Introduction

This case study is designed to enable you to compare your responses to the case study with other participating GPs and with our efforts to provide evidence-based feedback.  There are no clear correct answers because the interpretation of the options given is subjective.  We would be happy with any answers near ours.  One of us wrote the case scenario and questions.  We were not authors of any other part of NPS news 12: HRT.

Some doctors believe that the menopause is a disease like thyroid failure that always requires “replacement” therapy.  Others believe that the menopause is a normal life transition like puberty. Therapy recommendations should be based on evidence about whether or not benefit exceeds harm regardless of such personal beliefs.  We consider the term “Hormone Replacement Therapy” is a promotional rather than a neutral name for postmenopausal oestrogen and progestogen.[1]  However we will use the term HRT for consistency with NPS publications. 

Australian doctors have received conflicting messages about HRT.  Important new results were published during 2000 without much publicity.  The case scenario and questions were based on the HERS study because it is the largest randomised controlled trial of HRT published so far.[2]  HRT is effective for menopausal symptoms.  However, we chose to focus on potentially life threatening outcomes measured in the HERS trial such as fractures, cardiovascular disease and breast cancer.  There has been controversy about how to interpret the findings of the HERS trial.  Consequently it is no surprise that the responses to this case study include a wide range of views.  Differing beliefs often result from different ways of reaching conclusions so we will start by explaining the approach that we have used to evaluate the evidence and then discuss some of the controversy about the HERS trial.

Randomised Controlled Trials (RCTs) are more reliable than observational studies.

RCTs compare outcomes for people who were randomly allocated to take a therapy with those randomly allocated to a comparator (a placebo or an alternative therapy).  If therapy (A) changes the rate of an event (B) in an RCT then we can be confident that using (A) instead of the comparator for similar people causes the difference in the rate of (B).  By contrast, observational studies compare outcomes for people who are on a therapy versus those who are not because they decided not to take the therapy or just never decided.  Such studies suffer from many biases including the “healthy volunteer effect”.  Consequently, observational studies do not reliably tell us if (A) causes (B).  Observational studies can provide very useful information but require careful interpretation.

The study of women that found an association between agreeing to have mammography and lower heart attack rates is an example of an observational study that requires careful interpretation.  In this case (A) does not cause (B).  Instead (A) and (B) are caused by factor (C): women who do agree to having mammography are different e.g. they have more conscientious personalities and less depression so they look after themselves better and thus have less heart attacks.[3] This is called the “healthy volunteer effect”.

In most (but not all) observational studies it has been observed that women who take HRT have a lower rate of cardiovascular disease.  One possible explanation for that association is that HRT causes benefit ie (A) causes (B).  An alternative explanation is that the association due to confounding by the healthy volunteer effect i.e. healthy women are less likely to have cardiovascular disease and also are more likely to take HRT.[4] 

Another possible explanation is the “healthy survivor effect”.  Imagine a therapy that increases death rates during the first year but then makes no difference.  That pattern is likely to be detected by long term RCTs but long term observational studies are likely to falsely suggest benefit.  This is because subjects are usually not recruited into long term observational studies until after they have been taking the therapy for a year or so.  Consequently such studies may miss the initial harm.  Also subjects taking the therapy will appear to live longer than those who do not.  This is because only lower risk survivors will be recruited into the study because higher risk people will have died early (because of the therapy) before they could be recruited into the study.

The disadvantage of RCTs is that the subjects involved may be different, at higher or lower risk, than the patients involved in everyday clinical practice.  Consequently there is a need for care when generalising the results of RCTs to your patients. 

We can be more confident about the results of observational studies and RCTs if they are all consistent with each other.

Clinically important events are more reliable than surrogate events.

Studies of clinically important outcomes (eg fractures) are more reliable than studies of surrogates (eg bone density).  This is because surrogates are not always as accurate at predicting clinically important outcomes as hoped. 

HRT increases bone density but, as yet, there is no good RCT evidence of reductions in rates of important fractures.  It seems likely that HRT will be beneficial, given that studies of other agents such as bisphosphonates show that they increase bone density and modestly reduce rates of fractures.  However, fractures may result from a combination of many causes.  Bone density is just one factor along with balance, environment (eg slippery bathrooms), sedating medications, etc.  Consequently we do not know if the benefit from HRT for fracture rates is large or small.

All good quality RCT results available should be considered to reduce bias.

One form of bias arises from basing conclusions on only those trials that the author likes for whatever reason.  The safest way to reduce the risk of that type of bias is to consider all available relevant information from RCTs of reasonable quality that measured the outcome of interest.  Consequently, we have included the information available from the Womens’ Health Initiative trial despite the fact that it is very incomplete.[5]

Controversy about the HERS trial

The HERS trial studied postmenopausal US women with established coronary artery disease aged 55 to 80.[6]  The average age was 67.  1383 women took placebo and 1380 took conjugated equine oestrogen and medroxyprogestogen acetate.  The average duration of participation in the trial was 4.1 years.

It has been suggested that the results of the HERS trial can be dismissed because of various criticisms.

One criticism is that the results may not apply to women who do not have established coronary artery disease.  This objection is reasonable.  However the results of all the RCTs of women without coronary artery disease are consistent with HERS.[7]^,[8],[9]  Consequently, the possibility that the impact will be similar in both groups needs be considered seriously.  This case study addresses that criticism by focusing on Ann who has established coronary artery disease and would have met the entry criteria for the HERS trial so the results of that trial are relevant for her. 

Another criticism is that the trial terminated prematurely.  This is not true.  The trial ended on the scheduled date.  The investigators considered extending the trial because the average duration in the trial was lower than hoped for because of late recruitment.  However it was decided not to extend the trial because of uncertainty about gaining informed consent from the participants and because statistical calculations suggested that little would be gained by extending the trial.

Yet another criticism arises from the fact that women used other cardioprotective drugs including “statins” more often in the control group.  This occurred because HRT lowered lipid levels in the treatment group so “statins” were prescribed less often for women in that group.  However “statins” could not have accounted for the differences in cardiovascular events observed during the first year of the HERS because other trials of “statins” have shown that they do not make that much difference to cardiovascular events in the first year.

It has been suggested that there may have been unknown differences between the groups at randomisation.  However there were no significant differences in any of the many baseline parameters that were tested.

It has been suggested all the endpoints other than the primary endpoint (coronary death plus non-fatal infarct) can be dismissed.  It is true that if the treatment really makes no difference and if many endpoints are measured then and if the criteria for significance is set at P<0.05 then on average the findings for 5% of endpoints will be false positives.  This applies equally to all endpoints including the primary endpoint.  The best response to this is to accept uncertainty about the results of any trial in proportion to the P-value until those results are verified by at least one other trial.  If the P-value is very low then a lower level of uncertainty is needed.  Healthy flexible scepticism is appropriate.  Completely dismissing results for secondary endpoints is not appropriate.

Acknowledgements

We thank Alistair MacLennan, Amanda Newman, Angela Wai, Barbara Mintzes, Elina Hemminki, Joel Lexchin, Judith Mackson, Katrina Allen, Laureen Lawlor-Smith, Peter Lake and Sharon Sanders for their input.  However we accept responsibility for any deficiencies in this report.

Next: Fractures