Blog by: Peter Tennant and Tomasina Stacey
For most people, ‘taking a risk’ means doing something that might backfire. And if something is ‘high risk’ then you’ve taken a major gamble with a high chance of losing and/or losing big. So, if your doctor or midwife tells you that you’re at ‘high risk’, it seems fair to assume the worst and plan accordingly. Except you’d probably be wrong. Because what they meant – or thought they meant – would likely differ from most people’s idea of ‘high risk’.
This is because ‘risk’ can mean many things to many different people. So much that it’s
entirely possible (indeed common) for someone to be called ‘high risk’, yet still have a very low chance of anything bad happening! To understand this, we need to consider what a statistician means by ‘risk’. To a statistician, ‘risk’ (or more formally ‘absolute risk’) simply means the ‘chance’, ‘probability’, or ‘likelihood’ of something happening. To a statistician, ‘risk’ is neutral term, it doesn’t imply high or low. And it doesn’t imply a good or bad outcome.
In fact, understanding whether a ‘risk’ is high or low, or good or bad, is meaningless without
context (e.g. what we want to avoid) and comparison (i.e. ‘high risk’ compared with who or
what?). Imagine, for example, we said the ‘risk of rain on Sunday was 1%’. That would be good news if you were planning a BBQ. But pretty bad news if you lived near a forest that was currently on fire. Similarly, while one couple having ‘unprotected sex’ might be very worried about a 5% ‘risk’ of pregnancy, another ‘trying for a baby’ might consider the same ‘risk’ frustratingly low.
Absolute risks, relative risks, and risk differences
But surely it gets easier if we consider the risk of something universally bad happening, like stillbirth? Now we can all agree that any ‘risk‘ is bad, right? Well… Only in theory. Because the reality is that there is almost always some risk. So, what we describe as high or low or good or bad can only really be described in reference to something else. To do this, statisticians use both the ‘risk difference’ and the ‘relative risk’, which describes the extra risk (compared to a chosen reference group) in absolute terms or relative terms respectively.
To give an example, we’ll consider the risk of stillbirth in women with body mass index (BMI)
values above 30kg/m2 (commonly considered ‘obese’) compared with women with BMI values between 18.5kg/m2 and 25kg/m2 (commonly considered ‘healthy’). On average, women with ‘obese’ BMIs have a higher risk of stillbirth than women with ‘healthy’ BMIs.
Indeed, in the UK, the risk of stillbirth in women with obese BMIs is around 7.2 in 1000 (or 1 in 139) and the risk in women with healthy BMIs is around 3.35 in 1000 (or 1 in 299) [1]. So, if you go by ‘relative risk’, you’d say the risk of stillbirth is over twice as high in women with obese BMIs than women with healthy BMIs, which sounds quite dramatic. But if you go by ‘risk difference’, you’d say women with obese BMIs have a 0.4% higher risk than women with health BMIs, which sounds quite low.
Risk and the complex individual
Things get murkier the further you move from the sterile statistical space to the complex climate of the clinic. Because all these concepts – absolute risk, relative risk, risk difference – are statistical ideas. And statistics is about understanding and making comparisons
between groups of people, not individual people. So, while statistics is invaluable for predicting and planning services, and recognising better and worse practice overall, the relevance of a particular statistic for an individual doctor, midwife, family, or woman facing an individual-level decision is far more questionable.
Let’s return to the example of obesity, but this time to consider a specific woman with an obese BMI. Is she at ‘high risk’ of stillbirth compared to the general population? Well, the statistics reported above suggest she has twice the risk compared to an otherwise identical version of herself with a ‘healthy’ BMI. But BMI as a measure, isn’t actually meant to be used to estimate individual-level risks. And you’d need to know a lot more about her to have a remotely decent estimate of her individual-level ‘risk’. What’s her age? What’s her ethnicity? How much does she drink? How much does she smoke? Does she live with the father? Does he smoke? How old is he? And on and on it would go. And even if you knew everything there was to know and were able to calculate confidently that, for instance, she had four times the risk of stillbirth to an otherwise ‘healthy’ women, she’d still be 98+% likely to avoid a stillbirth.
Towards better terminology
All of which raises the question; is ‘risk’ the right word? And is it right to describe someone as
‘high risk’ if they’re still most likely to have a healthy and positive experience? We think there are serious conflicts between the statistical and common understandings of ‘risk’ that might be
reduced by using any of ‘chance’, ‘probability’, or ‘likelihood’. And headline-grabbing – but highly misleading – ‘relative risks’ need discarding entirely.
References
1 Statistics about stillbirth. Tommy’s, 2019. Available at: https://www.tommys.org/our-
organisation/charity-research/pregnancy-statistics/stillbirth
2 Chu et al. Maternal obesity and risk of stillbirth: a meta-analysis. Am J Obstet
Gynecol.2007;197(3):223-8.
3 Adult obesity, overweight and waist circumference, by region and sex. Health Survey for
England, 2017. Available at: https://digital.nhs.uk/data-and-
information/publications/statistical/health-survey-for-england/2017
Footnotes
[1] Estimated from a combination of the reported risk of stillbirth in 2017 (reference 1) together
with the relative risk of obesity (reference 2) and the distribution of BMI in adult women of
childbearing age (reference 3). This is an overestimate of the ‘risk’ as the baseline actually
includes ‘all women’, ie including those with a raised BMI- so the individual chance of stillbirth
for
Commentaires