Athlete biological passport

Why do drug testers now watch an athlete's own blood values over years instead of looking for the drug itself?

For most of a century, catching a doper meant finding the drug: take a sample, run it through a mass spectrometer, and either the banned molecule is there or it is not. That is clean, direct evidence of the kind a tribunal likes.

So it is strange that the most consequential anti-doping tool of the last two decades gives that up entirely. The biological passport does not look for a drug at all. It watches an athlete's own blood and steroid values, sample after sample, for years, and eventually says something like: this pattern is not one your body produces. Why would anyone trade direct evidence for an inference?

Because the direct evidence stopped being available.

Some doping substances are chemically identical to what the body already makes. Testosterone administered is testosterone. Worse, autologous transfusion — withdrawing your own blood and putting it back later — introduces no foreign substance whatsoever. There is nothing to find, in principle, not merely in practice.

And where there is something to find, the window is short. Micro-dosing exploits exactly that: small, frequent doses that clear within hours, timed around known testing patterns. Add designer compounds the assay was never written for, and the direct approach is chasing a moving target with a fixed net.

So ask the obvious question: if the substance disappears, does its effect disappear with it? No — and that asymmetry is the whole idea. The point of blood doping is to raise oxygen-carrying capacity, and that elevation persists for weeks after the last trace of the drug is gone. The effect outlives the cause, so measure the effect.

Now the difficulty that stalled this approach for years. Physiological markers vary enormously between people. One athlete's ordinary haemoglobin concentration is another's suspicious result, and there is no population threshold that separates them. Cycling's old haematocrit ceiling shows the bind: it had to be framed as a health precaution rather than a doping sanction, because it could not tell a doper from someone naturally high while letting dopers with low baselines rise a long way beneath it.

Here is the move that rescues it. Ask not whether a value is unusual among people, but whether it is unusual for this person. An individual's haemoglobin is far more stable across months than the spread across a population of athletes. So if you sample the same athlete repeatedly, you can build a predicted range for that athlete — and because within-person variation is the smaller quantity, that personal range is much tighter than any population range could be. Each new sample updates it, in a Bayesian fashion: the more history, the narrower the band, and the smaller a deviation has to be before it stands out.

One further refinement makes it sharper. Rather than watching one marker, the passport watches combinations. Haemoglobin concentration and the proportion of young red cells normally move together — if you gain red cells, you have been making them. A high haemoglobin with suppressed production, or a normal haemoglobin with a burst of production, are the fingerprints of adding cells from outside or withdrawing them for later. A derived index combining the two — the OFF-score, haemoglobin concentration minus sixty times the square root of the reticulocyte percentage, a formula I state from recall — was built to catch that dissociation.

This is a close relative of the reasoning in the collection's piece on screening false positives, and it is worth naming exactly where they part. There, the trouble was the base rate: a test judged against the population flags mostly healthy people because most people are healthy. The passport's central design move is to change the reference class altogether — from other athletes to the athlete's own past. That does not abolish the false-positive problem, but it shrinks the variance the inference must see through, which is what makes a modest deviation meaningful at all.

The false positives still have to be managed. Thresholds are set at demanding specificity — around the 99 per cent and 99.9 per cent individual limits, from recall — because with many markers, many samples, and thousands of athletes, even a rare accident becomes common in absolute number. And a flagged profile is not a finding: it goes to independent experts, who must rule out altitude, illness, sample handling, and plasma volume shifts before agreeing unanimously that no physiological explanation fits.

The load-bearing claim is the one about variance: that an athlete's marker values are substantially more stable over time than the spread across athletes. If a large cohort of athletes with no exposure to doping were sampled longitudinally and produced atypical profiles at a rate far above the model's nominal specificity, the individual-baseline premise would be refuted — the personal band would not in fact be tighter than the population's, and every finding built on it would be unsafe.

Two candid limits. Its sensitivity is modest and contested: careful micro-dosing, and plasma volume manipulation that shifts concentration without changing red cell mass, can keep a doped athlete inside their own limits. And the model needs a baseline worth having, which is weakest where it would help most — young athletes still developing, and those rarely tested. The passport is better understood as raising the cost of doping than as reliably catching it.

• Why long-term storage and retrospective re-analysis of old samples changes a doper's calculation even more than the passport does.
• What burden of proof a statistical inference should meet in a sporting tribunal, and how that differs from a criminal court.

Every term the collection defines is gathered in the glossary.