Imagine someone tells you: «The child is 108 centimetres tall.» Is that a lot? A little? Cause for concern? The question cannot be answered. Not because the number is imprecise — it is perfectly exact — but because the decisive part is missing: a comparison. Only the follow-up question «how old is the child?» turns the number into a statement. For a three-year-old, 108 centimetres is exceptionally tall; for an eight-year-old it is short.

Exactly this principle lies behind every percentile curve in a child health record — and behind the HBImed normative database. It is unspectacular, and that is why it is easily overlooked. But without it, every brain function analysis is blind.

A measurement is a number, not a verdict

From a resting EEG, a brain function analysis produces a whole series of measures: power in various frequency bands, connectivity measures, complexity and entropy values, microstate parameters. Each of these numbers is precise. And each on its own is as uninformative as «108 centimetres».

Because the decisive question is never: how high is the value? It is: how high is it compared with whom? An alpha value, a complexity measure, a connectivity measure become a finding only once one knows where the great majority of comparable people lie — and how far this one value departs from that.

That is the task of a normative database: it supplies the comparison group. For each measure it says how that measure is distributed across many other people. Only against this background can a single value be placed — as unremarkable, as at the edge, as clearly outside the usual.

Percentile instead of label

The result of this placement is ideally not a yes/no but a position. A value may lie in the uppermost range — higher than in the vast majority of the comparison group — or in the middle of the broad field in which almost everyone moves. This position is called the percentile: it states what proportion of the comparison group lies below the measured value.

The advantage of this way of thinking is that it stays honest. It does not claim «ill» or «healthy» but describes a position: «unusual» or «usual». What that position means — whether it means anything at all — is a second, clinical question. The normative database answers only the first. But without that first answer, the second hangs in the air.

Why the comparison group has to fit

Here lies the greatest source of error. A comparison group only helps if it fits the person measured — and with the brain it fits above all when the age is right. The child’s EEG changes dramatically across development; what is usual in a six-year-old would be conspicuous in an adult, and the other way round. Measuring a child’s findings against an adult norm leads one astray — as surely as one does not measure a toddler against a schoolchild’s growth curve.

A normative database is therefore not a single yardstick but a frame of reference structured by age (and, depending on the measure, by further characteristics). And that is why its size matters so much: the more people it contains, the more finely the comparison group can be cut, and the more robust the statement about a single value becomes. A norm resting on few people is a rough estimate; one that keeps growing becomes more accurate with every recording.

An example that makes this visible

How strong this age dependence is can be shown nicely with EEG microstates — those brief, stable voltage patterns in which the resting EEG rests before flipping abruptly into a different configuration. They are described by measures such as the duration of a state, how often it occurs and the share of time it takes up.

A longitudinal study by Kelsey and colleagues (Developmental Psychology, 2026) followed 383 children from infancy to about their seventh year across four measurement points. The result is instructive for our question: the duration of most microstates decreases systematically with age — the older child switches between states faster than the younger one. For other measures the changes run in different directions depending on the microstate class. And from about the age of three, differences between boys and girls appear in addition, and persist across all later measurement points.

The picture is therefore unambiguous: a microstate measure has no fixed «normal». What it means depends on age — and in part on sex. The same number can be unremarkable in a four-year-old and notable in a seven-year-old. Without a frame of reference structured by these characteristics, the value simply cannot be interpreted.

Incidentally, the authors draw from this no recommendation for ready-made norms but the opposite: further studies in more diverse populations are needed first. That too belongs to the picture — building a robust norm is work that is never quite finished.

What a normative database cannot do

Here as well, the value of the tool lies precisely in knowing its limits.

«Normal» does not mean «healthy». A percentile describes frequency, not health. A rare value can be harmless; a common value can be meaningful in a particular context. Statistics place a value; they do not diagnose.

A single conspicuous percentile is not yet a finding. If many measures are taken, one or another will lie far out purely by chance — that is the case in every healthy person. Only a coherent pattern, read in clinical context, carries a statement. A single extreme value picked out does not.

The norm is only as good as the people it is made of. Whoever is missing from the comparison group is less well served by the norm. A comparison group that is too small, or that does not represent a given population, leads to skewed placements. That is not an argument against normative databases but one for maintaining and extending them continuously.

And the last step stays with a person. The database says where a value lies. Whether that means anything is decided by a professional who brings the value together with everything else they know about the person.

The unremarkable core of personalised medicine

The whole idea of brain function analysis can be pinned to this one point. Personalised medicine does not mean putting every person into a drawer, but describing how this one brain works compared with many others. That description is not possible without a good basis for comparison. The normative database is therefore not an accessory to the method — it is its foundation.

It is also the reason why every new recording counts. Unlike a finished instrument, this foundation improves with every data set added. Whoever is examined today is measured against those who came before — and at the same time makes the placement more accurate for everyone who comes after.

Source

• Kelsey, C. et al. The development of EEG microstates from infancy into middle childhood and their associations with behavior. Developmental Psychology, 2026.