Probabilities of success by clinical phase
Of a hundred compounds that begin a Phase I trial, around eight reach approval. This page explains where that number comes from, what happens in between, and when there is good reason to depart from it.
What each phase tests
| Phase | Participants | Question | Typical duration |
|---|---|---|---|
| I | 20–100, mostly healthy | Is it tolerated? At what dose? | 1–2 years |
| II | 100–500 patients | Does it work at all? | 2–3 years |
| III | 300–3,000 patients | Does it work better than the standard of care? | 3–5 years |
| Filing | — | Is the dossier complete and the benefit demonstrated? | 1–2 years |
The transition rates
The source is the joint analysis by BIO, Informa Pharma Intelligence and QLS Advisors, Clinical Development Success Rates and Contributing Factors 2011–2020: 12,728 phase transitions from 9,704 development programmes across 1,779 companies. The observation window is not the full decade: it runs from 1 January 2011 to 30 November 2020. It is the broadest publicly available analysis of its kind — and the basis of our default assumptions.
| Transition | Success rate |
|---|---|
| Phase I → II | 52.0% |
| Phase II → III | 28.9% |
| Phase III → filing | 57.8% |
| Filing → approval | 90.6% |
| Phase I → approval (cumulative) | 7.9% |
Why Phase II is the hardest hurdle
Because it is the first to test efficacy. Phase I tests tolerability, and compounds fail there less often, since preclinical toxicology catches most of it beforehand.
In Phase II the hypothesis meets patients for the first time. It fails for three reasons: the target is less involved in the disease than assumed; the effect exists but is too small; or it cannot be separated cleanly from placebo.
The practical consequence for valuation: value a Phase II company and more than 70% of what you are valuing is a scenario that will not happen. The price of such a company is therefore not a discount on a value but an expected value across two very different futures.
Why Phase III still fails so often
57.8% means four in ten programmes fail after a positive Phase II result, and after an investment that usually exceeds all preceding phases combined. The usual causes:
- The Phase II population was more narrowly selected than the Phase III population.
- The primary endpoint was changed — or was a surrogate measure in Phase II.
- Phase II was single-arm, with no control group. The apparent effect was partly selection.
- Insufficient statistical power: the effect exists, but is smaller than planned.
Two tables that are constantly confused
The report contains two breakdowns by therapeutic area, and they measure different things. Figure 2 shows the success rate for each individual phase. Figure 5b shows the cumulative likelihood of approval from Phase I — the product of all four transitions.
| Therapeutic area | Cumulative from Phase I (Fig. 5b) |
|---|---|
| Haematology | 23.9% |
| Metabolic | 15.5% |
| All indications | 7.9% |
| Oncology | 5.3% |
| Urology | 3.6% |
Both tables are sorted the same way and look confusingly alike. Metabolic offers the clearest example: the Phase I success rate is 61.8%, while the cumulative road to approval stands at 15.5%. One is a single jump, the other the whole distance.
Swap the numbers and you are out by a factor of four. It is the commonest error made when citing this report.
When departing from the average is justified
The industry average is a starting point, not a truth about any individual programme. There is a case for adjusting upwards where there is:
- a known, validated mechanism of action — a second compound in a class whose first member is approved carries no mechanism risk;
- a biomarker-selected patient population, for which the report gives the only clearly quantified deviation (see below);
- a label extension of an already approved compound into a new indication.
And downwards, which is the more frequent case:
- Oncology, at 5.3% from Phase I, sits well below the average.
- First-in-class targets with no clinical history.
- Single-arm Phase II data as the sole evidence of efficacy.
The biomarker effect — and what supports it
The analysis identifies a factor that shifts the averages noticeably. Programmes whose trials preselect patients by biomarker reach a cumulative likelihood of approval from Phase I of 15.9% against 7.6% without that preselection — twice as high. In the report's own words: "Drug development programs with trials employing patient preselection biomarkers have a two-fold higher LOA (15.9%) than those that do not (7.6%)."
Anyone using that number should know what it rests on. The difference arises almost entirely in a single transition, Phase II to Phase III: 46.3% against 28.3%. And there it rests on 149 biomarker-supported transitions out of 767 analysed — around 6% of the whole data set. The report states the limitation itself: "this analysis only represents a small subset of the overall data."
That does not make the figure worthless, quite the opposite: it is the best-evidenced exception to the average that this data set offers. It is simply not a second average.
Beware the 9.6% still doing the rounds
The previous edition of the same series — BIO together with Biomedtracker and Amplion, covering 2006–2015 — reported 63.2%, 30.7%, 58.1% and 85.3% per transition and a cumulative 9.6%. Those values have travelled for years into presentations, trade articles and company materials, and they are still there.
Anyone citing 9.6% today is citing the old edition. The current figure is 7.9%.
Where the numbers mislead
They are averages over nearly ten years and every therapeutic area. For a specific programme none of these values holds exactly. They are the anchor from which one departs with a reason — not the result.
They contain a survivorship bias. Programmes quietly discontinued and never reported are missing from the analysis. The true rates are likely to be lower.
They say nothing about timing. A trial delayed by eighteen months costs present value and cash, even if it eventually succeeds. Which is why this page always sits alongside cash runway: the question is not only whether the programme works, but whether the company lasts that long.
Approval is not reimbursement. In Germany, approval is followed by benefit assessment and price negotiation. An approved drug without a recognised added benefit will not reach the peak sales assumed for it.
How we use the numbers
Every biotech analysis states the probability applied with source and rationale, and gives a range of scenarios rather than a point estimate. We also name which figure of the report the value comes from — cumulative or per phase — because the two are routinely confused. Where our assumption departs from the industry average, we say why.
See rNPV for the calculation and How we value biotech companies for the full framework.
Sources
- BIO, Informa Pharma Intelligence, QLS Advisors: Clinical Development Success Rates and Contributing Factors 2011–2020. Observation window 1 January 2011 to 30 November 2020; 12,728 phase transitions, 9,704 programmes, 1,779 companies. Phase rates from Figure 2, cumulative values by therapeutic area from Figure 5b, biomarker analysis in the section on contributing factors. go.bio.org
- BIO, Biomedtracker, Amplion: Clinical Development Success Rates 2006–2015 — the previous edition, source of the 9.6% still quoted today.
As of: 11 August 2026 · Version 2