Title: Explaining risk and study design

Article: » print this

Explaining risk

Margaret Rees
Heather Currie
20 April 2007

Publication of the results of the Women’s Health Initiative (WHI) and Million Women Study (MWS) since 2002 have led to considerable uncertainties about the role of hormone replacement therapy (HRT) among health professionals and women. This fact sheet will discuss how to explain risk and will describe different types of study design.

Explaining risk
Relative risk, absolute risk, attributable risk
Patients, the media and health professionals often confuse the terms “relative risk”, “absolute risk” and “attributable risk”. An understanding of the precise definitions of these terms is important in order to judge the actual magnitude of risks involved. Epidemiological data that examine the risk of a disease often report “relative risk” statistics to determine statistical significance. In other words, they report the risks run by a patient on a treatment relative to the risks run by a patient taking no treatment. This method provides substantial statistical power to detect effects of these agents, which might be quite small in magnitude. However it does not take into account the actual frequency of the condition in the untreated group. For example, a relative risk of 2 – often reported in the press as “a doubling of risk” – could describe something that increases the risk of a disease from one in a million to two in a million or something that increases the risk of a disease from four people in 10 to eight people in 10. Thus “absolute risk” and “attributable risk” which take into account the frequency of the condition are better methods of presenting the data.
Relative risk is defined as the rate of disease among the treated group divided by the rate of the disease among the untreated group. 
Absolute risk is determined by multiplying the usual rate of the condition in the untreated group by the relative risk.
Attributable risk is an absolute measure of the excess risk attributed to treatment.  It is calculated as the difference in risk of a particular condition between those who are treated and those who are not.

Hazard ratio and odds ratio
Risk also can be expressed as hazard and odds ratios. Hazard ratios (HR) are broadly equivalent to relative risk and are useful when the risk is not constant with respect to time. They use information collected at different times and were used in the publications from the WHI studies. The term typically is used in the context of survival over time. If the HR is 0.5, the relative risk of dying in one group is half the risk of dying in the other group.

The odds ratio is the odds of an event happening in the treated group expressed as a proportion of the odds of an event happening in the untreated group. When events are rare, the OR is analogous to the relative risk (RR), but as event rates increase, the OR and RR diverge.

Designs of clinical trials
Primary and secondary prevention trials
Primary prevention trials are those undertaken in healthy people, whereas secondary prevention trials involve participants with established medical conditions, such as cardiovascular disease and osteoporosis. It is important to distinguish between these two types of trials, as the responses may differ. For example, in women with cardiovascular disease, an endothelium with established atherosclerotic plaques is likely to respond differently from one without such plaques.

Randomized trials and cohort and case-control studies
Randomized trials
Randomized controlled trials (RCTs) are considered the best method for providing evidence on efficacy of an intervention. A perfectly randomized method to allocate participants to the study groups does not, however, protect an RCT from selection bias. Participants may not be typical of the whole population. In addition, differences may still exist between the groups studied – for example, in the number of smokers or types of drugs taken – and these can influence outcome. Randomized controlled trials, even if perfectly designed, can tell us which treatment is better, but they cannot tell us for whom it is better. How and whether to generalize the results of a single trial to an individual patient is one of the most complex issues in health care. The findings that relate to a single drug treatment in a particular population thus cannot necessarily be extrapolated to all people. For example, the results of the WHI studies of HRT undertaken in women older than 50 years cannot be applied to women with a premature menopause.

In addition, RCTs tend to be short-term trials, and some interventions may take several years to have an effect. Also the behaviour of volunteer patients in trials may differ from those treated among the general public. Some of these problems can be overcome by cohort studies.

Cohort studies
Cohort studies can be thought of as natural experiments in which outcomes are measured in the real world rather than in experimental settings. They can evaluate large groups of diverse people, follow them for long periods and provide information on a range of outcomes, including rare adverse events. The promise of cohort studies as a useful source of evidence needs to be balanced, however, against concerns about the validity of that evidence. Cohort studies are similar to randomized controlled trials in that they compare outcomes in groups that did and did not receive an intervention. The main difference is that allocation of individuals is not by chance, so cohort studies are vulnerable to selection bias, which leads to criticism of the “healthy user” effect, in which people with innately lower risks of an endpoint (such as myocardial infarction) as a result of good dietary and exercise habits may preferentially take the intervention under study – which then falsely gains credit for a favourable outcome. In cohort studies, therefore, factors that determine whether a person receives the intervention could result in the groups differing in factors related to the outcome – because people are selected preferentially to receive one treatment or because of choices they made. These baseline differences in prognosis could confound assessment of the effect of the intervention.

Case–control studies
In a case-control study, patients with a particular disease or condition (cases) are identified and “matched” with controls (patients with some other disease, the general population, neighbours or relatives). Data then are collected (for example, by examining these people’s medical records or asking them to recall their own history) about past exposure to a possible causal agent for the disease. Just as in cohort studies, case–control studies generally are concerned with the aetiology of a disease rather than its treatment. They lie lower down the hierarchy of evidence, but this design usually is the only  practical option when studying rare conditions.

All observational studies, be they case–control or cohort in design, cannot prove causality – only association.
 

Women’s Health Initiative and Million Women Studies
Women’s Health Initiative
The WHI is a large and complex clinical series of investigations of strategies for the prevention and control of some of the most common causes of morbidity and mortality among postmenopausal women, including cancer, cardiovascular disease and osteoporotic fractures. It was initiated in 1992, with a planned completion date of 2007. Postmenopausal women ranging in age from 50 to 79 years were enrolled at one of 40 clinical centres in the US into a clinical trial (CT) or observational study (OS). The CT was designed to allow randomized controlled evaluation of three distinct interventions.

• HRT (unopposed and combined) – hypothesized to reduce the risk of coronary heart disease (CHD) and other cardiovascular diseases and, secondarily, to reduce the risk of hip and other fractures, with increased risk of breast cancer being studied as a possible adverse outcome

• low-fat eating pattern – hypothesized to prevent breast cancer and colorectal cancer and, secondarily, to prevent CHD

• supplementation with calcium and vitamin D – hypothesized to prevent hip fractures and, secondarily, to prevent other fractures and colorectal cancer.

In the CT, the mean age of the women was 63 years.

The cohort of the OS comprised clinical trial screenees who were ineligible or unwilling to participate in the CT. The major clinical outcomes of interest in the OS are CHD, stroke, breast cancer, colorectal cancer, osteoporotic fractures, diabetes and total mortality.

Million Women Study
The MWS is an observational study that has evaluated the risk of breast cancer with respect to differences in HRT regimen and routes of administration (with the exception of vaginal preparations). Women invited to attend the NHS Breast Screening Programme (NHSBSP) were sent a self-administered questionnaire that asked them to document details about personal medical history and lifestyle factors, including the use of HRT. The study data were recorded from these questionnaires (which were returned before initial mammography), and women were followed to determine the incidence of cancer and death. A total of 1084,110 women were recruited between 1996 and 2001; about half had ever used HRT. The average duration of follow up was 2.6 years. Several publications have questioned its design, analysis and conclusions, and a selection of comments is given below.

• The study reported a lower risk of breast cancer for perimenopausal and postmenopausal women than premenopausal women, despite the well-established fact that the risk of breast cancer increases with age.
• Many differences were present when women who used and did not use HRT were compared, and this required multiple adjustments.
• Use or non-use of HRT was established only at study entry and changes were not recorded during follow up. Lack of prospective follow up of exposure to HRT means that uncertainty exists as to whether women switched preparations or started or stopped HRT during the study’s follow-up period.
• Validation of the questionnaire data was based on information obtained from only 570 women. Curiously, although the risk of endometrial cancer with unopposed oestrogen is well established, 14,024 non-hysterectomized women were documented as having taken HRT that contained oestrogen alone.
• Mortality from breast cancer was assessed after an average of 4.1 years of follow up and on the basis of a total of 517 deaths; however, breast cancer was diagnosed very rapidly, after a mean of 1.2 years, and deaths occurred swiftly (within an average of 1.7 years). This can be attributed to an underestimation of the total duration of exposure to HRT, as the risks presented were estimated on the basis of the use of HRT at recruitment. The study investigators did not adjust the total duration of use to account for the likely continued use of HRT in the period between recruitment and diagnosis of cancer (that is, a mean of 1.2 years). Current users and past users of HRT were compared with never users, and, although the risk of mortality was increased, it was not significant (RR  1.22, 95% confidence interval 1.00 to 1.48). This finding was highlighted because it differed from consistent reports in the literature over a decade that women who take hormones have better survival rates. The MWS calculated the risk of mortality by dividing deaths from breast cancer by the total number of women who used HRT or those who did not. If the risk is recalculated by dividing deaths from breast cancer by the total number of cases of breast cancer in women who did and did not use HRT, the results agree with the literature – the risk of mortality was reduced by about 27% in the women who took hormones.
• The higher estimates of risk reported compared with those from the randomized WHI study – especially the oestrogen alone arm, which found a reduced risk – probably reflects the observational nature of the MWS and suggests that the latter study probably grossly overestimated the risk of breast cancer.
• The data provided probably are representative of about 25% of all women aged 50–64 years in the UK (on the basis of uptake for the first prevalent NHSBSP round of 75%, completion of the MWS’ questionnaire by 50% of attendees and the number of screening centres in the UK that participated in the study (66/94)). Differences between women who did and did not attend the NHSBSP and between attendees who agreed – or declined – to participate in the study cannot easily be controlled for.

Further reading

Rees M, Purdie DW. Management of the Menopause. The Handbook. 4th ed. 2006 RSM press.

Whilst great care has been taken to ensure the accuracy of information contained in the fact sheets, the authors and the BMS cannot accept any responsibility for any errors omissions, mis-statements or mistakes or for any loss or damage arising from actions or decisions based on information contained in this publication. Ultimate responsibility for the treatment of patients and interpretation of published material lies with the medical practitioner. The opinions expressed are those of the authors, not necessarily those of the BMS. The inclusion in the publication of material relating to a particular product, method or technique does not amount to an endorsement of its value or quality, or of claims made by its manufacturer.
Margaret Rees and Sally Hope January 2008

« go back

Scientific papers, charts, presentations etc are available in the members area. » Join here