Case-control study of oestrogen replacement therapy and risk of cervical cancer

Introduction

Convincing epidemiological evidence exists that hormonal factors have a role in the development of cervical cancer. For example, several studies have shown that parity1 2 3 and use of oral contraceptives1 4 5 increase the risk of invasive cervical cancer. These findings suggest that oestrogen-progestin stimulation can favour, or accelerate, cervical carcinogenesis, possibly through glucocorticoid dependent oncogenic transformation by selected papilloma viruses.6 It has also been suggested that this may be a hormone mediated effect on one of the late stages of carcinogenesis.7

Little information is available, however, on the separate role of oestrogens and progestogens on cervical carcinogenesis. In particular, few data have been published on the potential relation between oestrogen replacement therapy and risk of invasive cervical cancer.8 We present the results of a case-control study of the effects of oestrogen replacement therapy on cervical cancer.

Subjects and methods

The design of this study has been described.3 Briefly, the study was a hospital based case-control investigation conducted in the greater Milan area, Northern Italy, on 645 women aged 40-75 years with histologically confirmed invasive cervical cancer who were admitted during 1981-93 to the obstetrics and gynaecology clinics of the University of Milan, the National Cancer Institute, and the Ospedale Maggiore of Milan (which includes the four largest hospitals in Milan). Of these women, 368 (57%) had squamous cancers and 116 (18%) adenocarcinoma; the 161 (25%) remaining women had another or undefined histological type. The comparison group consisted of 749 women aged 40-75 years with acute conditions judged to be unrelated to any of the known or suspected risk factors for cervical cancer who were admitted to the same hospitals where the cases had been identified (mainly the Ospedale Maggiore and several specialised university clinics). The control women were therefore from similar catchment areas to the cases. Controls were not individually matched but selected within comparable age strata of cases. They were not included if they were admitted for gynaecological, hormonal, or neoplastic diseases or had had a total hysterectomy. Of the 749 controls, 202 (27%) were admitted for trauma (mostly fractures and sprains), 270 (36%) had non-traumatic orthopaedic disorders (mostly lower back pain and disc disorders), 97 (13%) surgical conditions (mostly abdominal, such as acute appendicitis or strangulated hernia), and 180 (24%) other illnesses such as ear, nose, and throat or dental disorders. Less than 2% of eligible women (cases and controls) refused to be interviewed.

The structured questionnaire included information on personal characteristics and habits; education and other socioeconomic factors; general lifestyle habits such as smoking and alcohol and coffee consumption; a few indicators of sexual habits; gynaecological and obstetric data; number of cervical smears; related medical history; and lifetime use of oral contraceptives, hormonal replacement therapy in menopause, and female hormone preparations for other indications. The time and duration of each episode of use and brand name were registered, whenever available. Lists of the most common female hormone preparations (covering over 90% of those marketed over the past two decades) were provided to assist recall.

All interviews were conducted in hospital. The same questionnaire was used for cases and controls, and the same interviewers interviewed cases and controls.

Results

Table 1 gives the distribution of cases and controls according to age and selected risk factors for cervical cancer. Cases reported more sexual partners and births than controls and significantly fewer cervical smear tests.

Table 2 gives the relative risks for use of oestrogen replacement therapies. Forty cases versus 86 controls had ever taken oestrogens, and the corresponding multivariate odds ratio was 0.5 (95% confidence interval 0.3 to 0.8). Five cases and six controls reported combined oestrogen and progestogen use. The odds of cervical cancer fell with duration of use, being 0.6 (0.4 to 1.1) for use for under 12 months and 0.5 (0.2 to 1.0) for 12 months or more compared with never users. The protection tended to be somewhat stronger for women reporting first oestrogen use before the age of 50. The odds ratio was 0.9 (0.5 to 1.7) for women who had taken oestrogens within the past 10 years and 0.4 (0.2 to 0.7) for those who had taken them 10 or more years ago.

Discussion

This is one of few epidemiological studies on a meaningful number of subjects that provides reassuring information on oestrogen use and cervical carcinogenesis. The potential limitations of this study should, however, be considered. Firstly, the study was hospital based, with subjects collected from the main general and teaching hospitals in the greater Milan area. Although the study protocol indicated that all new consecutive cases should be interviewed, the design was not strictly population based, and it is likely that some subjects did not enter the study (for instance, because they were not present in the ward at the time of the interviewer's visit). Furthermore, women admitted to general and teaching hospitals are different from those treated in private or smaller institutions in terms of sociodemographic characteristics, and social class is related to use of oestrogen replacement therapy.11

Another important concern is the inclusion in the comparison group of subjects with orthopaedic diseases or trauma. Women with fractures or sprains may represent a group unusually active and therefore fitter than the cases. Alternatively, they may be selectively oestrogen deficient. These potential biases might, however, tend to reduce the protective effect of oestrogen observed in this analysis. We found no difference in the estimated odds ratio when the analysis was conducted after exclusion of controls with orthopaedic diseases or trauma (multivariate odds ratio for ever v never users 0.6, 95% confidence interval 0.3 to 1.1). Another potential limitation of this study is the low prevalence of menopausal replacement treatment in the population, which hampered detailed analysis of subgroups and interactions and might be the cause of selective mechanisms. However, the proportion of women taking oestrogen in the series was consistent with national data in Italy.12

With regard to other potential sources of bias, participation was almost complete, and there is no reason to suggest differential recall of hormone use by cases and controls. If anything, women with cervical cancer would be more likely (rather than less) to recall taking female hormones. Furthermore, the hospital based design may improve the comparability of drug recall by cases and controls.13 In addition, adjustment for several potential confounding factors did not substantially modify the odds ratios.

More accurate adjustment for confounding could further reduce the apparent association between oestrogens and risk of cervical cancer. However, the inclusion of terms for education and age at menopause, which may well imply overadjustment, gave an odds ratio of 0.7 (0.4 to 1.0).