METHODS

Participants and setting

120 adults attending cardiac rehabilitation clinics at two Leeds hospitals following a recent admission for coronary artery bypass surgery or myocardial infarction who were taking either simvastatin or atorvastatin were recruited to the study. Those who could not read or whose first language was not English were excluded. Forty five patients attending the clinic who were eligible for the study declined to participate.

Design and procedure

A randomised controlled method was used with two independent between subject variables: format of information (numerical or verbal) and side effect (constipation or pancreatitis). The allocation order was randomly predetermined and the questionnaires stored in order before use. The allocation order was not concealed but only after participant recruitment did the researcher take the next questionnaire from the pack.

After giving their written consent to take part in a study of “different ways of giving patients information about their medicines”, participants were asked to read a short piece of information about a side effect associated with the medicine they were taking (box 2). Half the participants received information about constipation, a frequently occurring mild side effect (both medicines); the rate of 2.5% was taken from the 4S trial (simvastatin)¹³ and a systematic review of trials of atorvastatin.¹⁴ The other half were given information about pancreatitis, a less frequent and more severe side effect (those taking atorvastatin only) with an incident rate of 0.04%.¹⁴ Within each side effect, half the participants received the information in a numerical form (“This side effect occurs in 0.04% (that is, 4 out of 10 000) people who take this medicine”) and half in a verbal form (“This is a rare side effect of the medicine”).

After reading the risk information, participants completed the answer sheet. They could refer back to the risk information when answering the questions. Participants were asked to give a percentage probability of the likelihood of having the side effect. They were also asked to respond to five questions on Likert scales (range 1–6) about the overall likelihood of their experiencing a side effect, their perception of the risk to their health, the severity of the side effect, their satisfaction with the information and, whether the information would affect their decision to take the medicine. The full information given and the questions are shown in box 2.

Analysis of data

The main outcome measure was the estimate of the likelihood of the side effect occurring. Secondary measures related to Likert scale responses to the five additional questions. The data were analysed using independent t tests after ensuring approximately normal distributions. The sample size of 30 in each of the four groups (verbal/numerical and constipation/pancreatitis) was calculated using data from our previous scenario based studies. The sample size had 90% power to detect a difference at the 5% significance level in the estimated probability of side effect occurrence. The study was approved by the local NHS research ethics committee.

RESULTS

One hundred and twenty patients were recruited (76 men) of median age 63 years (range 35–74) who had been taking a statin for a median of 6 months (range 1–70). They came from a range of educational backgrounds although the majority (n = 67, 56%) had no formal educational qualifications. The patients recruited to each of the four groups were comparable with respect to demographic data (table 1).

Pancreatitis (“rare” versus 0.04%)

The mean estimate of the likelihood of having pancreatitis was 18.0% in the verbal group and 2.1% in the numerical group (p<0.001, table 2). Participants in the verbal group also thought they were more likely to experience a side effect (p = 0.006) and thought that the side effect would pose a greater risk to their health (p = 0.002). The two groups showed no statistically significant difference in their rating of the severity of the side effect (p = 0.20). Participants in the verbal group were less satisfied with the information provided (p = 0.048), but there was no difference between the groups in their rating of whether the information would affect their decision to take the medicine (p = 0.16).

Constipation (“common” versus 2.5%)

The mean estimate of the percentage of people who would experience constipation was 34.2% in the verbal group and 8.1% in the numerical group (p<0.001, table 3). Participants in the verbal group also thought they would be more likely to experience a side effect (p<0.001) and that the constipation would pose a greater risk to their health (p = 0.041). The two groups showed no statistically significant difference in their rating of its severity (p = 0.39). Participants in the verbal group were more likely to say that the information would affect their decision to take the medicine (p = 0.037), but there was no difference between the groups in their satisfaction with the information provided (p = 0.06).

DISCUSSION

The principal findings are that participants given verbal terms to describe the likelihood of side effects of medicines they were taking thought those effects would be far more frequent than people given the numerical percentage equivalents. In the case of the term “common” (suggested by the EU for incident rates of 1–10%), the probability of the side effect occurring was said to be 34%—more than three times the upper end of the designated frequency band. The term “rare” (intended for incidences of 0.01–0.1%) was interpreted as having a probability of almost 20%—18 times the upper end of the frequency band.

This is the first study to examine this effect in the clinical setting—that is, ascribing these verbal terms to side effects for medicines which the participants were taking. Our previous scenario-based studies found mean estimates in the range 45–56.6% for “common” and 6.3–21.5% for “rare”.¹⁰ Thus, the patients taking real medicines followed the same pattern with a greater degree of overestimation for the less frequent (and much more serious) side effect. This lends further weight to the conclusion that use of the verbal terms in the EU guideline leads to a gross overestimation of the probability of harm. Our results are in agreement with earlier Dutch and German studies which showed that verbal terms are not always interpreted in the way intended.^15,16

Participants who received the verbal descriptors of risk thought they would be more likely to suffer side effects themselves, with ratings some 40–60% higher than in the numerical group. The verbal descriptors also resulted in greater perceived risk to health from the medicine, with ratings 40% higher than in the numerical group. Verbal descriptors led to participants being generally less satisfied with the information provided (one statistically significant difference and one borderline difference).

The provision of side effect risk information using the verbal descriptors also showed potential effects on the decision to take the medicine. This may be the most important aspect, since it indicates a possible change in the patient’s behaviour in medicine taking rather than simply a difference in perception about the safety of the medicine. Patients given the verbal information about the constipation side effect were significantly more likely to say that it would affect their decision to take the medicine. Patients given similar information about pancreatitis showed a similar trend but this did not reach statistical significance.

One limitation of the study is that the patients had been taking the medicines for variable periods, in some cases for many months. They are likely to have based their responses on personal experience as well as on the information provided. Testing the verbal descriptors with people about to take a medicine for the first time would give further insight into the effects on perception of risk and willingness to take the medicine. It is also likely that those attending cardiac rehabilitation clinics are not typical of the population of patients taking these medicines.

Another limitation may be the method of group allocation as the researcher was potentially aware of the next allocation at the time of recruitment. A lack of allocation concealment can be associated with exaggerated group differences, but it is most likely to occur when the researcher can anticipate how the prospective participant will respond to any allocated intervention. There is only a limited possibility that lack of concealment introduced bias here, as the researcher is unlikely to have been able to anticipate the patient’s response to verbal or numerical information about a side effect. Furthermore, the comparability of the groups is confirmed in table 1.

Some participants asked to estimate the frequency of the side effect pancreatitis may not have understood the term, although no participant asked what it was. The estimated severity of pancreatitis (table 2) was much greater than the estimated severity of constipation (table 3), suggesting that most participants had some understanding of it. However, it would be useful for further research to explore participants’ understanding of any side effects being researched.

In addition, asking participants to make an estimate of low frequency incident rates (such as the rate of pancreatitis in this study) by using a percentage may predispose them to making estimates of at least 1%. This would have the effect of producing larger mean estimates. We have examined this effect in previous studies¹⁷ and found that telling participants that they can make estimates of less than 1% had no significant effect. Furthermore, any measurement weakness in the study that might influence participants’ estimates applies equally to the verbal and numerical groups.

It is clear that, although we have found deficiencies with verbal descriptors, numerical information also has problems with some people finding it difficult to discuss risk in terms of numbers.^18,19 Indeed, verbal descriptors have some advantages²⁰ including the ability to divide a long list of side effects into manageable sections (based on frequency of occurrence). Also, there is often uncertainty about the precise level of risk for each side effect, as incident rates tend to vary between trials and many trials are too small to calculate low frequency incident rates. Verbal terms can allow for such variation and uncertainty. One conclusion is that we should examine which descriptors are most stable and determine the approximate bands of risk associated with those descriptors by the public. This and the previous scenario-based work would suggest that the term “very common” might be used for side effects occurring in over 50% of the population and “common” for 10–50%. However, much more research is needed before such recommendations could be implemented.

Calman proposed different verbal labels (including “high”, “moderate” and “low”) to describe the risk of side effects of medicines,⁷ and these terms also need testing in a clinical setting. Clearly, a significant problem in using any standardised scale applied to all patients is the variability in people’s estimates of what a verbal descriptor (such as “common”) means. The findings of this study, in line with others we have conducted with hypothetical scenarios, have revealed not only that the verbal terms are associated with overestimation, but also the high degrees of variation in estimates. Similarly, other studies have shown that estimates vary according to the context in which the term is interpreted.²¹

Progress might be possible through using appropriate combinations of verbal and numerical information or using graphical presentation methods.^20,22 Such flexibility might be needed to meet the variety of individual preferences or need.²³ However, the large number of different side effects listed for most medicines makes this problematic—for example, 17 for simvastatin and 26 for atorvastatin. This is particularly relevant when space is limited, such as on a medicine leaflet. In addition, presenting each side effect with a percentage frequency will result in a long list of percentages (some to three or four decimal places). This is likely to deter many patients from reading the information. The use of the terminology “x in 100 000” is another option. However, we found that the wordings recommended by the Committee on Safety of Medicines on the risk of thromboembolism with oral contraceptives (which used this terminology) were fully understood by less than one in eight women in higher education.²⁴ Another possibility is to use scales based on community risk—for example, one person in your street, town, etc^25,26—but again these would need testing in a clinical setting.

We conclude that, regardless of the impact on behaviour, the basic requirement that patients have a true understanding of the level of risk of side effects is far from being met. An accurate understanding of the likelihood of benefit and harm is fundamental to the concept of partnership in medicine taking, but we are some way from a standardised language of risk.²⁵ The problems associated with communicating information about risk and uncertainty related to complex medical information are significant. A lack of shared understanding of risk frequencies between patient and practitioner (whatever method of information provision is used) reduces the possibility of informed patient participation in decisions about treatments. Furthermore, it reduces the quality of information that practitioners provide, since information that is misunderstood is of little use to patients. Indeed, misunderstanding information (by exaggerating the risk of a negative outcome, for example) might cause a patient to decide not to take a medicine which they might otherwise have taken if the information had been more usable. Until there is further research into people’s interpretation of methods of describing the risk of side effects, we suggest that the EU amends this part of the guideline.