How many advanced therapies are authorised in Europe?

Counting conventions · Advanced therapies · 27 July 2026

Ask a simple question of three serious sources, and watch what happens.

How many advanced therapy medicinal products — gene therapies, cell therapies, tissue engineering — are authorised in the European Union? The trade publication Labiotech writes that twenty are available in Europe. A review published in Frontiers in Medicine in 2025 counts twenty-seven. The quarterly report of the European Medicines Agency's Committee for Advanced Therapies, June 2026 edition, counts thirty-three.

Three answers. And, more unsettling still, none of them is wrong.

I spent several weeks rebuilding this count row by row, against the Agency's assessment reports, the Commission's union register and the Official Journal. The conclusion fits in one sentence: the three sources count different objects, at different dates, according to conventions none of them fully states.

Four conventions are enough to produce the entire gap.

The cut-off date. The most obvious, the most neglected. Two authorisations were granted in the first half of 2026 alone. A count stopped at the end of 2024 and one stopped in mid-2026 cannot coincide — and a figure published without a date becomes unusable six months later.

Granted, or in force. Do we count everything ever authorised, or only what still is? Nine authorisations out of thirty-three have ended. Depending on the convention, you write thirty-three or twenty-four, and both are exact.

Authorisations, or products. The finest trap. Zolgensma and Itvisma share the same active substance but are the subject of two distinct authorisations — one intravenous, one intrathecal, for different populations. A count by substance finds one; a count by authorisation finds two. Symmetrically, Durveqtix and Beqvez designate one and the same authorisation, simply renamed along the way: a database built on brand names will count it twice.

Authorised, or available. The one that matters to a patient, and the one counts almost always ignore. Roctavian retains a perfectly valid authorisation today although its holder announced a worldwide commercial withdrawal, explicitly unrelated to efficacy or safety. Skysona and Beqvez were authorised then withdrawn without ever having been marketed in the Union. Counting authorisations in force therefore overstates the supply actually accessible.

A count with no stated convention is not a figure: it is an opinion about a population that was never defined.

Once the conventions are set, the count closes with no remainder. As at 27 July 2026, counting one row per centralised authorisation: thirty-three granted since 2009, twenty-four in force, nine ended, twenty-two actually marketed. And the twenty-seven authorisations dated before 30 November 2024 correspond exactly to the Frontiers in Medicine count at that same date. The two reconcile — they diverged only by date.

The count is only the preliminary. The interest begins when you look at the exits.

Nine authorisations ended out of thirty-three. Seven of those nine exits are commercial. Only one is about efficacy — Alofisel, whose confirmatory trial failed to meet its primary endpoint. Only one is about manufacturing — MACI, whose sole authorised manufacturing site closed.

In other words: in this segment, when a product leaves the market after being authorised, the obstacle is almost never scientific. It is economic. The product works, the authorisation is there, and it leaves anyway.

A word on the 27 % figure, because it invites a misreading I would rather defuse myself. It is a crude rate: it mixes authorisations followed for seventeen years with others followed for a month. It reads neither as a five-year nor as a seven-year risk. The Kaplan-Meier estimate, which accounts for that censoring, gives 18 % at five years and 45 % at seven years. Likewise, the thirty-eight-month median between grant and termination is conditional on exits — it is not median survival, which is not reached.

Most valuation models for a pharmaceutical project stop at the probability of obtaining an authorisation. That is the event we model, the one that switches on the revenue in the spreadsheet.

Yet in this segment, the authorisation is not the end of the story in one case out of four. A model carrying no parameter for commercial survival after the grant structurally overstates the value of an advanced therapy asset. Three parameters follow directly from the list: the probability of termination after grant, the time to that termination, and the distribution of reasons.

This is precisely an assumption in the sense I mean it elsewhere — something that has to be true for the plan to hold, and that can now be confronted with a documented precedent rather than an intuition.

With one reservation I want to state plainly: thirty-three observations permit no fine estimation by subgroup. On so thin a sample, what is worth selling is not an average — it is a collection of named cases, a precedent defensible before a committee. That is also why the list is published in full, product by product, with the source of every field: the complete dataset is here, filterable, with all thirty-three authorisations and parameters that recalculate on the current selection.

Next time someone hands you a market figure, before asking whether it is right, ask what it counts. You will often find the question had never been put.