groups. tsignificantly associated with OC use. *Last two digits of the year. Qmestigations such as x-ray, ultrasound, MRT, CT etc. during last 5 years: yes or Rio. #Work in jobs with potential exposure to liver cancer risk factors: yes or Rio, * +Personal history [interview]: yes or no. ¶History of biliary or any liver disease or surgery: any vs none
groups (no blood samples taken from population controls). The odds ratios (OR) calculated for other reported risk factors for HCC in the aggregate, such as having ever been exposed to occupational hazards, hepatotoxic drugs, alcohol, smoking, diabetes, and family history of cancer were not found to be statistically significant (data not shown), but nevertheless were considered as confounders. Comparability of Various Subgroups The odds ratio for ever having used any OC was found to be 0.75 (0.54 to 1.03) when all cases were compared with all controls (Table 2), and compared to hospital and population controls separately: 1.13 (0.86 to 1.48) and 0.78 (0.59 to 1.03), respectively. The estimate does not differ appreciably from that of 0.81 (0.51 to 1.28) based on data obtained only from personal interviews, justifying the decision that data from
proxies will not be shown separately in the results to be subsequently presented. The comparable OR based on subjects without hepatitis serology (0.72 [0.37 to 1.631) was also similar to that based on data from the whole group. Types of OC and Risk for HCC The variables chosen for inclusion in the final analytic model were age, center, catchment area, year of birth, year of diagnosis or interview, history of chronic diseases, and occupation. Table 2 shows no significant alteration in risk for developing HCC in women who have ever used CPA containing OC, or for women ever using any other type of OC. The results of the unconditional and conditional logistic regression analysis are not significantly different from unity for either of the OC subgroups. Adjusting for hepatitis B or C had little effect on the overall risk estimates, and these variables were not
Contraception 1997;56:275-284
Table
2. Types
Liver Cancer
and OC Use
277
of OC used and risk of HCC OR (95% CI)
Comparison All cases and controls Never used Any OC Non-CPA CPA group CPA alone
Cases
corl~rols
145 148 123 25 10
693 1086 907 179 71
Adjusted
Effects of Duration and Last OC Use
informa-
of OC Use and Time Since First
3. Effect of duration
history age,
1 .OO 1.05 1.02 1.19 1.41
of chronic
center,
area,
&eases, adjusted
for
year
(referent) (0.80;1.37) [0.77;1.36) (0.69;2.05) (0.57;3.46] N = 24 cases
of birth,
year
of
of OC use on OR for HCC OR (95% Cases
Comparison All cases and controls Never use Any OC l-2 years 3-5 years 6+ years CPA group l-2 years 3-5 years 6+ years No cirrhosis and no hepatitis and C* Never use Any OC l-2 years 3-5 years 6+ years CPA group l-2 years 3-5 years 6+
occupation;
(B)
increasing time since first or last OC use (data not shown). In a small subgroup of women (5 1 cases) with no evidence of hepatitis B or C infection or liver cirrhosis (removing 83.9% of cases and 67.6% of the controls with liver cirrhosis or hepatitis B or C from the analysis, and those with missing information), the OR were higher for long term users of any type of OC, that is, statistically significant for over 6 years duration [matched OR 2.29 [1.05 to 5.0211 with a statisti-
Table 3 shows that the OR for HCC did not significantly increase with increased duration of use for all OC together, for the subgroup with CPA-group gestagens, or for CPA alone (data not shown). Similarly, no significant OR or trends in risk were found with
Table
Matched
1 .OO (referent) 0.75 (0.54;1.03) 0.74 [0.53;1.03J 0.89 (0.49;1.61) 0.89 (0.37;2.18)
use refers to Llse at any pmnt. Unconditional analysis: adjusted OR [A) for age, area, center, year of birth, year of diagnosis/interview, excluded from the model due to tmssing information OII confounders. As sensitivity analysis the conditional logistic regression (B] was performed (matched for diagnosis/mtervicw, history of chronic diseases, occupation).
included in the models because of missing tion for population controls.
(A)
Controls
Adjusted
(A)
CI) Matched
(B)
145
693
1.00 [referent)
1.00 (referent)
26 26 90
238 201 638
0.78 (0.47;1.31)+ 0.59 (0.33;1.06) 0.77 (0.54;1.12)
0.94 (0.60;1.48)* 0.87 (0.55;1.37) 1.13 (0.83;1.54)
7 7 9
57 51 58
0.93 (0.36;2.40)* 0.88 [0.33;2.36) 0.85 (0.35;2.04)
1.11 (0.46;2.69)* 1.01 (0.41;2.52] 1.21 [0.55;2.67)
16
250
1.00 (referent)
1.00 (referent)
5 5 25
74 57 193
1.26 (0.40;4.00)~ 1.77 (0.52;6.04) 2.81 (1.25;6.32)
1.14 (0.38;3.38)t 1.42 (0.48;4.23) 2.29 [1.05;5.02)
3 2 I
20 16 24
2.53 (0.57;11.25)* 1.58 (0.26;9.58) 0.64 (0.05;4.13)
2.71 (0.59;12.41]* 1.95 (0.34;11.23) 0.63 (0.07;5.47)
B
Unconditional logistic regression: adjusted (A] for age, center, area, year of diagnosis/interview, year of birth, history of chronic disease, occupation. Conditional logistic regressmn [Bj as sensitivity analysis: matched for age, centre, area, and adjusted for year of birth, year of diagnosis/interview, history of chronic disease, occupation, *No significant trend of duration. Significant trend of duration (p < 0.05]. *Matched for age only, due to small numbers, but adjusted for center, area, year of diagnosis/interview, year of birth, history of chronic diseases, occupation.
280
MILTS Project
Team
tally significant trend with duration (Table 3), but not significantly increased for long-term users of OC in the CPA group (matched OR 0.63 [0.07 to 5.471). However, the numbers of cases represented in the analysis of OC with progestins of the CPA group were very small. We found no convincing evidence of changing risk estimates among different categories of duration of OC use with time since first or last use (data not shown). When analyses were restricted to the small subgroup of HCC cases with evidence of no prior infections with HBV or HCV, and no liver cirrhosis in the course of the disease, the OR in women who had used OC for ~6 years, and who had first used them ~20 and 220 years previously were 1.70 (0.51; 5.75) and 3.18 (1.34; 7.58), respectively. There was a statistically significant trend of increasing risk with duration of use in the group of ~20 years since first use (p = 0.01). Women who had used OC >6 years or had last used them < 15 years ago also had a significantly increased odds ratio: (2.94 [ 1.18 to 7.351) and a positive trend for duration (p = 0.01). However, no such results were found in women who had long-term use of OC of the CPA group.
Discussion HCC Risk and OC Use in Aggregate In previous case-control studies,5-14 the risk of HCC in users of sex hormone contraceptives varied. In regions with low incidence of HCC (nonendemic for hepatitis B or C), a statistically significant increased risk was observed in women taking OC. Because, only two of these studies contained more than 35 liver cancer cases, however, the precision of the risk estimates was low. In addition, none of the case-control studies accounted for the possible confounding influence of HBV/HCV infections, and few of them controlled for other potential confounders, such as alcohol, smoking, hepatotoxic drugs, and occupational exposures. In studies conducted in areas with high rates of HCC, which are endemic for HBV/HCV, no associations of risk with oral or injectable contraceptives were observed. Additional limitations of the earlier studies were the inconsistent exclusion of HCC cases with cirrhotic livers among studies, inconsistent eligibility criteria for cases (histologic vs only clinical evidence), and lack of attention to the OC type or formulation. As derived from published literature, the working hypotheses that we tested in the current study were that the relative risk of HCC is increased in women who have ever used combined OC and in users of OC containing CPA or similar gestagens, and that risk is particularly increased in women with long-term use.
Contraception 1997;56:275-284
However, we did not find an increased liver cancer risk after adjusting for age, center, year of birth, year of diagnosis or interview, history of chronic disease, and occupation, although the study had adequate power to detect an increase in risk.’ No significantly increased risk of HCC was found with longer duration of OC use. Although previous publications15 had raised the issue that risk may be increased in longterm users only after a period of time since first or last use, we found no convincing evidence of changing risk estimates among different categories of duration of OC use with time since first or last use, except some weak evidence in the subgroup of cases without liver cirrhosis and negative hepatitis serology. In the small subgroup of cases without liver cirrhosis and negative hepatitis B or C serology (Table 3), we did observe in long-term users in the aggregate a significantly increasing trend in risk with duration of use. Although the numbers were small, the risk estimate for the duration category of >6 years was statistically significant, but the risk estimates may not be stable.” The increase in risk was observed within 15 years since initial exposure and persisted for >20 years since last use. This special subgroup of liver cancer, however, is uncommon (16% of all HCC cases in our studyJ and usually occurs at young ages. The results derived from this subgroup may, therefore, not be applicable to other forms for HCC, but they are consistent with prior reports of an association of long term OC use with liver cancer in areas of the world where this disease is rare. Our overall results are consistent with those of the World Health Organization (WHO) study,5 an investigation in South Africa,” and a study in the US” that showed no association with the duration or time since first or last OC use. Two very large cohort studies also found no association between liver cancer and oral contraceptive use?J2 Our findings strongly suggest that OC in aggregate do not enhance the carcinogenic effects of HBV or HCV for the human liver. The five studies from the early 1980&” that found much higher risk estimates than we did for use at any point, and particularly for long-term use, were conducted in countries where hepatitis B and C are nonendemic, as was a more recent study in Italy16 (continuation of a previous study”) that found a more modest increase in risk. Figure 1 shows the risk estimates from the various studies. The pooled results from the six studies in developed countries,6-12 where hepatitis B and C are less prevalent (nonendemic), produce a pooled risk estimate for use of OC at any time of 2.0 (1.4; 3.0). The comparable pooled estimate for ever use from the two studies in developing countries5,i3 is 1.02 (0.5; 2.2).
Contraception 1997;56:275-284
Liver Cancer
and OC Use
281
1983 Henderson
1986 Neuberger 1987 Forman
1989 WHO 1990 Kew 1991 WHO 1992 Hsing 1993 Tavan i
lm Ever
l7I Long-term
1997 MILTS 0 Figure
1. Relative
Use
risk estimates
10
5 for liver
cancer
in women;
The respective pooled OR for long-term use are 4.3 (2.3; 8.0) and 0.7 (0.2; 2.5); the tests for homogeneity within the two sets of studies were statistically nonsignificant. None of the early studies in developed countries could adequately determine whether the apparent effect of OC on HCC risk is independent of other HCC risk factors, particularly HBV/HCV. Some of the previous studies with higher OR had no histological confirmation of the diagnosis; many had no face-toface interview to gather valid information and almost none tried to validate the information about the exposure. Progressive correction of these problems may explain the lower risk estimates from the studies in the 1990s than from the earlier studies (Figure 1). Cyproterone Acetate Containing OC In the toxicological studies that raised suspicion about CPA and risk of liver cancer in humans, DNA adduct formation was observed after treatment with CPA in animals and human liver cell cultures.zp4 However, in our study, no association was observed between HCC and CPA (or other products in the same chemical class), either for use at any point or for duration of use or with respect to time since first or last use. This result is consistent with what was
15 exposure
Use
20 to any type of oral contraceptives.
found in an active surveillance study of persons with long-term or very high dose CPA use in a series of European centers. ” In a cohort followed for more than 17,000 treatment-years, no cases of HCC were found, although six were expected according to the age specific incidence rates.23 This result is also consistent with those from a case-control study that found no association between HCC and use of depot medroxyprogesterone acetate, which belongs to the CPA family. l4 On the other hand, in a report from Japan, three cases of hepatoma in teenagers (serious underlying defects) were reported after treatment with extremely high doses of CPA (and other drugs) at a very young age. ” A recent review of HCC case series in the world literature showed no association with CPA treatment.25 Limitations Our study had certain limitations. One problem is the different prevalence of OC use among different control groups (higher in population than hospital controls). If the cases represented all cases from a defined population, the relative risk estimates would seem most reliable if based on comparisons of cases vs population controls. On the other hand, for a hospital based case series, hospital controls might be preferred
282
MILTS Project
Contraception 1997;56:275-284
Team
to control for factors related to health surveillance issues. Our results showed (Table 1) that cases resemble hospital controls more than population controls with regard to history of diabetes, gallbladder disease, varicose veins, and chronic disease history, all of which have a somewhat lower OC prevalence. This distinction may explain the lower risk estimates for HCC in relation to OC use, when based on population controls that are a population random sample (data not shown). Although hospital controls with a diagnosis of chronic liver diseases that led to the hospital admission were excluded (but not so in population controls), this condition was rarely the reason for admission. Moreover, preferential prescribing of OC to HCC cases or controls testing positive for hepatitis B or C or with a history of cirrhosis is a problem. The lifetime exposure (use of any OC at any point) was not as different as expected, and was still fairly high: 38% and 48% in cases and controls with positive HBV/ HCV compared with about 60% with negative hepatitis tests. On the other hand, there is no good approach to correct this inherent problem, neither in this study nor in all previously published studies. A balance must be found between the results of a rather atypical group of liver cancer cases (removing persons with positive hepatitis serology or history of cirrhosis, which is part of the “natural history”) on one side and, on the other side, the effect of a preferential use of OC on the liver cancer risk estimates, which were not very impressive in our study compared with those in previous reports (Figure 1). But certainly, the hypothesis might be generated from our subgroup analysis that the subset of cases unrelated to HBV or HCV may be the type of HCC that is associated with OC use. Two other potential limitations are the different time periods when cases were accrued (prospective/ retrospective), and different sources of information on exposures (face-to-face interview/proxy information). The point estimates of the OR, however, did not show impressive differences when based on data from the retrospectively or concurrently ascertained controls, or when the cases with data from proxy interviews were excluded from the analyses (data not shown). Another limitation is the absence of hepatitis serology in population controls (rejected by one ethics committee).
estimate of OR = 1.05 for all OC, age specific OC prevalence rates, incidence rates of HCC, and population data for Germany, this would lead to one extra case per 3 million women-years that could be attributable to OC use, and similarly in other European countries. In the atypical subgroup of HCC cases without evidence of prior hepatitis and cirrhosis (16% of all cases in our study), about one extra case per 1.5 million women-years might be attributable to OC use. Thus, fear of liver cancer should not affect a woman’s decision about whether to use OCs.
Conclusion The risk of liver cancer was not significantly increased in women who ever took combined OCs; and no trend in risk was found with duration of use or with time since first or last use. The absence of an increasing risk with years of use cannot be attributed to insufficient statistical power or to insufficient attention to confounders, although the power became low for detecting an increased risk with duration of use in certain subgroups of women or for some OC formulations. We found no evidence that oral contraceptives in general, or specifically those containing CPA and chemically similar substances, increase the overall risk of liver cancer, even after long-term use or use >20 years in the past. This conclusion applies to countries with relatively low HCC incidence, that were the source of most cases in this study, but also seems applicable to countries in the southern rim of Europe, where HCC is more frequent. The hypothesis can be drawn from one subgroup analysis that rare liver cancer cases without previous hepatitis B or C or cirrhosis might be associated with long-term use of any OC, but the association probably has no public health importance.
Acknowledgments This research was funded by an unconditional grant from Schering AG, Berlin (Germany). The scientific work was exclusively discussed with and reviewed by the Scientific Advisory Committee. Special thanks is deserved for the work of the members of the editing committee: A.F. Feinstein and D. Thomas. The study was funded by Schering AG, Berlin.
References Absolute
and Population
Attributable
Risk of OC
The absolute risk of HCC in women under 6.5 years in Germany, as in many other European countries, is about 10 cases per 1 million women.26 Using the risk
1. Heinemann LAJ, Thomas DB, MGhner M, for the MILTS Collaborative Study Team. Multicentric international liver tumour study. Protocol of the case-control study on liver cancer. Pharmacoepidemiol Drug Safety 1996;5:173-86.
Contraception 1997;56:275-284
2. Neumann I, Thierau D, Andrae U, Greim H, Schwarz LR. Cyproterone acetate induces DNA damage in cultured rat hepatocytes and preferentially stimulates DNA synthesis in g-glutamyl-transpeptidase-positive cells. Carcinogenesis 1992;13:373-8. 3. Topinka J, Andrae U, Schwarz LR, Werner S, Schramm RJ, Wolff T. Persistence and accumulation of DNA adducts by the synthetic steroid cyproterone acetate in rat liver in vivo (poster). 23rd Annual Meeting of the European Environmental Mutagen Society (EEMS) in Barcelona, Sept. 27-Oct. 2, 1993. 4. Marelli A, Glua M, Mattioli F, Mereto E, Andrae U. Experimental evidence for a carcinogenic risk of cyproterone acetate to humans (poster). Congress0 Nationale di Farmacologia in Turin, Sept. 5-6, 1994. 5. WHO Collaborative Study of Neoplasia and Steroid Contraceptives. Combined oral contraceptives and liver cancer. Int J Cancer 1989;43:254-9. 6. Henderson BE, Preston-Martin S, Edmondson HA, Peters RL, Pike MC. Hepatocellular carcinoma and oral contraceptives. Br J Cancer 1983;48:437-40. 7. Neuberger J, Forman D, Doll R, Williams R. Oral contraceptives and hepatocellular carcinoma. Br Med J 1986;292: 135.5-7. 8. Forman D, Vincent TJ, Doll R. Cancer of the liver and the use of oral contraceptives. Br Med J 1986;292:13.5761. 9. La Vecchia C, Negri E, Parazzini F. Oral contraceptives and primary liver cancer. Br J Cancer 1989;59:460-1. 10. Palmer JR, Rosenberg L, Kaufmann DW, Warshauer ME, Stolley P, Shapiro S. Oral contraceptive use and liver cancer. Am J Epidemiol 1989;130:878-82. RN, McLaughlin JK, et al. Oral 11. Hsing AW, Hoover contraceptives and primary liver cancer among young women. Cancer Causes Control 1992;3:43- 8. 12. Tavani A, Negri E, Parazzini F, Franceschi S, La Vecchia C. Female hormone utihsation and risk of hepatocellular carcinoma. Br J Cancer 1993;67:635-7. 13. Kew MC, Song E, Mohammed A, Hodkinson J. Contraceptive steroids as a risk factor for hepatocellular carcinoma: a case/control study in South African black women. Hepatology 1990;11:298302. 14, WHO Collaborative Study of Neoplasia and Steroid Contraceptives. Depot-medroxyprogesterone acetate (DMPA) and risk of liver cancer. Int J Cancer 1991; 49:182-5. 15. Thomas DB. Exogenous steroid hormones and hepatocellular carcinoma, in: Tabor E, Di Bisceglie AM, Purcell RH (eds): Etiology, Pathology, and Treatment of Hepatocellular Carcinoma in North America. Advances in Applied Biotechnology Series, Vol 13. Houston: Gulf Printing Co., 1991:77-89. 16. Prentice RL. Epidemiologic data on exogenous hormones and hepatocellular carcinoma and selected other cancers. Prev Med 1991;20:38-46. 17. World Health Organization. The World Health Organization Collaborative Study of Cardiovascular Disease and Steroid Hormone Contraceptives: description and validation of methods. J Clin Epidemiol 1996;346:1582-8. 18. Nischan P, Ebeling K, Thomas DB, Hirsch U. Comparison of recalled and validated oral contraceptive histories. Am J Epidemiol 1993;138:697-703. 19. STATA. Statistical software package, version 4.0. Stata Press, Texas. TR, Feinstein 20. Peduzzi P, Conato J, Kemper E, Holford
Liver Cancer
21. 22. 23.
24.
25. 26.
and OC Use
283
AR. A simulation study of the number of events per variable in logistic regression analysis. J Clin Epidemiol 1996j49:1373-9. Colditz GA. Oral contraceptive use and mortality during 12 years of follow-up: the Nurses’ Health Study. Ann Intern Med 1994j20:821-6. Hannaford PC, Clifford RK, Vessey MP, Painter R, Mant J. Combined oral contraceptives and liver disease. Contraception 1997j55:145-51. Heinemann LAJ, Will-Shahab L, van Kesteren P, Gooren LJG. Safety of cyproterone acetate: report of active surveillance. Pharmacoepidemiol Drug Safety 1997j6:169-78. Watanabe S, Yamasaki S, Tanae A, Hibi I, Honna T. Three cases of hepatocellular carcinoma among cyproterone users. (letter to the editor). Lancet 1994jii(344): 1567-8. Raabe T, Feldmann K, Heinemann LAJ, Runnebaum B. Cyproterone acetate. Is it hepato- or genotoxic? Drug Safety 1996jl4:25-38. Muir C, Waterhouse J, Mack T, Powell J, Whelan S, eds. Cancer incidence in five continents. Vol. V. Lyon, France: International Agency for Research in Cancer, 1987.
Appendix Collaborating Centers (Key Personnel) Germany: H. Lochs, W. Wermke, B. Eckert
(Humboldt University Teaching Hospital Charite, Berlin); P. Neuhaus, W.O. Bechstein, J. Tio (Free University Teaching Hospital R. Virchow, Berlin); J. Boese-Landgraf, A. Wagner (Free University Teaching Hospital B. Franklin, Berlin);
H.H.
Pahlig,
H.H.
Greiner
(Hospital
Frieclrichs-
ham, Berlin); H. Koop, P. Flach, U. Masisus (Klinikum Berlin-Buch Teaching Hospital, Internal Medicine); U. Gottschalk (St. Hedwigs Teaching Hospital, Berlinj; T. Sauerbruch, U. Spengler (RFW-University Teaching Hospital, BO~)j F. Kunzelmann, W. Pile, M. Hoppe (Municipal Hospital, Chemnitz); H.D. Saeger, M. Kellner (Carl-Gustav-Cams University Teaching Hospital, Dresden); H. Porst, L. Hahn (Municipal Hospital Friedrichstadt, Dresden); E.G. Hahn, A.C. Herold, I. Kopp (University Teaching Hospital, Erkrngen); H. Goebell, B. Breuer-Katschinski, A. Marr (University Teaching Hospital, Essen); A. Encke, M. Lorenz, J. Richolt, H. Petrowsky (Johann-Wolfgang-Goethe University Teaching Hospital, Frankfurt-am-Main]; B. Ringe, F.P. Schulze (Georg-August University Teaching Hospital, Gottingen]; C. Broelsch, X. Rogiers, C. Brunken, C. Hillert, K. Pueschel, A. Heinemann (Universitatskrankenhaus Eppendorf, Universitat Hamburg Teaching Hospital, Hamburg); R. Pichlmayr, A. Weimann, K.J. Oldhafer, F.W. Schwartz, M. Schlaud, M.P. Manns, A. Schuler, A. Schauerte, A. Klein (Hannover Medical School, Hamrover]; G. Otto, H.G. Krumm (University Teaching Hospital, Heidelberg); A. Hoffmann, U. Merkel, H. Bosseckert, J. Scheele, 0. Wiesinger, K. Schmidt (University Teaching Hospital, Jena); D. Hemre-Bruns,
284
MILTS Project
Contraception 1997:56:275-284
Team
B. Kremer, T. Kuchler, G. Bottcher, U.R. Folsch, C. Stoffregen, V. Bohme, G. Lorentz (Christain-Albrecht University Teaching Hospital, Kiel); V. Diehl, C. Pohl, D. Scholten (University Teaching Hospital, Koln); J. Hauss, F. GeiI3ler (University Teaching Hospital, Leipzig); K.H. Meyer zum Btischenfelde, G. Gerken, M. Schmitgen (Johannes-Gutenberg University Teaching Hospital, Mainz); J. Sturm, K. Schuster, K. Ostreich (University Teaching Hospital, Mam-iheim); G. Paumgartner, G.R. Pape, W.H. Caselmann, F.W. Schildberg, K.W. Jauch, K.U. Grutzner (Ludwig-Maximilian University Teaching Hospital GroBhadem, Munchen); A.S. Kekule, G. Froesner (LM-University Max von Pettenkofer Institute, Munchen); M. Lustermann (Municipal Hospital, Nordhausen); G. Schott, M. Hoppe, S. Boethig (Municipal Hospital H. Braun, Zwickau). France: C. Brechot, C. Bernard, M.C. de Lamberterie (INSERM U370, Faculte de Medicine, Necker Enfants-Malades and Liver Unit, Institut Pasteur, Paris); Y. Calmus (Hopital Cochin, Paris); C. Vons (Hopital Antoine Beclere, Paris); J.C. Trinchet (Hopita Jean Verdier, Paris). United Kingdom: R. Williams, N.V. Naoumov, M.E. Cramp, H.A. Hodges (Kings College Hospital Institute of Liver Studies, London); C. McCollum, G.A. Riley (University Hospital of South Manchester, Manchester). Italy: M. Colombo, R. Romeo, L. Contu (University Institute for Internal Medicine, Milan). Greece: S.J. Hadziyamris, A. Papakonstantinou (Academic Department of Medicine, Hippokration General Hospital, Athens). Spain: J. Rod&, J.M. Llovet (Hospital Clinic I Provincial de Barcelona); S. Sans, G. Paluzie (Institut d’ Estudis de la Salut, Barcelona). Scientific Advisory Committee A.R. Feinstein (Yale University, New Haven, CT); C. B&hot (INSERM U370, Paris, France); V. Beral (ICRF Cancer Epidemiology Unit, Oxford, UK)j M.P. Manns (Medizinische Hochschule, Hannover, Germany); D.B. Thomas (Hutchinson Cancer Research Center, Seattle, WA)j S. Watanabe (National Cancer Center Research Institute, Tokyo Japan); 0. Meirik (World Health Organization Special Programme of Human W.O. Spitzer Reproduction, Geneva, Switzerland); (McGill University, Montreal, Quebec, Canada). MILTS Steering Committee Kekule (Ludwig-Maximilian A,S. Munchen, Germany); G. Gerken
Universitat, (Johannes-Guten-
berg Universitat, Mainz, Germany); W.O. Bechstein (Virchow Klinikum, Humboldt University, Berlin, Germany); K.W. Jauch (Ludwig-Maximilian Universitat, Munchen, Germany); M. Lorenz (JohannesGutenberg Universitat, Frankfurt-am-Main, Germany); T. Vogl (Rudolf-Virchow Universitatsklinikum, Berlin, Germany); A. Weimann (Hannover Medical School, Germany].
MILTS International Coordinating Center L.A. Heinemann, C. Thiel, S. Hollmann, K.H. Gunther, S. Mohner, W. Barth, C. Barth, K. Martin (Center for Epidemiology and Health Research [ZEG] Berlin, Zepernick, Germany).
MILTS Data Management Center T. DoMinh, P. Nischan, M. Mohner, E. Classen, L. Gravens (Centre for Epidemiology and Health Research [ZEG] Berlin, Zepernick, Germany).
Statistical Advisory Group I. Guggenmoos-Holzmann (Free University, Institute Medical Statistics and Data Processing, Berlin, Germany); H. Brenner (Unit of Epidemiology and Biostatistics, University of Ulm, Germany); J. Hanley (Department of Epidemiology and Biostatistics, McGill University, Montreal, Quebec, Canada); M. Mohner (Institute for Epidemiology, Forschungszentrum fur Umwelt and Gesundheit, Munchen, Germany).
Liver Pathology Advisors C. Wittekind (Department of Pathology, University of Erlangen, Germany); M. Dieted (Department of Pathology, Humboldt University Teaching Hospital Charite, Berlin, Germany).
Diagnosis Expert Panel H. Koop (Klinikum Berlin-Buch, Teaching Hospital, Internal Medicine, Berlin, Germany); A. Decker (Klinikum Berlin-Buch, Pathologisches Institut, Berlin, Germany).
Oral Contraceptives
Results of the Multicentre study (MILTS) The Collaborative
and Liver Cancer
International
MILTS Project Teams
Many, but not all, previous epidemiological studies indicated a greater risk of hepatocellular cancer (HCC) in women who have used combined oral contraceptives for a long period of time, but no one has analyzed this risk based upon use of different formulations. It was decided to analyze specifically the risk of OC containing cyproterone acetate (CPA) after toxicological experiments in animals found hints for a potential genotoxicity. This report describes the risk associated with ever having used combined oral contraceptives (OC) among 317 cases of primary hepatocellular cancer (HCC) in women under age 65, compared with 1060 age-matched hospital and 719population controls in a case-control study, which was conducted in six European countries. The adjusted odds ratio (unconditional logistic regression) for ever having used any OC was found to be 0.75 (0.54 to 1.03) when all cases were compared w.th all controls, and compared to hospital and population controls separately: 1.13 (0.86 to 1.48) and 0.78 (0.59 to 1.03), respectively. The adjusted odds ratios for OC containing all progestins of the CPA group were 0.89 (0.49 to 1.61); and 0.89 (0.37 to 2.18) for OC containing only CPA. There was no increase in risk for HCC with increasing duration of OC use among the different groups of OCs in the total group of cases with pooled controls. The risk estimates were not related to time since first or last use of any of the types of OCs considered. The most important risk factors for HCC were confirmed as a prior history of hepatitis B and C (adjusted odds ratio
6 This article was oreuared bv L.A.J. Heinemann.?# T. DoM1nh.t I. Guaaenboos-Holzmann (ie&ased),i C. Thiel,t and E. &&be.*t# Scleniific Ad%ory Committee, A.R. Felnstein (chair), D. Thomas, C. Brechot, W.0. Spltzer, S. Watanabe, V Beral, 0. Meink (see the Appendix); Collaborating centers and key personnel, other advisory committees, and reference expert groups were published elsewhere and are listed in the Appendix *Potsdam Institute of Pharmacoepldemiology and Technology Assessment (PIPTA), Potsdam, Germany, #Department of Epidemiology and Biostatistics, McGill University, Montrbal, Quhbec, Canada, tCentre for Epidemiology and Health Research (ZEG), Zepernick, Germany; *institute of Medical Statistics and lnformatics, Free University Berlin, Berlin, Germany. Name and address for correspondence, Prof. Dr. Lothar A.J. Helnemann, Centre for Epidemiology and Health Research Berlin, Schoenerlinder Strasse I I, 16341 Zepernick, Germany, Tel. 4930 9451 ,OiZO; Fax, 4930 9451 0123; e-mail zegr&nx.de Submitted for publication July 18, 1997 Revised August 22, 1997 Accepted for publication August 25, 1997
@ 1997 Elsevier Science lnc 655 Avenue of the Americas,
Liver Tumor
All rights reserved. New York, NY 10010
3.1 (2.2; 4.3) and 37.9 (20.2: 70.9) for HBV and HCV, respectively). In the small subgroup of HCC cases without liver cirrhosis and with negative serology for HBV and HCV, there was evidence of an association with duration of OC use. No such trend was observed for the CPA group of OCs. Altogether, there is no evidence for an increased risk of HCC associated with CPA or CPA-like OCs. Oral contraceptives in the aggregate may enhance the risk of liver carcinomas not associated with HBV or HCV infection, but if so, this is an extremely rare adverse effect of their use, CONTRACEPTION ~997i56~275-284 0 1997 Elsevier Science Inc. All rights reserved. WORDS: hepatocellular cancer, oral contraceptive, cyproterone acetate, case-control study, MILTS Study
KEY
Introduction he Multicentre International Liver Tumour Study (MILTS)’ was initiated after experimental reports suggested that cyproterone acetate (CPA] is genotoxic in rat hepatocytesz,3 and carcinogenic,4 but there were many other reasons for a new study. Although nine case-control studie+l4 and some reviews15J6 had been published on steroid contraceptives and liver cancer, no pharmacoepidemiological studies had been carried out to distinguish between the various formulations [estrogens, progestins] of oral contraceptives (OC). Also, studies in hepatitis B-endemic area&l3 generally did not show a relationship between OC and risk of hepatocellular cancer (HCC), whereas studies in nonendemic areaP2 tended to show an increase in risk in long term users, suggesting the need for a new study. In addition to hepatitis B virus (HBV), other risk factors for HCC have been identified. These include: hepatitis C, liver cirrhosis, possibly diabetes mellitus, alcohol abuse, possibly smoking, and nutritional factors (for example, aflatoxin), employment in the rubber or polyvinyl chloride (PVC) industry, occupational contact with oil products, hydrocarbons, dyes, and other chemicals, and use of potentially hepatotoxic drugs (for example, stanozolol, clofibrate, or
T
ISSN 0010.7624/97/$17.00 Pll SOOlO-7824(97)00158-3
276
MILTS Project
Team
azathioprine). These have not consistently been considered as possible confounders in prior studies of liver cancer and OC. The current study was aimed mainly at elucidating possible associations between risk of HCC and use of various types of OC with special emphasis on formulations containing CPA or progestins of the same family. Secondary aims were to investigate possible associations with OC use in subgroups of women with and without other risk factors for HCC, including HBV and hepatitis C virus (HCV).
Methods The study methods have been published in detail e1sewhere.l In brief, this hospital based case-control study (with hospital and community controls) was undertaken in six European countries between July 1994 and June 1996, designed to get a mix of population with lower and higher prevalence of OC use, and variation in the incidence of HCC as well hepatitis B and C. Women under the age of 6.5 with and without HCC were recruited. The project involved 64 clinics in 31 mainly university related settings, and combined for analyses in 12 regions (seven in Germany and one each in the United Kingdom, France, Italy, Greece, and Spain). The study covered two time periods: “historical” accrual for the 4 years before implementation of the study (July 1990 to June 1994) and “concurrent” accrual from July 1994 to June 1996. A monitoring system was established in each collaborating hospital to identify all female cases under the age of 65. Cases were classified as having a probable or definite diagnosis of HCC. A probuble diagnosis of HCC was based on a valid imaging result (sonography, computerized tomography, magnetic resonance tomography), and a clearly elevated blood a-fetoprotein level [500+ units). A definite diugnosis required histological confirmation. An expert group [including a reference pathologist) evaluated and classified all candidate cases and decided on inclusion or exclusion of equivocal cases by consensus. On average, four controls were originally to be obtained for each HCC case: two general hospital controls without malignancy; one hospital control with an eligible tumor diagnosis (as noted in the previously published list’); and one population control. Frequency matching was done in the same 5 year age groups as the cases; living controls were obtained for cases in which patients were deceased. Cases were identified in specified liver centers in the participating countries by monitoring all admission records and regular visits to relevant hospital wards. Similarly, the hospital controls were selected
Contraception 1997;56:275-284
from among new admissions with prespecified diagnoses by regular visits to specific wards on a rotating basis. Population controls were randomly selected from a complete population register in the respective and visited in their regions, invited to participate homes for a face-to-face interview (details have been published elsewhere’). All cases and controls were interviewed in a standardized manner. The questionnaire was designed to obtain information on demographic variables, personal disease history, family history of cancer, reproductive history, including lifetime history of OC use, lifetime use of particular drugs, smoking and alcohol use, and occupational history and related exposure indicators. Three other forms were used to abstract relevant information from the patients’ medical records (various diagnostic information, data on potential exposures or confounders). An extensive system of quality control measures was applied (details have been published elsewherel). For deceased cases or those who could not be interviewed for other reasons, a next-of-kin was interviewed. Proxies were individuals who were sufficiently well acquainted with the case to be able to respond to the interview, and included the person’s mother, sister, husband, or friend. When a proxy of the case was interviewed, a concurrent control was interviewed; that is, neither historical controls nor proxies of controls were used in this study. This method, which avoids high costs, is justified by the high congruence reported in the literature between results of interviews of proxies and interviews with controls.“~‘* Differences in duration of observation and time of possible exposure were controlled for in the analyses. Oral contraceptive use was the primary exposure variable of interest. We also distinguished in the analyses exposure to the subgroup of OC containing CPA-like progestins, that is, the chemically similar substances: cyproterone acetate (CPA), medroxyprogesterone acetate (MPA), and chlormadinone acetate (CMA), as well as CPA alone. The history of a hepatitis B (HBV) or C (HCV) virus infection was defined as negative if all serological tests that were made (HBsAG, AntiHBs, AntiHB=, AntiHC) were negative, and as positive if any of the tests were positive. Persons with missing data were excluded from this analysis. The lab tests were performed with commercially available test kits in the individual collaborating liver centers. Details will be published elsewhere. Unconditional logistic regression models were used to estimate crude and adjusted odds ratios using STATA@ software.19 Conditional logistic regression was performed as a second choice to make cases and
Contraception 1997:56:275-284
controls as similar as possible regarding the frequency matching variables (age, center, and catchment area). Variables were included in regression models if their inclusion altered the odds ratio by 10% or more, or if previous findings in the literature suggested that they were likely to be confounders.
Results Response Rates Of 368 eligible cases of liver cancer that were identified, 317 (86.1?’ 0 ) were included in the study: 179, 40, 29, 15, 14, and 40 came from Germany, UK, France, Italy, Greece and Spain, respectively. A total of 1060 hospital controls and 719 population controls were included in the study. The response rates for hospitalized patients (cases and hospital controls) ranged between 68% and 100% among the participating centers; the lowest rates were for cases that either were not identified while hospitalized or who died with no next-of-kin willing to participate. The response rates for population controls were lower than for hospitalized women and ranged from about 80% (UK, Italy, Greece) to between 60% and 70% (Germany, France, Spain).
Comparability of Case Subgroups Some of the HCC cases (who were either severely ill or deceased) were not able to give a personal interview; therefore, the information was obtained from a relative, friend, or other source (136 out of 317 cases). Cases with proxy information had a slightly higher prevalence of OCs at any time compared with those with personal interview (55.9% vs 48.1%, 46.3% vs 39.2%, and 7.4% vs 8.8% for all OC types, non-CPA groups, and CPA groups, respectively). The type of OC was unknown in three cases without a personal interview (2.2%), but for none with a personal interview. Both subgroups were not significantly different regarding exposure and the types of OC analyzed. As prespecified in the protocol, we pooled the subgroups with and without personal interview in the main analysis and do not present them separately here.
Comparability of Control Groups Table 1 shows some differences between hospital and population controls in such features as history of gallbladder disease, history of chronic disease, previous major surgery, and year of birth that are related to OC use. Blood samples were only taken from cases and hospital controls.
Liver Cancer
and OC Use
277
Prevalence of OC Use The percentage of control women who ever used any OC, those of the CPA group, and those that contain CPA alone differed among the participating countries. These percentages for all controls, respectively, were as follows: Germany (69%, 15%, and 5%), UK (71%, 2%, and 2%), France (50%, 3%, and 2%), Italy (49%, 4%, and 4%), Greece (21%, 4%, and 4%), and Spain (23%, 0%, and 0%).
Validation of OC Use Differential recall of the OC type was considered as a potential problem in German centers because of a mass media campaign about an increased liver cancer risk for users of CPA containing OC. There was much less attention in other European countries regarding the risks of CPA. Therefore, a validation of the OC brands ever used according to the recall of the respondents was performed against the lifelong prescription data in a sample of cases and controls in Germany (125 of 285 invited persons who had ever used OC and who agreed to return to their prescribing doctors). Problems of the study included: no recall of doctor’s name/address; up to four doctors per respondent needed to be found and contacted; doctor moved away/died; offices closed; old files vanished or destroyed; and others. From the information obtained, we can summarize the following points. The congruence of information from doctor’s files and interview is high (8 1% and 9 1% for ever use in a certain period, for cases and controls, respectively). There was neither a significant difference between information obtained from personal interview or proxy information nor among OC types, particularly regarding CPA. The degree of congruence also did not significantly change with the time since last use, although the amount of information available from patient’s files declined the more we went back in time. A fairly high correlation coefficient was found between the recalled duration of use and the information available from physician’s files of prescriptions. Details will be published elsewhere (doctoral thesis of S. Hollmann in preparation; personal communication).
Risk Factors for HCC and Potential Confounders We combined all hospital and community controls for the main analysis. The odds ratios were significantly elevated in persons testing positive for hepatitis B (HBsAG, AntiHBs, AntiHBJ, and especially hepatitis C (AntiHC): 3.05 (2.18 to 4.25) and 37.88 (20.24 to 70.89), respectively. In data not shown here, these odds ratios did not differ appreciably when calculated separately using the two different hospital control
278
MILTS Project
Table 1. Frequency rows due to missing
Contraception 1997;56:275-284
Team
distribution information
and means (SD) for cases and two control
groups;
the total numbers
(N = 317)
~25 25-34 3544 45-54 55-64 Mean age (SD) t Mean year of birth (SD) t * Mean year of diagnosis/interview (SD)t+ Region: North (GER, FRA, UK)t Hepatitis serology available: yes-t Abdominal imaging tests: yes5 Occupational exposure: yest# Alcohol [regular use): yest Smoker (ever): yest Family history any cancer: yes Diabetes: yest l l Varicose veins: yest l * Gallbladder disease: yest l * Jaundice: yes* * Liver cirrhosis: yest l l Other liver disease: yes* l Surgery: yest * l Chronic diseases historyt’j Educational level: basict Town size: smallt
in some
Controls
Cases
Age:
differ slightly
Population
(N = 719)
Hospital
(N = 1060)
N or Mean
% or (SD)
N or Mean
% or (SD)
N or Mean
% or (SD)
20 21 47 74 155 50.4 42.2
6.3 6.6 14.8 23.3 49.9
2.5 8.5 9.3 26.3 53.4
;;;:;j
18 61 67 189 384 52.1 42.4
67 126 179 285 403 47.7 46.9
6.3 11.9 16.9 26.9 38.0* (12.6)* (12.6)*
93.1 248 260 317 125 253 114 139 26 76 62 109 116 16 205 237 179 52
(1.71 78.2 82.0 100.0 42.2 80.0 37.2 48.4 8.2 25.1 20.1 35.2 37.7 5.3 64.9 76.0 63.9 18.5
95.1 638 0 322 285 605 299 344 38 223 109 105 1 18 448 472 395 82
95.1 901 789 803 386 780 464 512 79 257 247 127 6 22 817 838 604 139
85.0* 74.4* 75.8* 36.4 73.6* 43.8 49.7 7.5 24.3* 23.4* 12.0 0.6 2.1 77.1* 79.1* 59.5 13.1
0 44.8 39.6 84.1 41.6 49.3 5.3 31.0 15.2 14.6 0.1 2.5 62.3 65.7 56.5 11.4
il.11
*Significant difference between control groups. tsignificantly associated with OC use. *Last two digits of the year. Qmestigations such as x-ray, ultrasound, MRT, CT etc. during last 5 years: yes or Rio. #Work in jobs with potential exposure to liver cancer risk factors: yes or Rio, * +Personal history [interview]: yes or no. ¶History of biliary or any liver disease or surgery: any vs none
groups (no blood samples taken from population controls). The odds ratios (OR) calculated for other reported risk factors for HCC in the aggregate, such as having ever been exposed to occupational hazards, hepatotoxic drugs, alcohol, smoking, diabetes, and family history of cancer were not found to be statistically significant (data not shown), but nevertheless were considered as confounders. Comparability of Various Subgroups The odds ratio for ever having used any OC was found to be 0.75 (0.54 to 1.03) when all cases were compared with all controls (Table 2), and compared to hospital and population controls separately: 1.13 (0.86 to 1.48) and 0.78 (0.59 to 1.03), respectively. The estimate does not differ appreciably from that of 0.81 (0.51 to 1.28) based on data obtained only from personal interviews, justifying the decision that data from
proxies will not be shown separately in the results to be subsequently presented. The comparable OR based on subjects without hepatitis serology (0.72 [0.37 to 1.631) was also similar to that based on data from the whole group. Types of OC and Risk for HCC The variables chosen for inclusion in the final analytic model were age, center, catchment area, year of birth, year of diagnosis or interview, history of chronic diseases, and occupation. Table 2 shows no significant alteration in risk for developing HCC in women who have ever used CPA containing OC, or for women ever using any other type of OC. The results of the unconditional and conditional logistic regression analysis are not significantly different from unity for either of the OC subgroups. Adjusting for hepatitis B or C had little effect on the overall risk estimates, and these variables were not
Contraception 1997;56:275-284
Table
2. Types
Liver Cancer
and OC Use
277
of OC used and risk of HCC OR (95% CI)
Comparison All cases and controls Never used Any OC Non-CPA CPA group CPA alone
Cases
corl~rols
145 148 123 25 10
693 1086 907 179 71
Adjusted
Effects of Duration and Last OC Use
informa-
of OC Use and Time Since First
3. Effect of duration
history age,
1 .OO 1.05 1.02 1.19 1.41
of chronic
center,
area,
&eases, adjusted
for
year
(referent) (0.80;1.37) [0.77;1.36) (0.69;2.05) (0.57;3.46] N = 24 cases
of birth,
year
of
of OC use on OR for HCC OR (95% Cases
Comparison All cases and controls Never use Any OC l-2 years 3-5 years 6+ years CPA group l-2 years 3-5 years 6+ years No cirrhosis and no hepatitis and C* Never use Any OC l-2 years 3-5 years 6+ years CPA group l-2 years 3-5 years 6+
occupation;
(B)
increasing time since first or last OC use (data not shown). In a small subgroup of women (5 1 cases) with no evidence of hepatitis B or C infection or liver cirrhosis (removing 83.9% of cases and 67.6% of the controls with liver cirrhosis or hepatitis B or C from the analysis, and those with missing information), the OR were higher for long term users of any type of OC, that is, statistically significant for over 6 years duration [matched OR 2.29 [1.05 to 5.0211 with a statisti-
Table 3 shows that the OR for HCC did not significantly increase with increased duration of use for all OC together, for the subgroup with CPA-group gestagens, or for CPA alone (data not shown). Similarly, no significant OR or trends in risk were found with
Table
Matched
1 .OO (referent) 0.75 (0.54;1.03) 0.74 [0.53;1.03J 0.89 (0.49;1.61) 0.89 (0.37;2.18)
use refers to Llse at any pmnt. Unconditional analysis: adjusted OR [A) for age, area, center, year of birth, year of diagnosis/interview, excluded from the model due to tmssing information OII confounders. As sensitivity analysis the conditional logistic regression (B] was performed (matched for diagnosis/mtervicw, history of chronic diseases, occupation).
included in the models because of missing tion for population controls.
(A)
Controls
Adjusted
(A)
CI) Matched
(B)
145
693
1.00 [referent)
1.00 (referent)
26 26 90
238 201 638
0.78 (0.47;1.31)+ 0.59 (0.33;1.06) 0.77 (0.54;1.12)
0.94 (0.60;1.48)* 0.87 (0.55;1.37) 1.13 (0.83;1.54)
7 7 9
57 51 58
0.93 (0.36;2.40)* 0.88 [0.33;2.36) 0.85 (0.35;2.04)
1.11 (0.46;2.69)* 1.01 (0.41;2.52] 1.21 [0.55;2.67)
16
250
1.00 (referent)
1.00 (referent)
5 5 25
74 57 193
1.26 (0.40;4.00)~ 1.77 (0.52;6.04) 2.81 (1.25;6.32)
1.14 (0.38;3.38)t 1.42 (0.48;4.23) 2.29 [1.05;5.02)
3 2 I
20 16 24
2.53 (0.57;11.25)* 1.58 (0.26;9.58) 0.64 (0.05;4.13)
2.71 (0.59;12.41]* 1.95 (0.34;11.23) 0.63 (0.07;5.47)
B
Unconditional logistic regression: adjusted (A] for age, center, area, year of diagnosis/interview, year of birth, history of chronic disease, occupation. Conditional logistic regressmn [Bj as sensitivity analysis: matched for age, centre, area, and adjusted for year of birth, year of diagnosis/interview, history of chronic disease, occupation, *No significant trend of duration. Significant trend of duration (p < 0.05]. *Matched for age only, due to small numbers, but adjusted for center, area, year of diagnosis/interview, year of birth, history of chronic diseases, occupation.
280
MILTS Project
Team
tally significant trend with duration (Table 3), but not significantly increased for long-term users of OC in the CPA group (matched OR 0.63 [0.07 to 5.471). However, the numbers of cases represented in the analysis of OC with progestins of the CPA group were very small. We found no convincing evidence of changing risk estimates among different categories of duration of OC use with time since first or last use (data not shown). When analyses were restricted to the small subgroup of HCC cases with evidence of no prior infections with HBV or HCV, and no liver cirrhosis in the course of the disease, the OR in women who had used OC for ~6 years, and who had first used them ~20 and 220 years previously were 1.70 (0.51; 5.75) and 3.18 (1.34; 7.58), respectively. There was a statistically significant trend of increasing risk with duration of use in the group of ~20 years since first use (p = 0.01). Women who had used OC >6 years or had last used them < 15 years ago also had a significantly increased odds ratio: (2.94 [ 1.18 to 7.351) and a positive trend for duration (p = 0.01). However, no such results were found in women who had long-term use of OC of the CPA group.
Discussion HCC Risk and OC Use in Aggregate In previous case-control studies,5-14 the risk of HCC in users of sex hormone contraceptives varied. In regions with low incidence of HCC (nonendemic for hepatitis B or C), a statistically significant increased risk was observed in women taking OC. Because, only two of these studies contained more than 35 liver cancer cases, however, the precision of the risk estimates was low. In addition, none of the case-control studies accounted for the possible confounding influence of HBV/HCV infections, and few of them controlled for other potential confounders, such as alcohol, smoking, hepatotoxic drugs, and occupational exposures. In studies conducted in areas with high rates of HCC, which are endemic for HBV/HCV, no associations of risk with oral or injectable contraceptives were observed. Additional limitations of the earlier studies were the inconsistent exclusion of HCC cases with cirrhotic livers among studies, inconsistent eligibility criteria for cases (histologic vs only clinical evidence), and lack of attention to the OC type or formulation. As derived from published literature, the working hypotheses that we tested in the current study were that the relative risk of HCC is increased in women who have ever used combined OC and in users of OC containing CPA or similar gestagens, and that risk is particularly increased in women with long-term use.
Contraception 1997;56:275-284
However, we did not find an increased liver cancer risk after adjusting for age, center, year of birth, year of diagnosis or interview, history of chronic disease, and occupation, although the study had adequate power to detect an increase in risk.’ No significantly increased risk of HCC was found with longer duration of OC use. Although previous publications15 had raised the issue that risk may be increased in longterm users only after a period of time since first or last use, we found no convincing evidence of changing risk estimates among different categories of duration of OC use with time since first or last use, except some weak evidence in the subgroup of cases without liver cirrhosis and negative hepatitis serology. In the small subgroup of cases without liver cirrhosis and negative hepatitis B or C serology (Table 3), we did observe in long-term users in the aggregate a significantly increasing trend in risk with duration of use. Although the numbers were small, the risk estimate for the duration category of >6 years was statistically significant, but the risk estimates may not be stable.” The increase in risk was observed within 15 years since initial exposure and persisted for >20 years since last use. This special subgroup of liver cancer, however, is uncommon (16% of all HCC cases in our studyJ and usually occurs at young ages. The results derived from this subgroup may, therefore, not be applicable to other forms for HCC, but they are consistent with prior reports of an association of long term OC use with liver cancer in areas of the world where this disease is rare. Our overall results are consistent with those of the World Health Organization (WHO) study,5 an investigation in South Africa,” and a study in the US” that showed no association with the duration or time since first or last OC use. Two very large cohort studies also found no association between liver cancer and oral contraceptive use?J2 Our findings strongly suggest that OC in aggregate do not enhance the carcinogenic effects of HBV or HCV for the human liver. The five studies from the early 1980&” that found much higher risk estimates than we did for use at any point, and particularly for long-term use, were conducted in countries where hepatitis B and C are nonendemic, as was a more recent study in Italy16 (continuation of a previous study”) that found a more modest increase in risk. Figure 1 shows the risk estimates from the various studies. The pooled results from the six studies in developed countries,6-12 where hepatitis B and C are less prevalent (nonendemic), produce a pooled risk estimate for use of OC at any time of 2.0 (1.4; 3.0). The comparable pooled estimate for ever use from the two studies in developing countries5,i3 is 1.02 (0.5; 2.2).
Contraception 1997;56:275-284
Liver Cancer
and OC Use
281
1983 Henderson
1986 Neuberger 1987 Forman
1989 WHO 1990 Kew 1991 WHO 1992 Hsing 1993 Tavan i
lm Ever
l7I Long-term
1997 MILTS 0 Figure
1. Relative
Use
risk estimates
10
5 for liver
cancer
in women;
The respective pooled OR for long-term use are 4.3 (2.3; 8.0) and 0.7 (0.2; 2.5); the tests for homogeneity within the two sets of studies were statistically nonsignificant. None of the early studies in developed countries could adequately determine whether the apparent effect of OC on HCC risk is independent of other HCC risk factors, particularly HBV/HCV. Some of the previous studies with higher OR had no histological confirmation of the diagnosis; many had no face-toface interview to gather valid information and almost none tried to validate the information about the exposure. Progressive correction of these problems may explain the lower risk estimates from the studies in the 1990s than from the earlier studies (Figure 1). Cyproterone Acetate Containing OC In the toxicological studies that raised suspicion about CPA and risk of liver cancer in humans, DNA adduct formation was observed after treatment with CPA in animals and human liver cell cultures.zp4 However, in our study, no association was observed between HCC and CPA (or other products in the same chemical class), either for use at any point or for duration of use or with respect to time since first or last use. This result is consistent with what was
15 exposure
Use
20 to any type of oral contraceptives.
found in an active surveillance study of persons with long-term or very high dose CPA use in a series of European centers. ” In a cohort followed for more than 17,000 treatment-years, no cases of HCC were found, although six were expected according to the age specific incidence rates.23 This result is also consistent with those from a case-control study that found no association between HCC and use of depot medroxyprogesterone acetate, which belongs to the CPA family. l4 On the other hand, in a report from Japan, three cases of hepatoma in teenagers (serious underlying defects) were reported after treatment with extremely high doses of CPA (and other drugs) at a very young age. ” A recent review of HCC case series in the world literature showed no association with CPA treatment.25 Limitations Our study had certain limitations. One problem is the different prevalence of OC use among different control groups (higher in population than hospital controls). If the cases represented all cases from a defined population, the relative risk estimates would seem most reliable if based on comparisons of cases vs population controls. On the other hand, for a hospital based case series, hospital controls might be preferred
282
MILTS Project
Contraception 1997;56:275-284
Team
to control for factors related to health surveillance issues. Our results showed (Table 1) that cases resemble hospital controls more than population controls with regard to history of diabetes, gallbladder disease, varicose veins, and chronic disease history, all of which have a somewhat lower OC prevalence. This distinction may explain the lower risk estimates for HCC in relation to OC use, when based on population controls that are a population random sample (data not shown). Although hospital controls with a diagnosis of chronic liver diseases that led to the hospital admission were excluded (but not so in population controls), this condition was rarely the reason for admission. Moreover, preferential prescribing of OC to HCC cases or controls testing positive for hepatitis B or C or with a history of cirrhosis is a problem. The lifetime exposure (use of any OC at any point) was not as different as expected, and was still fairly high: 38% and 48% in cases and controls with positive HBV/ HCV compared with about 60% with negative hepatitis tests. On the other hand, there is no good approach to correct this inherent problem, neither in this study nor in all previously published studies. A balance must be found between the results of a rather atypical group of liver cancer cases (removing persons with positive hepatitis serology or history of cirrhosis, which is part of the “natural history”) on one side and, on the other side, the effect of a preferential use of OC on the liver cancer risk estimates, which were not very impressive in our study compared with those in previous reports (Figure 1). But certainly, the hypothesis might be generated from our subgroup analysis that the subset of cases unrelated to HBV or HCV may be the type of HCC that is associated with OC use. Two other potential limitations are the different time periods when cases were accrued (prospective/ retrospective), and different sources of information on exposures (face-to-face interview/proxy information). The point estimates of the OR, however, did not show impressive differences when based on data from the retrospectively or concurrently ascertained controls, or when the cases with data from proxy interviews were excluded from the analyses (data not shown). Another limitation is the absence of hepatitis serology in population controls (rejected by one ethics committee).
estimate of OR = 1.05 for all OC, age specific OC prevalence rates, incidence rates of HCC, and population data for Germany, this would lead to one extra case per 3 million women-years that could be attributable to OC use, and similarly in other European countries. In the atypical subgroup of HCC cases without evidence of prior hepatitis and cirrhosis (16% of all cases in our study), about one extra case per 1.5 million women-years might be attributable to OC use. Thus, fear of liver cancer should not affect a woman’s decision about whether to use OCs.
Conclusion The risk of liver cancer was not significantly increased in women who ever took combined OCs; and no trend in risk was found with duration of use or with time since first or last use. The absence of an increasing risk with years of use cannot be attributed to insufficient statistical power or to insufficient attention to confounders, although the power became low for detecting an increased risk with duration of use in certain subgroups of women or for some OC formulations. We found no evidence that oral contraceptives in general, or specifically those containing CPA and chemically similar substances, increase the overall risk of liver cancer, even after long-term use or use >20 years in the past. This conclusion applies to countries with relatively low HCC incidence, that were the source of most cases in this study, but also seems applicable to countries in the southern rim of Europe, where HCC is more frequent. The hypothesis can be drawn from one subgroup analysis that rare liver cancer cases without previous hepatitis B or C or cirrhosis might be associated with long-term use of any OC, but the association probably has no public health importance.
Acknowledgments This research was funded by an unconditional grant from Schering AG, Berlin (Germany). The scientific work was exclusively discussed with and reviewed by the Scientific Advisory Committee. Special thanks is deserved for the work of the members of the editing committee: A.F. Feinstein and D. Thomas. The study was funded by Schering AG, Berlin.
References Absolute
and Population
Attributable
Risk of OC
The absolute risk of HCC in women under 6.5 years in Germany, as in many other European countries, is about 10 cases per 1 million women.26 Using the risk
1. Heinemann LAJ, Thomas DB, MGhner M, for the MILTS Collaborative Study Team. Multicentric international liver tumour study. Protocol of the case-control study on liver cancer. Pharmacoepidemiol Drug Safety 1996;5:173-86.
Contraception 1997;56:275-284
2. Neumann I, Thierau D, Andrae U, Greim H, Schwarz LR. Cyproterone acetate induces DNA damage in cultured rat hepatocytes and preferentially stimulates DNA synthesis in g-glutamyl-transpeptidase-positive cells. Carcinogenesis 1992;13:373-8. 3. Topinka J, Andrae U, Schwarz LR, Werner S, Schramm RJ, Wolff T. Persistence and accumulation of DNA adducts by the synthetic steroid cyproterone acetate in rat liver in vivo (poster). 23rd Annual Meeting of the European Environmental Mutagen Society (EEMS) in Barcelona, Sept. 27-Oct. 2, 1993. 4. Marelli A, Glua M, Mattioli F, Mereto E, Andrae U. Experimental evidence for a carcinogenic risk of cyproterone acetate to humans (poster). Congress0 Nationale di Farmacologia in Turin, Sept. 5-6, 1994. 5. WHO Collaborative Study of Neoplasia and Steroid Contraceptives. Combined oral contraceptives and liver cancer. Int J Cancer 1989;43:254-9. 6. Henderson BE, Preston-Martin S, Edmondson HA, Peters RL, Pike MC. Hepatocellular carcinoma and oral contraceptives. Br J Cancer 1983;48:437-40. 7. Neuberger J, Forman D, Doll R, Williams R. Oral contraceptives and hepatocellular carcinoma. Br Med J 1986;292: 135.5-7. 8. Forman D, Vincent TJ, Doll R. Cancer of the liver and the use of oral contraceptives. Br Med J 1986;292:13.5761. 9. La Vecchia C, Negri E, Parazzini F. Oral contraceptives and primary liver cancer. Br J Cancer 1989;59:460-1. 10. Palmer JR, Rosenberg L, Kaufmann DW, Warshauer ME, Stolley P, Shapiro S. Oral contraceptive use and liver cancer. Am J Epidemiol 1989;130:878-82. RN, McLaughlin JK, et al. Oral 11. Hsing AW, Hoover contraceptives and primary liver cancer among young women. Cancer Causes Control 1992;3:43- 8. 12. Tavani A, Negri E, Parazzini F, Franceschi S, La Vecchia C. Female hormone utihsation and risk of hepatocellular carcinoma. Br J Cancer 1993;67:635-7. 13. Kew MC, Song E, Mohammed A, Hodkinson J. Contraceptive steroids as a risk factor for hepatocellular carcinoma: a case/control study in South African black women. Hepatology 1990;11:298302. 14, WHO Collaborative Study of Neoplasia and Steroid Contraceptives. Depot-medroxyprogesterone acetate (DMPA) and risk of liver cancer. Int J Cancer 1991; 49:182-5. 15. Thomas DB. Exogenous steroid hormones and hepatocellular carcinoma, in: Tabor E, Di Bisceglie AM, Purcell RH (eds): Etiology, Pathology, and Treatment of Hepatocellular Carcinoma in North America. Advances in Applied Biotechnology Series, Vol 13. Houston: Gulf Printing Co., 1991:77-89. 16. Prentice RL. Epidemiologic data on exogenous hormones and hepatocellular carcinoma and selected other cancers. Prev Med 1991;20:38-46. 17. World Health Organization. The World Health Organization Collaborative Study of Cardiovascular Disease and Steroid Hormone Contraceptives: description and validation of methods. J Clin Epidemiol 1996;346:1582-8. 18. Nischan P, Ebeling K, Thomas DB, Hirsch U. Comparison of recalled and validated oral contraceptive histories. Am J Epidemiol 1993;138:697-703. 19. STATA. Statistical software package, version 4.0. Stata Press, Texas. TR, Feinstein 20. Peduzzi P, Conato J, Kemper E, Holford
Liver Cancer
21. 22. 23.
24.
25. 26.
and OC Use
283
AR. A simulation study of the number of events per variable in logistic regression analysis. J Clin Epidemiol 1996j49:1373-9. Colditz GA. Oral contraceptive use and mortality during 12 years of follow-up: the Nurses’ Health Study. Ann Intern Med 1994j20:821-6. Hannaford PC, Clifford RK, Vessey MP, Painter R, Mant J. Combined oral contraceptives and liver disease. Contraception 1997j55:145-51. Heinemann LAJ, Will-Shahab L, van Kesteren P, Gooren LJG. Safety of cyproterone acetate: report of active surveillance. Pharmacoepidemiol Drug Safety 1997j6:169-78. Watanabe S, Yamasaki S, Tanae A, Hibi I, Honna T. Three cases of hepatocellular carcinoma among cyproterone users. (letter to the editor). Lancet 1994jii(344): 1567-8. Raabe T, Feldmann K, Heinemann LAJ, Runnebaum B. Cyproterone acetate. Is it hepato- or genotoxic? Drug Safety 1996jl4:25-38. Muir C, Waterhouse J, Mack T, Powell J, Whelan S, eds. Cancer incidence in five continents. Vol. V. Lyon, France: International Agency for Research in Cancer, 1987.
Appendix Collaborating Centers (Key Personnel) Germany: H. Lochs, W. Wermke, B. Eckert
(Humboldt University Teaching Hospital Charite, Berlin); P. Neuhaus, W.O. Bechstein, J. Tio (Free University Teaching Hospital R. Virchow, Berlin); J. Boese-Landgraf, A. Wagner (Free University Teaching Hospital B. Franklin, Berlin);
H.H.
Pahlig,
H.H.
Greiner
(Hospital
Frieclrichs-
ham, Berlin); H. Koop, P. Flach, U. Masisus (Klinikum Berlin-Buch Teaching Hospital, Internal Medicine); U. Gottschalk (St. Hedwigs Teaching Hospital, Berlinj; T. Sauerbruch, U. Spengler (RFW-University Teaching Hospital, BO~)j F. Kunzelmann, W. Pile, M. Hoppe (Municipal Hospital, Chemnitz); H.D. Saeger, M. Kellner (Carl-Gustav-Cams University Teaching Hospital, Dresden); H. Porst, L. Hahn (Municipal Hospital Friedrichstadt, Dresden); E.G. Hahn, A.C. Herold, I. Kopp (University Teaching Hospital, Erkrngen); H. Goebell, B. Breuer-Katschinski, A. Marr (University Teaching Hospital, Essen); A. Encke, M. Lorenz, J. Richolt, H. Petrowsky (Johann-Wolfgang-Goethe University Teaching Hospital, Frankfurt-am-Main]; B. Ringe, F.P. Schulze (Georg-August University Teaching Hospital, Gottingen]; C. Broelsch, X. Rogiers, C. Brunken, C. Hillert, K. Pueschel, A. Heinemann (Universitatskrankenhaus Eppendorf, Universitat Hamburg Teaching Hospital, Hamburg); R. Pichlmayr, A. Weimann, K.J. Oldhafer, F.W. Schwartz, M. Schlaud, M.P. Manns, A. Schuler, A. Schauerte, A. Klein (Hannover Medical School, Hamrover]; G. Otto, H.G. Krumm (University Teaching Hospital, Heidelberg); A. Hoffmann, U. Merkel, H. Bosseckert, J. Scheele, 0. Wiesinger, K. Schmidt (University Teaching Hospital, Jena); D. Hemre-Bruns,
284
MILTS Project
Contraception 1997:56:275-284
Team
B. Kremer, T. Kuchler, G. Bottcher, U.R. Folsch, C. Stoffregen, V. Bohme, G. Lorentz (Christain-Albrecht University Teaching Hospital, Kiel); V. Diehl, C. Pohl, D. Scholten (University Teaching Hospital, Koln); J. Hauss, F. GeiI3ler (University Teaching Hospital, Leipzig); K.H. Meyer zum Btischenfelde, G. Gerken, M. Schmitgen (Johannes-Gutenberg University Teaching Hospital, Mainz); J. Sturm, K. Schuster, K. Ostreich (University Teaching Hospital, Mam-iheim); G. Paumgartner, G.R. Pape, W.H. Caselmann, F.W. Schildberg, K.W. Jauch, K.U. Grutzner (Ludwig-Maximilian University Teaching Hospital GroBhadem, Munchen); A.S. Kekule, G. Froesner (LM-University Max von Pettenkofer Institute, Munchen); M. Lustermann (Municipal Hospital, Nordhausen); G. Schott, M. Hoppe, S. Boethig (Municipal Hospital H. Braun, Zwickau). France: C. Brechot, C. Bernard, M.C. de Lamberterie (INSERM U370, Faculte de Medicine, Necker Enfants-Malades and Liver Unit, Institut Pasteur, Paris); Y. Calmus (Hopital Cochin, Paris); C. Vons (Hopital Antoine Beclere, Paris); J.C. Trinchet (Hopita Jean Verdier, Paris). United Kingdom: R. Williams, N.V. Naoumov, M.E. Cramp, H.A. Hodges (Kings College Hospital Institute of Liver Studies, London); C. McCollum, G.A. Riley (University Hospital of South Manchester, Manchester). Italy: M. Colombo, R. Romeo, L. Contu (University Institute for Internal Medicine, Milan). Greece: S.J. Hadziyamris, A. Papakonstantinou (Academic Department of Medicine, Hippokration General Hospital, Athens). Spain: J. Rod&, J.M. Llovet (Hospital Clinic I Provincial de Barcelona); S. Sans, G. Paluzie (Institut d’ Estudis de la Salut, Barcelona). Scientific Advisory Committee A.R. Feinstein (Yale University, New Haven, CT); C. B&hot (INSERM U370, Paris, France); V. Beral (ICRF Cancer Epidemiology Unit, Oxford, UK)j M.P. Manns (Medizinische Hochschule, Hannover, Germany); D.B. Thomas (Hutchinson Cancer Research Center, Seattle, WA)j S. Watanabe (National Cancer Center Research Institute, Tokyo Japan); 0. Meirik (World Health Organization Special Programme of Human W.O. Spitzer Reproduction, Geneva, Switzerland); (McGill University, Montreal, Quebec, Canada). MILTS Steering Committee Kekule (Ludwig-Maximilian A,S. Munchen, Germany); G. Gerken
Universitat, (Johannes-Guten-
berg Universitat, Mainz, Germany); W.O. Bechstein (Virchow Klinikum, Humboldt University, Berlin, Germany); K.W. Jauch (Ludwig-Maximilian Universitat, Munchen, Germany); M. Lorenz (JohannesGutenberg Universitat, Frankfurt-am-Main, Germany); T. Vogl (Rudolf-Virchow Universitatsklinikum, Berlin, Germany); A. Weimann (Hannover Medical School, Germany].
MILTS International Coordinating Center L.A. Heinemann, C. Thiel, S. Hollmann, K.H. Gunther, S. Mohner, W. Barth, C. Barth, K. Martin (Center for Epidemiology and Health Research [ZEG] Berlin, Zepernick, Germany).
MILTS Data Management Center T. DoMinh, P. Nischan, M. Mohner, E. Classen, L. Gravens (Centre for Epidemiology and Health Research [ZEG] Berlin, Zepernick, Germany).
Statistical Advisory Group I. Guggenmoos-Holzmann (Free University, Institute Medical Statistics and Data Processing, Berlin, Germany); H. Brenner (Unit of Epidemiology and Biostatistics, University of Ulm, Germany); J. Hanley (Department of Epidemiology and Biostatistics, McGill University, Montreal, Quebec, Canada); M. Mohner (Institute for Epidemiology, Forschungszentrum fur Umwelt and Gesundheit, Munchen, Germany).
Liver Pathology Advisors C. Wittekind (Department of Pathology, University of Erlangen, Germany); M. Dieted (Department of Pathology, Humboldt University Teaching Hospital Charite, Berlin, Germany).
Diagnosis Expert Panel H. Koop (Klinikum Berlin-Buch, Teaching Hospital, Internal Medicine, Berlin, Germany); A. Decker (Klinikum Berlin-Buch, Pathologisches Institut, Berlin, Germany).