The Percentage and the Thing It Measures: Reading Dementia Headlines Honestly
A note before anything else: this is not medical advice, and it is not a substitute for a doctor. Anyone worried about memory — their own or someone else’s — should see a physician. Memory problems have many causes, and a meaningful number of them are treatable. Nobody should start, stop, or change a medication based on an article.
What follows is about how to read the numbers, not what to do about them.
Two claims circulate widely in coverage of dementia prevention. Both come from good research. Both are stated accurately. And both are routinely understood to mean something much larger than what was actually measured.
”48% less cognitive decline” from treating hearing loss
This comes from the ACHIEVE trial (Lin FR, et al., The Lancet 2023;402:786–797), a well-designed randomized study of 977 adults aged 70–84 with untreated hearing loss, followed for three years.
The part that rarely travels with the headline: the trial’s primary result, across everyone in it, was null. The difference between the hearing-aid group and the control group was 0.002 standard deviations, with a p-value of 0.96 — about as close to nothing as a result gets.
The 48% comes from a subgroup. The trial recruited from two populations: 238 people drawn from an existing cardiovascular study, who were older and had more chronic illness, and 739 healthier volunteers recruited from the community. Analyzing those groups separately was planned in advance, not invented after the fact — that matters, and it is to the researchers’ credit. In the higher-risk group, hearing intervention was associated with 48% less change on a cognitive test score.
Three qualifications belong with that number:
- It describes a subgroup of about a quarter of the participants.
- It measures the slope of a test score, not whether anyone developed dementia.
- On the harder question — did fewer people actually develop cognitive impairment — the answer was no, even in that subgroup (hazard ratio 0.94, 95% CI 0.54–1.64).
A six-year extension is running specifically to look at dementia diagnoses. It has not published.
”27% slowing” from a $26,500-a-year drug
Lecanemab’s pivotal trial (van Dyck CH, et al., NEJM 2023;388:9–21) measured decline on a scale called CDR-SB, which runs from 0 to 18. Over 18 months, the placebo group worsened by 1.66 points and the treated group by 1.21.
The difference is 0.45 points on an 18-point scale. Divide 0.45 by 1.66 and you get 27%.
“27% slowing” and “0.45 points out of 18” are the same fact. They land very differently.
Whether 0.45 points is meaningful to a patient is genuinely unsettled among researchers. One widely cited analysis (Andrews JS, et al., 2019) proposed thresholds of 0.98 and 1.63 points for a change to count as clinically important — both above 0.45. Defenders note those thresholds were derived over a shorter timeframe, making direct comparison shaky. A 2024 review put it plainly: thresholds for meaningful differences on these measures “have not been established.” Anyone stating confidently that the benefit is real, or confidently that it is negligible, has picked one paper and not mentioned the others.
The pattern worth carrying
A percentage improvement is a ratio. On its own it says nothing about size, because it inherits its scale from whatever it is a percentage of. A 27% reduction in a 1.66-point change is 0.45 points. A 27% reduction in something enormous would be enormous.
Three questions turn most health headlines back into information:
- Percent of what? Find the raw numbers. If a story about a trial never gives them, that is itself informative.
- Did the trial’s main question get answered? Subgroup findings can be legitimate and interesting, and they are not the same as the result the study was built to produce.
- What was actually measured? A test score is not a diagnosis. Slowing a rate of change is not preventing an outcome.
One more, from a different corner of this literature: group averages can hide who something happened to. A 2009 study of people learning their genetic risk found no overall increase in anxiety or depression — but within that group, people who received bad news showed significantly more distress than those who did not. Both statements are true. Only one is about the person reading.
And a genuinely encouraging note
None of this argues against treating hearing loss. It argues for treating it for the hearing — for conversation, for music, for not withdrawing from a dinner table because following it is exhausting. Those benefits are immediate and certain. Over-the-counter hearing aids have been available in the US since October 2022 for perceived mild-to-moderate hearing loss, at a fraction of previous prices.
That is the quiet advantage of the well-evidenced basics: their value does not depend on winning an argument about a distant outcome. It shows up the same week.
Sources: Lin FR, et al., ACHIEVE trial, The Lancet 2023 (PMID 37478886). van Dyck CH, et al., NEJM 2023 (PMID 36449413). Andrews JS, et al., Alzheimer’s Dement (N Y) 2019;5:354–363. Green RC, et al., NEJM 2009 (REVEAL). FDA, OTC Hearing Aids guidance, effective 17 October 2022.
Again: not medical advice. Talk to a real professional who knows your body.
Get the next one in your inbox
When a new essay is ready, we’d like to send it to you — the whole essay, free. That’s the entire arrangement: we write the letters, you read the ones that earn it. No funnels dressed up as friendship, no “act now.” Unsubscribe anytime and we’ll assume you had good reasons.
Free. No spam, no selling your address. Unsubscribe anytime.