The most reliable source of information about the effects of treatments comes from appropriately powered randomised clinical trials (RCTs). RCTs measure the effects of randomly assigned interventions and the randomisation is the method by which bias and confounding are minimised. Communicating the results of clinical trials to professional groups, for example through peer reviewed publication, imposes a responsibility upon researchers regarding the accurate conveyance of the benefits and risks of treatments. With this responsibility in mind recommendations have been made to improve the quality of reporting randomised clinical trials [consolidated standards of reporting trials (CONSORT)] (Altman et al, 2001), and also observational studies [The strengthening the reporting of observational studies in epidemiology (STROBE)] (von Elm et al, 2007). Some journals have adopted editorial policies requiring submitted manuscripts to comply with these recommendations.
The interpretation of reported results with reference to trial design is complex and controversy may arise amongst professionals due to differences in interpretation and understanding, an example being the prevention of venous thrombosis by low dose heparin in relation to major orthopaedic surgery (Baglin, 2008). When potential benefits and risks are communicated to patients they may have difficulty understanding what might happen to them depending on the treatment choice they make and they may not regard population statistics as personally applicable. A patient’s interpretation of information may be influenced by education, previous experience, social class or race and so the communication of benefit and risk must be with reference to the individual. The traditional ‘paternalistic’ approach to clinical decision making, with the clinician being the decision-maker, is no longer considered appropriate in many societies and the modern approach is shared and informed decision making. However, one of the major challenges in clinical practice is applying evidence-based information to decisions for individual patients (Thomson et al, 2005). Patients’ preferences vary widely and clinicians may be poor at predicting these in individual patients. Doctors and patients show considerable variability when weighing up potential outcomes. For example, in a study utilising probability trade-offs, the number of strokes that need to be prevented in 100 patients over 2 years for anticoagulant treatment with warfarin to be justified was significantly lower for patients than doctors and the number of excess bleeds acceptable in 100 patients over 2 years was higher for patients than doctors (Devereaux et al, 2001). Clearly patients were more concerned with avoidance of stroke than risk of bleeding.
There are a variety of methods to present the likelihood of benefit and risk and similar considerations apply to the communication of information between health care professionals and between doctors and patients. A measure called the ‘number needed to treat’ (NNT) has been used for two decades to present the benefit of an intervention. The ‘number needed to harm’ (NNH) was subsequently introduced and used in parallel to present the risk of an intervention. In this issue of the bjh Jane Hutton explains the limitations of the NNT method and recommends ‘absolute risk reduction’ as a preferable measure (Hutton, 2009). As well as statistical concerns with reporting NNTs, for example in relation to measuring the certainty of an effect, there are concerns associated with NNTs as a means of communicating risk in an informative way. Hutton’s arguments are relevant to reporting the results of clinical trials, but also to how doctors explain benefits and risks to patients. The absolute risk reduction can be confusing to doctors and patients (Elwyn et al, 2004) but Hutton (2009) qualifies her recommendation by illustrating how the absolute risk reduction should be presented with reference to the absolute baseline risk of the disease in order for it to be meaningful. Information on absolute risk is crucial to patients involved in shared decision making and the issue of absolute risk is particularly important when explaining a risk that is very low, a common example in practice being the risk of venous thrombosis attributable to exposure to an oestrogen-containing combined oral contraceptive (COC).
A relative risk does not indicate how likely it is than an outcome will occur, only how much relatively more or less likely it is. It is the absolute risk that indicates how likely an event will be. For example the relative risk of venous thrombosis is increased twofold to sixfold in a female aged between 20 and 40 years old who takes an oestrogen-containing COC, depending on her age and the COC preparation. It is accepted that COC exposure increases the risk of venous thrombosis and some women presented with this information might choose not to use a COC as contraception, despite its proven efficacy and attractions. When informed that a risk would be 35-fold higher than the population baseline risk if the woman had the F5 R506Q (factor V Leiden) mutation and used a COC, some women would again choose to avoid use of a COC. However, in order to make an informed decision it is necessary for the woman to know how likely it is she would suffer venous thrombosis if she did not use the COC and how much the absolute risk would change if she did use a COC. The population baseline risk of thrombosis is about 1/10 000 per year under the age of 40 years. If a woman used a COC it would increase about fourfold (best guess) to about 4/10 000 per year. Clearly the risk of venous thrombosis associated with COC use is very low. For a woman with the F5 R506Q mutation her baseline risk of thrombosis would be about 1/2000 per year under the age of 40 years, a risk that is increased about sixfold (best guess) compared to a woman without this mutation. If this woman were to take the COC, her absolute risk would increase from 1/2000 to 1 in 300, a risk that is increased about sevenfold for her but that is 35-fold greater than the population baseline risk. For many women an annual risk of 1 in 300 is acceptable given the proven efficacy, ease of use and other obvious advantages of a COC.
A range of terms is used for communicating risks; low/moderate/high; likely/unlikely; common/uncommon. Interpretation of these terms varies and whilst standardisation of terminology in relation to risk has been advocated, for example low risk is when risk is <1 in 100, (Calman & Royston, 1997) the presentation of the actual data is likely to be more meaningful than the terminology. When presenting absolute risks a meaningful denominator should be used. For example, a risk of 4/10 000 might be more easily understood as 1/2500 when presented as: ‘during one year only 1/2500 women aged 20–40 years would suffer venous thrombosis or, alternatively, if 2500 women used a COC for 1 year one would suffer an event’. This information is sometimes presented as a percentage. However, if percentages are used to convey risk it is useful to cheque that the patient has interpreted the risk correctly by asking them to explain the risk in a different way, as many patients (and some clinicians) are confused by percentages. It is probably safer to communicate risk as a probability rather than a percentage. In the example of the woman with the F5 R506Q mutation who was considering using a COC the annual risk of venous thrombosis would be 0·3%, or 1 in 300, if she used the COC.
Presentation of risk not only requires a numerator and denominator but also a given time frame that is meaningful. For example, patients with atrial fibrillation considered a 5-year risk-period of stroke relevant to their ages and expectations (Thomson et al, 2002). In the context of a young woman who might anticipate taking a COC, information on risk might influence how long she elected to use a COC rather than whether or not to use it at all. For example, in the case of the woman with the F5 R506Q mutation who was considering using a COC she might decide to use it until she was 25 years old rather than 30 years old, as this might be the time it would be most beneficial to her. After the first year of use the risk would be lower (Bloemenkamp et al, 2000) and the increase in risk associated with age thereafter would be minimal up to the age of 25 years.
In addition to relative and absolute rates, several related measurements are used to communicate benefit and risk. The absolute risk (or rate) difference (benefit or risk) is the difference in incidence with and without treatment (or exposure). In the COC example, the absolute risk difference due to the COC is 3/10 000 (from 1 in 10 000 to 4 in 10 000). Assuming there is a causal relationship between the COC exposure and venous thrombosis this can be referred to as the attributable risk, i.e. three cases per 10 000 person years of COC use. The attributable fraction indicates the proportion of disease that is due to an exposure. The attributable fraction is calculated from the attributable risk difference divided by the incidence in the exposed (or treated patients). In the COC example, the attributable fraction is 0·75, calculated from (4–1)/4. Therefore, in women who use a COC, three-quarters of episodes of venous thrombosis are attributable to the COC exposure. A confusing feature of attributable fractions is that they can add up to more than 100% when there are multiple interacting causes of a disease. For example, in the case of women with the F5 R506Q mutation who develop venous thrombosis whilst using a COC, the attributable fraction of disease due to the genetic mutation is 80% (risk increase from 1/10 000 to 5/10 000) and that due to the COC is 86% (risk increase from 5/10 000 to 35/10 000).
The NNT is the average number of persons who need to be treated over a defined time period to prevent one outcome. Similarly the NNH is the average number required to result in one adverse outcome. The NNT and NNH are calculated as the inverse of the absolute rate differences [either benefit (NNT) or risk (NNH)]. Hutton (2009) has used the meta-analysis of randomised trials of extended-duration of prophylaxis for venous thrombosis after hip or knee replacement as a working example and illustrated the statistical limitations, particularly when the measurement is applied to meta-analysis. The NNT is also dependent on the absolute risk of the disease. For example if a treatment reduces the incidence of disease by 50% in two examples where the incidence of disease is 2/1000 or 20/1000, the NNTs will be 1000 and 100, respectively. The NNT is also determined by the duration of treatment. A longer duration will result in a lower NNT. However, the cost of treatment will increase with a longer duration and the NNH will increase. Therefore, neither the NNT or absolute risk reduction should be presented without reference to the absolute baseline risk of the disease or the proposed duration of treatment.
A large proportion of genetic counselling focuses on presenting risk information. Making sense of information can be difficult for patients (Saukko et al, 2007). Whenever a patient is offered an intervention there is a trade-off between benefit and risk. The benefit and risk information must be presented in ways that are relevant and understandable. In addition to the quantitative information the qualitative aspect of risk is equally important. For example, in the analysis of benefit and risk of long-term (lifelong) anticoagulation after a first episode of venous thrombosis, a recurrence might range from a fatal pulmonary embolism to a small clot in the calf whilst a major bleed may range from a gastrointestinal haemorrhage with full recovery to a central nervous system bleed with permanent neurological damage or death (Baglin, 2007; Kearon, 2007). A clinician’s perspective on the qualitative aspect of benefits and risks will inevitably be different to a patient’s and therefore the territory of ‘paternalism’ is again faced. Objective advances in resolving the ‘gap’ between clinicians and patients in this regard are being made, for example through probability trade-off scenarios (Protheroe et al, 2000; Locadia et al, 2004), but, even with information derived from groups of patients, the application of the information in the setting of an individual patient will remain a challenge (Smeeth, 2000). It remains to be determined if trade-off scenarios can be used with individual patients in the routine setting of a consultation for the patient to determine their ‘best option’ rather than accept the clinician’s ‘best guess’.