How to Read a Scare Headline

TL;DR. The most useful chapter in this book, because it is the one that keeps working after the specifics go out of date. Four questions dismantle almost every food and chemical scare story: is this hazard or risk, is the number relative or absolute, what kind of study was it, and compared with what? A "50 percent increase" in a risk of 2 in 10,000 is 1 extra case in 10,000. An IARC Group 2B classification means somebody found some evidence somewhere. A mouse fed a hundred times the human dose is not you.

1. Hazard or risk?

The distinction from Chapter 87, because it defuses more headlines than anything else:

$$\text{Risk} = \text{Hazard} \times \text{Exposure}$$

IARC classifies hazard. Its groups say how confident the working group is that something can cause cancer under some circumstances, not how much cancer it causes.

Group 1 contains processed meat and plutonium. Their risks differ by orders of magnitude. When a headline says "X is a Group 1 carcinogen, like asbestos," it is technically accurate and communicating nothing useful.

Regulatory agencies (EFSA, FDA, JECFA, EPA) assess risk, which is why they can classify something as a hazard and simultaneously say current intakes are safe. The aspartame case in July 2023 is the clean example: IARC said Group 2B, JECFA reaffirmed the acceptable daily intake unchanged on the same day, and both were correct (Chapter 57).

2. Relative or absolute?

The single most common distortion in health reporting.

"Eating bacon daily increases your risk of bowel cancer by 18 percent."

That is a relative risk increase. To make it meaningful you need the baseline.

• Lifetime bowel cancer risk in the UK: roughly 6 in 100.
• An 18 percent relative increase: roughly 7 in 100.
• Absolute increase: about 1 extra case per 100 people.
• Number needed to harm: about 100.

Both descriptions are true. One sounds alarming and one sounds manageable, and the second is the one that should inform a decision.

The same trick works in the other direction with benefits. "This drug halves your risk" is impressive until you learn the risk was 2 in 1,000, and 1,000 people must take it for one to benefit.

Always ask for four numbers:

1. The baseline risk
2. The absolute change
3. The number needed to treat or harm
4. Over what time period

If a report gives only a percentage change and no baseline, it has not told you anything.

3. What kind of study was it?

The hierarchy, roughly:

Three questions to ask about any study:

"Was it in humans?" A great many alarming headlines are about mice or cells. Mice are not small humans; the majority of drugs that work in mice fail in people.

"At what dose?" Rodent studies deliberately use doses many multiples of human exposure to detect effects in small groups. The saccharin bladder tumour finding involved feeding rats the equivalent of hundreds of cans of diet drink a day, through a mechanism specific to rat urine chemistry that does not apply to humans. Saccharin was delisted as a carcinogen in 2000.

"By what route?" Injected into the abdomen is not eaten.

On observational studies specifically: they are indispensable and they cannot establish causation. People who eat more vegetables also exercise more, smoke less, drink less, and are wealthier. Statistical adjustment helps and never fully resolves it. The clearest demonstration is hormone replacement therapy: observational studies consistently showed cardiovascular benefit, and the randomised Women's Health Initiative did not (Chapter 79). Healthy user bias was doing the work.

When observational studies are more convincing: a large effect size, a dose-response relationship, consistency across different populations and designs, a plausible mechanism, and temporal order. Those are essentially the Bradford Hill considerations, and smoking and lung cancer satisfy all of them, which is why that conclusion was reached without a randomised trial.

4. Compared with what?

Every choice has a counterfactual, and headlines almost never state it.

• Not using preservatives means more food poisoning.
• Not using pesticides means substantially less food.
• Avoiding fish for mercury appears worse for child neurodevelopment than eating the right fish (Chapter 92).
• Avoiding vaccines means the diseases return (Chapter 74).
• Not taking a statin means the baseline cardiovascular risk.
• Organic instead of conventional means roughly 20 percent more land for the same food (Chapter 7).

Substitution matters in nutrition especially. "Lower carbohydrate is associated with lower mortality" depends entirely on what replaced it: plant protein and fat, favourable; animal fat, not (Chapter 11).

5. The specific tricks to recognise

"Contains chemicals." Everything does.

"Detected in." Modern analytical chemistry detects parts per trillion, which is roughly a second in 30,000 years. Detection is a statement about instruments, not about risk. The number that matters is the amount as a fraction of the safety limit.

"Linked to." Almost always means an observational association. It is doing a great deal of work in a very small phrase.

"Scientists say" / "experts warn." Which scientists? A single preprint? A press release?

The press release problem. Studies of health news found that most exaggeration in reporting originates in university press releases, not in journalists. A cell study becomes "scientists discover cure."

Single-study syndrome. One study rarely changes anything. Replication is the whole point. Ioannidis's 2005 paper "Why Most Published Research Findings Are False" is the standard reference, and the replication crisis is real across nutrition, psychology, and biomedicine.

Publication bias. Positive findings are more likely to be published than null ones, which inflates apparent effects. Pre-registration and results-reporting requirements exist to combat it.

P-hacking and multiple comparisons. Test enough variables and something reaches p < 0.05 by chance. Nutritional epidemiology, which asks about hundreds of foods and dozens of outcomes, is structurally vulnerable to this.

The "everything causes cancer" effect. Jonathan Schoenfeld and John Ioannidis picked 50 ingredients at random from a cookbook and found that 80 percent had published studies claiming a link with cancer risk, in both directions, and that effect sizes were mostly weak and inconsistent. That is a finding about the literature, not about food.

Absolute confidence. Real findings come with confidence intervals and limitations. "Proven" and "miracle" are marketing words.

Appeal to nature. Chapter 87.

Conflict of interest, in both directions. Industry funding genuinely biases results, and so does funding and reputation attached to finding harms. Look at the data, not just the funder.

6. Worked examples from this book

"Bacon causes cancer, says WHO." Hazard classification, Group 1. Absolute effect: roughly 1 extra colorectal cancer per 100 people at 50 g a day. Real, small, worth reducing, not asbestos (Chapter 55).

"Aspartame possibly causes cancer." IARC Group 2B, the weakest positive category, alongside pickled vegetables. JECFA reaffirmed the safe intake the same day. You would need roughly 9 to 14 cans of diet drink daily to reach it (Chapter 57).

"Sunscreen chemicals absorbed into the blood." Correct, and the FDA said explicitly it did not indicate harm and that people should keep using sunscreen. It meant safety data were required (Chapter 94).

"Statins cause muscle pain." In practice, 10 to 25 percent report it. In blinded trials the same symptoms occur on placebo. Nocebo, and rhabdomyolysis is real and rare (Chapter 85).

"Coffee causes cancer." IARC removed coffee from the "possibly carcinogenic" list in 2016. What survived is that very hot drinks above 65 °C are probably carcinogenic to the oesophagus, whatever they are (Chapter 60).

"MMR causes autism." Fraudulent, retracted, author struck off, and refuted in cohorts totalling millions (Chapter 74).

7. A checklist

Ten questions, in order of usefulness:

1. Hazard or risk? Can it, or does it at real exposures?
2. What is the absolute change, and the baseline?
3. Humans, animals, or cells?
4. Randomised or observational?
5. At what dose, and by what route, compared with realistic exposure?
6. How many people, for how long?
7. Is this one study, or a body of evidence?
8. Compared with what? What is the counterfactual?
9. Who is saying it, and what is the primary source? Find the paper, or at least the abstract.
10. Does the effect size make biological sense, or is it implausibly large for a small exposure?

Where to check: Cochrane reviews, the NHS "Behind the Headlines" archive, Science Media Centre expert reactions, EFSA and FDA statements, and, for nutrition, the underlying paper's abstract, which usually reports the absolute numbers the headline omitted.

8. What this does not license

Scepticism is not the same as dismissal, and this chapter can be misused in the other direction.

Some things are settled, and treating everything as contested is its own error:

• Smoking causes cancer.
• Vaccines work and do not cause autism.
• LDL cholesterol is causal in atherosclerosis.
• Trans fats are harmful.
• There is no safe level of lead.
• Alcohol causes cancer.
• Air pollution kills.
• Antibiotic resistance is real and growing.

The intellectually honest position is not "nobody knows anything." It is that different claims carry different evidential weight, and the skill is telling them apart.

Consensus is not proof and it is not nothing either. When multiple independent bodies reviewing the same evidence reach the same conclusion, that is meaningful. When a single researcher contradicts them, occasionally they are Barry Marshall drinking H. pylori (Chapter 71) and usually they are not.

9. The bottom line

• Hazard is what something can do; risk is what it does at your exposure. IARC classifies hazard, which is why processed meat and plutonium share a group.
• Always convert relative to absolute. A percentage without a baseline is not information.
• Ask whether it was humans, at what dose, by what route. Cell and rodent studies at enormous doses generate most alarming headlines.
• Observational studies cannot establish causation, and the HRT reversal is the standard cautionary tale.
• Ask what the counterfactual is. Every avoidance has a cost, and headlines never mention it.
• One study rarely changes anything. Press releases exaggerate more often than journalists do.
• Scepticism is not dismissal. Smoking, lead, vaccines, LDL, and air pollution are settled, and pretending otherwise is the mirror-image error.

Sources and notes

IARC's classification system and its explicit hazard framing follow the Monographs preamble. The aspartame hazard-versus-risk pairing follows the joint WHO release of July 2023. Relative versus absolute risk communication follows Gigerenzer's work on risk literacy and the Bradford Hill considerations for causation follow Hill, Proceedings of the Royal Society of Medicine, 1965. The evidence hierarchy and its limitations follow standard epidemiology texts and GRADE. The HRT observational-versus-randomised divergence follows the Nurses' Health Study analyses and the Women's Health Initiative. The saccharin rat bladder mechanism and 2000 delisting follow the US National Toxicology Program. Replication problems follow Ioannidis, PLoS Medicine, 2005. The cookbook ingredient analysis is Schoenfeld and Ioannidis, American Journal of Clinical Nutrition, 2013. Press release exaggeration as the origin of most health news distortion is Sumner et al., BMJ, 2014. Publication bias and trial registration follow the AllTrials campaign and ICMJE requirements. Worked examples draw on the sources cited in their respective chapters.

Open questions. There is no agreed method for communicating uncertainty to the public that has been shown to work, which is why this chapter offers questions rather than a formula.

👉 Next: allergies and intolerances.