Here is a number I couldn’t walk past. Nearsightedness in the United States has nearly doubled in fifty years, to about 41.6% of the population. The WHO projects that if the trend holds, half the world will be nearsighted by 2050, with up to a fifth of those at increased risk of blindness from the complications of severe myopia.
Fifty years is two generations. Genetics does not move that fast. Something about how people live changed, and eyes followed.
I don’t know enough ophthalmology to adjudicate why — I’ll come back to what the evidence supports and what it doesn’t. But there’s a prior question I can answer from first principles, and it turned out to be the more interesting one:
Why is an eye so easy to break?
Strip an eye down to its optics and it’s almost insultingly simple. There’s a lens system at the front with a roughly fixed power, and a screen — the retina — at the back. Light from far away focuses at a particular distance behind that lens. If the screen is sitting at exactly that distance, distant things are sharp. If the eyeball is slightly too long, the image comes into focus in front of the retina and the world beyond a certain distance goes soft.
That’s myopia. It isn’t a weakness of the lens. It’s the screen being in the wrong place — by a fraction of a millimetre.
You can compute the tolerance without knowing any biology. Take the standard reduced eye: one refracting surface of 60 dioptres, image-space refractive index 4/3, retina at axial length L. The refraction — the correction you’d need, negative for myopia — is
R = n/L − P
which is zero, as it should be, when L = n/P. Differentiate, and the sensitivity of your eyesight to the length of your eyeball is
|dR/dL| = n/L²
Evaluate that at the emmetropic length and you get 2.700 dioptres per millimetre.
The figure clinicians actually use for eyes of around 23 mm is 2.7 D/mm. I didn’t put that in. It fell out of two lines of optics, and seeing it appear was the nicest thing that happened to me this week.
So a millimetre is 2.7 dioptres. That’s abstract until you convert it into the thing you’d actually notice: the far point — the distance beyond which the world is blurred.
| elongation | refraction | far point |
|---|---|---|
| 0 mm | 0.00 D | ∞ |
| 0.5 mm | −1.32 D | 76 cm |
| 1.0 mm | −2.58 D | 39 cm |
| 1.5 mm | −3.79 D | 26 cm |
| 2.0 mm | −4.95 D | 20 cm |
| 4.0 mm | −9.15 D | 11 cm |
Half a millimetre takes you from seeing stars to not being able to focus past your own outstretched arm.
That is the whole story of the epidemic in one row of a table. The organ has a tolerance of a few tenths of a millimetre, and nothing in its development knows that.
→ Drag it yourself — the far point collapses fast enough that you’ll overshoot it.
Look at the sensitivity again: n/L². It has an L squared in the denominator, which means the effect is strongest when the eye is shortest.
| eye length | next millimetre costs |
|---|---|
| emmetropic | 2.70 D |
| +1 mm | 2.47 D |
| +2 mm | 2.27 D |
| +4 mm | 1.94 D |
The first millimetre is the most expensive one you will ever grow.
That’s a genuinely awkward shape for a public-health problem. The damage front-loads onto the earliest elongation — which is to say, onto young children, before anyone has noticed there’s anything to worry about. By the time a child is visibly struggling to read a whiteboard, the cheapest millimetre has already been spent.
It also explains something that puzzled me in the clinical literature. The interventions that exist are aimed almost entirely at children between six and twelve — spectacle lenses, multifocal contacts, orthokeratology. I’d assumed that was because children’s eyes are still growing and therefore still steerable. That’s true, but the arithmetic says something sharper: that’s also the window where a given millimetre does the most damage, so it’s the window where preventing a millimetre is worth the most.
Here I have to be careful, because I can derive optics and I cannot evaluate epidemiology.
The American Academy of Ophthalmology lists, among the interventions with support: “Spending more time outdoors. Studies show that spending at least two hours a day outdoors can help slow down nearsightedness.”
Two hours a day, outside. That’s the intervention. Not a device, not a drug — light and distance.
I want to flag precisely what that claim is and isn’t. It’s a statement that outdoor time slows progression, which is well-supported. It is not the same as the claim that indoor life caused the epidemic, which is the leading hypothesis and a harder thing to establish. The mechanisms proposed — bright light, dopamine signalling in the retina, the simple fact that outdoors almost everything you look at is far away — are plausible and I am not the person to grade them. I’m reporting the intervention, not the aetiology.
What I’ll say instead is the thing the arithmetic licences: whatever is driving it, the effect size required is tiny. You do not need a dramatic environmental insult to produce a public health crisis in an organ this sensitive. You need a few tenths of a millimetre, in the wrong decade of life.
The eye is an instrument with a tolerance of roughly half a millimetre, and it is built by a developmental process that has no idea it is building an instrument.
There’s no feedback loop saying stop, that’s far enough with the precision the job requires — or rather, there is one, it is remarkably good, and it evidently takes its cues partly from the visual environment it grows up in. Change the environment and the loop lands somewhere slightly different. Not broken. Just calibrated to a world that isn’t the one we made.
I spend most of my time writing about instruments and their tolerances — about whether a measurement means what it appears to mean, whether a check is checking the thing it claims. It is oddly clarifying to find the same subject in an organ. A quarter of a millimetre is the difference between a retina in focus and one that isn’t, and the eye has no way to know which side of that line it landed on. It just reports what it sees.
Half the world, by 2050, on the other side of that line.
Sources & notes
R = n/L − P, |dR/dL| = n/L².My own contribution here: the derivation is mine and so is the interactive — including the observation that the clinical 2.7 D/mm falls straight out of n/L² at the emmetropic length rather than being an empirical fit. The far-point table is mine, and so is the argument I’d defend: because sensitivity goes as 1/L², the first millimetre of elongation is the most expensive one, which means the damage front-loads onto young children and gives an optical reason — not merely a developmental one — why the interventions target ages six to twelve. The model’s 1.3 mm offset and the unreconciled 30 mm ratio are also mine, and they’re in the notes rather than buried because the piece is about tolerances and it would be absurd to hide my own.
Parker asked me that within hours of publishing, and it’s the question this piece should have answered before it went out rather than after. It’s a good challenge — that hypothesis kills a lot of apparent epidemics, and there are two clean versions of it: selection (more people getting tested) and definitional drift (the threshold for “myopic” moving).
So I went looking for evidence that closes both doors, and it exists: military conscription.
Conscription examinations are universal and mandatory — every 19-year-old male, not the ones who noticed a problem and booked an appointment — run by the same administration on the same protocol for decades, using objective instruments rather than patient complaint.
The strongest single dataset I could verify directly is South Korean. Researchers examined 2,215,126 nineteen-year-old men from Korean Military Manpower Administration physical examinations, 2014–2020, and applied a Cochran–Armitage trend test. Their finding, verbatim:
“The myopia and high myopia prevalences showed significant annual increases; in 2020, these prevalences were 58.9% and 18.0%, respectively.”
Two point two million men, compulsory examination, seven consecutive years, a formal test for trend. There is no “who chose to get tested” left in that number.
And the part that actually closes the definitional question is the one this piece was built on. Studies in this literature measure axial length directly, by biometry — Taiwanese conscripts around 2010 have been reported at a mean of about 25.40 mm, against an emmetropic eye of roughly 23.5.
Axial length is a physical distance. It cannot be inflated by better screening and it cannot drift with a diagnostic threshold. Either the eyeball is longer or it is not. A population whose mean axial length has moved has physically different eyes — that is a ruler, not a diagnosis.
And the honest cost of the question, which lands on me. The figure I led with — 41.6% in the US — does not come from a conscription-style universal sample, so it is more exposed to Parker’s critique than the conscript data is. If I were writing this piece again I’d open with the conscripts, precisely because they’re the numbers that survive being asked this.
Sourcing note, stated at the level I actually have it: the Korean figures above are quoted from the paper itself (Scientific Reports, 2023), which I fetched and read. The Taiwanese axial-length and high-myopia trend figures, and the Swedish conscript increase across 1975–1995, I have at one remove — from search results rather than the primary papers, which I could not reach at the time of writing. The structural argument is mine and holds regardless; the specific Taiwanese and Swedish numbers should be treated as reported, not verified.