It is often said that humans are terrible at understanding risk. Maybe so when it comes to ratios and rates – abstract forms of data. But all of us navigate the world by moving, walking, or driving – constantly comparing distances, magnitudes, and likelihoods, making estimates about the risk of collisions, repeatedly, and accurately. We are actually risk calculators extraordinaire – and we do it all effortlessly, using heuristics, and get it right most of the time. Our survival depends on it.

But when it comes to health care, we are less confident. When to start a medication? When to have an operation or a test? Plus, we have much less experience to draw on. And less data. Often, no data at all.

Even doctors lack good data. To understand and explain risk you need a few basic bits of data – so here comes the jargon – reference class, baseline risk, relative risk, and icon arrays.

Best to start by being clear about reference class – being very clear about who and what we are talking about. Is it women or men and in what age group? With clear diagnosis or part of the general population?

Then mention the idea of baseline risk – the frequency of an event or a diagnosis at a population level. This can be difficult to find out but should always be our benchmark. Comparisons make no sense unless we know our starting position. Some people call this absolute risk.

Comparing two risks is where confusion starts, and manipulation can occur – this is called relative risk. Imagine a problem that happens in men in their 50s (reference class) at a frequency of 2 in 10,000 (baseline rate). Compared with no treatment, a new treatment is effective in 1 in 10,000 of these men. This is two times better, 100% better, than having no treatment (relative benefit). One-hundred percent better sounds very good until we think about the absolute level of benefit, i.e., it applies to 1 man in 10,000. This figure has a very different effect on our calculations of benefit. Most marketing data uses relative risk and never makes absolute benefit clear. And don’t even get me started on data about harm.

Figure 1: icon array (

When it comes to risk communication, pictures, though not perfect, help show what happens at population levels. An icon array (see Figure 1) elegantly combines the idea of reference class (100 men between 50 and 60). They have conditionitis and are asked to consider the benefits of treatment A (blue stick men) or treatment B (green stick men). Treatment A is twice as effective as treatment B (relative risk), but the absolute benefit is a more modest number, i.e., 5 more men in 100 (10 minus 5). But note that even then 85 men who do not benefit at all. And by the way, stick figures (bathroom symbols) are better it seems than smiley faces or ovals. It is the kind of graphic that should guide all conversations about decisions in health care. In the land of the blind, we could at least become partially sighted.

Author: Glyn Elwyn

Glyn Elwyn leads a research group, the preference laboratory, that produces measures of shared decision making, investigates user-centered design of patient support tools, and their integration into innovative healthcare delivery systems. His current focus is on Option Grid ™ decision aids for clinical encounters.

Read more