Background
Previous research has demonstrated that the perceived risk of side effects from taking a medicine can influence people's decision to take it. This is problematic given that most people overestimate their personal side effect risk, even when presented with numerical risk estimates1. The EU guidelines recommend presenting risk as one of 5 verbal descriptors to aid patient understanding (from 'very common' to 'very rare'). However, these descriptors lead to even greater inflation of perceived risk1. A possible solution is to present risk in terms of 'natural frequencies' (i.e. "x in 100 people get this side effect") that provide a concrete reference rather than the abstract concept of probability2. This study compared the effectiveness of presenting risks in natural frequency formats as opposed to percentages or verbal descriptors. The study used the web to collect data from people searching the Cancer Research UK's (CRUK) patient information site, www.cancerhelp.org.uk. This novel web-based methodology provided a useful way of recruiting actual users of CRUK's internet based information on chemotherapy.

Method
People searching the CancerHelp UK site were invited via a pop-up window to take part in the study. Participants were presented with a hypothetical scenario where they were to take the drug paclitaxel (Taxol). They were then randomly allocated to receive information about the risk of four potential side effects of Taxol in one of three different presentation formats. In the verbal format, two side effects were classed as "very common" and two as "common". In the other two conditions the corresponding percentages or frequencies were given (e.g. "50%" or "50 people in 100"). All participants were then asked "What do you think is the chance you will have [side effect] from taking this medicine?"

Results
148 (119 female/19 male) people took part in the study with an average age of 43 (SD=12.8). In all conditions participants significantly overestimated the risk of each side effect (p <.01). Participants in the verbal condition gave significantly higher estimates of side effect risk than those in the percentage or frequency conditions (p<.01). However, there was no difference in risk estimates between the percentage and frequency conditions. There was no significant difference in personal risk estimates between people who had experienced the side effect and those who had not.

Conclusions
In line with previous findings this study found that people consistently overestimate their personal risk of side effects from taking medicines, in this case a chemotherapy drug. Even when presented with natural frequencies participants inflated their chance of suffering an adverse event. The findings suggest that people maintain a high level of perceived personal risk in the face of concrete evidence to the contrary. Future research needs to explore the mechanisms by which people arrive at their estimate of side effect risk from medicine.