Health

Cadmium, aspartame, phones: how to tell virality apart from impact

Every year, a handful of health scares take off on social media, driven by a mix of legitimate concern and headline-grabbing shortcuts. Scented candles, aspartame, mobile phones: these "scandals" share a common feature — they generated intense coverage for risks that were often very modest. Cadmium, at first glance, looked like more of the same. A heavy metal. Fertilisers. Alarming headlines. Except this time, the data tells a different story.

---

The sceptic's reflex, or why I no longer react to health alerts

Candles, aspartame, phones: when media coverage far outstrips the evidence

Every year for the past decade or so, one or two health scares do the rounds on social media with the same mechanics: a chemical compound detected, an alarming headline, and a wave of anxiety that lasts a few weeks before giving way to the next story.

Three recent examples.

First, scented candles. When they burn, they release volatile organic compounds, some of which are genuinely carcinogenic at high doses: benzene, formaldehyde, benzo[a]pyrene. What alarmist articles consistently fail to mention is the question of dose. Studies conducted in real-use conditions measured formaldehyde concentrations of around 3.5 µg/m³ in the air of a closed room — 28 times lower than the threshold recommended by the WHO.[1] A Danish study from 2022, which followed people burning candles more than four times a week, found no link with cardiovascular or respiratory incidents. The risk exists in theory, under conditions of extreme exposure. In normal use, it is negligible. Mieux Donner covered this in a dedicated article.

Next, aspartame. In July 2023, the International Agency for Research on Cancer (IARC) classified this sweetener as "possibly carcinogenic to humans" (Group 2B).[2] Press headlines were immediate and unambiguous. What most did not report: the IARC itself noted the evidence was "limited". The FAO/WHO joint expert committee on food additives, which was simultaneously assessing the actual risk, kept the acceptable daily intake unchanged at 40 mg per kilogram of body weight. To exceed that threshold, a 70 kg adult would need to drink more than nine cans of diet soda a day. At normal consumption levels, no risk has been demonstrated. Group 2B, it is worth noting, also includes kimchi, aloe vera, and the caffeic acid found in coffee.

Finally, mobile phones, placed in that same Group 2B since 2011, on the basis of a single study with "limited evidence" — the IARC's own wording.[3] Since then, a systematic review published in 2024, including the large prospective COSMOS study conducted across 264,000 participants in five European countries, found no link with brain tumours.[4] The WHO is preparing a new monograph for 2028, expected to reflect this updated evidence.

Three alerts, three cases where the media coverage and the anxiety generated were vastly out of proportion to what the evidence justified.

The intensity of an alert is not a measure of its severity

This is not a coincidence. Health scares that spread most effectively tend to share certain features: they involve an everyday product, they attach that product to a word like "carcinogenic", and they rest on laboratory studies conducted at doses nobody encounters in real life. This type of content spreads because it triggers fear and touches on family and food.

After years spent analysing causes and interventions through an evidence-based lens, I have internalised a simple reflex when faced with any new health alert: before reacting, ask three questions. What is the quality of the evidence? What is the size of the effect? And can we act on it?

Those are the three questions I asked about cadmium in the spring of 2026, when it began filling news feeds. And for the first time in a long while, the answers surprised me.

---

Three questions to avoid being misled by a health risk

First question: what is the quality of the evidence?

Not all studies carry equal weight, and the hierarchy of evidence in health sciences is well established.

At the top: meta-analyses of randomised controlled trials, which pool results from multiple independent studies on the same question. Below that: prospective cohort studies, which follow populations over time. Further down: retrospective case-control studies, prone to recall bias. And at the bottom: laboratory studies on animals or cells, which frequently detect effects at doses entirely unrelated to real human exposure.

The first question to ask is therefore not "is this classified as carcinogenic?" but "what type of studies support that claim, and at what doses?"

Second question: what is the size of the effect?

A relative risk is a ratio, not an absolute risk. An RR of 2 means the risk is doubled relative to a reference population: if the baseline risk is 0.01%, doubling it gives 0.02% — an absolute difference of one hundredth of a percentage point. If the baseline risk is 10%, doubling it means 10 additional percentage points.

This is why comparisons between relative risks, without an anchor in absolute risk, can be highly misleading. Some useful benchmarks:

Effect size is the question the media asks least often, and the one that most changes the reading of a health alert.

Third question: how many people are affected, and can we act?

A risk can be well documented and genuine, yet affect a very limited number of people or have no identifiable lever for action. Conversely, a modest risk applied across an entire national population can represent a considerable public health burden. This is the difference between individual effect size and the scale of the problem.

The lever for action is equally important. Some risks are individually modifiable through simple behavioural changes. Others require regulatory or industrial action. Others still concern only a very specific fraction of the population.

These three dimensions — quality of evidence, size of effects, scale and tractability — form the framework I apply to every health alert. Let us see what it produces when applied to cadmium.

---

Cadmium, or how a sceptic had to revise his judgement

Evidence of a rarely achieved level of certainty

Cadmium is not classified as Group 2B. It has been classified as Group 1 by the IARC since 2012: definitively carcinogenic to humans, the highest category, the one shared by tobacco and asbestos.[5] This classification rests on multiple epidemiological datasets, documented biological mechanisms, and consistent animal studies.

The data on contamination levels in the French population come from the national Esteban study conducted by Santé publique France, with a representative cohort of adults and children. This is not a laboratory study conducted at extreme doses: it is a direct measurement of cadmium levels in the urine of ordinary French people.[7] The Anses report published in March 2026, drawing on more than 400 pages of international scientific literature, concluded that 47.6% of French adults exceed the renal toxicological reference value.[6]

On evidence quality: the signal is solid. First point confirmed.

A real risk on both cancer and kidneys

The media coverage of cadmium broadly reflected reality: a serious problem, documented for years, and unjustifiably ignored. The framework's most useful contribution here is distinguishing between two effects that carry different levels of certainty.

On cancer, the reference meta-analysis (Cho et al., 2013, PLOS ONE) gives a relative risk of 1.15 in Western populations for those most exposed through diet.[8] This is a real, statistically significant effect, but its order of magnitude is comparable to that of unprocessed red meat on colorectal cancer. A systematic review published in 2024 suggests probable causal links with pancreatic, lung, and bladder cancers, but without quantifying the total burden attributable to dietary exposure in France.[9]

Cadmium accumulates in renal tissue with a half-life of 10 to 30 years and is never fully eliminated. The 47.6% of French adults who exceed the toxicological reference value do not all have clinical kidney disease: they are at an exposure level beyond which the risk of silent deterioration of renal function over several decades is documented. This is a long-term cumulative risk, alongside the cancer effect.

Massive exposure, a 15-year regulatory delay, a solution whose cost was exaggerated

French exposure is structurally higher than in other European countries: Esteban 2021 data show average contamination levels two to four times higher than in Germany, Italy, or the United States depending on age group. This difference is not a geological accident: it is the product of regulatory choices.

For years, France maintained a standard allowing 90 mg of cadmium per kilogram of phosphate fertiliser, while countries such as Finland, Hungary, and Slovakia had long imposed a ceiling of 20 mg/kg without weakening their agriculture as a result. The European Union itself adopted a regulation in 2019 setting out a trajectory towards 20 mg/kg, which France was not implementing. This fifteen-year delay is documented: Anses had been recommending 20 mg/kg "as soon as possible" since 2019.

The economic argument advanced to justify this delay deserves scrutiny. The FNSEA cited a cost increase of €4,000 to €7,000 per cereal farm if Moroccan fertilisers were entirely replaced by Scandinavian imports. This is the most expensive scenario: full substitution using scarce alternative sources. According to European Commission research, the technical scenario — decadmiation of Moroccan phosphate — yields an estimate of around €2 per hectare, or a national annual additional cost in the order of €52 million.[10] Since February 2025, OCP, France's main Moroccan supplier, has declared that all fertilisers it exports to France already meet the 20 mg/kg threshold.

This is not a problem without a solution. It is a problem with identified solutions, whose cost has been manifestly overstated in public debate, and whose implementation has been delayed by more than a decade.

---

Similar tools to guide our generosity

The three questions I have just applied to cadmium — what is the quality of the evidence, what is the size of the effects, and can we act — are exactly the same ones I use to assess the effectiveness of charities and guide our recommendations at Mieux Donner. This is not a coincidence: they form a general framework for distinguishing real problems from false signals, and effective interventions from well-intentioned ones with low impact.

Intermediate results confused with final impact

Many organisations track activity indicators: number of meals distributed, number of workshops run, number of people "made aware". These measures are real, verifiable, and often sincere. They are also the kind of numbers that appear prominently on websites or in press headlines. But they say nothing about final impact. A charity distributing school supplies may look impressively productive on paper. If those supplies would have been purchased by families anyway, or if they do not change educational trajectories, the real impact is close to zero. The question of what would have happened without the intervention — what is known as counterfactual reasoning — is rarely asked.

In France, the vast majority of charity evaluations rest on bespoke indicators, built project by project, with no reference to comparable international standards such as DALYs or cost per life saved. This makes any comparison between interventions impossible, and deprives funders of the only information that truly matters: does this euro do more here than elsewhere?

Actions already known, often, not to work

Worse than not measuring impact: funding interventions that existing literature has already shown produce little or no effect.

It is, in a sense, the philanthropic equivalent of the cases discussed earlier: raising alarm about a risk backed by largely reassuring evidence, mobilising attention and resources, without producing any positive effect. The energy expended is not neutral — it crowds out space that could have gone to genuinely effective causes and interventions.

The Scared Straight programme, which aimed to reduce juvenile delinquency by taking teenagers on prison visits, was massively funded for years. Rigorous evaluations showed not only that it did not work, but that it slightly increased the likelihood of reoffending. PlayPumps, water pumps activated by children's playground equipment in developing countries, attracted enthusiastic press coverage and tens of millions of dollars in funding. They proved less effective, more expensive to maintain, and less suited to actual needs than the hand pumps they replaced.

These cases illustrate a structural problem: without any obligation to draw on available evidence before acting, substantial funding can flow to interventions that research had already assessed negatively. As we explain in our article on counterproductive donations, a poorly directed donation can be useless, or even harmful.

The media profile of a cause is unrelated to its effectiveness

The most visible cause is not the most urgent. The most funded is not the most effective. And the best-known charity is not necessarily the one where each pound or euro will have the greatest impact on real lives.

Studies on philanthropic behaviour consistently show that donations are heavily influenced by emotional proximity, organisational name recognition, and the vividness of the narratives presented. None of these factors correlates with the cost-effectiveness of an intervention. A charity distributing bed nets to prevent malaria saves lives at a documented cost, but generates neither heart-rending images nor viral campaigns.

The gaps in effectiveness between charities are nonetheless documented and considerable: according to a study cited by Our World in Data, the most effective charities can have up to a hundred times more impact per pound invested than the average.[11] A hundredfold. This gap is invisible if you choose where to give based on how much coverage a cause receives, or on the feeling that "something must be done".