Editor's Choice


One million women

In the last decade, it was third-generation oral contraception that aroused scientific and legal passions; in this decade, it will be hormone replacement therapy. This is why the survey reported by the Million Women Collaborators (pages 1319–1330) is so important, the aims of the survey being to investigate the effects of hormone replacement therapy on health. The authors administered questionnaires to more than one million women aged 50 to 64 years who were attending breast screening clinics in the UK between 1996 and 1999. We can be confident that this sample of women is typical of all women of this age in the UK, since their use of drugs for the common illnesses of middle age is the same as the general population.

The most striking feature of the survey is that there were strong variations in the use of hormone replacement therapy according to medical history, and that these variations were more important than socio-economic factors and lifestyle. Overall, about one-third of the women were current users of hormone replacement therapy, but this fraction increased to two-thirds in women who had undergone bilateral oophorectomy and to one-half in women who had undergone hysterectomy without oophorectomy. Nearly all the women with a history of bilateral oophorectomy and hysterectomy received oestrogen only, while nearly all the women with no history of these procedures received combined oestrogen and progestogen preparations.

More startling still is the frequent use of hormone replacement therapy in women with a history of breast cancer, stroke, thromboembolism or heart disease. The Journal received this article before the results of the Women's Health Initiative trial1 were known. This randomised trial, involving 16,000 women, showed that a continuous combined oestrogen–progestogen preparation increased the risk of breast cancer by 26%, of stroke by 41%, of coronary heart disease by 29% and of venous thomboembolism by a factor of two. Valerie Beral and her colleagues2 have summarised the evidence from randomised trials, confirming the risks of hormone replacement therapy. In view of this recent evidence, we should be concerned that in this survey 1 in 20 women with a history of breast cancer, one-quarter of women with a history of stroke or thromboembolism and nearly one-third of women with a history of heart disease should be taking hormone replacement therapy. These proportions are greater in women who have undergone bilateral oophorectomy or hysterectomy without oophorectomy. This emphasises the importance of the Million Women Study, for we should follow this cohort to determine not only changes in the use of hormone replacement therapy, but also the effect of different formulations of hormone replacement therapy on the development of disease, and on mortality.

Not a systematic review

Have you ever wondered why in one study the relative risk of cervical neoplasia in association with the human papillomavirus was 1.2 and in another it was 327? With a hint of understatement, Ciaran Woodman and Stuart Collins (pages 1311–1318) suggest that this variation cannot be explained by biological factors alone. The authors reviewed all the available cohort studies of the association of the human papillomavirus and cervical neoplasia, but did not conduct a systematic review; rather, their purpose is to describe the limitations of the various cohort studies that lead to such wild variations in the estimates of the risk of cervical neoplasia associated with the human papillomavirus. In doing this, they remind us of the important principles of cohort studies.

The association of cervical neoplasia and the human papillomavirus was established by case-control studies, but in order to determine causation, the temporal nature between the exposure, the human papillomavirus, and the outcome, cervical neoplasia, should be ascertained. This is why cohort studies are important. Thus, at the beginning of the study, neither cases nor controls should have cervical neoplasia, and at the end of the study, the frequency of cervical neoplasia in women exposed to the human papillomavirus is compared with the frequency of cervical neoplasia in women not so exposed. Allowance is made for the duration of exposure to the onset of the disease, and because women are exposed for varying lengths of time, the statistical techniques of survival analysis should be used. A nested case-control design is commonly employed, where the cytological samples are obtained from the cervix of every woman in the cohort and are stored. They are analysed for the human papillomavirus in the women who develop cervical neoplasia. Controls are then selected and matched for age, duration of follow up and the date of the initial sample, and their cervical samples similarly analysed for the human papillomavirus. By definition, the controls should be free of cervical neoplasia. The frequency of human papillomavirus in the women with cervical neoplasia is compared with the frequency of human papillomavirus in the controls.

In the cohort studies they have analysed, Woodman and Collins describe limitations in virtually all aspects of the design of cohort studies. These methodological problems must account for much of the wide variation in the risk of human papillomavirus in association with cervical neoplasia. For the physician without a background in epidemiology, their paper may be hard going, but it makes rewarding reading. Thirteen years ago, we published guidelines for the reporting of observational studies in the Journal1, and Woodman and Collins in their article make 11 recommendations concerning the design of cohort studies. We can use these recommendations in our own appraisal of the strengths and limitations of cohort studies published in medical journals.

ORACLE: divine ordinance or a subject for debate?

The ORACLE studies published in 20011,2 were the largest randomised trials to investigate the effect of three antibiotic regimens in preterm labour and prelabour rupture of the membranes. There were no benefits with antibiotics in preterm labour and only a modest benefit with erythromycin in a reduction of a composite neonatal outcome. Six months after publication of the trial, Sara Kenyon and David Taylor (pages 1341–1343) performed a survey of maternity units in the UK to determine the change in clinical practice brought about by the trial. In prelabour rupture of the membranes, fewer maternity units prescribed antibiotics overall, erythromycin was prescribed more frequently and co-amoxiclav less frequently. One-third of the maternity units stopped prescribing co-amoxiclav, in view of the association of necrotizing enterocolitis with this drug. In preterm labour, there was a slight increase in the number of women prescribed erythromycin. About half the maternity units had changed their guidelines for the treatment of prelabour rupture of the membranes.

The authors emphasise how slowly clinical practice changes in the light of evidence, citing examples as diverse as the Royal Navy in 1601 and a systematic review of corticosteroids in preterm labour in 1990. Several reasons are given for this slowness, but one major factor not mentioned by the authors is the culture in which the evidence from research accumulates. Thus, in the Royal Navy between 1601 and 1795, the health of the ships' crew was considered of lesser importance to the Admiralty than the construction and ordnance of ships-o'-the-line. The 13-year delay between the demonstration of the effectiveness of thrombolytic therapy for myocardial infarction can be accounted for by physicians' lack of awareness of the importance of randomised trials—after all, the evidence accumulated before the invention of meta-analysis and concepts of evidence-based medicine. The surprising feature about Kenyon and Taylor's survey is not that so few maternity units changed their clinical practice within six months, but that so many did so. We live in a culture of randomised trials, meta-analysis, clinical practice guidelines, targets and league tables. There is now a dangerous tendency that the results of randomised trials are accepted as divine ordinance, to be acted upon at once without question. One difficulty with the ORACLE trials is their two-by-two factorial design, resulting in many statistical tests and the danger that an outcome of borderline statistical significance may have occurred by chance. The evidence supporting the use of erythromycin in prelabour rupture of the membranes and condemning co-amoxiclav may not be as secure as we thought. The reaction of physicians to the results of randomised trials should not be to accept them uncritically and incorporate them into guidelines, but to analyse the trial for themselves and judge the results according to their relevance to their own clinical practice.

Active management of the first and second stages of labour

Prelabour rupture of the membranes at term is common, occurring in one in twelve pregnancies, and is a nuisance, because of uncertainty about its treatment. The largest randomised trial1 showed that there were no substantive differences in maternal or neonatal outcome with active compared with conservative treatment, a finding which has resulted in varying policies in labour wards. A conservative policy is attractive, since it minimises the discomforts associated with induction of labour; on the other hand a conservative policy taken too far will result in the rare case of serious maternal or neonatal infection. Labour wards therefore tend to adopt a middle course, where induction of labour is performed after a period of conservative treatment, taking into account the woman's preference for immediate induction or a conservative policy.

Misoprostol is a prostaglandin which has been shown to be effective in induction of labour where the membranes have not ruptured, but little is known of its efficacy in prelabour rupture of the membranes at term. A. Shetty and colleagues (pages 1354–1358) carried out a randomised trial to determine the effectiveness of a regimen of oral misoprostol in inducing labour, and to compare active management using oral misoprostol with their standard conservative regimen. More than nine-tenths of the women randomised to misoprostol were in labour within 24 hours, a much greater proportion than with the conservative regimen. There was no instance of hyperstimulation of the uterus, and the frequencies of pyrexia, nausea and vomiting were no different. Although more women expressed a preference for active management with misoprostol, the difference was not statistically significant. The conclusion of this trial is that oral misoprostol is a suitable method of induction of labour in women wishing active management.

With the widespread use of epidural analgesia in labour the practice has crept in, unobserved, of delaying the active second stage to allow the infant's head to descend into the pelvis and so reduce the risk of instrumental delivery. But is this practice justified? This was the question asked by Myra Fitzpatrick and her colleagues (pages 1359–1365), who conducted a randomised trial to test the hypothesis that delaying the active second stage by sixty minutes would reduce the rate of instrumental vaginal delivery in primiparae in labour with epidural analgesia. Secondary outcomes were the pressures on anal manometry, the findings on endosonography and the effects on the neurophysiology of the pudendal nerve. Delaying the active second stage of labour made no difference to any of these outcomes, and the frequency of operative vaginal delivery may even be increased by delay (relative risk 1.14; 95% confidence interval 0.81, 1.61). There is no evidence in this trial to support a policy of delaying the active second stage in primiparae in labour with epidural analgesia.