Mental Illness Has a Critical Period — and It's Over by 25

Most people assume mental health conditions arrive with adult life: the job, the mortgage, the divorce, the bereavement. The data says the opposite. For the majority who will ever have a condition, it has already begun before those pressures exist — often before they can vote, sometimes before they can drive.

The model: critical and sensitive periods

The idea that explains this comes from developmental biology, not psychiatry. In the 1930s the ethologist Konrad Lorenz showed that newly hatched goslings would bond permanently to whatever moving object they saw first — him, in the famous case — but only within a narrow window after hatching. Miss the window, and the bonding never happened. He called it imprinting; the general principle became the critical period: a bounded developmental stage in which a system is unusually open to being shaped, after which the same input has little or no effect.

The hard version (a window that slams shut) has softened over time into the sensitive period — a stretch when a system is most plastic, with influence tailing off rather than vanishing. The evidence base is broad and cross-species: Hubel and Wiesel's Nobel-winning work showed a kitten deprived of vision in one eye during an early window stayed functionally blind in it for life, though vision was mechanically fine. The brain had moved on.

The claim this model makes is specific and testable: risk and plasticity are concentrated in a developmental window, not spread evenly across life. That is exactly the shape the mental-health data shows.

The live case

A meta-analysis by Marco Solmi and colleagues in Molecular Psychiatry pooled 192 epidemiological studies to answer one question: at what age does a mental health condition first appear in someone who will eventually have it?

Among everyone who will ever experience a disorder, roughly a third have their first onset by age 14, and two-thirds by 25. This is "age of onset" — when it starts among those affected — not prevalence, not how many are affected. The conditions sort into three shapes:

• Flat across life. Generalized anxiety (persistent excessive worry), depression, and bipolar disorder (swings between depressive lows and manic highs) begin fairly evenly from the teens into old age. No single window.
• Sharp early window. Social anxiety (intense fear of being judged) and eating disorders cluster in adolescence — nearly 80% of social anxiety begins before 18; eating disorders peak around 16. Survive youth without them and risk falls steeply. This is the critical-period signature.
• A young-adult peak with a long tail. Substance use disorders and schizophrenia (hallucinations, delusions, disordered thinking) cluster around 20, then decline gradually — a sensitive period rather than a hard window.

The model earns its keep here: it tells you where the leverage is. If risk concentrates in a window, so should screening and intervention. For eating disorders and social anxiety that means schools and pediatricians, not workplace wellness programs. Catching a condition near onset — before it hardens into decades of recurrence — is where prevention has traction, and the model tells you when that moment is.

What the model doesn't license

Three cautions, because this is easy to over-read.

"Onset" is measured inconsistently. Some studies count first symptoms, others first diagnosis or hospitalization. Shifting depression from symptoms to diagnosis moved its median onset from 26 to 31. People suffer before the label arrives, so true first experience likely skews younger than diagnosis data shows.

These curves are not prevalence. Each is scaled to those who will ever get the condition. Schizophrenia's curve sits high not because it's common (it's rare) but because onset is squeezed into a narrow band.

Onset is not duration. The curves show when conditions begin, not who carries one now. Recurrence and persistence mean the diagnosed population is older and broader than onset suggests.

And the honest limit: nearly all the underlying studies come from high-income countries. Whether these windows hold elsewhere is genuinely unknown.

Where the same model applies

Once you see "risk and plasticity concentrated in a window," it is everywhere, and it always implies the same lesson: intervene inside the window or work against the grain.

• Second languages. Learn one before puberty and native fluency is routine; start at 40 and accent and grammar rarely fully arrive — a textbook sensitive period.
• Bone density. Largely set by the mid-twenties. The habits that protect you from osteoporosis at 70 are the ones built as a teenager; calcium at 65 is swimming upstream.
• Firms and institutions. Organizations imprint early too — the culture and hiring norms set in a company's first years prove strangely durable, which is why turnarounds are so much harder than founding.

The transferable move: when something matters, ask whether its outcome is being decided in a window you haven't noticed, and whether that window is still open.

What to watch

The "youth onset" story will be tested by studies that standardize how onset is measured and widen the geographic net. If consistent symptom-based definitions push median onset down, the critical-period reading strengthens.

Falsifiable expectation: as low- and middle-income data arrives, the broad shape — most onset before 25 — holds, but specific peak ages shift, because access to diagnosis varies so much by place. Roughly 70% likely the under-25 majority finding survives the next decade of better data.

The practical takeaway: for most people, mental illness is a condition of youth. The systems built to catch it are mostly built for adults — intervening, as the model warns, after the window has started to close.

Distilled from Our World in Data