Relative Absolute Risk – the discussion
This is a follow up discussion on the recent post on risk – its two flavours being “relative” and “absolute”. This is a quick look at the basics of describing risk and how it can lead one astray!
Lets start by looking at how “relative risk” can be used to sell stuff – newspapers, health messages, whatever.
In February 2014 the Daily Mail in the UK published this report on “Deadly Risk of Pill used by 1 million women”
Sounds pretty scary – these 3rd generation OCPs can nearly double the risk of a woman developing a serious thromboembolic problem. In fact there is in Australia a class-action against the manufacturer of a few brands of these “new Gen pills”. Now lets look at the raw numbers and some data. There have been a number of trials and all with varying rigidity in terms of how they followed-up who got a DVT or PE. So the numbers here are ranges given by a Report from the FDA n the US in 2012. The numbers are the rates of VTE per 10,000 woman-years exposure:
- Normal women, no OCP, not-pregnant – between 1 and 5 clots
- Women taking “2nd gen, or conventional OCPs” – 3 – 9 clots
- Women taking drospirenone-based OCPs (3rd / 4th gen) – 10 – 20 clots
- Pregnant women – 5 – 20 clots
- Post-partum period (12 weeks) – 40 – 65 clots
If you round those numbers off a bit and say that the absolute risk difference between the “old”pill and the “new” pill is at worst 10 clots per 10,000 woman years. Which is 0.1% per year. Or…. a really, really small number. Not the sort of numbers that sells newspapers or makes for a good scare-story on the evening news. Hardly scary at all!
So what happens when the newspapers run stories like these? Well bad things happen. In the UK after a similar story a few years ago – many women stopped taking their OCP and relied on other less-efficacious means of contraception and more became pregnant, more had babies and more had surgical termination of pregnancy. So as you can see from the higher rates in pregnancy – there were more clots, more morbidity and mortality…. and why? Because a news editor opted for the drama of “relative risk” over the reality of “absolute risk”.
In the end we recently saw a subsequent prospective, controlled cohort study released in Contraception , April 2014. more than 200,000 woman-years of exposure was analysed. This paper showed no significant difference between the various generations of OCP for VTE or other serious adverse effects. This is unsurprising when you look at the small effects in absolute terms form previous studies – such a small effect can easily evaporate when more data is added.
Relative risk is a poor way to represent data. In your mind when somebody tells you of a relative risk – your instinct should be to ask: relative to what?! This is often not done.
In Australia (as in many places) GPs and other doctors are bombarded by drug companies with print, digital and even face-face promotion of their products. If you actually read the glossy brochures [I don’t recommend it] you will notice that the big print numbers heralding the efficacy of the product are invariably the “relative benefit” numbers. To find the comparator or the raw absolute numbers one must read the infinitely tiny print on the last page of the promo pages. As the man, Tom Waits sang in Step Right Up (1976): “The large print giveth, and the small print taketh away!”
Now lets go back to the last post on Relative Absolute Risk – where I gave you 3 cases of long term cardiovascular risk. The Lancet [August 2014] published an interesting meta-analysis on the effect of BP-reducing therapy on cardiovascular risk. They looked at individual patient data and divided the patients into 4 risk-strata at baseline.
They then looked at the groups and what the relative and absolute benefits were in terms of cardiovascular events – and here are the basic outcomes:
GROUP Baseline 5 yr risk Rel Risk reduction Abs event reduct
Low 6 18% 14
Medium 12.1 15% 20
High 17.7 13% 24
V. high 26.8 15% 38
The reason I love this paper is that it demonstrates really nicely the uselessness of relative risk in terms of real-world outcomes. All 4 groups had the same “relative risk reduction” from taking BP-lowering medications…. and yet there was a dramatic difference in the actual numbers of CV events prevented.
Absolute risk reduction is highly dependent on the initial risk conditions. If you are at low risk of outcome X to start with – then even a massive RRR will still leave you basically where you started. So when it comes to deciding on treatment for “Cardiovascular Risk” – we really need to look at Absolute risk and then target the patients who have a high starting risk – much more bang for your buck!
The Australian Heart Foundation clearly state that Absolute risk is the primary target of therapy – which should replace the individual BP, cholesterol and other targets as individual risk factors. But now the problem is communicating this to a patient! Most people can grasp a high BP reading as being “bad”, i.e. requiring a tablet, but find it tougher to take a pill for a “10% 5-year CV event risk”
Hence my bemusement when my 30 year old brother was put on both a BP-lowering agent and a statin – despite being 30, thin and having no other risk factors other than eating too much take-out food! Clearly the message is not getting through! According to the calculators his baseline risk is about 2% / 5 years. And yet taking 2 drugs with all the side-effects will reduce this to…. about 2%. Really he just needs to be banned from McDonald’s and all is well!
So, should we just simply stop using relative risk?
No. It does has it uses. I like relative risk when I am trying to sell (or scare) a patient into action. Is this naughty? Or is it a good use of a bad statistic?
For example smoking cessation. My favourite trick is to use actual data but present it in a scary manner. For men, there really seems to be a strong fear of rectal carcinoma. Not sure why – it just seems to be a highly disturbing form of cancer. The incidence of rectal cancer is around 12 per 100,000 people on average. Smokers have a higher rate – nearly twice that of “never smokers”. So instead of saying – “ou should Quit for your lungs”, I say: “you know smoking doubles your chance of rectal cancer”…. Leave that visual image hanging for a bit and then discuss smoking cessation strategies.
OK I will leave you with that.
– when given “relative risk” ask what the raw numbers are – work out the absolute risk before getting too much further.
– recall that baseline risk is the best predictor of a benefit from any intervention that works
– beware of any advertising / promotion that uses “relative risk” as a selling point
– feel free to use relative risk when you need to change behaviour – commercial media do this everyday, why can’t we do the same?
Relative risk and absolute risk as descriptions, both have their uses.
Absolute risk are (much) better when we are trying to describe the magnitude of an effect. As per your examples, if we are trying to understand the importance of an intervention, we need to know what it is actually likely to do. A nice concrete example: a drug that is likely to lower your blood pressure by “20%” isn’t meaningful, compared to “20 mmHg”. Implicit in absolute risk is context in which the description is applied. Relative risks can be misleading in this setting when we make incorrect assumptions about the context of the statistic – e.g., “doubling” of VTE risk sounds bad because most of us will intuitively inflate the actual (very low) baseline risk to risks we experience on a day-to-day level.
Relative risk is useful when we are talking about an effect, independent of a clinical context. For instance, your BP lowering data on CVD risk. We can say that the relative risk reduction is the effect of the DRUG, on the DISEASE OUTCOME. As long as we remember that this isn’t about individual patients, then it is a useful statistics. This can be helpful when we are making generalisations, rather than individualised advice. For instance, what is the effect on CVD risk of quitting smoking, compared to BP lower and lipid lowering (model: male, 65 yo, BP=160mmHg systolic, smoker, tot.chol=7, HDL=1)? We could say that smoking cessation would lead to a relative risk reduction of CVD of 33%, compared to 18% and 27% for BP lowering and lipid lowering respectively. Reducing all three would lead to a relative risk reduction of CVD of 66%. Described this way we can see that quitting smoking is the single most important risk factor that can be changed, by itself accounting for half the reducible risk.
That is a good tip esoecially to those fresh to the blogosphere.
Short but very accuurate information… Appreciate yoir sharinhg tnis one.
A must read article!