A headline says a treatment cuts your risk by half. That sounds enormous. Then you find the underlying numbers: two people in a hundred had the event, and with the treatment it was one. The relative reduction really is fifty per cent. The absolute reduction is one percentage point. Both numbers are true and they leave you with completely different impressions. Relative risk is the larger, more quotable figure, so it is the one that reaches the headline. When you see a percentage reduction, the useful question is always: reduction from what, to what?
Trials are expensive and slow, so they often measure a marker that stands in for the outcome you care about - cholesterol rather than heart attacks, bone density rather than fractures, tumour shrinkage rather than survival. Sometimes the marker tracks the outcome faithfully. Sometimes it does not, and the history of medicine contains treatments that moved a marker in the right direction while doing nothing for patients, or harming them. When a report says a drug 'improved' something, check whether the something is the outcome or a proxy for it.
This is the single most useful number in medical reporting and it almost never appears. It answers: how many people have to take this for one of them to benefit? If it is five, that is a powerful intervention. If it is two hundred, then a hundred and ninety-nine people take it, absorb any side effects and cost, and get nothing. Neither figure is inherently good or bad - it depends on the seriousness of what is being prevented - but knowing it changes the conversation completely.
A result found in 60-year-old men with existing heart disease may not transfer to a healthy 30-year-old woman. Trials frequently enrol populations that do not resemble the readership of the article describing them. Age, sex, ethnicity, comorbidity and baseline risk all matter. A finding is not wrong because the population was narrow; it is just narrower than the headline implies.
A great deal of health coverage originates from a university press office rather than from a journal. Press releases are written to be picked up, and studies have repeatedly found that overstatement in coverage often traces back to overstatement in the release. Preprints - posted before peer review - are valuable for speed and have not yet been checked by anyone independent. Neither is disqualifying. Both are worth knowing about before you change anything you do.
Industry funding does not make a study wrong, and academic funding does not make one right. But funding and author conflicts are disclosed for a reason, and they belong in your assessment alongside everything else. Reputable journals publish the disclosure statement; it takes ten seconds to read and it is often the most informative paragraph in the paper.
Absolute numbers, not just percentages. The actual outcome, not a surrogate. How many people were studied and who they were. Whether it was peer reviewed. Who funded it. And the question that catches the most nonsense: does the headline describe a finding in people, or in cells, or in mice? A remarkable proportion of miracle-cure coverage is describing something that happened in a dish.
About this article. This is an explainer, not a report of a news event. It contains no invented statistics, studies or quotations. Every factual claim is either a public regulatory framework, a general statistical concept, or a statement about this publication — and the sources below are where to check them. Nothing here is medical advice; talk to a clinician about your own situation.
Primary references for the claims above.