Coffee and cancer: what Starbucks might have argued

David Spiegelhalter
Published in
7 min readApr 2, 2018

This week a judge in California ruled that coffee outlets are going to have to tell customers their product could cause cancer due to the presence of acrylamide. The case was brought in 2010 against Starbucks and others by activists citing California’s Proposition 65, which requires that consumers are given warnings “before exposure to chemicals known to cause cancer or reproductive toxicity”.

My attempt at a dangerous coffee. I really need barista training.

Acrylamide is a compound formed when toast is burnt, roast potatoes are crisped, meat is barbecued, coffee is roasted, and so on. The chemical is listed by the International Agency for Research on Cancer (IARC) as a Class 2A ‘probable carcinogen’, due to the occurrence of tumours in rodents given extremely high doses, but when it comes to the levels consumed by humans in their normal diet, Cancer Research UK say that “At the moment, there is no strong evidence linking acrylamide and cancer.”. But California decided in 1990 that acrylamide should count as a carcinogen.

Starbucks might have avoided the need for warning labels if they could have shown that an average coffee consumer had ‘no significant risk’, which is taken to mean that an average consumer has less than 1 in 100,000 of developing cancer over a 70 year lifetime because of the acrylamide. They failed to prove this in the first phase of the trial, and so part of their defence in this latest phase was that coffee was beneficial, and so they should be allowed a less stringent risk level. They again failed to prove this.

This might be dismissed as just another example of Californian neuroticism, but it illustrates both a worrying attention to unimportant hazards rather than real risks, and the poor use of statistical arguments in legal judgements. So the argument is worth looking at in more detail.

What the judge found

The judgement was scathing of Starbuck’s defence, finding that their expert, Dr Rhomberg, ‘failed to undertake the type of quantitative risk assessment that is necessary to quantify the risk of cancer from exposure to acrylamide in coffee’. He also found that the defendants had not ‘quantitatively compared any alleged health benefits with any adverse effects of coffee consumption.’

Finally, the judge said that the defendants’ measurements of acrylamide in their coffee were ‘scientifically unreliable and inadmissible’ and so rejected all the following expert testimony from that company.

What I would have argued on behalf of Starbucks.

I have not seen the expert evidence and so don’t know what they said. But if I had been asked (which I wasn’t) and had I accepted (which I would not have), my argument would have gone as follows.

The judge required the defendants to prove that any increased cancer risk from coffee led to less than 1 extra cancer in 100,000 lifetime coffee drinkers. It is impossible to prove this empirically: we would expect around 50,000 of these coffee drinkers to get cancer in their lifetime anyway from other causes, and we have to be sure that this number is 50,000 rather than 50,001 (the extra 1 being the one that would make acrylamide a ‘significant risk’ in the view of the court). This should make it clear how difficult, to put it mildly, ‘proving’ this would be. If the acrylamide in coffee truly had no effect on cancer rates, then according to a standard online sample size calculator you would need a study of 85 billion people to prove it to the California judge’s satisfaction. That’s around 10 times the number of people currently alive on this planet. Better start collecting data now.

We therefore have to use toxicological models, which are full of assumptions that can never be checked. Toxicologists are also notoriously poor risk communicators, with reports full of jargon and with almost no attempt at clear explanation. For example, the US Environmental Protection Agency’s IRIS database summarises the evidence of the carcinogenic potential of acrylamide in terms of an estimate of the ‘oral slope factor’ of 0.5 per mg/kg/day, or equivalently 0.0005 per µg/kg/day, where ‘µg’ denotes a microgram, or a millionth of a gram.

Some effort reveals that this number gives an estimated upper bound for the chance, given a daily amount of acrylamide in your diet, that it will give you cancer during a 70 year lifetime. The standard assumption is a 70kg adult, who therefore has, for each µg of acrylamide they consume every day, at most a 0.0005/70 = 1 in 140,000 chance of developing cancer.

So now we need to know how much acrylamide we get from a coffee habit. A Swedish study estimated that an average single (160ml) cup of roasted coffee contains 0.45 µg of acrylamide. So this means that, if you drink one cup of such coffee each day, the EPA estimates your chance of developing cancer increases by at most 0.45 x 0.0005/70 = 0.000003, or 1 in 300,000. To put this in perspective, the average lifetime risk of developing cancer is around 1 in 2: the acrylamide from a lifetime’s daily cup of coffee is therefore at most a 150,000th part of your overall cancer risk.

Remember that the judge’s criteria for an acceptable risk is less than 1 in 100,000 chance of cancer. Since each daily cup of coffee raises the risk by 1 in 300,000, the ‘safe’ limit is therefore three cups of coffee. Since only 30% of US coffee drinkers drink more than this, and the legal criterion concerns an average consumer, then the acrylamide in coffee appears to require no warning under Proposition 65. My case rests.

Comparing benefits and harms

I’ve noted the judge’s criticism that the defence had not compared the benefits and harms of coffee. Even though this seems a rather separate issue, it’s still an illuminating comparison to make.

Assuming, very pessimistically, that cancer shortens your life expectancy by 20 years, then the 1 in 300,000 chance that the acrylamide in that daily cup will give you cancer works out at a lowered life expectancy of 0.0008 months, or a total of half an hour lost on average due to a lifetime’s consumption. When it comes to the benefits, extensive evidence exists that routinely drinking coffee improves overall health, with a massive recent review concluding that ‘high versus low consumption was associated with an 18% lower risk of incident cancer’. The data from a study of over 500,000 people suggests that a habit of a daily cup is associated with around 3 months increased life expectancy for men, and around 1 month in women. Therefore the benefits of coffee may outweigh the acrylamide risks by a factor of at least 1/0.0008 = 1,250, or say a thousand, an extraordinary benefit/harm ratio.

As I said, I have not seen the expert testimony, and maybe they said all this — after all, Dr Rhomberg is an ex-EPA scientist. But if these arguments were made, and the court still found against them, it makes the judgement even more bizarre.

Hazard vs risk

What are the general lessons that can be learnt from this case?

First, acrylamide can be considered a hazard with possibly some potential for harm in very extreme cases, but it is not a risk in normal life, even for people who consume a lot. The distinction between hazard and risk is important: a lion in a cage is a hazard, but not a risk, as long as the door is kept shut. Flying in a plane is a hazard, but is a negligible risk (again provided the door is kept shut).

Nevertheless there have been repeated efforts to warn us of the perils of acrylamide in our food. This was illustrated in the ill-fated 2017 campaign by the UK Food Standards Agency to warn us of the dangers of burnt toast and crispy roast potatoes, which was widely criticised and essentially withdrawn within hours of its start.

So a clear lesson is that regulation of hazards, rather than risks, can lead to absurd legal conclusions.

Best to be on safe side — the different meanings of precaution.

In this legal case, rather than the plaintiff having to prove harm, the defendants had to prove that the acrylamide had a negligible cancer risk. Coffee was guilty until proven innocent. This is an example of the ‘strong precautionary principle’ — something is assumed harmful until proven safe. This is a long way from the original formulation of the precautionary principle, which said that the evidence of harm from something does not have to be utterly overwhelming before taking some precautionary measure. Just think of climate change.

Another example of the weaker precautionary principle concerns advice not to drink alcohol in pregnancy. The evidence that low levels of alcohol cause harm is very weak, but we do know that excessive alcohol consumption in pregnancy can harm the foetus, so it seems reasonable to advise women not to drink ‘just to be on the safe side’.

Getting things in proportion

Presumably Proposition 65 was passed by citizens with the best of intentions, in particular shaming companies into making their products safer. But judgements such as this can bring the whole idea of public health messages into disrepute — why should anyone trust health ‘experts’ when ludicrous warnings on coffee start appearing? Either people will reduce their coffee consumption, with possible negative effects on their health, or they will ignore such messages in future, which again could be harmful.

This might not improve the happiness of the meal

There are estimated to be 180,000 deaths due to obesity each year in the US, and since California has an eighth of the US population, presumably this includes over 20,000 Californian deaths. Maybe this is a more important concern than warnings about coffee.

Excessive attention to hazards, rather than real risks, can waste resources, threaten trust, and damage health.



David Spiegelhalter

Statistician, communicator about evidence, risk, probability, chance, uncertainty, etc. Chair, Winton Centre for Risk and Evidence Communication, Cambridge.